Andreas Bernhard Zeidler

Abstract Algebra
Rings, Modules, Polynomials, Ring Extensions, Categorical and Commutative Algebra February 15, 2012 (488 pages)

If you have read this text I would like to invite you to contribute to it: Comments, corrections and suggestions are very much appreciated, at abzeidler@gmx.de, or visit my homepage at www.mathematik.uni-tuebingen.de/ab/algebra/index.html

This book is dedicated to the entire mathematical society. To all those who contribute to mathematics and keep it alive by teaching it.

Contents
0 Prelude 0.1 About this Book . . . . . . . . . . . . . . . . . . . . . . . . . 0.2 Notation and Symbols . . . . . . . . . . . . . . . . . . . . . . 0.3 Mathematicians at a Glance . . . . . . . . . . . . . . . . . . . 1 Groups and Rings 1.1 Dening Groups 1.2 Permutations . . 1.3 Dening Rings . 1.4 Examples . . . . 1.5 First Concepts . 1.6 Ideals . . . . . . 1.7 Homomorphisms 1.8 Ordered Rings . 1.9 Products . . . . . 5 5 9 15 19 19 25 32 36 46 52 65 72 78 83 83 88 92 97 103 114 124 132 135 147 153 159 159 170 175 182 186 187 191 197 203 207

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

2 Commutative Rings 2.1 Maximal Ideals . . . . . . . . . 2.2 Prime Ideals . . . . . . . . . . . 2.3 Radical Ideals . . . . . . . . . . 2.4 Noetherian Rings . . . . . . . . 2.5 Unique Factorisation Domains . 2.6 Principal Ideal Domains . . . . 2.7 Lasker-Noether Decomposition 2.8 Finite Rings . . . . . . . . . . . 2.9 Localisation . . . . . . . . . . . 2.10 Local Rings . . . . . . . . . . . 2.11 Dedekind Domains . . . . . . . 3 Modules 3.1 Dening Modules . . . . 3.2 First Concepts . . . . . 3.3 Direct Sums . . . . . . . 3.4 Ideal-induced Modules . 3.5 Block Decompositions . 3.6 Dependence Relations . 3.7 Linear Dependence . . . 3.8 Homomorphisms . . . . 3.9 Isomorphism Theorems 3.10 Rank of Modules . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

3.11 Noetherian Modules . . . . . . . . . . . . . . . . . . . . . . . 213 3.12 Localisation of Modules . . . . . . . . . . . . . . . . . . . . . 221 4 Linear Algebra 4.1 Matix Algebra . . . 4.2 Elementary Matrices 4.3 Linear Equations . . 4.4 Multilinear Forms . 4.5 Determinants . . . . 4.6 Rank of Matrices . . 4.7 Canonical Forms . . 5 Spectral Theory 5.1 Metric Spaces . . 5.2 Banach Spaces . 5.3 Operatoralgebras 5.4 Diagonalisation . 5.5 Spectral Theorem 224 . 224 . 232 . 238 . 252 . 252 . 252 . 252 253 253 253 253 253 253 254 254 254 254 254

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

6 Structure Theorems 6.1 Associated Primes . . . 6.2 Primary Decomposition 6.3 The Theorem of Pr ufer . 6.4 Length of Modules . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

7 Polynomial Rings 7.1 Monomial Orders . . . . . 7.2 Graded Algebras . . . . . 7.3 Dening Polynomials . . . 7.4 The Standard Cases . . . 7.5 Roots of Polynomials . . . 7.6 Derivation of Polynomials 7.7 Algebraic Properties . . . 7.8 Gr obner Bases . . . . . . 7.9 Algorithms . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

255 . 255 . 261 . 268 . 268 . 268 . 268 . 268 . 268 . 268 269 . 269 . 269 . 269 . 269 . 269 . 269 . 269 . 269 270

8 Polynomials in One Variable 8.1 Interpolation . . . . . . . . . . . . 8.2 Irreducibility Tests . . . . . . . . . 8.3 Symmetric Polynomials . . . . . . 8.4 The Resultant . . . . . . . . . . . . 8.5 The Discriminant . . . . . . . . . . 8.6 Polynomials of Low Degree . . . . 8.7 Polynomials of High Degree . . . . 8.8 Fundamental Theorem of Algebra . 9 Group Theory

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

10 Multilinear Algebra 10.1 Multilinear Maps . . . . . . 10.2 Duality Theory . . . . . . . 10.3 Tensor Product of Modules 10.4 Tensor Product of Algebras 10.5 Tensor Product of Maps . . 10.6 Dierentials . . . . . . . . . 11 Categorial Algebra 11.1 Sets and Classes . . . . 11.2 Categories and Functors 11.3 Examples . . . . . . . . 11.4 Products . . . . . . . . . 11.5 Abelian Categories . . . 11.6 Homology . . . . . . . . 12 Ring Extensions 13 Galois Theory 14 Graded Rings 15 Valuations 16 Proofs - Fundamentals 17 Proofs - Rings and Modules

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

276 . 276 . 276 . 276 . 276 . 276 . 276 277 . 277 . 277 . 278 . 278 . 278 . 278 279 280 281 282 284 306 413

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

18 Proofs - Commutative Algebra

Chapter 0

Prelude
0.1 About this Book

The Aim of this Book Mathematics knows two directions - analysis and algebra - and any mathematical discipline can be weighted how analytical resp. algebraical it is. Analysis is characterized by having a notion of convergence that allows to approximate solutions (and reach them in the limit). Algebra is characterized by having no convergence and hence allowing nite computations only. This book now is meant to be a thorough introduction into algebra. Likewise every textbook on mathematics is drawn between two pairs of extremes: (easy understandability versus great generality) and (completeness versus having a clear line of thought). Among these contrary poles we usually chose (generality over understandability) and (completeness over a clear red line). Nevertheless we try to reach understandability by being very precise and accurate and including many remarks and examples. At last some personal philosophy: a perfect proof is like a perfect gem unbreakably hard, spotlessly clear, awlessly cut and beautifully displayed. In this book we are trying to collect such gemstones. And we are proud to claim, that we are honest about where a proof is due and present complete proofs of almost every claim contained herein (which makes this textbook very dierent from most others). This Book is Written for many dierent kinds of mathematicians: primarily is meant to be a source of reference for intermediate to advanced students, who have already had a rst contact with algebra and now closely examine some topic for their seminars, lectures or own thesis. But because of its great generality and completeness it is also suited as an encyclopaedia for professors who prepare their lectures and researchers who need to estimate how far a certain method carries. Frankly this book is not perfectly suited to be a monograph for novices to mathematics. So if you are one we think you can greatly prot from this book, but you will probably have to consult additional monographs (at a more introductory level) to help you understand this text.

Prerequisites We take for granted, that the reader is familiar with the basic notions of naive logic (statements, implication, proof by contradiction, usage of quantiers, . . . ) and naive set theory (Cantors notion of a set, functions, partially ordered sets, equivalence relations, Zorns Lemma, . . . ). We will present a short introduction to classes and the NBG axioms when it comes to category theory. Further we require some basic knowledge of integers (including proofs by induction) and how to express them in decimal numbers. We will sometimes use the eld of real numbers, as they are most probably wellknown to the reader, but they are not required to understand this text. Aside from these prerequisites we will start from scratch. Topics Covered We start by introducing groups and rings, immediately specializing on rings. Of general ring theory we will introduce the basic notions only, e.g. the isomorphism theorems. Then we will turn our attention to commutative rings, which will be the rst major topic of this book: we closely study maximal ideals, prime ideals, intersections of such (radical ideals) and the relations to localisation. Further we will study rings with chain conditions (noetherian and artinian rings) including the Lasker-Noether theorem. This will lead to standard topics like the fundamental theorem of arithmetic. And we conclude commutative ring theory by studying discrete valuation rings, Dedekind domains and Krull rings. Then we will turn our attention to modules, including rank, dimension and length. We will see that modules are a natural and powerful generalisation of ideals and large parts of ring theory generalises to this setting, e.g. localisation and primary decomposition. Module theory naturally leads to linear algebra, i.e. the theory of matrix representations of a homomorphism of modules. Applying the structure theorems of modules (the theorem of Pr ufer to be precise) we will treat canonical form theory (e.g. Jordan normal form). Next we will study polynomials from top down: that is we introduce general polynomial rings (also known as group algebras) and graded algebras. Only then we will regard the more classical problems of polynomials in one variable and their solvability. Finally we will regard polynomials in several variables again. Using Gr obner bases it is possible to solve abstract algebraic questions by purely computational means. Then we will return to group theory: most textbooks begin with this topic, but we chose not to. Even though group theory seems to be elementary and fundamental this is not quite true. In fact it heavily relies on arguments like divisibility and prime decomposition in the integers, topics that are native to ring theory. And commutative groups are best treated from the point of view of module theory. Never the less you might as well skip the previous sections and start with group theory right away. We will present the standard topics: the isomorphism theorems, group actions (including the formula of Burnside), the theorems of Sylow and lastly the p-q -theorem. However we are aiming directly for the representation theory of nite groups. The rst part is concluded by presenting a thorough introduction to what is called multi-linear algebra. We will study dual pairings, tensor products of modules (and algebras) over a commutative base ring, derivations and the module of dierentials. 6 0 Prelude

Thus we have gathered a whole bunch of separate theories - and it is time for a second structurisation (the rst structurisation being algebra itself). We will introduce the notions of categories, functors, equivalence of categories, (co-)products and so on. Categories are merely a manner of speaking nothing that can be done with category theory could not have been achieved without. Yet the language of categories presents a unifying concept for all the dierent branches of mathematics, extending far beyond algebra. So we will rst recollect which part of the theory we have established is what in the categorical setting. The categorial language is the right setting to treat chain complexes, especially exact sequences of modules. Finally we will present the basics of abelian categories as a unifying concept of all those separate theories. We will then aim for some more specialised topics: At rst we will study ring extensions and the dimension theory of commutative rings. A special case are eld extensions including the beautiful topic of Galois theory. Afterwards we turn our attention to ltrations, completions, zeta-functions and the Hilbert-Samuel polynomial. Finally we will venture deeper into number theory: studying the theory of valuations up to the theorem of RiemannRoche for number elds. Topics not Covered There are many instances where, dropping a niteness condition, one has to introduce some topology in order to pursue the theory further. Examples are: linear algebra on innite dimensional vector-spaces, representation theory of innite groups and Galois theory of innite eld extensions. Another natural extension would be to introduce the Zariski topology on the spectrum of a ring, which would lead to the theory of schemes directly. Yet the scope of this text is purely algebraic and hence we will usually stop at the point where topology sets in (but give hints for further readings).

The Two Parts Mathematics has a peculiarity to it: there are problems (and answers) that are easy to understand but hard to prove. The most famous example is Fermats Last Theorem - the statement (for any n 3 there are no nontrivial integers (a, b, c) Z3 that satisfy the equation an + bn = cn ) can be understood by anyone. Yet the proof is extremely hard to provide. Of course this theorem has no value of its own (it is the proof that contains deep insights into the structure of mathematics), but this is no general rule. E.g. the theorem of Wedderburn (every nite skew-eld is a eld) is easy and useful, but its proof will be performed using a beautiful trick-computation (requiring the more advanced method of cyclotomic polynomials). Thus we have chosen an unusual approach: we have separated the truth (i.e. denitions, examples and theorems) from their proofs. This enables us to present the truth in a stringent way, that allows the reader to get a feel for the mathematical objects displayed. Most of the proofs could have been given right away, but in several cases the proof of a statement can only be done after we have developed the theory further. Thus the sequel of theorems may (and will) be dierent from the order in which they are proved. Hence the two parts.

0.1 About this Book

Our Best Advice It is a well-known fact, that some proofs are just computational and only contain little (or even no) insight into the structure of mathematics. Others are brilliant, outstanding insights that are of no lesser importance than the theorem itself. Thus we have already included remarks of how the proof works in the rst part of this book. And our best advice is to read a section entirely to get a feel for the objects involved - only then have a look at the proofs that have been recommended. Ignore the other proofs, unless you have to know about them, for some reason. At several occasions this text contains the symbols () and ( ) . These are meant to guide the reader in the following ways: () As we have assorted the topics covered thematically (paying little attention to the sequel of proofs) it might happen that a certain example or theorem is far beyond the scope of the theory presented so far. In this case the reader is asked to read over it lightly (or even skip it entirely) and return to it later (after he has gained some more experience). ( ) On some very rare occasions we will append a theorem without giving a proof (if the proof is beyond the scope of this text). Such an instance will be marked by the black box symbol. In this case we will always give a complete reference of the most readable proof the author is aware of. And this symbol will be hereditarily, that is once we use a theorem that has not been proved any other proof relying on the unproved statement will also be branded by the black box symbol.

0 Prelude

0.2

Notation and Symbols

Conventions We now wish to include a set of the frequently used symbols, conventions and notations. In particular we clarify the several domains of numbers. First of all we employ the nice convention (introduced by Halmos) to write i as an abbreviation for if and only if. We denote the set of natural numbers - i.e. the positive integers including zero - by N := { 0, 1, 2, 3, . . . }. Further for any two integers a, b Z we denote the interval of integer numbers ranging from a to b by a . . . b := { k Z | a k b }. We will denote the set of integers by Z = N (N), and the rationals by Q = { a/b | a, b Z, b = 0 }. Whereas Z will be taken for granted, Q will be introduced as the quotient eld of Z. The reals will be denoted by R and we will present an example of how they can be dened (without proving their properties however). The complex numbers will be denoted by C = { a + ib | a, b R } and we will present several ways of constructing them. () We will sometimes use the Kronecker-Symbol (a, b) (in the literature this is also denoted by a,b ), which is dened to be (a, b) = a,b := 1 0 if a = b if a = b

In most cases a and b Z will be integers and 0, 1 Z will be integers, too. But in general we are given some ring (R, +, ) and a, b R. Then the elements 0 and 1 R on the right hand side are taken to be the zero-element 0 and unit-element 1 of R again. We will write A X to indicate that A is a subset of X and A X will denote strict inclusion (i.e. A X and there is some x X with x A). For any set X we denote its power set (i.e. the set of all its subsets) by P (X ) := { A | A X }. And for a subset A X we denote the complement of A in X by CA := X \ A. Listing several elements x1 , . . . , xn X of some set X , we do not require these xi to be pairwise distinct (e.g. x1 = x2 might well happen). Yet if we only give explicit names xi to the elements of some previously given subset A = { x1 , . . . , xn } X we already consider the xi to be pairwise distinct (that is xi = xj implies i = j ). Note that if the xi (not the set { x1 , . . . , xn }) have been given, then { x1 , . . . , xn } may hence contain fewer than n elements! Given an arbitrary set of sets A one denes the grand union A and the grand intersection A to be the set consisting of all elements a that are contained in one (resp. all) of the sets A A, formally A := { a | A A : a A } A := { a | A A : a A } 0.2 Notation and Symbols 9

Note that A only is a well-dened set, if A = is non-empty. A wellknown special case of this is the following: consider any two sets A and B and let A := { A, B }. Then A B = A and A B = A. This notion is just a generalisation of the ordinary union and intersection to arbitrary collections A of sets. If X and Y are any sets then we will denote the set of all functions from X to Y by F (X, Y ) = Y X = { f | f : X Y }. And for any such function f : X Y : x f (x) we will denote its graph (note that from the set-theoretical point of view f is its graph) by (f ) := { (x, f (x)) | x X } X Y

Once we have dened functions, it is easy to dene arbitrary cartesian products. That is let I = be any non-empty set and for any i I let Xi be another set. Let us denote the union of all the Xi by X X :=
i I

Xi = { x | i I : x Xi }

Then the cartesian product of the Xi consists of all the functions x : I X such that for any i I we have xi := x(i) Xi . Note that thereby it is customary to write (xi ) in place of x. Formally Xi := { x : I X : i xi | i I : xi Xi }
iI

If I = is any index set and Ai X is a non-empty Ai = subset of X (where i I ) then the axiom of choice states that there is a function a : I X such that for any i I we get a(i) Ai . In other words the product of non-empty sets is non-empty again. It is a remarkable fact that this seemingly trivial property has far-ung consequences, the lemma of Zorn being the most remarkable one. Xi =
iI

i I : Xi =

Let X = be a non-empty set, then a subset of the form R X X said to be a relation on X . And in this case it is customary to write xRy instead of (x, y ) R. This notation will be used primarily for partial orders and equivalence relations (see below). Consider any non-empty set X = again. Then a relation on X is said to be an equivalence relation on X , i it is reexive, symmetric and transitive. Formally that is for any x, y and z X we get x=y xy x y, y z = = = xy yx xz

10

0 Prelude

And in this case we dene the equivalence class [x] of x to be the set of all y X being equivalent to x, that is [x] := { y X | x y }. And the set of all equivalence classes is denoted by X/ := { [x] | x X }. Example: if f : X Y is any function then we obtain an equivalence relation on X by letting x y : f (x) = f (y ). Then the equivalence class of x X is just the bre [x] = f 1 (f (x)). Consider a non-empty set X = once more. Then a family of subsets P P (X ) is said to be a partition of X , i for any P , Q P we obtain the statements X P P =Q = = = P Q= P

Example: if is an equivalence relation on X , then X/ is a partition of X . Conversely if P is a partition of X , then we obtain an equivalence relation on X by letting x y : P P : x P and y P . Hence there is a one-to-one correspondence between the equivalence relations on X and the partitions of X given by X/. Consider any non-empty set I = , then a relation on I is said to be a partial order on I , i it is reexive, transitive and anti-symmetric. Formally that is for any i, j and k I we get i=j i j, j k i j, j i = = = ij ik i=j

And is said to be a total or linear order i for any i, j I we also get i j or j i (that is any two elements contained in I can be compared). Example: for any set X the inclusion relation is a partial (but not linear) order on P (X ). If now is a linear order on I , then we dene the minimum and maximum of i, j I to be i j := i j if i j if j i i j := j i if i j if j i

Now consider a partial order on the set X and a subset A X . Then we dene the set A of minimal respectively the set A of maximal elements of A to be the following A := { a A | a A : a a = a = a } A := { a A | a A : a a = a = a } And an element a A is said to be a minimal element of A. Likewise a A is said to be a maximal element of A. Note that in general it may happen that A has several minimal (or maximal) elements or even none at all (e.g. N = { 0 } and N = ). For a linear order minimal (and maximal) elements are unique however. 0.2 Notation and Symbols 11

 Finally is said to be a well-ordering on the set X , i is a linear order on X such that any non-empty subset A X has a (already unique) minimal element. Formally that is = A X a A such that a A : a a Let X be any set, then the cardinality of X is dened to be the class of all sets that correspond bijectively to X . Formally that is |X | := { Y | : X Y bijective } Note that most textbooks on set theory dene the cardinality to be a certain representative of our |X | here. However the exact denition is of no importance to us, what matters is comparing cardinalities: we dene the following relation between cardinals: |X | |Y | |X | = |Y | : : : X Y injective : Y X surjective : X Y bijective |X | |Y | and |Y | |X |

Note that the rst equivalence can be proved (as a standard exercise) using equivalence relations and a choice function (axiom of choice). The second equivalence is a rather non-trivial statement called the equivalence theorem of Bernstein. However these equivalences grant that has the properties of a partial order, i.e. reexivity, transitivity and anti-symmetry. Suppose x1 , x2 , . . . , xn are pairwise distinct elements. Then the set X := { x1 , . . . , xn } clealy has n elements. Using cardialities this would be expressed, as | { x1 , . . . , xn } | = |1 . . . n| In this case 1 . . . n X : i xi could be used as a bijection. Yet this is a bit cumbersome and hence we introduce another notation - if x1 , x2 , . . . , xn are pairwise distinct (that is xi = xj = i = j ) we let # { x1 , . . . , xn } := n In particular # = 0. Hence for any nite set X we have dened the number of elements #X . And for innite sets we simply let #X := Thereby a set X is said to be nite i there is some n N such that X (1 . . . n), that is i #X = n. And we usually denote the set of all nite subsets of X by (X ) := { A X | #A < }

12

0 Prelude

Likewise X is said to be innite, i N can be embedded into X , formally N X or in other words |N| |X |. Note that thereby we obtain a rather astounding equivalency of the statements (a) X is innite (b) |X | = |X X | (c) |X | = |(X )| If X and Y are nite, disjoint sets, then clearly #(X Y ) = (#X ) + (#Y ). Likewise for any nite sets X and Y we have #(X Y ) = (#X ) (#Y ), #P (X ) = 2#X and #F (X, Y ) = (#Y )#X . We will use these properties to extend the ordinary arithmetic of natural numbers to the arithmetic of cardinal numbers, by dening |X | + |Y | := |X Y|

|X | |Y | := |X Y | 2|X | := |P (X )| |Y ||X | := |F (X, Y )| for arbitrary sets X and Y . Note that refers to the disjoint union of X any Y , that is X Y := ({ 1 } X ) ({ 2 } Y ). This is not necessary, if X and Y are disjoint, in this case we could have taken X Y . However in general we have to provide a way of seperating X and Y and this nicely performed by X Y . It is a non-surprising but also non-trivial fact, that the power set of X is truly larger that X |X | < |P (X )| () We will introduce and use several dierent notions of substructures and isomorphy. In order to avoid eventual misconceptions, we emphasise the kinds of structures regarded by applying subscripts to the symbols and of substructures and = of isomorphy. E.g. we will write R r S to indicate, that R is a subring of S , a i R to indicate that a is an ideal of R and R =r S to indicate that R and S are isomorphic as rings. Note that the latter is dierent from R =m S (R and S are isomorphic as modules). And this can well make sense, if R r S is a subring, then S can be regarded as an R-module, as well. We will use the same subscripts for generated algebraic substructures, i.e. r for rings, i for ideals and m for modules.

0.2 Notation and Symbols

13

Notation A, B , C D, E , F G, H I, J, K L, M , N P, Q R, S , T U, V , W X, Y , Z a, b, c d, e e f , g, h i, j , k , l m, n p, q r, s s, t, u u, v , w x, y , z , , , , , , , , ,
a, b, c u, v, w p, q r, t m, n f, g , h

matrices, algebras and monoids elds groups and monoids index sets modules subsets and substructures rings (all kinds of) vectorspaces, multiplicatively closed sets arbitrary sets elements of rings degree of polynomials, dimension neutral element of a (group or) monoid functions, polynomials and elements of algebras integers and indices natural numbers residue classes elements of further rings polynomial variables elements of vectorspaces elements of groups and modules multi-indices (canonical) monomorphisms eigenvalues (canonical) epimorphisms permutations homomorphisms isomorphisms (xed) nite set ideals other ideals prime ideals other prime ideals maximal ideals fraction ideals (ordered) bases (canonical = euclidean) bases

A, B , C E, F

14

0 Prelude

0.3

Mathematicians at a Glance

In the following we wish to give a short list of several mathematicians whos names are attributed to some particular denition or theorem in honour of their outstanding contributions (to the eld of algebra). Naturally this list is far from being exhaustive, and if any person has been omitted this is solely due to the authors uninformedness. Likewise it is impossible to boil down an entire life to a few short sentences without crippling the biography. Nevertheless the author is convinced that it would be even more ignorant to make no dierence between words like noetherian and factorial. This is the reason why we chose not to say nothing about these avatars of mathematics. ?? Abel ?? Akizuki ?? Andr e Emil Artin ?? Auslander ?? Bernstein ?? B ezout ?? Brauer Nicolas Bourbaki (??) ?? Buchsbaum ?? Cauchy ?? Cayley ?? Chevalley ?? Cohen ?? Dedekind ?? Eisenbud ?? Euclid Evariste Galois ?? Fermat Carl Friedrich Gauss ?? Gorenstein ?? Goto ?? Gr obner ?? Grothendieck 0.3 Mathematicians at a Glance 15

 ?? Hensel David Hilbert ?? Hironaka ?? Hochster ?? Hopf ?? Jacobi ?? Jacobson ?? Kaplansky ?? Kei-ichi ?? Kronecker ?? Krull ?? Lagrange ?? Lasker ?? Legendre ?? Leibnitz ?? Macaulay ?? Mori ?? Nagata ?? Nakayama Johann von Neumann Isaac Newton Emmy Noether ?? Northcott ?? Ostrowski ?? Pr ufer ?? Pythagoras ?? Quillen ?? Ratli ?? Rees ?? Riemann ?? Rotthaus 16 0 Prelude

 ?? Schur ?? Serre ?? Sono ?? Stanley Bernd Sturmfels ?? Sylvester ?? Watanabe ?? Weber ?? Wedderburn ?? Weierstrass Andr e Weil ?? Yoneda ?? Zariski ?? Zorn

0.3 Mathematicians at a Glance

17

Part I

The Truth

Chapter 1

Groups and Rings

1.1 Dening Groups

The most familiar (and hence easiest to understand) objects of algebra are rings. And of these the easiest example are the integers Z. On these there are two operations: an addition + and a multiplication . Both of these operations have familiar properties (associativity for example). So we rst study objects with a single operation only - monoids and groups - as these are a unifying concept for both addition and multiplication. However our aim solely lies in preparing the concepts of rings and modules, so the reader is asked to venture lightly over problems within this section until he has reached the later sections of this chapter. (1.1) Denition: Let G = be any non-empty set and a binary operation on G, i.e. is a mapping of the form : G G G : (x, y ) xy . Then the ordered pair (G, ) is said to be a monoid i it satises the following properties (A) x, y, z G : x(yz ) = (xy )z (N) e G x G : xe = x = ex Note that this element e G whose existence is required in (N) then already is uniquely determined (see below). It is said to be the neutral element of G. And therefore we may dene: a monoid (G, ) is said to be a group i any element x G has an inverse element y G, that is i (I) x G y G : xy = e = yx Note that in this case the inverse element y of x is uniquely determined (see below) and we hence write x1 := y . Finally a monoid (or group) (G, ) is said to be commutative i it satises the property (C) x, y G : xy = yx (1.2) Remark: We have to append some remarks here: rst of all we have employed a function : G G G. The image (x, y ) of the pair (x, y ) G G has been written in an unfamiliar way however xy := (x, y ) 19

If you have never seen this notation before it may be somewhat startling, in this case we would like to reassure you, that this actually is nothing new - just have a look at the examples further below. Yet this notation has the advantage of restricting itself to the essential. If we had stuck to the classical notation such terms would be by far more obfuscated. E.g. let us rewrite property (A) in classical terms x, (y, z ) = (x, y ), z

It is easy to see that the neutral element e of a monoid (G, ) is uniquely determined: suppose that another element f G would satisfy x G : xf = x = f x, then we would have ef = e by letting x = e. But as e is a neutral element we have ef = f by (N) applied with x = f . And hence e = f are equal, i.e. e is uniquely determined. Hence in the following we will reserve the letter e for the neutral element of the monoid regarded without specically mentioning it. In case we apply several monoids at once, we will name the respective neutral elements explictly. Property (A) is a very important one and hence it has a name of its own: associativity. A pair (G, ) that satises (A) only also is called a groupoid. We will rarely employ these objects however. Suppose (G, ) is a monoid with the (uniquely determined) neutral element e G. And suppose x G is some element of G, that has an inverse element. Then this inverse is uniquely determined: suppose both y and z G satisfy xy = e = yx and xz = e = zx. Then the asociativity yields y = z , as we may compute y = ye = y (xz ) = (yx)z = ez = z Another important consequence of the associativity is the following: consider nitely many elements x1 , . . . , xn G (where (G, ) is a groupoid at least). Then any application of parentheses to the product x1 x2 . . . xn produces the same element of G. As an example consider x1 (x2 (x3 x4 )) = x1 ((x2 x3 )x4 ) = (x1 (x2 x3 ))x4 = ((x1 x2 )x3 )x4 In every step of these equalities we have only used the associativity law (A). The remarkable fact is that any product of any n elements (here n = 4) only depends on the order of the elements, not on the bracketing. Hence it is costumary to omit the bracketing altogether x1 x2 . . . xn := (x1 x2 ) . . . xn G

And any other assingment of pairs (without changing the order) to the elements xi would yield the same element as x1 x2 . . . xn . A formal version of this statement and its proof are given in chapter 16 of this book. The proof clearly will be done by induction on n 3, the foundation of the induction precisely is the law of associativity. 20 1 Groups and Rings

 Suppose (G, ) is any groupoid, x G is an element and 1 k N, then we abbreviate the k -nary product of x by xk , i.e. we let xk := xx . . . x (k times) If (G, ) even is a monoid (with neutral element e) it is customary to dene x0 := e. Thus in this case xk G is dened for all k N. Now suppose that x even is invertible (e.g. if (G, ) is a group), then we may even dene xk G for any k Z. Suppose 1 k N, then xk := x1
k

In a commutative groupoid (G, ) we may even change the order in which the elements x1 , . . . , xn G are multiplied. I.e. if we are given a bijective map : 1 . . . n 1 . . . n on the indices 1 . . . n we get x(1) x(2) . . . x(n) = x1 x2 . . . xn The reason behind this is the following: any permutation can be decomposed into a series of transpositions (this is intuitively clear: we can generate any order of n objects by repeatedly interchanging two of these objects). In fact any transposition can be realized by interchanging adjacent objects only. But any transposition of adjacent elements is allowed by property (C). A formal proof of this reasoning will be presented in chapter 16 again. (1.3) Example: The most familiar example of a (commutative) monoid are the natural numbers under addition: (N, +). Here the neutral element is given to be e = 0. However 1 N has no inverse element (as for any a N we have a + 1 > 0) and hence (N, +) is no group. The integers however are a (commutative) group (Z, +) under addition. The neutral element is zero again e = 0 and the inverse element of a Z is a. Thus N is contained in a group N Z. Next we regard the non-zero rationals Q := { a/b | 0 = a, b Z }. These form a group under multiplication (Q , ). The neutral element is given to be 1 = 1/1 and the inverse of a/b Q is b/a. Consider any non-empty set X = . Then the set of maps from X to X , which we denote by F (X ) := { | : X X }, becomes a monoid under the composition of maps (F (X ), ) (as the composition of functions is associative). The neutral element is the identity map e=1 1X which is given to be 1 1X : x x. Consider any non-empty set X = again. Then the set of bijective maps SX := { : X X | bijective } F (X ) on X even becomes a group under the composition of maps (SX , ). The neutral element is the identity map e = 1 1X again and the inverse of SX is the inverse function 1 . This will be continued in the next section. 1.1 Dening Groups 21

 () A special case of the above is the set of invertible (n n)-matrices gln (E ) := { A matn (E ) | det(A) = 0 } over a eld (E, +, ). This is a group (gln (E ), ) under the multiplication of matrices.

(1.4) Remark: Consider a nite groupoid (G, ), that is the set G = { x1 , . . . , xn } is nite. Then the composition can be given by a table of the following form x1 x2 . . . x1 x1 x1 x2 x1 . . . x2 x1 x2 x2 x2 . . . ... ... ... xn x1 xn x2 xn . . . x n xn

xn xn x1 xn x2 . . .

Such a table is also known as the Cayley diagram of G. As an example consider a set with 4 elements K := { e, x, y, z }, then the following diagram determines a group structure on K ((K, ) is called the Klein 4-group ). e x y z e e x y z x x e z y y y z e x z z y x e

(1.5) Proposition: (viz. 286) Let (G, ) be any group (with neutral element e), x, y G be any two elements of G and k , l Z. Then we obtain the following identities e1 = e x1 xy
l 1

= x
1

= y 1 x1

xk xl = xk+l xk = xkl

In particular the inversion i : G G : x x1 of elements is a selfinverse, bijective mapping i = i1 on the group G. If now xy = yx do commute then for any k Z we also obtain xy = yx = (xy )k = xk y k

22

1 Groups and Rings

(1.6) Denition: If (G, ) and (H, ) are any two groups, whose neutral elements are e G and f H respectively, then a map : G H is said to be a homorphism of groups or group-homorphism, i it satises x, y G we get (xy ) = (x)(y ) And in this case already links the neutral elements and maps inverse elements to the inverse, that is it satises (e) = f and x G we get (x1 ) = (x)1 Let us introduce the set of all group-homorphisms from G to H to be called ghom(G, H ) := { : G H | x, y G : (xy ) = (x)(y ) } Prob clearly (e) = (ee) = (e)(e) but as H is a group we may multiply by (e)1 to nd f = (e). Therefore f = (e) = (xx1 ) = (x)(x1 ). Likewise we get f = (x1 )(x) and hence (x1 ) = (x)1 . (1.7) Proposition: (viz. 287) Let (G, ) be a group with neutral element e G, then a subset P G is said to be a subgroup of G (written as P g G) i it satises e x, y P xP = = P xy P x1 P

In other words P G is a subgroup of G i it is a group (P, ) under the operation inherited from G. And in this case we obtain an equivalence relation on G by letting (for any x, y G) xy : y 1 x P

And for any x G the equivalence class of x is thereby given to be the coset xP := [x] = { xp | p P }. We thereby dene the index [G : P ] of P in G to be the following cardinal number := G [ G : P ] := G P P And thereby we nally obtain the following identity of cardinals which is called the Theorem of Lagrange (and which means G (G/P ) P ) | G| = [ G : P ] | P | G

1.1 Dening Groups

23

(1.8) Proposition: (viz. 288) Let (G, ) be any monoid with neutral element e and X G be any subset of G. Then the intersection of all submonoids of G that contain X is a submonoid of G again X
o

:=

{ P o G | X P }

We call X o the monoid generated by X G. And letting Xe := X { e } we can even give an explicit description of X o to be the following set X
o

= { x1 x2 . . . xn | n N, x1 , . . . , xn Xe }

If G even is a group, then we can regard the intersection of all subgroups of G that contain X and this is a subgroup of X again X
g

:=

{ P o G | X P }

Likewise ee call X g the group generated by X G. And letting X := X { e } x1 | x X we can even give an explicit description of X g to be the following set X
g

= { x1 x2 . . . xn | n N, x1 , . . . , xn X }

(1.9) Proposition: (viz. 289) If (G, ) is a group and x G is any element of G, then the above proposition tells us that the subgroup generated by x is given to be x
g

xk | k Z

If now G is nite, that is n := #G N, then x g (being a subset) is nite too. In particular we may dene the order of x to be number of elements of x g . And thereby we nd ord(x) := # x
g

= min{ 1 k N | xk = e }

That is k = ord(x) is the minimal number 1 k N such that xk = e. And the theorem of Lagrange even tells us that k divides n, that is ord(x) | #G

24

1 Groups and Rings

1.2

Permutations

In this section we will study the most important and most general example of groups: permutations. Permutation groups will not become important until we reach determinants in chapter 4. So a novice might well skip the entire section and return to it only later. In fact we even recommend this approach, as we deem the basics of ring theory to be easier than the basics of group theory. Also we will use some notation (namely products and the faculty) that will only be introduced in sections 1.3 and 1.5 respectively. Later in section ?? we will introduce group actions, that are a generalisation of the permutation action. Also permutations are used to study cojugation in a group and they are occasionally useful in other areas as well. In fact they will be applied in Galois, representation and invariant theory. (1.10) Denition: (viz. 289) If X = is any non-empty set then we dene the group SX of permutations of X to be the set of all bijective maps on X , that is SX := { : X X | is bijective } Thereby SX truly becomes a group (SX , ) under usual the composition of functions. An in case of X = 1 . . . n we will write Sn := { : (1 . . . n) (1 . . . n) | is bijective }

(1.11) Denition: Fix n N with n 1 and consider the perumation group (Sn , ). Then, for any permutation Sn and any i 1 . . . n, we dene x( ) := { i 1 . . . n | (i) = i } dom( ) := { i 1 . . . n | (i) = i } cyc(, i) := k (i) | k N

(, i) := #cyc(, i) Thereby x( ) is called the xed point set of and its complement dom( ) is called the domain of . Also cyc(, i) is called the cycle of i under and (, i) is said to be the length of that cycle. (1.12) Proposition: (viz. 289) (i) If X is a nite set, then the group (SX , ) is nite again, in fact it contains precisely (#X )! elements, formally that is #SX = (#X )!

1.2 Permutations

25

(ii) If (G, ) is any group, then G can be embedded into (SG , ). To be precise the following map is well-dened and injective L : G SG : g Lg Lg : G G : x gx and a homorphism of groups, that is: Le = 1 1G and for any g , h G we get Lgh = Lg Lh . Nota in this sense the permutation groups SX are the most general groups whatsoever.

(1.13) Proposition: (viz. 290) Fix n N with n 1 and consider the perumation group (Sn , ). Then, for any permutations and Sn and any i 1 . . . n, we nd the following statements (i) As k | k Z is a subgoup of the nite group Sn we nd that the cycle cyc(, i) if nite, more precisely (, i) ord( ) n! (ii) The domain of is stable under , that is for any a dom( ) we nd that (a) dom( ) again, such that (dom( )) = dom( ) In particular, if i dom( ) is contained in the domain of , then k (i) is contained in the domain of , for any power k N, such that cyc(, i) dom( ) (iii) If and share a common domain D 1 . . . n and agree on D, then they already are equal. Formally that is the equivalence D := dom( ) = dom( ) a D : (a) = (a) (iv) If the domains of and are disjoint, then = =

and commute, formally =

dom( ) dom( ) =

(1.14) Denition: Fix n N with n 1 and consider an -tuple (i1 , . . . , i ) of natural numbers ik 1 . . . n that are pariwise distinct (that is ij = ik = j = k ). Then the tuple (i1 , . . . , i ) gives rise to a permuation Sn by letting a ik+1 i1 if a (1 . . . n) \ { i1 , . . . , i } if a = ik for some k 1 . . . 1 if a = i 1 Groups and Rings

(a) :=

26

This permutation Sn is called a cycle of length and (by slight abuse of notation, as n is not specied) we will denote this cycle as (i1 i2 . . . i ) := A cycle of length 2 is said to be a transposition of Sn . And by denition it interchanges two numbers i = j . That is, if = (i j ) then (i) = j and (j ) = i and (a) = a for any other a 1 . . . n. (1.15) Remark: (viz. 291) The notation of cycles is ambigious as it is of no eect with which element the cycle starts, as long as the ordering of elements is not changed. That is (i1 i2 . . . i ) and (i2 i3 . . . i i1 ) and so on until (i i1 i2 . . . i 1 ) all yield the same permutation. In particular we have (i j ) = (j i) for any transposition. Any cycle can be written as a composition of transpositions. To be precise: given the cycle (i1 i2 . . . i ) we can rewrite this one, as (i1 i2 . . . i ) = (i1 i )(i1 i
1 ) . . . (i1 i2 )

Any transposition can be written as a composition of transpositions of adjacent elements. To be precise given any i, j 1 . . . n with i < j we can rewrite (i j ) = (i i + 1)(i + 1 i + 2) . . . (j 1 j )(j 2 j 1) . . . (i i + 1) If = (i1 i2 . . . i ) Sn is a cycle of length , then for any other permutation Sn the conjugate 1 is another cycle of length . In fact we get 1 = ( (i1 ) (i2 ) . . . (i )) (1.16) Proposition: (viz. 292) Fix n N with n 1 and consider the perumation group (Sn , ). If now Sn is any permutation with = 1 1 then the following ve statements are equivalent: (a) is a cycle, that is there is some 2 and some pairwise distict i1 , i2 , . . . , i 1 . . . n (that is ij = ik = j = k ) such that = ( i1 i2 . . . i ) (b) Any two elements a, b dom( ) in the domain of are linked, i.e. there is some k N such that b = k (a). (c) For any element i dom( ) the domain of already is generated by the orbit of i under , that is dom( ) = k (i) | k N 27

1.2 Permutations

(d) There is some i 1 . . . n such that the domain of is the orbit of i under , that is dom( ) = k (i) | k N

(e) If := ord( ) denotes the order of , then there is some i 1 . . . n such that the domain of is given to be dom( ) = k (i) | k 0 . . . ( 1)

And in this case the length of the cycle = (i1 i2 . . . i ) is precisely the order of that is we get the following equality ord( ) = in (a)

(1.17) Proposition: (viz. 294) Fix n N with n 1 and consider the perumation group (Sn , ). Then, for permutation Sn and any i 1 . . . n, we nd the following statements (i) We obtain an equivalency relation on 1 . . . n by letting i j if and only if there is some k N such that j = k (i). And the equivalence classes under this relation are precisely [i] = cyc(, i) = i, (i), . . . , k(i) (i)

where k (i) := (, i) 1. In particular cyc(, i) is -invariant, formally that is (cyc(, i)) = cyc(, i). Let us now regard i : 1 . . . n 1 . . . n (a) a if a cyc(, i) if a cyc(, i)

Then i is precisely the cycle induced by on i 1 . . . n, formally again i = (i (i) . . . k(i) (i)) And if i(1), . . . , i(r) is a system of representants of dom( ) under the relation , that is dom( ) is the disjoint union of the cycles cyc(, i(1)) to cyc(, i(r)), then can be written as the composition of the cycles i(i) to i(r) (in any order, as all the cycles are mutually commutative)
r r

dom( ) =
j =1

cyc(, i(j ))

=
j =1

i( j )

(ii) Let 1 , . . . , k Sn and 1 , . . . , t Sn be two sets of cycles, with pairwise disjoint domains, that is i = j 1...k r = s 1...t 28 = = dom(i ) dom(j ) = dom(r ) dom(s ) = 1 Groups and Rings

If now the compositions 1 . . . k = 1 . . . t are equal, then we already get k = t and there is some permutation Sr such that for any i 1 . . . k we get i = (i)
k t

i =
i=1 r=1

i 1 . . . k : i = (i)

(iii) For any permutation Sn with = 1 1 there is a uniquely determined cyclic decomposition. That is there are uniquely determined cycles 1 , . . . , r Sn with pairwise disjoint domains (for any i, j 1 . . . r that is i = j = dom(i ) dom(j ) = ), such that can be written as a composition of these cycles
r

=
i=1

(iv) Any permutation Sn is a product of transpositions of two adjacent numbers. That is there are nitely many i0 , . . . is N such that for any j 1 . . . r we get |ij ij 1 | = 1 and also
s

(ij 1 ij )
j =1

(v) Two permutations and Sn share the same cyclic structure if and only if they are conjugates of each other. Formally that is the equivalence of the following two statements (a) If we decompose = 1 . . . k and = 1 . . . t into cycles, i and r respectively, with pairwise disjoint domains (as in (ii)), then we get k = t and these cycles are of the same lengths. That is there is a permutation Sk such that i 1 . . . k : ord(i ) = ord((i) ) (b) There is some permutation Sn such that = 1 . (1.18) Example: Cyclic Decomposition: Let us present a short example of a cyclic decomposition in 1 . . . 9. To dene or permutation S9 we present a short table 1 2 3 4 5 6 7 8 9 i (i) 3 4 1 8 7 6 9 2 5 First of all it is clear that 6 is xed under , hence it will not appear in our cyclic decomposition (it would yield a cycle of length 1, namely (6)). Now begin by 1 and extract its cycle. We see that 1 3 1 so our rst cycle is (1 3). The next number not dealt with is 2, which is cycled like 2 4 8 2. So the second cycle is (2 4 8). As 3 and 4 occured in the rst and second cycles respectively, we continue with 5 which is moved like 5 7 9 5. Thus the third cycle is (5 7 9). 1.2 Permutations 29

As 6 is xed and 7 . . . 9 have already been taken care of we have exhausted the whole domain of . Altogether we have seen that = (1 3)(2 4 8)(5 7 9) Note that these cycles are uniquely determined even if there are several ways of putting them, e.g. (2 4 8) = (4 8 2) = (8 2 4). Also it is possible to commute these cycles arbitrarily, without changing . Finally we know, that any other permutation S9 is conjugated to , if and only if it consists of 3 cycles of lengths 2,3,3 respectively. (1.19) Denition: Fix any n N with n 1. If now is any permutation then we dene its signum to be the following number sgn( ) :=
1i<j n

(j ) (i) ji

It will turn out that the signum is contained in sgn( ) Z = { 1, +1 }. A permutation with sgn( ) = 1 is said to be even, whereas a permutation with sgn( ) = 1 is said to be odd. The set of all even permutations will be denoted, by An := { Sn | sgn( ) = 1 }

(1.20) Remark: The product, that has been used to dene the signum of might be a little obfuscated at a rst glance. But in fact the indices simply run over the set := (i, j ) (1 . . . n)2 | i < j . And as the multiplication in Q is commutative the ordering of this set has no eect on the result of the product. In yet other words we have
n1 n

sgn( ) =
i=1 j =i+1

(j ) (i) ji

As an example let us consider the case = (1 3) S4 . Then the signum of the transposition can be computed, as ( ) = 23 13 43 12 42 41 = 1 21 31 41 32 42 43

(1.21) Proposition: (viz. 297) (i) If Sn is any permutation, then the signum of can be evaluated by couning the number of inversions of . That is the signum is given to be sgn( ) = (1)I () where I ( ) := # (i, j ) (1 . . . n)2 | i < j and (i) > (j )

30

1 Groups and Rings

(ii) Let Sn be a permutation and = 1 . . . r be its decomposition into cycles with pairwise disjoint domains. Let (i) := ord(i ) be the length of the i-th cycle. Then the order of is the least common multiple of the (i) and we can easily compute the signum of as well ord( ) = lcm( (1), . . . , (r))
r

sgn( ) =
i=1

(1)

(i)1

(iii) If Sn is decomposed into the transpositions i that is = 1 2 . . . r where (for any i 1 . . . r) i is a cycle of length 2, then the signum of is just sgn( ) = (1)r (iv) For any n N with n 1 the signum is a group homorphism of the form sgn : Sn Z . That is for any , Sn we get sgn( ) = sgn( ) sgn( ) (v) () For any n N with n 2 we nd that An n Sn is a normal subgroup of Sn and we can even give its number of elements explictly # An = n! 2

1.2 Permutations

31

1.3

Dening Rings

In the previous section we have introduced an atomic component - groups (and their generalisations monoids and groupoids). We have also introduced some notation, that we employ whenever we are dealing with these objects. We will glue two groupoids over the same set R together - that is we consider a set equipped with two binary relations + and that are compatible in some way (by distributivity). Note that we again write a + b instead of +(a, b) and likewise a b (or ab only) for (a, b). These objects will be called rings and we will dedicate the entire rst part of this book to the study of what structures rings may have. (1.22) Denition: Let R = be any non-empty set and consider two binary operations + called addition and called multiplication on R + : R R R : (a, b) a + b : R R R : (a, b) a b Then the ordered tripel (R, +, ) is said to be ringoid i it satises both of the following properties (A) and (D) (A) (R, +) is a commutative group, that is the addition is associative and commutative, admits a neutral element (called zero-element, denoted by 0) and every element of R has an additive inverse. Formally a, b, c R : a + (b + c) = (a + b) + c a, b R : a + b = b + a 0 R a R : a + 0 = a a R n R : a + n = 0 Note that the zero-element 0 thereby is already uniquely determined and hence the fourth property makes sense. Further, if we are given a R, then the element n R with a + n = 0 is uniquely determined, and we call it the negative of a, denoted by a := n. (D) The addition and multiplication on R respect the following distributivity laws, i.e. a, b, c R we have the properies a (b + c) = (a b) + (a c) (a + b) c = (a c) + (b c)

32

1 Groups and Rings

(1.23) Denition: In the following let (R, +, ) be a ringoid, then we consider a couple of additional properties (S), (R) and (C), that R may or may not have (C) The multiplication on R is commutative, that is we have the property a, b, c R : a b = b a (S) The multiplication on R is associative, that is we have the property a, b, c R : a (b c) = (a b) c (R) The multiplication on R admits a neutral element 1, that is we have 1 R a R : 1 a = a = a 1 Note that the neutral element 1 of R in this case already is uniquely determined - it will be called the unit-element of R. (F) Let us assume properties (S) and (R), then we dene property (F): every non-zero element R has an inverse element, that is 0 = a R i R : a i = 1 = i a Note that 1 is given by (R) and due to (S) the element i R is uniquely determined by a. We call i the inverse of a, denoted by a1 := i. Using these properties we dene the following notions: let (R, +, ) be a ringoid, then (R, +, ) is even said to be commutative resp. called a semiring, ring, skew-eld of even eld, i commutative semi-ring ring commutative ring skew-eld eld : : : : : : (C) (S) (S) and (R) (S), (R) and (C) (S), (R), (F) and 0 = 1 (S), (R), (C), (F) and 0 = 1

Nota in the mathematical literature the word ring may have many dierent meanings, depending of what topic a given text pursues. While some authors do not assume rings to have a unit (or even not be associative) others assume them to always be commutative. Hence it also is customary to speak of a non-unital ring in case of a semi-ring and of a unital ring in case of a ring. (1.24) Remark: Let (R, +, ) be a ringoid, then we simplify our notation somewhat by introducing the following conventions The zero-element of any ringoid R is denoted by 0. If we deal with several ringoids simultaneously then we will write 0R to emphasise, which ringoid the zero-element belongs to. Analogously the unit element of any ring R is denoted by 1 and by 1R if we wish to stress the ring R. For some specic rings we will use a dierent symbol than 1, e.g. 1 or 1 1. 1.3 Dening Rings 33

 In order to save some parentheses we agree that the multiplication is of higher priority than the addition +, e.g. ab + c abbreviates (a b) + c. The additive inverse is sometimes used just like a binary operation, i.e. a b stands shortly for a + (b). We have introduced the multiplicative inverse a1 of a and we use the convention that the inversion is of higher priority, than the multiplication itself, i.e. ab1 stands shortly for a (b1 ). If a1 , . . . , an R are nitely many elements of R, then we denote the sum and product of these to be the following
n

ai :=
i=1 n

(a1 + a2 ) + . . . + an (a1 a2 ) . . . an

ai :=
i=1

For some integer k Z we now introduce the following abbreviations

k i=1 a

ka :=

0
k i=1 (a) k i=1 a

for k > 0 for k = 0 for k < 0 for k > 0 for k = 0 for k < 0

:=

1
k 1 i=1 (a )

For the denition of ak we, of course, required R to be a ring in the case k = 0 and even that a is invertible (refer to the next section for this notion) in the case k < 0. (1.25) Remark: Consider any ringoid (R, +, ), then the addition + on R is associative by denition. Hence the result of the sum a1 + + an R does not depend on the way we applied parentheses to it (cf. (1.2)). Therefore we may equally well omit the bracketing in the summation, writing
n

a1 + + an :=
i=1

ai

The same is true for the multiplication in a semi-ring (R, +, ). So by the same reasoning we omit the brackets in products, writing
n

a1 . . . an :=
i=1

ai

34

1 Groups and Rings

 Yet by denition the addition + of a ringoid (R, +, ) also is commutative. Thus the sum a1 + + an R does not even depend on the ordering of the ai (refer to (1.2) again). Now take any nite set and regard a mapping a : R. Then we may even dene the sum over the unordered set . To do this x any bijection of the form : 1 . . . n (in particular n := #). Then we may dene a(i) :=
i

0
n i=1 a( (i))

if = if =

The same is true for the multiplication in a commutative semi-ring (R, +, ). So given a : R and as above we dene (note that the case = even requires R to be a commutative ring) a(i) :=
i

1
n i=1 a( (i))

if = if =

If I is innite but the set := { i I | a(i) = 0 } is nite still we write an innte sum over all i I , that actually is dened by a nite sum a(i) :=
i I i

a(i)

If R is a commutative ring and analogously := { i I | a(i) = 1 } is nite then the same is true for the innite product a(i) :=
iI i

a(i)

1.3 Dening Rings

35

1.4

Examples

(1.26) Example: The most well-known example of a commutative ring are the integers (Z, +, ). And they are a most beautiful ring indeed - e.g. Z does not contain zero-divisors (that is ab = 0 implies a = 0 or b = 0). And we only wish to remark here that Z is an UFD (that is any a Z admits a unique primary decomposition) - this will be the formulated and proved as the fundamental theorem of arithmetic. Now regard the set 2Z := { 2a | a Z } of even integers. It is easy to see, that the sum and product of even numbers again is even. Hence (2Z, +, ) is a (commutative) semi-ring under the operations inherited from Z. Yet it does not contain a unit element 1 and hence does not qualify to be a ring. Next we will regard a somewhat strange ring, the so called zero-ring Z . As a set it is dened to be Z := { 0 } (with 0 Z if you wish). As it solely contains one point 0 it is clear how the binary operations have to be dened 0 + 0 := 0 and 0 0 := 0. It is esy to check that (Z, +, ) thereby becomes a commutative ring in which 1 = 0. We will soon see that Z even is the only ring in which 1 = 0 holds true. In this sense this ring is a bit pathological. Now consider an arbitrary (commutative) ring (R, +, ) and a nonempty set I = . Then the set RI of functions from I to R can be turned into another (commutative) ring (RI , +, ) under the pointwise operations of functions RI := { f | f : I R } f + g : I R : i f (i) + g (i) f g : I R : i f (i) g (i) Let X be any set and denote its power set by P (X ) := { A | A X }. And for any subsets A, B X we dene the symmetric dierence as A B := (A B ) \ (A B ) = (A \ B ) (B \ A) X . Thereby (P (X ), , ) becomes a commutative ring with zero-element and unit element X . And if A X is any subset, then the additive inverse of A is just A = A again. The next ring we study are the rationals (Q, +, ). It is well-known that Q even is a eld - if a/b = 0 then a = 0 and hence (a/b)1 = (b/a) is invertible. We will even present a construction of Q from Z in a seperate example in this section.

36

1 Groups and Rings

 () Another well-known eld are the reals (R, +, ). They may be introduced axiomatically (as a positively ordered eld satisfying the supremum-axiom), but we would like to sketch an algebraic construction of R. Let us denote the sets of Cauchy-sequences and zerosequences over the rationals Q respectively C := (qn ) Q k N n(k ) N m, n n(k ) : |qn qm | < 1/k k N n(k ) N n n(k ) : |qn | < 1/k

z :=

(qn ) Q

Then C becomes a commutative ring under the pointwise operations of mappings, that is (pn ) + (qn ) := (pn + qn ) and (pn )(qn ) := (pn qn ). And z i C is a maximal ideal of C (cf. to section 2.1). Now the reals can be introduced as the quotient (cf. to section 1.6) of C modulo z R := C z As z has been maximal R becomes a eld. Thereby Q can be embedded into R, as Q R : q (q ). And we obtain a positive order on R by letting (pn )+ z (qn )+ z : m N n m : pn qn . Note that this completion of Q (that is Cauchy-sequences modulo zero-sequences) can be generalized to arbitrary metric spaces. Its eect is that the resulting space becomes complete (that is Cauchy sequences are already convergent). In our case this also guarantees the supremum-axiom. Hence this construction truly realizes the reals. Most of the previous examples have been quite large (considering the number of elements involved). We would now like to present an example of a nite eld consisting of precisely four elements F = { 0, 1, a, b }. As F is nite, we may give the addidion and multiplication in tables + 0 1 a b 0 1 a b 0 0 1 a b 0 0 0 0 0 1 1 0 b a 1 0 1 a b a a b 0 1 a 0 a b 1 b b a 1 0 b 0 b 1 a

1.4 Examples

37

(1.27) Example: Having introduced Z, Q and the reals R we would like to turn our attention to another important case - the residue rings of Z. First note, that the integers Z allow a division with remainder (cf. to section 2.6). That is if we let 2 n N and a Z then there are uniquely determined numbers q Z and r 0 . . . (n 1) such that a = qn + r Thereby r is called the remainder of a in the integer division of a times n. Formally one denotes (which is common in programming languages) a div n := q a mod n := r And by virtue of this operation mod we may now dene the residue ring (Zn , +, ) of Z modulo n. First let Zn := { 0, 1, . . . , n 1 } as a set. And if a, b Zn , then we dene the operations a + b := (a + b) mod n a b := (a b) mod n Thereby (Zn , +, ) will in fact be a commutative ring. And Zn will be a eld if and only if n is a prime number. We will soon give a more natrual construction of Zn as the quotient Z/nZ, as this will be far better suited to derive the properties of this ring. (1.28) Example: Let (R, +, ) be an integral domain, that is R = 0 is a non-zero commutative ring such that for any two elements a, b R we have the implication ab = 0 = a = 0 or b = 0

Then we take to the set X := R (R \ { 0 }) and obtain an equivalence relation on X by letting (where (a, u) and (b, v ) X ) (a, u) (b, v ) : av = bu

If now (a, u) X then we denote its equivalence class by a/u, and let us denote the quotient set of X modulo by Q. Formally that is a u := { (b, v ) X | av = bu } a u a, b R, b = 0

Q :=

Thereby Q becomes a commutative ring (with zero-element 0/1 and unit element 1/1) under the following (well-dened) addition and multiplication a + u a u 38 a u a u := := av + bu uv ab uv 1 Groups and Rings

In fact (Q, +, ) even is a eld, called the quotient eld of R. This is due to the fact that the inverse of a/u = 0/1 is given to be (a/u)1 = u/a. This construction will be vastly generalized in section 2.9. Note that this is precisely the way to obtain the rationals Q := Q by starting with R = Z. (1.29) Example: Consider an arbitrary ring (R, +, ) and 1 n N, then an (n n)-matrix over R is a mapping of the form A : (1 . . . n) (1 . . . n) R. And we denote the set of all such by matn R := { A | A : (1 . . . n) (1 . . . n) R }. Usually a matrix is written as an array of the following form (where ai,j := A(i, j )) a1,1 . .. = . . . an,1 a1,n . . . an,n

A =

ai,j

And using this notation we may turn matn R into a (non-commutative even if R is commutative) ring by dening the addition and multiplication ai,j + bi,j := ai,j + bi,j
n

ai,j bi,j

:=
s=1

ai,s bs,j

It is immediately clear from the construction, that the zero-element, resp. the unit-element are given by the zero-matrix and unit-matirx respectively 0 . .. 0 = . . . 0 1 = 0 1 .. . 1 0 . . . 0 0

(1.30) Example: Let now (R, +, ) be any commutative ring, the we dene the (commutative) ring of formal power series over R, as the following set R[ t] := { f | f : N R } And if f R[ t] then we will write f [k ] instead of f (k ), that is f : k f [k ]. Further we denote f as (note that this is purely notational and no true sum)

f [k ] tk := f
k=0

1.4 Examples

39

And thereby we may dene the sum and product of two formal power series f , g R[ t] , by pointwise summation f + g : k f [k ]+ g [k ] and an involution product f g : k f [0]g [k ] + f [1]g [k 1] + + f [k ]g [0]. In terms of the above notation that is

f + g :=
k=0

f [k ] + g [k ] tk
i+j =k

f [i]g [j ] tk

f g :=
k=0

An elementary computation (cf. section 7.3) shows that (R[ t] , +, ) truly becomes a commutative ring. The zero-element of this ring is given by 0 : k 0 and the unit element is 1 = t0 : k (k, 0). The element t : k (k, 1) plays another central role. An easy computation shows that for any n N we get tn : k (k, n). If now 0 = f R[ t] is a non-zero formal power series we dene the degree of f to be deg(f ) := sup{ k N | f [k ] = 0 } N { } That is deg(f ) < just states that the set { k N | f [k ] = 0 } is a nite subset of N. And thereby we dene the polynomial ring over R to be R[t] := { f R[ t] | deg(f ) < } By denition R[t] R[ t] is a subset, but one easily veries, that R[t] even is another commutative ring under the addition and multiplication inherited from R[ t] . Further note that every polynomial f R[t] now truly is a nite sum of the form

f =
k=0

f [k ] tk

(1.31) Example: () The next example is primarily concerned with ideals, not with rings, and should hence be postponed until you want to study an example of these. Now x any (commutative) base ring (R, +, ) and a non-empty index set I = . We have already introduced the ring S := RI = F (I, R) of functions from I to R under the pointwise operations. For any subset A I we now obtain an ideal of S by letting
v(A) := { f S | a A : f (a) = 0 }

And for any two subsets A, B I it is easy to prove the following identities
v(A) v(B ) = v(A) v(B ) = v(A B ) v(A) + v(B ) = v(A B )

40

1 Groups and Rings

Clearly (if I contains at least two distinct points) S contains zero-divisors. And if I is innite then S is no noetherian ring (we obtain a stricly ascending chain of ideals by letting ak := v(I \ { a1 , . . . , ak }) for a seqence of points ai I ). In this sense S is a really bad ring. On the other hand this ring can easily be dealt with. Hence we recommend looking at this ring when searching for counterexamples in general ring theory. Finally if A I is a nite set then the radical of the ideal v(A) can also be given explictly
v(A) = { f S | a A : f (a) nilR }

Prob in the rst equalty v(A) v(B ) = v(A B ) is clear and the containment v(A) v(B ) v(A) v(B ) is generally true. Thus suppose h v(A B ) f (a) := 0 1 if a A if a A and g (a) := 0 h(a) if a B if a B

then f p(A) and g v(B ) are clear and also h = f g . Thus we have proved v(A B ) v(A) v(B ) and thereby nished the rst set of identities. In the second equality v(A) + v(B ) v(A B ) is clear. And if we are conversely given any h v(A B ) then let us dene f (a) := 0 h(a) if a A if a A and g (a) := h(a) 0 if a A \ B if a A \ B

thereby it is clear that f v(A) and g v(B ) and an easy distinction of cases truly yields h = f + g . Now consider some f S with f k v(A). That is for any a A we get f (a)k = f k (a) = 0. In other words f (a) nilR for any a A. Conversely suppose that A is nte and that for any a A we get f (a) nilR. That is for any a A there is some k (a) N such that f k(a) (a) = 0. By taking k := max{ k (a) | a A }, we nd f k v(A). (1.32) Example: The complex numbers (C, +, ) also form a eld (that even has a lot of properties that make it better-behaved than the reals). We want to present three ways of introducing C now (note that all are isomorphic): (1) First of all dene C := R2 to be the real plane. For any two complex numbers z = (a, b) and w = (c, d) we dene the operations a c + b d a c b d := := a+c b+d ad bc ac + bd

The advantage of this construction is, that it is purely elementary. The disadvantage is that one has to check that (C, +, ) truly becomes a eld under these operations (which we leave as an excercise). We only present the inverse of (0, 0) = z = (a, b) C. Dene the complex conjugate of z to be z := (a, b) and let (z ) := zz = a2 + b2 R, then z 1 = (a (z )1 , b (z )1 ). Let us now denote 1 = (1, 0) and i = (0, 1) C, then it is clear, that 1 is the unit element of C and that i2 = 1. Further any z = (a, b) C can be written uniquely, as z = a1 + bi. It is costumary to write z = a + ib for this however. 1.4 Examples 41

(2) () The next construction will imitate the one in (1) - yet it realizes (a, b) as a real (2 2)-matrix. This has the advantage, that addition and multiplication are well-known for matrices C := a b b a a, b R

It is straighforward to see that C thereby becomes a eld under the addition and multiplication of matrices, that coincides with the one introduced in (1) (check this, its neat). Note that thereby the determinant plays the role of (z ) = det(z ) and the transposition of matrices is the complex conjugation z = z . (3) () The third method implements the idea that i is introduced in such a way as to i2 = 1. We simply force a solution of t2 + 1 = 0 into C C := R[t] (t2 + 1)R[t] Comparing this construction with the one in (1) we have to identify (a, b) with a + bt + (t2 + 1)R[t]. In particular i = t + (t2 + 1)R[t]. Then it again is easy to see, that addition and multiplication coincide under this identication. The advantage of this construction is that it focusses on the idea of nding a solution to t2 + 1 = 0 (namely i), that C immediately is a eld (as t2 + 1 is prime, R[t] is a PID and hence (t2 + 1)R[t] is maximal). The disadvantage is that it already requires the algebraic machinery, which remains to be introduced here. (1.33) Example: () Now x any square-free 0 = d Z (d being sqare-free means that there is no prime number p Z such that p2 divides d). Then we dene the following subset of the (reals or) complex numbers Z[ d] := a + b d | a, b Z

Note that Z[ d] is contained in the reals if and only if 0 < d, and in the case of d < 0 we have d = i d C. In any case Z[ d] becomes a commutative ring under the addition and multiplication inherited from C (a + b d) + (e + f d) = (a + e) + (b + f ) d (a + b d) (e + f d) = (ae + dbf ) + (af + be) d It will be most interesting to see how the algebraic porperties of these rings vary with d. E.g. for d { 2, 1, 2, 3 } Z[ d] will be an Euclidean domain under the Euclidean function (a + b d) := |a2 db2 |. Yet Z[ d] will not even be an UFD for d 3.

42

1 Groups and Rings

(1.34) Example: We have just introduced the complex numbers as a two-dimensional space R over the reals. The next (and last sensibly possible, cf. to the theorem of Frobenius) step in the hirarchy leads us to the quaternions H. These are dened to be a four-dimensional, space H := R4 over the reals. And analogous to the above we specically denote four elements 1 := (1, 0, 0, 0) i := (0, 1, 0, 0) j := (0, 0, 1, 0) k := (0, 0, 0, 1) That is any element z = (a, b, c, d) H can be written uniquely, in the form z = a + ib + jc + kd. The addition of elements of H is pointwise and the multiplication is then given by linear expansion of the following multiplication table for these basis elements 1 i j k 1 1 i j k i i -1 -k j j j k -1 -i k k -j i -1

For z = a + ib + jc + kd and w = p + iq + jr + ks H these operations are z + w = (a + p) + i(b + q ) + j (c + r) + k (d + s) zw = (ap bq cr ds) + i(aq + bp + cs dr) +j (ar bs + cp + dq ) + k (as + br cq + dp) Note that we may again dene z := a ib jc kd and thereby obtain (z ) = zz = a2 + b2 + c2 + d2 R such that any non-zero 0 = z H has an inverse z 1 = (z )1 z = (a (z )1 , b (z )1 , c (z )1 , d (z )1 ). Yet H clearly is not commutative, as ij = k = ji. Thus the quaternions form a skew-eld, but not a eld. Using the complex numbers C we may also realize the quaternions as (2 2)-matrices over C. Given a + ib + jc + kd H we pick up the complex numbers z = a + ib and w = c + id C and identify a + ib + jc + kd with a (2 2)-matrix of the following form H =r z w w z z, w C

(1.35) Example: () Let (R, +, ) be an arbitrary ring, we will later employ the opposite ring (R, +, )op of R. This is dened to be R as a set, under the same addition + but with a multiplication that reverses the order of elements. Formally (R, +, )op := (R, +, ) where a b := b a Note that thereby (R, +, )op is a ring again with the same properties as (R, +, ) before. And if (R, +, ) has been commutative, then the opposite ring of R is just R itself (even as a set), i.e. (R, +, )op = (R, +, ). 1.4 Examples 43

(1.36) Example: Sometimes it is helpful to introduce innity as a number. In this example we want to present how this can be done formally. The advantage of this construction is the formal ease by which innity can be handled, the disadvantage is that the extensions loose may important algebraic properties. (i) Arithmetic in N := N { }: the set N of natural numbers is wellknow. We now pick up a new symbol , which we call innity, and extend the natural order on N by making a maximal element: k N : k<

Also we can extend the usual addition + and multiplication on N by dening the addition and multiplication with any a N to be a + = + a := a = a := 0 if a = 0 if a = 0

Thereby N doesnt loose much properties of N, still (N, +) and (N, ) are commutative monoids, and we also retain the distributivity law between addition and multiplication (it is straightforward to check all this), i.e. for any a, b and c N we get: a (b + c) = a b + a c (a + b) c = a c + b c (ii) () Arithmetic in R := R { , }: the eld R of real numbers has been introduced earlier in this section and is probably well-known to the reader already. Again we pick up a new symbol , which we call innity, and extend the natural order on R by making a maximal and a minimal element: x R : < x and x <

Again we extend the usual addition + and multiplication by dening the addition and multiplication with resp. with a R {} : a R { } : a R \ {0} a R \ {0} := := : a = a := : a () = () a := = = a+ a +a + a

if if if if

a>0 a<0 a>0 a<0

44

1 Groups and Rings

Note that we did neither dene + () nor did we dene 0 . Hence the set R is no algebraic structure, as the operations + and are not fully dened! In fact any such denition would only result in contradictions: Suppose we had = 0, as could have been expected, then since 1 = = 2 we had = (2 1) = = 0. Hence = 0 wouldnt be a new element. In conclusion we just cannot dene + () or 0 , these terms have to be evaded - that is they may not occur within a computation. This is tho one and only reason why computations with innity have to be handled with great care. The reason behind this is an analytic one: we would like to write (n) , as the sequence (n) R does not converge in R but increases strictly monotonous. Yet the dierence of two determined divergent sequences can be just about anything: for any a R we get (n + a) (n) = (a) a, whereas (n2 ) (n) . Thus the result of is completely arbitrary, as anything might occur.

1.4 Examples

45

1.5

First Concepts

Some Combinatorics In the following proposition we want to establish some basic computational rules, that are valid in arbitrary rings. But in order to do this we rst have to recall some elementary facts from combinatorics: First consider a set containing 1 k N elements. If we want to put a linear order to this set, we have to choose some rst element. To do this there are k possibilities. We commence by choosing the second element - now having (k 1) possibilities, and so on. Thus we end up with a number of linear orders on this set that equals k ! the faculty of k , that is dened to be k ! := # { : 1 . . . k 1 . . . k | bijective } = k (k 1) 2 1

If k = 0 we specically dene 0! := 1. Next we consider a set containing n N elements. If we want to select a subset I of precisely k 0 . . . n elements we have to start by choosing a rst element (n possibilities). In the next step we may only choose from (n 1) elements and so on. But as a subset does not depend on the order of its elements we still have to divide this by k ! - the number of linear orders on I . Altogether we end up with n k := # { I 1 . . . n | #I = k } = = n n1 nk+1 k k1 1 n! k ! (n k )!

Note that this denition also works out in the case n = 0 and k = 0 (in which the binomial coecient (n k ) equals one. That is we get (n 0) = 1 = (n n). Furthermore the binomial coecents satisfy a famous recursion formula n+1 k = n n + k k1

Finally consider some multi-index = (1 , . . . , k ) Nk . Then it will be useful to introduce the following notations (where n := ||) || := 1 + + k ! := (1 !) (k !) n n! := !

46

1 Groups and Rings

(1.37) Proposition: (viz. 306) Let (R, +, ) be any semi-ring and a, b R be arbitrary elements of R. Then the following statements hold true 0 = 0 a0 = 0 = 0a (a)b = (ab) = a(b) (a)(b) = ab If now a1 , . . . , am R and b1 , . . . , nn R are nitely many, arbitrary elements of R, then we also obtain the general rule of distributivity
m

ai

n j =1

bj =

ai bj
i=1 j =1

i=1

And if J (1), . . . , J (n) are nite index sets (where 1 n N) and a(i, j ) R for any i 1 . . . n, j J (i) then we let J := J (1) J (n) (and j = (j1 , . . . , jn ) J ) and thereby obtain
n n

a(i, ji ) =
i=1 ji J (i) j J i=1

a(i, ji )

Now suppose (R, +, ) is a ring, n N and a, b R are elements, that mutually commute (that is ab = ba), then we get the binomial rule
n

(a + b)n =
k=0

n k nk a b k

And if (R, +, ) even is a commutative ring a1 , . . . , ak R are nitely many elements and n N then we even get the polynomial rule (a1 + + ak )n =
| |= n

n a

where the above sum runs over all multi-indices Nk that satisfy || = 1 + + k = n. And a abbreviates a := (a1 )1 (ak )k .

(1.38) Remark: Thus there are many computational rules that hold true for arbitrary rings. However there also are some exceptions to integer (or even real) arithmetic. E.g. we are not allowed to deduce a = 0 or b = 0 from ab = 0. As a counterexample regard a = 2 + 6Z and b = 3 + 6Z in Z6 . A special case of this is the following: we may not deduce a = 0 from a = a. As a counterexample regard a = 1 + 2Z in Z2 .

1.5 First Concepts

47

(1.39) Remark: An exceptional role pays the zero-ring R = { 0 }, it clearly is a commutative ring (under the trivial operations) that would even satisfy condition (F). But it is the only ring in which 0 = 1. More formally weve got the eqivalency R = {0} 0=1

[as = is clear and in the converse direction regard a = 1a = 0a = 0]. This equality 0 = 1 will lead to some peculiarities however. This is precisely why we required 0 = 1 for skew-elds and elds. And several theorems will require R to be not the zero-ring, which we will abbreviate by R = 0. (1.40) Remark: (viz. 310) As a neat little application of the binomial rule let us present a generalisation of the recursion formula for the binomial coecients: regarding the expression (x + y )n = (x + y )k (x + y )nk in Z[x, y ] and comparing coecients, one easily veries, that for any k , n N and any r 0 . . . k we get
r i=0

k i

nk ri

n r

(1.41) Denition: (viz. 310) Let (R, +, ) be any semi-ring and b R be any element of R. Then we dene the set zdR of zero-divisors, the set nzdR of non-zero-divisors, the nil-radical nilR and the annulator ann(R, b) of b to be zdR := { a R | 0 = b R : ab = 0 } nzdR := { a R | b R : ab = 0 = b = 0 } nilR := a R | k N : ak = 0 ann(R, b) := { a R | ab = 0 } An element a nilR contained in the nil-radical is also said to be nilpotent. If now R even is a ring we dene the set R of units (or invertible elements) and the set R of relevant elements of R to be R := { a R | b R : ab = 1 = ba } R := R \ { 0 } R And (R, +, ) is said to be integral, i it does not contain any non-zero zero-divisors, that is i it satises one of the following equivalent statements (a) zdR = { 0 } (b) for any a, b and c R such that a = 0 we get ab = ac = b = c (b) for any a, b and c R such that a = 0 we get ba = ca = b = c (c) for any a, b R such that a = 0 we get ab = a = b = 1 (c) for any a, b R such that a = 0 we get ba = a = b = 1 48 1 Groups and Rings

Now (R, +, ) is said to be an integral domain, i it is a commutative ring that also is integral. I.e. a commutative ring with zdR = {0}. (1.42) Remark: Consider any semi-ring (R, +, ), two elements a, b R and a nonzerodivisor n nzdR. Then we obtain the equivalence a=b na = nb

Prob = is clear, and if na = nb then n(a b) = 0 which implies a b = 0 (and hence a = b), since n is a non-zero-divisor. Let (R, +, ) be a ring, a R be any element and n nzdR be a non-zerodivisor. Then we obtain the implication na = 1 = an = 1

Prob since n(an) = (na)n = n and hence n(an 1) = 0. But as n is a non-zerodivisor this implies an 1 = 0 and hence an = 1. Consider a commutative ring (R, +, ) and a1 , . . . , an R nitely many elements. Now let u := a1 . . . an R, then we obtain the implication u R = a1 , . . . , an R

Prob choose any j 1 . . . n and let aj := i=i ai , then by construction we get 1 = u1 (a1 . . . an ) = (u1 aj )aj . And hence we see aj R . If (R, +, ) is a skew-eld, then R already is integral (i.e. zdR = { 0 }) and allows division: that is for any a, b R with a = 0 we get ! u, v R : ua = b and av = b Prob consider a = 0, since R is a skew-eld there is some i R such that ai = 1 = ia. Now suppose ab = 0, then b = (ia)b = i(ab) = 0 and hence a nzdR. This proves zdR = { 0 }. Now consider a, b R with a = 0. Again choose i R such that ai = 1 = ia and let u := bi and v := ib. Then ua = (bi)a = b(ia) = b and av = a(ib) = (ai)b = b. Now suppose u , v R with ua = b = u a and av = b = av . Then a(v v ) = 0 but as a = 0 we have seen that this implies v v = 0 and hence v = v . Likewise (u u )a = 0 but as u = 0 this can only be, if u u = 0 and hence u = u . The sets of zero-divisors zdR and of relevant elements R rarely carry any noteworthy algebraic structure. Yet the units R form a group, the nonzero-divisors nzdR are multiplicatively closed, the nil-radical nilR is an (even perfect) ideal of R and the annulator ann(R, b) of some element b R is a submodule. Though we have not introduced these structures yet, we would like to present a formal proposition (and proof) of these facts already.

1.5 First Concepts

49

(1.43) Proposition: (viz. 309) () (i) Let (R, +, ) be any semi-ring, then R is the disjoint union of its zerodivisors and non-zero divisors, formally that is nzdR = R \ zdR (ii) In any non-zero (R = 0) ring (R, +, ) the nilradical is contained in the set of zero-divisors. And the zero-divisors are likewise contained in non-units of R. That is weve got the inclusions nilR zdR R \ R

(iii) Let (R, +, ) be any semi-ring, then the set nzdR R of non-zerodivisors of R is multiplicatively closed, that is we get 1 nzdR a, b nzdR = ab nzdR

(iv) Consider any ring (R, +, ), then (R , ) is a group (where denotes the multiplicaton of R), called the multiplicative group of R. (v) If (R, +, ) is any ring then R is a skew-eld if and only if the multiplicative group of R consists precisely of the non-zero elements R skew-eld R = R \ { 0 }

(vi) If (R, +, ) is a commutative ring, then the nil-radical is an ideal of R nilR i R

(vii) Let (R, +, ) be any ring and b R an element of R. Then the annulator ann(R, b) is a left-ideal (i.e. submodule) of R ann(R, b) m R

(viii) Let (R, +, ) be a commutative ring and u R be a unit of R. Then u + nilR R . To be precise, if a R with an = 0 then we get
n1

(u + a)

=
k=0

(1)k ak uk1

50

1 Groups and Rings

(1.44) Example: We present a somewhat advanced example that builds upon some results from linear algebra. Let us regard the ring R := mat2 (Z) of (2 2)-matrices over the integers. We will need the notions of determinant and trace det tr a b c d a b c d a b c d := ad bc Z := a + d Z := d b c a

Then we can explictly determine which matrices are invertible (that is units), which are zero-divisors and which are nilpotent. We nd R = { A R | det A = 1 } zdR = { A R | det A = 0 } nilR = { A R | det A = trA = 0 } () Prob rst note that an easy computation yields AA = (det A)1 1 = A A. Hence if A is invertible, then A1 = (det A)1 A . Thus A is invertible if and only if det A is a unit (in Z), but as the units of Z are given to be Z = { 1 } this already is the rst identity. For the second identity we begin with det A = 0. Then AA = (det A)1 1 = 0 and hence A is a zerodivisor. Conversely suppose that A zdR is a zero-divisor, then AB = 0 for some B = 0 and hence (det A)B = A AB = 0. But as B is non-zero this implies det A = 0. Finally if det A = 0 then an easy computation shows A2 = (trA)A, thus if also trA = 0 then A2 = 0 and hence A nilR. And if conversely A nilR zdR, then det A = 0 and hence (by induction on n) An = (trA)n1 A. As A is nilpotent there is some 1 n N such that 0 = An = (trA)n1 A and hence trA nilZ = { 0 }.

1.5 First Concepts

51

1.6

Ideals

(1.45) Denition: Let (R, +, ) be a semi-ring and P R a subset. Then P is said to be a sub-semi-ring of R (abbreviated by P s R), i (1) 0 P (2) a, b P = a + b P (3) a P = a P (4) a, b P = ab P And if (R, +, ) is a ring having the unit element 1, then a subset P R is called a sub-ring of R (abbreviated by P r R), i (S) P s R (5) 1 R Finally if (R, +, ) even is a (skew)eld, then a subset P R is said to be a sub-(skew)eld of R (abbreviated P f R), i (R) P r R (6) 0 = a P = a1 P Let (R, +, ) be a semi-ring again, then a subset a R is said to be a left-ideal (or submodule) of R (abbreviated by a m R), i (1) 0 a (2) a, b a = a + b a (3) a a = a a (4) a a, b R = ba a And a R is said to be an ideal of R (abbreviated by a i R), i (M) a m R (5) a a, b R = ab a We will sometimes use the set of all subrings (of a ring R), the set of all sub(skew)elds (of a (skew)eld R), resp. the set of all (left-)ideals (of a semi-ring R) - these will be denoted, by subr R := { P R | P r R } subf R := { P R | P f R } subm R := { a R | a m R } ideal R := { a R | a i R }

52

1 Groups and Rings

(1.46) Remark: Consider any semi-ring (R, +, ) and a sub-semi-ring P s R of R. Then we would like to emphasise, that the notion of a sub-semi-ring was dened in such a way that the addition + and multiplication of R induces the like operations on P +
P P

: P P P : (a, b) a + b : P P P : (a, b) ab

And thereby (P, + P , P ) becomes a semi-ring again (clearly all the properties of such are inherited from R). And in complete analogy we nd that subrings of rings are rings again and sub(skew)elds of (skew)elds are (skew)elds again, under the induced operations. Nota as + P and P are just restrictions (to P P ) of the respective functions + and on R, we will not distinguish between + P and + respectively between P and . That is we will speak of the semiring (P, +, ), where + and here are understood to be the restricted operations + P and P respectively. Beware: it is possible, that a sub-semi-ring P s R is a ring (P, +, ) under the induced operations, but no subring of R. E.g. consider R := Z Z under the pointwise operations and P := Z { 0 } R. Then P s R is a sub-semi-ring. But it is no subring, as the unit element (1, 1) R of R is not contained (1, 1) P in P . Nevertheless P has a unit-element, namely (1, 0) P and hence is a ring. If (R, +, ) is a ring, then property (3) of left-ideals is redundant, it already follows from property (4) by letting b := 1. Let (R, +, ) be a semi-ring, then trivially any left-ideal a of R already is a sub-semi-ring. Formally put for any subset a R we get
a m R

a s R

Consider a commutative semi-ring (R, +, ), then the property (4) of left-ideals already implies property (5) of ideals, due to commutativity. That is in a commutative semi-ring R we nd the following equivalence for any subset a R
a m R

a i R

Now let (R, +, ) be any ring and let a m R be a left-ideal of R. Then we clearly obtain the following equivalence
a=R

1a

Prob = is clear, as 1 R and if converely 1 a then due to property (4) we have a = a1 a for any a R. Hence R a R.

1.6 Ideals

53

 Consider any semi-ring (R, +, ) and some element a R. Then we obtain a left-ideal Ra in R (called the principal ideal of a), by letting Ra := { ba | b R } m R

Prob clearly 0 = 0a Ra and if ba and ca R, then we see that ba + ca = (b + c)a Ra and (ba) = (b)a Ra. Finally we have c(ba) = (cb)a Ra for any c R. In a commutative semi-ring (R, +, ) (where a R again) we clearly get Ra = aR, where the latter is dened to be aR := { ab | b R } i R

And as R is commutative Ra = aR already is an ideal of R. And it is customary to regard aR instead of Ra in this case. We will later see (cf. section 2.6), that Z is a PID - that is every ideal of Z is of the form aZ for some a Z. Formally that is ideal Z = { aZ | a Z } In any semi-ring (R, +, ) we have the trivial ideals { 0 } and R i R. If R is a skew-eld, then these even are the the only ideals of R. R skew-eld = subm R = { { 0 } , R }

Prob let a m R be any left-ideal of R, if there is some 0 = a a then a1 R and hence 1 = a1 a a. And from this we get a = R. Now consider some non-zero (R = 0) commutative ring (R, +, ). Then we even get: R is a eld if and only if the only ideals of R are the trivial ones. That is equivalent are R eld ideal R = { { 0 } , R }

Prob = has already been shown above, for = regard any 0 = a R, then aR i R is an ideal of R. But as aR = { 0 } this only leaves aR = R and hence there is some i R such that ai = 1, which means i = a1 . Altogether R is a eld. We could have introduced a dual notion of a left-ideal - a so-called right-ideal. This is a subset a R such that the properties (1), (2), (3) and (5) of (left-)ideals hold true. That is we have replaced (4) by (5). However this would lead to a completely analogous theory. Also refer to section 3.1 for a little more comments on this. (1.47) Lemma: (viz. 318) For the moment being let abbreviate any one of the words semi-ring, ring, skew-eld or eld and let (R, +, ) be a . Further consider an arbitrary family (where i I = ) Pi R of sub- s of R. Then the intersection Pi is a sub- of R again Pi R is a subiI

54

1 Groups and Rings

Likewise let us abbreviate by the word left-ideal or the word ideal. Now suppose (R, +, ) is a semi-ring and consider an arbitrary family (where i I = ) ai R of s of R. Then the intersection of the ai is a of R again ai R is a
iI

The proof of this lemma is an easy, straightforward verication of the properties of a (sub-) in all the seperate cases. But though this is most easy it has far-reaching consequences. Whenever we are given a subset X R we may consider the smallest of R containing X . Just take the intersection over all s containing X (this is possible since at least R is a with X R). The following denition formalizes this idea. And it is not much of a surprise, that this smallest then can be described explictly by taking all the elements of X and performing all the operations allowed for a . (1.48) Denition: Let now (R, +, ) be a semi-ring and consider an arbitrary subset X R of R. Then the above lemma allows us to introduce the sub-semi-ring, left-ideal, resp. ideal generated by X to be the intersection of all such containing X X X X
s m i

:= := :=

{ P R | X P s R } { a R | X a m R } { a R | X a i R }

Likewise if R even is a ring (or (skew)eld) we may dene the subring (or sub-(skew)eld respectively) to be the intersection of all such containing X X X
r f

:= :=

{ P R | X P r R } { P R | X P f R }

Nota in case that there may be any doubt concerning the semi-ring X is contained in (e.g. X R S ) we emphasise the semi-ring R used in this construction by writing X R i or X i i R or even R X i . (1.49) Remark: In case X = it is clear that { 0 } is an ideal containing X . And as 0 is contained in any other ideal (even sub-semiring) of R we nd
s

= {0}

The case of the subring resp. sub(skew)eld generated by usually is less trivial however. Let us regard the case R = Q for example. As any subring has to contain 1 and allow sums and negatives, we nd that Z = r Q. And as a subeld even allows to take inverses we nd Q = f Q. However in C we also get Q = f C only.

1.6 Ideals

55

Hence the intersection of all subrings resp. all sub(skew)elds deserves a special name, it is called the prime-ring resp. prime(skew)eld of R
r f

= =

{ P R | P r R } { P R | P f R }

(1.50) Proposition: (viz. 319) Consider a ring (R, +, ) and a non-empty subset = X R. Then we can give an explicit description of the (left)-ideal generated by X
n

=
i=1 n

ai xi

1 n N, xi X, ai R

=
i=1

ai xi bi

1 n N, xi X, ai , bi R

Suppose R is any semi-ring and denote X := { x | x X } { x | x X } then we can also give an explicit description of the sub-semiring of R generated by X (and if R is a ring also of the subring of R generated by X )
m n

xi,j
i=1 j =1

1 m, n N, xi,j X X {1}

Finally suppose that R even is a eld, refering to the generated subring we can also give an explicit description of the subeld of R generated by X X
f

ab1 | a, b X r , b = 0

Ideals have been introduced into the thoery of rings, as computations with ordinary elements of rings may have very strange results. Thus instead of looking at a R one may turn to aR i R (in a commutative ring). It turns out that it is possible to dene the sum and product of ideals, as well and that this even has nicer properties that the ordinary sum and product of elements. Hence the name - aR is an ideal element of R. Nowadays ideals have become an indespensible tool in ring theory, as they turn out to be kernels of ring-homomorphisms.

56

1 Groups and Rings

(1.51) Proposition: (viz. 320) Consider any semi-ring (R, +, ) and arbitrary ideals a, b and c i R. Then these induce further ideals of R by letting
a b := a + b := a b :=
i=1

a R | a a and a b a + b | a a, b b
n

ai bi

1 n N, ai a, bi b

Note however that a b need not be an ideal of R. And thereby we have the following chains of inclusions of ideals of R
ab

a b a, b a + b

(1.52) Proposition: (viz. 320) Let (R, +, ) be any semi-ring, then the compositions , + and of ideals are associative, that is if a, b and c i R are ideals of R, then we get (a b) c = a (b c) (a + b) + c = a + (b + c) (a b) c = a (b c) Further and + are commutative, and if R is commutative then so is the multiplication of ideals, formally again (with a and b i R)
ab = ba a+b = b+a ab = ba

Recall that the last equality need not hold true, unless R is commutative! The addition + and multiplication of ideals even satises the distributivity laws, that is for any ideals a, b and c i R we get
a (b + c) = (a b) + (a c)

(a + b) c = (a c) + (b c)

(1.53) Remark: It is clear that for any ideal a i R we get a R = a. Thus (ideal R, ) is a commutative monoid with neutral element R. Likewise we get a + 0 = a for any ideal a i R and hence (ideal R, +) is a commutative monoid with neutral element 0. Finally a R = a = R a for any ideal a i R and hence (ideal R, ) is a monoid with neutral element R, as well. And if R is commutative, so is (ideal R, ). In this sense it is understood that a0 := R and ai := a . . . a (i-times) for 1 i N.

1.6 Ideals

57

(1.54) Proposition: (viz. 321) Now consider a ring (R, +, ) and two arbitrary subsets X and Y R. Let a = X i and b = Y i be the ideals generated by these. Then the sum a + b is generated by the union X Y , formally that is
a+b =

X Y

And if R is commutative, then the product a b is generated by the pointwise product X Y := { xy | x X, y Y } of these sets, formally again
ab =

XY

(1.55) Remark: (viz. 321) (i) Due to (1.54) it oftenly makes sense to regard sums a + b or products a b of ideals. And in (1.61) we will also the notion b/a of a quotient of ideals. Thus one might also be led to think that there should be something like a dierence of ideals. A good candidate would be the following: consider the ideals a b i R in some semi-ring (R, +, ). Then we might take b \ a i to be the dierence of b minus a. Yet this yields nothing new, since
a b

b= b\a

(ii) Let (R, +, ) be a ring again and a1 , . . . , an i R be a nite family of ideals of R. Then using (1.54) and induction on n it is clear that
n n n

ai :=
i=1 i=1

ai

ai ai

=
i=1

ai

This property inspires us to dene the sum of an arbitrary family (i I ) of ideals ai i R of R. As it is not clear what we should substitute for the nite sum of elements we instead dene
ai :=
i I i I

ai

(iii) And thereby it turns out that the innite sum consists of arbitarily large but nite sums over elements of the ai , formally that is
ai =
i I iI

ai ai
i n

ai ai , # { i I | ai = 0 } < ai ai , I, # < n N, i(k ) I, ak ai(k)

=
k=1

ak

58

1 Groups and Rings

(iv) Let now b i R be another ideal of R, then the distributivity rule of (1.52) generalizes to the case of innite sums, that is
ai
iI

b =
iI

(ai b)

(1.56) Proposition: (viz. 322) Let (R, +, ) be a commutative ring and consider the ideals a, b, a1 , . . . , ak and b, . . . , bk i R of R (where 1 k N). Then we obtain the statements (i) a and b are said to be coprime i they satisfy one of the following three equivalent conditions (a) a + b = R (b) (1 + a) b = (c) a a, b b such that a + b = 1 (ii) Suppose that a and b are coprime and consider any i, j N, then the j powers ai and b are coprime as well, formally that is
a+b=R

ai + b = R

(iii) Suppose that a and bi (for any i 1 . . . k ) are coprime, then a and b1 . . . bk are coprime as well, formally this is i 1 . . . k : a + bi = R =
a + (b1 . . . bk ) = R

(iv) Suppose that the ai are pairwise coprime, then the product of the ai equals their intersection, formally that is i = j 1 . . . k : ai + aj = R =
a1 ak = a1 . . . ak

(1.57) Denition: (viz. 323) Let (R, +, ) be any semi-ring and a m R be a left-ideal of R, then a induces an equivalence relation on R by virtue of ab : aba

where a, b R. And the equivalence classes under this relation are called cosets of a, and for any b R this is given to be b + a := [b] = { b + a | a a } Finally we denote the quotient set of R modulo a (meaning the relation ) R 1.6 Ideals R a := = {a + a | a R} 59

If a i R even is an ideal (e.g. when R is commutative) then R/a can even be turned into a semi-ring (R/a, +, ) again under the following operations (a + a) + (b + a) := (a + b) + a (a + a) (b + a) := (a b) + a Thereby (R/a, +, ) is also called the quotient ring or residue ring of R modulo a. Clearly, if R is commutative, then so is R/a. And if R is a ring, so is R/a, where the unit element is given to be the coset 1 + a. (1.58) Remark: It is common that beginners in the eld of algebra encounter problems with this notion. But on the other hand the construction of taking to the quotient ring is one of the three most important tools of algebra (the others being localizing and taking tensor products). Hence we would like to append a few remarks: We have dened R/a to be the quotient set of R under the equivalence relation a b a. And the equivalence classes have been the cosets b + a. Hence two cosets are equal if and only if their representatives are equivalent and formally for any a, b R this is a+a=b+a aba

And if a i R is an ideal of R we even were able to turn this quotient set R/a into a ring again, under the operations above. Now dont be confused, we sometimes also employ the pointwise sum A + B := { a + b | a A, b B } and product A B := { ab | a A, b B } of two subsets A, B R. But this a completely dierent story - the operations on cosets b + a have been dened by referring to the representatives b. (1.59) Example: (viz. 325) Let 1 n N be a positive integer and consider the ideal nZ i Z. Then we wish to present an example of the above remark: If a and b Z are any two integers, then the following three statements are equivalent (a) a + nZ = b + nZ (b) a mod n = b mod n (c) there is some k Z such that b = a + kn (1.60) Remark: Intuitively speaking the residue ring R/a is the ring R where all the information contained in a has been erased. Thus R/a only retains the information that has been stored in R but not in a. Hence it is no wonder that R/a is smaller than R (|R/a| |R| to be precise). However R/a is not a subring of R! Let us study some special cases to enlighten things: In the case a = R the residue ring R/R contains precisely one element, namely 0 + R. Hence R/R = { 0 + R } is the zero-ring. This matches the intuition all information has been forgotten, what remains carries the trivial structure only. 60 1 Groups and Rings

 The other extreme case is a = {0}. In this case a induces the trivial relation a b a = b. Hence the residue class of a is just a + {0} = {a}. Therefore we have a natural identication R R {0} : a { a } Again this matches the intuition: {0} contains no information hence R/{0} retains all the information of R. In fact we have just rewritten any a R as {a} R/{0}. Now consider any commutative ring (R, +, ) and build the polynomial ring R[t] upon it. Then the ideal a = tR[t] allows us to regain R from R[t]. Formally there is a natural identication R R[t] tR[t] : a at0 + tR[t] Hence one should think that tR[t] contains the information we have added by going to the polynomial ring R[t]. Forgetting it we return to R. Further it is nice to note that the same is true for the formal power series R[ t] instead of R[t]. The same amount of information that has been added is lost. Prob it is clear that R R[t] : a at0 is injective and for a = b R we even get at0 bt0 = (a b)t0 tR[t]. This means that even a at0 +tR[t] is injective. And if we are given an arbitrary polynomial f = an tn + + a1 t + a0 t0 then f + tR[t] = a0 t0 + tR[t] which also is the surjectivity. Note that this all is immediate from the rst isomorphism theorem: R[t] R : f f [0] has kernel tR[t] and image R. Likewise (for any 1 n N) we obtain a natural identication between the ring Zn (as dened in (1.27)) and the residue ring Z/nZ via the isomorphism Z
nZ : a + nZ a mod n

This example teaches that the properties of the residue ring (here Z/nZ) may well dier from those of the ring (here Z). E.g. Z is an integral domain, whereas Z/4Z is not ((2 + 4Z)(2 + 4Z) = 0 + 4Z). On the other hand Z/2Z is a eld, whereas Z is not. Prob rst of all (a + nZ) = a mod n is well-dened and bjective, according to (1.59). The inverse map is just the canonial projection (r) = r + nZ. For the homorphism property of let us write (by division with remainder) a = pn + r and b = qn + s where r = a mod n and s = b mod n. Then a + b = (p + q )n + (r + s) and ab = (pqn + ps + qr)n + rs. Hence we can compute, using (1.59) again ((a + nZ) + (b + nZ)) = (a + b) mod n = (r + s) mod n ((a + nZ) (b + nZ)) = (ab) mod n = (rs) mod n

1.6 Ideals

61

But by denition of Zn we have the compositions r + s = (r + s) mod n and rs = (rs) mod n hence truly transfers the tructure of Z/nZ onto Zn . In particular Zn truly is a commutative ring again.

(1.61) Lemma: (viz. 325) Correspondence Theorem Consider any semi-ring (R, +, ) and an ideal a i R, then we obtain a 1-to-1 correspondence between the ideals of the residue ring R/a and the ideals of R containing a by virtue of (where b/a := b + a | b b ) ideal R a u b a
b b i R | a b

{b R | b + a u}

And this correspondence even is compatible with intersections, sums and products. That is if consider any two ideals b and c i R with a b c
bc a = a b +c b+c a a = a b c b c a a = a a b c

Finally the correspondence also is compatible with the generation of ideals. That is consider an arbitrary subset X R and denote its set of residue classes by X/a := { x + a | x X }. Then we obtain the identity X
i

+a

a =

(1.62) Remark: The following graphic is supposed to illustrate the situation: going from R to R/a all the ideals below a are deleted, while those ideals b i R with a b are preserved. Those ideals that were not comparable to a are lost uncontrollably.

R/a

other ideals ab

ab a erased ideals ba 0

62

1 Groups and Rings

(1.63) Remark: Let (R, +, ) be a semi-ring and a, b i R be two ideals, such that a b. As above let us denote the ideal of R/a induced by b by
b a :=

b+a|bb

Then it is easy to see that for any x R we even get the following equivalence xb x+ab a

Prob if x b then x + a b/a by denition. And if conversely x + a b/a then there is some b b such that x + a = b + a. That is x b a b and hence x + b = b + b = 0 + b again. But the latter already means x b. (1.64) Remark: We will soon introduce the notion of a homomorphism, an example of such a thing is the following map that sends b R to its equivalence class b + a : R R
a : bb+a

It is clear that this mapping is surjective. And it also is clear that for any ideal b i R with a b the image of b under is given to be (b) = b+a|bb = b a acting on ideals
1 (u).

Hence the above correspondence of ideals is just given by (i.e. on certain subsets of R). That is b (b) and u

(1.65) Example: Let us give a well/known, neat little application of the theory: the usual way of denoting integers a Z is using the decimal representation, e.g. 273 = 2 100 + 7 10 + 3 1. Generally we write a = an . . . a1 a0 i ak 0 . . . 9 and
n

a =
k=0

ak 10k

As 2 divides 10 we get a + 2Z = a0 + 2Z and hence it is clear that 2 divides a if and only if 2 divides a0 (the same holds true if we replace 2 by 5) 2 | a 2 | a0

Also it is customary to dene the cross-sum of a to be the sum of the digits ak in the decimal representation of a, formally that is
n

(a) :=
k=0

ak N

1.6 Ideals

63

E.g. the cross sum of 273 is 2 + 7 + 3 = 12. Now, as 10 = 3 3 + 1 we have 10 + 3Z = 1 + 3Z, and thereby 10k + 3Z = (10 + 3Z)k = (1 + 3Z)k = 1 + 3Z. This enables us to compute
n n

a + 3Z =
k=0 n

ak 10 + 3Z =
k=0 n

ak (10k + 3Z) ak + 3Z = (a) + 3Z

k=0

=
k=0

ak (1 + 3Z) =

That is a + 3Z = (a) + 3Z and in particular we have 3 | a if and only if (a) + 3Z = a + 3Z = 0 + 3Z which is equivalent to 3 | (a) again. That is we can check wether a is divisible by 3 just by checking wether the cross-sum (a) of a is divisible by 3. E.g. (273) = 12 is divisible by 3 and hence 273 is divisible by 3, as well - in fact 273 = 3 91. We can even iterate the cross summation, e.g. ((273)) = 3 which is easier to divide by 3. (Again, as 10 + 9Z = 1 + 9Z, the same reasonings holds true if we replace 3 by 9) 3 | a 3 | (a)

(1.66) Remark: () Let now (R, +, ) be a commutative ring, then the above correspondence of ideals interlocks maximal ideals, prime ideals and radical ideals: ideal R a R srad a R spec a R smax a { p spec R | a p } { m smax R | a m }
b srad R | a b b ideal R | a b

That is the ideal b i R with a b is radical if and only if the corresponding ideal b/a i R/a is radical, too. Likewise a b i R is prime (maximal) i b/a i R/a is prime (maximal). In fact we even have the following equality of radical ideals for any b i R with a b
b

a =

(1.67) Example: () Consider the ring (Z, +, ), as Z is an Euclidean domain all its ideals are of the form aZ for some a Z (confer to section 2.6 for a proof). We now x some ideal a := aZ, then aZ bZ is equivalent, to the fact that b divides a (see section 2.5 for details). Hence by the correspondence theorem the ideals of Za = Z/aZ are preciely given to be ideal Z aZ 64 = bZ aZ b | a 1 Groups and Rings

1.7

Homomorphisms

(1.68) Denition: Let (R, +, ) and (S, +, ) be arbitrary semi-rings, a mapping : R S is said to be a homomorphism of semi-rings (or shortly semi-ring-homomorphism ) i for any a, b R it satises the following two properties (1) (a + b) = (a) + (b) (2) (a b) = (a) (b) And if R and S even are rings having the unit elements 1R and 1S respectively, then is said to be a homomorphism of rings (or shortly ring-homomorphism i for any a, b R it even satises the properties (1) (a + b) = (a) + (b) (2) (a b) = (a) (b) (3) (1R ) = 1S And we denote the set of all semi-ring-homomorphisms respectively of all ring-homomorphisms from R to S (that is a subset of F (R, S )) by shom(R, S ) := { : R S | (1) and (2) } rhom(R, S ) := { : R S | (1), (2) and (3) } And if : R S is a homomorphism of semi-rings, then we dene its image im S and kernel kn R to be the following subsets im() := (R) = { (a) | a R } kn() := 1 (0S ) = { a R | (a) = 0 }

(1.69) Remark: In the literature it is customary to only write hom(R, S ) instead of what we denoted by shom(R, S ) or rhom(R, S ). And precisely which of these two sets is meant is determined by the context only, not by the notation. In this book we try to increase clarity by adding the letter s or r. From a systematical point of view it might be appealing to write homs (R, S ) for shom(R, S ) (and likewise homr (R, S ) for rhom(R, S )), as we have already used these indices with substructures. Yet we will require the position of the index later on (when it comes to modules and algebras). Hence we chose to place precisely this index in front of hom instead.

1.7 Homomorphisms

65

(1.70) Example: The easiest example of a homomorphism of semi-rings is the so-called zero-homomorphism which we denote by 0 and which is given to be 0 : R S : a 0S Let (S, +, ) be any semi-ring and R s S be a sub-semi-ring of S . Then the inclusion R S gives rise the the inclusion homomorphism : R S : aa Clearly this is a homomorphism of semi-rings. And if S even is a ring and R r S is a subring of S , then is a homomorphism of rings. In the special case R = S we call 1 1 := the identity map of R. Now let (R, +, ) be any (semi-)ring and consider some ideal a i R. Then we have already introduced the quotient (semi-)ring R/a in (1.57). And this gives rise to the following homomorphism of (semi-) rings (which is called canonical epimorphism ) : R R
a : bb+a

Note that the denition of the addition and multiplication on R/a are precisely the properties of being a homomorphism. And also by denition the kernel of this map precisely is kn( ) = a. Of course is surjective and hence its image is the entire ring im( ) = R/a. Consider any (semi-)ring (R, +, ) again and let a, b i R be two ideals of R such that a b. Then we obtain a well-dened, surjective homomorphism of (semi-)rings by (further note that the kernel of this map is given to be the quotient ideal b/a) R
a

b : a+aa+b

Prob if a + a = b + a then a b a b whence we get a + b = b + b again. This already is the well-denedness and the surjectivity and homorphism properties are clear. Further b + b = 0 + b i b a which again is equivalent, to b + a b/a. And this also is kn() = b/a. Finally we would like to present an example of a homomorphism of semi-rings, that is no homomorphism of rings. The easiest way to achieve this is by regarding the ring R = Z2 . Then we may take : Z2 Z2 : (a, b) (a, 0)

66

1 Groups and Rings

(1.71) Proposition: (viz. 334) (i) Let (R, +, ) and (S, +, ) be semi-rings and : R S be a homomorphism of semi-rings again. Then already satises (for any a R) (0R ) = 0S (a) = (a) (ii) And if (R, +, ) and (S, +, ) are rings, u R is an invertible element of R and : R S is a homomorphism of rings, then (u) S is an invertible element of S , too with inverse (u1 ) = (u)1 (iii) Let (R, +, ), (S, +, ) and (T, +, ) be (semi-)rings, : R S and : S T be homomorphisms of (semi-)rings. Then the composition after is a homomorphism of (semi-)rings again : R S : a ((a)) (iv) Let (R, +, ) and (S, +, ) be (semi-)rings and : R S be a bijective homomorphism of (semi-)rings. Then the inverse map 1 : S R is a homomorphism of (semi-)rings, too. (v) Let (R, +, ) and (S, +, ) be (semi-)rings and : R S be a homomorphism of (semi-)rings, then the image of is a sub-(semi-)ring of S and its kernel is an ideal of R im() kn() s i S R

(vi) More generally let (R, +, ) and (S, +, ) be any two (semi-)rings containing the sub-(semi-)rings P s R and Q s S respectively. If now : R S is a homomorphism of (semi-)rings, then (P ) and 1 (Q) are sub-(semi-)rings again (P )
1

s s

S R

(Q)

(vii) Analogously let (R, +, ) and (S, +, ) be any two semi-rings containing the (left-)ideals a m R and b m S respectively. If now : R S is a homomorphism of semi-rings then 1 (b) is a (left-)ideal of R again. And if even is surjective, then (a) is a (left-)ideal of R 1 (b) surjective m R

(a) m R

1.7 Homomorphisms

67

(viii) Let (R, +, ) be a ring again and (S, +, ) even be a skew eld. Then any non-zero homomorphism of semi-rings from R to S already is a homomorphism of rings (i.e. satises property (3)). Formally S skew-eld = shom(R, S ) = rhom(R, S ) { 0 }

(ix) Now let (R, +, ) be a skew-eld and (S, +, ) be an arbitrary semi-ring, then any nonzero homomorphism : R S of semi-rings is injective R skew-eld = injective or = 0

(1.72) Proposition: (viz. 336) Let (R, +, ) be an arbitrary semi-ring and a i R be any ideal in R. Then we regard the canonical epimorphism from R to R/a, that is : RR a : r r+a If now b i R is any other ideal in R then the image of b under to be (a + b)/a and the preimage of this set are given to be a + b (b) = a + b a 1 ( (b)) = a + b is given

(1.73) Denition: Consider any two (semi-)rings (R, +, ) and (S, +, ) and a homomorphism : R S of (semi-)rings. Then is said to be a mono-, epi- or isomorphism of (semi-)rings, i it is injective, surjective or bijective respectively. And a homomorphism of the form : R R is said to be an endomorphism of R. And is said to be an automorphism of R i it is a bijective endomorphism of R. Altogether we have dened the notions is called monomorphism epimorphism isomorphism endomorphism automorphism i is injective surjective bijective R=S R = S and bijective

And if (R, +, ) is any ring, then we denote the set of all ring automorphisms on R by (note that this is a group under the composition of mappings) raut(R) := { : R R | bijective homomorphism of rings } In the case that : R S is an isomorphism of semi-rings we will abbreviate this by writing : R =s S . And if : R S even is an isomorphism of rings we write : R =r S . And thereby R and S are said to be isomorphic if there is an isomorphism between them. And in this case we wite R =s S R =r S 68 : : : : R =s S : : R =r S 1 Groups and Rings

(1.74) Proposition: (viz. 336) (i) Let (R, +, ) and (S, +, ) be semi-rings and : R S be a homomorphism of semi-rings. Then is injective, resp. surjective i injective surjective kn() = { 0 } im() = S

(ii) Let (R, +, ) and (S, +, ) be (semi-)rings and : R S be a homomorphism of (semi-)rings, then the following statements are equivalent (and in this case we already get = = = 1 ) (a) is bijective (b) there is some homomorphism : S R of (semi-)rings such that = 1 1R and = 1 1S (c) there are homomorphisms : S R and : S R of (semi-) rings such that = 1 1R and = 1 1S (iii) Let (R, +, ) and (S, +, ) be any two semi-rings and : R =s S be an isomorphism between them. Then R is a ring if and only if S is a ring. And in this case : R =r S already is an isomorphism of rings. (iv) The relation =r has the properties of an equivalence relation on the class of all rings (and the same is true for =s on the class of all semirings). To be precise, for all rings (R, +, ), (S, +, ) and (T, +, ) we obtain the following statements 1 1R : R =r R :R =r S = 1 : S =r R = : R =r T

:R =r S and : S =r T

(1.75) Remark: Intuitively speaking two (semi-)rings (R, +, ) and (S, +, ) are isomorphic if and only if they carry precisely the same structure. A bit lax that is the elements may dier by the names they are given but nothing more. Therefore any algebraic property satised by R also holds true for S and vice versa. We give a few examples of this fact below, but this is really true for any algebraic property. Thus from an algebraic point of view, isomorphic (semi-)rings are just two copies the same thing (just like two 1 Euro coins). And the dierence between equality and isomorphy is just a settheoretical one. We verify this intuititive view by presenting a few examples - let (R, +, ) and (S, +, ) be any two semi-rings that are isomorphic under :R =s S . Then we get In the above proposition we have already shown R is a ring if and only if S is a ring. And in this case already is an isomorphism of rings.

1.7 Homomorphisms

69

 R is commutative if and only if S is commutative. Prob suppose that R is commutative and consider any two elements x, y S . Then let a := 1 (x) and b := 1 (y ) R and compute xy = (a)(b) = (ab) = (ba) = (b)(a) = yx. Hence S is commutative, too. And the converse implication follows with instead of 1 . R is an integral domain if and only if S is an integral domain. Prob suppose that S is an integral domain and consider any two elements a, b R. If we had ab = 0 then 0 = (0) = (ab) = (a)(b). But as S is an integral domain this means (a) = 0 or (b) = 0. And as is injective this implies a = 0 or b = 0. Hence R is an integral domain, too. And the converse implication follows with 1 instead of . R is a skew-eld if and only if S is a skew-eld. In fact for any two rings R and S we even get the stronger correspondence (R ) = S Prob if a R is a unit of R then (a1 ) = (a)1 and hence (a) S is a unit, too. And if conversely x S is a unit of S , then let a := 1 (x) and b := 1 (x1 ). Thereby ab = 1 (x)1 (x1 ) = 1 (xx1 ) = (1S ) = 1R and likewise ba = 1R . Hence we have a1 = b such that a R and x (R ) is the image of a unit of R. (1.76) Remark: () In the chapters to come we will introduce the notion of a category (in our case this would be the collection of all semi-rings or of all rings respectively). And in any category there is the notion of monomorphisms, epimorphisms and isomorphisms. That is let (R, +, ) and (S, +, ) be two (semi-)rings. Then : R S will be called a monomorphism in the category of (semi)rings, i it is a homorphism of (semi-)rings, such that for any (semi-)ring (Q, +, ) and any two homomorphisms , : Q R of (semi-)rings we would have the implication = = =

Likewise : R S will be called a epimorphism in the category of (semi-) rings, i it is a homorphism of (semi-)rings, such that for any (semi-)ring (T, +, ) and any two homomorphisms , : S T of (semi-)rings we would have the implication = = = Finally : R S is said to be an isomorphism in the category of (semi-) rings, i it is a homomorphism of (semi-)rings and there is another homomorphism : S R of (semi-)rings that is inverse to , formally : S R : = 1 1R and = 1 1S

70

1 Groups and Rings

It is immediately clear that an injective homomorphism of (semi-)rings thereby is a monomorphism in the categorial sense. And likewise a surjective homomorphism of (semi-rings) is an epimorphism in the categorial sense. The converse implications need not be true however. Yet we will see in the subsequent proposition that the two notions of isomorphy are equivalent. (1.77) Theorem: (viz. 337) Isomorphism Theorems (i) Let (R, +, ) and (S, +, ) be any two rings, a i R be an ideal of R and : R S be a homomorphism of rings. If now a kn() then we obtain a well-dened homomorphism of rings by virtue of : R a S : b + a (a) Note if R and S are semi-rings and is a homomorphism of such, then still is a well-dened homomorphism of semi-rings. (ii) First Isomorphism Theorem Let (R, +, ) and (S, +, ) be any two rings and : R S be a homomorphism of rings. Then the kernel of is an ideal kn() i R of R and the its image im() r S is a subring of S . And nally we obtain the following isomorphy of rings R kn() =r im() : a + kn() (a)

Note if R and S are semi-rings and is a homomorphism of such, then we still get kn() i R, im() s S and R/a =s im(). (iii) Second Isomorphism Theorem Let (S, +, ) be a ring, R r S be a subring of S and b i S be an ideal of S . Then b R i R is an ideal of R, the set b + R := b + a | b b, a R r S is a subring of S and b i b + R is an ideal of b + R. And thereby we obtain the following isomorphy of rings R b+R : a+bRa+b b R =r b

Note if S is a semi-ring and R s S is a sub-semi-ring of S , then still b + R s S , b i b + R and we retain the isomorphy R/b R =s b + R/b. (iv) Third Isomorphism Theorem Let (R, +, ) be any ring containing the ideals a b i R, then b/a := b + a | b b i R/a is an ideal of the quotient ring R/a and we obtain the following isomorphy of rings R/a R : (a + a) + b/a a + b b/a =r b

Note if R is a semi-ring then everything remains the same, except that the above isomorphy is an isomorphy of semi-rings (and not of rings).

1.7 Homomorphisms

71

1.8

Ordered Rings

In this section we will regard commutative rings that are subjected to a (as we call it) positive order. However it is completely independent of the other sections within this book. Its results will neither be used nor continued elsewhere. But as it is a nice generalisation of the classical examples of rings - foremost Z, as these carry a natural order - we chose to include it nevertheless. Enjoy. (1.78) Denition: Let (R, +, ) be a commutative ring, then a subset C R is said to be a positive cone in R, i it satises the following three properties (C1) 0, 1 C (C2) a, b C = a + b C (C3) a, b C = a b C A positive cone C R is said to be proper i it satises the following fourth property (where C := { a | a C }) (CP) C (C ) = { 0 } And a positive cone C R is said to be total i it satises another property (CT) C (C ) = R (1.79) Proposition: (viz. 340) (i) Let (R, +, ) be a commutative ring and (Ci ) be a family (i I ) of positive cones Ci R of R, then their intersection is a positive cone in R again Ci R is a positive cone
iI

(ii) Due to (i) we may dene the positive cone generated by any subset A R to be the smallest positive cone containing A, that is cone(A) := { C R | A C, C is a positive cone }

(iii) Let both (R, +, ) and (S, +, ) be commutative rings and : R S be a ring homomorphism, then we get the following implications: C R positive cone D S positive cone (1.80) Example: (viz. 340) (i) For any commutative ring (R, +, ) the whole set R R is a cone in R. In fact any subring S R is a cone in R. The dierence between these two notions is that a subring S has to be a subgroup of the additive group (R, +) of the ring, whereas a positive cone C R only is s submonoid of (R, +). In fact subrings yield rather uninteresting positive cones. 72 1 Groups and Rings = = (C ) S positive cone 1 (D) R positive cone

(ii) For any commutative ring (R, +, ) and any 1 k N we can interpret k as an element of R as kR := 1R + + 1R (k -times) and likewise for negative numbers: (k )R := (kR ). Then we can regard the ring homomorphism : Z R : k kR . The image of N under this homomorphism is a positive cone in R P := { kR | k N } In fact P is the smallest positive cone of R at all [that is P = cone()] and hence also called the prime cone of R. The homomorphism has another interesting property: the kernel of is an ideal in Z and hence there is an unique c N such that kn( ) = c Z. This c is called the characteristic of R and denoted by char(R) := min { k N | kR = 0 } (iii) A special case of the above is N Z, which obviously is a proper, total, positive cone. Note however that N Q is a proper, positive cone, that is not total. (iv) Another well-known example is R+ C, which is another proper, positive, but not total cone. (v) Generally let (R, +, ) be a commutative ring under a xed positive order (see below), then we obtain a proper, positive cone by letting R+ := { a R | a 0 } (vi) If (R, +, ) is any commutative ring, then we obtain a positive cone on R (not necessarily proper or total) by letting :=
2 a2 1 + + an | n N, ak R

(vii) Let now (R, +, ) be an integral domain under a xed positive order (see below) and consider the polynomial ring R[x] over R. If 0 = f R[x] is any non-zero polynomial, then there are uniquely determined k n N such that f is if the form: f (x) = an xn + + ak xk with an = 0 and ak = 0. Thereby an is called the leading coecient of f (x) whereas ak is said to be the the minor coecient of f (x) which we denote by lc (f ) := an mc (f ) := ak

1.8 Ordered Rings

73

And particularly for f = 0 we let lc (0) := 0 and mc (0) := 0. It is well known that both of these maps are multiplicative and therefore we obtain two proper, positive cones, by letting: R[x]+ := { f R[x] | lc (f ) 0 } R[x] := { f R[x] | mc (f ) 0 } Note that the order on R[x] induced by the cone R[x]+ is non-archimedian as for any 2 n N we get x < 1/n. This is not so with the cone R[x] however. (1.81) Denition: Let (R, +, ) be a commutative ring and consider a partial order on R. That is is a subset of R R satisfying the following three properties: (O1) a R we have: a a (O2) a, b R we have: if a b and b c then a c (O3) a, b R we have: if a b and b a then a = b Thereby we already used the customary notation a b for (a, b) . And as usual we write a < b to denote that a b but a = b. Now is said to be a positive order, i for any a, b R we get (P1) 0 1 (P2) a b 0 b a (P3) 0 a, b = 0 a + b (P4) 0 a, b = 0 ab A partial order is said to be total i it further satises the following property (T) a R we have a b or b a As usual, in this case, we may dene the minimum a b and maximum a b of a, b R to be a b := a if a b and a b := b if b a. Likewise a b := b if a b and a b := a if b a. And we can further dene the sign sgn(a) R and absolute value |a| R of an element a R 1 0 sgn(a) := 1 |a| := a a if a > 0 if a = 0 if a < 0 if 0 a if a 0

74

1 Groups and Rings

(1.82) Remark: If is a positive order on the commutative ring (R, +, ) then it already satises the following, stronger properties for any a, b and h R (P5) a < b 0 < b a (P6) a 0 0 a (P7) a b = a + h b + h (P8) a b, 0 h = ah bh (P9) a b, h 0 = ah bh Prob (P5) a < b means a b but a = b. By property (P2) we get 0 b a. Suppose 0 = b a, then a = b, which leaves 0 < b a only. (P6) this is property (P2) applied to b = 0, (P7) if a b then 0 b a = (b + h) (a + h) and hence - by (P2) again - a + h b + h. (P8) if a b then 0 b a and as also 0 h property (P4) yields 0 (b a)h = bh ah. Now by (P2) again ah bh. If h 0 then by (P6) we have 0 h and hence by (P8) a(h) b(h) and hence ah bh. By adding ah + bh and using (P7) this yields bh ah. (1.83) Proposition: (viz. 342) Let (R, +, ) be a commutative ring and be a total, positive order on R. Then for any a, b R we get the following statements (i) 0 |a| (ii) a |a| (iii) |a|2 = a2 (iv) |a| = sgn(a)a Now suppose that R even is an integral domain, then we get the additional properties for any a, b R (v) |ab| = |a||b| (vi) |a + b| |a| + |b| (vii) ||a| |b|| |a b| (viii) sgn(ab) = sgn(a)sgn(b) (ix) if a, b 0 then a b a2 b2 (1.84) Remark: (i) Due to property (P2) it suces to designate the positive elements of R, that is those a R with 0 a. This fully denes the order . Hence we denote the set of positive elements of R by R+ , formally R+ := { a R | a 0 } And as we have seen in the example (1.80) above, the set R+ is a proper positive cone of R. 1.8 Ordered Rings 75

(ii) If is a total positive order on the commutative ring R, then the positive cone R+ R satises another property a R : a2 R + Prob if a 0 then a2 = a a 0 is trivial by property (P4). Otherwise ( is total) if a 0 then by (P7) we get 0 a and hence 0 (a)(a) = a2 by (P2) again. Thus a2 0 is true for any a R. (iii) If R = 0 is a non-zero ring, that allows a positive order , then R already is of characteristic 0, that is the following map is injective Z R : k k 1R Prob suppose there was some n Z such that n1R = 0, even though n = 0. Then we may take n 1 to be minimal (i.e. n generates the kernel nZ = kn(k k 1R )). As 0R 1R we nd 1R 2R and so on until (n 1)R nR . Hence 0R 1R nR 0R , which yields 0R = 1R , a contradiction to R = 0. Hence the kernel of k k 1R is zero. (iv) () If R is a commutative ring, is a positive order on R and M is an R-module, then there is the notion of a cone C in M : that is a subset C M is said to be a cone in M , i (1) 0 C (2) for all x, y C we get x + y C (3) for all x C and all a 0 we get ax C If we consider R itself as an R-module then a positive cone C R in the ring R need not be a cone in the R-module R, however. As an example let us regard R = R under the usual order , then C = N R is a positive cone, but not a cone (in the sense here). (1.85) Example: (viz. 341) (i) If (R, +, ) is a commutative ring and C R is a proper positive cone in R, then we obtain a positive order on R by letting: ab : baC

(ii) Let R := Z2 be the 2-dimensional lattice under the pointwise operations, that is (a, b) + (c, d) := (a + c, b + d) and (a, b)(c, d) := (ac, bd). We install the following parial order on R: (a, b) (c, d) : a c and b d

Then is a positive order on R even though R is not an integral domain. (iii) The natural order 0 < 1 < 2 < 3 on R = Z4 is not positive, as we have 2 3 but 2 + 1 = 3 > 0 = 3 + 1. 76 1 Groups and Rings

(1.86) Proposition: (viz. 341) Let (R, +, ) be a commutative ring, then we obtain a 1-to-1 correspondence between the proper positive cones C in R and the positive orders on R, by virtue of { C R | C proper positive cone } C R = {a R | a 0}
+

{ R R | positive order } (a b : b a C )

(1.87) Proposition: (viz. 343) () Let (R, +, ) be a commtative ring, be a positive order on R and let N denote the set of positive non-zero divisors of R, that is N := { n R | n 0 and a R : an = 0 = a = 0 } Then N is multiplicatively closed and we denote the localisation of R in N by Q := N 1 R. Thereby we obtain a positive order on Q, by letting a b m n : an bm

1.8 Ordered Rings

77

1.9

Products

(1.88) Denition: Let = I be an arbitrary set and for any i I let (R, +, ) be a semiring. Then we dene the (exterior) direct product of the Ri to be just the carthesian product of the sets Ri Ri :=
i I

a:I
iI

Ri

i I : a(i) Ri

Let us abbreviate this set by R for a moment. Then we may turn this set into a semi-ring (R, +, ) by dening the following addition + and multiplication for elements a and b R a+b : I
i I

Ri : i a(i) + b(i)

ab : I
i I

Ri : i a(i) b(i)

Note that thereby the addition a(i) + b(i) and multiplication a(i)b(i) are those of Ri . Further note that it is customary to write (ai ) instead of a R, where the ai Ri are given to be ai := a(i). And using this notation the above denitions of + and in R take the more elegant form (ai ) + (bi ) := (ai + bi ) (ai ) (bi ) := (ai bi ) For any i I let us now denote the zero-element of the semi-ring Ri by 0i . Then we dene the (exterior) direct sum of the Ri to be the subset of those a = (ai ) R for which only nitely many ai are non-zero. Formally Ri :=
iI

a
i I

Ri

# { i I | a(i) = 0i } <

Let us denote this set by R for the moment, then R s R is a subsemi-ring. In particular we will always understand R and R as semi-rings under these compositions, without explictly mentionining it. (1.89) Example: Probably the most important case is where only two (semi-)rings (R, +, ) and (S, +, ) are involved. In this case the direct product and direct sum coincide, to be the carthesian product R S = R S = { (a, b) | a R, b S }

78

1 Groups and Rings

And if we write out the operations on R S explictly (recall that these were dened to be the pointwise operations) these simply read as (where a, b R and r, s S ) (a, r) + (b, s) = (a + b, r + s) (a, r)(b, s) = (ab, rs) Thereby it is clear that the zero-element of R S is given to be 0 = (0, 0). And if both R and S are rings then the unit element of R S is given to be 1 = (1, 1). It also is clear that R S is commutative i both R and S are such. The remarks below just generalize these observations and give hints how this can be proved. Another familiar case is the following: consider a (semi-)ring (R, +, ) and any non-empty index set I = . In the examples of section 1.4 we have already studied the (semi-)ring RI :=
i I

R = {f | f : I R}

In the previous example as well as in this denition here we have taken the pointwise operations. That is for two elements f , g : I R we have dened the operations f + g : I R = i f (i) + g (i) f g : I R = i f (i) g (i) And it is clear that the zero-element is given to be the constant function 0 : I R : i 0. And if R is a ring then we also have a unit element given to be the constant function 1 : I R : i 1. We continue with the example above. Instead of the direct product we may also take to the direct sum of R over I . That is we may regard RI :=
i I

R = { f : I R | # supp(f ) < }

where supp(f ) := {i I | f (i) = 0}. It is clear RI RI truly is a subset of the direct product. But be careful, as in in any non-zero ring R = 0 we have 1 = 0 we nd that for innite index sets I the unit element 1 : I R : i 1 is not contained in RI . That is RI need not be a ring, even if R is such. (1.90) Remark: As above let = I be an arbitray index set and for any i I let (Ri , +, ) be a (semi-)ring having the zero element 0i (and unit element 1i if any). Then let us denote the direct product of the Ri by R and the direct sum of the Ri by R. Then we would like to remark the following It is clear that (R, +, ) truly becomes a (semi-)ring under the operations above. And if we denote the zero-element of any Ri by 0i then the zero-element of R is given to be 0 = (0i ). 1.9 Products 79

Prob one easily checks all the porperties required for a semi-ring. As an example we prove the associativity of the addition: let a = (ai ), b = (bi ) and c = (ci ) R, then ((a + b) + c)i = (a + b)i + ci = (ai + bi ) + ci = ai + (bi + ci ) = ai + (b + c)i = (a + (b + c))i . Clearly R is commutative if and only if Ri is commutative for any i I . And R is a ring if and only if Ri is a ring for any i I . In the latter case let us denote the unit element of Ri by 1i . Then the unit element of R is given to be 1 = (1i ). Prob if any Ri is commutative, then for any a = (ai ), b = (bi ) R we get (ab)i = ai bi = bi ai = (ba)i which is the commutativity of R. And if R is commutative, then any Ri is commutative, by the same argument. If any Ri is a ring then for any a = (ai ) R we get (1a)i = 1i ai = ai and (a1)i = ai 1i = ai . And if R is a ring, then any Ri is a ring, by the same argument. By denition R R is a subset. But R even is a sub-semi-ring R s R of R. Yet R need not be a subring of R (even if any Ri is a ring), as (1i ) need not be contained in R. Prob consider any a = (ai ) and b = (bi ) R and let us denote the set S (a) := { i I | ai = 0i }. E.g. we get S (0) = and hence 0 R. Further it is clear that S (a + b) S (a) S (b), S (a) = S (a) and S (ab) S (a) S (b). In particular all these sets are nite again and hence a + b, a and ab R, which had to be shown. Note that we get R = R if and only if I is a nite set. And in the case of nitely many semi-rings R1 , . . . , Rn we also use the slightly lax notation (note that thereby R1 Rn = R1 Rn )
n

R1 Rn :=
i=1 n

Ri Ri
i=1

R1 Rn :=

It is clear that the product of (semi-)rings is associative and commutative up to isomorphy. That is if we consider the (semi-)rings (R, +, ), (S, +, ) and (T, +, ), then we obtain the following isomorphies of (semi-)rings RS =s S R : (a, b) (b, a) RS T : (a, b), c a, (b, c) =s R S T Prob it is immediately clear that the above mappings are homomorphisms of (semi-)rings by denition of the operations on the product. And likewise it is clear (this is an elementary property of carthesian products) that these mappings also are bijective. Note that R usually does not inherit any nice properties from the Ri (except being commutative or a ring). E.g. Z is an integral domain, but Z2 = Z Z contains the zero-divisors (1, 0) and (0, 1). 80 1 Groups and Rings

 For any j I let us denote the canonical projection from R (or R respectively) by j . Likewise let j denote the canonical injection of Rj into R (or R respectively). That is we let j : R Rj : (ai ) aj i aj 0i if i = j if i = j

j : Rj R : aj

Then it is clear that i and j are homomorphisms of semi-rings satisfying j j = 1 1j . In particular j is surjective and j is injective. And we will oftenly interpret Rj as a sub-semi-ring of R (or R respectively) by using the isomorphism j : Rj =s im(j ). (1.91) Proposition: (viz. 338) () Let (R, +, ) and (S, +, ) be any commutative rings and consider the ideals a i R and b i R. Then we obtain an ideal a b of R S by letting
a b :=

(a, b) | a a, b b

Conversely let us denote the canonical projections of R S to R and S respectively by : R S R : (a, b) a and : R S S : (a, b) b. If now u i R S is an ideal then we get the identity
u =

(u) (u)

And thereby we obtain an explicit description of all ideals (respecively of all radical, prime or maximal) ideals of R S , to be the following ideal R S = srad R S =
a b |, a ideal R, b ideal S a b |, a srad R, b srad S

spec R S = { p S | p spec R } { R q | q spec S } smax R S = { m S | m smax R } { R n | n smax S }

(1.92) Theorem: (viz. 343) (i) Consider any commutative ring (R, +, ) and two elements e, f R. Further let us denote the respective principal ideals by a := eR and b := f R. If now e and f satisfy e + f = 1 and ef = 0 then we get a + b = R and a b = { 0 }, formally that is e + f = 1, ef = 0 =
a + b = R, a b = { 0 }

(ii) Now let (R, +, ) be any (not necessarily commutative) ring and a, b i R be two ideals of R satisfying a + b = R and a b = { 0 }. Then we obtain the following isomorphy of rings R =r R a R b : x (x + a, x + b)

1.9 Products

81

(iii) Chinese Remainder Theorem Let (R, +, ) be any commutative ring and consider nitely many ideals ai i R (where i 1 . . . n) of R, that are pairwise coprime. That is for any i = j 1 . . . n we assume
ai + aj = R

Let us now denote the intersection of all the ai by a := a1 an . Then we obtain the following isomorphy of rings
n

a =r

R
i=1

ai : x + a (x + a1 , . . . , x + an )

(1.93) Remark: Combining (i) and (ii) we immediately see that, whenever we are given two elements e, f R in a commutative ring R, such that e + f = 1 and ef = 0, then this induces a decomposition of R into two rings R =r R eR R f R : x (x + eR, x + f R) Clearly (ii) is just a special case of the chinese remainder theorem (iii). Nevertheless we gave a seperate formulation (and proof) of the statement to be easily accessible. In fact (ii) talks about decomposing R into smaller rings, whereas (iii) is about solving equations: Suppose we are given the elements ai R (where i 1 . . . n). Then the surjectivity of the isomorphism in (iii) guarantees that there is a solution x R to the following system of congruencies x + a 1 = a1 + a 1 . . . x + a n = an + a n Now let x be any solution of this system, then due to the isomorphhy in (iii) we also nd that the set of all solutions is given to be x + a. () Let us consider an example of the above in the integers R := Z. Let a1 := 8Z and a2 := 15Z. As 8 and 15 are relatively prime we truly nd a1 + a2 = Z. And also a = a1 a2 = 120Z, since the least common multiple of 8 and 15 is 120. It is elementary to verify the following two systems of congruencies 105 + 8Z 105 + 15Z = = 1 + 8Z 0 + 15Z 16 + 8Z 16 + 15Z = = 0 + 8Z 1 + 15Z

Hence if we are given any a1 , a2 Z then we let x := 105a1 +16a2 Z to solve the general system of congruencies x + 8 Z = a1 + 8 Z x + 15Z = a2 + 15Z

82

1 Groups and Rings

Chapter 2

Commutative Rings
2.1 Maximal Ideals

(2.1) Remark: (viz. 302) (i) Let us rst recall the denition of a partial order on some non-empty set X = . This is a relation on X [formally that is a subset of the form X X ] such that for any x, y and z X we get x=y x y, y z x y, y x = = = xy xz x=y

[where we already wrote x y for the formal statement (x, y ) ]. And in this case we also call (X, ) a partially odered set. And in this case it is also customary to use the notation (for any x, y X ) x<y : x y and x = y

(ii) And is said to be a linear order on X (respectively (X, ) is called a linearly ordered set ), i is a partial order such that any two elements of X are comparable, formally x, y X : x y or y y And in this case we may dene the maximum x y and minimum of any two elements x, y X to me the following element of X x y := x y if x y if y x x y := y x if x y if y x

(iii) Example: if S is any set (even S = is allowed), then we may take the power set X := P (X ) = { A | A S } of X . Then the inclusion relation which is dened by (for any two A, B X ) A B : ( x : x A = x B )

83

is a partial order on S . However the inclusion almost never is a total order. E.g. regard S := { 0, 1 }. Then the elements { 0 } and { 1 } X of X are not comparable. In what follows we will usually consider a set X ideal R P (R) of ideals of a commutative ring R under the inclusion relation . (iv) Now let (X, ) be a partially ordered set and A X be a subset. Then we dene the sets A of minimal and A of maximal elements of A to be the following A := { a A | a A : a a = a = a } A := { a A | a A : a a = a = a } And an element a A is said to be a minimal element of A. Likewise a A is said to be a maximal element of A. Note that in general it may happen that A has several minimal (or maximal) elements or even none at all. Example: N = { 0 } and N = under the usual order on N. (v) If (X, ) is a linearly ordered set and A X is a subset, then maximal and minimal elements of A are uniquely determined. Formally that is a, b A a, b A = = a=b a=b

(vi) Let again (X, ) be a linearly ordered set and A = { a1 , . . . , an } X be a nite subset of X . Then there is a (uniquley) determined minimal element A (resp. maximal element a ) of A. And this is given to be A = { a } where a = (a1 a2 ) . . . an A = { a } where a = (a1 a2 ) . . . an (vii) Once again let (X, ) be any partially ordered set and A X be a subset of X . Then it is clear that induces another partial order A on A by restricting A := () (A A)

And we will write for A again, that is we write (A, ) instead of the more correct form (A, A ). Now a subset C X is said to be a chain in X i (C, ) is linearly ordered. Equivalently that is i any two elements of C are comparable, formally x, y C : x y or y x (viii) Lemma of Zorn Let (X, ) be a partially ordered set (in particular X = ). Further suppose that any chain C of X has an upper bound u in X , formally C X : C chain = u X c C : c u 84 2 Commutative Rings

Then X already contains a (not necessarily) unique maximal element x X x X : x x = x = x Nota though it may come as a surprise the lemma of Zorn surely is the single most powerful tool in mathematics! We will see its impact on several occasions: e.g. the existence of maximal ideals or the existence of bases (in vector-spaces). However we do ask the reader to refer to the literature (on set-theory) for a proof. In fact the lemma of Zorn is just one of the equivalent reformulations of the axiom of choice. Hence you might as well the lemma of Zorn as a set-theoretic axiom. (ix) In most cases we will use a quite simple form of the lemma of Zorn: consider an arbitrary collection of sets Z . Now verify that (1) Z = is non-empty (2) if C Z is a chain in Z (that is C = and for any A, B C we get A B or B A) then we get C Z again Then (by the lemma of Zorn) Z already contains a (not necessarily unique) -maximal element Z Z . That is for any A Z we get Z A = A=Z

(x) Of course the lemma of Zorn can also be applied to nd minimal elements. In analogy tho the above we obtain the following special case: consider an arbitrary collection of sets Z . Now verify that (1) Z = is non-empty (2) if C Z is a chain in Z (that is C = and for any A, B C we get A B or B A) then we get C Z again Then (by the lemma of Zorn) Z already contains a (not necessarily unique) -minimal element Z Z . That is for any A Z we get A Z = A=Z

(2.2) Denition: Let (R, +, ) be a commutative ring and = A ideal R be a non-empty family of ideals of R. Then we recall the deniton of maximal a A and minimal a A elements (concerning the order ) of A: A := { a A | a A : a a = a = a } A := { a A | a A : a a = a = a }

2.1 Maximal Ideals

85

(2.3) Denition: Let (R, +, ) be a commutative ring, then m is said to be a maximal ideal of R, i the following three statements hold true (1) m i R is an ideal (2) m = R is proper (3) m is a maximal element of ideal R \ { R }. That is for any ideal a i R
m a

a = m or a = R

The set smax R of all maximal ideals is called maximal spectrum of R and the intersection of all maximal ideals is called Jacobson radical of R smax R := { m R | (1), (2) and (3) } jacR := smax R

Nota in the light of this denition the maximal spectrum is nothing but the maximal elements of the set of non-full ideals: smax R = (ideal R \ { R }) . (2.4) Lemma: (viz. 360) (i) Let (R, +, ) be a commutative ring and = A ideal R be a chain of ideals (that is for any a, b A we have a b or b a). Then the union of all the a A is an ideal of R, formally = A ideal R chain = A i R ideal

(ii) Let R = a i R be a non-full ideal in the commutative ring (R, +, ), then there is a maximal ideal of R containing a, formally R = a i R = m smax R : a m

(iii) In a commutative ring (R, +, ) the group of units R of R is precisely the complement of the union of all maximal ideals of R, formally R = R \ smax R

(iv) The zero-ring is the one and only ring without maximal ideals, formally R=0 smax R =

86

2 Commutative Rings

(2.5) Proposition: (viz. 372) Let (R, +, ) be a commutative ring and m i R be an ideal of R. Then the following four statements are equivalent (a) m is a maximal ideal (b) the quotient ring R/m is a eld (c) the quotient ring R/m is simple - i.e. it only contains the trivial ideals
a i R m

a = { 0 + m } or a = R m

(d) m is coprime to any principal ideal that is not already contained in m, formally that is: for any a R we get the implication am =
m + aR = R

(2.6) Proposition: (viz. 372) Let (R, +, ) be any commutative ring, then we can reformulate the jacobson radical of R: for any j R the following two statements are equivalent (a) j jacR (b) a R we get 1 aj R And if a i R is an ideal of R, then a is contained in the jacobson radical i 1 + a is a subgroup of the multiplicative group. That is equivalent are (a) a jacR (b) 1 + a R (c) 1 + a g R (2.7) Proposition: (viz. 373) Let (R, +, ) be a commutative ring and let { m1 , . . . , mk } smax R be nitely many maximal ideals of R. Then we obtain the following statements i = j 1 . . . n : mi + mj = R
m1 mk = m1 . . . mk m1 m1 m2 . . . m1 m2 . . . mk

2.1 Maximal Ideals

87

2.2

Prime Ideals

(2.8) Denition: Let (R, +, ) be a commutative ring, then p is said to be a prime ideal of R, i the following three statements hold true (1) p i R is an ideal (2) p = R is proper (3) a, b R we get ab p = a p or b p And the set spec R of all prime ideals of R is called spectrum of R, formally spec R := { p R | (1), (2) and (3) } An ideal p i R is said to be a minimal prime ideal of R, i it is a minimal element of spec R. That is p spec R and for any prime ideal p spec R we get the implication p p = p = p. Let us nally denote the set of all minimal prime ideals of R by smin R := spec R

(2.9) Proposition: (viz. 373) Let (R, +, ) be a commutative ring and p i R be an ideal of R. Then the following four statements are equivalent (a) p is a prime ideal (b) the quotient ring R/p is a non-zero (R/p = 0) integral domain (c) the complement U := R \ p is multiplicatively closed, this is to say that (1) 1 U (2) u, v U = uv U (d) p = R is non-full and for any two ideals a, b i R we get the implication
ab p

a p or b p

(2.10) Remark: At this point we can reap an important harvest of the theory: maximal ideals are prime. Using the machinery of algebra this can be seen in a most beautiful way: if m is a maximal ideal, then R/m is a eld by (2.5). But elds are (by denition non-zero) integral domains. Hence m already is prime by (2.9). Of course it is also possible to give a direct proof - an example of such will be presented for (2.19).

88

2 Commutative Rings

(2.11) Proposition: (viz. 374) Let (R, +, ) be a commutative ring and p i R be a prime ideal of R, then (i) If a1 , . . . , ak R are nitely many elements then by induction it is clear that we obtain the following implication a1 . . . ak p = i 1 . . . k : ai p

(ii) Likewiese let a1 , . . . , ak i R be nitely many ideals of R, then by induction we also nd the following implication
k

ai p
i=1

i 1 . . . k : ai p

(iii) prime avoidance Consider some ideals b1 , . . . , bn i R where 2 n N such that b3 , . . . , bn spec R are prime. If now a i R is another ideal, then we obtain the following implication
n

a
i=1

bi

i 1 . . . n : a bi

(2.12) Example: In a less formalistic way the prime avoidance lemma can be put like this: if an ideal a is contained in the union of nitely many ideals bi of which at most two are non-prime, then it already is contained in one of these. In spite of the odd assumption the statement is rather sharp. In fact it may happen that a is contained in the union of three stricly smaller ideals (which of course are all non-prime in this case). Let us give an example of this: x 2 2 the eld E := Z2 and the ideal o := s, t 2 i = s , st, t i i E [s, t]. Then the ring in consideration is R := E [s, t] o Note that any residue class of f E [s, t] can be represented in the form f + o = f [0, 0] + f [1, 0]s + f [0, 1]t + o. That is R contains precisely 6 elements, namely a + bs + ct + o where a, b and c E = Z2 . Now take the following ideal
a := o { f + o | f E [s, t], f [0, 0] = 0 }

s + o, t + o

= =

{ 0 + o, s + o, t + o, s + t + o }

2.2 Prime Ideals

89

On the other hand let us also take the following ideals b1 := (s + o)R, b2 := (t + o)R and b2 := (s + t + o)R. Now an easy computation (e.g. for b1 we get (s + o)(a + bs + ct + o) = as + o) yields
b1 := { 0 + o, s + o } b2 := { 0 + o, t + o } b3 := { 0 + o, s + t + o }

And hence it is immediately clear that any bi a is a strict subset, but the union of these ideals precisely is the ideal a again
a = b1 b2 b3

(2.13) Lemma: (viz. 375) Let (R, +, ) and (S, +, ) be any two commutative rings and denote their prime spectra by X := spec (R) and Y := spec (S ) respectively. Now consider some ring-homomorphism : R S . Then induces a well-dened map on the spectra of S and R, by virtue of spec () : Y X : q 1 (q) For any element a R let us denote Xa := { p X | a p } and for any ideal a i R let us denote V(a) := { p X | a p }. If now a R, a i R and b i S , then spec () satises the following properties (spec )1 (Xa ) (spec ) V(b) (spec )1 (V(a)) = = Y(a) V 1 (b) V ( (a) i )

(2.14) Proposition: (viz. 375) (i) Let (R, +, ) be a commutative ring and P spec R be a chain of prime ideals of R (that is for any p, q P we have p q or q p). Then the intersection of all the p P is a prime ideal of R, formally = P spec R chain = P spec R prime

(ii) Let = P spec R be a non-empty set of prime ideals of the commutative ring (R, +, ). And assume that P is closed under , i.e. p spec R q P : p q = p P Then P already contains a minimal element, formally that is P = .

90

2 Commutative Rings

(iii) Let (R, +, ) be a commutative ring and consider an arbitrary ideal a i R and a prime ideal q i R. We now give a short list of possible conditions we may impose and the respective assumptions that have to be supposed for a and q in this case assumption condition(p) R=0 none nothing p q a p a=R a q a p q Then there is a prime ideal p of R that is minimal among all prime ideals satisfying the condition imposed. Formally that is = { p spec R | condition(p) } (iv) Let (R, +, ) be a commutative ring, a i R be any ideal and q i R a prime ideal with a q. Then there is a prime ideal p minimal over a that is contained in q. Formally that is p { p spec R | a p } with a p q

(v) Let (R, +, ) be a commutative ring, a i R be an ideal of R and U R be a multiplicatively closed set (that is 1 U and u, v U implies uv U ). If now a U = then the set of all ideals b i R satisfying a b R \ U contains maximal elements. And any such is maximal element is a prime ideal, formally =
b i R | a b R \ U

spec R

(2.15) Corollary: (viz. 470) (i) Let (R, +, ) be a non-zero R = 0, commutative ring, then the set of zero-divisors of R is the union of a certain set of prime ideals of R zdR = { p spec R | p zdR }

(ii) In a commutative ring (R, +, ) the minimal prime ideals are already contained in the zero-divisors of R, formally that is
p smin R

p zdR

2.2 Prime Ideals

91

2.3

Radical Ideals

(2.16) Denition: Let (R, +, ) be a commutative ring, a i R be an ideal of R and b R be an element. Then we dene the radical a and fraction a : b of a to be
a :=

a R | k N : ak a

a : b := { a R | ab a }

Now a is said to be a radical ideal (sometimes this is also called perfect ideal in the literature), i a equals its radical, that is i we have (1) a i R (2) a = a And thereby we dene the radical spectrum srad R of R to be the set of all radical ideals of R, formally that is srad R :=
a i R | a =

(2.17) Proposition: (viz. 377) Let (R, +, ) be a commutative ring, a i R be an ideal of R and b R be an element. Then we obtain the following statements (i) The fraction a : b is an ideal of R again containing a, formally that is
a

a:b

(ii) The radical

a even is a radical ideal of R containing a, formally again a

a srad R

(iii) Let (R, +, ) be a commutative ring, a, b i R ideals and b R, then

a b a b

= =

a:b b:b a b

(iv) Let (R, +, ) be a commutative ring, a i R an ideal and b R, then

a:b = R

b a

Thus if p i R even is a prime ideal then we have precisely two cases

p:b =

R p

if b p if b p

(v) The intersection of a collection A = of radical ideals is a radical ideal = A srad R 92 = A srad R 2 Commutative Rings

(vi) Let again (R, +, ) be a commutative ring, ai i R be an arbitrary family of ideals (i I ) and b R. Then we nd the equality (ai : b) =
iI i I

ai

:b

(2.18) Proposition: (viz. 378) Let (R, +, ) be a commutative ring and a i R be an ideal of R. Then the following four statements are equivalent (a) a a (b) a = a is a radical ideal (c) a R we get k N : ak a = a a (d) the quotient ring R/a is reduced, i.e. it contains no non-zero nilpotents nil R a = {0 + a}

(2.19) Corollary: (viz. 378) Let (R, +, ) be a commutaive ring and m, p and a i R be ideals of R respectively. Then we have gained the following table of equivalencies
m maximal p prime a radical

m eld R p non-zero integral domain

a reduced

where a ring S is said to be reduced, i sk = 0 (for some k N) implies s = 0. And subsuming the properties of the respective quotient rings we thereby nd the inclusions smax R spec R srad R

2.3 Radical Ideals

93

(2.20) Proposition: (viz. 378) (i) The radical of a is the intersection of all prime ideals containing a, i.e.
a =

{ p spec R | a p } { p spec R | a p }

(ii) In particular the nil-radical is just the intersection of all prime ideals nilR = 0 = spec R = smin R

(iii) In a commutative ring (R, +, ) the map a a is a projection (i.e. idempotent). That is for any ideal a i R we nd
a =

(iv) Let (R, +, ) be a commutative ring, a i R be an arbitrary and p i R be a prime ideal. If now k N, then we get the implication
ak p

p=

(v) Let (R, +, ) be a commutative ring, a, b i R be ideals, then we get

ab = ab =

(vi) And thereby - if (R, +, ) is a commutative ring, 1 k N and a i R is an ideal, then we obtain the equality
ak =

(vii) Let a i R be a nitely generated ideal in the commutative ring (R, +, ). And let b i R be another ideal of R. Then we get
a b

k N : ak b

(viii) Let again (R, +, ) be a commutative ring and ai i R be an arbitrary family of ideals (i I ), then we obtain the following inclusions
i I

iI

ai

ai
i I

i I

ai

94

2 Commutative Rings

(2.21) Remark: Due to (2.17.(iii)) the intersection of prime ideals yields a radical ideal. And in (2.19) we have just seen that maximal and prime ideals already are radical. Now recall that the jacobson-radical jacR has been dened to be the intersection of all maximal ideals. And we have just seen in (2.20.(ii)) above that the nil-radical is the intersection of all (minimal) prime ideals. This means that both are radical ideals (and hence the name) jacR = nilR = smax R srad R spec R srad R

(2.22) Example: Note however that taking sums and radicals of (even nitely many) ideals does not commute. That is there is no analogous statement to (2.20.(vi)) concernong sums a + b. As an example consider R := E [s, t], the polynomial ring in two variables, where (E, +, ) is an arbitrary eld. Now pick up
p := (s2 t)R and q := tR

It is clear that p and q are prime ideals (as the quotient is isomorphic to the integral domain E [s] under the isomorphisms induced by f (s, t) f (s, s2 ) and f (s, t) f (s, 0) respectively). However the sum p + q of both not even is a radical ideal, in fact we get
p + q = s2 R + tR sR + tR = p+q

Prob We rst prove a + q = s2 R + tR - thus consider any h = (s2 t)f + tg p + q, then h can be rewritten as h = s2 f + t(g f ) s2 R + tR. But conversely t q p + q and s2 = (s2 t) + t p + q. In particular s is contained in the radical of p + q and hence p + q sR + tR p + q. As sR + tR is prime (even maximal) this implies the equality of the radical ideal. So nally we wish to prove s p + q (and in particlar p + q is no radical ideal. Thus suppose s = s2 f + tg for some polynomials f , g R. Then (as s is prime and does not divide t) s divides g . To be precise we get g = sh for h = (1 st)/t R. And hence we get s = s2 f + sth. Eliminating s we nd 1 = sf + th, an equation that cannot hold true (as 1 is of degree 0 and sf + th of degree at least 1). Thus we got s2 s2 R + tR. (2.23) Example: (viz. 381) Be aware that radicals may be very large when compared with their original ideals. E.g. it may happen that a = aR is a principal ideal, but a is not nitely generated. In fact in the chapter of proofs we give an example where there even is no n N such that ( a)n a, as well.

2.3 Radical Ideals

95

(2.24) Remark: Combining (2.17.(iii)) and (viii) in the above proposition it is clear that for an arbitrary collection of ideals ai i R (where i I ) in a commutative ring (R, +, ) we obtain the following equalities
i I

a =
iI

ai ai
i I

iI

ai =

(2.25) Example: The inclusions in (2.20.(viii)) need not be equalities! As an example let us consider the ring of integers R = Z. And let us take I = N and ai = pi Z for some prime number p Z. As any non-zero element of Z is divisible by nite powers pi only the intersection of the ai is { 0 }. On the other hand ai = pZ due to (2.20.(vii)) and since pZ is a prime (and hence radical) ideal of Z. Thereby we found the counterexample
ai =
iN

0 = 0 pZ =
i N

pZ =
i N

ai

(2.26) Proposition: (viz. 381) Let now : R S be a ring-homomorphism between the commutative rings (R, +, ) and (S, +, ). And consider the ideals a i R and b i S . Then let us denote the transfered ideals of a and b
b R := 1 (b) i R aS

:=

(a)

i S

Note that in case that is surjective (but not generally), we get aS = (a). Using these notions we obtain the following statements (i)
bR = bR

(ii)

aS

aS =

aS

(iii) Now let us abbreviate by any of the words maximal ideal, prime ideal or radical ideal. Then we nd the equivalence
b is a

b R is a

And if : R

S even is surjective then we get another equivalence

a + kn() is a

aS is a

96

2 Commutative Rings

2.4

Noetherian Rings

In this section we will introduce and study noetherian and artinian rings. Both types of rings are dened by a chain condition on ideals - ascending in the case of noetherian and descending in the case of artinian rings. So one might expect a completely dual theory for these two objects. Yet this is not the case! In fact it turns out that the artinian property is much stronger: any artinian ring already is noetherian, but the converse is false. An while examples of noetherian rings are abundant, there are few artinian rings. On the other hand both kinds of rings have several similar properties: e.g. both are very well-behaved under taking quotients or localisations. Throughout the course of this book we will see that noetherian rings have lots of beautiful properties. E.g. they allow the Lasker-Noether decomopsition of ideals. However in non-noetherian rings peculiar things may happen. Together with the easy properties of inheritance this makes noetherian rings one of the predominant objects of commutative algebra and algebraic geometry. Artinian rings are of lesser importance, yet they arise naturally in number theory and the theory of algebraic curves. (2.27) Denition: (viz. 413) Let (R, +, ) be a commutative ring, then R is said to be noetherian i it satises one of the following four equivalent conditions (a) R satises the ascending chain condition (ACC) of ideals - that is any family of ideals ak i R of R (where k N) such that
a0 a1 . . . ak ak+1 . . .

becomes eventually constant - that is there is some s N such that

a0 a1 . . . as = as+1 = . . .

(b) Every non-empty family of ideals contains a maximal element, formally that is: for any = A ideal R there is some a A . (c) Every ideal of R is nitely generated, formally that is: for any ideal a i R there are some a1 , . . . , ak R such that we get
a = a1 , . . . , a k
i

= a1 R + + ak R

(d) Every prime ideal of R is nitely generated, formally that is: for any prime ideal p spec R there are some a1 , . . . , ak R such that
p = a1 , . . . , a k
i

= a1 R + + ak R

(2.28) Remark: (viz. 413) Parts of these equivalencies are no specialty of commutative rings, but a general fact of ordered sets. To be precise: if (X, ) is a partially ordered set, then the following two statements are equivalent: 2.4 Noetherian Rings 97

(a) Any non-empty A = subset A X of X has a maximal element. That is there is some a A such that a A : a a = a = a (b) Any ascending chain within X stabilizes, that is for any (xk ) X (where k N) such that x0 x1 x2 . . . we have s N i N : xs+i = xs We will now dene artinian rings (by property (a) for the formalists among us). And again we will nd an even longer list of equivalent statements. The similarity in the denition gives strong reason to study these objects simultaneously. But as often only one half of the equivalencies can be proved straightforwardly while the other half requires heavy-duty machinery that is not yet available. For a novice reader it might hence be wise to concentrate noetherian rings only. These will be the important objects later on. Artinian rings not only are far less common but they also are severely more dicult to handle. (2.29) Denition: (viz. ??) Let (R, +, ) be a commutative ring, then R is said to be artinian i it satises one of the following six equivalent conditions (a) R satises the descending chain condition (DCC) of ideals - that is any family of ideals ak i R of R (where k N) such that
a0 a1 . . . ak ak+1 . . .

becomes eventually constant - that is there is some s N such that

a0 a1 . . . as = as+1 = . . .

(b) Every non-empty family of ideals contains a minimal element, formally that is: for any = A ideal R there is some a A . (c) R is of nite length as an R-module - that is there is an upper bound on the maximal length of a strictly ascending chain of ideals, formally (R) := sup kN a0 , a1 , . . . , ak i R : a0 a1 . . . ak <

(d) R is a noetherian ring and any prime ideal already is maximal, formally smax R = spec R (e) R is a noetherian ring and even any minimal prime ideal already is a maximal ideal, formally again smax R = smin R 98 2 Commutative Rings

(f) R is a noetherian ring whose jacobson radical equals its nil-radical (i.e. jacR = nilR) and that is semi-local (i.e. #smax R < ). (2.30) Example: Any eld E is an artinian (and noetherian) ring, as it only contains the trivial ideals 0 and E . In particular the chain of ideals with strict inclusions is 0 E , and this is nite. Any nite ring - such as Zn - is artinian (and noetherian), as it only contains nitely many ideals (and in particular any chain of ideals has to be eventually constant). We will study principal ideal domains (PIDs) in section 2.6. These are integral domains R in which every ideal a i R is generated by some element a R (that is a = aR). In particular any ideal is nitely generated and hence any PID is noetherian. The integers Z form an easy example of a noetherian ring (as they are a PID). But they are not artinian, e.g. they contain the innitely decreasing chain of ideals Z 2Z 4Z 8Z . . . 2k Z . . . We will soon see (a special case of the Hilbert basis theorem) that the polynomial ring R = E [t1 , . . . , tn ] in nitely many variables over a eld E is a noetherian ring. And again this is no artinian ring, as it contains the innitely decreasing chain of ideals R t1 R t2 1R . . . Of course there are examples of non-noetherian rings, too. E.g. let E be a eld and consider R := E [ti | 1 i N] the polynomial ring in (countably) innitely many indeterminates. Then R is an integral domain but it is not noetherian. Just let pk := t1 , t2 , . . . , tk i be the ideal generated by the rst k indeterminates. Then we obtain an innitely ascending chain of (prime) ideals p1 p2 . . . pk . . . Non-noetherian rings may well lie inside of noetherian rings. As an example regard the ring R = E [ti | 1 i N] again. As R is an integral domain R is contained in its quotient eld F . But as F is a eld it in particular is noetherian. () We would nally like to present an example from complex analysis: let O := { f : C C | f holomorphic } be the ring of entire functions (note that this is isomorphic to the ring C{z } of globally convergent power series). Now let ak := { f O | k n N : f (n) = 0 }. Clearly the ak i O are ideals of O they even form a strictly ascending chain of ideals (this is a special case of the Weierstrass product theorem). Again O is an integral domain (by the identity theorem of holomorphic functions) but - as we have just seen - it is not noetherian.

2.4 Noetherian Rings

99

(2.31) Proposition: (viz. 416) Let (R, +, ) be a commutative ring and x as an abbreviation for either of the words noetherian or artinian. Then the following statements hold true: (i) If a i R is an ideal and R is , then the quotient R/a also is a ring.

(ii) Let (S, +, ) be any ring and let : R S be a surjective ringhomomorphism. If now R is a ring, then S is a ring, too. (iii) Suppose R and S are commutative rings. Then the direct sum R S is a ring if and only if both R and S are rings. (2.32) Remark: We are about to formulate Hilberts basis theorem. Yet in order to do this we rst staighten out a few thigs for the reader who is not familiar with the notion of an R-algebra. Consider a commutative ring (S, +, ) and a subring R r S of it. Further let E S be any subset, then we introduce the R-subring of S generated by E to be the following n m R[E ] := ai ei,j m, n N, ai R, ei,j E
i=1 j =1

Nota that 0 R[X ] as m = 0 and 1 R[E ] as n = 0 are allowed. And thereby R[E ] is a subring of S , containing R. In fact R[E ] is precisely the R-subalgebra of S generated by E in the sense of section 3.2. And for those who already know the notion of polynomials in several variables this can also be put in the following form R[E ] = { f (e1 , . . . , en ) | 1 n N, f R[t1 , . . . , tn ], ei E } In particular if E is a nite set - say E = { e1 , . . . , ek } - we also write R[e1 , . . . , ek ] := R[E ]. And this is just the image of the polynomial ring R[t1 , . . . , tn ] under the evaluation homomorphism ti ei . (2.33) Theorem: (viz. 417) Hilbert Basis Theorem Let (S, +, ) be a commutative ring and consider a subring R r S . Further suppose that S is nitely generated (as an R-algebra) over R, that is there are nitely many e1 , . . . , en S such that S = R[e1 , . . . , en ]. If now R is a noetherian ring, then S is a noetherian ring, too R noetherian = S = R[e1 , . . . , en ] noetherian

100

2 Commutative Rings

(2.34) Remark: The major work in proving Hilberts Basis Theorem lies in regarding the polynomial ring S = R[t]. The rest is just induction and (2.31.(iii)). And for this we chose a constructive proof. That is given a non-zero ideal 0 = u i S we want to nd a nite set of generators of u. To do this let
ak := { lc (f ) | f u, deg(f ) = k } { 0 }

As ak is an ideal of R it is nitely generated, say ak = ak,1 , . . . , ak,n(k) i where ak,i = fk,i for some fk,i u with deg(fk,i = k and ak,i = lc (fk,i ). Further the ak form an ascending chain and hence there is some s N such that as = as+1 = as+2 = . . . . Then a trick argument shows
u =

fk,1 | k 0 . . . s, i 1 . . . n(k )

(2.35) Denition: Let (R, +, ) be any commutative ring and q , q i R be a prime ideals of R. Then we say that q lies directly under q if q is maximal among the prime ideals contained in q. And we abbreviate this by writing q q. Formally that is
q q

q { p spec R | p q }

(1) q spec R with q q (2) p spec R : q p q = q = p or p = q

(2.36) Proposition: (viz. 419) Let (R, +, ) be a noetherian ring, then we obtain following three statements (i) Any non-empty set of ideals of R contains a maximal element. That is for any = A ideal R there is some a A such that a a we get a a =
a = a

(ii) Any non-empty set of prime ideals of R contains maximal and minimal elements. That is for any = P spec R we get P = and P = (iii) In particular for any two prime ideals p, q i R with p q there is a prime ideal q lying directly under q. Formally that is q spec R : p q q (iv) Let a i R be any ideal except a = R, then there only are nitely many prime ideals lying minimally over a. Formally that is 1 # { p spec R | a p } <

2.4 Noetherian Rings

101

(2.37) Proposition: (viz. ??) Let (R, +, ) be an artinian ring, then we obtain following seven statements (i) An artinian integral domain already is a eld, that is the equivalence R artinian, integral domain R eld

(ii) Any artinian ring already is a noetherian ring, that is the implication R artinian = R noetherian

(iii) Maximal, prime and minimal prime ideals of R all coincide, formally smax R = spec R = smin R (iv) R is a semilocal ring, that is it only has nitely many maximal ideals #smax R < (v) The jacobson radical and nil-radical coincide and are nilpotent, that is jacR = nilR and n N : (jacR)n = 0 (vi) The reduction R/nilR is isomorphic to the (nite) direct sum of its residue elds - to be precise we get R nilR =r
m smax R

(vii) R can be decomposed into a nite number of local artinian rings. That is there are local, artinian rings Li (where i 1 . . . r := #smax R) with R =r L1 Lr

(2.38) Corollary: (viz. 419) Let (R, +, ) be any commutative ring, then R is the union of a net of noetherian subrings. That is there is a family (Ri ) (where i I ) of subrings such that the following statements hold true (1) Ri r R is a subring for any i I (2) Ri is a noetherian ring for any i I (3) R =
iI

Ri

(4) i, j I k I such that Ri Rj Rk

102

2 Commutative Rings

2.5

Unique Factorisation Domains

The reader most probably is aware of the fact, that any integer n Z (apart from 0) admits a decomposition into primes. In this and the next section we will study what kinds of rings qualify to have this property. We will see that neotherian integral domains come close, but do not suce. And we will see that PIDs will allow prime decomposition. That is the truth lies somewhere in between but cannot be pinned down precisely. According to the good, old traditions of mathematics were in for a denition: a ring will simply said to be factorial, i it allows prime decomposition (a mathematical heretic might object: If you cant prove it, dene it!). Then PIDs are factorial. And the lemma of Gauss is one of the few answers in which cases the factorial property is preserved. So we have to start with a lengthy series of (luckily very easy) denitions. But once we have assembled the language needed for prime decomposition we will be rewarded with several theorems of great beauty - the fundamental theorem of arithmetic at their center. (2.39) Denition: (viz. 420) Let (R, +, ) be a commutative ring and a, b R be two elements of R. Then we say that a divides b (written as a | b) i one of the following three equivalent properties is satised (a) b aR (b) bR aR (c) h R : b = ah And we dene the order of a in b to be the highest number k N such that ak still divides b (respectively if ak divides b for any k N) a|b := sup{ k N | b ak R } N { }

Now let R even be an integral domain. Then we say that a and b are associated (written as a b) i one of the following three equivalent properties is satised (a) aR = bR (b) a | b and b | a (c) R : b = a

2.5 Unique Factorisation Domains

103

(2.40) Proposition: (viz. 421) (i) Let (R, +, ) be any commutative ring then for any a, b and c R the relation of divisibility has the following properies 1 | b a | 0 a | ab 0 | b a | 1 a | b = b=0 a R ac | bc

(ii) If a, b and u R such that u nzdR is a non-zero divisor of R, then we even nd the following equivalence a | b au | bu

(iii) Divisibility is a reexive, transitive relation that is antisymmetric up to associativity. Formally that is for any a, b and c R a | a a | b and b | c a | b and b | a = = a | c ab

(iv) Let (R, +, ) be an integral domain, then associateness is an equivalence relation on R. And for any a R its equivalence class [a] under is given to be aR . Formally that is [a] = aR := { a | R } (2.41) Remark: If (R, +, ) is an integral domain and a, b R such that a = 0 and a | b then the divisor h R satisfying b = ah is uniquely determined. As we sometimes wish to refer to the divisor, it deserves a name of its own. Hence if R is an integral domain, a, b R with a = 0 and a | b we let a := h such that b = ah b Prob we have to show the uniqueness: thus suppose b = ag and b = ah then a(g h) = 0. And as a = 0 and R is an integral domain this yields g h = 0 and hence g = h.

104

2 Commutative Rings

(2.42) Example: Let (R, +, ) be an commutative ring and a, b R. Let us denote the ideal generated by a and b by a := aR+bR. Now consider any polynomial f R[t], 1 k N and let d := deg(f ). Then a straightforward computation yields f (a)ak f (b)bk = (a b)
i=0 d k1+i

aj bkj

f [i]
j =0

A special case of this is the third binomial rule a2 b2 = (a b)(a + b) which is obtained by letting f (t) := t. Thus this cumbersome double sum on the right is the divisor of f (a)ak f (b)bk times a b. And in the case of an integral domain (and a = b) this even is uniquely determined. And from the explict representation of the divisor one also nds f (a)ak f (b)bk ab
d k1+i

=
i=0

f [i]
j =0

aj bkj ak1

(2.43) Remark: Let (R, +, ) be a commutative ring, recall that we have dened the set of relevant elements of R to be the non-zero, non-units of R R := R \ R { 0 } R is empty if and only if R = 0 or R is a eld, that is equivalent are R = R = 0 or R eld

Prob if R = 0 then trivially R = . Conversely R = if and only if R = R \ {0}. And this is just the denition of R being a eld. The complement of R is multiplicatively closed (recall that this can be put as: 1 R and a, b R = ab R ) R \ R = R { 0 } is multiplicatively closed Prob rst of all we have 1 R {0} = R \ R (wether R = 0 or not). Now consider a, b R \ R . If a = 0 or b = 0 then ab = 0 and hence ab R \ R . Else if a, b = 0 then a, b R and hence ab R , too. If R is an integral domain, then R is closed under multiplication, i.e. R integral domain = a, b R = ab R

Prob if a, b R then a, b = 0 and hence ab = 0, as R is an integral domain. Also we get a, b R and likewise this implies ab R (else we had a1 = b(ab)1 for example). Together this means ab R .

2.5 Unique Factorisation Domains

105

(2.44) Denition: Let (R, +, ) be a commutative ring and p R be an elements of R. Then p is said to be irreducible if p is a relevant element of R but cannot be further factored into relevant elements. Formally i (1) p R (2) a, b R : p = ab = a R or b R Likewise p R is said to be a prime element of R, i it is a relevant element of R such that it divides (at least) one of the factors in every product it divides. Formally again, i (1) p R (2) a, b R : p | ab = p | a or p | b (2.45) Denition: Let (R, +, ) be a commutative ring and a R and let abbreviate either the word prime or the word irreducible. Then a tupel (, p1 , . . . , pk ) (with k N and k = 0 allowed!) is said to be a decompostion of a, i (1) R (2) a = p1 . . . pk (3) i 1 . . . n : pi R is If (, p1 , . . . , pk ) at least satises (1) and (3) then let us agree to call it a series. Now two series (, p1 , . . . , pk ) and (, q1 , . . . , ql ) are said to be essentially equal (written as (, p1 , . . . , pk ) (, q1 , . . . , ql )), i k = l, both decompose the same element and the pi and qi are pairwise associated. Formally that is (, p1 , . . . , pk ) (, q1 , . . . , ql ) : (1), (2) and (3) where (1) k = l (2) p1 . . . pk = q1 . . . ql (3) Sk : i 1 . . . k : pi q(i) (2.46) Example: () Let (R, +, ) be an integral domain and a R, then the linear polyonmial p(t) := t a R[t] is prime. Note that this need not be true unless R is an integral domain. E.g. for R = Z6 we have t 1 = (2t + 1)(3t 1) but t 1 | 2t + 1, t 1 | 3t 1 Prob consider f , g R[t] such that p | f g . that is there is some h R[t] such that (t a)h = f g . In particular f (a)g (a) = (f g )(a) = (a a)h(a) = 0. As R is an integral domain this means f (a) = 0 or g (a) = 0. And from this again we get p = t a | f or p = t a | g .

106

2 Commutative Rings

(2.47) Denition: Let (R, +, ) be a commutative ring again and = A R be a nonempty subset of R. Then we say that m R is a common multiple of A, i every a A divides m, formally A | m : a A : a | m

And thereby m is said to be a least common multiple of A, i m is a common multiple that divides any other common multiple (1) A | m (2) n R : A | n = m | n And we denote the set of least common multiples of A by lcm(A) i.e. lcm(A) := { m R | (1) and (2) } Analogous to the above we say that an element d R is a common divisor of A, i d divides every a A, formally again d | A : a A : d | a

And thereby d is said to be a greatest common divisor of A, i d is a common divisor that is divided by any other common divisor (1) d | A (2) c R : c | A = c | d And we denote the set of greatest common divisors of A by gcd(A) i.e. gcd(A) := { d R | (1) and (2) } Finally A is said to relatively prime, i 1 is a greatest common divisor of A. And by (2.40) this can also be formulated, as A relatively prime : 1 gcd(A) c R : c | A = c R

By abuse of notation a nite family of elements a1 , . . . , ak R is said to be relatively prime, i the set {a1 , . . . , ak } is relatively prime. (2.48) Proposition: (viz. 421) (i) Let (R, +, ) be a commutative ring, p R be a prime element and a1 , . . . , ak R be arbitrary elements of R. Then we obtain
k

p |
i=1

ai

i 1 . . . k : p | ai

2.5 Unique Factorisation Domains

107

(ii) Let (R, +, ) be a commutative ring and p R. Then p is a prime element i p = 0 and the principal ideal pR i R is a prime ideal p R prime p = 0 and pR i R prime

(iii) Let (R, +, ) be a commutative ring, p R and R a unit of R. Let us again abbreviate the word prime or irreducible by , then p is p is

(iv) Let (R, +, ) be an integral domain 0 = a R and b R be a non-zero non-unit. Then we obtain the following strict inclusion of ideals (ab)R aR

(v) Let (R, +, ) be an integral domain. Then any prime element of R already is irreducible. That is for any p R we get p is prime = p is irreducible

(vi) Let (R, +, ) be an integral domain. Then the set D of all d R that allow a prime decomposition is saturated and multiplicatively closed. That is D := {p1 . . . pk | R , k N, pi R prime} satises 1 c, d D D cd D

(vii) Let (R, +, ) be an integral domain, p1 , . . . , pk , q1 , . . . , ql R be a nite collection of prime elements of R (where k , l N with k = 0 and l = 0 allowed) and , R be two units of R. Then we obtain p1 . . . pk = q1 . . . ql = k = l and Sk such that i 1 . . . k : pi q(i)

(viii) Let (R, +, ) be an integral domain and = A R be a non-empty subset of R. Further let d, m R be two elements of R, then m lcm(A) d gcd(A) lcm(A) = mR gcd(A) = dR

(2.49) Theorem: (viz. 423) Let (R, +, ) be a noetherian integral domain and x some elements a, b and p R where p is prime. Then we obtain the following statements (i) If a R then the order in which a divides b is nite, formally that is a|b 108 N 2 Commutative Rings

(ii) The order of a prime element p R turns multiplications into additions p | ab = p|a + p|b

(iii) Any a = 0 admits an irreducible decomposition. That is there are some R and nitely many irreducibles p1 , . . . , pk R such that a = p1 . . . pk (iv) If 0 = p i R is any prime ideal of R then p is generated by nitaly many irreducibles. That is there are p1 , . . . , pk R irreducible with
p =

p1 , . . . , p k

(v) Let P R be a set of parwise non-associate non-zero non-units of R. That is we assume that for any p, q P we get p q = p = q . Then for any a = 0 the number of p P that divide a is nite, i.e. # { p P | a pR } < (vi) Let b R be a non-zero, non-unit of R. Further consider a nite collection q1 , . . . , qk R of prime elements of R that are pairwise non-associate. That is for any i, j 1 . . . k we assume qi qj = i=j

If now p R is any prime element of R then we obtain the statements

k

a :=
i=1

qi

qi |b

p | b

p |

b xor i 1 . . . k : p qi a

(2.50) Denition: (viz. 426) We call (R, +, ) an unique factorisation domain (which we will always abbreviate by UFD), i R is an integral domain that further satises one of the following equivalent properties (a) Every non-zero element admits a decomposition into prime elements. That is for any 0 = a R we get the following statement (, p) = (, p1 , . . . , pk ) prime decomposition of a (b) Every non-zero element admits an essentially unique decomposition into irreducible elements. That is for any 0 = a R we have (1) (, p) = (, p1 , . . . , pk ) irreducible decomposition of a (2) (, p), (, q) irreducible decompositions of a = (, p) (, q) 2.5 Unique Factorisation Domains 109

(c) Every non-zero element admits a decomposition into irreducible elements and every irreducible element is prime. Formally that is (1) 0 = a R (, p) irreducible decomposition of a (2) p R : p prime p irreducible (d) The principal ideals of R satisfy the ascending chain condition and every irreducibe element is prime, Formally that is (1) for any family of elements ak R (where k N) of R such that a0 R a1 R . . . ak R ak+1 R . . . there is some s N such that for any i N we get as+i R = as R. (2) p R : p prime p irreducible (e) Any non-zero prime ideal of R contains a prime element of R, formally 0 = p spec R p P prime : p p (f) () Let D := {p1 . . . pk | R , k N, pi R prime} denote the set of all d R that allow a prime decomposition again. Then either R = 0 is the zero ring or the quotient eld of R is precisely the locaslisation of R in D, formally quotR = D1 R (2.51) Remark: The zero-ring R = 0 trivially is an UFD. As it does not contain a single non-zero element such that condition (a) of UFDs is void (hence true). Propery (b) guarantees that - in an UFD (R, +, ) - any nonzero element 0 = a R admits an irreducible (equivalently prime) decomposition a = p1 . . . pk , that even is essentially unique. And this uniqueness determines the number k of irreducible factors. Hence we may dene the length of a to be precisely this number (a) := k where (, p1 , . . . , pk ) prime decomposition of a Any eld (E, +, ) is an UFD, as we allowed k = 0 for a prime decomposition (, p1 , . . . , pk ). To be precise if 0 = a E then a E already is a unit and hence (a) is a prime decomposition of a. In the next section we will prove that any PID (i.e. an integral domain in which every ideal can be generated by a single element) is an UFD. If (R, +, ) is a noetherian integral domain in which any irreducible element is prime, then R is an UFD. This is clear from property (d). If (R, +, ) is an UFD then so is the polynomial ring R[t] (this is the one of the lemmas of Gauss that will be proved in (??)). () If (R, +, ) is an UFD and U R is multiplicatively closed, then U 1 R is an UFD too. In fact, if p R is a prime element then p/1 either is a unit or a prime in U 1 R. This will be proved in (2.110). 110 2 Commutative Rings

 () A subring O r C of the complex numbers is said to be an algebraic number ring, i any a O satises an integral equation over Z (that is there is a normed polynomial f Z[t] such that f (a) = 0). We will see that any algebraic number ring O is a Dedekind domain. And it is also true, that O is an UFD if and only if it is a PID. This list of UFDs is almost exhaustive. At the end of this section we will append a list of counter examples such that the reader may convince himself that the above equivalencies cannot be generalized. (2.52) Remark: It is oftenly useful not to regard all the prime (analogously irreducible) elements of R but to restrict to a representative set modulo associateness. That is a subset P R such that we obtain a bijection of the form P { p R | p prime } : p pR Prob this is possible since is an equivalence relation on R of which the equivalence classes are precisely aR by (2.40). In particular also is on equivalence relation on the subset P of prime elements of R. And by (2.48) we know that for any p P we have [p] = { q P | p q } = pR again. Hence we may choose P as a representing system of P/ by virtue of the axiom of choice. (2.53) Example: The units of the integers Z are given to be Z = {1, +1}. That is associateness identies all elements having the same absolute value, aZ = {a, a}. And thereby we can choose a representing system of Z modulo simply by choosing all positive elements a 0. Thus we nd a representing system of the prime elements of Z by P := { p Z | p prime, 0 p } () Now let (E, +, ) be any eld and consider the polynomial E [t]. Then we will see that the units of E [t] are given to be E [t] = E = {at0 | 0 = a E }. Thus associateness identies all polynomials having the same coecients up to a common factor. That is we get f E [t] = {af | 0 = a E }. Thus by requiring a polynomial f to be normed (i.e. f [deg(f )] = 1) we eliminate this ambiguity. Therefore we nd a representing system of the prime elements of E [t] by P := { p E [t] | p prime, normed } (2.54) Theorem: (viz. 427) Let (R, +, ) be an UFD, p R be a prime element and 0 = a, b R be any two non-zero elements. Then the following statements hold true (i) Suppose (, p1 , . . . , pk ) is a prime decomposition of a. Then for any prime p R the order of p in a can be expressed as p|a = # { i 1 . . . k | p pi } N 111

2.5 Unique Factorisation Domains

(ii) The order of a prime element p R turns multiplications into additions p | ab = p|a + p|b

(iii) a divides b if and only if for any prime p the order of p in a does not exceed the order of p in b. Formally that is the equivalence a | b p R prime : p | a p | b

(iv) Let P be a representing system of the prime elements modulo associateness. Then for any 0 = a R we obtain a unique representation ! R ! n = (n(p))
pP

N : a=
pP

pn(p)

and thereby we even have n(p) = p | a such that the sum of all n(p) is precisely the length of a (and in particular nite). Formally that is (a) =
pP

n(p) <

Nota that n p N, that is n : P N is a map such that the number of p P with n(p) = 0 is nite. And hence the product over all pn(p) (where p P) well-dened using the notations of section 1.3. (v) Let P be a representing system of the prime elements modulo associateness again. And consider a non-empty subset = A R such that 0 A. Then A has a greatest common divisor, namely pm(p) gcd(A) where m(p) := min{ p | a | a A }
pP

Likewise if A = {a1 , . . . , an } R is a non-empty, nite (1 n N) subset with 0 A then A has a least common multiple, namely pn(p) lcm(A) where n(p) := max{ p | a | a A }
pP

(vi) Let A := { a, b } and d gcd(A) be a greatest common divisor, m lcm(A) a least common multiple of a and b. Then we obtain ab dm (vii) If 0 = a, b, c R are any three non-zero elements of R, then the least common multiple and greatest common divisor of these can be evaluated recursively. That is if d gcd{a, b} and m lcm{a, b} then gcd { a, b, c } = gcd { d, c } lcm { a, b, c } = lcm { m, c }

112

2 Commutative Rings

(2.55) Proposition: (viz. 429) () Any UFD R is normal. Thereby a commutative ring (R, +, ) is said to be normal (or integrally closed), i R is an integral domain that is integrally closed in its quotient eld. That is R is normal, i (1) R is an integral domain (2) for any a, b R with a = 0 get: if there is some normed polynomial f R[t] such that f (b/a) = 0 quotR, then we already had a | b. (2.56) Example: (viz. 430) (i) UFDs need not be noetherian! As an example regard a eld (E, +, ), then the polynomial ring in countably many variables E [ti | i N] is an UFD by the lemma of Gauss (??)). But it is not noetherian (by the examples in section 2.4). (ii) Noetherian integral domains need not be UFDs! As an example we would like to present the following algebraic number ring Z[ 3] := a + ib 3 | a, b Z C

It is quite easy to see that in this ring 2 Z[ 3] is an irreducible element, that is not prime. Hence Z[ 3] cannot be an UFD. Never the less it is an integral domain (being a subring of C) and noetherian (by Hilberts basis theorem (2.33), as it is generated by 3 over Z). (iii) Residue rings (even integral domains) of UFDs need not be UFDs! As an example let us start with the polynomial ring R[s, t] in two variables over the reals. Then R[s, t] is an UFD by the lemma of Gauss (??). Now consider the residue ring R := R[s, t] p where p := (s2 + t2 1)R[s, t] Then p(s, t) = s2 + t2 1 R[s, t] is an irreducible (hence prime) polynomial, such that p is a prime ideal. Yet we are able to prove that R is no UFD (in fact we will prove that there is an irreducible, non-prime element in a certain localisation of R).

2.5 Unique Factorisation Domains

113

2.6

Principal Ideal Domains

(2.57) Denition: Let (R, +, ) be an integral domain and be a mapping of the following form : R \ { 0 } N { } Then the ordered pair (R, ) is said to be an Euclidean domain, i R allows division with remainder - that is for any a, b R with a = 0 there are q , r R such that we get (1) b = qa + r (2) (r) < (a) or r = 0 (2.58) Remark: We will employ a notational trick to eliminate the distinction of the case r = 0 in the denition above. To do this we pick up a symbol and set < n for any n N. Then will be extended to : R N { } : a (a) if a = 0 if a = 0

Thereby the property to allow division with remainder simply reads as: for any a, b R with a = 0 there are q , r R such that we get (1) b = qa + r (2) (r) < (a) If (R, ) is an Euclidean domain and a R satises (a) = 0, then a R already is a unit of R. The converse need not be true however. That is we have the incluison (but not necessarily equality) { a R | (a) = 0 } R

Prob using division with remainder we may nd q , r R such that 1 = qa + r and (r) < (a) = 0. But this can only be if (r) = and hence r = 0. But this means 1 = qa again and hence a R . (2.59) Example: (viz. 432) (i) If (E, +, ) is a eld, then (E, ) is an Euclidean domain under any function : E \ {0} N. Because for any 0 = a E and b E we may let q := ba1 and r := 0. (ii) The ring Z of integers is an Euclidean domain (Z, ) under the (slightly modied) absolute value as an Euclidean function a a if a > 0 if a < 0 if a = 0

: Z N { } : k

114

2 Commutative Rings

In fact consider 1 a Z and any b Z then for q := b div a and r := b mod a (see below for a denition) we have b = aq + r and 0 r < a (in particular (r) < (a)). And in case a < 0 we can choose q and r similarly to have division with remainder. b div a := max{ q Z | aq b } b mod a := b b div a a

(iii) If (E, +, ) is a eld then the polynomial ring E [t] (in one variable) is an Euclidean domain (E [t], deg) under the degree deg : E [t] \ { 0 } : f max{ k N | f [k ] = 0 } This is true because of the division algorithm for polynomials - see (??). Note that the division algorithm can be applied as any non-zero polynomial can be normed, since E is a eld. Hence R[t] will not be a Euclidean ring, unless R is a eld. Also, though the division algorithm for polynomials can be generalized to Buchbergers algorithm it is untrue that the polynomial ring E [t1 , . . . , tn ] is an Euclidean domain (for n 2) under some Euclidean function . (iv) Consider any d Z such that d Q (refer to (??) for this). Then we consider the subring of C generated by d, that is we regard Z[ d] := a + b d | a, b Z C

d] is unique (that is for Then the representation x = a + b d Z[ any a, b, f and g Z we get a + b d = f + g d (a, b) = (f, g )). And thereby we obtain a well-dened ring-automorphism on Z [ d] by virtue of x = a + b d x := a b d. Now dene : Z[ d] \ { 0 } N : x = a + b d |xx| = |a2 db2 | Then is multiplicative, that is (xy ) = (x) (y ) for any x, y Z[ d]. And it is quite easy to see that the units of Z[ d] are given to be Z[ d] = x Z[ d] (x) = 1

We will also prove that for d 3 the ring Z[ d] will not be an UFD, as 2 is an irreducible element, that is not prime. Yet for d {2, 1, 2, 3}, we nd that (Z[ d], ) even is an Euclidean domain.

(2.60) Proposition: (viz. 434) Let (R, ) be an Euclidean domain and 0 = a, b R be two non-zero elements. Then a and b have a greatest common divisor g R that can be computed by the following (so called Euclidean ) algorithm 2.6 Principal Ideal Domains 115

input initialisation

algorithm

output

0 = a, b R if (a) (b) then (f a, g b) else (f b, g a) while f = 0 do begin choose q , r R with (g = qf + r and (r) < (f )) g f, f r end-while g

That is given 0 = a, b R we start the above algorithm that returns g R and for this g we get g gcd{a, b}. If now g denotes the greatest common divisor g gcd{a, b} of a and b, then there are r and s R such that g = ra + sb. And these r and s can be computed (along with g ) using the following renement of the Euclidean algorithm input initialisation 0 = a, b R if (a) (b) then (f1 b, f2 a) else (f1 a, f2 b) k1 repeat k k+1 choose qk and fk+1 R with (fk1 = qk fk + fk+1 and (fk+1 ) < (fk )) until fk+1 = 0 n k , g fn r2 1, s2 0 r3 q2 , s3 1 for k = 3 to n 1 step 1 do begin rk+1 rk1 qk rk sk+1 sk1 qk sk end-for if (a) (b) then (r rn , s sn ) else (r sn , s rn ) g , r and s

algorithm

output

(2.61) Example: As an application we want to compute the greatest common divisor of a = 84 and b = 1330 in Z. That is we initialize the algorithm with f1 := b = 1330 and f2 := a = 84. Then we nd 1330 = 15 84 + 70 (at k = 2), which is q2 = 15 and f3 = 70. Thus in the next step (now k = 3) we observe 84 = 1 70 + 14, that is we may choose q3 := 1 and f4 := 14. Now (at counter k = 4) we have 70 = 5 14 + 0 which means q4 = 5 and f5 = 0. Thus for n := k = 4 we have nally arrived at fk+1 = 0 terminating the rst part of the algorithm. It returns g = fn = 14 the greatest common divisor of a = 84 and b = 1330. The following table summarizes the computations 116 2 Commutative Rings

k 2 3 4

fk 1 1330 84 70

fk 84 70 14

fk+1 70 14 0

qk 15 1 5

Now in the second part of the algorithm we initialize r2 := 1, s2 := 0, r3 := q2 = 15 and s3 := 1. Then we compute r4 := r2 q3 r3 = 16 and s4 := s2 q3 s3 = 1. As n = 4 this already is r = r4 = 16 and s = s4 = 1. And in fact ra + bs = 14 = g . Again we summarize these computations in k 2 3 4 qk 15 1 5 rk 1 -15 16 sk 0 1 -1

(2.62) Proposition: (viz. 435) Consider an integral domain (R, +, ) and n N, then we recursively dene R0 := R = { a R | b R : ab = 1 } Rn+1 := { 0 = a R | b R r Rn { 0 } : a | b r } (i) The sets Rn form an ascending chain of subsets of R, i.e. for any n N R0 R1 . . . Rn Rn1 . . . (ii) If R(, ) is an Euclidean domain that for any n N the set of a R with a = 0 and (a) n is contained in Rn , formally that is { a R | (a) n } Rn { 0 } (iii) Conversely if the Rn cover R, that is R \ {0} = n Rn , then (R, ) becomes an Euclidean domain under the following Euclidean function : R \ { 0 } : a min{ n N | a Rn } And thereby we obtain the following statements for any a, b R, b = 0 (a) (ab) (b) = (ab) a R

(iv) In particular we obtain the equivalency of the following two statements R \ {0} =
nN

Rn

: (R, ) Euclidean domain

(v) If R(, ) is an Euclidean domain, then (R, ) is an Euclidean domain, too and for any a R we get (a) (a). That is is the minimal Euclidean function on R.

2.6 Principal Ideal Domains

117

(2.63) Denition: Let (R, +, ) be any commutative ring, then an ideal a i R is said to be principal, i it can be generated by a single element, i.e. a = aR for some a R. And R is said to be a principal ring i every ideal of R is principal, that is i a i R a R : a = aR Finally R is said to be a principal ideal domain (which we will always abbreviate by PID), i it is an integral domain that also is principal, i.e. (1) R is an integral domain (2) R is a principal ring (2.64) Remark: Clearly the trivial ideals a = 0 and a + R are always principal, as they can be generated by the elements 0 = 0R and R = 1R respectively. If R is a principal ring and a i R is any ideal, then the quotient ring R/a is principal, too. Prob due to the correspondence theorem (1.61) any ideal u i R/a is of the form u = b/a for some ideal a b i R. Thus there is some b R with b = bR. Now let u := b + a, then clearly u = { bh + a | h R } = u(R/a) is a principal ideal, too. Let R1 , . . . , Rn be principal rings, then the direct sum R1 Rn is a principal ring, too. Prob by (1.91) the ideals of the direct sum are of the form a1 an for some ideals ai i Ri . Thus by assumption there are some ai Ri such that ai = ai Ri . And thereby it is clear, that a1 an = (a1 , . . . , an )R1 Rn is principal again. (2.65) Lemma: (viz. 437) (i) If (R, +, ) is a non-zero PID then all non-zero prime ideals of R already are maxmial (in a fancier notation kdimR 1), formally that is spec R = smax R { 0 } (ii) If (R, +, ) is a PID then R already is a noetherian ring and an UFD. That is any 0 = a R admits an (essentially uniqe) decomposition a = p1 . . . pk into prime elements pi - viz. (2.50). (iii) If (R, ) is an Euclidean domain then R already is a PID. In fact if 0 = a i R is a non-zero ideal then a = aR for any a a satisfying (a) = min{ (b) | 0 = b a } (iv) If (R, +, ) is an integral domain such that any prime ideal is principal, then R already is a PID. Put formally that is the equivalency R PID p spec R p R : p = pR

118

2 Commutative Rings

(2.66) Remark: In (2.59.(i)) we have seen that any eld E is an Euclidean domain (under any ). Hence any eld is a PID due to (iii) above. But this is also clear from the fact that the only ideals of a eld E are 0 and E itself. And these are principal due to 0 = 0E and E = 1E . Also (i) is trivially satised: the one and only prime ideal of a eld is 0. The items (i) and (ii) in the above lemma are very useful and will yield a multitude of corollaries (e.g. the lemma of Kronecker that will give birth to eld theory in book II). And (iii) provides one of the reasons why we have introduced Euclidean domains: combining (iii) and (ii) we see that any Euclidean domain is an UFD. And this yields an elegant way of proving that a certain integral domain R is an UFD. Just establish an Euclidean function on R and voil` a R is an UFD. Due to example (2.59.(ii)) we know that (Z, ) is an Euclidean domain. By (iii) above this means that Z is a PID and (ii) then implies that Z is an UFD. By denition of UFDs this means that Z allows essentially unique prime decomposition. This now is a classical result that is known as the Fundamental Theorem of Arithmetic. We have proved an even stronger version, namely (R, ) Euclidean domain = R is an UFD

In the subsequent proposition we will discuss how far a PID deviates from being an Euclidean domain (though this is of little practical value). To do this we will introduce the notion of a Dedekind-Hasse norm. We will then see that an Euclidean function can be easily turned into a Dedekind-Hasse norm and that R is a PID if and only if it admits a Dedekind-Hasse norm. This of course is another way of proving that any Euclidean domain is a PID. (2.67) Proposition: (viz. 438) (i) If (R, +, ) is a PID then R admits a multiplicative Dedekind-Hasse norm. That is there is a function : R N satisfying the following three properties for any a, b R (1) (ab) = (a) (b) (2) (a) = 0 a = 0 (3) if a, b = 0 then b aR or r aR + bR such that (r) < (a) Nota to be precise we may dene (0) := 0 and for any 0 = a R we let (a) := 2k for k := (a) the length of any decomposition of a = p1 . . . pk into prime elements. (ii) If (R, +, ) is an integral domain and is a Dedekind-Hasse norm on R (that is : R N is a function satisfying (2) and (3) in (i)), then R already is a PID. In fact if 0 = a i R is a non-zero ideal then we get a = aR for any 0 = a a satisfying (a) = min{ (b) | 0 = b a } 2.6 Principal Ideal Domains 119

(iii) If (R, ) is an Euclidean domain then we obtain a Dedekind-Hasse norm on R (that is : R N satises (2) and (3) in (i)) by letting : RN : a 0 (a) + 1 if a = 0 if a = 0

(2.68) Lemma: (viz. 438) of Bezout Let (R, +, ) be an integral domain and = A R be a non-empty subset of R. Then we obtain the following statements (i) The intersection of the ideals aR (where a A) is a principal ideal if and only if A admits a least common multiple. More precisely for any m R we obtain the following equivalency aR = mR
aA

m lcm(A)

(ii) If the sum of the ideals aR (where a A) is a principal ideal then A admits a greatest common divisor. More precisely for any d R we obtain the following equivalency aR = dR
aA

d gcd(A)

Nota the converse implication is untrue in general. E.g. regard the polynomial ring R = Z[t], then 1 gcd(2, t) but the ideal 2R + tR is not principal. The converse is true in PIDs however: (iii) If R even is a PID then the sum of the ideals aR (where a A) is the principal ideal generated by the greatest common divisor of A. More precisely for any d R we obtain the following equivalency aR = dR
aA

d gcd(A)

In particular in a PID a greatest common divisor d gcd(A) can be written as a linear combination of the elements a A (ba ) RA : d =
aA

aba

Nota in an Euclidean ring (R, ) (supposed that A is nite) the Euclidean algorithm (2.60) provides an eective method to (recursively, using (2.54.(viii))) compute these elements ba . (iv) Let R be a PID again and consider nitely many non-zero elements 0 = a1 , . . . , an R such that the ai are pairwise relatively prime (that is i = j 1 . . . n = 1 gcd(ai , aj )). If now b R then there are b1 , . . . , bn R such that we obtain the following equality (in the quotient eld quotR of R) b b1 bn = + + a1 . . . an a1 an 120 2 Commutative Rings

In an UFD R any collection = A R of elements of R has a greatest common divisor d, due to (2.54.(v)). And we have just seen that in a PID the greatest common divisor d, that can even be written as a linear combination d = a aba . In an Euclidean domain d and the ba can even be computed algorithmically, using the Euclidean algorithm. We now ask for the converse, that is: if any nite collection of elements of R has a greatest common divisor, that is a linear combination is R a PID already? The answer will be yes for noetherian rings, but no in general. (2.69) Denition: (viz. 439) Let (R, +, ) be an integral domain, then R is said do be a Bezout domain i it satises one of the following three equivalent properties (a) Any two elements a, b R admit a greatest common divisor, that can be written as a linear combination of a and b. Formally that is a, b R r, s R : ra + sb gcd { a, b } (b) For any a, b R the ideal aR + bR is principal, that is there is some d R such that we get dR = aR + bR. Formally again that is a, b R d R : dR = aR + bR (c) Any nitely generated ideal a i R of R is principal, formally that is a i R : a nitely generated =
a principal

(2.70) Corollary: (viz. 440) Let (R, +, ) be any ring ring, then the following statements are equivalent (a) R is a PID (b) R is an UFD and Bezout domain (c) R is a noetherian ring and Bezout domain (2.71) Example: (i) Let R := Z + tQ[t] := {f Q[t] | f (0) Z} Q[t] be the subring of polynomials over Q with constant term in Z. Then R is an integral domain [as it is a subring of Q[t]] having the units R = { 1, 1 } only [as f g = 1 implies f (0)g (0) = 1 and deg(f ) = 0 = deg(g )]. However R neither is noetherian, nor an UFD, as it contains the following innitely ascending chain of principal ideals tR 1 1 1 tR . . . n tR n+1 tR . . . 2 2 2

[the inclusion is strict, as 2 is not a unit of R]. (Also the ideal tQ[t] i R is not nitely generated, as else there would be some a Z such that atQ[t] tZ[t], which is absurd). Yet it can be proved that R is a Bezout domain (see [Dummit, Foote, 9.3, exercise 5] for hints on this). 2.6 Principal Ideal Domains 121

(ii) We want to present another example of a Bezout domain. Fix any eld (E, +, ) and let S := E [tk | k N] be the polynomial ring in countably many variables. Then we dene the following ideal
u :=

tk t2 k+1 | k N

Then R := S/u is a Bezout domain (refer to [Dummit, Foote, 9.2, exercise 12] for hints). However it is no noetherian ring (in particular no PID), as the following ideal a of R is not nitely generated
a :=

tk + u | k N

(iii) () Let us denote denote the integral closure of Z in C by O. That is if we denote the set of normed polynomials over Z by Z[t]1 (that is Z[t]1 := {tn + a1 tn1 + + an Z[t] | n N, ai Z}), then O := { z C | f Z[t]1 : f (z ) = 0 } It will be proved in book II that O is a subring of C (and in particular an integral domain, as C is a eld). In fact O is a Bezout domain that satises spec O = smax O {0} (see [Dummit, Foote, chapter 16.3, exercise 23] for hints how to prove this). However O neither is noetherian, nor an UFD, as it contains the following innitely ascending chain of principal ideals 2O 2 O . . . 21/2 O 21/2
n n+1

O ...

(iv) Consider := (1 + i 19)/2 C and consider the following subring Z[ ] := {a + b | a, b Z} C. Then we obtain a Dedekind-Hasse norm on Z[ ] by letting : Z[ ] N : a + b a2 + ab + 5b2 (this is proved in [Dummit, Foote, page 282]). In particular Z[ ] is a PID by virtue of (2.67.(ii)). However Z[ ] is not an Euclidean domain (under any function whatsoever). The latter statement is proved in [Dummit, Foote, page 277].

(2.72) Remark: By now we have studied a quite respectable list of dierent kinds of commutative rings from integral domains and elds over noetherian rings to Bezout domains. Thereby we have found several inclusions, that we wish to summarize in the diagram below 122 2 Commutative Rings

integral domains UFDs

Bezout domains PIDs Euclidean domains elds

noetherian rings artinitan rings

Note however that every single inclusion in this diagram is strict. E.g. there are integral domains that are not Bezout domains and there are PIDs that are not Euclidean domains. The following table presents examples of rings R that satisfy one property, but not the next stronger one R Z[ 3] Z + tQ[t] Q [s, t] Z[(1 + 19)/2] Z is noetherian integral domain Bezout domain noetherian UFD PID Euclidean domain is not UFD noetherian, UFD Bezout domain Euclidean domain eld, artinian ring

2.6 Principal Ideal Domains

123

2.7

Lasker-Noether Decomposition

(2.73) Denition: (viz. 440) Let (R, +, ) be a commutative ring and a i R be a proper (i.e. a = R) ideal of R, then a is said to be a primary ideal of R, i it satises one of the following equivalencies (a) The radical of a contains the set of zero-divisors of R/a (as R-module) zdR R a
a

(b) In the ring R/a the set of zero-divisors is contained in the nil radical zd R a nil R a

(c) The set of zero-divisors of R/a (as a ring) equals the nil-radical of R/a zd R a = nil R a

(d) For any two elements a, b R of R we obtain the following implication ab a, b a = a

a

(e) For any two elements a, b R of R we obtain the following implication ab a, b

a

aa

And if a is a primary ideal, then it is customary to say that a is associated to (the prime ideal) a. Finally the set of all primary ideals of R is said to be the primary spectrum of R and we will abbreviate it by writing spri R := { a i R | a = R, (a) } And if p spec R is a prime ideal of R, then we denote the set of all primary ideals of R associated to p, by sprip R :=
a spri R

a=p

Nota clearly associateness is an equivalency relation on spri R. That is for = b and thereby any primary ideals a and b i R we let a b : a is an equivalency relation on spri R. Further if p = a, then sprip R is precisely the equivalency class of a under .

124

2 Commutative Rings

(2.74) Remark: In the denition above there already occured the notion of zero-divisors of a module (which will be introduced in 3.2). Thus we shortly wish to present what this means in the context here. Let (R, +, ) be a commutative ring and a i R be an ideal. Then the set of zero-divisors of R/a as an R-module will be dened to be zdR R a := = aR aR 0 = b + a R a : a(b + a) = 0 0 = b + a R a : (a + a)(b + a) = 0

because of a(b + a) := ab + a = (a + a)(b + a) (this is just the denition of the scalar multiplication of R/a as an R-module). Now have a look at the set of zero-divisors of R/a zd R a = a+aR a 0 = b + a R a : (a + a)(b + a) = 0

We nd that the condition in both of these sets coincides. And it also is clear that a is contained in zdR (R/a) [unless a = R, as in this case we have ab a for b = 1 R]. Thus we nd zd R a = zdR (R/a) a

(2.75) Proposition: (viz. 441) Let (R, +, ) be a commutative ring and a i R be an ideal. Then we obtain the following statements (i) Any prime ideal of R already is primary, that is we obtain the inclusion spec R spri R

(ii) If a is a primary ideal of R, then the radical a is a prime ideal of R. In fact it is the uniquely determined smallest prime ideal containing a, that is we obtain the following implications
a spri R a spri R

=
a

a spec R

= { p spec R | a p }

(iii) If m := a is a maximal ideal of R, then a is a primary ideal of R associated to m, that is we obtain the following implication
a smax R

a spri R

(iv) Suppose that m smax R is a maximal ideal of R, such that there is some k N such that mk a m. Then a is a primary ideal of R associated to m. That is we obtain the following implication
mk a m

a sprim R

2.7 Lasker-Noether Decomposition

125

(v) If a1 , . . . , ak i R are nitely many primary ideals of R, all associated to the same pirme ideal p = ai . Then their intersection is a primary ideal associated to p, too
a1 , . . . , ak sprip R

a1 ak sprip R

(vi) If a i R is a primary ideal and u R \ a, then a : u i R is another primary ideal, associated to the same prime ideal, that is
a sprip R, u a

a : u sprip R

(vii) Consider a homomorphism : R S between the commutative rings (R, +, ) and (S, +, ). If now b i S is a primary ideal of S , then 1 a := (b) is a primary ideal of R. And if q := b denotes the ideal b is associated to, then a is associated to 1 (q). Formally that is
b spriq S

1 (b) spri1 (q) R

(viii) () Let U R be a multiplicatively closed set, p i R be a prime ideal of R with p U = and denote by q := U 1 p i U 1 R the corresponding prme ideal of U 1 R. Then we obtain a one-to-one correspondence, by sprip R
a bR

spriq U 1 R U 1 a
b

(2.76) Example: (viz. 442) (i) Let (R, +, ) be an integral domain and p R be a prime element of R. Then for any 1 n N the ideal a := pn R is primary and associated, to the prime ideal a = pR. (ii) In a PID (R, +, ) the primary ideals are precisely the ideals 0 and pn R for some p R prime. Formally that is the identity spri R = { pn R | 1 n N, p R prime } { 0 } (iii) Primary ideals do not need to be powers of prime ideals. As an example let (E, +, ) be any eld and consider R := E [s, t]. Then the ideal a := s, t2 i is primary and associated to the maximal ideal m := s, t i . In fact we get m2 a m and hence there is no prime ideal p i R such that a = pk for some k N. (iv) Powers of prime ideals do not need to be primary ideals. As an example let (E, +, ) be any eld again and consider R := E [s, t, u]/u where u := u2 st i . Let us denote a := s + u, b := t + u and c := u + u. Then we obtain a prime ideal p i R by letting p := b, c i . However we will see that a := p2 is no primary ideal.

126

2 Commutative Rings

(2.77) Denition: Let (R, +, ) be a commutative ring, then p is said to be an irreducible ideal of R, i it satises the following three properties (1) p i R is an ideal (2) p = R is proper (3) p is not the intersection of nitely many larger ideals. That is for any two ideals a, b i R we obtain the implication
p=ab

p = a or p = b

(2.78) Theorem: (viz. 443) (i) Let (R, +, ) be a commutative ring, then any prime ideal of R already is irreducible. That is for any ideal p i R we get the implication
p prime

p irreducible

(ii) Let (R, +, ) be a noetherian ring, then any irreducible ideal of R already is primary. That is for any ideal p i R we get the implication
p irreducible

p primary

(iii) Let (R, +, ) be a noetherian ring, then any proper ideal a i R, a = R admits an irreducible decomposition (p1 , . . . , pk ). That is there are ideals pi i R (where i 1 . . . k and 1 k N) such that (1) i 1 . . . k : pi is an irreducible ideal (2) a = p1 pk (2.79) Denition: Let (R, +, ) be a commutative ring and a i R be an ideal of R. Then a tupel (a1 , . . . , ak ) is said to be a primary decomposition of a, i we have (1) i 1 . . . k : ai spri R is a primary ideal of R (2) a = a1 ak And (a1 , . . . , ak ) is said to be minimal (or irredundant) i we further get (3) j 1 . . . k : a =
ai (4) i = j 1 . . . k : ai = aj
i= j

Finally if (a1 , . . . , ak ) is a minimal primary decomposition of a, then let us denote pi := ai . Then we dene the set of associated prime ideals and isolated and embedded components of (a1 , . . . , ak ) to be the following ass(a1 , . . . , ak ) := { p1 , . . . , pk } iso(a1 , . . . , ak ) := { ai | i 1 . . . k, pi ass(a1 , . . . , ak ) } =
ai

i 1 . . . k, j 1 . . . k : pj pi = j = i

emb(a1 , . . . , ak ) := { a1 , . . . , ak } \ iso(a1 , . . . , ak ) 2.7 Lasker-Noether Decomposition 127

Nota that is the isolated components belong to the minimal associated prime ideals. And a ai is said to be an embedded component if it is not an isolated component, that is if its prime ideal pi is not minimal. These notions have a geometric interpretation, whence the odd names come from. (2.80) Example: Let (R, +, ) be an UFD and a R with 0 = a R . Then we pick up a nk 1 primary decomposition of a, that is a = pn 1 . . . pk where 1 k N and for any i, j 1 . . . k we have pi R is prime, 1 ni N and pi R = pj R implies i = j . Then we obtain a minimal primary decomposition of a = aR
i (a1 , . . . , ak ) where ai := pn i R

And for this primary decomposition we nd pi =

ai = pi R and furthermore

ass(a1 , . . . , ak ) := { p1 , . . . , pk } iso(a1 , . . . , ak ) := { a1 , . . . , ak }

(2.81) Proposition: (viz. 444) Let (R, +, ) be a commutative ring and a i R be an ideal of R. If a admits any primary decomposion, then it already admits a minimal primary decomposition. More precisely let (a1 , . . . , ak ) be a primary decomposition of a. Then we select a subset I 1 . . . k such that #I = min #J J 1 . . . k, a = aj
j J

Now dene an equivalency relation on I by letting i j : ai = aj . And denote the quotient set by A := I/ . Then we have found a minimal primary decomposition (a ) (where A) of a by letting
a :=
i

ai

(2.82) Proposition: (viz. 445) Let (R, +, ) be a commutative ring, a i R be a proper a = R ideal and consider (a1 , . . . , ak ) a minimal primary decomposition of a. Let us further denote pi := ai , then we obtain the following statements (i) Clearly the number of associated prime ideals of the decomposition equals the number of primeary ideals of the decomposition. Formally #ass(a1 , . . . , ak ) = k (ii) For any u R we obtain the following identity (where by convention the intersection over the empty set is dened to be the ring R itself)
a:u =

{ pi | i 1 . . . k, u ai } 2 Commutative Rings

128

(iii) () If pi is minimal among ass(a1 , . . . , ak ) then we can recunstruct ai from a and pi . To be precise for any i 1 . . . k we get
pi ass(a1 , . . . , ak )

a : (R \ pi ) = ai

(iv) Let us introduce ass(a) := spec R { a : u | u R} the set of prime ideals associated to a. Then we obtain the following identity ass(a) = ass(a1 , . . . , ak ) (v) () Let us introduce iso(a) := {a : (R \ p) | p ass(a) } the set of isolated components of a. then we obtain the following identity iso(a) = iso(a1 , . . . , ak ) (2.83) Remark: () Let (R, +, ) be a commutative ring. In section 6.1 we will introduce the notion of a prime ideal associated to an R-module M . Thus if a i R is an ideal, then it makes sense to the regard the prime ideals associated to a as an R-module. And by denition this is just sassR (a) = spec R { ann(b) | b a } = spec R { 0 : b | b a } (where the equalty is due to ann(b) = 0 : b for any b R). Thus we nd that sassR (a) has absolutely nothing to do with the set ass(a) that has been dened in the proposition above sass(a) = spec R
a:u

uR

However it will turn out that the set ass(a) is closely related to the set of associated prime ideals of the R-module R/a. By denition this is sassR R a xR a = spec R { ann(b + a) | b R } = spec R ann(x) = spec R { a : b | b R } (the latter equality is due to ann(b + a) = a : b for any b R again). Thus the dierence between this set and ass(a) lies in the fact that ass(a) allows the taking of radicals. Hence it is easy to see that sassR (R/a) ass(a). In section 6.1 we will prove that for noetherian rings R we even have sassR R a = ass(a)

2.7 Lasker-Noether Decomposition

129

(2.84) Corollary: (viz. 446) Lasker-Noether Let (R, +, ) be a noetherian ring, then any proper ideal a i R, a = R admits a minimal primary decomposition. And in this decomposition the associated prime ideals and the isolated components are uniquely determined by a. Formally that is (1) There are ideals ai i R (where i 1 . . . k and 1 k N) such that (a1 , . . . , ak ) is a minimal primary decomposition of a. (2) If (a1 , . . . , ak ) and (b1 , . . . , bl ) are any two minimal primary decompositions of a then we obtain the following three identities k = l ass(a1 , . . . , ak ) = ass(b1 , . . . , bl ) iso(a1 , . . . , ak ) = iso(b1 , . . . , bl ) (3) In particular if a = a is a radical ideal, then the minimal primary decomposition (p1 , . . . , pk ) of a is uniqeuly determined (up to ordering) and consists precisely of the prime ideals minimal over a, formally { p1 , . . . , pk } = { p spec R | a p } (2.85) Example: Fix any eld (E, +, ) and let R := E [s, t] be the polynomial ring over E in s and t (note that R is noetherian due to Hilberts basis theorem). Now consider the ideal a := s2 , st i . Then we obtain two distinct primary decompositions (a1 , a2 ) and (b1 , b2 ) of a by letting i=1 i=2
ai s i s2 , st, t2 bi s i s2 , t pi s i s, t i

where pi := ai = bi . It is no coincidence that both primary decompositions have 2 elements and that the radicals pi are equal. Further p1 p2 , that is a1 = b1 are isolated components of a and a2 = b2 are embedded components (of the decompositions). In fact this example demonstrates that the embedded components may be dierent. ass(a) = { sR, sR + tR } iso(a) = { sR }

(2.86) Corollary: (viz. 448) Let (R, +, ) be a noetherian ring and a be a proper ideal of R, that is R = a i R. If now q i R is any prime ideal of R then we obtain (i) a is contained in q i q contains an associated prime ideal of a i q contains an isolated component of a. Formally that is
a q

p ass(a) : p q i iso(a) : i q 2 Commutative Rings

130

(ii) The set of primes lying minimally over a is precisely the set of primes belonging to the isolated components of a. Formally again { q spec R | a q } = ass(a) (iii) And thereby the radical of a is precisely the intersection of all (minimal) associated prime ideals of a. Put formally that is
a =

ass(a) =

ass(a)

2.7 Lasker-Noether Decomposition

131

2.8

Finite Rings

(2.87) Denition: (viz. 448) Let (R, +, ) be any ring, denote its zero element by 0R and its unit element by 1R . For any 1 k N let us denote kR := k 1R = 1R + +1R (k -times). Then there exists a uniquely determined homomrphism of rings from Z to R, and this is given to be kR 0R (k )R if k > 0 if k = 0 if k < 0

: ZR : k

The image of in R is said to be the prime ring of R and denoted by prr R. In fact it precisely is the intersection of all subrings of R. Formally prr R := im( ) = { P | P r R }

And the kernel of is an ideal in Z. Hence (as Z is a PID) there is a uniquely determined n N such that kn( ) = nZ. This number n is said to be the characteristic of R, denoted by charR := n where kn( ) = nZ That is if R has characteristic zero n = 0, then we get kR = 0 = k = 0. And if R has non-zero characteristic n = 0, then n is precisely the smallest number such that nR = 0. Formally that is charR = min{ 1 k N | kR = 0 }

(2.88) Remark: If (F, +, ) even is a eld, we may also introduce the prime eld of F (denoted by prf F ) to be thie intersection of all subelds of F . Formally prf F := { E | E f F }

And if is easy to see that prf F is precisely the quotient eld of the prime ring prr F . Formally that is the following identity prf F = ab1 | a, b prr F, b = 0

Prob as E f F implies E r F (by denition) we have prr F prf F . Let us now denote the quotioent eld of prr F by Q F . As prf F is a eld and prr prf F it is clear that Q prf F . On the other hand Q f F is a subeld and hence we have prf F Q by construction. Together this means prf F = Q as claimed.

132

2 Commutative Rings

(2.89) Proposition: (viz. 449) Let (R, +, ) be any ring, then the characteristic of R satises the properties (i) If R = 0 is a non-zero integral domain, then the characteristic of R is prime or zero. Formally that is the following implication R = 0 integral domain = charR = 0 or prime

(ii) R has characteristic zero if and only if the prime ring of R is isomorphic to Z. More precisely we nd the following equivalencies charR = 0 k N : kR = 0 = k = 0 Z =r prr R : k (k )

(iii) R has non-zero characteristic if and only if the prime ring of R is isomorphic to Zn (where n = charR). More precisely for any n N we nd the equivalency of the following statements charR = n = 0 n = min{ 1 k N | kR = 0 } r prr R : k + nZ (k ) n = 0 and Zn =

(iv) Now suppose that R is a eld, then the prime eld of R is given to be charR = 0 charR = n = 0 Q =r prf R Zn =r prf R

(v) Let (R, +, ) be a nite ring (that is #R < ), then the characteristic of R is non-zero and divides the number of elements of R, formally 0 = charR | #R (2.90) Proposition: (viz. 450) (i) Let (R, +, ) be a nite (that is #R < ), commutative ring. Then the non-zero divisors of R already are invertible, that is R = R \ zdR (ii) Let (S, +, ) be an integral domain and consider a prime element p S . Let further 1 n N and denote the quotient ring R := S/pn S . Then the set of zero-divisors of R is precisely the ideal generated by the residue class of p. Formally that is zdR = (p + pn S ) R

2.8 Finite Rings

133

(iii) Now let (S, +, ) be a PID, consider a prime element p S and let R := S/pn S again. Then we obtain the following statements nilR = zdR = R \ R = (p + pn S ) R spec R = smax R = { nilR } (iv) Let (R, +, ) be any noetherian (in particular commutative) ring and x any 1 m N. Then there only are nitely many prime ideals p of R such that R/p has at most m elements. Formally #
p spec R

#R p m

<

(2.91) Remark: () As in (iii) let (S, +, ) be a PID, p S be a prime element and dene R := S/pn S again. Then we have seen, that nil(R) = R \ R and we will soon call a ring with such a property a local ring. To be precise: R is a local ring with maximal ideal nil(R). Further - in a fancier language - the property spec R = smax R can be reformulated as R is a zero-dimensional ring. (2.92) Theorem: (viz. 453) of Wedderburn Let (F, +, ) be a ring with nitely many elements only (formally #F < ), then the following three statements are equivalent (a) F (b) F (c) F is a eld is a skew-eld is an integral domain

134

2 Commutative Rings

2.9

Localisation

Localisation is one of the most powerful and fundamental tools in commutative algebra - and luckily one of the most simple ones. The general idea of localisation is to start with a commutative ring (R, +, ), to designate a sucient subset U R and to enlarge R to the ring U 1 R in which the elements of u U R U 1 R now are units. And it will turn out, that in doing this the ideal structure of U 1 R will be easier than that of R. And conversely we can restore information of R by regarding suciently many localisations of R (this will be the local-global-principle ). These two properties gives the method of localizing its punch. In fact commutative algebra has been most successful with problems that allowed localisation. So let us nally begin with the fundamental denitions: (2.93) Denition: Let (R, +, ) be any commutative ring, then a subset U R is said to be multiplicatively closed i it satises the following two properties 1 u, v U = U uv U

And U R is said to be a saturated mutilplicative system, i we even get 1 u, v U uv U = = U uv U uU

(2.94) Proposition: (viz. 455) (i) In a commutative ring (R, +, ) the set R of units and the set nzdR of non-zero-divisors of R both are saturated multiplicatively closed sets. (ii) If U R is a saturated, multiplicatively closed set, then R U , as for any u R we get uu1 = 1 U and hence u U . (iii) If U , V R are multiplicatively closed sets in the commutative ring R, then U V := { uv | u U, v V } is multiplicatively closed, too. (iv) If a i R is an ideal, then 1+ a R is a multiplicatively closed subset. (v) Let : R S be a ring-homomorphism between the commutative rings R and S and consider U R and V V . Then we get U V V mult. closed mult. closed saturated = = = (U ) 1 (V ) 1 (V ) mult. closed mult. closed saturated

(vi) If U P (R) is a chain (under ) of multiplicatively closed subsets of R, then the union U R is a multiplicatively closed set again. 2.9 Localisation 135

(vii) If U P (R) is a non-empty (U = ) family of [saturated] multiplicatively closed subsets of R, then the intersection U R is a [saturated] multiplicatively closed set, too. Nota this allows to dene the multiplicative closure of a subset A of R to be { U R | A U, U multiplicatively closed }. Likewise we may dene the saturated multiplicative closure of A to be the intersetion of all saturated multiplicatively closed subsets containing A. (viii) Consider any u R, then the set U := uk | k N clearly is a multiplicatively closed set. In fact it is the smallest multiplicatively closed set containing u. That is U is the multiplicative closure of { u }. (ix) Let U R be a multiplicatively closed subset of R, then the saturated multiplicative closure U of U can be given explictly to be the set U = { a R | b R, u, v U : uab = uv }

(2.95) Proposition: (viz. 456) Let (R, +, ) be any commutative ring and U R be a subset. Then U is saturated multiplicatively closed, i it is the complement of a union of prime ideals of R. Formally that is the equivalence of (a) P spec R such that U = R \ P

(b) U R is a saturated multiplicatively closed subset (2.96) Denition: (viz. 457) Let (R, +, ) be any commutative ring and U R be a multiplicatively closed subset of R. Then we obtain an equivalence relation on the set U R by virtue of (with a, b R and u, v U ) (u, a) (v, b) : w U : vwa = uwb

And we denote the quotient of U R modulo by U 1 R and the equivalence class of (u, a) is denoted by a/u, formally this is a := { (v, b) U R | (u, a) (v, b) } u U 1 R := U R Now U 1 R is a commutative ring under the following algebraic operations a + u a u b v b v := := av + bu uv ab uv

It is clear that under these operations the zero-element of U 1 R is given to be 0/1 and the unit element of U 1 R is given to be 1/1. And we obtain a canonical ring homomorphism from R to U 1 R by letting : R U 1 R : a 136 a 1 2 Commutative Rings

In general this canonical homomorphism need not be and embedding (i.e. injective). Yet we do obtain the following equivalencies injective bijective U nzdR U R

(2.97) Remark: Though the equivalence relation in the above denition might look a little articial at rst sight nothing here is mythical or mere chance. To see this let us rst consider the case of an integral domain (R, +, ). Then we may divide by w and hence we get a b = u v va = ub

and this is just what we expect, if we multiply the left-hand equation by uv . To allow a common factor w is just the right way to deal with zero divisors in R. For convenience we repeat the dening property of the quotient a/u for general commutative rings (R, +, ) a b = u v w U : vwa = uwb

And from this equivalence it is immediately clear that fractions a/u in U 1 R can be reduced by common facors v of U . That is for any elements a R and u, v U we have the equality av a = uv u And this allows to express sums by going to a common denominator. That is consider a1 , . . . , an R and u1 , . . . , un U . Then we denote u := u1 . . . un U and ui := u/ui U (e.g. u1 = u2 . . . un ). Then we have just remarked that ai /ui = (ai ui )/u and thereby a1 an a1 u1 + + an un + + = u1 un u This equivalence also yields that 1/1 = 0/1 holds true in U 1 R if and only if the multiplicatively closed set contains zero 0 U . But as 0/1 is the zero-element and 1/1 is the unit element of U 1 R we thereby found another equivalence U 1 R = 0 0U

As we have intended, localizing turns the elements u U into units. More precisely if U R is multiplicatively closed, then we obtain U 1 R 2.9 Localisation

au v

a R , u, v U 137

Yet the equality need not hold true! As an example consider R = Z and U = 6k | k N = { 1, 6, 36, . . . }. Then 2/1 is invertible, with inverse (2/1)1 = 3/6, yet 2 is not of the form R U = 6k | k N . Prob consider any au/v such that a R and u, v U , then we get (au/v )(a1 v/u) = (aa1 uv )/(uv ) = (uv )/(uv ) = 1/1 and hence au/v is a unit, with inverse (au/v )1 = a1 v/u. If U R even is a saturated multiplicatively closed subset, then we even obtain a complete description of the units of U 1 R to be U 1 R

u v

u, v U

Prob clearly u/v is a unit of U 1 R, since (u/v )1 = v/u. Conversely if b/v is a unit of U 1 R there is some a/u U 1 R such that ab/uv = (a/u)(b/v ) = 1/1. That is there is some w U such that abw = uvw. Hence b(aw) = uvw U and as U is saturated this implies b U . Let (R, +, ) be a commutative ring and U R be a multiplicatively closed set, such that U 1 R is nitely generated as an R-module. That is we suppose there are b1 , . . . , bn R and v1 , . . . .vn U such that U 1 R = a1 b1 an bn + + v1 vn a1 , . . . , an R

Then the canonical mapping : R U 1 R : b b/1 is surjective. Prob consider any a/u U 1 R and denote v := v1 . . . vn U and vi := v/vi V (e.g. v1 = v2 . . . vn ). As uv V and U 1 R is generated by the bi /v1 there are a1 , . . . , an R such that a a1 b1 an bn a1 b1 v1 + + an bn vn = + + = uv v1 vn v That is there is some w U such that vwa = uvwb where we let b := a1 b1 v1 + + an bn vn R. That is we have found (vw)1a = (vw)ub, thus by denition we have (b) = b/1 = a/u. And as a/u has been arbitrary this is the surjectivity of .

(2.98) Example: Let now (R, +, ) be any commutative ring. Then we present three examples that are of utmost importance. Hence the reader is asked to regard these examples as denitions of the objects Ru , quotR and Rp . If R is an integral domain and U R is a multiplicatively closed set, then the construction of U 1 R can be simplied somewhat. First let B := (R \ {0}) R, then we obtain an equivalency relation on B by letting (a, u) (b, v ) : av = bu. Let us denote the eqivalency class of (a, u) by a/u := [(a, u)]. Then we have regained the quotient eld, that has already been introduced in section 1.4 quotR = 138 a b a, b R, b = 0 2 Commutative Rings

Note that this is precisely the construction that is used to obtain Q from Z. If now U R is an arbitrary multiplicatively closed set satisfying 0 U , then the localisation of R in U is canonically isomorphic to the following subring of the quotient eld: U 1 R =r a quotR b a R, u U : a a u u

Prob we will rst prove the well-denedness and injectivity: by denition a/u = b/v in U 1 R is equivalent to w(av bu) = 0 for some w U . But as 0 U we get w = 0 and hence av = bu, as R is an integral domain. This again is just a/u = b/v in quotR. The surjectivity of the map is obvious and it clearly is a homomorphism, as the algebraic operations are literally the same.

Let us generalize the previous construction a little: it is clear that the set U := nzdR of non-zero divisors of R is a saturated multiplicatively closed set. Thus we may dene the total ring of fractions of a commutative ring R to be the localisation in this set quotR := (nzdR)1 R Note that in case of an integral domain we have nzdR = R \ {0} and hence we obtain precisely the same ring as before. Further note that by the very nature of U = nzdR the canonical homomorphism is injective. That is we can always regard R as a subring in quotR R quotR : a a/1 is injective And as nzdR is saturated we can give the units of quotR explicitly (quotR) = u v u, v nzdR

In particular we nd that quotR is a eld if and only if nzdR = R\{ 0 } and this is precisely the property of R being an integral domain. That is we obtain the equivalence R integral domain quotR eld

Consider some eld (F, +, ) and a subring R r F . In particular R = 0 is a non-zero integral domain and hence quotR is a eld. Let us now denote the subeld of F generated by R by E , explictly E := ab1 | a, b R

Then is is straightforward to check that the algebraic operations on the quotient eld quotR coincide with those of the eld E . That is we obtain a well-dened isomorphism of rings a quotR ab1 =r E : b 2.9 Localisation 139

 Consider any u R, then clearly U := 1, u, u2 , u3 , . . . tively closed. Hence we obtain a commutative ring by Ru := uk kN
1

is multiplica-

in which u = u/1 is invertible (with inverse 1/u). By the remarks above it is clear that Ru = 0 is zero if and only if u is a nilpotent of R. And R Ru is embedded under if and only if u is a non-zero divisor of R, respectively isomorphic if u is a unit. Altogether Ru = 0 : R Ru :R =r Ru u nilR u nzdR u R

In (2.9) and (in a more general version) in (2.95) we have seen, that the complement R \ p of a prime ideal p i R is multiplicatively closed. Thus for prime ideals p we may always dene the localisation Rp := (R \ p)1 R And by going to complements it is clear that R is embedded into Rp under , i p is contained in the set of zero divisors of R : R Rp
p zdR

In any case - as R \ p is saturated multiplicatively closed subset we can explicitly describe the group of units of Rp to be the following Rp

u v

u, v p

If (R, +, ) is an integral domain and p i R is a prime ideal of R then we get an easy alternative description of the localised ring Rp : let E := quotR be the quotient eld of R, then we denote: Ep := a E b bp

And thereby we obtain a canonical isomorphy from the localised ring Rp to Ep by virtue of a/b a/b. (One might hence be tempted to say Rp = Ep , but - from a set-theoretical point of view - this is not quite true, as the equivalency class a/b Rp only is a subset of a/b Ep ) a a Rp =r Ep : b b Note that under this isomorphy the maximal ideal mp of Rp induced by p (that is mp := pRp = (R \ p)1 p) corresponds to
mp

a E b

a p, b p

140

2 Commutative Rings

(2.99) Lemma: (viz. 458) Local-Global-Principle Let (R, +, ) be any commutative ring and a, b R, then equivalent are a=b a b = Rp 1 1 a b m smax R : = Rm 1 1 p spec R :

Further if R is an integral domain and F := quotR is its quotient eld, then we can embed R into E , as R = { a/1 F | a R }. Likewise if p i R is a prime ideal then we embed the localisation into F as well, as Rp := { a/u F | a R, u p }. And thereby we obtain the following identities (as subsets of F ) R =
pspecR

Rp =
msmaxR

Rm

(2.100) Proposition: (viz. 459) Universal Property Let (R, +, ) and (S, +, ) be commutative rings, U R be a multiplicatively closed subset and : R S be a homomorphism of rings, such that (U ) S . Then there is a uniquely determined homomorphism of rings : U 1 R S such that = . And this is given to be : U 1 R S : a (a)(u)1 u

(2.101) Remark: Let (R, +, ) be a commutative ring and U , V R be multiplicatively closed sets, such that U V . Further consider the canonical homomorphism : R V 1 R : a a/1. Then it is clear that (U ) (V 1 R) , as the inverse of u/1 is given to be 1/u. Thus by the above proposition we obtain an induced homomorphism : U 1 R V 1 R : a a u u

Thereby (a/u) = 0/1 i there is some v V such that va = 0. And this is precisely the same property for (a) = 0/1. Hence the kernel of is just kn() = U 1 (kn()) = a u a kn(), u U

(2.102) Denition: (viz. 459) Let (R, +, ) be a commutative ring, U R be a multiplicatively closed set and denote the canonical homomorphism, by : R U 1 R : a a/1 again. If now a i R and u i U 1 R are ideals then we dene the transfered ideals u R i R and U 1 a i U 1 R to be the following
u R := 1 (u) =

aR a u

a u 1

U 1 a := 2.9 Localisation

(a)

a a, u U 141

(2.103) Example: Let (R, +, ) be any commutative ring, a R and U R be multiplicatively closed. Then an easy, straightforward computation shows U 1 aR = a 1 U R 1

(2.104) Denition: (viz. 460) Let (R, +, ) be a commutative ring, U R be a multiplicatively closed set and a i R be an ideal of R. Then we obtain another ideal a : U i R by
a : U := { b R | v U : vb a }

And clearly a : U i R thereby is an ideal of R containing a a : U and using the notation of (2.102) we nally obtain the equality U 1 a R = a : U

(2.105) Proposition: (viz. 460) Let (R, +, ) be a commutative ring and U R be a multiplicatively closed subset. Further let a, b i R be any two ideals of R, then we obtain U 1 a b U 1 a + b = = U 1 a U 1 b U 1 a + U 1 b

U 1 a b = U 1 a U 1 b U 1 a = U 1 a
a:U :U ab :U a:U

= a:U
a:U b:U = a:U

Conversely consider any two ideals u, w i U 1 R and abbreviate a := u R and b := w R i R. Then we likewise get the identities
uw R = ab u+w R = a+b :U uw R = ab : U uR = a

142

2 Commutative Rings

(2.106) Example: The equality (a + b) : U = (a : U ) + (b : U ) need not hold true. As an example consider R = Z, U := {1, 2, 4, 8, . . . }, a = 9Z and b = 15Z. Then it is clear that a : U = a and b : U = b such that (a : U ) + (b : U ) = 3Z (as 3 is the greatest common divisor of 9 and 15). Yet 13 2 = 26 = 9 + 15 a + b. Thus 13 (a + b) : U even though 13 3Z = (a : U ) + (b : U ). (2.107) Denition: (viz. 462) Let (R, +, ) be a commutative ring and U R be a multiplicatively closed set in R. Then we dene the following sets of ideals in R U 1 ideal R := { a ideal R | a = a : U } = U
1

{ a ideal R | u U, ua a = a a }

srad R := srad R U 1 ideal R = { p spec R | p U = } { m smax R | m U = }

U 1 spec R := spec R U 1 ideal R U

1

smax R := smax R U 1 ideal R =

(2.108) Proposition: (viz. 462) Let (R, +, ) be a commutative ring and U R be a multiplicatively closed set in R. Then we obtain an order-preserving one-to-one correspondence between the ideals of the localisation U 1 R and U 1 ideal R, formally ideal (U 1 R)
u

(U 1 ideal R)
uR a

This correspondence is order-preserving, that is for any two ideals a, b U 1 ideal R and u, w ideal U 1 R respectively, we get
a b u w

= =

U 1 a U 1 b
uR wR

And this correspondence correlates maximal, prime and radical ideals of U 1 R with U 1 smax R, U 1 spec R respectively with U 1 srad R ideal (U 1 R) srad (U spec (U smax (U
1

(U 1 ideal R) U 1 (srad R) U 1 (spec R) U 1 (smax R)

R) R) R)

2.9 Localisation

143

(2.109) Example: Thus the ideal structure of U 1 R is simpler than that of R itself. The most drastic example is the following Z has a multitude of ideals (namely all aZ where a N, which are countably many). Yet Q = quotZ if a eld and hence only has the trivial ideals 0 and Q itself. Let now (R, +, ) be any commutative ring and choose any u R, then the spectrum of Ru corresponds (under the mapping u u R of the proposition above) to the following subset of the spectrum of R spec (Ru ) { p spec (R) | u p } Prob this is clear as for prime ideals p of R we have the equivalence u p u, u2 , u3 , . . . p p u, u2 , u3 , . . . = such that the claim is immediate from (2.108). Let q spec R be a xed prime ideal of R, then (for any other prime ideal p of R) clearly p (R \ q) = is equivalent to p q. Using this in the above proposition (2.108) we nd the correspondence spec (Rq ) { p spec (R) | p q } Thus localizing in q simply preserves the prime ideals below q and deletes all the prime ideals beyond q. Note that this is just the opposite of the quotient R/q - here all the ideals below q are cut o. That is let P := { p spec R | p q or q p } be the set of all prime ideals of R comparable with q. Then the situation can be illustrated in the following way:

0 pq spec(Rq) q qRq

spec(R/q) qp

(2.110) Corollary: (viz. 466) Let (R, +, ) be a commutative ring and U R be a multiplicatively closed subset with 0 U . Further let us abbreviate by any one of the words integral domain, noetherian ring, artinian ring, UFD, PID, DKD or normal ring. Then we get the following implication R is a = U 1 R is a

144

2 Commutative Rings

(2.111) Corollary: (viz. 468) Let (R, +, ) be an integral domain and denote its quotient eld by E := quotR. Then the following three statements are equivalent (a) R is normal (b) p spec R : Rp is normal (c) m smax R : Rm is normal (2.112) Proposition: (viz. 463) Let (R, +, ) be any commutative ring, then we present a couple of useful isomorphies concerning localisations: x a multiplicatively closed set U R and some point u R then we obtain (i) The localisation of the polynomial ring R[t] is isomorphic to the polynomial ring over the localized ring, formally that is U 1 R[t] f u =r
=0

(U 1 R)[t]

f [ ] t u

(ii) The localized ring Ru is just a quotient of some polynomaial ring of R Ru a uk f (1/u) =r R[t] (ut 1)R[t]

atk + (ut 1)R[t] f + (ut 1)R[t]

(iii) Let (R, +, ) be an integral domain and 0 U R be a multiplicatively closed subset of R not containing 0. Then the quotient elds of R and U 1 R are isomorphic under a a/1 quotR =r quotU 1 R : b b/1 (iv) Let (R, +, ) be an integral domain and 0 = a, b R be two non-zero elements of R and let n N. Then we obtain the following isomorphy Ra =r Rab : x/ai xa(n+1)j bi (b/an )j (ab)i+j

b/an

(v) Consider U R multiplicatively closed and a i R any ideal of R such that U a = . Now denote U/a := { u + a | u U }, then we obtain the following isomorphy U 1 R U 1 a =r
a b+a v+a

b + U 1 a v 2.9 Localisation

145

(vi) In particular: let p i R be a prime ideal and a i R be any ideal of R such that a p. Then we denote ap := (R \ p)1 a and thereby obtain the following isomorphy Rp ap =r R
a

p/a

b b+a + ap u u+a

(vii) Let p i R be a prime ideal of R and denote p := (R \ p)1 p. If now a p is not contained in p then we obtain the isomorphy a r/1 Rp =r (Ra )p : u u/1

146

2 Commutative Rings

2.10

Local Rings

(2.113) Denition: (viz. 471) Let (R, +, ) be any commutative ring, then R is said to be a local ring, i it satises one of the following two equivalent properties (a) R has precisely one maximal ideal (i.e. # smax R = 1) (b) the set of non-units is an ideal in R (i.e. R \ R i R) Nota and in this case the uniquely determined maximal ideal of R is precisely the set R \ R = {a R | aR = R} of non-units of R. (2.114) Proposition: (viz. 471) Let (R, +, ) be a commutative ring and m i R be a proper ideal of R (that is m = R). Then the following four statements are equivalent (a) R is a local ring with maximal ideal m (b) a i R : a = R = a m (c) R \ m = R (d) R \ m R (2.115) Example: If (F, +, ) is a eld, then F \ {0} = F , in particular F already is a local ring having the maximal ideal {0}. Let (E, +, ) be a eld and consider the ring of formal power series R := E [ t] over E . Then E [ t] is a local ring having the maximal ideal
m := t E [ t] = { f E [ t] | f [0] = 0 }

To prove this it suces to check R \ m R (due to the above proposition (2.114)). But in fact if f E [ t] is any power series with f [0] = 0 then an elementary computation shows that we can iteratively compute the inverse of f , by f 1 [0] = f
1

1 f [0] 1 f [0]
k 1

[k ] =

f [k j ]f 1 [j ]
j =0

Let (R, +, ) be a commutative ring and p i R be any prime ideal of R. Then in (2.116) below we will prove that the localised ring Rp is a local ring having the maximal ideal
mp := (R \ p)1 p = { p/u | p p, u p }

2.10 Local Rings

147

 Let (S, +, ) be a PID, 1 n N and p S be a prime element of S . Then we have proved in (2.90.(iii)) that R := S/pn S is a local ring having the maximal ideal nilR = zdR = R \ R = (p + pn S ) R (2.116) Proposition: (viz. 471) Let (R, +, ) be a commutative ring and p i R be a prime ideal. Then Rp is a local ring (i.e. a ring with precisely one maximal ideal). Thereby the maximal ideal of Rp is given to be
mp := (R \ p)1 p = { p/u | p p, u p }

And the residue eld of Rp is isomorphic to the quotient eld of R/p via Rp
mp

=f

a+p a + mp quotR p : u u+p

(2.117) Proposition: (viz. 487) () Let (R, +, ) be a local, noetherian ring with maximal ideal m and denote the residue eld of R by E := R/m. We now regard m as an R-module, then m/m2 becomes an E -vector space under the following scalar multiplication: (a + m)(m + m2 ) := (am) + m2 And if we now denote by rankR (m) the minimal number k N such that there are elements m1 , . . . , mk m with m = Rm1 + + Rmk , then we get dimE m m2 = rankR (m)

(2.118) Denition: Let (R, +, ) be any non-zero (i.e. R = 0), commutative ring, then a mapping : R N { } is called a valuation (and the ordered pair (R, ) a valued ring) if for any a, b R we obtain the following properties (1) (a) = a = 0 (2) (ab) = (a) + (b) (3) (a + b) min{ (a), (b) } And a valuation is said to be normed i it further satises the property (N) a S : (a) = 1

148

2 Commutative Rings

(2.119) Proposition: (viz. 472) (i) Let (R, ) be a valued ring (in particular R = 0), then R and its valuatuion already satisfy the following additional properties (U) R = () = 0 (I) R is an integral domain (P) [ ] := {a R | (a) 1} spec R (ii) If (R, +, ) is a non-zero integral domain then R always carries the trivial valuation : R N { } dened by (a) := 0 i a = 0 and (a) := i a = 0. Note that this valuation is not normed. (iii) Let (R, +, ) be a noetherian integral domain and consider any p R prime. Then we obtain a normed valuation = p on R by : R N { } : a a[p] where a[p] := sup{ k N | pk | a } In particular we have (p) = 1 and [ ] = pR. And for any a, b R this valuation satises a fth property, namely we get (a) = (b) = (a + b) = min{ (a), (b) }

(iv) Let (R, +, ) be a PID then we obtain a one-to-one correspondence between the maximal spectrum of R and the normed valuations on R: smax R
m

{ : R N { } | (1), (2), (3), (N ) } p where m = pR

[ ]

(2.120) Denition: Let (R, +, ) be a commutative ring, then we call : R N {} a discrete valuation on R (and the ordered pair (R, ) is said to be a discrete valuation ring), i satises all of the following properties ( a, b R) (1) (a) = a = 0 (2) (ab) = (a) + (b) (3) (a + b) min{ (a), (b) } (4) (a) (b) = a | b (N) m R : (m) = 1

2.10 Local Rings

149

(2.121) Proposition: (viz. 473) Let (R, ) be a discrete valuation ring, then R and its valuation already satisfy all of the following additional properties (i) (R, ) is a valued ring, in particular R is a non-zero integral domain. (ii) R is a local ring with maximal ideal [ ] = {a R | (a) 1}. (iii) (R, ) is an Euclidean domain under : R \ {0} N : a (a). (iv) Fix any m R with (m) = 1, then for any 0 = a R there is a uniquely detemined unit R such that a = mk for k = (a). (v) According to (iii) and (iv) the set of ideal of R is precisely the following ideal R = mk R kN {0}

(vi) Let (F, +, ) be an arbitrary eld and : F Z {} be a function satisfying the following four properties (for any x, y F ) (1) (x) = x = 0 (2) (xy ) = (x) + (y ) (3) (x + y ) min{ (x), (y ) } (N) m F : (m) = 1 In this case is said to be a discrete valuation and further we obtain a subring R of F by letting R := {a F | (a) 0} r F . Also : R N {} is a discrete valuation on R. Finally F is the quotient eld of R, that is we get F = ab1 | a, b F, b = 0

(2.122) Example: We regard F = Q and any prime number p Z. Then we obtain a discrete valuation p : Q Z {} by (where a, b Z, b = 0) p a b := [a]p [b]p

[a]p := sup{ k N | pk | a } Thereby p is said to be the p-adic valuation of Q. And if we denote p := pZ, then the discrete valuation ring detemined by p is just Zp . More generally the above construction works out for any PID R: let F be the quotient eld of R and p R be any prime element. Then we may repeat the denition of p literally (except for a, b R) and thereby obtain a discrete valuation p : F Z {}. And the discrete valuation ring determined by p is Rp (where p := pR) again. 150 2 Commutative Rings

 The prototype of a discrete valuation ring is the ring of formal power series E [ t] over a eld E . And in this case we obtain a discrete valuation (note that = tE [ t] ), by : E [ t] Z { } : f sup{ k N | tk | f } (2.123) Theorem: (viz. 474) A commutative ring (R, +, ) is said to be a discrete valuation ring (abbreviated by DVR), i it satises one of the following equivalent properties (a) R admits a discrete valuation : R N {} (b) R is an integral domain and there is some element 0 = m R satisfying R \ mR = R and
kN

mk R = 0

(c) R is a PID and local ring but no eld. (d) R is an UFD but no eld and any two prime elements of R are associates (i.e. if p, q R are irreducible, then we get pR = qR). (e) R is a noetherian integral domain and local ring, whose maximal ideal m is a non-zero principal ideal (i.e. m = mR for some 0 = m R). (f) R is a noetherian, normal and local ring with non-zero maximal ideal m = 0 and the only prime ideals of R are 0 and m (i.e. spec R = {0, m}). (2.124) Denition: A commutative ring (R, +, ) is said to be a valuation ring, i it is a nonzero integral domain in which the divides relation is a total order. Formally that is, i R satises the following three properties (1) R = 0 (2) R is an integral domain (3) a, b R we get a | b or b | a (2.125) Remark: The last property (3) in the denition above might seem a little articial. We hence wish to explain where it comes from: as R is an integral domain we may regard its quotient eld E := quotR. Then property (3) can be reformulated, as x E \ { 0 } : x R or x1 R Thus if we are given a valuation (in the sense of elds) : E Z {0} on E , we may regard the subring R := {a E | (a) 0}. And it is clear that R satises the property x R or x1 R for any 0 = x E . Thus every is assigned a valuation ring R in E . Therefore valuation rings appear naturally whenever we regard valuations on elds.

2.10 Local Rings

151

(2.126) Proposition: (viz. 476) (i) If (R, +, ) is a valuation ring, then R already is a local ring and Bezout domain (that is any nitely generated ideal already is principal). (ii) If (R, +, ) is a valuation ring, then we get the following equivalencies (a) R is noetherian (b) R is a PID (c) R is a DVR or eld

152

2 Commutative Rings

2.11

Dedekind Domains

(2.127) Denition: Let (R, +, ) be an integral domain and let us denote its quotient eld by F := quotR. Then a subset f F is called a fraction ideal of R (which we abbreviate by writing f f R) i it satises (1) f m F is an R-submodule of F (2) 0 = r R such that rf R (2.128) Remark: In particular any ideal a i R is a fraction ideal a f R (Prob just let r = 1). By abuse of nomenclature ordinary ideals of R are hence also said to be the integral ideals of F . Conversely for any fraction ideal f f R we see that f R i R is an ideal of R. Prob because 0 f R and if f , g f and a R then clearly f + g f R and af f R as both, f and R are R-modules. And clearly for any x F the generated module f := xR is a fraction ideal f f R of R Prob because if x = r/s then s f = rR R. If R itself is a eld, then F = R and the only possible fraction ideals of R are 0 and R (Prob if 0 = x f f R then x has an inverse in R and hence 1 = x1 x f. Thus for any a R we have a = a1 f). (2.129) Proposition: (viz. 477) Let (R, +, ) be an integral domain and let us denote its quotient eld by F := quotR. As in the case of integral ideals, if f and g f R are fraction ideals of R, then the sum f + g, intersection f g and product f g of nitely many fraction ideals is a fraction ideal again. And if g = 0 we may also dene the quotient f : g of two fraction ideals. I.e. if f and g f R are fraction ideals of R then we obtain further fraction ideals of R by letting
f + g := f g := f : g :=
n

x + y | x f, y g x | x f and x g x F | xg f fi gi n N, fi f, gi g

f g :=
i=1

(2.130) Remark: Let now 0 = f and g f R be two non-zero fraction ideals of the integral domain (R, +, ) and let F = quotR. Then the product f g and quotient f : g = 0 are non-zero, too. Formally that is
f, g = 0

f g = 0 and f : g = 0

2.11 Dedekind Domains

153

Prob as f and g = 0 we may choose some 0 = f f and 0 = g g. Then f g f g and f g = 0, as F is an integral domain (yielding f g = 0). Now let sg R for some s = 0. Then (sf )g = f (sg) f R f and hence sf f : g. But as sf = 0 this means f : g = 0. For any fraction ideal f f R we always get the following two inclusions
f ( R : f)

f:f

Prob f (R : f) contains elements of the form x1 y1 + + xn yn where xi R : f and yi f. But by denition of R : f this means xi yi R and hence x1 y1 + + xn yn R. And R f : f is clear, as f m F . First note that the multiplication f g of fraction ideals is both commutative and associative. Further it is clear that for any fraction ideal f f R we get R f = f. That is the set of fraction ideals of R is a commutative monoid under the multiplication (dened in (2.129)) with neutral element R. This naturally leads to the denition of an invertible fraction ideal. Of course the non-zero fraction ideals als form a monoid and this is called the class monoid of R, denoted by C (R) :=
f f R | f = 0

(2.131) Denition: (viz. 477) Let (R, +, ) be an integral domain and let us denote its quotient eld by F := quotR. Then a fraction ideal f f R of R is said to be invertible, i it satises one of the following two equivalent conditions (a) g f R : f g = R (b) f (R : f) = R Note that in this case the fraction ideal g with f g is uniquely determined, to be g = R : f. And it is called the inverse of f, written as
f
1

:= g = R : f

(2.132) Proposition: (viz. 477) Let (R, +, ) be an integral domain and let us denote its quotient eld by F := quotR. If now 0 = a i R is a non-zero ideal of R, then the following statements are equivalent (a) a is invertible (as a fraction ideal of R), i.e. there is some fraction ideal g f R of R such that a g = R. (b) There are some a1 , . . . an a and x1 , . . . , xn F such that for any i 1 . . . n we get xi a R and a1 x1 + + an xn = 1. (c) a is a projective R-module, i.e. there is a free R-module M and a submodule P m M such that a P = M .

154

2 Commutative Rings

(2.133) Corollary: (viz. 479) Let (R, +, ) be a local ring, that also is an integral domain. If now 0 = a i R is a non-zero ideal of R then the following statements are equivalent (a) a is a principal ideal (i.e. a = aR for some a R) (b) a is invertible (as a fraction ideal a f R) (2.134) Remark: If aR is any non-zero (a = 0) principal ideal in an integral domain (R, +, ), then aR f R is an invertible fraction ideal of R. This is clear as its inverse obviously given by
1

aR

1 R a

Now consider an (integral) ideal a i R and elements a1 , . . . , an a and x1 , . . . , xn R : a F such that a1 x1 + + an xn = 1. Then a is already generated by the ai , formally that is
a =

a1 , . . . , an

Prob the inclusion is clear, as ai a. For the converse inclusion we consider any a a. As xi R : a we in particlar have xi a R. And because of a1 x1 + + an xn = 1 we are able to write a in the form a = (x1 a)a1 + + (xn a)an a1 , . . . , an i . We have seen in the proposition above, that an (integral) ideal a i R is invertible if and only if there are a1 , . . . , an a and x1 , . . . , xn F such that for any i 1 . . . n we get xi a R and a1 x1 + + an xn = 1 Nota as we have just seen this already implies, that the ai generate a. Hence any invertible ideal already is nitely generated! Now consider some multiplicatively closed subset 0 U R. Then F = quotR = quotU 1 R are canonically isomorphic, by (2.112.(iii)). If now f and g f R are any two fraction ideals of R then it is easy to see that U 1 f := a/ub | a/b f, u U f U 1 R is a fraction ideal of U 1 R and a straightforward computation shows U 1 f U 1 g = U 1 (f g)

In particular if f f R is invertible (over R) then U 1 f is invertible (over U 1 R) and vice versa. Formally that is the equivalence
f f R invertible

U 1 f f U 1 R invertible

2.11 Dedekind Domains

155

Prob (U 1 f) (U 1 g) is generated by elements of the form (a/ub)(c/vd) where a/b f, c/d g and u, v U . Now uv U and (a/ub)(c/vd) = (ac)/(uvbd) such that this is contained in U 1 (f g) again. This proves conversely let a/b f, c/d g and u U . Then U 1 (f g) is generated by elements of the form (ac)/(ubd). And as (ac)/(ubd) = (a/ub)(c/d) this is also contained in (U 1 f) (U 1 g). This proves the converse inclusion and hence the equality. And from this equality 1 it is clear that (U 1 f)1 = U 1 (f ) for invertible fraction ideals (and hence the equivalence). (2.135) Denition: (viz. 479) Let (R, +, ) be an integral domain and let us denote its quotient eld by F := quotR. Then R is said to be a Dedekind domain (shortly DKD) i it satises one of the following equivalent statements (a) every non-zero fraction ideal 0 = f f R is invertible (i.e. there is some fraction ideal g f R such that f g = R). (b) every non-zero (integral) ideal 0 = a i R is invertible, (i.e. there is some fraction ideal g f R such that a g = R). (c) every ideal a i R with a {0, R} allows a decomposition into nitely many prime ideals, i.e. p1 , . . . , pn spec R such that a = p1 . . . pn . (d) R is a notherian, normal ring in which every non-zero prime ideal already is maximal, that is spec R = smax R { 0 }. (e) R is noetherian and for any non-zero prime ideal 0 = p spec R the localisation Rp is a discrete valuation ring (i.e. a local PID). (2.136) Example: In particular any PID (R, +, ) is a Dedekind domain - it is noetherian, as any ideal even has only one generator, it is normal (as it is an UFD by the fundamental theorem of arithmetic) and non-zero prime ideals are maximal (which can be seen using generators of the ideals). (2.137) Theorem: (viz. 485) Let (R, +, ) be a Dedekind domain and 0 = a i R be a non-zero ideal of R. Then R already satises an extensive list of further properties, namely (i) Any ideal 0 = a i R admits a decomposition into nitely many (up to permutations) uniquely determined prime ideals, formally ! p1 , . . . , pn spec R \ { 0 } such that a = p1 . . . pn

(ii) For any ideal 0 = a i R the quotient ring R/a is a principal ring (i.e. a ring in which any ideal is generated by a single element). (iii) Any ideal of R can be generated by two elements already, we even get: 0 = a a b a such that a = aR + bR (iv) If R also is semilocal (i.e. R only has nitely many maximal ideals) then R already is a PID (i.e. every ideal is a principal ideal). 156 2 Commutative Rings

(v) If m i R is a maximal ideal of R and 1 k N then we obtain the following isomorphy of rings (where as usual Rm = (R \ m)1 R and mk Rm = (R \ m)1 mk i Rm denote the localisation of R resp. mk ) R a k k Rm k mk =r m Rm : a + m 1 + m Rm

(2.138) Denition: (viz. 483) Let (R, +, ) be a Dedekind domain, m smax R be a maximal ideal and 0 = a i R be a non-zero ideal of R. Then we dene the valuation m induced by m on the set of ideals of R, by m : ideal R N : a max{ k N | a mk } : 00 And for any a R we denote m (a) := m (aR) N. Let now a, b i R be arbitrary ideals of R. The we say that b divides a, written as b | a, i it satises one of the following equivalent properties:
b | a

c i R : a = b c
a b

m smax R : m (b) m (a)

(2.139) Proposition: (viz. 483) Let (R, +, ) be a Dedekind domain with quotient eld F := quotR. And let m i R be any maximal (i.e. non-zero prime) ideal of R. Then we obtain the following statements: (i) Consider M = { m1 , . . . , mn } smax R a nite collection of maximal ideals and k (1), . . . , k (n) N. Then for any m smax R we obtain m m1
k(1) (n) . . . mk n

k (i) 0

if m = mi if m M

(ii) Consider any two non-zero ideals 0 = a, b i R then the multiplicity function m acts additively under multiplication of ideals, that is m (a b) = m (a) + m (b) (iii) Let 0 = a, b i R be any two non-zero ideals of R and decompose a = k(1) k(n) l(1) l(n) m1 . . . mn and b = m1 . . . mn with pairwise distinct maximal ideals mi smax R and k (i), l(i) N (note that this can always be established, as k (i) = 0 and l(i) = 0 is allowed). Then we get
a + b = m1
m(1) m(n) . . . mn where m(i) := min{ k (i), l(i) }

2.11 Dedekind Domains

157

(iv) Consider any non-zero ideal 0 = a i R, as before we decompose k(1) k(n) a into a = m1 . . . mn (with pairwise distinct mi smax R and 1 k (i) N). Then we obtain chinese remainder theorem :
n

a =r

R
i=1

mi

k(i)

: b + a b + mi

k(i)

(v) Let a i R be any ideal in R, then there is some ideal b i R such that a + b = R are coprime and a b is principal (i.e. a R : a b = aR).

158

2 Commutative Rings

Chapter 3

Modules
3.1 Dening Modules

(3.1) Denition: Let (R, +, ) be a ring, then the triple (M, +, ) is said to be an R-module, i M = is a non-empty set and + resp. are binary operations of the form + : M M M : (x, y ) x + y and : R M M : (a, x) a x = ax that satisfy the following properties (G) (M, +) is a commutative group, that is the addition is associative and commutative, admits a neutral element (called zero-element, denoted by 0) and every element of M has an additive inverse. Formally x, y, z M x, y M 0 M x M x M x M : x + ( y + z ) = (x + y ) + z : x+y = y+x : x+0 = x : x+x = 0

(M) The addition + and scalar multiplication on M satisfy the following compatibility conditions regarding the operations of R a R x, y M a, b R x M a, b R x M x M (3.2) Remark: The notations in the denition above deserve some special attention. First note that - as always with binary operations - we wrote x + y instead of +(x, y ). Likewise a x instead of (a, x) and of course a + b instead of +(a, b), a b instead of (a, b) (as we already did with rings). Recall that for a ring (R, +, ) the operation + has been called addition and has been called multiplication of R. In analogy, if (M, +, ) is an R-module, then + is said to be the addition of M and is called the scalar multiplication of M . 159 : a (x + y ) = (a x) + (a y ) : (a + b) x = (a x) + (b x) : (a b) x = a (b x) : 1 x = x

 Next note that we have used two dierent binary operations +, namely + : R R R of (R, +, ) and + : M M M of (M, +, ). At rst it may be misleading to denote two dierent functions by the same symbol, but you will soon become used to it and appreciate the simplicity of this notation. It will not lead to ambiguities, as the neighbouring elements (a and b for a + b, resp. x and y for x + y ) determine which of the two functions is to be applied. As an excercise it might be useful to rewrite the dening properties (M) using +R for + in R and +M for + in M . As an example let us reformulate the rst two properties in this way a R x, y M a, b R x M a (x +M y ) = (a x) +M (a y ) (a +R b) x = (a x) +M (b x)

Note that the properties (G) of R-modules truly express that (M, +) is a commutative group. In particular we employ the usual notational conventions for groups. E.g. we may omit the bracketing in sums due to the associativity of the addition + of M x + y + z := (x + y ) + z
n

xi :=
i=1

(x1 + x2 ) + . . . + xn

And as always the neutral element 0 M is uniquely determined again (if 0 and 0 are neutral elements, then 0 = 0 + 0 = 0 + 0 = 0). Likewise given x M there is a uniquely determined x M such that x + x = 0 and this is said to be the negative element of x, denoted by x := x (if p and q are negative elements of x then p = p + 0 = p + (x + q ) = q + (x + p) = q + 0 = q ). Let (R, +, ) be any ring (with unit element 1) and M be an R-module, then we wish to note some immediate consequences of the axioms above. Let x M and recall that x denotes the negative of x, then 0 x = 0 (1) x = x Prob as 0 = 0+0 in R we get 0 x = (0+0) x = (0 x)+(0 x) from (M). And substracting 0 x once from this equality we nd 0 = 0 x. This may now be used to see 0 = 0 x = (1 + (1)) x = (1 x) + ((1) x). Substracting x = 1 x we nd x = (1) x. We have already simplied the notations for modules rigorously. However it is still useful to introduce some further conventions. Recall that for rings (R, +, ) we wrote ab instead of a b. And we wrote a b instead of a + (b) and ab + c instead of (a b) + c. The same is true for the addition in the module M : we will write x y instead of x + (y ). Likewise we write ax instead of a x and declare that the scalar multiplication is of higher priority that the addition +. That is we write ax + y instead of (a x) + y . Note that it is unambiguously to write abx instead of (a b) x = a (b x), as this equality has been one 160 3 Modules

of the axioms for R-modules. Thus we may also omit the brackets in mixed terms with multiplication and scalar multiplication. In contrast to rings (where we faithfully wrote (R, +, ) on every rst occurance of the ring) we will omit the operations + and when refering to the R-module (M, +, ). That is we will say: let M be an R-module and think of (M, +, ) instead. What we dened here is also called a left R-module, this is to say that in the scalar multiplication : R M M the ring R acts from the left. One might be tempted to regard right R-modules as well, i.e. structures (M, +, ) in which the ring R acts from the right : M R M : (x, a) x a that satisfy the analogous properites (x + y ) a = (x a) + (y a), x (a + b) = (x a) + (x b), x (a b) = (x a) b and x 1 = x. However this would only yield an entirely analogous theory: Recall the construction of the opposite ring (R, +, )op in section 1.4. If (M, +, ) is a (left) R-module, then we obtain a right Rop -module (M, +, )op by taking M as a set and + as an addition again, but using a scalar multiplication that reverses the order of elements. That is we obtain a right Rop -module by (M, +, )op := (M, +, ) where x a := a x

The theory of R-modules is particularly simple in the case that R is a eld. As this case historically has been studied rst there is some special nomenclature: suppose that V is an F -module, where (F, +, ) is a eld. Then we will use the following notions set theory (F, +, ) (V, +, ) elements of F elements of V (3.3) Example: Consider any commutative group (G, +), then G becomes a Z-module under the following scalar multiplication : Z G G
k i=1 x

linear algebra base eld vector space scalars vectors

k x :=

0 k i=1 (x)

for k > 0 for k = 0 for k < 0

Let (R, +, ) be any ring and a m R be a left-ideal of R (note that in particular a = R is allowed). Then a is an R module under the operations inherited from R, that is + : a a a : (x, y ) x + y : R a a : (a, x) ax

3.1 Dening Modules

161

That is: although we started with quite a lot of axioms for R-modules, examples are abundant - the theory of modules encompasses the theory of rings! And the proofs are full of examples how module theory can be put to good use in ring theory (e.g. the lemma of Nakayama). Consider any (i.e. not necessarily commutative) ring (S, +, ) and suppose R r S is a subring of S . If now M is an S -module, then M becomes an R-module under the operations inherited from S + : M M M : RM M : (x, y ) x + y : (a, x) ax

Let (R, +, ) and (S, +, ) be any two (not necessarily commutative) rings and consider a ring-homomorphism : R S . If now M is an S -module, then M also becomes an R-module under the operations + : M M M : RM M : (x, y ) x + y : (a, x) (a)x

Consider an arbitrary (i.e. not necessarily commutative) ring (R, +, ). If now 1 n N then Rn becomes a well dened R-module under the pointwise operations + : Rn Rn Rn and : R Rn Rn (x1 , . . . , xn ) + (y1 , . . . , yn ) := (x1 + y1 , . . . , xn + yn ) a (x1 , . . . , xn ) := (a x1 , . . . , a xn ) Note that the operations + and on the right hand side are those of the ring R, whereas the operations + and are thereby dened on the module Rn . Further note that in the case n = 1 we get Rn = R and the addition + on R1 by denition coincides with the addition + on R. This is precisely the reason why we did not need to distinguish between +R and +M - in case of doubt it doesnt matter.

162

3 Modules

(3.4) Denition: Let (R, +, ) be a ring, then the quadrupel (A, +, , ) is said to be an Rsemi-algebra, i A = is a non-empty set and +, resp. are binary opertations of the form + : A A A, : A A A and : R A A that satisfy the following properties (R) (A, +, ) is a semi-ring, that is the addition + is associative and commutative, admits a neutral element (called zero-element, denoted by 0), every element of A has an additive inverse, the multiplication is associative and satises the distributivity laws. Formally f, g, h A : f + (g + h) = (f + g ) + h f, g A : f + g = g + f 0 A f A : f + 0 = f f A f A : f + f = 0 f, g, h A : f (g h) = (f g ) h f, g, h A : f (g + h) = (f g ) + (f h) f, g, h A : (f + g ) h = (f h) + (g h) (M) (A, +, ) is an R-module, that is the addition + and scalar multiplication of A satisfy the following compatibility conditions a R f, g A : a (f + g ) = (a f ) + (a g ) a, b R f A : (a + b) f = (a f ) + (b f ) a, b R f A : (a b) f = a (b f ) f A : 1 f = f (A) The multiplication and scalar multiplication of A are compatible in the following sense (note that this is some kind of associativity) a R f, g A : (a f ) g = a (f g ) = f (a g ) If now (A, +, , ) is an R-algebra, then we transfer most of the notions of ring theory to the case of algebras, by refering to the semi-ring (A, +, ). To be precise introduce the following notions (A, +, , ) is called R-algebra commutative integral division algebra noetherian artinian local semi-local : : : : : : : : (A, +, ) is a(n) ring commutative integral domain skew-eld commutative, noetherian ring commutative, artinian ring commutative, local ring commutative, semi-local ring

3.1 Dening Modules

163

(3.5) Remark: By denition any R-semi-algebra (A, +, , ) has a (uniquely determined) neutral element of addition, denoted by 0 (that is f + 0 = f for any f A). To be an R-algebra by denition means that the multiplication also admits a neutral element. That is there is some 1 A such that for any f A we get f 1 = f = 1 f . And due to the associativity of this unit element 1 is uniquely determined again. If (A, +, , ) is an R-algebra, then by denition (A, +, ) is a semi-ring and (A, +, ) is an R-module. In particular A inherits all the properties of these objects and we also pick up all notational conventions introduced for these objects. That is we write 0 for the zero-element and 1 for the unit element (supposed A is an R-algebra). Likewise we write f for the additive inverse of f (that is f + (f ) = 0 and f 1 for the multiplicative inverse of f (that is f f 1 = 1 = f 1 f ), supposed such an element exists. In case you are not familiar with these conventions we recommend having a look at section 1.3. If A is a commutative R-semialgebra, then condition (A) boils down to (a f ) g = a (f g ) for any a R and f , g A. Because if this property is satised then we already get (a f ) g = f (a g ) from the commutativity f (a g ) = (a g ) f = a (g f ) = a (f g ) = (a f ) g . If A is an R-algebra, i.e. an R-semialgebra containing a unit element 1, then due to property (A) the scalar multiplication can be expressed in terms of the ordinary multiplication of A with elements of the form a 1. To be precise, let a R and f A, then we clearly obtain (a 1) f = a f Note that property (A) in the denition of R-semi-algebras enables us to not only omit the multiplicative symbol and scalar multiplication but also the bracketing. Whenever we have a term of the form af g then we may use either bracketing (af )g = a(f g ) due to (A). () Let (R, +, ) be a commutative ring and A be an R-semi-algebra, then we can embed A into an R-algebra A canonically: as a set we take A := R A and dene the following operations on A (a, f ) + (b, g ) := (a + b, f + g ) (a, f ) (b, g ) := (ab, f g + ag + bf ) a (b, g ) := (ab, af ) Then (A , +, , ) truly becomes an R-algebra (the verication of this is left to the interested reader) with zero element (0, 0) and unit element (1, 0). And of course we get a canonical embedding (i.e. an injective homomorphism of R-algebras), by virtue of A A : f (0, f )

164

3 Modules

(3.6) Example: Let (R, +, ) be a commutative ring then, R becomes an R-module (even an R-algebra) canonically under its addition + as an addition of vectors and its multiplication as both, its multiplication and scalarmultiplication. I.e. if we let A := R, then R becomes an R-algebra (A, +, , ), under the operations + : A A A : (f, g ) f + g : A A A : (f, g ) f g : R A A : (a, f ) af In the following - whenever we say that we regard some ring (R, +, ) as an R-module - we will refer to this construction. And in this case any subset a R is an ideal of R if and only if it is an R-submodule of R (where R is regarded as an R-module, as we just did)
a i R

a m R

Let (A, +, ) be a commutative ring again and I = be any non-empty set, then we take A := F (I, R) to be the set of functions from I to R. Note that this includes the case A = Rn by taking I := 1 . . . n. Then A becomes an R-algebra (A, +, , ), under the pointwise operations + : A A A : (f, g ) (i f (i) + g (i)) : A A A : (f, g ) (i f (i) g (i)) : R A A : (a, f ) (i af (i)) Likewise let (R, +, ) be a commutative ring and a i R be an ideal of R. Then we may consider the quotient ring A := R/a as an R-algebra (A, +, , ), under the following operations + : A A A : (b + a, c + a) (b + c) + a : A A A : (b + a, c + a) (bc) + a : R A A : (a, b + a) (ab) + a This is a special case of the following general concept: consider any two commutative rings (R, +, ) and (A, +, ) and regard an arbitrary ring-homomorphism : R A. Then A can be considered to be an R-algebra under its own addition + and its own multiplication as multiplication and scalar multiplication with the imported elements (a). That is A becomes an R-algebra under the operations + : A A A : (f, g ) f + g : A A A : (f, g ) f g : R A A : (a, f ) (a)f

3.1 Dening Modules

165

 The most general case of the above is the following: let (R, +, ) be any (that is even a non-commutative) ring. Then R still is an R-module under the above perations. But R no longer needs to be R-algebra. Likewise consider a second ring (S, +, ) and a ring-homomorphism : R S . Then S is an R-module under the above operations. + : S S S : (x, y ) x + y : R S S : (a, x) (a)y Again S , need not be an R-algebra, in general. However S becomes an R-algebra (under its own multiplication), if the image of R under is contained in the center of S , that is if (R) cen(S ) := { x S | y S : xy = yx } An example of the above situation will be the ring of n n matrices over a commutative ring (R, +, ). Recall the construction of matrices given in section 1.4 (for more details confer to section ??) and let S := matn R. Then we obtain a homomorphism of rings by letting : RS : a 0 a .. . a 0

And as R is commutative it is clear, that (R) cen(S ) (in fact it even is true that (R) = cen(S )) and hence S becomes a well-dened R-algebra (under the usual multiplication and addition of matrices). (3.7) Denition: Let (R, +, ) be a ring and M be an R-module, then a subset P M is said to be an R-submodule of M , i it satises the following properties (1) 0 P (2) x, y P = x + y P (3) a R, x P = ax P And in this case we will write P m M . Likewise if A is a R-semi-algebra and P A is a subset, then P is said to be an R-sub-semi-algebra, i it is both, a submodule and sub-semi-ring of A, formally i it satises (M) P m A (4) f , g P = f g P And in this case we will write P b A. If A even is an R-algebra (with unit element 1), then a subset P A is said to be an R-subalgebra, i it is both, a submodule and subring of A, formally i it satises (B) P b A (5) 1 P 166 3 Modules

And in this case we will write P a A. Let now A be an R-semi-algebra again. Then a subset a A is said to be an R-algebra-ideal of A. i it is both, an ideal and a submodule of A, formally i it satises (M) a m A (I) f a, g A = f g a and gf a And in this case we write a a A. Having dened all these notions the set of all submodules of an R-module M is denoted by subm M . Likewise subb A denotes the set of all R-sub-semialgebras, aideal A denotes the set of all algebra-ideals of a semi-algebra A. And nally the set of all R-subalgebras of an R-algebra A id denoted by suba A. Formally we dene subm M := { P M | P m M }

subb A := { P A | P b A } suba A := { P A | P a A } aideal A := { a A | P b A }

(3.8) Remark: If P m M is a submodule (of the R-module M , where (R, +, ) is an arbitrary ring), then P g M already is a subgroup of (M, +), too. That is 0 P and for any x, y P we get x P and x + y P . Thereby x P follows from (3) by regarding a := 1, as in this case we get x = (1) x P . Likewise if A is an R-algebra (over an arbitrary ring (R, +, )), then we only need to check properties (1), (2), (4) and (5), as (3) already follows from (4). Given f R and a A we get af = (a1A ) f P . Let A be an R-semi-algebra over the ring (R, +, ), then any sub-semialgebra of A already is a a sub-semi-ring of A, formally P b A = P s A

Prob properties (1), (2) and (4) of sub-semi-algebras and sub-semirings coincide. It only remains to check property (3) of sub-semi-rings: consider any f P , then f = (1)f P by (3) of sub-semi-algebras. Let A be an R-semi-algebra over the ring (R, +, ), then a is an algebra ideal of A i it is an ideal and R-sub-semialgebra of A, formally
a a A

a b A and a i A

Prob the implication = is clear: (M) of algebra-ideals is identical to (M) of sub-semi-algebras and (I) of algebra ideals just combines (4) and (5) of ideals. Concersely we rst check property (4) of sub-semialgebras: if f , g a, then in particular g A and hence f g a by (4) of algebra-ideals. Now properties (1) and (2) of ideals are contained in (M) of algebra ideals ((4) and (5) are (I) again). Thus it only remains to check property (3) of ideals. But if f a then, as before f = (1)f a by property (3) of algebra-ideals. 3.1 Dening Modules 167

 If A even is an R-algebra over the ring (R, +, ), then there is no dierence between ideals and algebra-ideals of A, formally
a a A

a i A

Prob we have already seen a a A = a i A above. Further we have already argued that it only remains to check property (3) of algebra-ideals: thus consider any a R and f a, then a1 A (where 1 A, as A is an R-algebra) and hence af = (a1)f a by (I). (3.9) Proposition: (viz. 318) Fix any ring (R, +, ) and for the moment being let appreviate any one of the words R-module, R-semi-algebra or R-algebra and let M be a . Further consider an arbitrary family (where i I = ) Pi M of sub- s of M . Then the intersection of the Pi is a sub- again Pi
i I

M is a sub-

Likewise let A be an R-semi-algebra and let ai a A be an arbitrary (that is i I = again) family of R-algebra-ideals of A, then the intersection of the ai is an R-algebra-ideal of A again
ai
i I

A is an R-algebra-ideal

(3.10) Denition: Fix any ring (R, +, ) again, let appreviate any one of the words R-module, R-semi-algebra or R-algebra again and let M be a . If now X M is an arbitrary subset then we dene the generated by X to be the intersection of all sub- s containing X X X X
m b a

:= := :=

{ P M | X P m M } { P M | X P b M } { P M | X P a M }

Nota in case there is any doubt concerning the module X is contained in, (e.g. X M N ) we emphasise the R-module M used in this construction by writing X M m or X m m M or even M X m . Let us nally dene the linear hull lh(X ) resp. lhR (X ) of X to be the following subset of M . If X = , then we let lh() := { 0 } and if X = then
n

lh(X ) =
i=1

ai xi

1 n N, ai R, xi X

168

3 Modules

(3.11) Proposition: (viz. 319) Fix any ring (R, +, ) and consider an R-module M containing the non-empty subset = X M . Then the submodule generated by X is precisely the linear hull of X in M , formally
n

=
i=1

ai xi

1 n N, ai R, xi X

And if A is an R-semi-algebra, = X A, then we can also describe the R-sub-semi-algebra generated by X explictly. Namely we get m n X b = 1 m, n N, ai R, xi,j X ai xi,j
i=1 j =1

Finally if A is an R-algebra, then the sub-algebra generated by X becomes X

a

X {1}

(3.12) Proposition: (viz. 361) Let (R, +, ) be an arbitrary ring and N be an R-module. Further consider a conite submodule P m N , that is there are some x1 , . . . , xn N such that P + lh { x1 , . . . , xn } = N If further P = N , then there is a maximal submodule M of N containing P . That is M m N is a submodule satisfying the properties (1) M = N (2) P M N (3) Q m N we get M Q N = M = Q or M = N

3.1 Dening Modules

169

3.2

First Concepts

In this and the following section we will introduce three seperate concepts, that are all of fundamental importance. First we will introduce quotient modules - we have already studied quotient rings in (1.57), now we will generalize this concept to modules. Secondly we will turn our attention to torsion and annihilation. This is an important concept when it comes to analyzing the structure of a module, but will not occur in linear algebra. And nally we will introduce direct sums and products - the major tool for composing and decomposing modules over a common base ring. (3.13) Denition: (viz. 324) Let (R, +, ) be an arbitrary ring, consider an R-module M an R-submodule P m M of M , then P induces an equivalence relation on M , by xy : yxP

where x, y M . The equivalency classes under this relation are called cosets of P and for any x M this is given to be x + P := [x] = { x + p | p P } Finally we denote the quotient set of M modulo P (meaning the relation ) M M P := = {x + P | x M }

And thereby M/P can be turned into an R-module (M/P, +, ) again under (x + P ) + (y + P ) := (x + y ) + P a (x + P ) := (ax) + P Thereby (M/P, +, ) is also called the quotient module or residue module of M modulo P . Now consider an R-semi-algebra A and an R-algebraideal a a A. Then A/a (in the sense above) not only is an R-module, but even becomes an R-semi-algebra (A/a, +, , ), under the multiplication (f + a)(g + a) := (f g ) + a Thereby (A/a, +, , ) is also called the quotient algebra or residue algebra of A modulo a. And if A even is an R-algebra, then (A/a, +, , ) is an R-algebra again, its unit element being given to be 1 + a.

170

3 Modules

(3.14) Proposition: (viz. 327) Correspondence Theorem Let (R, +, ) be an arbitrary ring, M be an R-module and L m M be any R-submodule of M . Then we obtain the following correspondence (i) Consider some R-submodule P m M such that L P , then we obtain an R-submodule P/L of the quotient module M/L by virtue of P L := { p + L | p P } m M L

and thereby for any element x M we obtain the following equivalency x+LP L xP

(ii) Now the R-submodules of M/L correspond to the R-submodules of M containing L and this correspondence can be explictly given to be subm M L U P L { P subm M | L P } {x M | x + L U } P

(iii) Consider an arbitrary family Pi m M (where i I ) of R-submodules of M such that for any i I we get L Pi . Then the intersection commutes with taking to quotients, that is we obtain Pi
iI

i I

Pi

(iv) As in (iii) consider any Pi m M where i I and L Pi , then the summation also commutes with taking to quotients, that is Pi
iI

i I

Pi

(v) More generally consider any family of sumbodules Pi m M (where i I ) such that L i Pi , then we obtain the following identity Pi + L
i I

iI

Pi

(vi) Finally for any subset X M let us denote X/L := { x + L | x X } that is X/L M/L, then we obtain the following identity X L
m

+L

3.2 First Concepts

171

(3.15) Denition: Let (R, +, ) be any ring and M be an R-module. Further consider an element x M and an arbitrary subset X M , then we introduce (i) We dene the annihilator of x resp. of X to be the following subsets ann(x) := { a R | ax = 0 } ann(X ) := { a R | x X : ax = 0 } =
xX

ann(x)

If we want to stress the base ring in which the annulator is generated, we also write annR (X ) instead of ann(x). It is easy to see that any annihilator ann(x) is a left-ideal in R. Finally the set of all all leftideals of R occuring as annihilators of non-zero elements of M is called the annihilator spectrum of M sann M := { ann(x) | 0 = x M } (ii) For X { 0 } we let zdR (X ) := { 0 } and if there is some x X with x = 0 then we dene the set of zero-divisors of X to be zdR (X ) := { a R | 0 = x X : ax = 0 } =
0=xX

ann(x)

(iii) The complement of the zero-divisors is called the set of non-zerodivisors of X , written as nzdR (X ) := R \ zdR (X ). And if there is some x X with x = 0 then this is given to be nzdR (X ) = { a R | 0 = x X : ax = 0 } (iv) We thereby dene the torsion submodule of M to be the collection of elements of M that have an non-trivial annihilator, formally that is tor(M ) := { x M | 0 = a R : ax = 0 } = { x M | ann(x) = 0 }

Now M is said to be a torsion module i tor(M ) = M and M is said to be faithful or torsion-free i tor(M ) = { 0 }. (v) A submodule P m M is said to be a pure submodule of M , i it satises one of the following two equivalent conditions (a) tor(M/P ) = { 0 + P } (b) for any a R with a = 0 and any x M we get the implication ax P = xP

172

3 Modules

Prob for any x M we have x + P tor(M/P ) if and only if there is some a R with a = 0 such that a(x + P ) = 0 + P and this again is ax P . Hence M/P is torsion-free if and only if ax P implies x + P = 0 + P , which in turn is x P . (3.16) Proposition: (viz. 328) Let (R, +, ) be any ring, M be an R-module and X M be an arbitrary subset of M . Then we obtain the following statements (i) The annihilator of an arbitrary set X is a left-ideal of R, the annihilator of an R-module M even is an ideal of R, formally that is ann(X ) ann(M ) m i R R

(ii) () Let x M be any element of the R-module M and recall that the R-submodule of M generated by { x } is given to be Rx = { ax | a R }. Then we obtain the following isomorphy of R-modules R ann(x) =m Rx : b + ann(x) bx

(iii) If a i R is an ideal contained in the annihilator ideal a annR (M ) of M , then M can be turned into a (R/a)-module using the following scalar-multiplication (where a R and x M ) (a + a)x := ax (iv) Let R = 0 be a non-zero integral domain, then the torsion submodule tor(M ) truly is an R-submodule of M , formally that is tor(M ) m M

and the quotient module of M modulo tor(M ) is torsion-free, formally tor M tor(M ) = {0}

(v) If R = 0 is not the zero-ring, then the following statements are equivalent (recall that in this case M was said to be torsion-free ) (a) tor(M ) = { 0 } (b) zdR (M ) { 0 } (c) a R, x M we get ax = 0 = a = 0 or x = 0 (vi) Suppose R is a commutative ring, then the annihilator of X equals the annihilator of the R-submodule generated by X , formally that is ann(X ) = ann X 3.2 First Concepts
m

173

Consequently if { xi | i I } M is a set of generators of the Rmodule M = lh { xi | i I } then the annihilator of M is given to be ann(M ) =
iI

ann(xi )

() And the R-module homorphism : R M I dened by (a) := (axi ) has the kernel kn() = ann(M ). In particular we obtain an embedding (i.e. an injective R-module homorphism) R
I ann(M ) M : a + ann(M ) (axi )

(vii) Let (S, +, ) be a skew-eld and M be an S -module, if now x M is any element of M , then the annihilator of x = 0 is 0. In particular M Sx : a ax. is torsion free and if x = 0 then S ann(x) = S 0 if x = 0 if x = 0

(viii) Let (R, +, ) be a commutative ring, then maximal, proper annihilator ideals of (the R-module) M are prime ideals of R, formally that is sann M \ { R }

spec R

(ix) () If (R, +, ) is a commutative ring and M is a simple, non-zero R-module, then the annihilator of M is a maximal ideal of R, formally ann(M ) smax R (x) If M = 0 is a nitely generated, non-zero, torsion-free tor(M ) = 0 R-module, then M contains a maximal pure submodule, formally { P m M | P = M, P is pure } = (3.17) Proposition: (viz. 331) Modular Rule Let (R, +, ) be an arbitrary ring, M be an R module and P , Q and U m M be R-submodules of M such that P Q. Then we obtain QU +P = Q P +U P U = Q U, P + U = Q + U = P =Q

174

3 Modules

3.3

Direct Sums

(3.18) Denition: Let (R, +, ) be an arbitrary ring and consider an R-module M . If we are given nitely many submodules P1 , . . . , Pn m M of M , then we dene their sum to be set of all sums of elements of the Pi P1 + + Pn := { x1 + + xn | xi Pi } If we are given an arbitrary collection of R-submodules Pi m M , where i I is any index set, then we generalize the nite case directly by dening the sum of the Pi to be the set of all nite sums within the Pi Pi :=
i I

x1 + + xn

n N, { i1 , . . . , in } I, xk Pi(k)

By denition this set consists of nite sums of elements x1 + + xn where xk Pi(k) for some i(k ) I . Thus if we denote the collection of nite subsets of I by I := { I | # < } then we may present a slightly more formal description of this set Pi =
i I i

xi

I , x i Pi

(3.19) Remark: (i) Suppose we are given a collection of elements (xi ) M such that for any i I we have xi Pi . More importantly suppose that among these xi only nitely many are non-zero, that is the set := { i I | xi = 0 } is nite. As in (1.25) we then denote xi :=
iI i

xi

That is: if only nitely many of the xi are non-zero, then there is a well-dened notion of the sum of these elements - just leave out the zeros and sum up these (nitely many) non-zero elements. (ii) It is clear that P1 + +Pn and - more generally - i Pi are submodules of M . Prob Given any x = i xi and y = i yi i Pi their sum is x + y = i (xi + yi ) and as Pi is a submodule this means xi + yi Pi and hence x + y i Pi again. Likewiese for any a R we have ax = i (axi ) i Pi , as also axi Pi . (iii) In fact, in the light of proposition (3.11) this is just the R-submodule of M generated by the Pi , i.e. P1 + + Pn = P1 Pn m m M . More generally: given an arbitrary collection of R-submodules Pi m M , where i I is any index set, then the sum of the Pi is just the submodule generated by the union of all the Pi Pi =
i I i I

Pi

3.3 Direct Sums

175

(iv) In particular, if x M is any single element, then the submodule of M generated by x is { x } m = lhR (x) = Rx := { ax | a R }. Hence for any x1 , . . . , xn M the R-linear hull of these elements is just lhR ({ x1 , . . . , nn }) = Rx1 + + Rxn More generally, as both, the R-linear hull and the sum of submodules is based on nite sums only it is clear that for an arbitrary collection of elements (xi ) M the liear hull of the xi is just the the sum of the submodules generated by the xi , formally that is lhR ({ xi | i I }) =
i I

Rxi

(v) If we are given a chain (Pi ) of submodules, that is Pi m M is a submodule for any i I such that for any i, j I we have Pi Pj or Pj Pi , then even the union of the Pi itself (not only their sum) is a submodule of M Pi m M
i I

Prob consider a R and x, y P where we abbreviate P = i Pi . That is x Pi and y Pj for some i resp. j I . Without loss of generality we may assume Pi Pj and that is x Pj as well. Then as Pi and Pj m M are submodules again we clearly have ax Pi and x + y Pj such that ax and x + y P again. That is P truly is a submodule of M again.

(3.20) Example: Sadly the summation and intersection of modules do not t together nicely. This is to say: if we are given submodules P and Qi m M of an R-module M , then in general we only get P+
i I

Qi

i I

(P + Qi )

Yet the equality of these sets need not hold true. As an example we regard the ring R := Z and pick up the Z-module M := Z again. If now P := 2Z and for any i I = { i N | i 3, i prime } we let Qi := iZ, then it is clear, that P + Qi = M for any i I [as 2Z + iZ = gcd(2, i)Z = Z due to the lemma (2.68) of Bezout]. However if x Qi for any i I , then i | x, such that x would have innitely many prime divisors. This forces x = 0 and hence the intersection i Qi = 0. Therefore we only have P+
i I

Qi = 2 Z

Z =
iI

(P + Qi )

176

3 Modules

(3.21) Denition: (viz. 348) Let (R, +, ) be any ring, I = be an arbitrary index set and for any i I let Mi (more explictly (Mi , +, )) be an R-module. Then we regard the Carthesian product of the Mi M :=
iI

Mi

This can be turned into another R-module - called the (exterior) direct product of the Mi - by installing the following (pointwise) algebraic operatons (where x = (xi ), y = (yi ) M and i runs in i I ) + : M M M : RM M : (x, y ) (xi + yi ) : (a, x) (axi )

Note that xi + yi is the addition of vectors in Mi and axi is the scalar multiplication in Mi and hence these operations are well-dened. It is easy to see that M thereby inherits the properties of an R-module. We have just introduced the direct product M of the Mi as an Rmodule. Let us now dene a certain R-submodule of M , called the (exterior) direct sum of the Mi . To do this let us rst denote the support of x = (xi ) M to be supp(x) := { i I | xi = 0 Mi } I , i.e. the set of indices i with non-vanishing xi . Then we let Mi :=
iI

x
i I

Mi

#supp(x) <

That is i Mi consists of all the x = (xi ) i Mi that contain nitely many non-zero coecients xi = 0 only. In particular the direct sum coincides with the direct product if and only if I is nite Mi =
i I i I

Mi

#I <

And i Mi becomes an R-module again under the same algerbaic operations that we have introduced for the direct product i Mi . Formally that is: the direct sum is an R-submodule of the direct product Mi
iI

m
iI

Mi

() Let now N be another R-module and for any i I consider an Rmodule homomorphism i : Mi N . Then we obtain a well-dened R-module-homomorphism from the direct sum to N by virtue of i :
i I i I

Mi N : (xi )
i I

i (xi )

3.3 Direct Sums

177

This is well-dened, as only nitely many xi Mi are non-zero (as (xi ) is contained in the direct sum of the Mi ) and hence the sum only contains nitely many non-zero summands. And the properties of an R-module-homomorphism are trivially inherited from the i . Now consider any R-module M and for any i I an R-submodule Pi m M . And for any j I let us denote by Pj the R-submodule generated by all Pi except Pj , formally that is Pj :=
i= j

Pi =
iI \{j }

Pi

Then M is said to be the (inner) direct sum of the Pi i the Pi satisfy one of the following two equivalent statements concerning M : (a) The Pi build up M but for any index j I the intersection of Pj and its complement Pj is trivial, formally that is M=
i I

Pi and j I : Pj Pj = 0

(b) Any x M has a uniquely determined representation as a (nite) sum of elements of the Pi , formally that is x M ! (xi )
i I

Pi : x =
iI

xi

And in case that M is the inner direct sum ot the Pi we will write (beware the similarity to the exterior direct sum above) M =
i I

Pi

Let M be any R-module and P m M and R-submodule again. Then P is said to be complemented or a direct summand of M , i there is another submodule P of M , such that M is the inner direct sum of P and P . Formally that is: P m M such that (1) M = P + P (2) P P = 0

(3.22) Remark: (viz. 350) Recall that in Carthesian products (and the direct product has been dened as such) for any j I there is a canonical projection j of the product i Mi onto Mj . Thereby the projection j is dened, to be j :
iI

Mi Mj : (xi ) xj

178

3 Modules

But in contrast to arbitrary Carthesisan products we also nd (for any j I ) a canonical embedding j into the direct sum (and product), by letting j : Mj
i I

Mi : xj (i,j xj )

where i,j xj := xj for i = j and i,j xj := 0 Mi for i = j . Then it is clear that j j = 1 1 is the identity on Mj and in particular j is surjective and j is injective. Further these maps are tightly interwoven with the structure of the direct product and sum respectively. To be precise, consider any x = (xi ) i Mi then we get x = i (x)

On the other hand for any x = (xi ) i Mi we obtain another identity (note that the sum thereby is well-dened, as only nitely many xi are non-zero, such that in truth the sum is a nite only) x =
iI

i (xi )

() By denition of the operations on the direct product/sum it is also clear that both projection and embedding are homomorphisms of R-modules.

(3.23) Example: Consider any ring (R, +, ), the R-module M := R2 and the submodules P1 := R(1, 0) = { (a, 0) | a R } and P2 := R(0, 1) = { (0, b) | b R }. Then clearly M is the inner direct sum of P1 and P2 M = P1 P2 To check wether M can be decomposed as a direct sum of some Pi it does not suce to verify, that the intersection of the Pi is pairwise trivial. In the example above let P3 := R(1, 1) = { (a, a) | a R }. Then it is clear that M = P1 + P2 + P3 and Pi Pj = 0 for any i = j 1 . . . 3, but M is not the inner direct sum of P1 , P2 and P3 . In fact x = (1, 1) M has a non-unique representation as x = (1, 0)+(0, 1) P1 + P2 and x = (1, 1) P3 .

(3.24) Example: The most common example of a direct sum is a simple multiple of some module: to be precise, let (R, +, ) be some ring and consider an R-module M . Then for any non-empty set I we denote the product (resp. sum) of a copy of M for any i I by MI M
I

:=
i I

M = { (xi ) = x : I M : i xi } M =
iI

:=

(xi ) M I | { i I | xi = 0 } is nite

3.3 Direct Sums

179

The most important example is the case M = R, i.e. RI consists of I identical copies of R. By denition the operations on M I (and M I ) are component-wise, that is for any (xi ), (yi ) M I and any a R we have (xi ) + (yi ) := (xi + yi ) a(xi ) := (axi ) A special case of this is M n - here we take n identical copies of M , where 1 n N is any natural number. In the formal terms above this can be put as M I , where the set is I = 1 . . . n. As I is nite there is no dierence between the direct sum and product
n n

M n := M 1...n =
i=1

M =
i=1

(3.25) Remark: There is a reason why we didnt bother to strictly seperate exterior and inner direct products. The reason simply is: there is not much of a difference, any inner direct product already is an exterior direct product (up to isomorphy) and any exterior direct product can be reformulated as an interior direct product. We will give a precise formulation of this fact in the proposition below. Thus the distinction between inner and exterior is a purely set-theoretic one, it doesnt aect the algebra. This is why we do not bother that our formalism doesnt really distinguish between the interior and exterior direct sums - the dierence is no structural one. (In general group theory there will be a dierence). (3.26) Proposition: (viz. 350) (i) () Let (R, +, ) be any ring, M be an R-module and Pi m M be an arbitrary (i I ) collection of R-submodules of M . If now M is the inner direct sum M = i Pi of the Pi , then M is already isomorphic (as an R-module) to the exterior direct sum of the Pi under Pi =m M : (xi )
iI i I

xi

(ii) Let (R, +, ) be any ring and Mi an arbitrary (i I ) collection of Rmodules. Now let Pi := i (Mi ) m i Mi , then the exterior direct sum of the Mi is just the interior direct sum of the Pi put formally Mi =
iI iI

Pi

180

3 Modules

(3.27) Proposition: (viz. 350) () Let (R, +, ) be an arbitrary ring and let I and J be arbitrary index sets. For any i I resp. j J let Mi and Nj be R-modules. Then we obtain the following isomorphy of R-modules mhom
iI

Mi ,
j J

Nj

=m
(i,j )I J

mhom(Mi , Nj )

We can even give the above isomorphism explictly: let us denote the direct sum of the Mi by M := i Mi and the direct product of the Nj by N := j Nj . Further denote the canonical embedding i : Mi M and the canonical projection j : N Nj . Then the isomorphy explictly reads as mhom(M, N ) i,j
iI

=m
(i,j )I J

mhom(Mi , Nj ) j i i,j

Let us explain this compact notation a bit: for any (I J )-tupel of homomorphisms i,j : Mi Nj the corresponding homomorphism := ( i i,j ) is explictly given to be the following: : M N : (xi ) ( i i,j (xi )). (3.28) Remark: Suppose we even have (i,j ) i j mhom(Mi , Nj ), that is for any i I the set { j J | i,j = 0 } is nite. Then the induced map even lies in ( i i,j ) mhom( i Mi , j Nj ). The converse need not be true however, as an example regard I = 0, J = N, M = RN and Nj = R. As a homomorphism we choose the identity = 1 1 : RN RN . Then induces the tuple 1 1 (j ), which has no nite support.

3.3 Direct Sums

181

3.4

Ideal-induced Modules

In this short section we will introduce the concept of the submodule aM generated by an ideal a of R. This is a way of passing on from a module M over a ring R to a related module over the quotient ring R/a. So as always with going to quotients, one tries to single out unnecessary complications, solve the problem in the reduced case and later lift it up again, solving the original problem. This is particularly useful in case the quotient ring R/m is a eld, i.e. m is a maximal ideal. While the construction is elementary it yields some very useful results - primarily the lemma of Nakayama. Though these results are of little value in their own right, they are powerful tools in providing proofs of more substantial theorems. Hence one might as well skip this section and postpone it, until one encounters the urge to pass on using a quotient construction. (3.29) Denition: Consider an arbitrary ring (R, +, ), an ideal a i R and an R-module M . Let further P m M be a submodule of M , then we denote the submodule of M generated by elements of the form ap, where a a and p P , by
aP

:=

{ ap | a a, p P }

(3.30) Proposition: (viz. 332) Let (R, +, ) be an arbitrary ring, a i R an ideal and consider a submodule P m M of the R-module M . Then we get M (i) The submodule aP of M can be given explictly to be the following set
n

aP

=
i=1

ai p i

1 n N, a i a , p i P

(ii) If P is generated by { pi | i I } P , then for any x aP there are some a1 , . . . , an a and p1 , . . . , pn { pi | i I } such that x = a1 p1 + + an pn (iii) It is clear that aP is a submodule of P and hence we obtain the following chain of inclusions
aP

(iv) The assignment P aP is order preserving, that is given any two submodules P and Q m M we obtain the implication P Q =
aP aQ

182

3 Modules

(v) Going to residues modulo P is compatible with inducing modules by ideals, that is given P m M and a i R we get the equality
a M P

= P + aM P

(vi) The quotient module M/aM becomes an (R/a)-module under the following operations: for any a R and x, y M we let (x + aM ) + (y + aM ) := (x + y ) + aM (a + a)(x + aM ) := (x + y ) + aM (vii) () If M and N are R-modules and : M N is an R-modulehomomorphism, then we obtain a well-dened (R/a)-module-homomorphism, by letting M N a : aM aN : x + aM (x) + aN

If is surjective or bijective, then so is the induced homomorphism /a. That is if abbreviates any of the words surjective or bijective then we obtain the implication is =
a is

Nota the injectivity of need not be preserved. E.g. consider R = Z and a = 3Z. Then Z Q : k k clearly is injective, but aQ = Q and hence the induced map Z/aZ Q/aQ cannot be injective. (viii) Thereby the linear hull of any subset Z M/aM of the quotient module does not depend on whether we take R or R/a as a base ring lhR (Z ) = lhR/a (Z ) (ix) Hence a subset Q M/aM of the quotient module is an R-submodule of M/aM if and only if it is an (R/a)-submodule of M/aM . Therefore the (R/a)-submodules of M/aM are given to be submR/a M aM = P
aM

P m M, aM P

(3.31) Example: (viz. 371) Let (R, +, ) be a commutative ring and x any ideal a i R. Then for any non-empty set I , the induced submodule aRI is nothing but aI
aRI = aI

And thereby we obtain an isomorphy of the (R/a)-modules RI /aRI and (R/a)I which can be given explictly, to be the following R
I

I RI aRI : (bi + a) (bi ) + aR

3.4 Ideal-induced Modules

183

(3.32) Lemma: (viz. 397) () Let (R, +, ) be a commutative ring, a i R be an ideal and consider an R-module M . Then the following statements are true: (i) If M is a free R-module, with basis { mi | i I }, then the quotient module M/aM is a free (R/a)-module with basis { mi + aM | i I, mi aM } And thereby the elements of this basis are pairwise distinct. To be precise: if the mi are pairwise distinct (i.e. for any j , j I we get mi = mj = i = j ) then the mi + aM are pairwise distinct too, that is: for any i, j I with mi and mj aM we get mi + aM = mj + aM = i=j

(ii) Lemma of Nakayama: Given any ideal a of R the following two statements are equivalent: (a) a jacR (b) For any nitely generated R-module M we get the implication
aM = M

M =0

(iii) If P m M is a submodule of M , such that the quotient M/P is nitely generated, and a jacR, then we obtain the implication: M = P + aM = P =M

(iv) If N is another R-module, and : M N is an R-module-homomorphism, such that N/im is nitely generated. If again a jacR, then we get the following implication is surjective
a is surjective

(v) If M is nitely generated, : M M even is an R-module-endomorphism, and again a jacR, then we get the following implication is bijective
a is bijective

(vi) Lemma of Dedekind: If M is a nitely generated R-module such that aM = M , then there is some element a a of a satisfying (1 a)M = 0

184

3 Modules

(3.33) Proposition: (viz. 402) Let (R, +, ) be a commutative ring and consider an R-modlue M . Then we obtain all of the following statements (i) Suppose that M is nitely generated, that is there are some x1 , . . . , xn M such that M = lhR { x1 , . . . , xn }. Then we get the implication m : M = mM = M =0

(ii) Suppose P m M is a conite module, that is there are some x1 , . . . , xn M such that M = P + lhR { x1 , . . . , xn }. Then we get the implication m : M = P + mM = P =M

(iii) If N is another R-module and : M N is an R-module-homomorphism such that im m N is conite. Then we get the equivalency is surjective m smax R : m is surjective

(iv) If M is nitely generated and : M M is an R-module-endomorphism of M , then we get the following equivalency is bijective m smax R : m is bijective

3.4 Ideal-induced Modules

185

3.5

Block Decompositions

direkte Summen und Produkte von Algebren Blockzerlegung von Algebren vgl. mein Skriptum zur Darstellungstheorie n pro: wenn sumn i=1 ei = 1 mit und ei ej = i,j dann A = i=1 Aei pfr: n n existenz: a = a 1 = a i=1 Aei = aei Ae eindeutigkeit: wenn i i=1 n n n n i=1 ai ei = 0 dann 0 = ( i=1 ai ei ) ej = i=1 ai ei ej = i=1 ai i,j = aj exp: Rn mit Euclidean basis e1 , e2 , . . . , en matn (R) funktioniert nicht!

186

3 Modules

3.6

Dependence Relations

In algebra there are two seperate notions of dependence: linear dependence and algebraic dependence. A subset S of a module is linearly dependent, if we can nd a linear equation with elements of S that turns out 0. A subset S of an algebra is algebraically dependent, if we can nd a polynomial equation with elements of S that turns out 0. Linear dependence yields the notion of a basis of a module and we will see that the modules structure is determined by its basis. The same is true with algebraic dependence: this yields the notion of a transcendence basis, which determines the structure of the algebra. It is customary to approach these problems directly and hence seperately. Yet there is the underlying concept of dependence. So we will try a little more abstract approach: rst we will introduce the general notion of a dependence relation. Then we will prove (among others) that any dependence relation admits a basis. This will be accomplished in this section. Consequently it suces to show that linear dependence (over a skeweld) is a dependence relation, respectively that algebraic dependence (over a eld) is a dependence relation. This will be discussed in the sections to come. This approach features the advantage that it neatly seperates settheoretic and algebraic concepts, which yields beautiful clarity and pinpoints precisely where the (skew-)eld is involved. (3.34) Denition: Let M be an arbitrary set, in the following we will regard relations of the following form P (M ) M . And for any such relation, elements x M and subsets S , T M we will use the following notations S S S sub(M ) T S x x : : := := : (S, x) (S x) x} {x M | S S T x S : T x

{ S |S M}

Now is said to be a dependence relation on M , i it is of the form P (M ) M and for any elements x M and subsets S , T M it satises the following four properties: (D1) any element x of S already depends on S , formally this can be put as xS = S x

(D2) if x depends on S and S depends on T , then x already depends on T T S, S x = T x

(D3) if x depends on S , then it already depends on a nite subset Sx of S S x = Sx S : #Sx < and Sx x 187

3.6 Dependence Relations

(D4) if x depends on S but not on S \ { s } (for some s S ), then we can interchange the dependencies of x and s, formally again S x, s S, S \ { s } x = (S \ { s }) { x } s

And in case that is a dependence realtion on M we further introduce the following notions: a subset S M is said to be independent (or more precisely independent ), i it satises s S : S \ {s} s

Clearly S is said to be dependent (or more precisely dependent ), i it is not independent (that is S \ { s } s for some s S ). And a subset B M is said to be a basis (or more precisely a basis ) i it is independent and generates M . Formally B is a basis, i it satises (B1) B is independent (B2) B = M

(3.35) Lemma: (viz. 362) Let M be an arbitrary set, S , T M be arbitrary subsets and be a dependence relation on M . Then we obtain all of the following statements (i) The generation of subsets under S T = preserves the inclusion of sets, i.e. S T

(ii) Recall that we denoted sub(M ) to be the set of all T for some subset T M . Then for any subset S M we obtain the identities S = S = { P sub(M ) | S P }

(iii) For any subset S M the following three statements are equivalent (a) S is dependent (b) T M : S T = T is dependent (c) S0 S such that #S0 < and S0 is dependent (iv) For any subset T M the following three statements are equivalent (a) T is independent (b) S M : S T = S is independent (c) T0 T we get #T0 < = T0 is independent (v) For any S M and x M the following statements are equivalent (a) S { x } is independent and x S (b) S is independent and S 188 x 3 Modules

(vi) Let Si M be a chain (where i I of idependent subsets of M , then the union of the Si is independent again. Formally that is i I : Si is independent i, j I : Si Sj or Sj Si =
iI

Si is independent

(vii) Let S M be an independent subset and T M be a generating subset M = T . If now S T , then there is some B X such that (1) S B T and (2) B is a basis of M (viii) Let B M be a basis of X , b B and x M be arbitrary. Then we denote by B := (B \ { b }) { x } the set B with b replaced by x. If now B b, then B is a basis of M , as well. (ix) A subset B M is a basis of M if and only if it is a minimal generating subset of M . And this again is equivalent to being a maximal independent subset. Formally that is the equivalency of (a) B is a basis of M (b) B generates M (that is M = B ) and for any S M we obtain S B = M= S

(c) B is independent and for any S M we obtain the implication B S = S is not independent

(x) All bases of M have the same number of elements, formally that is A, B M bases of M = | A| = | B |

(xi) If S M is independent and B M is an basis of M , then S contains no more elements than B , that is |S | |B |. (xii) Consider a independent subset S M and a basis B M of M . Further assume that #B < is nite, then the following two statements are equivalent (a) |S | = |B | (b) S is an basis of M Nota that in the proof of this lemma we did not always require all the properties of a dependence relation. Namely (i), (ii), (iii), (iv), the implication (a) = (b) in (v), (vi) and the equivalence (a) (b) in (ix) require the relation to satisfy properties (D1), (D2) and (D3) only.

3.6 Dependence Relations

189

(3.36) Remark: In item (vi) we may chose S = , as this always is independent respectively T = M , as always M = M . These special choices yield the so called basis selection respectively basis completion theorems (the converse implications follow immediately from (i) and (ii) respectively). And as these are used frequently let us append them here. Thus suppose M is an arbitrary set and is a dependence relation on M again. Then we get Basis Selection Theorem : a subset T M contains a basis of M if and only if it generates M , that is equivalent are (a) T = M (b) B T : B is a basis of M Basis Completion Theorem : a subset S M can be extended to a basis of M if and only if S is independent, that is equivalent are (a) S is independent (b) B M : S B and B is a basis of M

190

3 Modules

3.7

Linear Dependence

The primary task of any algebraic theory is to classify the objects involved. The same is true for module theory: what types of modules do exist? Surprisingly the answer will strongly depend on the modules base ring. A rst complete answer can be given in the case that the base ring is a skew-eld S , then all modules over S are of the form S I for some index set I . (We will later also nd complete answers in case the base ring is a PID or DKD. But this is how far the theory carries). In order to achieve this result we will have to dene the notion of a basis of a module. Modules with basis will be said to be free. However most of the work is done already - all we have to do is prove that linear dependence truly is a dependence relation, as we have introduced it in the section before. (3.37) Denition: Let (R, +, ) be an arbitrary ring an M be R-module. Further consider an arbitrary subset X M . Then we introduce the following notions (i) A subset X M is said to generate M as an R-module (or simply to R-generate M ), i the R-module generated by X is M . Formally that is i (where we also recalled proposition (3.11))
n

M =

=
i=1

ai xi

1 n N, ai R, xi X

Now X is said to ba a minimal R-generating set, i X R-generates M but any proper subset of X does not R-generate M , formally i W M : W X = W
m

=M

(ii) We thereby dene the rank - denoted by rank(M ) - to be the minimal cardinality of an R-generating subset X of M . Formally that is rank(M ) := min{ |X | | X M, X
m

=M}

Nota that this is well dened, as the cardinal numbers are wellordered - that is any non-empty set of cardinal numbers has a minimal element. Also the dening set above is non-empty, as M M and M = M m . In particular we have rank(M ) |M |. Hence this denition yields the following properties: if I = rank(M ) then we get (1) there is some X = { xi | i I } M such that M = X (2) for any X M with M = X : I X from I into X .
m m,

and

there is some injective map

(iii) M is said to be nitely generated, i it is R-generated by some nite subset X M . That is i it satises one of the equivalent conditions (a) rank(M ) < (b) X M : #X < and M = X
m

3.7 Linear Dependence

191

(iv) A submodule P m M is said to be conite (in M ), i it satises one of the following two equivalent conditions (a) M/P is nitely generated (b) X M : #X < and M = P + X
m

Nota if X = { x1 , . . . , xn } such that M = P + X m , then it is clear that M/P is generated by the set { xi + P | i 1 . . . n }. Vice versa, if { xi + P | i 1 . . . n } generates M/P , then we get M = P + X m , by taking X = { x1 , . . . , xn }. Let us prove this: given any x M there are ai R such that x + P = a1 x1 + + an xn + P and hence x = p + a1 x1 + + an xn P + X m for some p P . (v) Let us now introduce a relation of the form P (M ) M called the relation of R-linear dependence. For any subset X M and any element x M let us dene X x : n N, x1 , . . . , xn X, a1 , . . . , an R such that x = a1 x1 + + an xn

Nota that we allowed n = 0 in this denition. And by convention the empty sum is set to be 0. Thus we get X 0 for any subset X M . In fact we even get x x = 0 for any x M . (vi) A subset X M is said to be R-linearly dependent i for any 1 n N, any x1 , . . . , xn X and any a1 , . . . , an R we get a1 x1 + + an xn = 0 i, j 1 . . . n : xi = xj = i = j = a1 = = an = 0

If this is not the case (i.e. there are some a1 , . . . , an R \ { 0 } and pairwise distinct x1 , . . . , xn X such that a1 x1 + + an xn = 0), then X is said to be R-linearly dependent. Finally X is said to be a maximal R-linearly independent set, i X is R-linearly dependent, but any proper superset of X is R-linearly dependent. Formally Y M : X Y = Y is R-linearly dependent (vii) A subset B M is said to be a R-basis of M , i it is R-linearly independent (that is (iv)) and R-generates M (that is (i)). And M is said to be free, i it has a basis. That is M free B M : B is a R-basis

(viii) Finally an ordered tupel (xi ) M (where i I ) is said to be an ordered R-basis of M , i it satises the following two properties (1) { xi | i I } is a R-basis of M (2) i, j I : xi = xj = i = j

192

3 Modules

(3.38) Example: (viz. 366) (i) A very special case in the above denitions is the zero-module { 0 }. It is a free module, as { 0 } is an R-basis. And thereby for an arbitrary R-module M we obtain rank(M ) = 0 M = {0}

(ii) From the above denitions it is clear, that M is nitely generated, if and only if the zero-submodule 0 m M is conite (in M ). Also, if P and Q m M are submodules of M such that P is conite and P Q, then Q is conite as well. (iii) Consider an integral domain (R, +, ) and an R-submodule (that is an ideal) a m R. Then a is a free R-module, if and only if it is principal
a free

a R : a = aR

(iv) The standard example of a free module is the following: let (R, +, ) be an arbitrary ring and I = be any non-empty set. Then we consider RI =
i I

R =

(xi ) RI

# { i I | xi = 0 } <

Recall that this is an R-module under the component-wise operations (xi ) + (yi ) := (xi + yi ) and a(xi ) := (axi ) - where a, ai and yi R. For any j I let us now denote the j -th Euclidean vector ej := i,j R I

That is the i-th component of ej is 1 if and only if i = j and 0 otherwise. Then RI is a free R-module having the Euclidean basis E = { ej | j I } () In particular - if (S, +, ) is a skew-eld - we nd that for any set I = the S -vecor-space S I is of dimension dim(S I ) = |I | (3.39) Proposition: (viz. 366) Let (R, +, ) be an arbitrary ring and M be an R-module. Let us denote the relation of R-linear dependence by again (refer to (3.37.(iv)). Then the following statements hold true (i) The relation satises the properties (D1), (D2) and (D3) of a dependence relation (refer to (3.34) for the explicit denitions). (ii) If X M is any subset and we denote X again, then we obtain the following identities X = X
m

:= { y M | X

y}

= lhR (X )

3.7 Linear Dependence

193

(iii) For any subset X M the following three statements are equivalent (a) X is dependent (b) Y M : X Y = Y is dependent (c) X0 X such that #X0 < and X0 is dependent (iv) For any subset Y M the following three statements are equivalent (a) Y is independent (b) X M : X Y = X is independent (c) Y0 Y we get #Y0 < = Y0 is independent (v) A subset B M is a basis of M if and only if it is a minimal generating subset of M . Formally that is the equivalency of (a) B is a basis of M (b) M = B and S B = M = S (vi) A subset B M is an R-basis of M if and only if any x M has a unique representation as an R-linear combination over B . Formally that is the equivalency of (a) B is an R-basis of M (b) for any element x M there is a uniquely determined tupel (xb ) RB such that x can be represented in the form x =
bB

xb b

Nota that the sum occuring in (b) is well-dened, as only nitely many of the xb are non-zero. Further note that in fact the existence of such a representation is equivalent to M = B m and the uniqueness of the representation is equivalent to the R-linear independence of B . (3.40) Proposition: (viz. 367) Let (S, +, ) be a skew-eld and M be an S -module. Let us denote the relation of S -linear dependence by again (refer to (3.37.(iv))). Then we obtain the following statements in addition to (3.39) (i) is a dependence relation (in the sense of (3.34)), that even satises the following property (for any X M and any x, y M ) x X, X y, y = 0 = (X \ { x }) { y } x

(ii) Let X M be an arbitrary subset, then the following two statements are equivalent (for arbitrary rings we only get (b) = (a)) (a) X is independent (b) X is S -linearly independent

194

3 Modules

(iii) For any X M and y M the following statements are equivalent (a) X { y } is S -linearly independent and y X (b) X is S -linearly independent and X y

(iv) For any subset B M all of the following statements are equivalent (a) B is a basis of M (b) B is an S -basis of M (c) B is a minimal S -generating subset of M , that is B generates M (i.e. M = lhS (B )) and for any X M we obtain X B = M = lhS (X )

(d) B is a maximal S -linearly independent subset, that is B is S linearly independent and for any X M we obtain B X = X is not S -linearly independent

(e) for any element x M there is a uniquely determined tupel (xb ) S B such that x can be represented in the form x =
bB

xb b

(v) Let X M be a S -linearly independent subset and Y M be a generating subset M = lhS (Y ). If now X Y , then there is some B M such that (1) X B Y and (2) B is an S -basis of M (vi) All S -bases of M have the same number of elements, formally that is A, B M S -bases of M = |A| = |B |

Thus by (v) M has an S -basis (it is a free module) and by (vi) all S -bases of M share the same cardinality. Hence we may dene the dimension of M over S to be the cardinality of any S -basis of M . That is we let dimS (M ) := |B | where B is an S -base of M (vii) A set of S -linearly elements contains at most dimS (M ) many elements X M S -linearly independent = |X | dimS (M )

(viii) Suppose M is nite-dimensional (i.e. dimS (M ) < ) and consider an S -linearly independent subset X M . Then the following two statements are equivalent (a) |X | = dimS (M ) (b) X is an S -basis of M 3.7 Linear Dependence 195

(ix) Im particular if M is nite-dimensional and P m M is a submodule of equal dimension dim(P ) = dim(M ) < then already P = M . (3.41) Lemma: (viz. 368) Let (S, +, ) be a skew-eld and consider two S -modules V and W . Then we obtain the following formulae concerning their dimensions (i) If P and Q m V are any two S -submodules of V , then the dimension of the sum P + Q satises the following relation dim(P + Q) = dim(P ) + dim(Q) dim(P Q) (ii) If P m V is an S -submodule of V then the dimension of the quotient is just the dierence of the dimensions of V and P dim V P = dim(V ) dim(P )

(iii) If : V W is any S -linear map, then then we obtain the following dimension formula for the dimension of the kernel and image of dim(V ) = dim(kn()) + dim(im())

196

3 Modules

3.8

Homomorphisms

Homomorphism of modules - better known as linear maps - are one of the most common concepts of algebra. Examples are abundant: derivation and integration of functions both are linear, the arithmetical mean is linear, matrices are just another way to dene linear maps, etc. The advantage of linear maps is, that they are both, well-understood and well-behaved. In fact this concept is so successful, that many theories (e.g. chaos theory) thrive on nding linear substitutes for non-linear problems. (3.42) Denition: (i) Let (R, +, ) be any ring and consider two R-modules M and N . Then a map : M N is said to be an R-module-homomorphism or simply an R-linear map, i for any a R and any x, y M we get (ax) = a(x) (x + y ) = (x) + (y ) We will denote the set of all R-module-homomorphisms from M to N by mhom(M, N ) or, in order to emphasize the base ring R, by mhomR (M, N ). In linear algebra (i.e. primarily in case of a base eld R) it is also customary to simply write L(M, N ): L(M, N ) := { : M N | is an R-module-homomorphism } mhom(M, N ) := mhomR (M, N ) := L(M, N ) (ii) Consider any two rings (R, +, ) and (S, +, ) and x a ring-homomorphism : R S . Then a map : M N is said to be an -module-homomorphism or simply an -linear map, i for any a R and any x, y M we get (ax) = (a)(x) (x + y ) = (x) + (y ) We will denote the set of all R-module-homomorphisms from M to N by mhom (M, N ) or, in case is understood, simply by mhom(M, N ). In linear algebra (i.e. primarily in case of a base eld R) it is also customary to write L (M, N ): L (M, N ) := { : M N | is an -module-homomorphism } mhom(M, N ) := mhom (M, N ) := L(M, N ) (iii) In any case (i) and (ii) we dene the kernel kn and the image im to be the following subsets of M or N respectively kn() := { x M | (x) = 0 } im() := { (x) | x M } M N

3.8 Homomorphisms

197

And we establish the usual nomenclature: is said to be a mono- epior isomorphism, i it is injective, surjective or bijective respectively. is called monomorphism epimorphism isomorphism i is injective surjective bijective

(iv) In case (i), an R-module-homomorphism : M M acting solely on the R-module M is called an R-module-endomorphism. And it is called an R-module-automorphism, i it also is bijective is called endomorphism automorphism i is M =N M = N and bijective

(v) If M and N are any two R-modules, then M and N are said to be isomorphic, i there is an isomorphism of R-modules from M to N , formally we write M =m N (3.43) Example: (i) The most easy and completely trivial example is the zero-homorphism. If M and N are any two R-modules, then we obtain an R-modulehomorphism 0 by letting 0 : M N : x0 (ii) Another trivial exapmle of an R-module-homorphism is the identity map 1 1 on some R-module M . This simply is 0 : M M : xx (iii) A slight generalisation is the map of set theoretical inclusion, which (by abuse of notation) is written as again. If P m M is a submodule of the R-module M , then we get an R-module-homorphism by : P M : xx (3.44) Remark: (viz. 344) (i) The notion (ii) of -linearity clearly includes the case (i) of R-linearity - just take = 1 1 : R R, then an -linear map : M N is nothing but an R-linear map. (ii) If : M N is an R-module-homomorphism, then the kernel and image of are submodules of M and N respectively: kn() = { x M | (x) = 0 } im() = 198 { (x) | x M } m m M N 3 Modules : : M N R-module isomorphism

More generally: if : M N is an -module-homomorphism then kn is an R-submodule of M . And if is surjective, then im also is an S -submodule of N . (iii) If : M N is an R-module-homomorphism, then has image N and is injective if it has kernel 0, formally that is is surjective is injective im = N kn = { 0 }

(iv) If L, M and N all are R-modules and the maps : L M and : M N both are homorphisms of R-modules then the composition is a homomorphism of R-modules as well : L N : x ((x)) More generally: suppose L is an R-module, M is an S -module and N is an T -module, where (R, +, ), (S, +, ) and (T, +, ) are arbitrary rings. Further we x ring homorphisms : R S and : S T . If now : L M is an -module-homorphism and : M N is an -module-homorphism, then the composition : L N is an ()-module-homorphism. (v) If M and N are R-modules, where (R, +, ) is a commutative ring, then the set mhomR (M, N ) of all R-module-homorphisms from M to N becomes an R-module in itself. Thereby for any , mhomR (M, N ) and any a R we take the following operations + : M N : x (x) + (x) a : M N : x a(x) (vi) More generally: suppose (R, +, ) is an arbitrary and (S, +, ) even is a commutative ring, M is an R-module and N is an S -module. Further x a homorphism : R S of rings. Then mhom (M, N ) becomes an S -module under the above operations of (v). (vii) Consider any non-empty subset = X M of M . For now we will examine the canonical R-module-homomorphism : RX M : (ax )
x X

ax x

Note that is well-dened, as in any (ax ) RX only nitely many ax are non-zero. Also - as the compositions on RX are component-wise - it is clear, that truly is a homomorphism of R-modules. Yet also satises a number of interesting properties im is surjective is injective is bijective 3.8 Homomorphisms = lhR (X )

X generates M X is R-linearly independent X is an R-basis of M 199

In particular, i B M is a basis, then is an isomorphy of R M . That is all free R-modules are of the form modules : RB B R for some set B .

(3.45) Proposition: (viz. 346) Let (R, +, ) be an arbitrary ring and consider any two R-modules M and N . Then we get the following two statements: (i) If X M generates M , that is M = ln(X ), and , : M N are any two R-linear maps, then we nd the equivalency
X

(ii) And if B M even is an R-basis of M and n : B N is an arbitrary map from B to N , then there is a uniquely dened R-linear map : M N satisfying B = n. (It is customary to say that is generated by linear expansion of the map n). And this is given to be
xB

ax x

:=
x B

ax n(x)

(3.46) Denition: (viz. 346) (i) Fix any ring (R, +, ) and consider the three R-modules L, M and N . If : M N is an R-module-homorphism, then it induces another R-module-homorphism called the push-forward of by letting : mhomR (L, M ) mhomR (L, N ) : (ii) Fix any ring (R, +, ) and consider the three R-modules L, M and N . If : L M is an R-module-homorphism, then it induces another R-module-homorphism called the pull-back of by letting : mhomR (N, M ) mhomR (L, N ) : (iii) More generally: suppose L is an R-module, M is an S -module and N is an T -module, where (R, +, ), (S, +, ) and (T, +, ) are arbitrary rings. Further we x ring homorphisms : R S and : S T . If now : M N is an -module-homorphism, then it induces another -module-homorphism called the push-forward of by letting : mhom (L, M ) mhom (L, N ) : (iv) More generally: suppose L is an R-module, M is an S -module and N is an T -module, where (R, +, ), (S, +, ) and (T, +, ) are arbitrary rings. Further we x ring homorphisms : R S and : S T . If now : L M is an -module-homorphism, then it induces a T -module-homorphism called the pull-back of by letting : mhom (M, N ) mhom (L, N ) : 200 3 Modules

The following statement seems to be a mere example - barely worth mentioning: xing an endomorphism of a module M over a commutative ring R we can turn M into a module over the polynomial ring R[t]. Yet this probably is the most important construction of linear algebra. Analyzing the structure of this module will grant deep insights into the homorphism . In fact the entire theory of canonical forms of an endomorphism (e.g. Jordan canonical form) will arise by applying the theorem of Pr ufer to this module. (3.47) Lemma: (viz. 348) Let (R, +, ) be a commutative ring, consider an R-module (M, +, ) and x an R-module-endomorphism : M M . Then the set M can be turned into a module (M, +, ) over the polynomial ring R[t] by retaining the addition + of M and establishing a new scalar multiplication : R[t] M M : (f, x) f ()(x) Again we write f x := f x = f ()(x). More explictly: if f R[t] is any polynomial over R then f can be written in the form if f = an tn + + a1 t + a0 R[t]. If now x M is any element of M , then the scalar multiplication is set to be

f x := f ()(x) =
k=0

f [k ]k (x) = an n (x) + + a1 (x) + a0 x

(3.48) Lemma: (viz. 396) () Cayley-Hamilton Theorem (i) Let (R, +, ) be commutative ring and A matn (R) be an n n matrix over R. Pick up the characteristic polynomial of A, that is c(t) := det(t1 1n A) R[t] By construction c(t) is a normed polynomial of degree n that satises c(0) = (1)n det(A). Then the matrix A satises the characteristic polynomial, that is we get
n

c(A) =
k=0

c[k ]Ak = 0 matn (R)

(ii) If (R, +, ) is a commutative ring again, M is a nitely generated Rmodule and : M M is an R-module-endomorphism, then there is a polynomial f (t) R[t] of deg(f ) = rank(M ) such that satises f
deg(f )

f () =
k=0

f [k ]k = 0 end(M )

3.8 Homomorphisms

201

(3.49) Denition: Let (R, +, ) be any ring and consider a family (Mi ) (where i I ) of Rmodules. Let nally M be a submodule of the direct product M m
i I

Mi

(i) Let N be another R-module, then a map : M N is said to be R-multilinear, i for any i I the composition with the canonical embedding i : Mi N is R-linear. That is for any i I , any a R and any xi , yi Mi we get (. . . , axi , . . . ) = a(. . . , xi , . . . ) (. . . , xi + yi , . . . ) = (. . . , xi , . . . ) + (. . . , yi , . . . ) (ii) Let (S, +, ) be another ring, let N be an S -module and x a ringhomomorphism : R S . Then a map : M N is said to be -multilinear, i for any i I the composition with the canonical embedding i : Mi N is -linear. That is for any i I , any a R and any xi , yi Mi we get (. . . , axi , . . . ) = (a)(. . . , xi , . . . ) (. . . , xi + yi , . . . ) = (. . . , xi , . . . ) + (. . . , yi , . . . ) (3.50) Example: (viz. 346) Let (R, +, ) be any commutative ring, then we can simultaneously regard R2 = R R as an R-module, or as a product of the R-module R itself, twice. In particular we can consider both R-linear and R-bilinear maps on R2 . Our examples now are: (i) Consider the addition map on R, that is : R2 R : (a, b) a + b. Then is R-linear but need not be R-multilinear. (ii) Consider the multiplication map on R, that is : R2 R : (a, b) ab. Then is R-multilinear but need not be R-linear.

202

3 Modules

3.9

Isomorphism Theorems

In this section we will study several basic properties of R-module-homorphisms. First of all we will encounter the fundamental isomorphism theorems. Later on we will get some more advanced criteria whether an endomorphism is an isomorphism. A more general way to look at a homorphism is to study a chain of homorphisms. This notion will also be introduced here. (3.51) Theorem: (viz. 347) Let (R, +, ) be any ring and consider any two R-modules M and N . Let further : M N be an R-module-homomorphism, then we get (i) If P m M and Q m N are submodules of M and N respectively, such that P 1 (Q), then we obtain a well-dened homomorphism of R-modules, by letting : M P N Q : x + P (x) + Q (ii) 1st Isomorphism Theorem: induces an isomorphy of R-modules from M/kn to im by virtue of
im : x + kn (x) : M kn

(iii) 2nd Isomorphism Theorem: if P and Q m M are any two submodules of M , we obtain another isomorphy of R-modules, by virtue of Q
P +Q P Q P : x+P Q x+P

(iv) 3rd Isomorphism Theorem: if P and Q m M are submodules of M again, such that P Q, then we obtain one more isomorphy of R-modules, by virtue of M
M/P Q Q Q/P : x + Q (x + P ) + P

(3.52) Corollary: (viz. 400) (i) Consider a commutative ring (R, +, ) and a nitely-generated R-module M . If now : M M is an R-module-endomorphism, then the following two statements are equivalent: (a) is surjective (b) is bijective (ii) Consider a skew-eld (S, +, ) and a nitely generated S -module V . If now : V V is an S -module-endomorphism, then the following two statements are equivalent: (a) is injective (b) is surjective (c) is bijective

3.9 Isomorphism Theorems

203

(3.53) Denition: Fix any number 1 n N and a ring (R, +, ). Further consider the Rmodules Mi (where i 1 . . . n + 1) and the R-module-homomorphisms, i (where i 1 . . . n). Let us denote M = (Mi ) and = (i ), then the ordered pair (M, ) is called a (nite) sequence of R-modules, i we get (1) for any i 1 . . . (n + 1): Mi is an R-module (2) for any i 1 . . . n: : Mi Mi+1 is an R-module homorphism (3) for any i 1 . . . (n 1): we have i+1 i = 0. Nota an equivalent formulation of the latter property is: im(i ) kn(i+1 ) In this situation we will use a rather intuitive notation for the ordered pair (M, ), namely we write: M1 M2 M3 Mn Mn+1 And the nite sequence (M, ) of R-modules is said to be exact, i apart from properties (1) and (2) we even have the following, stronger property (3) for any i 1 . . . (n 1) im(i ) = kn(i+1 )
1 2 n

(3.54) Remark: (i) If Mi = 0, then we necessarily have i = 0 and i1 = 0, that is i : 0 Mi+1 : 0 0 and i1 : Mi1 0 : x 0. As these maps are already determined, we dont have to denote them in the sequence Mi2 Mi1 0 Mi+1 Mi+2 If this sequence is exact, then it even follows that i+1 is injective and i2 is surjective, formally put Mi = 0 Mi = 0 = = i+1 is injective i2 is surjective
i2 i+1

Prob due to the exactness of the chain we get im(i2 ) = kn(i1 ) = kn(0) = Mi1 and hence i2 is surjective. Likewise kn(i+1 ) = im(i ) = im(0) = 0 and hence i+1 is injective. (ii) In particular for any R-module homorphism : M N we get a chain of R-modules, simply by taking 0 M N 0 And this chain is exact if and only if is an isomorphism. Prob due to (i) the chain is exact, i is injective and surjective, that is i is bijective. And thats when is called an isomorphism. 204 3 Modules

(iii) In a similar fashion we nd a chain-version of the rst isomorphism theorem: given an exact sequence of R-modules 0 L M N 0 the rst isomorphism theorem of R-modules yields the following isomorphy (Prob as im( ) = N and kn( ) = im()) M
im() N : x + im() (x)

(3.55) Example: (i) Any R-module-homorphism : M N induces a canonical short exact sequence of R-modules, simply by taking 0 kn() M im() 0 (ii) But also induces another chain of R-modules. Let us denote the canonical projection : N N/im() : y y + im(), then we nd another exact chain of R-modules, by taking
0 kn() M N N im() 0

(iii) Another, most important example is the following: consider any two R-modules M and N . Let us denote the canonical embedding of M by : M M N : x (x, 0) and the canonical projection onto N by : M N N : (x, y ) y . Then it is clear that im() = M 0 = kn( ), such that we get a short exact chain of R-modules, by 0 M M N N 0 (iv) A special case of (iii) is the following: consider any two distinct prime numbers 2 p = q Z. By the chinese remainder theorem we get Zp Zq = Zpq . Thereby the canonical maps become : Zp Zpq : a + pZ aq + pq Z and : Zpq Zq : b + pq Z b + q Z. And hence we obtain the short exact sequence of Z-modules 0 Zp Zpq Zq 0

(3.56) Lemma: (viz. 403) Let (R, +, ) be a commutative ring then the following statements are true (i) If R = 0 is non-zero and I , J = are any two non-empty sets, then RI is isomorphic (as an R-module) to RJ i I and J have the same cardinality, formally that is RI =m RJ 3.9 Isomorphism Theorems |I | = |J | 205

(ii) Let a, b i R be two ideals in R, if now R/a and R/b are isomorphic as R-modules, then a and b already are equal R R a =m b =
a=b

(3.57) Remark: The statement of (i) need not be true for non-commutative rings R. As an example x any ring R, let M := RN and take to the ring of R-module endomorphisms of M S := end(M ) m S n are isomorphic Then for any m, n N we nd that S m = as S -modules. In fact it suces to give a basis of S consisting of two elements. And these can be constructed using the fact, that S has innite dimension - for details refer to [Scheja, Storch, Algebra 1, Kapitel 25 Beispiel 3 und Kapitel 36 Beispiel 2]. We would also like to sketch a less constructive approach: Let again M := ZN and S := end(M ). It is easy to see M =m M M and hence S = =m mhom(M, M ) =m mhom(M, M M ) mhom(M, M ) mhom(M, M ) = S 2

A ring R with the property of (i) is also called to have IBN (invariant basis number). Thus (i) may be expressed as commutative rings have IBN. Of course there is a multitude of non-commutative rings that have IBN as well. An important example is: if : R S is a ring-epimorphism and S has IBN then R has IBN, as well. For more comments on this, please refer to the rst chapter of [Lam]. Likewise the converse of (ii) is false in general - e.g. regard S := R[s, t]. If we now dene the ideals a := tS and b := stS then a and b clearly are isomorphic as R-modules, but we nd S S
a b

=a =a

R[s] R[s] tR[t]

In (3.62.(ii)) we will present a much stronger statement. In fact any Rbasis of M has cardinality rank(M ) - supposed R is commutative. And as the isomorphism RI =m RJ transfers bases we could conclude |I | = rank(RI ) = rank(RJ ) = |J | Yet the proof of this statement is far longer and uses substantially more tools (the theory of noetherian modules), hence we have chosen to give a more direct proof in advance.

206

3 Modules

3.10

Rank of Modules

Remember that in section 3.7 we dened the rank of an R-module to be the minimal cardinality of a generating set of M , formally rank(M ) := min{ |X | | X M, M = lh(X ) } That is: if I = rank(M ) then there is some generating subset { mi | i I } M such that M = lh(mi | i I ). And conversely, if X M generates M as an R-module M = lh(X ), then |I | |X |. Be careful however - a minimal set of generators, need not have minimal cardinality ! E.g. the set { 2, 3 } Z is a minimal set of generators of the Z-module Z. That is { 2, 3 } generates Z = 2, 3 m and of course neither { 2 } nor { 3 } generates Z. However we have rank(Z) = 1, as Z = 1 m is clear. As we have seen in (3.40.(iv)) this is quite dierent in the case of modules over skew-elds, but even the base ring Z (and what a better in is there to wish for?) gives rise to such peculiarities. Let us now study the basic properties of the rank of a module, but be ready for some nasty surprises, as the theory will take some twists and turns. Many properties one would think of as evident will be untrue unless we deal with nice rings and modules. Others will hold true in astounding generality. (3.58) Example: (viz. 404) (i) The ring Z is an UFD (in fact even a PID) that contains innitely many prime elements. That is the following set is innite P := { p Z | 2 p is prime } (ii) The set Q regarded as an Z-module is has no nite set of generators. Hence it is of countably innite rank: a generating set is given by P := 1 | p P, k N pk

(3.59) Proposition: (viz. 405) Let (R, +, ) be an arbitrary ring and consider any two R-modules M and N . If now : M N is a surjective R-linear map (an R-module epimorphism), then rank(N ) rank(M ) Likewise, if P m M is an R-submodule then we get the following inequality rank(M ) rank(P ) + rank M P

3.10 Rank of Modules

207

(3.60) Proposition: (viz. 409) Let (R, +, ) be a commutative, non-zero ring R = 0. If now M and N are free R-modules, then M and N are isomorphic as R-modules, if and only if they share the same rank, formally that is M =m N rank(M ) = rank(N )

In particular, if M is free of rank I = rank(M ), then M is isomorphic to the euclidean R-module M =m RI presented in (3.38). (3.61) Remark: The above proposition (3.59) enables us to control the rank of M by controling the rank of a submodule P m M (and of the quotient M/P ). On the other hand one is tempted to expect that the rank of a submodule P of M cannot exceed the rank of the containing module M . However nothing is farther from the truth. Consider the ring R = Q[ti | I ] where I is an arbitrary index set. If I = 1 . . . n then we have the usual polynomial ring Q[t1 , . . . , tn ] that is a very tame ring - a noetherian integral domain. If we now regard the ring M = R as an R-module itself, it is clear, that rank(M ) = 1. However, hell breaks loose, once we take a look at the ideal m := ti | i I i i R. If we take m as an R-submodule of M = R, then the rank of m is no less than rank(M ) = 1 |I | = rank(m) By primary decomposition it is easy to see, that T := { ti | i I } is a minimal generating set m = lhR T and in fact it hits the rank. That is the submodule m of M can have arbitrarily large rank, even though M is of rank 1! Modules can have astoundligly vast space within themselves! The reason is that all those ti do not take up much space, they are all linearly dependent: tj ti ti tj = 0. So the question arises how to control the space inside a module. One of the best answers one can give refers to free submodules of M - these cannot be larger than M itself - refer to (3.62.(iii)). Also note that as long as I is nite R is noetherian. And in fact submodules of nitely generated modules over noetherian rings are again nitely generated. This will be the topic of section 3.11. Finally note that in case of |I | = 1 we even had rank(m) = rank(M ). But in this case R even is a PID and in fact nitely generated modules over PIDs are very well-behaved. (3.62) Lemma: (viz. 406) Consider a commutative ring (R, +, ) and an R-module M . Further suppose, that R = 0 is non-zero. Then the following statements are true (i) If { x1 , . . . , xn } M is an R-generating set of M (where n N), than any n + 1 elements of M are R-linearly dependant. That is: given y1 , . . . , yn , yn+1 M there are some a1 , . . . , an , an+1 R such that ak = 0 for some k 1 . . . n + 1 and: a1 y1 + + an yn + an+1 yn+1 = 0 208 3 Modules

(ii) If B M is an R-basis of M then the cardinality of B equals the rank of M , formally that is rank(M ) = |B | (iii) If F m M is a free submodule of M , then its rank does not exceed the rank of M , formally that is rank(F ) rank(M ) (3.63) Remark: Dimension In general, for an arbitrary ring (R, +, ) and a free R-module M the dimension is set to be the minimal cardinality of a basis of M dim(M ) := min{ |B | | B M is a basis of M } From this denition it is clear that the rank of M does not exceed the dimension of M [because any base B of M already generates M = lh(B )] rank(M ) dim(M ) In this light lemma (3.62.(ii)) tells us that in case of a commutative ring R all bases of a free R-module M share the same cardinality, namely the rank of M . In particular - if R is commutative and M is free - we even have dim(M ) = rank(M ) Thus - for commutative rings - the dimension is not a new invariant. However it is customary to refer to the dimension, rather than the rank, when operating with free modules. Also we have seen in (3.40.(vi)) that for any skew-eld (S, +, ) and any S -vector-space V all bases share the same cardinality, as well. It is easy to see that this is the rank again dim(V ) = rank(V ) Prob as rank(V ) dim(V ) is always true, it remains to prove dim(V ) rank(V ). Thus let X M be a generating set of M with |X | = rank(M ). Then by the basis selection theorem (3.36) there is a subset B X such that B is a basis of M . In particular dim(S ) = |B | |X | = rank(X ). After these two important cases, commutative rings and skew-elds, on might be tempted to think that generally (over arbitrary rings) all bases share the same cardinality, preferably the rank of M . This is untrue however. There are some modules over non-commutative rings, that have bases of dierent size. We have given one such example in (3.57).

3.10 Rank of Modules

209

(3.64) Denition: Let (R, +, ) be an arbitrary ring and M be an R-module. Then a subset L M is said to be a maximal linear independent set of M , i it is Rlinearly independent and for any other R-linearly independent set L M we get the implication L L = L = L

As we will see in (3.65.(iv)), if R is an integral domain then all maximal linear independent subsets of M are of the same cardinality. Hence in this case we may dene the linear freedom of M to be free(M ) := |L| for some maximal linear independent set L M

(3.65) Corollary: (viz. 409) Let (R, +, ) be an arbitrary ring and M be an R-module. Then we nd the following statements concerning maximal linear independent subsets of M (i) Any linear independent set in M is contained in a maximal linear independent set. That is: if L M is an R-linearly independent set, then there is a maximal linear independent set L M such that L L (ii) If R = 0 and L is a maximal linear independent set and x M is any element of M , then there is some a R such that a = 0 and ax lh(L) (iii) If L M is a linear independent set, then the submodule F := lh(L) generated by L is a free R-module with basis L. In particular if R = 0 is a non-zero commutative ring then the cardinality of any any linear independent set L in M is bounded by the rank of M . That is |L| rank(M ) (iv) In a more fancy language (ii) can be put as: If R = 0 is a non-zero ring and L is a maximal linear independent set, with F := lh(L) then the quotient M/F is a torsion R-module. That is z M F a R such that a = 0 and az = 0 Conversely if F m M is a free submodule of M such that the quotient M/F is a torsion module, then any basis B F is a maximal linear independent set B M in M . (v) Let now R even be an integral domain and consider a linearly independent set L M . Further for any x L let ax R be any non-zero element ax = 0 of R. Then the set L := { ax x | x L } is linearly independent again and |L| = |L |. 210 3 Modules

(vi) If R even is an integral domain, then any two maximal linear independent sets have the same cardinality. That is: if K and L M are any two maximal linear independent sets, then |K | = |L| rank(M ) (vii) If R is an integral domain and P m M is any submodule of M , then linear freedom of M equals the sum of the linear freedoms of P and the quotient M/P , formally that is free(M ) = free(P ) + free M P (viii) () Let R be an integral domain and denote its quotient eld by E := quotR. Then the linear freedom of M is just the dimension of the scalar extension of M to E , formally that is freeR (M ) = dimE (E R M ) (3.66) Remark: (i) It is clear that in the case of a free R-module M over an integral domain R the linear freedom equals the dimension, as any basis of M already is a maximal linear independent subset M free = free(M ) = dim(M )

However, if M is not free, then the rank of M may well be larger, than the linear freedom. The following two examples (ii) and (iii) should enlighten the situation gererally = free(M ) rank(M )

(ii) Consider the Z-module M := Zf (Zn )k for some arbitrary numbers f , k N. As F := Zf 0 m M is free and M/F = 0 (Zn )k is a torsion module (3.65.(iv)) implies that the linear freedom of M is precisely f . However (Zn )k contributes to the rank of M and hence f = free(M ) rank(M ) = f + k This is to be interpreted in the following way: the rank yields a more accurate description of the module, hence (3.60) - it regards every part of M . The linear freedom only is the dimension of the largest free module inside M , it neglects components with torsion. This makes it is easier to control, whereas the rank can explode incontrollably by picking up torsion. Just compare (3.65.(vii)) to (3.59).

3.10 Rank of Modules

211

(iii) Consider the integral domain R := Z[t] and regard the (maximal) ideal m := 2, t i i R as an R-module. Then m has a linear freedom of 1, as { t } is a maximal linear independent set (any other f R would yield a linearly dependent set). The rank is 2 however (m is not a principal ideal). This can happen, as R/m =m Z2 has torsion. Also note that m is not a free module, just because it is not a principal ideal.

212

3 Modules

3.11

Noetherian Modules

In the previous section we have seen, that the size (in terms of the rank ) of a submodule P of M can be very hard to control. However there are those well-behaved modules for which the rank remains nite at last. As it turns out these modules have a whole bunch of nice propoerties that we will explore in this section: (3.67) Denition: (viz. 382) Let (R, +, ) be commutative ring and M be an R-module. Then M is said to be a noetherian module, i it satises one of the following three equivalent conditions: (a) M satises the ascending chain condition of submodules - that is every ascending chain of submodules of M becomes stationary. I.e. any sequence (Pk ) (where k N) of submodules Pk m M that satises P0 P1 . . . Pk Pk+1 . . . becomes eventually constant - that is there is some s N such that P0 P1 . . . Ps = Ps+1 = . . . (b) Every non-empty P = family P subm (M ) of submodules of M contains a maximal element. I.e. there is some P m M such that P m M : P P = P = P (c) Every submodule P of M is nitely generated. That is for any submodule P m M there are some { x1 , . . . , xn } P such that P = lh(x1 , . . . , xn ) (3.68) Denition: Let (R, +, ) be commutative ring and M be an R-module. Then M is said to be a artinian module, i it satises one of the following two equivalent conditions: (a) M satises the descending chain condition of submodules - that is every descending chain of submodules of M becomes stationary. I.e. any sequence (Pk ) (where k N) of submodules Pk m M that satises P0 P1 . . . Pk Pk+1 . . . becomes eventually constant - that is there is some s N such that P0 P1 . . . Ps = Ps+1 = . . . (b) Every non-empty P = family P subm (M ) of submodules of M contains a minimal element. I.e. there is some P m M such that P m M : P P = P = P 3.11 Noetherian Modules 213

Prob the equivalence of these statements is no specialty of modules, but a general truism for partial orders. In this case the order in question is P Q dened by Q P . Then this is just an application of (2.28). (3.69) Remark: The notion of a noetherian (resp. artinian) module is closely related to the notion of a noetherian (resp. artinian) ring. In fact if we regard a commutative ring R as a module over itself, then the ideals a i R are precisely the submodules a m R of R. Hence by property (a) we see, that R is noetherian (resp. artinian) as a module, if and only if it is noetherian resp. artinian) as a ring. (3.70) Example: Clearly Z is a noetherian ring (as even is a PID) and hence a noetherian Z-module. However Z is not artinian, as there is an innitely descending chain of ideals, that does not stabilize, given by Z 2Z 4Z . . . 2k Z . . . Conversely pick up any prime p Z and regard the subring Z[1/p] r Q of Q generated by 1/p Q. Clearly Z r Z[1/p] is a subring as well, and hence a submodule. Therefore we may regard the quotient module I (p) := Z[1/p] Z as an Z module. By the way: this is just the p-primary component of Q/Z, that is I (p) = x + Z | k N : pk x + Z = 0 + Z . Then I (p) is not noetherian, as there is an innitely ascending chain of submodules of I (p) that does not stabilize Z Z 1 +Z p Z 1 +Z p2 Z 1 +Z p3 ...

However it can be shown, that I (p) is a artinian module. Recall that in the case of commutative rings, any artinian ring already is noetherian. This is not the case for modules over commutative rings however! And I (p) is just an example for this. (3.71) Theorem: (viz. 383) Let (R, +, ) be a commutative ring and M , N be R-modules. We also regard two submodules P , Q m M of M . As the following statements will be true for noetherian and artinian modules alike, let us abbreviate { noetherian, artinian } in order to avoid giving two completely analogous lists of theorems. Then the following statements hold true:

214

3 Modules

(i) M is a module if and only if both P and the quotient M/P are modules. Formally that is M is P and M P are

(ii) If both quotient modules M/P and M/Q are modules, then the quotient M/(P Q) is a module again. Formally that is M M P and Q are = M P Q is

(iii) If both P and Q are modules, then so is their sum P + Q m M . Formally that is the implication P and Q are = P + Q is

(iv) If both M and N are modules, then the direct sum M N of M and M is a module again and vice versa. Formally that is M and N are M N is

(v) If : M N is a surjective homorphism of R-modules and M is a module, then so is its image N . Formally that is M N and M is = N is

(vi) If M is a nitely generated R-module and R is a ring, then M already is a module. Formally that is M nitely generated and R is = M is

(vii) If L, M and N are R-modules and 0 L M N 0 is an exact sequence of R-module homomorphisms, then we get the implication M is L and N are

(3.72) Corollary: (viz. 391) Let (R, +, ) be a commutative ring and M be an R-module, then we obtain the following statements (i) M is noetherian, if and only if M is nitely generated and R/ann(M ) is a noetherian ring, formally that is M noetherian rank(M ) < and R ann(M ) noetherian

(ii) Under the assumption that M is nitely generated: M is artinian, if and only if R/ann(M ) is an artinian ring, formally that is M artinian 3.11 Noetherian Modules R ann(M ) artinian 215

(iii) If M is a nitely generated, artinian R-module, then M already is an noetherian R-module, as well. Formally again M artinian and nitely generated = M noetherian

(iv) If M is noetherian and : M M is an R-module-endomorphism, then if is surjective, it already is bijective, that is is surjective is bijective

Nota in fact we have already proved this statement for any nitely generated module M over a commutative ring R in (3.52). However the proof of this requires much more eort than the proof of this statement here. (v) If M is artinian and : M M is an R-module-endomorphism, then if is injective, it already is bijective, that is is injective is bijective

(vi) () M is both artinian and noetherian, if and only of M is a module of nite length, formally that is M artinian and noetherian (M ) <

(vii) () If M is a semi-simple R-module (e.g. a vector space over a eld), then the following ve statements are equivalent: (a) M is artinian (b) M is noetherian (c) M is nitely generated (d) M has nite length (M ) < (e) M is a nite direct sum of simple R-modules, that is there are nitely many (i 1 . . . r) simple submodules Mi m M of M such that M = M1 Mr In case M even is a vector-space, that is R even is a eld, then we obtain the following sixth equivalent statement: (f) M is nite-dimensional dim(M ) < (viii) Suppose M is a nitely generated R-module, such that every ascending chain of ideal-induced submodules of M stabilizes, then M is noetherian. That is: suppose for every seqence of ideals (ak ) where ak i R and k N that satises
a0 M a1 M a2 M . . .

there is some s N such that for any i N we get as+i M = as M , then M already is a noetherian R-module. 216 3 Modules

(3.73) Denition: (viz. 385) Let (R, +, ) be an arbitrary ring and M be an R-module. Then M is said to be a simple R-module if M = 0 and M has no proper submodules. That is M = 0 and it satises one of the following two equivalent properties: (a) For any submodule P m M of M we either have P = 0 or P = M (b) Any non-zero element of M generates M as an R-module, that is for any x M with x = 0 we already get M = Rx. And if R is commutative, we obtain the following third equivalent statement (c) There is some maximal ideal m i R of R such that M is isomorphic to the quotient ring R/m as an R-module m smax R such that M =m R m Let (R, +, ) be an arbitrary ring and M be an R-module. Then M is said to be semi-simple, i M can be decomposed into a direct sum of simple submodules. That is there is an arbitrary family Mi m M of submodules (where i I ), such that Mi is a simple R-module (for any i I ) and M is the inner direct sum of the Mi M =
i I

Mi

(3.74) Denition: Let (R, +, ) be a commutative ring and M be an R-module. Then the tuple (M0 , M1 , . . . , Mr ) is said to be a composition series of M , i the Mi form an ascending chain of submodules Mi m M (for any i 0 . . . r) 0 = M 0 M1 . . . M r = M such that for any i 1 . . . r the quotient module Mi /Mi1 is simple. And by (3.73) this can be put as: for any i 1 . . . r and any submodule P m M of M we get the following implication Mi1 P Mi = P = Mi1 or P = Mi

If M admits a composition series (M0 , M1 , . . . , Mr ) then M is said to be of nite length. And we will see in (3.76) that in this case the length r of the series is uniquely determined. Hence this number will be called the length of the module M , denoted by (M ) := r N And in case M admits no composition series we say that M is of innite length and we denote this situation, by writing (M ) :=

3.11 Noetherian Modules

217

(3.75) Remark: (i) If (M ) = 0 then 0 = M0 = M , that is M is the zero-module. And clearly if M = 0 is zero, then (0) is the only available composition series. Hence there is the equivalency (M ) = 0 M = 0

(ii) If (M ) = 0 then 0 = M0 M1 = M , that is M = M/0 = M1 /M0 is a simple R-module. Conversely if M is simple then (0, M ) already is the one and only composition series of M , such that (M ) = 0 M is simple

(iii) We will oftenly use the following trick: if N is a simple R-module and : M N is an R-module monomorphism, then already M = 0 or N even is an isomorphism. :M Prob let Q := im() m N be the image of M in N . Then, as N is simple, we either have Q = 0 or Q = N . But as is injective im() = 0 implies M = 1 (0) = 0 and im() = N means that is surjective and N. hence an isomorphism : M (iv) Another trick is the following observation: if M is a simple R-module and : M N is an R-module-epimorphism, then already N = 0 or N even is an isomorphism. :M Prob let P := kn( ) m M be the kernel of in M . Then, as M is simple, we either have P = 0 or P = M . But kn( ) = 0 implies that N ) and kn( ) = M is injective (and hence an isomorphism : M implies N = im( ) = 0. (v) If V is a nite-dimensional vectorspace over the eld E , then the dimension of V is precisely the length of V , that is (V ) = dim(V ) Prob let { v1 , . . . , vn } V be a basis of V , then we denote the subspaces Vi := lh(v1 , . . . , vi ) v V for i 0 . . . n. Then we see E =v Vi /Vi1 : a avi + Vi1 such that Vi /Vi1 is simple. Hence (V0 , V1 , . . . , Vn ) is a composition series of V and in particular (V ) = n. (vii) If V is a vector-space over the skew-eld (S, +, ) and x V is any nonzero element x = 0 of V . Then clearly Sx V is a simple submodule of V . In particular V already is semi-simple. To see this simply choose a basis B V of V , then V can be decomposed, into V =
bV

Sb

218

3 Modules

(3.76) Theorem: (viz. 385) Jordan-H older Theorem Let (R, +, ) be a commutative ring and M be an R-module of nite length (M ) < . Then we obtain the following statements (i) If P m M is a submodule of M with P = M , then the length of M exceeds the length of P , formally that is (P ) < (M )

(ii) If M is an arbitrary R-module and P M is any submodule of M then M has nite length if and only if both P and the quotient M/P are of nite length, formally that is M has nite length P and M P have nite length

And in this case we even obtain the following equality for the length of the modules involved (M ) = (P ) + M P

(iii) Any two composition series of M are equivalent : That is, given any two composition series (M0 , M1 , . . . , Mr ) and (N0 , N1 , . . . , Ns ) of M , then we get r = s and there is a permutation Sr such that for any i 1 . . . r there is an isomorphy Mi N(i) Mi1 =m N(i)1

(iv) Let us denote the set of all strictly increasing chains of submodules of M , that start in 0 and terminate in M by C (0, M ), that is C (0, M ) := (P0 , P1 , . . . , Pr ) i 0 . . . r : Pi m M 0 = P0 P1 . . . Pr = M

Then we obtain a partial order on C (0, M ) by virtue of the following denition: given two chains (P0 , P1 , . . . , Pr ) and (Q0 , Q1 , . . . , Qs ) we say that (P0 , P1 , . . . , Pr ) (Q0 , Q1 , . . . , Qs ) i r s and there is an injective map : 0 . . . r 0 . . . s with (0) < (1) < < (r) such that for any i 0 . . . r we get Q(i) = Pi And thereby, if (M0 , M1 , . . . , Mr ) C (0, M ) is a series of submodules of M the following statements are equivalent (a) r = (M ) (b) (M0 , M1 , . . . , Mr ) is maximal (c) (M0 , M1 , . . . , Mr ) is a composition series

3.11 Noetherian Modules

219

(v) Any series of submodules of M can be rened into a composition series: That is, if P0 P1 . . . Pk is a strictly increasing chain of submodules Pi m M , then there is a composition series (M0 , M1 , . . . , Mr ) of M such that (in the sense of (iv) above) we get (P0 , P1 , . . . , Pk ) (M0 , M1 , . . . , Mr ) (vi) Consider a nite sequence of R-modules Mi where i 0 . . . n. That is for any i 0 . . . n we have i+1 i = 0 0 M0 M1 . . . Mn 0 If for any i 0 . . . n the module Mi is of nite length and Hi denotes the i-th homology module Hi := kn(i+1 )/im(i ) then we get
n n 0 1 2 n n+1

(1) (Mi ) =
i=0 i=0

(1)i (Hi )

220

3 Modules

3.12

Localisation of Modules

We have already studied the localisation of rings in section 2.9, now we would like to transfer this concept to modules. It turns out that the localisation of modules combines neatly with the localisation of modules. Also localisation will preserve all the nice properites a module might have. Hence localisation of modules is as powerful as the localisation of rings and almost as useful. Recall denition (2.93) which introduces the notion of a multiplicatively closed subset U R of a commutative ring (R, +, ). This is a subset U , such that 1 U and for any u, v U we have uv U , as well. We have already studied these sets in detail in section 2.9, so the reader is asked to refer to this section, in case of doubt. Let us now redo the construction of localisation for modules in precisely the same manner we used for rings: (3.77) Denition: (viz. 222) Let (R, +, ) be any commutative ring, U R be a multiplicatively closed subset and M be an R-module. Then we obtain an equivalence relation on the set U M by virtue of (with x, y M and u, v U ) (u, x) (v, y ) : w U : vwx = uwy

And we denote the quotient of U M modulo by U 1 M and the equivalence class of (u, x) is denoted by x/u, formally this is x u U 1 M := { (v, y ) U M | (u, x) (v, y ) } := U M

Now U 1 M becomes an R-module again, even an U 1 R-module, under the following addition and scalar multiplications (where a R) x y + u v x a u ay uv := := := vx + uy uv ax u ay uv

It is clear that under these operations the zero-element of U 1 M is given to be 0/1 and we obtain a canonical R-module homomorphism from M to U 1 M by letting x : M U 1 M : x 1 In general this canonical homomorphism need not be and embedding (i.e. injective). Yet we do obtain the following implications injective bijective = U nzdR U R

3.12 Localisation of Modules

221

Proof of (3.77): We will rst prove that is an equivalence relation: the reexivity is due to 1 U , as u1a = u1a implies (u, a) (u, a). For the symmetry suppose (u, a) (v, b), i.e. there is some w U such that vwa = uwb. As = is symmetric we get uwb = vwa and this is (v, b) (u, a) again. To nally prove the transitivity we regard (u, a) (v, b) and (v, b) (w, c). That is there are p, q U such that pva = pub and qwb = qvc. Since p, q and v U we also have pqv U . Now compute (pqv )wa = qw(pva) = qw(pub) = pu(qwb) = pu(qvc) = (pqv )uc, this yields that truly (u, a) (w, c) are equivalent, too. Next we need to check that the operations + and are well-dened. Suppose a/u = a /u and b/v = b /v that is there are p and q U such that pu a = pua and qv b = qvb. Then we let w := pq U and compute u v w(av + bu) = vv q (pu a) + uu p(qv b) = vv q (pua ) + uu p(qvb ) = uvw(a v + u b ) u v wab = (pu a)(qv b) = (pua )(qvb ) = uvwa b That is (a/u) + (b/v ) = (av + bu)/uv = (a v + b u )/u v = (a /u ) + (b /v ) and (a/u)(b/v ) = ab/uv = a b /u v = (a /u )(b /v ) respectively. And this is just the well-denedness of sum and product. Next it would be due to verify that U 1 R thereby becomes a commutative ring. That is we have to check that + and are associative, commutative and satisfy the distributivity law. Further that for any a/u U 1 R we have a/u + 0/1 = a/u and a/u + (a)/u = 0/1 and nally that a/u 1/1 = a/u. But this can all be done in straightforward computations, that we wish to omit here. It is obvious that thereby becomes a homomorphism of rings, (0) = 0/1 is the zero-element and (1) = 1/1 is the unit element. Further (a + b) = (a + b)/1 = (a1 + b1)/1 1 = a/1 + b/1 = (a) + (b) and (ab) = ab/1 = ab/1 1 = (a/1)(b/1) = (a)(b). Let us now prove that is injective if and only if U nzdR: note that is injective, i kn() = 0 that is if x/1 = 0/1 implies that x = 0. But by denition x/1 = 0/1 means wx = w0 = 0 for some w U . That is is injective if and only if U and the set R \ nzd(M ) = { a R | x M such that x = 0 and ax = 0 } are disjoint. In other words that is U nzd(M ). Let us nally assume that U R . Then (as M = 0) it is clear that U nzd(M ) such that already is injective. It remains to prove the surjectivity. Hence consider any y/u U 1 M . Then we chose w = 1 and x = u1 y from which we nd wux = uu1 y = y = wy such that (x) = x/1 = y/u. Together we derived the bijectivity of . 222 3 Modules

3.12 Localisation of Modules

223

Chapter 4

Linear Algebra
4.1 Matix Algebra

Martices serve two purposes: rst of all they can represent systems of linear equations. And secondly (by xing ordered bases) they can be used to represent homorphisms between free modules. Using matrices these concepts are interlocked: the set of solutions is the kernel of the homorphism (resp. an ane shift of the kernel in case of an inmomogeneous equation). Also matrices provide a method of keeping track of the manipulations of a system of linear equations - the elementary matrices. Hence the algorithm how to solve a system of linear equations (Gauss-Algorithm) can be expressed in terms of matrices. This ultimately leads to a criterion whether a system of linear equations is (unambiguously) solvable - the determinant. So, as a rst step, let us introduce the notion of a matrix and dene the operations on matrices that will be used later on. (4.1) Denition: Let (R, +, ) be any ring and x a natural number 1 n N. As before we will concern ourselves with the set Rn of n-tuples over R Rn := { (x1 , . . . , xn ) | x1 , . . . , xn R } In linear algebra it is customary (and occasionally useful) to write the ntuple x = (x1 , . . . , xn ) in vertical order as well x1 . (x1 , . . . , xn ) = . . xn We already turned Rn into an R-algebra by establishing the pointwise operations on Rn , let us recall this: given any x = (x1 , . . . , xn ) and y = (y1 , . . . , yn ) Rn and any a R we have dened the operations: (x1 , . . . , xn ) + (y1 , . . . , yn ) := (x1 + y1 , . . . , xn + yn ) (x1 , . . . , xn ) (y1 , . . . , yn ) := (x1 y1 , . . . , xn yn ) a(x1 , . . . , xn ) := (ax1 , . . . , axn )

224

As a new operation on Rn we dene the scalar product (do not confuse this with the scalar multiplication ) Rn Rn R : (x, y ) x, y which is dened to be (as always x = (x1 , . . . , xn ) and y = (y1 , . . . , yn ))
n

x, y

:=
i=1

xi yi

Finally we need to x an easy notation for the canonical projections i of the product Rn that is i : Rn R : x xi . Recall that the i thereby are R-algebra homomorphisms. Given any i 1 . . . n we denote [x]i := xi where x = ( x1 , . . . , x n )

(4.2) Remark: (i) There are a multitude of ways how to embed R Rn as a subalgebra. The most reasonable one (due to its symmetry) is to interpret a R as the tuple (a) = (a, . . . , a) Rn . The advantage of this is evident: the multiplication then coincides with the scalar multiplication: (a) x = ax. Hence we x R Rn : a (a) := (a, . . . , a) (ii) Trivially Rn is a free module, having E = { e1 , . . . , en } the Euclidean R-basis. Thereby the elements ei Rn are given to be (refer to (3.38) for details) e1 := (1, 0, 0, . . . , 0, 0) e2 := (0, 1, 0, . . . , 0, 0) . . . en := (0, 0, 0, . . . , 0, 1) (iii) Clearly the scalar product on Rn has some neat computational properties: in fact it is a bilinear form of Rn , that is for any x, y , z Rn and any a R we get ax + y, z = a x, z + x, z x, ay + z = a x, y + x, z R-linearity (in the 1st argument) R-linearity (in the 2nd argument)

And if R even is a commutative ring, then the scalar product even is a symmetric bilinear form, that is in this case we get a third property: x, y = y, x

Prob all these properties are straightforward applications of the axioms of rings in conjunction with the pointwise operations: ax + y, z = i (axi + yi )zi = a i (axi yi + xi zi ) = a i x i yi + i xi zi = a x, y + x, z . And the R-linearity in the second argument can be proved 4.1 Matix Algebra 225

in complete analogy. If R is commutative then the symmetry of the scalar product is evident: x, y = i xi yi = i yi xi = y, x . (iv) Note that in general there is a dierence between the scalar product x, y and an inner product x | y . The latter is dened in the case of R = R or R = C only. And in the case R = C there is a subtle distinction between the scalar and inner products - the inner products uses complex conjugates. However this dierence has dire consequences: the scalar product over C is not positively denite, in fact we see that 1 i , 1 i = 12 + i2 = 1 1 = 0

(4.3) Denition: Let (R, +, ) be any ring again and x two natural numbers 1 m, n N. Then a (m n)-matrix is dened to be an n-tuple of m-tuples over R. Formally we will denote this set, by matm,n (R) := (Rm )n A matrix usually is written as an m times n block of elements of R. That is if we have the m-tuples aj = (a1,j , . . . , am,j ) Rm and j 1 . . . n then we write the matrix A = (a1 , . . . , an ) in the following form a1,1 a2,1 . . . a1,2 a2,2 . . . a1,n a2,n := (ai,j ) := A . . . am,n

am,1 am,2

That is the second index j of A = (ai,j ) designates the j th m-tuple aj of A and this is written as the j th column of A. Complementary the rst index i of A = (ai,j ) designates the ith component within the m-tuple aj and these are written as the ith row of A. To introduce a formal notation let us x the (m n)-matrix A = (ai,j ) as above. Then - for any i 1 . . . n and any j 1 . . . n - we denote the (i, j )th entry of A, the ith row of A and the j th column of A respectively by enti,j (A) := ai,j R rowi (A) := (ai,1 , . . . , ai,n ) Rn colj (A) := (a1,j , . . . , am,j ) Rm The transpose A of an (m n)-matrix A = (ai,j ) is dened to be the (n m)-matrix in which colums and rows are interchanged, that is a1,1 a1,2 := . . . a2,1 a2,2 . . . am,1 am,2 matn,m (R) . . . am,n 4 Linear Algebra

a1,n a2,n 226

If m = n then the rectangular block A of elements becomes a square. As this is an important special case we abbreviate our notation somewhat: matn (R) := matn,n (R) = (Rn )n A special case of such a matrix is a diagonal matrix. That is a square matrix A = (ai,j ) in which all entries o the diagonal are zero, i.e. ai,j = 0 for any i = j 1 . . . n. And if we are given n-elements a1 , . . . , an R then the diagonal matrix with the entries ai on its diagonal is denoted by a1 0 0 a2 . .. dia(a1 , . . . , an ) := . . . 0 0 0 0 0 0 an1 0 0 0 . . matn (R) . 0 an

(4.4) Remark: (i) By denition of a (m n)-matrix, every n m-tuples aj = (ai,j ) Rm (where j 1 . . . n) determine a matrix A := (a1 , . . . , an ). To stress that the aj build up the columns of A we write a1 an := (ai,j )

Conversely, if we are given m n-tuples ai = (ai,j ) Rn (where i 1 . . . m) we can build up a (m n)-matrix A = (ai,j ) by taking the ai to be the rows of A. In this case we write a1 . . := (ai,j ) . am (ii) Recall that an (m n)-matrix A = (ai,j ) is dened to be an n-tuple of m-tuples of R. And the (i, j )th entry ai,j of A is the ith component of the j th component of A. So in a somewhat formal way that is [A]i,j := ai,j = [[A]j ]i

(4.5) Denition: Let (R, +, ) be any ring again and consider the natural numbers 1 m and n N. By construction the set matm,n (R) of (m n)-matrices of R forms an R-module under the pointwise operations. That is: consider a R and two (m n)-matrices A = (ai,j ) and B = (bi,j ) matm,n (R). Then we dene the addition and scalar-multiplication A + B := (ai,j + bi,j ) aB := (a bi,j ) 4.1 Matix Algebra 227

Now consider another natural number 1 l N and suppose that A matl,m (R) is a (l m)-matrix, whereas B matm,n (R) is an (m n)matrix. Then we can dene the matrix-multiplication AB of A and B to be the following (l n)-matrix AB := ( rowi (A), colk (B ) ) =
j =1 m

ai,j bj,k matl,n (R)

A special case of this is the matrix-vector multiplication: given an (m n)matrix A matm,n (R) and an n-tuple x Rn over R we obtain an m-tuple Ax Rn by virtue of Ax := ( row1 (A), x , . . . , rowm (A), x ) Rm

(4.6) Remark: (i) Let us reformulate the above denitions of addition, scalar multiplication and vector multiplication in terms of the entries of the resulting matrices. That is we take A, B matm,n (R), a R and x Rn . Then for any i 1 . . . m and any j 1 . . . n we have enti,j (aA + B ) = a enti,j (A) + enti,j (B )
n

[Ax]i =
j =1

enti,j (A) [x]j

And if A matl,m (R) and B matm,n (R) we can also reformulate the matrix multiplication for all the entries i 1 . . . l and k 1 . . . n
m

enti,k (AB ) =
j =1

enti,j (A) entj,k (B )

(ii) Note that the matrix-matrix multiplication AB is nothing but a matrixvector multiplication Abk for all the column vectors bk of B . That is given A matl,m (R) and bk Rm (where k 1 . . . n) we have A b1 Abm

bm = Ab1

(iii) Let us also reformulate the transposition A A of matrixes in terms of the rows, columns and entries. Then it is clear that for any A matm,n (R) and any i 1 . . . m, j 1 . . . n we get entj,i (A ) = enti,j (A) coli (A ) = rowi (A) rowj (A ) = colj (A) Prob recall that entj,i (A ) = enti,j (A) is just the denition of the trans228 4 Linear Algebra

pose A . And thus we nd coli (A ) = (ent1,i (A ), . . . , entn,i (A )) = (enti,1 (A), . . . , enti,n (A)) = rowi (A). Likewise we compute rowj (A ) = (entj,1 (A ), . . . , entj,m (A )) = (ent1,j (A), . . . , entm,j (A)) = colj (A).

4.1 Matix Algebra

229

(4.7) Proposition: (viz. 311) Consider any ring (R, +, ) and x the natural numbers 1 l, m, n, p and q N. Then all of the following statements hold true (i) The set matm,n (R) of (m n)-matrices over R is an R-module under the addition and scalar multiplication of matrices, as dened in (4.5). The zero-element thereby is the zero-matrix 0 0 0 0 0 = (0) = . . . . . . 0 0 0 0 . . . 0

In fact matm,n (R) is a free R-module of rank mn. We can easily provide the Euclidean R-basis E := { Er,s | r 1 . . . m, s 1 . . . n } of matm,n (R). Thereby Er,s is the (m n) matrix that is entirely 0, except its (r, s)th entry, which is 1. Formally Er,s := (i,r j,s ) matm,n (R) (ii) The multiplication of matrices is associative and distributive (but in general not commutative). Suppose that A, B and C are of compatible sizes, then (AB )C = A(BC ) A(B + C ) = AB + AC (A + B )C = AC + BC To be of compatible sizes means A matm,n (R), B matn,p (R) and C matp,q (R) for the associativity, A matl,m (R), B and C matm,n (R) for the left-distributivity and A, B matl,m (R) and C matm,n (R) for the right-distributivity. (iii) In particular the set matn (R) of (nn)-matrices over R is an R-algebra under the addition, scalar-multiplication and matrix-multiplication of matrices as dened in (4.5). Its unit-element is the identity-matrix 1 0 0 1 1 1 = (i,j ) = . .. . . . 0 0 0 0 1

(iv) Recall the denition of the center cen(R) of a ring R to be the subring of all elements that commute with all other elements, that is cen(R) = { a R | b R : ab = ba }. Then the center of matn (R) is just the center of R embedded as diagonal matrices: a 0 0 a cen (matn (R)) = cen(R)1 1 = . .. . . . 0 0 230 0 0 a 4 Linear Algebra

a cen(R)

(v) The transposition A A of matrices is an isomorphism of R-modules mat matm,n (R) n,m (R). In particular for any a R and any A, B matm,n (R) we have (aA + B ) = aA + B (vi) If R is commutative, then the transposition of matrices swaps the order of matrix-multiplications. That is given any matrices A matl,m (R) and B matm,n (R) we nd that (AB ) = B A (vii) If R is commutative then taking inverses in matn (R) commutes with the transposition. To be precise: if A matn (R) is a square matrix, then A is invertible, if and only if A is invertible. And in this case the inverse of A is given to be (A )1 = A1

(viii) If R is a commutative ring, then for any vectors x Rn , y Rm and matrix A matm,n (R) we get the following identity in R Ax, y = x, A y

4.1 Matix Algebra

231

4.2

Elementary Matrices

(4.8) Denition: Let (R, +, ) be any ring and 1 n N. Then we will dene the set gln (R) of invertible matrices, the set trin (R) of (upper) triangular matrices and the set dian (R) of diagonal matrices to be the following subsets of matn (R) gln (R) := { A matn (R) | B matn (R) : AB = 1 1n = BA } trin (R) := { A matn (R) | 1 i < j n : enti,j (A) = 0 } a1 0 0 0 a2 0 dian (R) := a1 , . . . , a n R . . .. . . 0 0 an

(4.9) Proposition: (viz. 314) Let (R, +, ) be any ring and 1 n N. Then the invertible matrices gln (R) is precisely the multiplicative group of the ring matR . In particular gln (R) is a (non-commutative) group under the multiplication of matrices gln (R) = (matn (R)) And the upper trianguar matrices form a subring of the ring matn (R), in particular trin (R) is closed under the addition and multiplication of matrices trin (R) r matn (R)

(4.10) Denition: Let (R, +, ) be any ring and 1 n N. Then we will introduce a number of special (n n)-matrices. First of all, for any r, s 1 . . . n we recall the denition of the Euclidean matrices Er,s := (i,r j,s ) matn (R) And is a1 , . . . , an R are any scalars we can thereby introduce the diagonal matrix having the entries ai to be the following
n

dia(a1 , . . . , an ) :=
i=1

ai Ei,i

a1 0 0 a2 = . .. . . . 0 0

0 0 an

Another important matrix will simply be denoted by Nn , it is an upper triangular matrix whose only non-zero entries are 1s, one step above the 232 4 Linear Algebra

diagonal Nn := (i+1,j ) = 0 1 0 .. . trin (R) .. . 1 0

Now, if a R and i, j 1 . . . n, we dene a dilatation matrix Di (a) and a translocation matrix Ti,j (a) to be (n n)-matrices of the following form Ti,j (a) := 1 1n + aEi,j Di (a) := 1 1n + (a 1)Ei,i Consider any i, j 1 . . . n again, then we dene a transposition matrix Pi,j to to be an (n n)-matrices of the following form Pi,j = 1 1n Ei,i Ek,k + Ei,k + Ek,i More generally, let us denote the Euclidean basis of Rn by ej = (i,j ) again. That is e1 = (1, 0, . . . , 0), e2 = (0, 1, 0, . . . , 0) and so on until en = (0, . . . , 0, 1). Then for any function : 1 . . . n 1 . . . n we dene permutation matrix P of to be the following P := ((i),j ) = e(1) . . . e(n) The set of elementary matrices nally consists of all dilatation matrices Di (a), all translocation matrices Ti,j (a) for i = j and all transposition matrices Pi,j with i = j . It will be denoted by elmn (R) := { Di (a) | i 1 . . . n, a R } { Ti,j (a) | i = j 1 . . . n, a R } { Pi,j | i = j 1 . . . n } matn (R)

(4.11) Remark: The above denitions are formally correct but some of them are a little bit obfuscated. Hence we wish to enlighten them by rewriting the respective formulas in matrix notation: First of all the dilatation matrix Di (a) is nothing but a diagonal matrix - start with the identity matrix 1 1n but replace the (i, i)-th coecient with a. That is 1 .. . a Di (a) = dia( 1, . . . , a, . . . , 1 ) = . . . a at i-th position 1 4.2 Elementary Matrices 233

 Next - for i = j - the transposition matrix Ti,j (a) is just the identity matrix 1 1n with the (i, j )-th entry being a instead of 0. That is Ti,j (a) = 1 .. . 1 a .. . 1 .. . 1

By denition the transposition matrix Pi,j is the identity matrix in which the (i, i)-th entry 1 is moved to (j, i) and the (j, j )-the entry 1 is moved to (i, j ). In that we nd that Pi,j is the permutation matrix that belongs to (i j ), i.e. the permutation that interchanges i and j 1 .. . 0 1 . .. = 1 0 .. . 1

Pi,j = P(i j )

(4.12) Proposition: (viz. 314) Let (R, +, ) be any ring and 1 m, n N. Then we obtain the following formulae when multiplying one of the above matrices with another matrix (i) For any r, s 1 . . . let Er,s matn (R) denote the respective Euclidean matrix. Then for any (m n)-matrix A matm,n (R) and any (n m)matrix B matn,m (R) we get colk (AEr,s ) = rowi (Er,s B ) = colr (A) 0 rows (B ) 0 if k = s if k = s if i = r if i = r

(ii) Consider the square matrix A = (ai,j ) matn (R), then for any four indices i, j , k and l 1 . . . n we have the following identity Ei,j A Ek,l = aj,k Ei,l Using this equation on the identity matrix A = 1 1n = (i,j ) we also get Ei,j Ek,l = j,k Ei,l (iii) For any r, s 1 . . . let Er,s matn (R) denote the respective Euclidean matrix again. Also let a R, A matm,n (R) and B matn,m (R). 234 4 Linear Algebra

Multiplying with the translocation matrix Tr,s (a) from the left adds a-times the sth row to the rth row of B . Likewise, multiplying with Tr,s (a) from the right adds a-times the rth column to the sth column of A. That is for any i, j 1 . . . n we get colk (ATr,s (a)) = rowi (Tr,s (a)B ) = colk (A) + colr (A) a colk (A) rowi (B ) + a rows (B ) rowi (B ) if k = s if k = s if i = r if i = r

4.2 Elementary Matrices

235

(iv) Consider any a1 , . . . , an R and any x1 , . . . , xn Rm , then the multiplication the diagonal matrix of the aj is nothing but the scalar multiplication with these elements, formally dia(a1 , . . . , an ) x1 x1 . . . xn a1 x1 . . . an xn xn an

xn dia(a1 , . . . , an ) = x1 a1

(v) Multiplying Nn from the right shifts the rows up one step, multiplying with Nn from the left shifts the columns right one step. Formally put: consider any x1 , . . . , xn Rm then we get Nn x1 x2 x1 x2 . . . xn x2 . . . xn 0 xn1

xn Nn = 0 x1

(vi) If x1 , . . . , xn Rm again and : 1 . . . n 1 . . . n is an arbitrary function, then multiplying with the permutation matrix P from the left permutes the rows with , formally P x1 . . . xn x(1) . . . x(n)

Now suppose, that : 1 . . . n 1 . . . n even is bijective. Then multiplying with the permutation matrix P from the right permutes the columns with 1 , formally x1 x1 (m)

xn P = x1 (1)

(4.13) Corollary: (viz. 316) Let (R, +, ) be any ring and 1 n N, then we summarize our results concerning the inverse matrices of the various elementary matrices (i) The set of diagonal matrices is a R-subalgebra of matn (R), that is isomorphic to the R-algebra Rn . The isomorphism thereby is given by
dia (R) : (a , . . . , a ) dia(a , . . . , a ) Rn n 1 n 1 n

In particular the matrix dia(a1 , . . . , an ) is invertible if and only if all the ai are invertible in R, i.e. ai R . And in this case the inverse 236 4 Linear Algebra

matrix is given to be
1 1 dia(a1 , . . . , an )1 = dia(a 1 , . . . , an )

(ii) In particular for any i 1 . . . n and any a, b R we have the identity Di (a) Di (b) = Di (ab) And Di (a) is invertible if and only if a R is a unit in R. And in this case the inverse of Di (a) is given to be Di (a)1 = Di a1 (iii) Consider any i = j 1 . . . n and any a, b R. Then the matrix multiplication of Ti,j (a) and Ti,j (b) simply is Ti,j (a) Ti,j (b) = Ti,j (a + b) In particular any transposition matrix Ti,j (a) with i = j is invertible and the inverse matrix of Ti,j (a) is given to be Ti,j (a)1 = Ti,j (a) (iv) Let us denote the identity map by id : 1 . . . n 1 . . . n : i i. Then clearly the permutation matrix of id is the identity matrix Pid = 1 1n . And if and : 1 . . . n 1 . . . n are any two functions on 1 . . . n, then P P = P In particular the permutation matrix P is invertible, if and only if is bijective, and in this case the inverse is given to be
1 P = P1

4.2 Elementary Matrices

237

4.3

Linear Equations

A system of equations is a (nite) collection of equations for several variables x1 , . . . , xn over some ring R. That is we are given some functions fi : Rn R and some bi R (where i 1 . . . m) and consider the equations f1 (x1 , . . . , xn ) = b1 f2 (x1 , . . . , xn ) = b2 . . . . . . . . . fm (x1 , . . . , xn ) = bm A solution of this system of equations is an n-tuple (x1 , . . . , xn ) Rn that satises the equations above. The set of solutions us sometimes denoted by L = { x Rn | i 1 . . . m : fi (x) = bi } However the question remains how to solve such a system of equations. In this generality the problem is absolutely unsolvable. In fact very little can be said for sure: in the case R = R or R = C there is the local inversion theorem which (under certain conditions) guarantees the existence of a solution. To truly nd one there are a multitude of numerical methods (e.g. the newtonian algorithm ). If all the fi are polynomial and R is a(n algebraically closed) eld, then the vast theory of algebraic geometry studies the structure of the set of solutions. The one and only case in which a complete theory of solving a system of equations exists is the case of linear equations. Thereby the system is said to be linear, i all the fi are multi-linear, i.e. any xk fi (. . . , xk , . . . ) is R-linear. In this case the system of equations is of the form a1,1 x1 + a2,1 x1 + . . . ... ... +a1,n xn +a2,n xn . . . = b1 = b2 . = . .

am,1 x1 + . . .

+am,n xn = bm

for some xed coecients ai,k R and bi R (where i 1 . . . m and k 1 . . . n). Note that matrices provide a very compact notation for linear systems of equations. Just let A = (ai,k ) matn (R) and b = (b1 , . . . , bm ) Rm , then the above system boils down to s single equation for vectors (that is nite tuples over R) Ax = b Still a notation does not provide an algorithm of solving the equations. Now this is the purpose of this section and we start by giving an example: (1) 2x1 + 2x2 3x3 + 2x4 = 8 (2) 2x1 + 4x2 + 8x3 4x4 = 6 (3) 3x1 4x2 + x4 = 1 This is a (3 4)-system and we seek all 4-tuples (x1 , x2 , x3 , x4 ) satisfying all 3 equations above. The problem is that these equations are interlocked 238 4 Linear Algebra

with each other, so let us try to disentangle these equations. We see that subtracting equation (1) from equation (2) the coecient of x1 vanishes, so lets try (2) (1) (2). With equation (3) the case is not so simple, but it works out ne with 3(1) + 2(3) (3) (1) 2x1 + 2x2 3x3 + 2x4 = 8 (2) 2x2 + 11x3 6x4 = 2 (3) 2x2 9x3 + 8x4 = 26 So we have reduced the problem by the variable x1 at the cost of equation (1) - if we continue using (1), then x1 will reappear. So we restrict ourselves to (2) and (3). To eliminate x2 let us add (2) and (3), i.e. (3) + (2) (3) (1) 2x1 + 2x2 3x3 + 2x4 = 8 (2) 2x2 + 11x3 6x4 = 2 (3) 2x3 + 2x4 = 24 Let us suppose R = Q, then dividing by 2 we found x3 + x4 = 12 - this is a multitude of solutions, every x4 gives rise to another x3 . Let us denote s := x4 , then x3 = 12 s due to equation (3). But by equation (2) this now means 2x2 = 2 11(12 s)+6s = 134+17s and hence x2 = 67+17s/2. This can be inserted into equation (1) 2x1 = 8(134+17s)+3(12s)+2s = 178 22s and hence x1 = 89 11s. Altogether any solution x of this system of equations is of the form x1 89 x2 67 x3 12 x3 0 11 + Q 17/2 1 1

(4.14) Proposition: (viz. 359) Let (R, +, ) be an arbitrary ring, 1 m, n N and consider any matrix A matm,n (R) over R. Finally let x Rn and b Rm , then we obtain the following statements (i) Let us denote L(Ax = b) := { x Rn | Ax = b } the set of solutions of the linear system Ax = b of equations. If b im(A) then there is no solution L(Ax = b) = . Conversely if b im(A) then there is some special solution u Rn with Au = b. And the set of solutions is then given to be L(Ax = b) = u + kn(A) (ii) If S gln (R) and T glm (R) are any two invertible matrices, let us denote D := SAT matm,n (R). Thereby the following two statements are equivalent (a) Ax = b (b) there is some z Rn such that x = T z and Dz = Sb (iii) If we use the same notation as in (i) and (ii), then the set of solutions of Ax = b can be reformulated as L(Ax = b) = T L(Dx = Sb) 4.3 Linear Equations 239

(iv) Let us continue with (iii), if Sb im(D) then there is no solution L(Ax = b) = . Conversely if Sb im(D) then there is some u Rn with Du = Sb. And the set of solutions is then given to be L(Ax = b) = T u + T kn(D) This proposition provides a strategy for solving linear systems Ax = b of equations: nd invertible matrices S and T such that D = SAT becomes an easy matrix. Well, D is easy if the equation Dx = Sb can be solved easily. This of course is the case, if D is something like a diagonal matrix - one entry per column - as in this case we do not have an interlocked system of equations, but simply n seperate equations of the form ai xi = bi . So we will have to dene, when the matrix D is easy (this will be called the Gauss-form ) and we have to provide a way of nding S and T - this will be the Gauss-algorithm. Then we only have to solve Du = Sb (supposed such an u exists), as then L(Ax = b) = T u + T kn(D). (4.15) Denition: Let (R, +, ) be an arbitrary ring and 1 m, n N, then for any n-tuple a = (a1 , . . . , an ) Rn with a = 0 we denote (a) := min{ k 1 . . . n | ak = 0 } 1 . . . n and specically for a = 0 we let (a) := . That is (a) is the index of the rst non-zero entry of a. Analogously for an (m n)-matrix A matm,n R with A = (ai,k ) = 0 we dene (A) to be the index of the rst non-zero entry of any of its rows, that is (A) := min{ (rowi A) | i 1 . . . m : ak = 0 } = min{ k 1 . . . n | i 1 . . . m : ai,k = 0 }

and specically for A = 0 we let (A) := again. Thus if A = 0 we trivially have (A) 1 . . . n, as before. (4.16) Denition: Let (R, +, ) be an arbitrary ring and 1 m, n N again, then a matrix A matm,n (R) is said to be in echelon-form, i either A = 0 is zero, or A = 0 and there is some r 1 . . . m such that: (1) (row1 A) < < (rowr A) and (2) i N with r < i m we have rowi A = 0 That is the non-zero entries of A form a descending (from top left to bottom right) stair. In particular the zero-rows of A are at the bottom of A. Now A is said to be in Gauss-form, i A is in echelon-form and any non-zero row begins with a 1 R (not just any non-zero element of R). That is: let r N be as before, r := max{ i 1 . . . m | rowi A = 0 } and let us abbreviate k (i) := (rowi A) for any i 1 . . . r, then A is in Gauss-form, i also (3) i 1 . . . r we get ai,k(i) = 1 (4.17) Example: The following (3 4)-matrix features r = 2, (row1 A) = 1, (row2 A) = 3 240 4 Linear Algebra

and (row3 A) = and hence is in echelon-form. However it is not in Gauss-form, as ent2,3 A = 2 = 1 1 0 1 4 A = 0 0 2 0 0 0 0 0 The next example of a (4 4)-matrix is not in echelon-form, as the second column features a double-step. Formally that is (row2 B ) = 2 and (row2 B ) = 2 which contradicts the strict inequality (row2 B ) < (row2 B ) 1 0 B = 0 0 0 1 1 0 0 0 1 0 1 0 1 1

(4.18) Example: If a system of linear equations is in Gauss form, then it is truly possible to solve this system. As an example, let us consider the following system 0 0 0 0 1 0 0 0 2 3 1 4 0 0 0 0 1 5 7 5 x = 2 1 0 0

Written explicitly - where we denote x = (x1 , x2 , x3 , x4 , x5 ) as usual - this system abbreviates the following set of equations (1) x2 + 2x3 3x4 + x5 (2) x3 4x4 + 5x5 (3) x5 (4) 0 = = = = 5 7 2 0

First of all it is clear that equation (4) is a void condition, the set L of solutions is solely determined by equations (1) to (3). This would have been entirely dierent, if we had the equation 0 = 1, in this case we had L = , as there would be no specic solution u. Also the variable x1 does not occur in the system at all - due to col1 A = 0. Consequently the set L of solutions will be of the form R L for some subset L R4 . Let us now turn our attention to solving this system: In this case we already know x5 = 2 and this allows to deduce x3 = 7+4x4 5x5 = 3+4x4 from equation (2). This again can be used in equation (1) to obtain x2 = 5 2x3 + 3x4 x5 = 7 2(3 + 4x4 ) + 3x4 = 1 5x4 . Altogether the solutions x are of the form x1 0 1 0 1 5x4 1 0 5 = 3 + x1 0 + x4 4 3 + 4 x x = 4 x4 0 0 1 2 2 0 0 4.3 Linear Equations 241

So as could have been expected from (4.14) the set of solutions L is of the form L = u + K where u = (0, 1, 3, 0, 2) is a specic solution to the system and K is the kernel of A, here K is a (rank 2) submodule of R5 .

242

4 Linear Algebra

(4.19) Lemma: (viz. 359) Gaussian Algorithm Let (R, +, ) be a commutative ring, 1 m, n N and consider any (m n)matrix A = (ai,j ) matm,n (R) over R. First recall the denition of the elementary matrices (4.10), then proceed: There is an (m m)-matrix S matm (R) such that S is a product of elementary matrices S = S1 . . . S where Sk elmm R and E := S A matm,n (R) is in echelon-form. And this matrix S can be found using the following Gaussian-algorithm, that iterates on the number m of columns of A: (1) Initialisation: Pick up a working matrix E = (ei,j ) := A and begin with S := 1 1m . The algorithm starts with row number r := 1. (2) Termination: If the r-th row and all rows below that are zero (that is rowi (E ) = 0 for any i r . . . m), then E is in echelon form and hence we may terminate. Else we commence with step (3) (3) Targetting: Among the r-th to last row of E seek out the row with the rst non-zero entry, say this is the s-th row. To be precise we take u := min{ (rowv E ) | v r . . . m } s := min{ t r . . . m | (rowt E ) = u } (4) The Pivot: Swap the r-th and the s-th row of E , that is we move the row with the rst non-zero entry up to the top row of our current step S := Pr,s S E := Pr,s E (5) The Procedure: Now, for any i (r + 1) . . . m proceed as follows: if ei,u = 0 then ignore this i, else - if ei,u = 0 - replace S := Ti,r (ei,u ) Di (er,u ) S E := Ti,r (ei,u ) Di (er,u ) E (6) The Recursion: Supposed r < m consider the next row, that is replace r := r + 1 and iterate to step (2). If m = r, then terminate the algorithm - E is in echelon-form. Note that step (2) may also terminate the algorithm. In any case we nd E = SA the matrix in echelon-form. (4.20) Remark: (i) Let us explain the Gaussian algorithm a bit: the main step (5) is where the actual computations are done. If we decode the operations of the elementary matrices they actually become quite simple: In fact for any i (r + 1) . . . m the algorithm simply multiplies row number i with er,u and then subtracts ei,u times row number r, that is rowi E := er,u (rowi E ) ei,u (rowr E ) 4.3 Linear Equations 243

(ii) As the algorithm is iterated at most m times and the r-th iteration requires at most m r row-operations (as described in (i) above) the algorithm needs at most m(m 1)/2 row-operations in total. (iii) If R is a eld, then all the elementary matrices are invertible elmm R glm R and hence S is invertible, as a product of such matrices. (iv) If R is a eld, the matrix E = (ei,j ) in echelon form can easily be processed into a matrix G in Gauss-form: let r be the number of nonzero rows of E , that is r = # { i 1 . . . m | rowi E = 0 }. Further let k (i) := (rowi E ) 1 . . . n and ei := ei,k(i) and D := dia G := D E Then G is in Gauss-form, T = DS is a product of elementary matrices still, and G = T A. And as we will see in (v) this allows to solve the system of linear equations Ax = b. 1 1 , . . . , , 1, . . . , 1 q1 qr

(v) Suppose we are given a system of linear equations Ax = b over a eld R. We then use the Gaussian algorithm and (iv) to nd a matrix T glm R such that G = T A is in Gauss form. Then by (4.14) the set of solutions of Ax = b is given to be: L(Ax = b) = L(Gx = T b) Hence it remains to solve the equation Gx = T b. But as G = (gi,j ) is in Gauss form this is feasible: see the example above. For a more formal approach let us denote r = # { i 1 . . . m | rowi G = 0 } 0 . . . m, k (i) := (rowi G) 1 . . . n and gi := gi,k(i) again. Then we we can iteratively compute the solution beginning with i = r down to i = 1
n

xk(i) = [T b]i
j =k(i)+1

gi,j xj

Thereby all the variables { x1 , . . . , xn } \ xk(1) , . . . , xk(r) remain undetermined. That is any choice of these variables produces a solution, whereas the variables xk(1) , . . . , xk(r) are determined by these. In other words the kernel of G is (n r)-dimensional subspace of Rn and the set of solutions is of the form L(Ax = b) = u + kn(G) (vi) It might be helpful to reformulate the Gaussian algorithm in pseudocode, so thats what we want to do now: rst of all we will need the following variables: A = (ai,j ) and E = (ei,j ) matn,m R, S = (si,j ) matm R also i, r, s and u N and nally swap Rn . 244 4 Linear Algebra

S1 1m EA r1 while (r < m) and ( s r . . . m : rows E = 0) do begin u min{ (rowv E ) | v r . . . m } s min{ t r . . . m | (rowt E ) = u } if (s = r) then begin swap rowr E rowr E rows E rows E swap end-if for i := r + 1 to m step 1 do begin rowi S sr,u (rowi S ) si,u (rowr S ) rowi E er,u (rowi E ) ei,u (rowr E ) end-for r r+1 end-while (4.21) Example: We would like to enlighten things a little, by presenting an example of the workings of the Gaussian algorithm in the case of the following example 0 1 0 1 A = 0 2 0 1 2 3 4 7 4 6 1 3 1 6 2 0

First of all the Gauss-algorithm initializes E := A, S := 1 1m and starts with row r = 1 now, as A is non-zero the algorithm does not terminate right away. As the 1-st column of E has a minimal non-zero entry we get u = 2 and s = 1. Now as r = s there is no need to pivot (we have e2,1 = 1 = 0). In the central step, for i = 2, we get S = T2,1 (1) D2 (1) and multiply E with T2,1 (1) D2 (1) from the left. That is the 2-nd row is multiplied with 1 (which changes nothing) and then 1 times the 1-st row is added to the 2-nd row. Likewise for i = 3 we multiply with T3,1 (2) D3 (1) from the left again. That is the 3-rd row is multiplied with 1 (void again) and then 2 times the 1-st row is added to the 3-rd row. Finally for i = 4 we multiply with T4,1 (1) D4 (1) from the left, that is we add the 1-st row to the 4-th row. So far weve got S = T4,1 (1) D4 (1) T3,1 (2) D3 (1) T2,1 (1) D2 (1) and E = SA which amounts to 1 0 0 0 0 1 2 3 2 0 1 2 3 2 1 1 0 0 0 1 4 8 4 = 0 0 2 4 5 0 0 0 0 0 2 0 1 0 0 2 4 0 1 1 0 0 1 0 1 2 3 0 0 0 3 0 1 Now the Gauss-algorithm iterates with r = 2. The rst column with nonzero entries (from row 2 on) is s = 2 with u = 3, so again there is no need to pivot (e3,2 = 2 = 0). However for i = 3 we nd e3,2 = 0 and hence we continue with i = 4 already. Hereby we multiply with T4,2 (3) D4 (2) from the left which is: rst multiply row 4 with 2 and then add 3 times the second row to row 4. So S has risen to S = T4,2 (3) D4 (2) T4,1 (1) T3,1 (2) T2,1 (1) 4.3 Linear Equations 245

(omitting identity matrices) and this step yielded the matrices 1 0 0 1 0 0 0 3 0 0 1 0 0 0 0 0 0 0 0 2 1 0 0 0 2 3 2 4 0 0 3 0 0 2 0 5 = 0 0 0 1 1 0 0 0 2 3 2 2 4 5 0 0 0 0 12 13

As we have a zero-row row3 E = 0 the following step r = 3 will be the last one to take. But lets stick to the letter: here we get u = 4 and s = 4 so we have to swap rows 3 and 4 by multiplying P3,4 from the left. So by now we found the matrices 1 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 1 0 0 1 1 0 0 1 1 0 0 S= 0 0 0 1 0 0 1 0 2 0 1 0 = 5 3 0 2 2 0 1 0 1 0 0 1 0 3 0 2 0 0 1 0 1 0 1 1 E= 5 3 2 0 0 0 0 1 0 0 1 0 0 1 2 0 2 0 0 1 2 3 4 7 4 6 1 3 1 0 6 0 = 2 0 0 0 1 0 0 0 2 3 1 2 4 5 0 12 13 0 0 0

In step r = 4 we see that row4 E = 0 already, hence the termination condition (2) of the Gaussian algorithm is satised. And in fact, we see that E is in echelon-form already. Note that this does not represent the modied linear system of equations that we computed in the introductory example. The reason is that in the manual computation we have evaded the common multiple of 2 in the rst step (2) (1) (2). But the Gauss-algorithm in its general form does not check for common multiples. So we see that for UFDs there has to be a rened algorithm. LR-Zerlegung (4.22) Lemma: (viz. 247) Let (R, +, ) be a commutative ring, m, n N and A, B matmn R. Then the following statements hold true (i) For any 1 r m, n the r-minors are invariant under elementary eqivalence up to the sign, formally that is A B = minorr A = minorr B

where for M R we let M := { m | m M } { m | m M } (ii) If (R, ) even is an euclidean domain and m n, then there are scalar factors d1 , . . . , dm R that form a divisibly ascending chain, i.e. d1 | d2 , . . . , dm1 | dm such that A 0 d1 .. . 0 0 ... . . . 0 . . . 0

dm 0 . . .

And for any size r 1 . . . m we nd that the scalar factors d1 , . . . , dm 246 4 Linear Algebra

can be expressed in terms of minors of A d1 . . . dr gcd(minorr A) In particular the di are uniquely determined by A up to multiplication with units of R. (iii) If (R, ) even is an euclidean domain but n m then we can nd scalar factors d1 , . . . , dn R that form a divisibly ascending chain, i.e. d1 | d2 , . . . , dn1 | dn such that A d1 .. 0 0 . . . 0 . dn 0 . . . 0 0

... ...

And for any size r 1 . . . n we nd that the scalar factors d1 , . . . , dn can be expressed in terms of minors of A d1 . . . dr gcd(minorr A) In particular the dj are uniquely determined by A up to multiplication with units of R. Proof of (4.22): The proof of the rst statement is a tedious distinction of several cases, and the second is a non-algorithmical trick computation. However both are non-trivial and hence presented here. Now (iii) is clear, simply apply (ii) to the transpose matrix of A. (i) The proof will be conducted in several steps - the rst two steps are the technical part, the later parts will be applications of this that will yield the claim by induction We will rst prove that for any permutation matrix P matm R minorr (P A) minorr (A)

Thus choose any A(r, m) and A(r, n) and suppose P = P ( ). Then dene A(r, m) to be the index with the range (1 . . . r) = ((1 . . . r)) and dene Sr to be the permutation that satises = . Then it is clear that ent, P ( ) A = P ( ) ent , A And hence the minor of P A selected by (, ) is just the minor of A selected by ( , ) (up to the sign of P ( )). Now we will prove that for any transvection matrix T matm R minorr (T A) minorr (A) 4.3 Linear Equations 247

Thus choose any A(r, m) and A(r, n), then the cauchybinet-formula reads as det(ent, T A) =
A(r,m)

det(ent, T ) det(ent, A)

It is clear that the submatrix ent, T is a triangular matrix, since T was such. Thus the determinant only depends on the entries in the diagonal. In the case = all these diagonal entries are 1 and hence det(ent, T ) = 1. In the case = there is (at least one) 0 on the diagonal and hence det(ent, T ) = 0. Thus the sum over collapses to the case = and hence det(ent, T A) = det(ent, A) In other words the respective minor just remains unchanged. In the previous two steps we have seen that for any elementary matrix E matm R we get minorr (E A) minorr (A)

But with E also E 1 is an elementary matrix and A = E 1 E A. Therefore we can use this inclusion twice to obtain minorr (A) minorr (E A) minorr (A)

Thus we have seen that for any elementary matrix E matm R minorr (E A) = minorr (A) As the determinant is invariant under transposition of matrices we see that for any elementary matrix F matn R we also get minorr (A F ) = minorr (A) Thus if A and B are elementary equivalent by virtue of the matrices E1 , . . . , Er matm R and F1 , . . . , Fs matn R A B = E1 . . . Er A F1 . . . Fs Then we may easily use induction on the number r + s of elementary matrices taking the previous part as the induction step. (ii) We begin by proving the existence of such a representation of A by a divisibly ascending chain of scalar factors. This will be the major part of the proof, the representation of the scalar factors in terms of minors then is simple First of all we dene - for any matrix 0 = A = (ai,j ) matmn R (A) := min{ (ai,j ) | i 1 . . . m, j 1 . . . n, ai,j = 0 } Now we regard the set of all (B ) where B is a matrix that is 248 4 Linear Algebra

elementaryly eqivalent to A { (B ) | B matmn R, A B } N

As N is well-ordered this set has a minimum and we choose such a matrix M for which (M ) is minimal.

Suppose that the (i, j )-th entry mi,j of M takes the minimal value (M ) = (mi,j ). Then we may rearrange M to shift mi,j to the position (1, 1) and we denote this rearranged matrix by M := P (1 i) M P (1 j )

I.e. ent1,1 M = enti,j M = mi,j and hence (M ) = (ent1,1 M ). Thus we may equally well assume that (M ) = (m1,1 ) We now claim that m1,1 divides all the entries mi,1 in the rst column of M . To see this we use division with remainder mi,1 = pi m1,1 + ri where 0 (ri ) < (m1,1 )

Suppose ri = 0 then we would obtain a matrix B := Ti,1 (pi ) M being elementary equivalent to M and hence to A, too. Clearly the (i, 1)-entry of B is given to be [rowi B ]1 = [rowi M pi row1 M ]1 = mi,1 pi m1,1 = ri Therefore (B ) (ri ) < (m1,1 ) = (M ), but M was chosen to be minimal, so this cannot occur.

Analogously we see that m1,1 divides all the entries m1,j in the rst row of M . We use division with remainder again m1,j = qj m1,1 + si where 0 (sj ) < (m1,1 )

Suppose sj = 0 then we would obtain a matrix B := M T1,j (qj ) that is elementary equivalent to A again. Clearly the (1, j )-entry of B is then given to be [colj B ]1 = [colj M qj col1 M ]1 = m1,j qj m1,1 = sj Therefore (B ) (sj ) < (m1,1 ) = (M ), but M was chosen to be minimal, so this cannot occur.

We now use elementary transpositions to replace the entries of the rst column mi,1 ri = 0 and of the rst row m1,j sj = 0. And this obviously is accomplished by N := T2,1 (p2 ) . . . Tm,1 (pm ) M T1,2 (q2 ) . . . T1,n (qn ) 4.3 Linear Equations 249

As we have seen this matrix N now is of the following form N = m1,1 0 . . . 0 ... . . . . . . 0 ... 0 . . .

Let us abbreviate N = S M T where S and T denote the respective products of transposition matrices. Then we will have a look at the entries of N for enti,j N = [colj S M T ]i = [colj S M qj col1 S M ]i = enti,j S M qj enti,1 S M = mi,j pi m1,j = mi,j pi qj m1,1

We nally prove that m1,1 = n1,1 even divides any other entry ni,j = mi,j pi qj m1,1 (with i 2 . . . m and j 2 . . . n) of N as well. First we use division with remainder once more ni,j = hi,j n1,1 + ki,j where 0 (ki,j ) < (n1,1 )

Now regard the matrix N := Ti,1 (1) N , clearly the i-th row of N assumes the form rowi N = rowi N + row1 N = n1,1 ni,2 . . . ni,n

Thus dening N := N T1,j (hi,j ) yields another matrix with colj N . . .

= colj N hi,j col1 N = ki,j . . .

Hence (N ) (ki,j ) < (n1,1 ) = (M ). But N is elementarily equivalent to N and hence to A and as M was chosen to minimal among these matrices we nd ki,j = 0 again.

Thus we have decomposed N into a scalar block (m1,1 ) and a scalar multiple of a block A1 of the size (m 1) (n 1) N = (m1,1 ) (m1,1 A1 ) Thus we let d1 := m1,1 and recurse the entire process starting with A1 . Using induction we nd after m 1 steps that A truly is equivalent to a matrix D of the claimed form. 250 4 Linear Algebra

 Until now we have shown that A is elementarily equivalent to D A D = 0 d1 .. . 0 0 ... . . . 0 . . . 0

dm 0 . . .

As the determinant of a submatrix of D always is a product of the coecients of the submatrix (and hence of D) we nd { 0 } d(1) . . . d(r) | A(r, m) = minorr D

Since the di form a divisibly ascending chain it is clear that the greatest common divisor of these elements is just d1 . . . dr . But by (ii) the gcd of the minors is invariant under elementary equivalence and therefore d1 . . . dr gcd(minorr D) = gcd(minorr A) P

4.3 Linear Equations

251

4.4

Multilinear Forms

rem: Permutationsgruppe Sn rem: signum, als Zahl der Transpositionen, aus Zyklenzerlegung, durch Abzaehlen def: Multilinearform, Semi-Algebra unter punktweiser Verknuepfung, symmetrisch, alternierend pro: L symmetrisch, dann L(x(1) , . . . , x(n) ) = L(x1 , . . . , xn ) L alternierend, dann L(x(1) , . . . , x(n) ) = sgn( )L(x1 , . . . , xn ) pro: symmetrische MLF bilden Unter-Halb-Algebra aller MLF alternierende MLF bilden R-Untermodul aller MLF def: aufsteigende Indices pro: symmetrische Multilinearformen auf (Rm )n alternierende Multilinearformen auf (Rm )n Eindeutigkeit der alternierende Multilinearformen auf (Rn )n

4.5 4.6

Determinants Rank of Matrices

Beispiel: A = (ui vj ) ist Ax = x | v u, insbesondere rank(A) = 1

4.7

Canonical Forms

252

4 Linear Algebra

Chapter 5

Spectral Theory
5.1 Metric Spaces

metric convergence relation completeness completion linear metrics innite sums

5.2

Banach Spaces

normed space sub-multiplicative norm Banach space Banach algebra

5.3 5.4 5.5

Operatoralgebras Diagonalisation Spectral Theorem

253

Chapter 6

Structure Theorems
6.1 Associated Primes

Skripten von Dimitrios und Benjamin Matsumura Kapitel 6 (in Teil 2) Bourbaki (commutative algebra) IV.1, IV.2 Eisenbud 3.1, 32., 3.3, Aufgaben 3.1, 3.2, 3.3 Dummit, Foote: Kapitel 15.1: Aufgaben 29, 30, 31, 32, 33, 34 und 35 Kapitel 15.4: Aufgaben 30, 31, 32, 33, 34, 35, 36, 37, 38, 39 und 40 Kapitel 15.5: Aufgaben 25, 26, 28, 29 (bemerke assR ass(a) geht viel einfacher direkt). und 30

6.2 6.3 6.4

Primary Decomposition The Theorem of Pr ufer Length of Modules

254

Chapter 7

Polynomial Rings
7.1 Monomial Orders

We have already introduced the polynomial ring R[t] in one variable (over a commutative ring R) as an example in section 1.4. And we have also used this concept in the chapter on linear algebra (e.g. the characteristic polynomial of a linear mapping). In this chapter we will now introduce a powerful generalisation of this ring - the polynomial ring (also called group ring) R[A]. Thereby R is a commutative ring once more and A will be any commutative monoid. On the other hand polynomial rings are natural examples of graded algebras. So it may be advantegeous to rst study graded algebras as a general concept. This will be done in the subsequent section. There are a multitude of special cases of this concept which we will regard in this chapter, as well. The most important special case will be the polynomial ring R[t1 , . . . , tn ] in (nitely many) variables. The polynomials f R[t1 , . . . , tn ] are the working horses of ring theory, as they describe all the compositions (of additions and multiplications) that can be performed in a ring. This makes the study of the polynomial ring a task of vast importance and likewise gives the theory its punch. The polynomial ring R[t] is included into this general theory as the polynomial ring R[A] where A is chosen to be the the natural numbers A = N. Yet the algabraic structure of N is not the sole important property for R[t]. The natural order on N is of no lesser importance. Therefore we will rst study moniods A that carry an adequate order, as well. So rst recall the denitition of a linear order (also called total order ). If you are not familiar with this notion take a look at the introduction (section 0.2) or (re)read the beginning of section 2.1 for even more comments. So let uns now introduce the basic concepts of ordered monoids:

255

(7.1) Denition: Let (A, +) be a commutative monoid and denote its neutral element by 0. Then we introduce the following notions concerning A A is said to be integral i for any , , A we get the implication + = + = =

Further we A is said to be solution-nite i for any xed A the equation + = has nitely many solutions only, formally that is A : # (, ) A2 | + = <

(7.2) Denition: (i) The triple (A, +, ) is said to be a positively ordered monoid, i it satises both of the following two properties (1) (A, +) is a commutative monoid (with neutral element 0) (2) is a positive, linear order on A. That is is a linear order such that for any elements , and A we get the implication = + +

(ii) The triple (A, +, ) is said to be a strictly positively ordered monoid, i it satises both of the following two properties (1) (A, +) is a commutative monoid (with neutral element 0) (2) is a strictly positive, linear order on A. That is is a linear order such that for any elements , and A we get < = + < +

Note that thereby we used the convention a < b : a b and a = b. Thus it is clear that a strictly positively ordered monoid already is a positively ordered monoid (if = , then + = + is clear). (iii) Finally (A, +, ) is said to be an monomially ordered monoid and is said to be a monomial order on (A, +) i we get (1) (A, +, ) is a positively ordered monoid (2) is a well-ordering on A, that is i any non-empty subset M of A has a minimal element, formally this can be written as = M A M such that M we get (7.3) Remark: If now (A, +, ) is a positively ordered monoid, we extend A to A by formally picking up a new element (i.e. we introduce a new symbol to A) A := A { } 256 7 Polynomial Rings

We then extend the composition + and order of A to A dening the following operations and relations for any A () () =

() + := + () := Nota clearly (A, +) thereby is a commutative monoid again and if has been a positive or monomial order on A then the extended order on A is a positive resp. monimial order again. (7.4) Example: The standard example of a monoid that we will use later is Nn under the pointwise addition as the composition. However we will already discuss a slight generalisation of this monoid here. Let I = be any non-empty index set and let us dene the set NI := NI | # { i I | i = 0 } <

If I is nite, say I = 1 . . . n, then it is clear, that N(1...n) = Nn . We now turn this set into a commutative monoid by dening the composition + to be the pointwise addition, i.e. for any = (i ) and = (i ) we let (i ) + (i ) := (i + i ) And it is clear that this composition turns NI into a commutative monoid, that is integral and solution-nite (for any NI the set of solutions (, ) of + = is determined given to be { (, ) | i i }, which is nite, as NI ). For any i I we now dene i : I N : j i,j where i,j denotes the Kronecker symbol (e.g. refer to the notation and symbol list). From the construction it is clear, that any = (i ) NI is a (uniquely determined) nite sum of these i since (i ) =
iI

i i

This sum in truth is nite as only nitely many coecients i are non-zero and we can omit those i I with i = 0 from the sum. This enables us to introduce some notation here that will be put to good use later on. Let us rst dene the absolute value and norm of = (i ) NI || :=
iI

:= max{ i | i I }

For a xed weight NI we introduce the weighted sum of (note 7.1 Monomial Orders 257

that this is well-dened, as NI only has nitely many non-zero entries and hence the sum is nite only) || :=
i I

i i

Another oftenly useful notation (for any NI and any k N with k ) are the faculty and binomial coecients ! :=
i I

i ! k i

:=
i I

And if x = (xi ) RI is an n-tupel of elements of a commutative ring (R, +, ) we nally introduce the notations (note that these are well-dened again, as NI has only nitely many i = 0) x :=
i I

i xi
i x i

x :=
i I

(7.5) Proposition: (viz. 301) (i) Let A = be a non-empty set and A A be a linear order on A. Then any nite, non-empty subset = M A has uniquely determined minimal and a maximal elements. I.e. for any = M A with #M < we get ! M : M ! M : M

: :

(ii) Let A = be a non-empty set and A A be a well-ordering on A. Then any non-empty subset = M A has a uniquely determined minimal element. I.e. for any = M A we get ! M : M : Nota we will refer to these minimal and maximal elements of M by writing min M := and max M := respectively. (iii) Let (A, +) be a commutative monoid and let A A be a linear order on A. Then the following three statements are equivalent (a) A is integral and is a positive order on A (b) (A, +, ) is a strictly positively ordered monoid (c) for any , , A we get < + < + 258 7 Polynomial Rings

(iv) Let (A, +, ) be a positively ordered monoid, such that (A, +) is integral. Let further M, N A be any two subsets and M respecitvely N be minimal elements of M and N , i.e. M : and N :

Then for any elements M and N we obtain the implication + = + = = and =

(7.6) Example: (viz. 303) Let now (I, ) be a well-ordered set (i.e. I = is a non-empty set and is a linear well-orderering on I ) and x any weight NI . Then we introduce two dierent linear orders on the commutative monoid (NI , +). Of these the -graded lexicographic order will be the standard order we will employ in later sections. Thus consider = (i ) and = (i ) NI and dene Lexicographic Order lex : ( = ) or ()

() := k I such that k < k and i < k : i = i It will be proved below, that the lexicographic order lex is positive, i.e. the triple (NI , +, lex ) is a positively ordered monoid. In the case of I = 1 . . . n we will see that lex even is a well-ordering, i.e. the triple (Nn , +, lex ) is a monomially ordered monoid. -Graded Lexicographic Order : (|| < | | ) or (|| = | | and lex )

Analogously to the above the -graded lexicographic order is positive, i.e. the triple (NI , +, ) is a positively ordered monoid. And in the case I = 1 . . . n it even is a well-ordering, i.e. the triple (Nn , +, ) is a monomially ordered monoid once more.

7.1 Monomial Orders

259

(7.7) Remark: ( ) We wish to append two further linear orders on (NI , +). However we will not require these any further and hence abstain from proving any details about them. The interested reader is asked to refer to [Cox, Little, OShaea; Using Algebraic Geometry; ??] for more comments on these. Reverse Graded Lexicographic Order rgl : ( = ) or (|| < | |) or (|| = | | and (2))

(2) := k I such that k > k and i < k : i = i Just like the reverse graded lexicographic order rgl is positive and in the case of I = 1 . . . n it even is a well-ordering again. Bayer-Stillman Order For the fourt order we will require I = 1 . . . n already. Then we x any m 1 . . . n and obtain another monomial order on Nn bs(m) : (3) or ((4) and gl )

(3) := 1 + + m < 1 + + m (4) := 1 + + m = 1 + + m

260

7 Polynomial Rings

7.2

Graded Algebras

(7.8) Denition: Let (D, +) be a commutative monoid and (R, +, ) be a commutative ring. Then the ordered pair (A, deg) is said to be a D-graded R-(semi)algebra, i it satises all of the following properties (1) A is a commutative R-(semi)algebra (2) deg : hom(A) D is a function from some subset hom(A) A to D, where we require 0 hom(A). Thereby an element h A is said to be homogeneous, i h hom(A) { 0 }. (3) for any d D let us denote Ad := deg1 (d) { 0 } A. Then we require that Ad m A is an R-submodule of A. And we assume that A has a decomposition as an inner direct sum of submodules A =
dD

Ad

(4) given any two c, d D we require that the product gh of homogeneous elements g of degree c and h of degree d is a homogeneous element of degree c + d. That is letting Ac Ad := { gh | g Ac , h Ad } we assume Ac Ad (7.9) Remark: By property (2) Ad is a submodule of A, that is the sum g + h of two homogeneous elements of degree d again is a homogeneous element of degree d. And likewise the scalar multiple ah is a homogeneous element of degree d as well. Thus for any g , h hom(A) with deg(g ) = deg(h) and any a R such that ag + h = 0 we get deg(ag + h) = deg(g ) = deg(h) Property (4) asserts that the product of homogeneous elements is homogeneous again, in particular hom(A) { 0 } is closed under multiplication. And thereby the degree even is additive, that is for any two homogeneous elements g , h hom(A) we get gh = 0 = deg(gh) = deg(g ) + deg(h) Ac+d

7.2 Graded Algebras

261

 According to (3) any element f A has a unique decomposition in terms of homogeneous elements. To be precise, given any f A, there is a uniquely determined sequence of elements (fd ) (where d D) such that fd Ad (that is fd = 0 or deg(fd ) = d) the set { d D | fd = 0 } is nite and f = fd
dD

Thereby fd is called the homogeneous component of degree d of f . And writing f = d fd we will oftenly speak of the decomposition of f into homogeneous components. It is insatisfactory to always distinguish the cases h = 0 when considering a homogeneous element h hom(A) 0 A. Hence we will add a symbol to D as outlined in (7.3). Then the degree can be extended canonically to all homogeneous elements deg : hom(A) { 0 } D { } : h deg(h) if h = 0 if h = 0

If A is an integral domain this turns deg into a homorphism of semigroups, from hom(A) { 0 } (under the multiplication of A) to D. That is for any g , h hom(A) { 0 } we now get deg(gh) = deg(g ) + deg(h) In the literature it is customary to call the decomposition A = d Ad itself a D-graded R-algebra, supposed Ac Ad Ac+d . But this clearly is equivalent to the denition we gave here. Just let hom(A) :=
dD

Ad \ { 0 }

Then we obtain a well-dened function deg : hom(A) D by letting deg(h) := d where d D is (uniquely) determined by h Ad . And thereby it is clear that (A, deg) becomes a D-graded R-algebra, with deg1 (d) = Ad \ { 0 } again. (7.10) Example: If (R, +, ) is any commutative ring and (D, +) is any commutative monoid (with neutral element 0). Then there is a trivial construction turning R into a D-graded R-algebra. Just let hom(R) := R \ 0 and dene the degree by deg : hom(R) D : h 0. That is R0 = R, whereas Rd = 0 for any d = 0. Thus examples of graded algebras abound, but of course this construction wont produce any new insight. (7.11) Remark: Given some commutative ring (A, +, ) (and a commutative monoid (D, +)) there are two canonical ways how to regard A as an R-algebra (1) We could x R := A, in this case the submodules Ad m A occuring in property (3) would be ideals Ad i A. So this would yield the quite peculiar property Ac Ad Ac+d Ad = { 0 } (if c + d = d). 262 7 Polynomial Rings

(2) Secondly we could x R := Z, in this case the submodules Ad m A occuring in property (3) would only have to be subgroups Ad g A. Thus there no longer is a reference to the algebra structure of A. Hence we will also speak of a D-graded ring in this case. (7.12) Example: Rees Algebra () Let (R, +, ) be a commutative ring and a i R be an ideal of R. Note that for any n N the quotient an /an+1 becomes an R/a-module under the following, well-dened scalar multiplication (a + a)(f + an+1 ) := af + an+1 . Let us now take to the exterior direct sum of these R/a-modules R(a) :=
nN

an

an+1

Then R(a) can be turned into an R/a-algebra (the so called Rees algebra of R in a), by dening the following multiplication (fn + an+1 )(gn + an+1 ) :=
i+j =n

fi gj + an+1

Let us denote R(a)n := an /an+1 where we consider R(a)n as a subset R(a)n R(a) canonically (as in (3.22)). Further we dene the homogeneous elements of R(a) to be hom(R(a)) := n R(a)n \ { 0 }. Then (R(a), deg) becomes an N-graded R/a-algebra under the graduation deg : hom(R(a)) N : fn + an+1 n

(7.13) Denition: Let (D, +) and (E, +) be commutative monoids, (R, +, ) be a commutative ring, (A, deg) be a D-graded and (E, deg) be an E -graded R-algebra. Then the ordered pair (, ) is said to be a homorphism of graded algebras from (A, deg) to (B, deg) (or shortly a graded homorphism ) i it satises (1) : A B is a homorphism of R-algebras (2) : D E is a homorphism of monoids (3) h hom(A) we get deg((h)) = (deg(h))

7.2 Graded Algebras

263

(7.14) Remark: If (A, deg) is a D-graded R-algebra and a i A is a graded ideal of (A, deg), then (a, deg) clearly becomes a D-graded R-semi-algebra under the graduation inherited from (A, deg) hom(a) := hom(A) a deg := deg
hom(a)

If both (A < deg) and (B, deg) are D-graded R-algebras, then a homorphism : A B of R-algebras is said to be graded i (, 1 1) is a homorphism of graded algebras, that is i for any h hom(A) we get deg((h)) = deg(h) (7.15) Denition: (viz. 352) Let (D, +) be a commutative monoid, (R, +, ) be a commutative ring and (A, deg) be a D-graded R-algebra. Then a subset h A is said to be a graded ideal (or homogeneous ideal ) of (A, deg), i h i A is an ideal, that satises any one of the following three equivalent properties (a) h can be decomposed as an inner direct sum of the R-submodules h Ad
h =
dD

h Ad

(b) h is closed under going to homogeneous components. That if given any f A with homogeneous decomposition f = d fd we get f=
dD

fd h

d D : fd h

(c) h is generated by homogeneous elements, that is there is some set H of homogeneous elements of A generating h. Formally that is H hom(A) such that h = H
i

264

7 Polynomial Rings

(7.16) Proposition: (viz. 353) Let (D, +) be a commutative monoid (with neutral element 0), (R, +, ) be a commutative ring and (A, deg) be a D-graded R-algebra. Then the following statements are true (i) If D is an integral monoid, then 1 A0 (ii) If (B, deg) is another D-graded R-algebra and : A B is a graded homorphism, then the kernel of is a graded ideal in A kn() =
dD

kn() Ad

(iii) If a i A is a graded ideal of (A, deg), then the quotient A/a becomed a D-graded R-algebra again under the induced graduation hom A a := { h + a | h hom(A) \ a }

where deg(h + a) := deg(h). And thereby the set of homogeneous elements of degree d D is precisely given to be the following A
a

= Ad + a a

(iv) Let (D, +) even be a commutative group and consider a multiplicatively closed subset U A such that U hom(A). Then the localisation U 1 A (as a ring) is a D-graded R-algebra again. Thereby hom U 1 A := h u h hom(A), u U \ 0 1

Then the graduation is dened to be deg(h/u) := deg(h) deg(u) and we nd the modules of homogeneous elements (for any d D) U 1 A = h u u U, h Ad+deg(u)

(v) Suppose D = N and consider some ideal a i A. If now a is a graded ideal, then its radical a i A is a graded ideal, too.

7.2 Graded Algebras

265

(7.17) Remark: The above statement (iv) has an easy generalisation to the case where (D, +) only is an integral, commutative monoid. In this case U 1 A becomes a grDgraded R-algebra, where grD denotes the extension of D to a group. Formally that is grD := (D D)/ where (c, d) (c , d ) : c + d = c + d. Then grD becomes a well-dened group under (c, d)+(c , d ) := (c + c , d + d ) with neutral element (0, 0) and (c, d) = (d, c). And D is included canonically into grD by virtue of D grD : c (c, 0). (7.18) Denition: Let (R, +, ) be a commutative ring, (D, +, ) be a positively ordered monoid and (A, deg) be a D-graded R-algebra. If now f A with f = 0 has the homogeneous decomposition f = d fd (where fd Ad ) then we dene deg(f ) := max{ d D | fd = 0 } ord(f ) := min{ d D | fd = 0 } Note that this is well-dened, as only nitely many fd are non-zero and is a linear order on D. Thus we have dened two funtions on A deg : A \ { 0 } D ord : A \ { 0 } D

(7.19) Proposition: (viz. 357) Let (R, +, ) be a commutative ring, (D, +, ) be a positively ordered monoid and (A, deg) be a D-graded R-algebra. Then we obtain (i) The newly introduced function deg : A \ { 0 } D is an extension of the degree original function deg : hom(A) D. Formally that is h Ad , h = 0 = deg(h) = d

(ii) It is clear that the order and degree of some homogeneous element h hom(A) will agree. But even the converse is true, that is for any non-zero element f \ { 0 } we get the equivalency f hom(A) ord(f ) = deg(f )

(iii) For any elements f , g A such that f g = 0 we obtain the estimates ord(f ) deg(f ) deg(f g ) deg(f ) + deg(g ) ord(f g ) ord(f ) + ord(g ) (iv) And if (D, +, ) is a strictly positively ordered monoid and A is an integral domain, then for any f , g A \ { 0 } we even get the equalities deg(f g ) = deg(f ) + deg(g ) ord(f g ) = ord(f ) + ord(g ) 266 7 Polynomial Rings

(v) Thus if (D, +, ) is a strictly positively ordered monoid and A = 0 is a non-zero integral domain again, then the invertible elements of A already are homogeneous A hom(A)

(vi) If (D, +, ) is a strictly positively ordered monoid, A is an integral domain and h hom(A) is a homogeneous element of A, then we get f | h = f hom(A)

7.2 Graded Algebras

267

7.3 7.4 7.5 7.6 7.7 7.8 7.9

Dening Polynomials The Standard Cases Roots of Polynomials Derivation of Polynomials Algebraic Properties Gr obner Bases Algorithms

268

7 Polynomial Rings

Chapter 8

Polynomials in One Variable

8.1 8.2 8.3 8.4 8.5 8.6 8.7 8.8 Interpolation Irreducibility Tests Symmetric Polynomials The Resultant The Discriminant Polynomials of Low Degree Polynomials of High Degree Fundamental Theorem of Algebra

269

Chapter 9

Group Theory
normal subgroups, kernel+image, isomorphism theorems, quotient groups, group actions, conjugation, sylow theorems, p-q theorem, ordered groups, representation theory, invariant theory (9.1) Denition: Let (G, ) be a group and X be an arbitrary set, then a map : G X X (written as : (g, x) gx) is said to be an operation of G on X , i it satises one of the following two equivalent conditions (a) G SX : g Tg where Tg : X X : x gx is a group-homorphism (b) g , h G and x G we get the properties ex = x and (gh)x = g (hx) And in this case we dene the following notions: for any x X we dene the orbit Gx X and the stabilizer Gx G. And for any g G we also dene the xed point set X g X and the set X G of invariants Gx := { gx | g G } Gx := { g G | gx = x } GX X X
g G

:= { g G | x X : gx = x } := { x X | gx = x } := { x X | g G : gx = x }

Finally a subset A X is said to be G-invariant i for any g G we get gA := { gx | x A } A

(9.2) Example: (i) If X = is any non-empty set then any subgroup S g SX of the permutation group SX operates on X in a most obvious way S X X : (, x) (x) (ii) By (1.12.(ii)) any group G operates on itself, by multiplication from the left, that is we get an operation on G by G G G : (g, h) gh 270

(iii) Finally any group G operates on itself by conjugation, that is we obtain another action of G on G by letting G G G : (g, h) hg := ghg 1 Prob rst of all it is clear that for any x G we have xe = exe1 = x. Also for any g , h G we can compute xgh = (gh)x(gh)1 = ghxh1 g 1 = g (xh )g 1 = (xh )g , so that both properties are satised. (9.3) Proposition: (viz. 300) (i) It is easy to see that an operation G X X denes an equivalency relation on X by letting x y : g G : y = gx. And the equivalency classes of this relation clearly are the orbits Gx. The quotient set of X under this relation is denoted by X G := { Gx | x X }

(ii) It also is immedeately clear that the stabilizer Gx g G is a subgroup of G. Another well-known fact is the following 1-to-1 correspondence Gx G Gx : gx gGx (iii) If A X is G-invariant, then for any g G we have gA = A. And we can thereby restrict to an operation of G on A, that is : G A A : (g, x) gx (iv) If A is G-invariant, then the complement X \ A is G-invariant, too. (v) Orbit Formulas: If (G, ) is a nite group, that operates on the nite set X , then (for any x X ) the following equations are true #Gx = #G # Gx
GxX/G

#X = # G

1 #Gx

(vi) Formula of Burnside: If (G, ) is a nite group, that operates on the nite set X , then the following equation holds true #G #X G =
g G

#X g =
x X

# Gx

(9.4) Denition: Let G be a group and F be any eld. Then a group-homorphism : G F from G to the multiplicative group F is called a character of G. (9.5) Proposition: (viz. 273) Let G be any group and F be any eld. If now 1 , . . . , r : G F are 271

pairwise characters of G, then the set { 1 , . . . , r } F (G, F) already is F-linear independant (as F-valued functions). (9.6) Denition: Let G be any group and R be a commutative ring. Then : G R R is said to be an algebraic operation, i we get the properties (1) : G R R is an operation of G on R (2) for any g G the map Tg : R R : a ga is a homomorphism And in this case we dene the invariant ring under the operation to be RG := { a R | g G : ga = g }
1 = T Nota as Tg is bijective with Tg g 1 we nd that (2) already implies that any Tg even is an automorphism of the ring R. And it is immediately clear that the invariant ring RG r R is a subring of R.

(9.7) Proposition: (viz. 275) Let G be a group, R be a commutative ring and A be a commutative monoid. If now : G R R is an algebraic operation, then this canonically induces an algebraic operation G R[A] R[A] on the group ring R[A], by G R[A] R[A] : (g, f )
A

(gf [])t

And for this operation we obtain the following identity of the invariant rings R[A]G = RG [A]

(9.8) Proposition: (viz. 273) Let G R R : a ga be an algebraic operation of the group G on the commutative ring R. Let now a i R be any ideal, then we get (i) aG := { a a | g G : ga = a } = a RG i RG (ii) If a is an G-invariant ideal (that is g a a for any g G) then we obtain an algebraic operation of G on the quotient ring, via G R a R a : (g, b + a) gb + a (iii) And in case a is an G-invariant ideal, then we obtain a well-dened, injective homorphism of rings by virtue of RG
aG

: b + aG b + a

(9.9) Proposition: (viz. 274) Let G X X : (g, x) gx be a group operation and let R be a commutative ring. Further consider A := F (X, R) the R-algebra of functions from 272 9 Group Theory

X to R. Then we obtain an algebraic operation of G on A by letting G A A : (g, f ) x f (g 1 x) And if M X is a G-invariant subset (that is gM M for any g G), then I(M ) := { f A | x M : f (x) = 0 } is a G-invariant ideal (that is I(M ) i A is an ideal and g I(M ) I(M ) for any g G). And in this case we nally we obtain the isomorphy AG
I(M )G

I(M )

: f + I(M )G f + I(M )

Proof of (9.5): If r = 1 then the linear independance is clear, as 1 = 0. So pick up the minimal number k 2 such that there are a1 . . . , ak F and (pairwise distinct) i1 , . . . , ik 1 . . . r satisfying a1 i1 + + ak ik = 0 renumbering the i we may assume i1 = 1 and i2 = 2 and so on until ik = k . Let us now denote bi := ai /ak then we obtain the central equation
k 1

bi i + k = 0
i=1

As 1 = k (remember k 2) there is some h G such that 1 (h) = k (h). For any g G let us now insert gh into the central equation, this yields
k 1

bi i (g )i (h) + k (g )k (h) = 0
i=1

On the other hand (for any g G we may rst insert g into the central equation and afterwards multiply with k (h). This yields another equation
k 1

bi i (g )k (h) + k (g )k (h) = 0
i=1

Subtraction these two equations we obtain another identity (for any g G)

k1

bi k (h) k (h) i (g ) = 0
i=1

As any ci := bi (k (h) k (h)) F is scalar we have found a linear dependance with fewer that k elements. As k has been minimal this can only be if ci = 0 for any i 1 . . . (k 1). But by construction we have c1 = 0, a contradiction. Thus k could not have be chosen in the rst place and this is just the linear independance of the i . P

273

Proof of (9.8): (i) From the denition of aG it is clear that aG = a RG . Yet R : RG is a ring extension and hence it is a general fact that a RG i RG . (ii) We rst prove the well-denedness of the operation: consider b, b R with b + a = b + a. That is b b a and hence (as a is G-invariant) gb gb = g (b b) g a a. And this again yields gb + a = gb + a. Now as b gb + a is nothing but the composition of b gb gb + a we also nd that b + a gb + a is a homorphism. That is the operation even is algebraic. (iii) Let us rst consider RG (R/a)G : b b + a. This clearly is a well-dned homorphism as g (b + a) = gb + a = b + a. And the kernel of this homorphism clearly is b RG | b a = a RG = aG . Hence the map in the claim is well-dened and injective. P Proof of (9.9): As ef (x) = f (e1 x) = f (x) we clearly have ef = f for any f A. And given g , h G, f A and x X it is straightforward to compute (h(gf ))(x) = (gf )(h1 x) = f (g 1 h1 x) = f ((gh)1 x) = ((gh)f )(x) Thus G A A is a group action. And the operations of A are pointwise it is clear that this operation also is algebraic. Next we prove that I(M ) is a G-invariant ideal. The properties of an ideal are immediatley clear. Let us consider f I(M ), g G and any x M . As M is G-invariant we get g 1 x g 1 M M . And as f I(M ) this yields (gf )(x) = f (g 1 x) = 0. As x has been arbitary, this proves gf I(M ), which also yields the Ginvariance g I(M ) I(M ). It remains to prove AG I(M )G

I(M )

: f + I(M )G f + I(M )

We have already proved the well-denedness and injectivity of this map in (9.9). For the surjectivity we are given some f + I(M ) which is G-invariant. That is gf + I(M ) = f + I(M ) for any g G. In other words that is f (g 1 x) = (gf )(x) = f (x) for any g G. Let us denote 1M : X R the characteristic function of M and let f := 1M f . If x M then we clearly have f (x) = 1M (x)f (x) = 0. And as X \ M is G-invariant as well, we get 1M (g 1 x) = 0 and hence (g f )(x) = f (g 1 x) = 1M (g 1 x)f (x) = 0, too. Thus f (x) = 0 = g f (x) for any x X \ M . Conversely if x M then 1M (x) = 1 = 1M (g 1 x) which yields g f (x) = f (g 1 x) = 1M (g 1 x)f (g 1 x) = f (g 1 x) = (gf )(x) = f (x) = 1M (x)f (x) = f (x) as x M . Altogether we have proved f = g f , (for any x X ) which is the Ginvariance of f . And we have already seen that f (x) = f (x) for any x M . But we have also seen, that f + I(M ) = f + I(M ), which means nothing but f + I(M )G f + I(M ) under the map in question. P 274 9 Group Theory

Proof of (9.7): We will rst prove that G R[A] R[A] truly is a group action. To do this consider any g and h G and any f R[A]. By denition we clearly have ef = f . And further we nd g (hf ) = g
A

(hf [])t

=
A

g (hf [])t =
A

(gh)f []t = (gh)f

From the homorphism property of R R : a ga we nd that (g, f ) gf truly is an algebraic operation (we omit tedious computations). So it only remains to prove R[A]G = RG [A]. The inclusion is trivial. Conversely consider any f R[A]G . Then we have to show that for any A and any g G we get gf [] = f []. Yet to see this it suces to compare coecients (gf [])t = gf = f =
A A

f []t P

275

Chapter 10

Multilinear Algebra
10.1 10.2 10.3 10.4 10.5 10.6 Multilinear Maps Duality Theory Tensor Product of Modules Tensor Product of Algebras Tensor Product of Maps Dierentials

Matsumura - commutative ring theory - chapter 9

276

Chapter 11

Categorial Algebra
11.1 Sets and Classes

App: Cantors denition of sets, App: Russels antinomy, Rem: formalisation - the language Lset , Def: axiomatisation of set theory (ZFC) and consequences (the foundation axiom, the axiom of choice), Lem: AC Zorn well ordering theorem, Def: Peanos axioms, App: construction of N, Def: axioms of the arithmetic, Def: construction Z, Q and R, App: Heuristical denition of classes (2-nd level sets), Exp: the universe of sets, Rem: formalisation - the language Lclass , Def: axioms of classes (NBG)

11.2

Categories and Functors

denition of categories,(full) subcategories, small categories, ordinary categories, monomorphisms, epimorphisms, isomorphisms, denition of functors Rem: covariant, contravariant via opposite category, composition of functors, isofunctors, representable functors, Yonedas lemma, natural equivalence, equivalence of categories (11.1) Denition: Let C be any category, then an object U obj(C ) is said to be universal, i for any other object X obj(C ) there is a uniquely determined morphism greekph from U to X . Formally i X obj(C ) : ! C (U, X )

(11.2) Lemma: (viz. 277) If C is an arbitrary category, than any two universal objects U and V obj(C ) are isomorphic U = V. Proof of (11.2): As U is universal there is a unique morphism : U V , likewise as V is universal there is a unique morphism : V U . Hence : U U is a morphism and as 1 1U : U U is another morphism the uniquess (U is universal) implies = 1 1U . Likewise we nd = 1 1V and hence : U V is an isomorphism, with inverse morphism : V U . 277

11.3

Examples

categories: Set, Top, Ring, Dom, Mod(R), Loc, CS , C , Bun(), opposite category, functor categroy, functors: /a, U 1 , hom, spec

11.4

Products

denition of (co)products, examples: , , , bered product, injective and projective objects, denition of (co)limits

11.5

Abelian Categories

additive categories and their properties, additive functors, examples, kernel, cokernel and their properties, examples, canonical decomposition of morphisms, denition of abelian categories, left/right-exactness, examples: /a, U 1 , hom

11.6

Homology

chain (co)complexes, (co)homology modules, (co)homology morphisms, examples

278

11 Categorial Algebra

Chapter 12

Ring Extensions
basics, transfer of ideals, integral extensions, dimension theory, [Eb, exercise 9.6] gibt Beispiel f ur -dimensionalen noetherschen Ring

279

Chapter 13

Galois Theory

280

Chapter 14

Graded Rings
graduations, homogeneous ideals, Hilbert-Samuel polynomial, ltrations, completions

281

Chapter 15

Valuations
siehe Bourbaki (commutative algebra) VI, mein Skriptum

282

Part II

The Proofs

Chapter 16

Proofs - Fundamentals
In this chapter we will nally present the proofs of all the statements (propositions, lemmas, theorems and corollaries) that have been given in the previous part. Note that the order of proofs is dierent from the order in which we gave the statements, though we have tried to stick to the order of the statements as tightly as possible. So we begin with proving the statements in section 1.1. In this section however we have promised to ll in a gap rst: we still have to present a formal notion of applying brackets to a product. And also we have to establish a simple fact from elementary number theory: Division with Remainder: Consider any d N with d 1 which we call the divisor. Then for any x Z there are uniquely determined q , r Z such that (1) x = qd + r (2) r 0 . . . (d 1) This result will be largely generalized in section 2.6 and proved in greater generality, but as we use this fact for Z we have to give a proof beforehand. The numbers q and r in the statement above can even be given explictly q = x div d := max{ k Z | kd x } d = x mod d := x (x div d) d Prob Existence: as x |x| |x| d then set { k Z | kd x } is bounded from above by |x|. Hence the maximum q := max{ k Z | kd x } is well dened. Also take r := x qd Z. Then (1) x = qd + r is clear from the denition of r and as qd x we also have r 0. Now suppose r d then we would have 0 r d = x qd d = x (q + 1)d such that (q + 1)d x. But as q has been maximal, this is absurd, which only leaves r < d, such that (2) is established, as well. Uniqueness : suppose x = qd + r and x = ud + v for some q , r, u and v Z with r and v 0 . . . (d1). Then 0 = xx = qd+r (ud+v ) = (q u)d+r v . That is d divides (q u)d = v r. But as 0 v , r < d we have |v r| < d and hence v r can only be divided by d, if v r = 0. That is r = v and therefore qd + r = ud + v implies qd = ud which also yields q = u. P 284

Formal Bracketing () If we multiply (or sum up) several elements x1 , . . . , xn G of a groupoid (G, ) weve got (n 1) possibilities, with which two elements we start even if we keep their ordering. And we have to keep choosing until we nally got one element of G. In this light it is quite clear, what we mean by saying any way of applying parentheses to the product of the xi results in the same element. It just says, that the result actually doesnt depend on our choices. However - as we do math - we wish to give a formal way of putting this. To do this rst dene a couple mappings for 2 n N and k 2 . . . (n 2) b(n,1) : Gn Gn1 : (x1 , . . . , xn ) (x1 x2 , x3 , . . . , xn ) b(n,k) : Gn Gn1 : (x1 , . . . , xn ) (x1 , . . . , xk xk+1 , . . . , xn ) b(n,n1) : Gn Gn1 : (x1 , . . . , xn ) (x1 , . . . , nn2 , xn1 xn ) Now we can determine the sequel of choosing the parentheses simply by selecting an (n 1)-tupel k in the following set Dn := { k = (k1 , . . . , kn1 ) | ki 1 . . . (n i) for i 1 . . . (n 1) } For xed k Dn the bracketing (by k ) now simply is the following mapping Bk := b(2,kn1 ) b(n,k1 ) : Gn G

Proof of (1.2) (associativity): () We now wish to prove the fact that in a groupoid (G, ) the product of the elements x1 , . . . , xn is independent of the order in which we applied parentheses to it. Formally this can now be put as: for any k Dn we get Bk (x1 , . . . , xn ) = x1 x2 . . . xn We will prove this statement by induction on the number n of elements multiplied. In the cases n = 1 and n = 2 there only is one possibility how to apply parentheses and hence the statement is trivial. The case n = 3 there are two possibilities, which are equal due to the associativity law (A) of (G, ). Thus we assume n > 3 and regard two ways of applying parentheses k, l Dn , writing them in terms of the last multiplication used a := Bk (x1 , . . . , xn ) = Bp (x1 , . . . , xi )Bp (xi+1 , . . . , xn ) b := Bl (x1 , . . . , xn ) = Bq (x1 , . . . , xj )Bq (xj +1 , . . . , xn ) where i := k1 , j := l1 and p := (k2 , . . . , kn1 ), q := (l2 , . . . , ln1 ). We may assume i j and if i = j we are already done because of the induction hypothesis. Thus we have i < j and by the induction hypothesis the parentheses may be rearranged to a = b = (x1 . . . xi ) (xi+1 . . . xj )(xj +1 . . . xn ) (x1 . . . xi )(xi+1 . . . xj ) (xj +1 . . . xn ) 285

Now the associativity law again implies a = b and as k and l were arbitrary this implies a = x1 x2 . . . xn for any bracketing k chosen. P Proof of (1.5): Let (G, ) be any group, then ee = e = ee and hence e = e1 by denition of the inverse element. Likewise if y = x1 then xy = e = yx and hence x = y 1 . Then we prove (xy )1 = y 1 x1 by (xy )(y 1 x1 ) = x y (y 1 x1 ) = x ex1 = x (yy 1 )x1

= xx1 = e = y 1 (x1 x)y

(y 1 x1 )(xy ) = y 1 x1 (xy ) = y 1 ey

= y 1 y = e

Next we will prove (xk )1 = (x1 )k by induction on k 0. If k = 0 then the claim is satised, as e1 = e. Now we conduct the induction step, using what we have just proved: xk+1
1

xk x

= x1 xk

= x1 x1

x1

k+1

Now assume that xy = yx do commute. Then it is easy to see that x1 y 1 = y 1 x1 = (xy )1 do commute as well, just compute y 1 x1 = (xy )1 = (yx)1 = x1 y 1 And by induction on k 0 it also is clear, that xk y = yxk . Now we will prove (xy )k = xk y k by induction on k . In the case k = 0 everything is clear (xy )0 = e = ee = x0 y 0 . Now compute
k+1 k

xy

xy

xy

xk y k

yx

= xk y k (yx) = xk xy k+1

= xk y k+1 x = xk+1 y k+1

And for negative k we regard k where k 0, then we easily compute

k

xy

= =

(xy )1 x1
k

= y 1
k

x1 y 1

= xk y k

Note: in the last step of the above we have used (x1 )k = xk . But we still have to prove this equality: as for any x G we nd that xx1 = e = x1 x do commute we also obtain xk = (xk )1 = (x1 )k 286 16 Proofs - Fundamentals

for any k 0, just compute xk xk = x1

k

xk =
k

x1 x xx1

= ek = e = ek = e

xk xk = xk x1

This also allows us to prove (xk )l = xkl for any k , l Z. We will distinguish four cases - to do this assume k , l 0 then xk xk xk xk
l l l

= xk xk . . . xk (l times) = xkl (x1 )k

l

= =

=
l

x1

kl

= x(kl) = x(k)l
l

(xk )1

=
l

(x1 )k

= x(kl) = xk(l)
l

= ((x1 )k )1

= ((xk )1 )1

= xkl = x(k)(l)

It remains to prove xk xl = xk+l for any k , l Z. As above we will distinguish four cases, that is we regard k , l 0. In this case xk xl = xk+l is clear by denition. Further we can compute xk xl = x1
k

x1

x1

k +l

= x(k+l) = x(k)+(l)

For the remaining two cases we rst show x1 xl = xl1 . In case l 0 this is immediately clear. And in case l 0 we regard l with l 0 x1 xl = x1 x1
l

x1

l+1

= x(l+1) = x(l)1

We are now able to prove xk xl = x(k)+l by induction on k . The case k = 0 is clear, for the induction step we compute x(k+1) xl = (x1 )k x1 xl = (x1 )k x1 xl = xk xl1 = x(k+1)+l The fourth case nally is xk xl = xk+(l) , but as xk and xl commute (use inductions on k and l) this readily follows the above. P

Proof of (1.7): The rst statement that needs to be veried is that a subgroup P g G truly invokes an equivalence relation x y y 1 x P . This relation clearly is reexive: x x due to x1 x = e P . And it also is transitive because if x y and y z then y 1 x P and z 1 y P . Thereby z 1 x = z 1 (yy 1 )x = (z 1 y )(y 1 x) P too, which again means x z . Finally it is symmetric as well: if x y then y 1 x P and hence x1 y = (y 1 x) P 287

too, which is y x. Next we wish to prove that truly [x] = xP [x] = { y G | y x } = y G | x1 y = p P

= { y G | p P : y = xp } = { xp | p P } = xP So it only remains to prove the theorem of Lagrange. To do this we x a system P of representatives of G/P . That is we x a subset P G such that P G/P : q qP (which is possible due to the axiom of choice). Then we obtain a bijection by letting G G
1 P P : x xP, q x where xP = qP

This map is well dened: as xP G/P there is a uniquely determined q P such that qP = xP . And for this we get q x which means q 1 x P . And it is injective: if xP = yP = qP then q 1 x = q 1 y yields x = y (by multiplication with q ). Finally it also is surjective: given (qP, p) we let x := qp qP and thereby obtain q x and hence (xP, q 1 x) = (qP, p). P Proof of (1.8): We rst have to prove, that X o is a well-dened submonoid of G. As X G and G is a submonoid, the intersection runs over a non-empty set and hence is well-dened. Also, as e Xe , we clearly have e X o . And if x and y X o then for any P o G with X P we have x, y P . But as P is a submonoid this implies xy P again and as this is true for any P we also get xy X o again. Likewise if G is a group, then (by the same arguments as above) X g is a submonoid of G. It remains to prove that it even is a subgroup: But as P is a subgroup this implies x1 P again and as this is true for any P we also get x1 X o again. Next we will prove the explict representation of X o , to to this let us denote the set of nite products of elements of Xe by Q := { x1 x2 . . . xn | n N, x1 , . . . , xn Xe } As e Xe it is clear that e Q. And if x = x1 . . . xm and y = y1 . . . yn Q then by construction we also have xy Q. That is: Q is a submonoid of G and as X Q is clear we have X o Q. Conversely, as X o is a monoid with X X o we have e and x X o for any x X . That is Xe X o . Therefore, if x1 , . . . , xn Xe then x1 . . . xn X o again and this also is Q X o . Together we have Q = X o . We nally have to prove the explicit representation of X g , to do this let us abbreviate the set of nite products of elements of X by Q := { x1 x2 . . . xn | n N, x1 , . . . , xn X } If x X we also have x1 X and as (x1 )1 = x we nd that x1 X for any x X . That is X is closed under the inversion of elements. Also e X is trivial. And if x = x1 . . . xm and y = y1 . . . yn Q then 1 1 by construction we also have xy Q. But also x1 = x m . . . x1 Q, as 288 16 Proofs - Fundamentals

1 with xi X we also have x X . Hence Q is a subgroup of G and i as X Q is clear we have X g Q. Conversely, as X g is a group with X X g we have e, x and x1 X g for any x X . That is X X g . Therefore, if x1 , . . . , xn X then x1 . . . xn X g again and this also is Q X g . Together we have Q = X g .

P Proof of (1.9): It is clear that Q := xk | k Z is a subgroup of G, as for any i, j Z we have xi xj = xi+j Q and (xi )1 = xi Q. Also, as x Q we get x g Q. But as x x g and x g is a group we clearly have xk x g for any k Z and hence Q x g as well. Altogether x g = Q. Now suppose n := #G N is nite, then Q G is nite, too, say k := #Q. If now i, j N with i < j then (by multiplication with xi from the left) we clearly have xi = xj xj i = e

Thus if we had xp = e for any p N then xi = xj for any i < j and hence x g had innitely many elements. As this is untrue, there has to be some p N with 1 p such that xp = e. Let m N be the minimal such number m := min{ 1 p N | xp = e } As xm = e is the rst occurance of e we have xi = e for any 1 i < m. And this implies xi = xj for any i, j N, with i = j and i, j < m. Hence we have found m distinct elements in x g , namely xi | i 0 . . . (m 1) x
g

Conversely consider any xq x g and write q = pm + i by division of q by m with remainder i 0 . . . (m 1). Then xq = xpm+i = (xm )p xi = ep xi = xi . That is x g only consists of elements of the form xi for some 0 i < m and hence the above sets even are equal. In particular we have m = k . And as x g is a subgroup of G the theorem of Lagrange (1.7) truly implies k | n. P Proof of (1.10): If and : X X then the composition : x ( (x)) is another map of the form : X X . And if both and are bijective, so is with inverse ( )1 = 1 1 . Hence SX is closed under the compositon . And the associativity of the composition is well-known. Also 1 1X : x x is bijective and hence 1 1X SX . Hence (SX , ) is a monoid. And as 1 is the inverse element of and bijective again we have 1 SX again, such that SX is a group. P Proof of (1.12): 289

(i) By induction on the number of elements n := #X : In case n = 1 we have X = { x } for some single element x X . Hence the one any only map on X is given to be (x) = x. In particular #SX = 1 = 1! = (#X )!. Now for the induction step we consider a set with n + 1 elements X = { x1 , . . . , xn , xn+1 }. Then we can decompose SX into the following disjoint union
n+1

SX =
i=1

F (i) where F (i) := { SX | (xn+1 ) = xi }

Clearly, if (xn+1 ) = xi , then induces a bijection between the sets X (i) := X \ { xi } and X (n + 1). As X (i) always contains n elements we can choose a bijection i : X (i) X (n + 1). And this now gives rise to a bijection between F (i) SX (n+1) : i
X (n+1)

In particular for any i 1 . . . n + 1 the set F (i) contains as many elements, as SX (n+1) . And by induction hypothesis, as #X (n + 1) = n this is n!. So altogether
n+1

#SX =
i=1

#F (i) = (n + 1) n! = (n + 1)!

(ii) First of all we see that for any g , h and x G the associativity of G gives birth to the following equality: Lgh (x) = (gh)x = g (hx) = Lg (hx) = Lg (Lh (x)) = Lg Lh (x). And as x G has been arbitrary this is Lgh = Lg Lh . Also it is clear that Le (x) = ex = x = 1 1G (x) such that Le = 1 1G . In particular Lg1 is the inverse map of Lg since Lg1 Lg = Lg1 g = Le = 1 1G Lg Lg1 = Lgg1 = Le = 1 1G In particular Lg : G G not only is a map, but even a bijective one. Hence Lg SG and this makes L well-dened. It remains to show the injectivity of L but this is easy: if Lg = Lh then g = ge = Lg (e) = Lh (e) = he = h. P Proof of (1.13): (i) As S := k | k Z is a subgoup of the nite group Sn (this has been shown in (1.12.(i))), we know from (1.9), that there is some k N with k = 1 1, say k = ord( ). In particular k (i) = i such that the cycle cyc(, i) if nite, with (, i) k = ord( ). (ii) Suppose we had b := (a) x( ), then we had (a) = b = (b). But as is injective this implies a = b x( ) which is absurd, since we started with a dom( ). Hence we can restrict to : dom( ) dom( ) : a (a) 290 16 Proofs - Fundamentals

But as is injective and dom( ) is a nite set (it is contained in 1 . . . n) this implies that the restriction of even is surjective, in particular (dom( )) = dom( ). If now i dom( ) then we will use induction on k to prove that any k (i) lies within dom( ): The case k = 0 is true by assumption 0 (i) = i dom( ). Now k+1 (i) = ( k (i)) and by induction hypothesis we know k (i) dom( ). Hence k+1 (i) (dom( )) = dom( ) again. (iii) If = then dom( ) = dom( ) and (a) = (a) for any a D are trivial, as we even have (a) = (a) for any a 1 . . . n. Conversely let D = dom( ) = dom( ). If a D then (a) = (a) by assumption. But if a D then we have a x( ) such that (a) = a. Likewise a x( ) such that (a) = a = (a). Hence we have (a) = (a) for any a and this is = as claimed. (iv) If a dom( ) then by assumption we have a dom( ) such that a x( ). In particular we get (a) = (a)

And as we know (a) dom( ) again, because of (ii), we likewise have (a) x( ) such that we also nd (a) = (a)

That is (a) = (a) = (a) for any a dom( ). In complete analogy (by interchanging the roles of and ) we nd (for any a dom( )) (a) = (a) = (a)

And if nally a (1 . . . n) \ (dom( ) dom( )) = x( ) x( ) then it is clear that (a) = (a) = a and also (a) = (a) = a. In any case we have seen (for any a 1 . . . n) that (a) = (a) which is the commutativity = . P Proof of (1.15): As long as the order of the ik is maintained, we still have ik ik+1 (and i i1 ) no matter with wich element ik we begin. Hence (i1 i2 . . . i ) = (i2 i3 . . . i i1 ) = . . . = (i i1 i2 . . . i
1 )

Let us denote := (i1 i )(i1 i 1 ) . . . (i1 i2 ), then given any a = iq let us regard from the right: iq is unaected by any (i1 ip ) with p < q . Hence iq it rst hits (i1 iq ) which maps iq i1 . Now i1 immediatley strikes (i1 iq+1 ) again mapping i1 iq+1 . And iq+1 passes through unhampered, as it only meets (i1 ir ) whith r > q + 1 from now on. Altogether we have : iq i1 iq+1 . A special case is a = i as this only strikes (i1 i ) at the very end, such that : i i1 . Therefore = (i1 i2 . . . i ). 291

 By induction on d := j i N. The case d = 1 is trivial, as (i j ) = (i i + 1) in this case. For d 2 we have to regard (i j ) = (i i + d). To do this we rst prove that (i i + d) = (i i + 1)(i + 1 i + d)(i i + 1) by explicitly computing (from right to left): (i i + 1) (i + 1 i + d) (i i + 1) i+d i+1 i i i i+1 i+1 i+d i+d

i+d i+1 i

As the domain of = (i i + 1)(i + 1 i + d)(i i + 1) is dom( ) = { i, i + 1, i + d } it is clear, that (a) = a for any a { i, i + 1, i + d }. And as (i) = i + d, (i + 1) = i + 1 and (i + d) = i we have truly shown that = (i i + d). But by induction hypothesis (i + 1 i + d) is of distance d 1 and can hence be written as (i + 1 i + d) = (i + 1 i + 2) . . . (i + d 1 i + d) . . . (i + 1 i + d) Inserting this into (i i + d) = (i i + 1)(i + 1 i + d)(i i + 1) we have also proved the case d 2 and hence for any transposition, as d may be arbitrarily large. Let us denote the cycle := ( (i1 ) (i2 ) . . . (i )). Then is is clear that dom( ) = { (i1 ), (i2 ), . . . , (i ) } and for any k 1 . . . 1 we have ( (ik )) = (ik+1 ), also ( (i )) = (i1 ). But on the other hand it is easy to compute 1 ( (ik )) = (ik ) = (ik+1 ) and likewise 1 ( (i )) = (i ) = (i1 ). Hence we have seen 1 (a) = (a) for any a = (ik ) (where k 1 . . . ). If now a 1 . . . n is none of the (ik ) (that is a dom( )) then 1 (a) { i1 , i2 , . . . , i } = dom( ). Therefore we have 1 (a) = ( 1 (a)) = a = (a) again. That is 1 and share the same domain and agree on it, such that by (1.13.(iii)) they are equal. P Proof of (1.16): (a) = (b): by assumption we have = (i1 i2 . . . i ) for some pairwise distinct ik 1 . . . n. Therefore it is clear that dom( ) = { i1 , i2 , . . . , i } Let us abbreviate i := i1 , then by induction on k it is clear that ik = k1 (i), because k (i) = ( k1 (i)) = (ik ) = ik+1 . If we are now given any a, b dom( ), then there are r and s 1 . . . such that a = ir = r1 (i) and b = is = s1 (i). If s r then we can readily take k := s r N to see k (a) = sr r1 (i) = s1 (i) = b And if s < r then we take k := + s r N (which is positive, since r ). Note that (i) = 1 (i) = (i ) = i1 such that we can 292 16 Proofs - Fundamentals

likewise compute k (a) =

+sr r1

(i) = s1 (i) = = s1 (i) = b

Thus in any case we have seen that b = k (i)(a) which had to be shown. (b) = (c): choose any i dom( ) then by assumption we have b = k (i) for any b dom( ) for some sucient k N. That is dom( ) k (i) | k N

But the converse inclusion is a generat truth, for any permutation, as we have shown in (1.17.(ii)). And hence the above inclusion even is an equality of sets. Now (c) = (d) is trivial - note that dom( ) = is nonempty, as we assumed = 1 1. (d) = (e): all we have to do is prove, that the higher powers k of k (i) are already contained in { r (i) | r 0 . . . 1}. So consider any k , then we use division with remainder to write k as k = m + r for some m N and r 0 . . . 1. As = ord( ) we have = 1 1 such that we are enabled to compute k (i) = m
+r m

(i) =

r (i) = 1 1m r (i) = r (i)

Hence any k (i) is contained in { r (i) | r 0 . . . 1} such that we get dom( ) = { r (i) | r 0 . . . 1} due to the assumption (d). (e) = (a): let us take ik := k1 (i) for any k 1 . . . . We rst prove that there elements are truly pairwise distinct: Suppose ir = is for some r = s, without loss of generality we can take r < s. Then r1 (i) = s1 (i) implies sr (i) = i (by multiplication with 1r from the left). That is there would have to be some h = s r N with h (i) = i. Thereby h = s r < s such that h 1 . . . 1. As = ord( ) is the minimal number with = 1 1 this implies h = 1 1, h that is there has to be some a 1 . . . n with (a) = a. In particular a dom( ) as else we would also have h (a) = a. By assumption there hence is some k 0 . . . 1 such that a = k (i). Now compute h (a) = h k (i) = k h (i) = k (i) = a in contradiction to h (a) = a. Hence the assumption ir = is must have been false, that is the ik are pairwise distict. Hence we are entitled to dene := (i1 i2 . . . i ). Then for any k 1 . . . 1 it is clear that (ik ) = ik+1 = k (i) = ( k1 (i)) = (ik ). However we also nd (i ) = i1 = i = (i) = ( 1 (i)) = (i ). That is we have shown that (a) = (a) for any a dom( ). But as dom( ) = { i1 , i2 , . . . , i } = k (i) | k 0 . . . 1 = dom( )

is true by assumption, this implies = . That is we have shown that truly is the cycle generated by the ik . 293

 The fact that = ord( ) can be easily seen directly. But the implication (a) = (e) already yields ord( ) = , as the length of the cycle appears in (e) as the order of , since k1 (i) = ik . P Proof of (1.17): (i) Let us abbreviate X := 1 . . . n and S := k | k Z . Then this proof could be abbreviated by noting that S X X : (, i) (i) is a group action with the orbits Si = cyc(, i). Anyhow we want to give a direct proof that does not require group actions: Clearly is an equivalency relation, i i is true, due to i = 0 (i) and if i j with j = p (i) and j k with k = q (j ) then clearly k = q ( p (i)) = p+q (i) such that i k again. Also let z := ord( ) N (also see (i)), then z = 1 1 such that ( k )1 = k = z k = z k . Hence if i j with j = k (i) then i = z k (j ) such that j i again. And the equivalence class is obviously given to be [i] = { j X | j i } = k (i) | k N = cyc(, i)

If now j cyc(, i), then j is of the form j = k (i) for some k N. In particular (j ) = k+1 (i) cyc(, i) again. And hence we get a well-dened map i : cyc(, i) cyc(, i) : a (a) Once more, as is injective and cyc(, i) is a nite set we even get (cyc(, i)) = cyc(, i) such that the above i bocomes bijective. We now expand i to 1 . . . n trivially by letting i (a) := a for any a cyc(, i). Then it is clear, that dom(i ) = cyc(, i) And as i acts on cyc(, i) = i, (i), . . . , k(i) (i) just like does, it also is clear, that i = (i (i) . . . k(i) (i)). Now, as restricts to an equivanlence relation on dom( ) [for a x( ) the equivalence classe is just [a] = { a }] we may pick up a system of representants i(1), . . . , i(r). As the cycles are disjoint cyc(, i(p)) cyc(, i(q )) = for p = q 1 . . . r we nd from (1.13.(iv)) that the respective cycles commute p = q 1...r = i(p) i(q) = i(q) i(p)

Also if a dom( ) then there is precisely one p 1 . . . r such that a cyc(, i(p)) = dom(i(p) ). In particular we have a x(i(q) ) for any q = p. And hence we get
j =1 r

i(j ) (a) = i(p)

q =p

i(q) (a) = i(p) (a) = (a)

294

16 Proofs - Fundamentals

Thereby we only used the fact, that we can bring i(p) in front by successively interchanging two cycles (which is obvious by induction on r). And if a x( ) then by consruction we also have a x(i(p) ) for any p 1 . . . r such that (i(1) . . . i(r) )(a) = a = (a). In any case we have found (a) = (i(1) . . . i(r) )(a) such that the equality = i(1) . . . i(r) is nally established, as well. (ii) Let us abbreviate := 1 . . . k and := 1 . . . t . We will prove this statement by induction on m := min{ k, t }. For m = 1 we already have 1 = 1 such that there is nothing to prove. For m 2 consider any a dom(1 ). By assumption this implies a dom(i ) for any i = 1. Hence we get (a) := 1 . . . k (a) = 1 (a) In particular a dom( ) = dom( ). Likewise there is a uniquely determined r 1 . . . t such that a dom(r ) as for any s = r we get a dom(s ). Hence - by (ii) - we have r (a) dom(r ) again such that r (a) dom(s ) for any s = r, as well. This implies (a) := 1 . . . t (a) = r (a) Altogether 1 (a) = (a) = (a) = r (a). We will now prove dom(1 ) = dom(r ): given any b dom(1 ) there is some p N such that b = p p (a) = (a) by virtue of (1.16.(b)), as 1 is a cycle. Therefore b = 1 1 p r (a) dom(r ) by (ii) again. Altogether this is dom(1 ) dom(r )

Reversing the roles of 1 and r we nd the converse inclusion, such that the domains are equal. In particular, given any a dom(1 ) we can repeat the above calculation, to nd a dom(1 ) = dom(r ) : 1 (a) = r (a) Clearly this implies 1 = r . Recall that the domains of the s were assumed to be pairwise disjoint, such that by (1.13.(iv)) the s are mutually commutative. Then we can compute
1 1 2 . . . k = 1 = r = s= r

Hence by induction hypotheseis we nd k 1 = t 1 and some permutation such that i = (i) for any i 2 . . . k . Clearly this implies k = t and we can append to Sr by (1) := r. (iii) By (iv) can be written as a composition of cycles j := i(j ) with pairwise disjoint domains. And by (v) these cycles are uniquely determined. Hence this item (vi) is an immediate result of (iv) and (v). (iv) By (iv) any permutation Sn can be written as a composition of cycles = 1 . . . r . And according to (1.15) any cycle i can be written 295

as a composition of transpositions i =
j J ( i)

i,j

Again, according to (1.15), any transposition i,j can be written as a composition of adjacent transpositions i,j =
kK (i,j )

i,j,k

Altogether we can decompose into a - admittably very large - composition of adjacent cycles i,j,k by putting all this together.

(v) The implication (b) = (a) is quite easy: pick up the cyclic decomposition = 1 . . . k as in (iii) and let i := i 1 for any i 1 . . . r. Then by (1.15) i is a cycle of the same length ord(i ) = ord(i ) again. And as is bijective the cycles i are pairwise disjoint again. Now compute = 1 = 1 2 . . . k 1 = 1 1 2 1 . . . k 1 = 1 2 . . . k That is we have found a cyclic decomposition of again. And by construction this decomposition satises the properties of (a). But as the cyclic decomposition of is unique (up to sequence, as we have seen in (ii)) (a) is true for any cyclic decomposition of . The converse implication (a) = (b) only requires little more eort: Let us abbreviate (i) := ord(i ) the length of i . That is there are some numbers ai,j 1 . . . n such that i can be written as i = (ai,1 ai,2 . . . ai,
(i) )

And by assumption we have ord((i) ) = ord(i ) = (i) such that there also are numbers bi,j 1 . . . n such that (i) can be written as (i) = (bi,1 bi,2 . . . bi,
(i) )

As the domains of the i and the r are pairwise disjoint we obtain a disjoint union of sets (that compose the domain of )
k

D :=
i=1

{ ai,j | j 1 . . . (i) }

And this enables us to dene a permutation Sn by letting (ai,j ) := bi,j for any i 1 . . . k and j 1 . . . (i) and (a) := a for any a D. 296 16 Proofs - Fundamentals

Then by (1.15) it is clear that i 1 = (ai,1 ai,2 . . . ai, = (bi,1 bi,2 . . . bi,
(i) ) 1 (i) ))

= ( (ai,1 ) (ai,2 ) . . . (ai,

(i) )

= (i)

Hence we nd (reversing the computation in the case (b) = (a) above) that and are truly conjugates under
k k k

1 =
i=1

1 =
i=1

(i 1 ) =
i=1

i = P

Proof of (1.2) (commutativity): We will now prove that in a commutative groupoid (G, ) the order in which the elements x1 , . . . , xn G are multiplied has no impact on the result of the composition. That is for any permutation Sn we have x(1) x(2) . . . x(n) = x1 x2 . . . xn As we have seen in (1.17.(iv)) any permutation Sn can be expressed as a composition of adjecent transpositions = 1 2 . . . s We will now prove the claim by induction on the number of transpositions s. In case s = 1 we have to deal with one adjacent transposition, say = (j j + 1). Then we have
n

x(i) = x1 . . . xj 1 xj +1 xj xj +2 . . . xn
i=1

But by the commutativity rule (C) we can interchange xj +1 and xj in this product to nd x(1) x(2) . . . x(n) = x1 . . . xj xj +1 . . . xn . In case s 2 we let := 1 and = 2 . . . s . Then it is clear that = . And as in the case s = 1 resp. by induction hypothesis we get
n n n n

x(i) =
i=1 i=1

(i)

=
i=1

( i)

=
i=1

xi P

Proof of (1.21): (i) Let us denote the set := {(i, j ) (1 . . . n)2 | i < j } and the map : (1 . . . n)2 Z : (i, j ) j i. For p, q N we let p q be the minimum and p q be the maximum of p and q . At last we dene : : (i, j ) ( (i) (j ), (i) (j )) 297

Thereby is well dened, since 1 (i) (j ) < (i) (j ) n. In fact even is bijective, as we can explictly give its inverse to be 1 : : (i, j ) ( 1 (i) 1 (j ), 1 (i) 1 (j )) As is nite it suces to check 1 = 1 1 : let (p, q ) := (i, j ), then p and q { (i), (j ) } and p < q . Hence 1 (p) and 1 (q ) { i, j }, so ordering them we get 1 (p, q ) = (i, j ) again, which is the bijectivity of . Also we easily compute (i, j ) = = (j ) (i) (i) (j ) (j ) (i) if (j ) > (i) if (j ) < (i) 1 1 if (j ) > (i) if (j ) < (i)

After all these preparations we can proceed to the main computation: we use the fact that (due to the commutativity of Z) the product can be permuted arbitraily without changing its result. Here we have (j i) =
(i,j ) (i,j )

(i, j ) =
(p,q )()

(i, j ) (j ) (i)
(i,j )

=
(i,j )

(i, j ) = (1)I ()

And from here we are almost done, note that the product (j i) is non-zero, as we always have i < j . Also 1/(1)I () = (1)I () as this number is either +1 or 1. Hence dividing by the product and (1)I () we nally get (1)I () =
(i,j )

(j ) (i)

(i,j ) (j i)

=
(i,j )

(j ) (i) ji

(iv) Let us continue with the notation introduced in (i) above. Then we have already shown the following identity for any permutation (1)I ()
(i,j )

(j i) =
(i,j )

(j ) (i)

Using this identity for the composition and the permutations and seperately we can again compute (1)I (
) (i,j )

(j i) =
(i,j )

(j ) (i) (j ) (i)
(i,j )

= (1)I () = (1)
I ( )

(1)I (

) (i,j )

(j i)

Hence dividing by the product 298

(j i) we have established the ho16 Proofs - Fundamentals

morphism property of the signum, expressed in terms of (i): sgn( ) = (1)I (

)

= (1)I () (1)I (
)

= (1)I ( ) (1)I (

= sgn( ) sgn( )

(iii) Consider the permutation = (a b) where (without loss of generality) a < b. Then it is easy to see, that has the following inversions { (i, j ) | (i) > (j ) } = { (a, a + s) | 1 s b a } { (a + r, b) | 1 r < b a } In particular we get I ( ) = (b a) + (b a 1) = 2(b a) 1 and thereby we get - using variant (i) of computing the signum sgn( ) = (1)I ( ) = (1)2(ba)1 = (1)2
ba

(1)1 = 1

That is any transposition is an odd permutation! Thus, if = 1 . . . r is a composition of r transpositions, then the homorphism property (iv) of the signum implies
r r

sgn( ) =
i=1

sgn(i ) =
i=1

(1) = (1)r

(ii) By (1.15) we know that any cycle i can be written as a composition of (i) 1 transpositions. Hence - by (iv) combined with (iii)
r r

sgn( ) = sgn(1 . . . r ) =
i=1

sgn(i ) =
i=1

(1)

(i)1

And as the domains of the i are pairwise disjoint (1.13.(iv)) implies, that they are mutually commutative. Hence by (1.5) we nd that any power k N of is just
r

k =
i=1

ik

Also as dom(ik ) dom(i ) (this is (1.13.(ii))) and the dom(i ) are pairwise disjoint we have (for any a 1 . . . n) (a) = a if and only if ik (a) = a for any i 1 . . . r. In particular we see the equivalency of k = 1 1 i 1 . . . r : ik = 1 1

1 As i is a cycle its order is given to be (i) = ord(i ). Hence ik = 1 if and only if (i) divides k (by division with remainder again, write (i) k = m (i) + r then ik = (i )m ir = ir = 1 1 i r = 0). Therefore k = 1 1 i 1 . . . r : (i) | k

Now ord( ) = k is the minimal number 1 k N such that k = 1 1. 299

And likewise lcm( (1), . . . , (r)) is the minimal number that is divided, by all the (i). In particular both numbers are equal. (v) By denition An is just the kernel of the group-homorphism sgn : Sn Z . Therefore the rst isomorphism theorem (of groups) yields Sn
An Z : An sgn( )

In particular these two sets have the same cardinality. Now use the theorem of Lagrange and (1.12.(i)) to compute #Sn n! 2 = #Z = #Sn #A = = n # An # An P Proof of (9.3): (i) Clearly for any x X we have x x, since x = ex. Also if x y , say y = gx for some g G, then x = ex = (g 1 g )x = g 1 (gx) = g 1 y such that we also have y x. And if also y z , say z = hy for some h G, then we also have z = hy = h(gx) = (hg )x such that x z . Altogether is an equivalency relation on X . And for x X the equivalency class of x is the orbit Gx, since [x] = { y X | x y } = { y X | g G : y = gx } = { gx | g G } = Gx (ii) We will rst prove that the stabilizer Gx = { g G | gx = x } is a subgroup of G. First of all e Gx since ex = x. Also if g , h Gx then gx = x and hx = x such that (gh)x = g (hx) = gx = x, and this means gh Gx again. Finally g Gx implies x = ex = (g 1 g )x = g 1 (gx) = g 1 x such that g 1 Gx again. Altogether Gx is a subgroup of G. It remains to prove the well-denedness and bijectivity of Gx G Gx : gx gGx Let us rst check the well-denedness: if gx = hx then (h1 g )x = h1 (gx) = h1 (hx) = (h1 h)x = ex = x such that h1 g Gx . And this again is gGx = hGx by denition of G/P . Now the surjectivity is clear: for g G then gx Gx is mapped to gGx . And for the injectivity suppose we had some g , h G with gGx = hGx . Undoing the above computation we then have h1 g Gx such that gx = e(gx) = (hh1 )gx = h((h1 g )x) = hx again. (iii) By denition we have gA A for any g G. But this implies A = eA = (gg 1 )A = g (g 1 A) gA A such that gA = A. (iv) Given g G and x X \ A we have to show gx X \ A. Thus suppose gx A, then x = ex = (g 1 g )x = g 1 (gx) g 1 A A. 300 16 Proofs - Fundamentals

(v) By (ii) we have #Gx = #(G/Gx ) and (as both, G and X are nite) by the theorem of Lagrange (1.7) this implies #Gx = [G : Gx ] = #G # Gx

Now the second orbit formula is a simple consequence: as is an equivalency relation the orbits Gx form a partition of X and hence #X =
GxX/G

#Gx = #G
GxX/G

1 # Gx

(vi) The basic idea of this proof is the following identity of the disjoint unions of the stabilizers Gx = { g G | gx = x } and the xed point sets X g = { x X | gx = x }. Thus taking to the unions we nd { g } X g = { (g, x) G X | gx = x } =
g G x X

Gx { x }

And as these unions are disjoint (by construction) we hence already have proved the second part of Burnsides formula #X g =
g G xX

#G x

For the rst part we use the partition of X into orbits Gx again, that is X = Gx where the union runs over all orbits Gx X/G. This can be used to split the sum over all x X into two seperate sums the rst over Gx X/G the second over y Gx # Gx =
xX GxX/G y Gx

# Gx =
GxX/G

#Gx #Gx

But by the orbit fomula in (iv) we have #Gx #Gx = #G identically, such that this sum is reduced once again, leaving only #Gx =
xX GxX/G

#G = # G #X G P

Proof of (7.5): (i) Recall the denition of the minimum and maximum of any two elements , A (this is truly well-dened, since is a linear order) := if if := if if

In the case that M = { } contains one element only, we clearly get the minimal and maximal elements = = . If now M is given 301

to be M = { 1 , . . . , n } where n 2 then we regard := (. . . (1 2 ) . . . ) n := (. . . (1 2 ) . . . ) n Then and are minimal, resp. maximal elements of M , which can be seen by induction on n: if n = 2 then is minimal by construction. Thus for n 3 we let H := { 1 , . . . , n1 } M . By induction hypothesis we have H = { } for = ((1 2 ) . . . ) n1 . Now let := n then i for i < n and n by construction. Hence we have i for any i 1 . . . n, which means M . And the uniqueness is obvious by the anti-symmetry of . And this also proves the the claim for the maximal element , by taking to the inverse order : . (ii) The existence of is precisely the property of a well-ordering. The uniqueness is immediate from the anti-symmetry of again: suppose 1 and 2 both are minimal elements of M , in particular 1 2 and 2 1 . Ant this implies 1 = 2 , as is a (linear) order on A. (iii) In the direction (a) = (b) we consider < . By the positivity of this implies + + . If the equality + = + would hold true then - as A is integral - we wolud nd = in contradiction to the assumption < . In the converse direction (b) = (a) we are given . The positivity + + (for any A) of is clear in this case. Now asume + = + then we need to show = to see that A is integral. But as is total we have or , where we may assume without loss of generality. If we now had = then < and hence + < + in contradiction to + = + , which rests this case, too. Now (c) = (b) is trivial such that there only remains the direction (b) = (c). But this is clear - assume + < + but , then by the positivity (a) we would get the contradiction: + + . (iv) Assume = , then by the minimality of this would yield < and hence (by property (b) in point (iv) above) + = + < + Thus we see that = and hence (by the minimality of ) we get < again. Using (b) once more this yields a contradiction + < + < + Thus we have seen = but in complete analogy we nd that = leads to another contradiction and hence = . P Proof of (2.1): (v) As is a total order, we may assume a b without loss of generality. If a, b A , then b A, a b and a A implies a = b. Likewise if a, b A then a A, a b and a A implies a = b. 302 16 Proofs - Fundamentals

(vi) By induction on n: if n = 1 then a = a1 is trivial and if n = 2 then a is minimal by construction. Thus for n 3 we let H := { a1 , . . . , an1 } A. By induction hypothesis we have H = { h } for h = ((a1 a2 ) . . . ) an1 . Now let a := h an then a h ai for i < n and a an by construction. Hence we have a ai for any i 1 . . . n, which means a A . And the uniqueness has already been shown above. This also proves the like statement for A by taking to the inverse ordering b a : a b. (ix) Regard Z as a partially ordered set (Z , ) under the inclusion relation (this is allowed, due to (1)). If now C Z is a chain, then by (2) we get U := C Z . And clearly we have C U for any C C . Thus U is an upper bound of C and hence we may apply the lemma of Zorn to (Z , ), to nd a maximal element Z Z . (x) Let us dene the partial order B A : A B on Z . If now C Z is a chain, then by (2) we get U := C Z . And clearly U C (which translates into C U ) for any C C . Thus U is an upper bound of C and hence we may apply the lemma of Zorn to nd a -maximal element Z . But being -maximal trivially translates into being -minimal again. P Proof of (7.6): In the following we will regard arbitrary elements = (i ), = (i ) and = (i ) A := NI . In a rst step we will show that - assuming (I, ) is a linearly ordered set - the lexicographic order is a positive partial order on A. The reexivity lex is trivial, however, and for the transitivity we are given lex and lex and need to show lex . The case where two (or more) of , and are equal is trivial and hence the assumption reads as k < k l < l and and i < k : i = i i < l : i = i

for some k, l I . We now let m := min{ k, l } then it is easy to see, that i < m : i = i = i and m < m which establishes lex . It remains to prove the anti-symmetry, i.e. we are given lex and lex and need to show = . Suppose we had = then again k < k l < l and and i < k : i = i i < l : i = i

In the case k l this implies k < k = k and in the case l k we get l < l = l . In both cases this is a contradiction. Hence lex is a partial order, but the positivety is easy. Assume lex , then we need to show + lex + . If even = , then there is nothing to prove. Hence we may assume, that there is some k I such that k < k and i < k : i = i 303

But then + lex + is clear, as from this we immediately get k + k < k + k and i < k : i + i = i + i In a second step we will show that - assuming (I, ) is a well-ordered set - the lexicographic order truly is a positive linear order on A. By the rst step it only remains to verify that lex is total, i.e. we assume that lex is untrue and need to show lex . By assupmtion we now have for any k I k k or i < k : i = i It is clear that = and as (I, ) is well-ordered we may choose k := min{ l I | l = l } Then i < k we get i = i (as k was chosen minimally) and k k (by assumption). Yet k = k is false by construction which only leaves k > k and hence establishes lex . In the third step we assume that I is nite, i.e. I = 1 . . . n without loss of generality. Note that I clearly is well-ordered under its standard ordering and hence it only remains to prove the following assertion: let = M A be a non-empty subset, then there is some M such that for any M we get lex . To do this we dene 1 := min{ 1 N | M } 2 := min{ 2 N | M, 1 = 1 } ... ... n := min{ n N | M, 1 = 1 , . . . , n1 = n1 } Hereby the set { 1 | M } is non-empty, since M = was assumed to be non-empty. And { 2 | M, 1 = 1 } is non-empty, as there was some M with 1 = 1 and so forth. Hence is well dened and it is clear that M , as = for some M . But the property lex is then clear from the construction. Now we will proof, that - assuming (I, ) is a linearly ordered set the -graded lexicographic order is a positive partial order as well. It is clear that inherits the porperty of being a partial order from lex so it only remains to verify the positivity: let , then we distinguish two cases. In the case || < | | we clearly get | + | = || + | | < | | + | | = | + | If on the other hand || = | | then by assumption lex and as we have seen already this implies + lex + . Thus in both cases we have found + + . We now assume that (I, ) even is a well-ordered set and prove that is a total order in this case. Hence we need to regard the case that is untrue. Of course this implies || | | and in the case 304 16 Proofs - Fundamentals

|| < | | we are done already. Thus we may assume || = | | , but in this case the assumption reads as: lex . As we have seen above the lexicographic order is total and hence <lex which conversely implies < . In the nal step we assume that I is nie, i.e. I = 1 . . . n and we need to show that is a well-ordering. Thus let = M A be a nonempty subset again, then we have to verify that there is some M such that for any M we get < . To do this we dene m := min{ || | M } := min{ | M, || = m } where the second minimum is taken under the lexicographic order. As M is non-empty m is well-dened and as lex has been shown to be a well-ordering is well-dened, too. But M is trivial and is clear from the construction. P

305

Chapter 17

Proofs - Rings and Modules

Proof of (1.37): The statement 0 = 0 is clear from 0 + 0 = 0. And a0 = 0 follows from a0 = a(0 + 0) = a0 + a0. Likewise we get 0a = 0. And from this an easy computation shows (a)b = (ab), this computation reads as ab + (a)b = (a + (a))b = 0b = 0 (and a(b) = (ab) follows analogously). Combining these two we nd (a)(b) = ab, since ab = (ab) = ((a)b) = (a)(b). Next we will prove the general rule of distributivity: we start by showing that (a1 + + am )b = (a1 b) + + (am b) by induction on m. The case m = 1 is clear and in the induction step we simply compute a1 + + am + am+1 b =
m m

(a1 + + am ) + am+1 b
m+1

=
i=1

ai

b + am+1 b =
i=1

ai b

+ am+1 b =
i=1

ai b

Likewise it is clear, that a(b1 + + bn ) = (ab1 ) + + (abn ). And combining these two equations we nd the generale rule
m

ai

n j =1

bj =

n j =1 n

ai
i=1 m

bj

i=1

=
j =1 i=1 m n

ai bj ai bj
i=1 j =1

Using induction on n we will now prove the generalisation of the general law of distributivity. The case n = 1 is trivial (here J = J (1) so there is nothing to prove). Thus consider the induction step by adding 306

J (0) to our list J (1), . . . , J (n) of index sets. Then clearly

n i=0 ji J (i)

a(i, ji ) =
j0 J (0)

a(0, j0 )

n i=1 ji J (i)

a(i, ji )

Hence we may use the induction hypothesis in the second term on the right hand side and the general law of distributivity (note that we use J = J (1) J (n)) again) to obtain
n i=0 ji J (i)

a(i, ji ) =
j0 J (0)

a(0, j0 )

n j J i=1

ai,ji

=
j0 J (0) j J

a(0, j0 )
i=1

a(i, ji )

To nish the induction step it suces to note that the product of n a(0, j0 ) with n i=0 a(i, ji ) (due to the i=1 a(i, ji ) can be rewritten associativity of the multiplication). And the two sums over j (0) J (0) resp. over j J can be composed to a single sum over (j0 , j ) J (0)J . Thas is we have obtained the claim
n n

a(i, ji ) =
i=0 ji J (i) (j0 ,j )J (0)J i=0

a(i, ji )

Thus we have arrived at the binomial rule, which will be proved by induction on n. In the case n = 0 we have (a + b)0 = 1 = 1 1 = a0 b00 . Thus we commence with the induction step (a + b)n+1 = (a + b)n (a + b)
n

=
k=0 n

n k nk a b k

(a + b)
n k=0

=
k=0 n+1

n k+1 nk a b + k

n k (n+1)k a b k
n k=0

=
k=1

n ak b(n+1)k + k1
n k=1 n k=1

n k (n+1)k a b k ak b(n+1)k + n 0 n+1 a b 0

= =

n n+1 0 a b + n

n n + k1 k

n + 1 0 n+1 a b + 0
n+1

n + 1 k (n+1)k n + 1 n+1 0 a b + a b k n+1

=
k=0

n k nk a b k

It remains to verify the polynomial rule, which will be done by induction on k . The case k = 1 is clear again, and the case k = 2 is the 307

ordinary binomial rule. Thus consider the induction step a1 + + ak + ak+1 = =

k+1 =0 n n n

ak+1 + (a1 + + ak )
n

n k+1

k+1 (a1 + + ak )nk+1 ak+1

n a k+1 k+1 k+1 n k+1

||=nk+1

=
k+1 =0 n

n k+1 a

=
k+1 =0 ||=nk+1

n k+1 k+1 a ak+1

=
|(,k+1 )|=n

n (a, ak+1 )(,k+1 ) (, k+1 ) P

308

17 Proofs - Rings and Modules

Proof of (1.43): The equality of sets nzdR = R \ zdR is immediately clear from the denition: the negation of the formula b R : (b = 0) (ab = 0) is just b R : ab = 0 = b = 0. If a nilR then we choose k N minimal such that ak = 0. Suppose k = 0, then 0 = a0 = 1 which would mean that R = 0 has been the zero-ring. Thus we have k 1 and may hence dene b := ak1 = 0. Then ab = ak = 0 and hence a zdR, which proves nilR zdR. And if ab = 0, then a Rast would imply b = a1 0 = 0. Thus if a zdR then a R which also proves zdR R \ R . It is clear that 1 nzdR, as 1b = b. Now suppose a and b nzdR and consider any c R. If (ab)c = 0 then (because of the associativity a(bc) = 0 and as a nzdR this implies bc = 0. As b nzdR this now implies c = 0 which means ab nzdR. It is clear that 1 R - just choose b = 1. Now the associativity and existence of a neutral element e = 1 is immediately inherited from (the multiplication) of R. Thus consider a R , by denition there is some b R such that ab = 1 = ba. Hence we also have b R and thereby b is the inverse element of a in R . = rst suppose 0 = 1 R then we would have R = 0 and hence R \ {0} = and R = R = , in contradiction to the assumption R = R \ {0}. Thus we get 0 = 0. If now 0 = a R then a R by assumption and hence there is some b R such that ab = 1 = ba. But this also is porperty (F) of skew-elds. = if 0 = a R then as R is a skew eld - there is some b R such that ab = 1 = ba. But this already is a R and hence R \ {0} R . Conversely consider a R , that is ab = 1 = ba for some b R. Suppose a = 0, then 1 = ab = 0 b = 0 and hence 0 = 1, a contradiction. Thus a = 0 and this also proves R R \ {0}. Consider a and b nilR such that ak = 0 and bl = 0. Then we will prove a + b nilR by demonstrating (a + b)k+l = 0, via
k+l

(a + b)

k+l

=
i=0 k

k + l i k+li ab i k + l i l+(ki) ab + i
k l j =1 l j =1

=
i=0

k + l k+j lj a b k+j k + l j l j a b k+j

= bl
i=0

k + l i k i ab + ak i

= 0+0 = 0 And if c R is an arbitrary element, then also (ca)k = ck ak = ck 0 = 0 such that ca nilR. As a has been chosen arbitarily, this implies R (nilR) R and hence nilR i R is an ideal. Consider x and y ann(R, b). Then it is clear that (x + y )b = xb + yb = 0 + 0 = 0 and hence x + y ann(R, b). And if a R is arbitrary then 309

(ax)b = a(bx) = a0 = 0 such that ax ann(R, b). In particular x = (1)x ann(R, b) and hence ann(R, b) is a submodule of R. Let u R and a R with an = 0, then we can verify the formula for the inverse of u + a in straightforward computation (see below). And in particular we found u + a R . But as a has been an arbitrary nilpotent, this has truly been u + nilR R .
n1

(a + u)
k=0 n1

(1)k ak uk1
n1 k k+1 (k+1)

=
k=0 n

(1) a

+
k=0

(1)k ak uk

n1

=
k=1

(1)k1 ak uk
k=0 n1 n n

(1)k ak un(k1)

= (1)

a u

+ (1)0 a0 u0 = 1 P

Proof of (1.41): We now prove of the equivalencies for integral rings: for (a) = (b) we are given a, b and c R with a = 0 and ba = ca. Then we get (c b)a = 0 which is c b zdR. By assumption (a) this implies c b = 0 and hence b = c. Now the implication (b) = (c) is trivial, just let c = 1. And for the nal step (c) = (a) we consider 0 = b R such that ab = 0. Then (a + 1)b = ab + b = b such that by assumption (c) we get a + 1 = 1 and hence a = 0. Thereby we have obtained zdR { 0 } and the converse inclusion is clear. Next we consider (a) = (b), again we are given a, b and c R with a = 0 and ab = ac. This yields a(c b) = 0 and as a = 0 we have a zdR such that a nzdR. Hence c b = 0 which is b = c. The implication (b) = (c) is clear again, just let c = 1. And for (c) = (a) we consider 0 = a R such that ab = 0. Then a(b + 1) = ab + a = a. By assumption (c) this implies b + 1 = 1 and hence b = 0. Therefore R \ { 0 } nzdR which is R \ { 0 } = nzdR and hence zdR = { 0 }. P Proof of (1.40): Let us regard the polynomial ring R = Z[x, y ] in two variables x and y . As R is commutative it is clear, that (x + y )n = (x + y )k (x + y )nk . By the binomial rule (1.37) we may evaluate these terms to
n r=0

n r nr x y = r =

k i=0 n

k i ki xy i k i k i

nk j =0

n k j nkj x y j

r=0 i+j =r n

n k i+j ki+nkj x y j n k r nr x y j 17 Proofs - Rings and Modules

=
r=0 i+j =r

310

where we also used the general rule of distributivity. Now we may compare coecients (which is allowed by construction of R) to nd our claim (as j = r i) n r =
i+j =r

k i

nk j

=
i=0

k i

nk ri P

Proof of (4.7):

(i) It is clear that matm,n (R) is an R-module, as a matrix in nothing but a peculiar way of arranging an mn-tuple A Rmn . If this doesnt convince you try the following reasoning: M := Rm is a R-module under the pointwise operations. Hence matm,n (R) = M n is an R-module under the pointwise operations, too. But the pointwise operations are precisely what we have taken for the addition and scalar-multiplication of matrices. And the fact that E is an R-basis follows immediately from (3.39.(vi)) once we notice, that any matrix A = (ai,j ) can be written uniquely, as
m n

A =
i=1 j =1

ai,j Ei,j

(ii) We will rst prove the associativity of the matrix multiplication. To do this x any index i 1 . . . m and l 1 . . . q and compute
p

enti,l ((AB )C ) = =

enti,k (AB ) entk,l (C )

k=1 p n

enti,j (A) entj,k (B ) entk,l (C )

k=1 j =1 n p

=
j =1 n

enti,j (A)
k=1

entj,k (B ) entk,l (C )

=
j =1

enti,j (A) entj,l (BC )

= enti,l (A(BC )) The distributivity can be proved by a direct computation as well. In 311

this case we prefer to e4mploy the scalar product however: enti,k (A(B + C )) = = = rowi (A), colk (B + C ) rowi (A), colk (B ) + colk (C ) rowi (A), colk (B ) + rowi (A), colk (C )

= enti,k (AB ) + enti,k (AC ) = enti,k (AB + AC ) enti,k ((A + B )C )) = = = rowi (A + B ), colk (C ) rowi (A) + rowi (B ), colk (C ) rowi (A), colk (C ) + rowi (B ), colk (C )

= enti,k (AC ) + enti,k (BC ) = enti,k (AC + AC )

(iii) As matn is an R-module by (i) and the multiplication is associative and distributive by (ii) we already know that matn (R) is an R-semialgebra. It remains to prove that 1 1 truly is a unit element in matn (R). To see this consider any A matn (R) and compute
n

enti,k (1 1 A) =
j =1 n

enti,j (1 1) entj,k (A) i,j entj,k (A) = enti,k (A)

j =1 n

enti,k (A 1 1) =
j =1 n

enti,j (A) entj,k (1 1) enti,j (A) j,k = enti,k (A)

j =1

(iv) We will rst prove cen(R)1 1 cen(matn (R)), that is we consider any a cen(R) and B = (bi,j ) matn (R). Then we have to check that (a1 1)B = B (a1 1). But as a commutes with all other elements of R this is quite easy to see: (a1 1)B = a(1 1B ) = aB = (abi,j ) = (bi,j a) = B (a1 1) Conversely consider any A = (ai,j ) cen(matn (R)), that is A is a matrix, that commutes with all other matrices. In particular we have AEr,s = Er,s A for any r, s 1 . . . n. Now evaluate the (i, k )th entry
n

enti,k (A Er,s ) =
j =1 n

ai,j j,r k,s = ai,r k,s i,r j,s aj,k = i,r as,k
j =1

enti,k (Er,s A) =

That is for any i, k , r and s 1 . . . n we have ai,r k,s = i,r as,k . Let 312 17 Proofs - Rings and Modules

us choose s := k then this equality specializes to ai,r = i,r ak,k In particular - by taking r := i - we nd ai,i = ak,k for any indices i, k . And if r = i, then we have ai,r = 0. That is A is a diagonal matrix, in which all diagonal elements are equal. In short: A is of the form A = a1 1 for some a R. Now take B := b1 1, then clearly AB = ab1 1 and BA = ba1 1. But as AB = BA and b R is arbitrary, this implies a cen(R). Hence we also have cen(matn (R)) cen(R)1 1. (v) It is clear, that the transposition A A is bijective - the inverse map is the transposition matn,m (R) matm,n (R) : B B again. And the R-linearity of the transposition is straightforward again entj,i ((aA + B ) ) = enti,j (aA + B ) = a enti,j (A) + enti,j (B ) = a entj,i (A ) + entj,i (B ) = entj,i (aA + B ) (vi) Recall that R is assumed to be commutative here. Let us compare all the entries of the matrices (AB ) and B A , i 1 . . . l and k 1 . . . n. Then by denition of the matrix-multiplication we nd that all entries are equal, just compute entk,i (B A ) = = rowk (B ), coli (A ) = colk (B ), rowi (A) rowi (A), colk (B ) = enti,k (AB )

= entk,i ((AB ) ) (vii) Let us rst assume that A is invertible, then an easy computation 1 1 yields 1 1n = 1 1 n = (AA ) = (A ) A where the latter equality is due to (vi). Likewise 1 1n = (A1 A) = A (A1 ) . Hence we have seen that in this case A is invertible, too and that the inverse is given to be (A1 ) . That is we have proved the equality (A )1 = A1

If conversely A is invertible, then we have just seen that (A ) = A is invertible. So this already establishes the equivalency of the invertibilities of A and A . (viii) Let us denote x = (x1 , . . . , xn ) Rn and y = (y1 , . . . , ym ) Rm , then it is straightforward to compute
m m n

Ax, y

=
i=1 n

[Ax]i yi =
i=1 j =1 m

enti,j (A) xj yi
n m

=
j =1 n

xj
i=1

enti,j (A) yi =
j =1

xj
i=1

entj,i (A ) yi

=
j =1

xj [A y ]j = x, A y

313

Proof of (4.9): The deniton of gln (R) is precisely that of the multiplicative group of matn (R), hence there is nothing to prove. It is clear that the unit element 1 1n of matn (R) is contained in trin (R). Also if A and B trin (R) are any two upper triangular matrices, then for any a R and any i, j 1 . . . n we have enti,j (aA + B ) = a enti,j (A) + enti,j (B ) In particular if i < j we see that enti,j (aA + B ) = 0, such that aA + B is an upper triangular matrix again. By taking a = 1 we see that A + B trin (R) and by taking a = 1 and B = 0 we also have A trin (A). That is we have proved, that (trin (R), +) is a subgroup of (matn (R), +) - in fact we have shown that trin (R) is an R-submodule of matn (R). It remains to prove that trin (R) is closed under multiplication of matrices as well. Hence consider any two upper triangular matrices A = (ai,j ) and B = (bi,j ) trin (R) again. then for any i, k 1 . . . n we have
n

enti,k (AB ) =
j =1

ai,j bj,k

But since A trin (R) we have ai,j = 0 for any j > i. And as B trin (B ) as well, we also have bj,k = 0 of any j < k . Hence the only summands which might be non-zero are those from j = k to j = i. That is the sum breaks down to
i

enti,k (AB ) =
j =k

ai,j bj,k

Hence if i < k there are no non-zero elements left at all and the sum is 0. But this precisely means that AB trin (R) is an upper triangular matrix again, and this had to be shown. P Proof of (4.12): (i) Let us denote A = (ai,j ) matm,n (R) and B = bi,j matn,m (R). By denition we have Er,s = (i,r j,s ), hence - for any i, k 1 . . . n we may simply compute the matrix multiplication
n

enti,k (AEr,s ) =
j =1

ai,j j,r k,s = ai,r k,s

That is the (i, j )-th entry of AEr,s is zero, if k = s and if k = s, then it simply is ai,r . For the k -th column this implies colk (AEr,s ) = k,s (a1,r , . . . , ai,r ) = k,s colr (A) 314 17 Proofs - Rings and Modules

Likewise we compute the (i, k )-th entry of Er,s B and use this to evaluate the i-th row of this matrix. This time we get
n

enti,k (Er,s B ) =
j =1

i,r j,s bj,k = i,r bs,k

rowi (Er,s B ) = i,r (bs,1 , . . . , bs,n ) = i,r rows (B ) (ii) This identity is a straightforward computation - just be aware that i,j commutes with any other scalar of R and that the sum j i,j xj breaks down to xi . Then we pick up any p, s 1 . . . n and compute entp,s (Ei,j A Ek,l ) =
r=1 n

entp,r (Ei,j A) entr,s (Ek,l ) entp,q (Ei,j ) aq,r entr,s (Ek,l )

r=1 q =1 n

=
r=1 q =1

i,p j,q aq,r k,r l,s

= i,p aj,k l,s = aj,k entp,s (Ei,l ) (iii) This is an ultimate consequence of (i): clearly Tr,s B = (1 1n + aEr,s )B = B + aEr,s B . Hence the i-th row of Tr,s (a)B is given to be rowi (Tr,s (a)B ) = rowi (B ) + a colk (Er,s B ) = rowi (B ) + k,s a rowi (B ) Likewise we have ATr,s = A(1 1n + aEr,s ) = A + (Aa)Er,s . And hence the k -th column of ATr,s (a) is given to be colk (ATr,s (a)) = colk (A) + colk ((Aa)Er,s ) = colk (A) + k,s colr (Aa) = colk (A) + k,s colr (A) a (iv) Let us denote xj = (x1,j , . . . , xm,j ) Rm , that is xj is the j -th column of the matrix X = xi,j matm,n (R). By denition the diagonal matrix D := dia(a1 , . . . , an ) has i,j aj as its (i, j )-th entry. We begin by proving the second equality by simple matrix-multiplication
n

enti,k (XD) =
j =1

xi,j j,k ak = xi,k ak

Therefore the k -th colums of XD is colk (XD) = (x1,k ak , . . . , xm,k ak ) which is xk ak as we had claimed. In the rst equality the xj are used as rows, that is we actually deal with the transpose X of X . Hence the correct ansatz is
n

enti,k (DX ) =
j =1

i,j aj xk,j = ai xk,i

315

Therefore the i-th row of X is rowi (DX ) = (ai x1,i , . . . , ai xm,i ) which is ai xi . Hence we also have shown the rst equality. (v) These equations are straightforward computations using the denition Nn = (i+1,j ). Let us denote xj = (x1,j , . . . , xm,j ) Rm , then X = (xi,j ) is the (m n)-matrix composed of the columns xj . Hence the transposed matrix X has the rows xj . Thereby for any i, k 1 . . . n
n

enti,k (XNn ) =
j =1 n

xi,j j +1,k = i+1,j xk,j =

j =1

xi,k1 0 xk,i+1 0

if k > 1 if k = 1 if i < n if i = n

enti,k (Nn X ) =

Once the entries are established, it is easy to evaluate the rows and colums of these matrix products colk (Nn X ) = (x1,k1 , . . . , xm,k1 ) = rowi (Nn X ) = (x1,i+1 , . . . , xm,i+1 ) = xk1 0 xi+1 0 if k > 1 if k = 1 if i < n if i = n

(vi) Recall the denition of the permutation matrix P = ((i),j ). Let us denote xj = (x1,j , . . . , xm,j ) Rm and X = (xi,j ) again. Then the (m n)-matrix composed of the columns xj and the transposed matrix X has the rows xj . Thereby for any i, k 1 . . . n
n

enti,k (P X ) =
j =1

(i),j xk,j = xk,(i)

Hence the i-th row of P X is (x1,(i) , . . . , xm,(i) ) which is nothing but x(i) . Now we supposed that is bijective, that is the equation (j ) = k has the unique solution j = 1 (k ). Then we get
n

enti,k (XP ) =
j =1

xi,j (j ),k = xi,1 (k)

Thereby the k -th column of XP is (x1,1 (k) , . . . , xm,1 (k) ) which nothing but x1 (k) . Altogether we have proved the equalities given. P

Proof of (4.13): (i) It is clear that the map given is an R-module homorphism, as the operations are pointwise, in both cases. The bijectivity is trivial as well. As also (1, . . . , 1) dia(1, . . . , 1) = 1 1n it only remains to prove, that the map also transfers the respective multiplications. Hence consider 316 17 Proofs - Rings and Modules

any a1 , . . . , an and b1 , . . . , bn R. Using (4.12.(ii)) we nd dia(a1 , . . . , an ) dia(b1 , . . . , bn )

n n

=
i=1 n

ai Ei,i
n

j =1

bj Ej,j

=
i=1 j =1 n n

ai bj Ei,i Ej,j ai bj i,j Ei,j

i=1 j =1 n

=
i=1

ai bi Ei,i = dia(a1 b1 , . . . , an bn )

That is we have shown that the map given truly is an isomorphism of R-algebras. Also dia(a1 , . . . , an ) dia(b1 , . . . , bn ) = 1 1n = dia(1, . . . , 1) is satised if and only if ai bi = 1 for any i 1 . . . n. Likewise bi ai = 1 for any i 1 . . . n. But this is equivalent to ai R being a unit in R 1 with inverse bi = a i . Hence dia(a1 , . . . , an ) is invertible if and only ai R for any i 1 . . . n. (ii) This is just a special case of (i) - the i-th entry is xed to be a and the other n 1 entries are 1, which always is a unit 1 R of R. (iii) Using (4.12.(ii)) again, the multiplication property of the transposition matrices is an easy computation Ti,j (a) Ti,j = (1 1n + aEi,j )(1 1n + bEi,j )
2 = 1 1n + (a + b) Ei,j + ab Ei,j

= 1 1n + (a + b) Ei,j = Ti,j (a + b)
2 = 0, as we have assumed i = j to be distinct. Now, Hereby Ei,j as Ti,j (0) = 1 1n we see that Ti,j (a)Ti,j (a) = 1 1n and of course also Ti,j (a)Ti,j (a) = 1 1n . That is Ti,j (a) is invertible and the inverse is given to be Ti,j (a).

(iv) Let E = { e1 , . . . , en } be the Euclidean basis of Rn . Then we use the denition of the permutation matrix P and the multiplication formula (4.12.(vi)) for permutation matrices P P = P e (1) . . . e (n) e (1) . . . e (n) = P

Again it is clear that truly Pid = 1 1n . Hence if is bijective, then 1 P = P = 1 P P1 = Pid = 1 1n and analogously P 1n such that P id is invertible and the inverse is given to be P1 . Conversely suppose P P = 1 1n and P P = 1 1n . By comparing coecients we nd that = id and = id, that is is an inverse function of and in 317

particular is bijective. P Proof of (1.47): ( = semi-ring) as any Pi is a sub-semiring we have 0 Pi and hence 0 i Pi again. Now consider a, b i Pi that is a, b Pi for any i I . As any Pi is a sub-semiring this implies that a + b, ab and a Pi (for any i I ). And thereby a + b, ab and a i Pi again. ( = ring) in complete analogy we have 1 Pi (for any i I , as any Pi has been assumed to be a subring. Hence we also get 1 i Pi . ( = (skew)eld) consider 0 = a i Pi . That is a Pi for any i I and as any Pi is a subeld we get a1 Pi such that a1 i Pi . ( = (left)ideal) consider any b R and a i Pi - that is a Pi for any i I . As Pi is a left-ideal we nd bai Pi again and hence ba i Pi . In the case of ideals Pi we analogously nd ab i Pi . Note that thereby the generation of is well-dened - the set of all s containing X is non-empty, as R itself is a (e.g. R is an ideal of R). And by the lemma we have just proved, the intersection is a again. P Proof of (3.9): First suppose M is an R-module and Pi m M are R-submodules and let P := i Pi M . In particular we get 0 Pi for any i I and hence 0 P . And if we are given any x, y P this means x, y Pi for any i I . And as any Pi m M is an R-submodule this implies x + y Pi which is x + y P again, as i has been arbitrary. Likewise consider any a R, then axi Pi again and hence ax P . Altogether P m M is an R-submodule of M . Now suppose M is an R-semi-algebra and Pi b A are R-sub-semialgebras and let P := i Pi M again. In particular Pi m M and, as we have already seen, this implies P m M . Thus consider any f , g P that is f , g Pi for any i I . As any Pi is an R-subsemi-algebra we get f g Pi again and hence f g P , as i has been arbitrary. Thus P is an R-sub-semi-algebra again. Next suppose M is an R-algebra with unit element 1, Pi a M are R-subalgebras and let P := Pi once more. Then we have just seen that P b M again. But as Pi a M we have 1 Pi again. And as this is true for any i I this implies 1 P , such that P a M . Finally consider the R-semialgebra A and the R-algebra-ideals ai a A and let a := i ai A. As our rst claim we have already shown a m A, thus consider any f a and any g M . Then we have f ai for any i I and as this is an R-sub-semialgebra this implies f g and gf ai again. And as i is arbitrary this nally is f g and gf a such that a a A, again. 318 17 Proofs - Rings and Modules

P Proof of (1.50): Let us denote the set a := { i ai xi } of which we claim a = X m . First of all it is clear that X a, as x = 1x a for any x X . Further it is clear that a m R is a left-ideal of R: choose any x X , then 0 = 0 x a, and if b R, x = a1 x1 + + am xm and y = b1 y1 + + bn yn a then bx = (ba1 )x1 + + (bam )xm a and x + y = a1 x1 + + am xm + b1 y1 + + bn yn a. Thus by denition of X m we get X m a. Now suppose b m R is any left-ideal of R containing X b, Then for any ai R and any xi X b we get ai xi b and hence i ai xi b, too. That is a b, and as b has been arbitrary this means a X m , by the denition of X m . Denote the set P := { i j xi,j } of which we claim P = X s again. Trivially we have X P and further it is clear that P s R is a sub-semialgebra of R: choose any x X , then 0 = 0 x P , and if a = i j xi,j and b = k l yk,l , then a = i (xi,1 ) j 2 xi,j P , a + b = i j xi,j + k l yk,l P and by general distributivity also ab = i j ( j xi,j l yk,l ) P . Thus by denition of X s we get X s P . Now suppose Q s R is any sub-semi-ring of R. Then for any and any x X Q we also get x Q and hence X Q. Now consider any xi,j X Q then j xi,j Q and hence j xi,j Q. That is P Q, and as Q has been arbitrary, this means P X s , by denition of X s . By construction P := X { 1 } s is a sub-semiring containing 1, in other words a subring. And as X P this implies X r P . And as any subring Q r R contains 1 we get X Q = X { 1 } Q = P Q. This also proves X r P , as Q has been arbitrary. Now suppose R is a eld, and let P := ab1 | a, b X r , b = 0 . First of all 0 = 0 11 and 1 = 1 11 P . And if ab1 and cd1 P , then we get (ab1 + cd1 = (ad + bc)(bd)1 P and (ab1 )(cd1 ) = (ab)(cd)1 P . And if ab1 = 0 then we in particular have a = 0 and hence (ab1 )1 = ba1 P . Clearly X P , as even X X r . Altogether P is a eld containing X and hence X f P . And if conversely Q R is a eld containing X , then X r Q. Thus is a, b Q with b = 0 then b1 Q and hence ab1 Q, as Q is a eld. But this proves P Q and thus P X f . P Proof of (3.11): Let us denote the set P := { i ai xi } of which we claim P = X m . First of all it is clear that X P , as x = 1x P for any x X . Further it is clear that P m M is a submodule of M : choose any x X , then 0 = 0 x P , and if a R, x = a1 x1 + + am xm and y = b1 y1 + + bn yn P then ax = (aa1 )x1 + + (aam )xm P and x + y = a1 x1 + + am xm + b1 y1 + + bn yn P . Thus by denition of X m we get X m P . Now suppose Q m M is any 319

submodule of M containing X Q. Then for any ai R and any xi X Q we get ai xi Q and hence i ai xi Q, too. That is P Q, and as Q has been arbitrary this means P X m , by the denition of X m . Denote the set P := { i ai j xi,j } of which we claim P = X b . Again it is clear, that X P , as x = 1x P for any x X . Further it is clear that P b A is a sub-semialgebra of A: choose any x X , then 0 = 0 x P , and if a R, f = i ai j xi,j and g = k bk l yk,l , then af = i (aai ) j xi,j P , f + g = i ai j xi,j + k bk l yk,l P and by general distributivity f g = i j (ai bj )( j xi,j l yk,l ) P . Thus by denition of X b we get X b P . Now suppose Q b A is any sub-semi-algebra of A. Then for any ai R and any xi,j X Q we get j xi,j Q, hence ai j xi,j Q and nally i ai j xi,j Q. That is P Q, and as Q has been arbitrary, this means P X b , by denition of X b . By construction P := X { 1 } b is a sub-semi-algeba containing 1, in other words a subalgebra. And as X P this implies X a P . And as any subalgebra Q a A contains 1 we get the implications X Q = X { 1 } Q = P Q. This also proves X a P , as Q has been arbitrary. P Proof of (1.51): We rst have to prove, that a b, a + b and a b truly are ideals of R. In the case of a b this has already been done in (1.47). For a + b it is clear that 0 = 0 + 0 a + b. and if a + b and a + b a + b then also (a + b) + (a + b ) = (a + a )+(b + b ) a + b and (a + b) = (a)+(b) a + b. Thus consider any f R, as a and b are ideals we have f a, af a and f b, bf b again. Hence we get f (a + b) = (f a) + (f b) and (a + b)f = (af ) + (bf ) a + b. Likewise 0 = 0 0 a b. And if f = i ai bi and g = j cj dj a b then it is clear from the denition that f + g a b. Further f = i (ai )bi a b and for any arbitrary h R we also get hf = i (hai )bi and f h = i ai (bi h) a b by the same reasoning as for a + b. Thus it remains to prove the chain of inclusions a b a b a a + b. Hereby a a + b is clear by taking b = 0 in a + b a + b. And a b a is true trivially. Thus it only remains to verify a b a b. That is we consider f = i ai bi a b. Since ai a and a is an ideal we have ai bi a and hence even f a. Likewise we can verify f b such that f a b as claimed. P Proof of (1.52): The associativity of is clear (from the associativity of the locical and ). The associativity of ideals (a + b) + c = a + (b + c) is immediate from the associativity of elements (a + b) + c = a + (b + c). The same is true for the commutativity of and +. Next we prove (a b) c a (b c). That is we are given an element of (a b) c which is of the form
n j =1 m m n

ai,j bi,j
i=1

cj =
i=1 j =1

ai,j (bi,j cj )

320

17 Proofs - Rings and Modules

For some ai,j a, bi,j b and cj c. However from the latter representation it is clear that this element is also contained in a (b c), which establishes the inclusion. The converse inclusion can be shown in complete analogy, which proves the associativity (a b) c = a (b c). And if R is commutative the commutativity of ideals a b = b a is immediate from the commuativity of the elements ab = ba. Thus it remains to check the distributivity
a (b + c) = (a b) + (a c)

For the inclusion we are given elements ai a, bi b and ci c (where i 1 . . . m) such that i ai (bi + ci ) a (b + c). But it is already obvious that i ai (bi + ci ) = i ai bi + i ai ci (a b) + (a c). And for the converse inclusion we are given ai , aj a, bi b and cj c such that i ai bi + j aj cj (a b)+(a c). This however can be rewritten into the form i ai bi + j aj cj = i ai (bi + 0) + j aj (0 + cj ) a (b + c). In complete analogy it can be shown, that (a + b) c = (a c) + (b c). P Proof of (1.54): By assumtion we have a = X i and b = Y i , in particular X a and Y b. Hence we get X Y a b a + b and therefore X Y i a + b (as a + b is an ideal of R). For the converse inclusion we regard f = i ai xi bi a and g = j cj yj dj (note that this representation with xi X and yj Y is allowed, due to (1.50)). And hence we get f + g = j cj yj dj X Y i . Together we have found a + b = X Y i . i ai xi bi + For a b = X Y i we have assumed R to be commutative. The one incluion is clear: as X a and Y b we trivially have X Y a b. And as a b is an ideal of R this implies X Y i a b. For the converse inclusion we regard f a and g b as above. Then f g = i j ai bi xi yj cj dj X Y i (due to the commutativity of R) which also establishes X Y i = a b. P Proof of (1.55): (i) First of all it is clear that b \ a b. And hence we nd b \ a b i = b (as b is an ideal already). Thus we have proved
b\a
i i

Now x any b b \ a (this is possible by assumption a b) and consider any a a. Then a b b but a b a (else let := a b a, then b = a a, a contradiction). Thus a b b \ a and hence a = b +(a b) b \ a i . As a has been arbitrary this yields a b \ a i and thereby we have also proved the converse inclusion
b = a (b \ a) a b \ a
i

= b\a

(ii) By (1.50) the ideal generated by the union of the ai consists of nite sums of elements of the form ai xi bi where ai , bi R and xi ai . But as ai is an ideal we get ai xi bi ai already. And conversely any nite sum of elements xi ai can be realized, by taking ai = bi = 1. 321

(iii) Here we just rewrote the the equality in (ii) in several ways: given a sum iI ai where ai ai and := { i I | ai = 0 } is nite then iI ai = i ai by denition of arbitrary sums. And likewise this is just k ak where n := #, = { i(1), . . . , i(n) } and ak = ai(k) . (iv) The general distributivity rule can be seen by a straightforward computation using the general rules of distributivity and associativity. First note that by denition and (iii) we have
ai b =
i I

n(i)

ai,k bi,k ai,k ai , bi,k b

k=1

Now let n := max{ n(i) | i } and for any k > n(i) let ai,k := 0 ai . Then we may rewrite the latter set in the form
n

ai b =
i I k=1 i

ai,k bi,k

ai,k ai , bi,k b

Let us now regart the other set involved. Again we will compute what this set means explictly - and thereby we will nd the same identity as the one wh have just proved n ai b = ai,k bk ai,k ai , bk b
iI k=1 i(k)

Then (in analogy to the above) we take := (1) (n), ai,k := 0 for i (k ) and bi,k := bk . Then this turns into
n

ai
i I

b =
k=1 i

ai,k bi,k

ai,k ai , bi,k b

Proof of (1.56): (i) (a) = (b): 1 R = a + b = a + b | a a, b b is clear. Hence there are a a and b b such that 1 = a + b. Now let a := a a then 1 = (a) + b and hence b = 1 + a (1 + a) b. (b) = (c): let b (1 + a) b, that is there is some a a such that b = 1 + a . Again we let a := a a, then a + b = 1. (c) = (a): by the same reasoning as before we nd 1 = a + b a + b. And as a + b is an ideal of R, this implies a + b = R. (ii) As a and b are coprime there are some a a and b b such that 322 17 Proofs - Rings and Modules

a + b = 1 - diue to (i). Using these we compute

i+ j i+ j i+j

1 = 1

= (a + b)

=
h=0 i+j

i + j h i+j h a b h ah bi+j h

=
h=0 i

i + j h i+j h a b + h

h=i+1 j

= bj
h=0

i + j h i h a b + ai h
i

h=1

i + j h j h a b i+h

b +a

(iii) We will use induction on k . The case k = 1 is trivial and hence we found the induction at k = 2. In this case let us rst choose a1 , a2 a, b1 b1 and b2 b2 such that a1 + b1 = 1 and a2 + b2 = 1. Then b1 b2 = (1 a1 )(1 a2 ) = 1 a1 a2 + a1 a2 This implies 1 = (a1 + a2 a1 a2 ) + b1 b2 a + b1 b2 which rests the case k = 2. So we now assume k 2 and are given a + bi = R for any i 1 . . . k + 1. By induction hypothesis we in particular get a + b1 . . . bk = R. And together with a + bk+1 = R and the case k = 2 this yields the induction step a + (b1 . . . bk )bk+1 = R. (iv) We will prove the two inclusions by induction on k seperately. The case k = 1 is trivial, so we will start with k = 2 only. In this case the inclusion a1 a2 a1 a2 has already been proved. And if k 2 then (a1 . . . ak )ak+1 (a1 . . . ak ) ak+1 (a1 ak ) ak+1 due to the case k = 2 and the induction hypothesis respectively. For the converse inclusion we start with k = 2 again. And as a1 and a2 are coprime we may nd a1 a1 and a2 a2 such that a1 + a2 = 1. If now x a1 a2 is an arbitrary element then x = x1 = x(a1 + a2 ) = xa1 + xa2 = a1 x + a2 x a1 a2 This proves a1 a2 a1 a2 and hence we may commence with k 2. In this case we are given some x a1 ak ak+1 . In particular this is x a1 ak and hence x a1 . . . ak by induction hypothesis. Thus we get x (a1 . . . ak ) ak+1 . But because of (iii) (using a := ak+1 and bi := ai ) we know that a1 . . . ak and ak+1 are coprime. Therefore we may apply the case k = 2 again to nally nd x (a1 . . . ak )ak+1 . P Proof of (1.57): We rst prove that is an equivalence relation: Reexivity a a is clear, as a a = 0 a. Symmetry if a b then a b a and hence b a = (a b) a which proves b a. Transitivity if a b and b c then 323

a c = (a b) + (b c) a such that a c. If now a R then [a] = { b R | a b } = { b R | a b a } = {b R | h a : a b = h} = {a + h | h a} = a + a Now assume that a i R even is an ideal, then we have to prove, that R/a is a ring under the operations + and as we have introduced them. Thus we prove the well-denedness: suppose a + a = a + a and b + a = b + a then a a and b b a and hence (a + b) (a + b ) = (a a ) + (b b ) a. This means a + b a + b and hence (a + b) + a = (a + b ) + a. Analogously ab a b = ab a b + a b a b = (a a )b + a (b b ) As a is an ideal both (a a )b and a (b b ) are contained in a and hence also the sum ab a b a. This means ab a b and hence ab + a = a b + a. Hence these operations are well-dened. And all the properties (associativity, commutativity, distributivity and so forth) are immediate from the respective properties of R. This also shows that 1 + a is the unit element of R/a (supposed 1 is the unit element of R). P Proof of (3.13): We rst prove that truly is an equivalence relation, the reexivity x x is clear, as x x = 0 P . For the symmetry suppose x y , that is y x P and hence x y = (y x) = (1)(y x) P . For the transitivity we nally consider x y and y z . That is y x P and z y P and this yields z x = (z y ) + (y x) P . As in the proof of (1.57) it is clear that [x] = x + P , as [x] = { y M | y x P } = x + P Next we will prove the well-denedness of the addition and scalarmultiplication. Thus consider x + P = x + P and y + P = y + P , that is x x P and y y P . As P is an R-submodule we get (x + y ) (x + y ) = (x x) + (y y ) P and this again is (x + y ) + P = (x + y ) + P . Likewise for any a R we get (ax ) (ax) = a(x x) P , which is ax + P = ax + P . Now we immediately see, that M/P is an R-module again. The associativity and commutativity of + are inherited trivially from M . And the same is true for the compatibility properties of . The neutral element of M/P is given to be 0 + P , as for any x + P M/P we get (x + P )+(0+ P ) = (x +0)+ P = x + P . And the inverse of x + P M/P is just (x) + P , as (x + P ) + ((x) + P ) = (x + (x)) + P = 0 + P . Finally suppose A is an R-(semi)algebra, then we want to show that A/a is an R-(semi)algebra again. As the R-algebraideal a is an Rsubmodule A/a already is an R-module, by what we have already shown. Next we have to verify the well-denedness of the multiplication. Thus consider f + a = f + a and g + a = g + a, then 324 17 Proofs - Rings and Modules

(f g ) (f g ) = f (g g ) + (f f )g a, by assumption on a. Now it is clear that A/a inherits the associativity of the multiplication from A. Likewise the compatibility properties of , + and are inherited trivially. As an example let us regard the identity of a(f + a) (g + a) = (af + a)(g + a) = (af )g + a = a(f g ) + a (f + a) a(g + a) a (f + a)(g + a) = (f + a)(ag + a) = f (ag ) + a = a(f g ) + a = a(f g + a) = a(f g ) + a P Proof of (1.59): First of all we have (a) = (c) because b b + nZ = a + nZ gives some c nZ such that b = a + c. And c nZ means nothing but c = kn for some k Z. For (c) = (b) we use division with remainder (1.27) to write a = pn + r and b = qn + s such that r = a mod n and s = b mod n. Then we can compute qn + s = b = a + kn = pn + r + kn = (p + k )n + r which yields s r = (p + k q )n such that |s r| = |p + k q | n. As r and s both are positive, this implies that 0 |s r| < max{ s, r } < n. But this can only be, if |p + k q | = 0 such that q = n + k and r = s. And the latter just is the claim (b). It remains to prove (b) = (a): continueing with the notation we have r = s and hence b a = qn + s qn r = (q p)n. Letting k := q p we have b a = kn nZ. That is a and b are equivalent and hence have the same equivalency classes a + nZ = b + nZ. P Proof of (1.61): We rst prove that b/a truly is an ideal of R/a. Thus regard a + a and b + a b/a, that is a, b b. Then we clearly get a + b and a b such that (a + a) + (b + a) = (a + b) + a b/a and (a + a) = (a) + a b/a. Further if f + a R/a (that is f R) then f b, bf b, as b is an ideal. And hence (f + a)(b + a) = (f b) + a b/a, likewise (b + a)(f + a) b/a. Next we prove that b := {b R | b + a u} i R is an ideal, supposed u i R/a is an ideal. Thus consider a, b b then (a + b) + a = (a + a) + (b + a) u and (a) + a = (a + a) u which proves a + b and a b. And if f R then (f b) + a = (f + a)(b + a) u which proves f b b (likewise bf b. Hence b is an ideal, and a b is clear, as for any a a we get a + a = 0 + a u. Thus we have proved that the maps b b/a and u b are well dened. We now claim that they even are mutually inverse. For the rst composition we start with a b i R. Then
b b a

bR|b+ab a

b|bb

= b 325

For the converse composition we start with an ideal u i R/a, which is mapped to b := {b R | b + a u}. Now it is easy to see that
u b b a = {b + a | b + a u} = u

Next we prove that this correspondence respects intersections, sums and products of ideals as claimed. That is b and c i R are ideals with a b and a c. Then
b b c c+ c c c = c =

b+a

bbc

= bc c b b, c c = b+c c

(b + a) + (c + a) (b + c) + a
n

=
b c c c =

b b, c c

(bi + a)(ci + a)
i=1 n

bi b, ci c = bc c

=
i=1

bi ci

+a

bi b, ci c

The nal statement of the lemma is concerned with the generation of ideals. For the inclusion we consider any u X/a i . That is there are ai + a, bi + a R/a and xi + a X/a such that
k

u =

(ai + a)(xi + a)(bi + a)

i=1 k

=
i=1

ai xi bi

+a X

+a

And for the converse inclusion we are given some a a and y X i . This again means that there are some ai , bi R and xi X such that y = i ai xi bi . We now undo the above computation
k

(a + y ) + a = y + a =
i=1

(ai + a)(xi + a)(bi + a) X a

We now prove that this correspondence even interlocks radical ideals. First note that b i R is radical i b = b. Now remark that (b + a)k = bk + a b/a is (by denition) equivalent to bk b. Thus we found
b

a =

And in particular we see that b/a is a redical ideal if and only if b is a radical ideal. And this has been claimed. If p i R/a is prime then regard any a, b R. Then (a + a)(b + a) = ab + a p/a is equivalent, to ab p which again is equivalent, to a p or b p (as p is prime). And again this is equivalent, to a + a p/a or b + a p/a. Hence p/a is prime, too. For the converse implication 326 17 Proofs - Rings and Modules

just pursue the same arguments in inverse order. As the correspondence b b/a maintains the inclusion of subsets, it is clear that maximal elements mutually correspond to each other. That is m/a is maximal, if and only if m/a is maximal. P Proof of (3.14): (i) We will rst prove the eqivalency x + L P/L x P . The one implication is trivial: if x P , then x + L P/L by denition. Conversely suppose there is some p P such that x + L = p + L P/L. By denition of M/L this means x p L, that is x p = for some L P . And therefore x = p + P , as p, P and P m M . Now we are ready to prove P/L m M/L, clearly 0 = 0 + L P/L, as we have 0 P . Now consider x + L and y + L P/L, by what we have just seen this means x and y P . As P m M this implies x + y P and hence (x + L) + (y + L) = (x + y ) + L P/L again. Finally consider any a R, as x P and P m M this implies ax P and hence also a(x + L) = (ax) + L P/L. Altogether P/L is a submodule of M/L. (ii) In (i) we have seen that P P/L is a well-dened map, so next we will point out that U (U ) := { x M | x + L U } is well-dened, too. That is we have to verify (U ) m M and L (U ). The containment L (U ) is clear, if L then + L = 0 + L = 0 U and hence (U ). In particular 0 (U ). Next suppose we age given x, y (U ), that is x + L, y + L U . Then (x + y ) + L = (x + L) + (y + L) U as U m M/L and hence x + y (U ) again. Likewise for any a R we get (ax) + L = a(x + L) U and hence ax (U ), as x + L U and U m M/L. Thus both maps given are well-dened. It remains to verify that they are mutually inverse: the one direction is easy, consider any U m M/L, then U (U ) (U )/L = { x + L | x + L U } = U . Conversely consider any L P m M , then P P/L (P/L), yet using (i) we obtain (P/L) = P from the following computation P L = xM |x+LP L = {x M | x P } = P

(iii) To prove this claim we just have to iterate the equivalency in (i): x + L ( i Pi )/L is equivalent to x i Pi , which again is equivalent to i I : x Pi . Now we use (i) once more to reformulate this into the equivalent statement i I : x + L Pi /L, which obviously translates into x + L i (Pi /L) again. (iv) We want to verify i (Pi /L) = ( i Pi )/L, to do this recall the explicit description of an arbitrary sum of submodules for the Pi /L m M/L Pi
iI

=
i

(xi + L)

I nite, xi + L Pi L 327

Next recall the equivalency in (i), this allows us to simply substitute xi + L Pi /L by xi Pi . Further - by denition of + in M/L - we can rewrite the sum i (xi + L) = ( i xi ) + L. Together this yields Pi
i I

=
i

xi

+L

I nite, xi Pi

That is the term on the right hand side is (apart from the +L in ( i xi ) + L) nothing but the sum i Pi again (by the explicit description of arbitrary sums of the submodules Pi . That is we have already arrived at our claim i (Pi /L) = ( i Pi )/L. (v) Let us denote Pi := Pi + L, then it is clear that L Pi for any i I . But on the other hand i Pi = i (Pi + L) = ( i Pi ) + L as the nitely many occurances of elements i L can be composed to a single occurance of := i i L. But by assumption L is already contained in i Pi and hence ( i Pi ) + L = i Pi . That is we have obtained i Pi = i Pi . Now statement (iv), that has just been proved, yields ( i Pi )/L = ( i Pi )/L = i (Pi /L), as claimed. (vi) To prove this identity we directly refer to the denition of the Rsubmodule generated by X/L M/L. And this is given to be X L
m

U m M L

L U

By (ii) the submodules U of M/L correspond bijectively to the submodules P of M containing L. I.e. we may insert P/L for U , yielding X L
m

L P m M, X L P L

Clearly X/L P/L translates into x X : x + L P/L, and due to the equivalency in (i) this is x X : x P , i.e. X P , therefore X L
m

L P m M, X P

Of course L P and X P can be written more compactly as X L P . And due to (iii) the intersection commutes with taking to quotients, such that we may reformulate X L
m

{ P | P m M, X L P }

That is we have arrived at X/L m = X L m /L. But by the explicit representation of the generated submodules (3.11) it is clear, that X L m = X m + L m . And as L already is a submodule of M we have L m = L. Altogether X/L m = ( X m + L)/L. P Proof of (3.16): (i) We will rst prove that ann(X ) is a left-ideal of R. As this is just the intersection of the annihilators ann(x) (where x X ) it suces to 328 17 Proofs - Rings and Modules

verify ann(x) m R for any x M , due to (1.47). Now 0 ann(x) is clear, as 0 x = 0 for any x M . Now consider any two a, b ann(x), that is ax = 0 and bx = 0. Then (a + b)x = (ax) + (bx) = 0 + 0 = 0 and hence a + b ann(x) again. Finally consider any r R, then (ra)x = r(ax) = r 0 = 0 and hence ra ann(x), as well. Altogether we have ann(x) m R and hence ann(X ) m R is a left-ideal, as it is an intersection of such. (i) Next we want to prove that ann(M ) i R even is an ideal. As we already know ann(M ) m R it only remains to prove that ar ann(M ) for any a ann(M ) and any r R. As a ann(M ) we have ay = 0 for any y M . Now consider any x M , then rx M , too as M is an R-module. Thus (ar)x = a(rx) = 0 and as x has been arbitrary this means ar ann(M ) again. (ii) Let us denote : R/ann(x) Rx : b := b + ann(x) bx, we will rst prove the well-denedness and injectivity of this map: ax = (a) = (b) = bx is equivalent to (a b)x = 0, that is a b ann(x) or in other words a = a + ann(x) = b + ann(x) = b. The surjectivity of is clear, as any b R is allowed. And also is a homomorphism of R-modules, as by denition of the algebraic operations of R/ann(x) we have (a + b) = (a + b) = (a + b)x = (ax) + (bx) = (a) + (b) and (ab) = (ab) = (ab)x = a(bx) = a(b). Nota that this claim could have also been proved by regarding the epimorphism : R Rx : b bx, noting that its kernel is kn() = ann(x) and invoking the rst isomorphism theorem of R-modules to get . (iii) We rst have to check the well-denedness of the scalar multiplication. To do this consider any b + a = c + a, that is a := c b a annR (M ). If now x M is an arbitrary element, then cx = (a + b)x = ax + bx = 0 + bx = bx, as a annihilates any x M . Hence (b + a, x) bx does not depend on the specic b R chosen. As we have only changed the scalar-multiplication on M , it remains to prove, that it still satises the properties (M) in the denition of modules. Thus consider any a, b R and any x, y M . Then we can compute (a + a)(x + y ) = a(x + y ) = ax + ay = (a + a)x + (a + a)y ((a + a) + (b + a))x = ((a + b) + a)x = (a + b)x = ax + bx = (a + a)x + (b + a)x ((a + a)(b + a))x = ((ab) + a)x = (ab)x = a(bx) = (a + a)(bx) = (a + a)((b + a)x) (1 + a)x = 1x = x (iv) As R = 0 we have 1 = 0 and as also 1 0 = 0 we already have 0 tor(M ). Now consider any two x, y tor(M ), that is there are some 0 = a, b R such that ax = 0 and by = 0. As R is an integral domain we also get ab = 0. Now compute (ab)(x + y ) = (ab)x +(ab)y = (ba)x + (ab)y = b(ax) + a(by ) = b 0 + a 0 = 0 + 0 = 0. That is (ab)(x + y ) = 0 and hence x + y M , as ab = 0. Finally consider any 329

r M , then we need to show rx tor(M ). But this is clear from a(rx) = (ar)x = (ra)x = r(ax) = r 0 = 0. (iv) Let us abbreviate T := tor(M ), now consider any x + T tor(M/T ). That is there is some 0 = a R such that (ax)+ T = a(x + T ) = 0+ T . This is just ax T , that is there is some 0 = b R, such that (ba)x = b(ax) = 0. As R is an integral domain, we get ba = 0 and this means x T , such that x + T = 0 + T . Thus we have just proved tor(M/T ) { 0 + T } = { 0 } and the converse containment is clear. (v) (a) = (c): Consider any a R and any x M such that ax = 0. If a = 0, then we are done, else we have x tor(M ) = { 0 } (by denition of tor(M ), as ax = 0 and a = 0) and this is x = 0 again. (v) (c) = (b): Consider any a zdR (M ), that is there is some x M such that x = 0 and ax = 0. Thus by assumption (c) we nd a = 0, and as a has been arbitrary, this means zdR (M ) { 0 }, as claimed. (v) (b) = (a): As R = 0 we have 1 = 0 and as 1 0 = 0 this implies 0 tor(M ). Conversely consider any x tor(M ), that is there is some a R such that a = 0 and ax = 0. Suppose we had x = 0, then a zdR (M ) { 0 } (by denition of zdR (M ), as x = 0 and ax = 0) which is a = 0 again. But as a = 0 this is a contradiction. Thus we nexessarily have x = 0 and as x has been arbitrary this means tor(M ) { 0 }, as well. (vi) Let us abbreviate P := X m , then by denition it is clear that X P and hence ann(P ) ann(X ). For the converse inclusion consider any a ann(X ), then we have to show a ann(P ). That is ap = 0 for any p P . By (3.11) any p P is given to be p = i ai xi for some ai R and xi X . Therefore ap = i (aai )xi = i ai (axi ) = i ai 0 = 0 as R is commutative. Altogether that is ann(X ) = ann(P ). Now suppose M = X m where X = { xi | i I } and consider the map : R M I dened by (a) := (axi ) (where i I ). If now a ann(M ) = ann(X ) then axi = 0 for any i I and hence (a) = 0 is clear. Therefore we have ann(M ) kn(). Conversely (a) = 0 means axi = 0 for any i I , that is a ann(X ) = ann(M ) and hence even kn() = ann(M ). Thereby R/ann(M ) M I is well-dened and injective according to the rst isomorphism theorem (3.51.(ii)) (vii) First suppose that x = 0, then ax = a0 = 0 is clear for any a S and hence ann(x) = S . Secondly consider x = 0 and a ann(x), that is ax = 0. Suppose we had a = 0, too, then (as S is a skew-eld) there was some a1 S such that a1 a = 1. Then we would be able to compute x = 1x = (a1 a)x = a1 (ax) = a1 0 = 0, a contradiction to x = 0, such that a = 0 and therefore ann(x) = 0. (viii) Let p = ann(x) be a maximal annihilator ideal of M with p = R. Further consider any a, b R with ab p but b p, then it remains to verify a p. As b p we have bx = 0. Now consider a := ann(bx), as bx = 0 we have 1 a and hence a = R. Also if p p, then p(bx) = b(px) = b0 = 0 and hence p a, which means p a. But by maximality of p this is p = a. Now as ab p we have a(bx) = (ab)x = 0 such that a a = p, as has been claimed. 330 17 Proofs - Rings and Modules

(ix) Choose any x M , yet x = 0. Then Rx = { bx | b R } is a non-zero submodule of M . But as 0 = Rx m M and M is simple we get Rx = M . Hence we get
m := annR (M ) = { a R | y Rx : ay = 0 }

= =

{ a R | b R : abx = 0 } { a R | ax = 0 } = ann(x)

Here { a R | b R : abx = 0 } ann(x) is clear - just take b = 1. And the converse inclusion follows easily: if ax = 0, that abx = b(ax) = b0 = 0 as well. So by (ii) we obtain an isomorphy of R-modules R
m M : a + m ax

Now, as M is simple, so is R/m. That is 0 = m/m is a maximal ideal of R/m. That is R/m is a eld and this again means that m is a maximal ideal of R. (x) Let us denote the set of all proper, pure submodules P of M by P , that is P := { P m M | P = M, P is pure }. As M is torsion free 0 m M is a pure submodule of M (ax = 0 implies a = 0 or x = 0). Also 0 = M such that P is non-empty. Now consider a chain (Pi ) P in P , then we let Q :=
i I

Pi

As (Pi ) is a chain we know that Q is a submodule of M again. Now suppose we had Q = M . As M is nitely generated we nd some mk M such that M = lh(x1 , . . . , xn ). And as (for any k 1 . . . n) xk M = Q there is some i(k ) I such that xk Pi(k) . Taking to the maximum of the Pi(k) we nd xk Pi for some i I . But this would imply Pi = M in contradiction to our assumption. Hence Q = M . In fact Q even is pure again: suppose we have ax Q, that is ax Pi for some i I . And as this Pi has been pure we nd x Pi which yields x Q again. Hence every chain (Pi ) in P has an upper bound Q P and therefore P has a maximal element, due to the lemma of Zorn. P Proof of (3.17): Let us rst prove the identity part in the modular rule, starting with (Q U )+ P Q (P + U ). Thus we consider any x Q U and p P and have to verify x + p Q (P + U ). As x Q and p P Q we have x + p Q. And as also x U we have x + p U + P hence the claim. Conversely for (Q U ) + P Q (P + U ) we are given any p P and u U such that q := p + u Q and we have to verify p + u (Q U ) + P . But as u = q p and p, q Q we nd u Q and hence u Q U . And this already is p + u = u + p (Q U ) + P . 331

 So it ramains to verify the implication in the modular rule. As P Q by assumption it remains to prove Q P . Thus consider any q Q, then in particular q Q + U = P + U . That is there are p P and u U such that q = p + u. And as p P Q this yields u = q p Q, such that u Q U = P U . In particular u P and hence q = p + u P , as claimed. P Proof of (3.30): (i) Recall that aP has been dened to be the submodule generated by elements of the form p with a and p P . In the light of (3.11) this turns into
aP

= lh { p | a, p P }
n

=
i=1

bi i pi

1 n N, bi R, i a, pi P

But as a is an ideal and i a we nd that ai := bi i a again. And hence aP is contained in the set given in claim (i). Conversely any a1 p1 + + an pn can be realized by taking b1 = = bn = 1 and i = ai . Thus the set given in claim (i) is also contained in aP :
n

aP

=
i=1

ai pi

1 n N, ai a, pi P

(ii) As x aP there are some 1 , . . . , m a and q1 , . . . , qm P such that x = 1 q1 + + m qm as we have seen in (i). But as P is generated by { pi | i I } any qk can be written in the form qk =
iI

bk,i pi

for some bk,i R of which only nitely many are non-zero. Inserting these equations into our presentation of x we nd
m m m

x =
k=1

k qk =
k=1

k
i I

bk,i pi

=
iI k=1

k bk,i

pi

As a is an ideal and any k a we also nd ai := k k bk,i a. And as there are only nitely many i I such that bk,i = 0, all but nitely many ai are zero. Let us denote these non-zero ai a by a1 , . . . , an and the corresponding pi by p1 , . . . , pn , then we have arrived at the claim x = a1 p1 + + an pn . (iii) As ap P for any a a R and any p P , so is the submodule generated by these elements - and this is aP . (iv) Considering the submdules P and Q with P Q we nd that for any a a and any p P Q we get ap aQ and hence aP aQ. 332 17 Proofs - Rings and Modules

(v) We rst prove that a(M/P ) is contained in (aM + P )/P , to do this, consider any element of a(M/P ), by denition this is of the form (where ak a and xk M )
n n

ak (xk + P ) =
k=1 k=1

ak xk

+ P aM + P P

That is a(M/P ) (aM + P )/P , conversely consider any element of aM + P , that is we are given any ak a, xk M and p P , then
n n

ak xk
k=1

+p+P

=
k=1 n

ak xk

+P

=
k=1

ak (xk + P ) a M P

That is (aM + P )/P a(M/P ) as well and hence we have proved the equality of these sets, which had been claimed. (vi) By denition (3.13) of the quotient module M/aM is a well-dened R-module. Hence all we have to do is checking the well-denedness of the scalar-multiplication. We will present two ways of demonstrating this: (1) clearly we get a(x + aM ) = ax + aM = 0 + aM for any a a and any x M . That is a annR (M/aM ), such that we could simply cite (3.16.(iii)). The more instructive way (2) is a direct computation however: consider b + a = c + a and x + aM = y + aM , that is a := c b a and p := y x aM . Then cy bx = cy by + by bx = (c b)y + b(y x) = ay + bp. But as a a we nd that ay aM and as p aM already we see that cy bx aM . But this is bx + aM = cy + aM and hence (b + a, x + aM ) bx + aM does not depend on the specic choice of b R or x M . (vii) We rst have to check the well-denedness of /a: to do this we consider any x + aM = y + aM in M/aM , that is p := y x aM . By definition of aM this means that p can be written as p = a1 p1 + + an pn for some ai a and pi M . Then we can compute
n

(y ) (x) = (p) =
i=1

ai (pi ) aN

That is (y ) + aN = (x) + aN , such that the map /a truly is welldened. The properties of being a homomorphism are straightforward: consider any a R and any x, y M , then ((a + a)(x + aM )) = (ax + aM ) = (ax) + aN = a(x) + aN = (a + a) ((x) + aN ) = (a + a) (x + aN ) ((x + a) + (y + aM )) = ((x + y ) + aM ) = (x + y ) + aN = (x) + (y ) + aN = ((x) + aN ) + ((y ) + aN ) 333

(vii) If : M N is surjective, then clearly /a : M/aM N/aN is surjective as well: given any y + aN N/aN we choose any x M with (x) = y , then (/a)(x + aM ) = (x) + aN = y + aN . If even is bijective, it remains to check the injectivity of /a. To do this we regard the following homomorphism of R-modules : M N aN : x (x) + aN That is = , where denotes the canonical projection of N onto N/aN . Now by denition of we see kn = { x M | (x) aN } and therefore aM kn . If conversely (x) aN , then (x) = a1 y1 + + an yn for some ai a and yi N . Yet as is bijective we may dene xi := 1 (yi ), that is yi = (xi ) and hence (x) = a1 (x1 ) + . . . an (xn ) = (a1 x1 + + an xn ) (aM ) By the bijectivity of this means x aM again, such that we have proved kn = aM . Now note that - by deniton of - we have (/a)(x + aM ) = (x). Therefore kn a = kn aM = aM aM = 0 M aM That is /a has zero-kernel and hence is injective. As the surjectivity has already been proved we found the bijectivity of /a. (viii) Consider any elements z1 , . . . , zn Z , then by denition of the scalar multiplication on M/aM we already see the identity
n n

(ai + a)zi
i=1 i=1

ai zi

And as in both cases the ai R can be chosen completely arbitrary this implies that the R-linear hull of Z equals the (R/a)-linear hull. (ix) Clearly Q is an R-submodule of M/aM if and only if Q equals its R-linear hull. Likewise Q is an (R/a)-submodule of M/aM if and only if Q equals its (R/a)-linear hull. But by (viii) these linear hulls are equal, and hence we get Q submR M aM lhR (Q) = Q = lhR/a (Q) Q submR/a M aM

Now the explict description of the (R/a)-submodules of M/aM is just an application of the correspondence theorem (3.14) of modules. P Proof of (1.71): (i) Clearly (0R ) = (0R + 0R ) = (0R ) + (0R ) and subtracting (0R ) from this equation (i.e. adding (0R ) to both sides of the equation) 334 17 Proofs - Rings and Modules

we nd 0S = (0R ). Hence (a) + (a) = (a a) = (0R ) = 0S which proves that (a) is the negative (a). (ii) Analogous to the above we nd 1S = (1R ) = (uu1 ) = (u)(u1 ) and 1S = (1R ) = (u1 u) = (u1 )(u) which proves that (u1 ) is the inverse (u)1 . (iii) If a, b R then (a + b) = ((a) + (b)) = (a) + (b) and likewise for the multiplication. Hence is a homomorphism of semirings again. And in the case of rings and homomorphisms of rings we also get (1R ) = (1S ) = 1T . (iv) Consider any x, y S , as is surjective there are a, b R such that x = (a) and y = (b). Now 1 (x + y ) = 1 ((a) + (b)) = 1 (a + b) = a + b = 1 (x) + 1 (y ). Likewise we nd 1 (xy ) = 1 (x)1 (y ). And if is a homomorphism of rings, then 1 (1S ) = 1R is clear from (1R ) = 1S . (v) Since im() = (R) the rst statement is clear from (vi) by taking P = R. So we only have to prove, that kn() is an ideal of R. First of all we have 0R kn() since (0R ) = 0S by (i). And if a, b kn() then (a + b) = (a) + (b) = 0S + 0S = 0S and hence a + b kn(). Analogously (a) = (a) = 0S = 0S yields a kn(). Now let a kn() again and let b R be arbitrary. Then (ab) = (a)(b) = 0S (b) = 0S and hence ab kn() again. Likewise (ba) = (b)(a) = (b)0S = 0S which yields ba kn(). (vi) As P is a sub-semi-ring of R we have 0R P and therefore 0S = (0R ) (P ). Now assume that x = (a) and y = (b) (P ) for some a, b P . As P is a sub-semi-ring of R we have a + b, a and ab P again. And hence x + y = (a) + (b) = (a + b) (P ), x = (a) = (a) (P ) and xy = (a)(b) = (ab) (P ). Thus we have proved that (P ) s R is a sub-semi-ring again. And if R and S even are rings, P is a subring of R and is a homomorphism of rings, then also 1S = (1R ) (P ) as 1R P . (vi) And as Q is a sub-semi-ring of S we have 0S Q and therefore 0S = (0R ) implies 0R 1 (Q). Now suppose a, b 1 (Q), that is (a), (b) Q. Then (a + b) = (a) + (b), (a) = (a) and (ab) = (a)(b) Q, as Q s S . This means that a + b, a and ab 1 (Q) and hence 1 (Q) s S is a sub-semi-ring of S . And in the case of rings we also nd 1R 1 (Q) as in (vi) above. (vii) First assume that b m S is a left-ideal of S . We have already proved in (vi) that 1 (b) s S is a sub-semi-ring of S (since Q := b s R). Thus it only remains to verify the following: consider any a R and b 1 (b). Then (b) b and as b m S we hence get (ab) = (a)(b) b. Thus ab 1 (b) which means that 1 (b) is a leftideal of R again. And in the case that b i S even is an ideal, then by the same reasoning (ba) = (b)(a) b and hence ba 1 (b). (vii) Next assume that a m R is a left-ideal of R and that is surjective. Then (a) s S is a sub-semi-ring of S by (vi) for P := a s R. Thus it again remains to prove the following property: Consider any 335

y S and x = (a) (a) (where a a). As is surjective there is some b R with y = (b). And as ba a we hence nd yx = (b)(a) = (ba) (a). Thus (a) m S is a left ideal of S . And in the case that a i R even is an ideal, then by the same reasoning xy = (a)(b) = (ab) (a). (viii) Let : R S be a homomorphism of semi-rings. If (1R ) = 0S then for any a R we get (a) = (1R a) = 0S (a) = 0S and hence = 0. Else we have (1R ) = (1R 1R ) = (1R )(1R ) and dividing by 0 = (1R ) (i.e. multiplying by (1R )1 ) we hence nd 1S = (1R ). (ix) As we have seen in (v) kn() i R is an ideal of R. And as R is a skew-eld, it has precisely two ideals: { 0 } and R. In the rst case kn() = R we trivially have = 0. And in the second case kn() = { 0 } we will soon see, that is injective. P Proof of (1.72): If b b a + b then (b) = b + a (a + b)/a is clear and hence (b) (a + b)/a. Conversely for any a a and b b we get (a + b)+ a = b + a = (b) and hence (a + b)/a (b) as well. Now consider any r 1 ( (b)) that is (r) (b). That is there is some b b such that r + a = (r) = (b) = b + a. That is r b a and hence r b = a for some a a. In other words r = a + b a + b which proves 1 ( (b)) a + b. Conversely consider any a a and b b. Then (a + b) = a + b + a = b + a = (b) (b) and hence a + b 1 ( (b)). P Proof of (1.74): (i) By denition is surjective i any x S can be reached as x = (a) for some a R. And this is just S (R) = im(), or equivalently S = im(). And if is injective then there can be at most one a R such that (a) = 0S . And as (0R ) = 0S this implies kn() = { 0R }. Conversely suppose kn() = { 0R } and consider a, b R with (a) = (b). Then 0S = (a) (b) = (a b). Hence a b kn() such that a b = 0R by assumption. Hence is injective. (ii) The implication (a) = (b) is clear: just let := 1 then we have already seen, that is a homomorphism of (semi-)rings again. And the converse (b) = (a) is clear (in particular is an inverse mapping of ). The implication (b) = (c) is trivial - just let := and := . And for the converse (c) = (b) we simply compute: = 1 1S = ( ) = () = 1 1R = . Thus letting := = we are done. (iii) Suppose R is a ring, then let f := (1R ). If now x S is any element then we choose a R such that x = (a). And thereby we get f x = (1R )(a) = (1R a) = (a) = x and likewise xf = x. Hence f = 1S is the unit element of S and hence S is a ring, too. Further we have seen (1R ) = 1S such that is a homomorphism of rings. 336 17 Proofs - Rings and Modules

Suppose S is a ring, then let e := 1 (1S ). If now a R is any element, then (a) = (a)1S = (a)(e) = (ae) and hence a = ae. Likewise we see ea = a and hence e = 1R is the unit element of R. Hence R is a ring, too and (1R ) = 1S is a homomorphism of rings. (iv) Clearly the identity map 1 1R : R R : a a is both, bijective and a homomorphism of (semi-)rings for any (semi-)ring (R, +, ). But this is nothing but the claim 1 1R : R =r R. And if : R S is an isomorphism of (semi-)rings, then 1 is a bijective (this is clear) homomorphism of (semi-)rings again (this has already been proved). Hence the second implication. And if both : R =r S and : S =r T then is bijective and a homomorphism of (semi-)rings (as and are such). And this has been the third implication. P

Proof of (1.77): (i) We rst need to check, that is well-dened: thus consider b, c R such that b + a = c + a. That is c b a kn. And hence 0 = (c b) = (c) (b) which is the well-denedness (b) = (c). Now it is straightforward to see, that also is a homomorphism (b + a) + (c + a) (b + a)(c + a) = (b + c) + a = (bc) + a = (b + c)

= (b) + (c) = (b + a) + (b + a) = (bc) = (b)(c) = (b + a)(b + a) Thus always is a homomorphism of semi-rings. And if if R and S even are rings, then R/a is a ring as well with the unit element 1 + a. Thus if (1) = 1 is a ring-homomorphism, then (1 + a) = (1) = 1 is a ring-homomorphism, too. (ii) Let a := kn() and := , then by (i) is a well-dened (semi-)ringhomomorphism and the image of certainly is the image of im = R a = (R) = im

Hence the mapping : R/a im() is surjective by denition. But it also is injectitve (and hence an isomorphism) due, to (a) = (b) 0 = (a) (b) = (a b) a b kn() a + kn() = b + kn()

(iii) We will rst prove that b + R r S is a subring of S : clearly we have 0 = 0 + 0 b + R and (if R is a subring of S ) 1 = 0 + 1 b + R. Now 337

let b + a and d + c b + R, then (b + a) + (d + a) = (b + d) + (a + c) b + R (b + a) = (b) + (a) b + R (b + a)(d + c) = (bd + bc + ad) + ac b + R And hence b + R is a subring of S . And as b i R is an ideal of S with b b + R it trivially also is an ideal b i b + R. Now let us denote the restriction of the canonical epimorphism : S S/b to R by : RS b : aa+b Then is a homomorphism with kernel kn( ) = kn( ) R = b R. In particular this proves that b R i R is an ideal of R (as it is the kernel of a homomorphism). And it is straightforward to see, that im() = (b + R)/b. [For any a R we have a b + R and hence (a) (b + R)/b. And if conversely x = b + a b + R then (a) = a + b = x + b]. Thus we may apply the rst isomorphism theorem (ii) to obtain R R kn( ) = bR a+bR =r im( ) = b + R b (a) = a + b

(iv) Suppose c and d R with c + a = d + a. Then c d a b and hence c + b = d + b. Therfore we may dene the following mapping (which clearly is a homomorphism of (semi-)rings) : R aR b : c+ac+b And clearly kn() = b + a | b + b = 0 + b = b + a | b b = b/a. Hence we immediately obtain (iv) from the rst isomorphism theorem. P Proof of (1.91): (1) We will rst prove that a b i R S is an ideal. As a and b are ideals we have 0 a and 0 b and hence 0 = (0, 0) a b. Now consider (a, b) and (p, q ) a b. Then we have a + p a, b + q b and hence (a, b)+(p, q ) = (a + p, b + q ) a b. Finally consider any (r, s) R S then ra a and sb b such that (r, s)(a, b) = (ra, sb) a b again. And as R and S are commutative rings, so is R S such that this has been enough to guarantee that a b is an ideal of R S . (2) Next we will prove u = (u) (u). If (a, b) u then it is clear that a = (a, b) (u) and b = (a, b) (u) such that (a, b) (u) (u). Conversely if (a, b) (u) (u) then there are some r R and s S such that (a, s) u and (r, b) u. But as u is an ideal we also have (a, 0) = (1, 0)(a, s) u again and analogously (0, b) u. Thererby (a, b) = (a, 0) + (0, b) u proving the equality. 338 17 Proofs - Rings and Modules

(3) Now we are ready to prove ideal R S = {a b | a i R, b i S }. We have already seen in (1) that a b is an ideal of R S (supposed a and b are ideals). Conversely if u i R S is an ideal, then we let a := (u) and b := (u). Then a i R and b i S are ideals, since and are surjective homomorphisms. And u = a b by (2). (4) Now let a i R and b i S . Then we will prove a b = a b. First suppose that (a, b)k = (ak , bk ) a b for some k N. Then ak a k and b b such that a a and b b and thereby (a, b) a b. Conversely suppose ak a and bl b for some k , l N. Then we dene n := max {k, l} and thereby get (a, b)n = (an , bn ) a b. And this means that (a, b) is contained in the radical of a b again. (5) This enables us to prove srad R S = {a b | a = a, b = b}. First suppose that u i R S is a radical ideal and let a := (u) and b := (u) again. Then by (4) we may compute
ab = u =

u =

ab =

This proves that a = a and b = b are radical ideals of R and S respectively. And if conversely a and b are radical ideals then a b is a radical ideal of R S by virtue of
ab =

b =

ab

(6) Next we will prove spec R S = {p S } {R q}. If p i R is a prime ideal then we have already proved p S i R S in (1). And it is clear that p S = R S as (1, 0) p S . Thus suppose (a, b)(p, q ) = (ap, bq ) p S . Then ap p and as p is prime this means a p or p p. Without loss of generality we may take p p and thereby (p, q ) pS . Altogether we have seen that pS is a prime ideal of R S . And analogously one may prove that R q is prime, supposed q i S is prime. Thus conversely consider a prime ideal u i R S let p := (u) and q := (u). Clearly (1, 0)(0, 1) = (0, 0) u. And as u is prime this means (1, 0) u or (0, 1) u. We will assume (0, 1) u, as the other case can be dealt with in complete analogy. Then 1 q and hence q = S . That is u = p S and as u = R S this means p = R. Thus consider a and p R such that ap p. Then (a, 0)(p, 0) = (ap, 0) p S = u. And as u is prime this means (a, 0) u or (p, 0) u (which we assume w.l.o.g.). Thereby p p, altogether we have found that u = p S for some prime ideal p i R. (7) It remains to prove smax R S = {m S } {R n}. Thus consider a maximal ideal m i R and suppose m S u i R S . Then S = (m S ) b := (u) i S and hence b = S . Likewise we nd m = (m S ) a := (u) i R and as m is maximal this implies a = m or a = R. Thus we get u = a b = m S or u = a b = R S . And this means that m S is a maximal ideal of R S . Likewise one can see that R n is a maximal ideal for any n i S maximal. Conversely consider a maximal ideal u i R S . In particular u is prime and hence - by (6) - u = m S or u = R n for some prime ideal m i R or n i S respectively. We only regard the case u = m S 339

again, as the other is completely analogous. Thus consider some ideal m a i R, then u = m S a S i R S . By maximality of u that is u = m S = a S or a S = R S . Again this yields m = a or a = R and this means that m is maximal. P Proof of (1.79): (i) Let C denote the intersection C = { a R | i I : a Ci }, then it is clear that 0, 1 C , as any Ci is a positive cone and hence 0, 1 Ci . Likewise, if a, b C then for any i I we have a, b Ci and hence a + b Ci and a b Ci , as Ci was supposed to be a positive cone. Yet this again means a + b C and a b C , such that C is a positive cone again. (ii) It remains to show, that cone(A) is well-dened. However, as R itself is a positive cone and C R, hence R is an element of the dening set { C R | A C, C is a positive cone }. In particular this set is non-empty and hence the intersection over all its elements exists. (iii) Recall the denitions of the image (C ) = { (a) | a C } and the preimage 1 (D) = { a R | (a) D }. Now, as 0 = (0) and 1 = (1) we get both 0, 1 (C ) and 0, 1 1 (D). So given any u, v (C ) there are some a, b C such that u = (a) and v = (b). Thereby u + v = (a) + (b) = (a + b) (C ) and likewise u v = (a) (b) = (a b) (C ) again. Altogether this means that (C ) is a positive cone again. Similarly, if a, b 1 (D) we have (a), (b) D and therefore (a + b) = (a) + (b) D which implies a + b 1 (D). In complete analogy we nd (a b) = (a) (b) D and hence a b 1 (D). P Proof of (1.80): In (i) to (iv) there is nothing to prove - everything is either trivial or wellknown. But we wish to give a short prove for (v): First of all, we have 0 0 due to the reexivity (O1) of . Furthermore 0 1 is property (P1) of . Thereby 0, 1 R+ . If now a, b R+ , then 0 a and 0 b and hence 0 a + b and 0 a b due to properties (P3) and (P4). But this again means a + b and a b R+ such that R+ is a positive cone. Suppose a R+ (R+ ) that is 0 a and 0 a. The latter is 0 0 a and hence a 0 due to (P2). But as also 0 a the anti-symmetry (O2) implies a = 0. Claim number (vi) is fairly easy to see: rst of all 0 = 02 and 1 = 12 such that 0, 1 is satised. Property (C2) of cones is satised trivially and for 2 2 2 property (C3) we have to regard a = a2 1 + + am and b = b1 + + bn . Then it is clear that also ab , as
m n 2 a2 i bj = i=1 j =1 (m,n)

ab =

(ai bj )2
(i,j )=(1,1)

340

17 Proofs - Rings and Modules

We will now prove (vii) that R[x]+ is a proper cone: As 0 R+ and lc (0) = 0 we get 0 R[x]+ as well. Likewise 1 R[x]+ . If now f and g are two polynomials in R[x]+ then we have to show that f + g and f g are in R[x]+ again. If one of them - say f = 0 - then f + g = g and f g = 0 and hence there is nothing to prove. Otherwise let f (x) = am xm + + ak xk g (x) = bn xn + + bl xl By assumption f , g R[x]+ we have am and bn 0. Therefore, if m = n then lc (f + g ) = an + bn 0. And if m = n, say m < n then lc (f + g ) = bn 0. In any case we get lc (f + g ) 0 and hence f + g R[x]+ again. Also by denition of the multiplication in R[x] we have

(f g )(x) =
u=0 p+q =u

f [p] g [q ] xu

Thus if u > m + n then either p > m or q = u p > n, in any case we have f [p] g [q ] = 0. And for the same reason all combinations (p, q ) with p + q = u for u = m + n yield f [p] g [q ] = 0 except f [m] g [n] = am bn = 0, as R is an integral domain. Hence lc (f g ) = am bn and as both have been positive, so is lc (f g ) 0. And this again means f g R[x]+ such that R[x]+ is a positive cone as well. To prove that R[x]+ is proper consider some f R[x] such that f R[x]+ and f R[x]+ . Let a := lc (f ) then this means a 0 and a 0. But due to (v) this implies a = 0 (see above). And the only polynomial with lc (f ) = 0 is f = 0. In complete analogy to the arguments above one can show that R[x] is a proper, positive cone of R[x], as well. P Proof of (1.85): First of all for any a R we have a a = 0 C and hence (O1). Also if a b and b c this means b a and c b C and hence c a = (c b) + (b a) C such that (O2). In particular, if a b and b a that b a and (b a) = a b C , but as C is proper this means b a = 0 and hence (O3). As 1 C we have 0 1 which is (P1), (P2) is satised trivially: a b b a C (b a) 0 C 0 b a. And at last, for any a and b R with 0 a and 0 b we have a, b C such that a + b and ab C which again means a + b resp. ab 0. P Proof of (1.86): We have already proved that: given a proper positive cone C we can dene a positive order a b by letting b a C . And conversely, if is a positive order on R, then R+ = { a R | a 0 } is a proper positive cone in R. Hence both of these maps are well-dened. It remains to show that they also are mutually inverse. Let us rst start with a proper positive cone C , then we have to show that R+ = C . By denition a 0 means a 0 = a C , so this is true 341

already. Conversely, given a positive order we have to show that a b is equivalent to b a R+ . But by denition this is b a 0 such that this is guaranteed by property (P2) of positive orders. P Proof of (1.83): (i) If a 0 then |a| = a 0 is clear. Otherwise, if a 0 then |a| = a but as a 0 (P6) implies a 0 and hence |a| 0. (ii) If a 0 then |a| = a a is clear. Otherwise, if a 0 then |a| = a but as a 0 (P6) implies a 0 a and hence |a| a. (iii) If a 0 then |a| = a and hence |a|2 = a2 is clear. Otherwise, if a 0 then |a| = a such that |a|2 = (a)2 = a2 again. (iv) We have to consider three cases: (1) if a > 0 then |a| = a and sgn(a) = 1 such that |a| = a = 1 a = sgn(a)a. (2) if a = 0, then |a| = 0 and sgn(a) = 0 such that |a| = 0 = 0 0 = sgn(a)a. (3) lastly, if a < 0 then |a| = a and sgn(a) = 1 such that |a| = a = (1) a = sgn(a)a. Thus the equality holds true in any of these three cases. (ix) If a b then (as 0 a) we get a2 ab due to (P8). Likewise (as 0 b as well) we nd ab b2 . Thus a2 ab b2 which implies a2 b2 due to the transitivity of . Conversely we start with a2 b2 . Suppose we had a > b then in particular b a such that b2 a2 b2 . This implies a2 = b2 and hence 0 = b2 a2 = (a b)(a + b). As R is an integral domain this would mean a = b (in contradiction to a > b) or a + b = 0. But again 0 a = b 0 implies a = b = 0 in contradiction to a > b. Thus a > b is false, which only leaves a b. (viii) If a = 0 or b = 0 then ab = 0 and hence sgn(ab) = 0 = sgn(a)sgn(b). Now suppose a = 0 and b = 0 and distinguish 4 cases: (1) if a > 0 and b > 0 then ab > 0 due to (P4) and hence sgn(ab) = 1 = 1 1 = sgn(a)sgn(b). (2) if 0 < a and b < 0 then 0 > ab due to (P9) such that sgn(ab) = 1 = 1 (1) = sgn(a)sgn(b). (3) the case a < 0 and b > 0 is completely analogous to case (2), by ipping a and b. (4) if a < 0 and b < 0 then a > 0 and b > 0. Now by (1) we get sgn(ab) = sgn((a)(b)) = 1 = (1)(1) = sgn(a)sgn(b). Thus the equality holds true in any of these cases. (v) Once (iv) and (viii) have been established this is straightforward: |ab| = sgn(ab)ab = sgn(a)a sgn(b)b = |a| |b|. (vi) Due to (iii) we know |a + b|2 = (a + b)2 = a2 + 2ab + b2 . But according to (ii) ab |ab| such that |a + b|2 = a2 + 2ab + b2 a2 + 2|ab| + b2 . We now invoke (v) and (iii) again to nd |a + b|2 a2 + 2|ab| + b2 = |a|2 + 2|a||b| + |b|2 = (|a| + |b|)2 . As both |a| and |b| 0 we can nally use (ix) to deduce |a + b| |a| + |b| as claimed. (vii) First of all |a| = |b + (a b)| |b| + |a b| according to (vi). And from this we get |a| |b| |a b|. Now in complete analogy we have |b| = |a + (b a)| |a| + |b a| and hence (|a| |b|) = |b| |a| |b a| = |a b|. Combining these two cases we see ||a| |b|| |a b|. 342 17 Proofs - Rings and Modules

P Proof of (1.87): A word in advance: we have N = R+ nzdR. It is well known (and easy to see) that the set nzdR of non-zero divisors of R is multiplicatively closed. And R+ is multiplicatively closed due to (1) and (4). Hence N is multiplicatively closed, as an intersection of such sets. To begin with we have to prove the well-denedness of the relation. Thus we consider a/m = a /m and b/n = b /n . That is u(am a m) = 0 for some u N . In particular u is a non-zero divisor and hence am = a m. Likewise bn = b n. We now have to show that a /m b /n implies (and by universality hence already is equivalent to) a/m b/n. By denition a b m n a n b m 0 b m a n

Multiplying the latter inequality with mn 0 we get 0 mn(b m a n ) = b nmm a mnn = bn mm am nn . This again yields am nn bn mm , which once again iturns into a(nm n ) (bm n )m a bm n b = m nm n n

Thus we truly got a/m b/n and hence the well-denedness of . In a next step we have to prove that this relation also is a partial order. The reexivity a/m a/m is clear, as am am. The anti-symmetry is easy as well: consider a/m b/n such that a/m b/n as well. This means an bm and bm an such that an = bm. This again is a/m = b/n. It remains to verify the transitivity : consider a/m b/n and b/n c/u. That is an bm and bu cn. As u and m 0 we nd aun bum and bum cnm such that aun cnm. This again yields a/m (cn)/(un) = c/u. Now were ready for the actual topic of the claim: the positivity of the partial order . First note that by denition we have 0 b n 0b

Thereby property (1) is obvious, as 0/1 1/1 0 1. And for (2) we consider a/m b/n, which is an bm and hence 0 bm an. As we have just seen this is 0 (bm an)/(mn) = b/n a/m, such that the equivalency is established. For properties (3) and (4) we are given a/m and b/n 0. That is a, b 0, hence ab 0 and thus (a/m)(b/n) = (ab)/(mn) 0. Likewise (as m, n 0 as well) we have an + bm 0 such that a/m + b/n = (an + bm)/(mn) 0. P Proof of (1.92): (i) From 1 = e + f a + b i R it is clear that a + b = R. Now consider any x a b. That is there are a and b R such that x = ae = bf . Then we get xe = (bf )e = b(f e) = 0 and xf = (ae)f = a(ef ) = 0. And thereby x = x1 = x(e + f ) = xe + xf = 0 such that a b = { 0 }. 343

(ii) It is clear that : x (x + a, x + b) is a well-dened homomorphism of rings (as it bers into the canonical epimorphisms x x + a and x x + b). Now suppose (x + a, x + b) = (x) = 0 = (0 + a, 0 + b). This means x a and x b and hence x a b = { 0 }. Thus we have x = 0 which implies kn() = { 0 }, which means that is injective. It remains to prove, that also is surjective: Since a + b = R there are some e a and f b such that e + f = 1. Now suppose we are given any (y + a, z + b). Then let x := yf + ze and compute (x) = (yf + ze + a, yf + ze + b) = (yf + a, ze + b) = (yf + ye + a, ze + zf + b) = (y (f + e) + a, z (e + f ) + b) = (y + a, z + b) (iii) We will now prove the chinese remainder theorem in several steps. (1) Let us rst consider the following homomorphism of rings
n

: R
i=1

ai : x (x + ai )

We now compute the kernel of : (x) = 0 = (0 + ai ) holds true if and only if for any i 1 . . . n we have x + ai = 0 + ai . And of course this is equivalent, to x ai for any i 1 . . . n. And again this can be put as x a1 an . Hence we found kn() = a. Now we aim for the surjectivity of : (2) As a rst approach we prove the existence of a solution x R of the system (x) = (1 + a1 , 0 + a2 , . . . , 0 + an ). Since a1 + aj = R for any j 2 . . . n we also nd a1 + a2 . . . an = R due to (1.56.(iii)). That is there are elements e1 a1 and f1 a2 . . . an such that e1 + f1 = 1. Hence we nd f1 + aj = 0 + aj for any j 2 . . . n. And also f1 + a1 = (1 e1 ) + a1 = 1 + a1 . That is x := f1 does the trick. (3) Analogous to (2) we nd f2 , . . . , fn R such that for any j 1 . . . n we obtain the system (fj ) = (i,j + ai ) (4) Thus if we are now given any a1 , . . . , an R then we let us dene x := a1 f1 + + an fn and we thereby obtain
n n

x + ai =
j =1

aj fj + ai =
j =1

aj i,j + ai = aj + ai

Hence (x) = (ai + ai ) and as the ai have been arbitrary this means that is surjective. (5) Thus (1.77.(ii)) yields the desired isomorphy. P Proof of (3.44): (ii) If x, y kn() then (ax) = (a)(x) = (a)0 = 0 and hence ax kn() for any a R. Likewise (x + y ) = (x)+ (y ) = 0+0 = 0, 344 17 Proofs - Rings and Modules

such that x + y kn() as well. Hence kn() is an R-submodule of M . Now we have the assumption that is surjective and consider any u, v im, that is u = (x) and v = (y ) for some x, y M . Then u + v = (x) + (y ) = (x + y ) im. And if b S then - due to the surjectivity of - we may choose some a R with b = (a). Then bu = (a)(x) = (ax) im again, and hence im() is an S -submodule of N . (iii) is surjective i for any y N there is some x M such that (x) = y . In other words that is N im. But as im() N is true in any case, this is equivalent to im() = N . Likewise is injective, i for any x, y M we get the implication (x) = (y ) = x = y . But since is R-linear, this can also be put as (y x) = 0 = y x = 0. Let us now substitute p = y x, then this is equivalent to: (p) = 0 = p = 0 for any p M . In other words that is kn() { 0 }. But as as { 0 } kn() is true in any case, this is equivalent to kn() = { 0 }. (iv) It suces to check the general case, as the case of R-modules is included via = = 1 1, that is by leaving out and . Consider any x, y L and any a R, then it is easy to compute (ax + y ) = ((ax + y )) = ((a)(x) + (y )) = ((a)) ((x)) + ((y )) = (a)(x) + (x). (v) We have to prove that + and a are homomorphisms of R-modules again. To do this consider any x, y M and b R, then it is easy to compute ( + )(bx + y ) = (bx + y ) + (bx + y ) = b(x) + (y ) + b (x)+ (y ) = b((x)+ (x))+ (y )+ (y ) = b( + )(x)+( + )(y ). And as R is commutative, we also get (a)(bx + y ) = a(b(x)+ (y )) = ab(x) + a(y ) = ba(x) + a(y ) = b(a)(x) + (a)(y ). (vi) We have to prove that + and a are -module-homomorphisms again. To do this consider any x, y M , a S and b R, then it is easy to compute ( + )(bx + y ) = (bx + y ) + (bx + y ) = (b)(x) + (y ) + (b) (x) + (y ) = (b)((x) + (x)) + (y ) + (y ) = (b)( + )(x) + ( + )(y ). And as S is commutative, we also get (a)(bx + y ) = a((b)(x) + (y )) = a(b)(x) + a(y ) = (b)a(x) + a(y ) = (b)(a)(x) + (a)(y ). (vii) The properties of an R-module-homorphism are easy to see, if a R and (ax ), (bx ) RX then (a(bx )) = (abx ) = x (abx )x = X a x bx x = a (bx ), by construcion of R , refer to (3.38) for details. Likewise we compute ((ax ) + (bx )) = (ax + bx ) = x (ax + bx )x = x ax x + x bx x = (ax ) + (bx ). And im = lhR (X ) is clear from the denition of lhR (X ) and the construction of . Now is surjective, i lhR (X ) = im = N and in the light of (3.39.(ii)) this is precisely the deniton that X generates M . Also we have seen that is injective, i kn( ) = 0 and by construction of this means x X : ax = 0. This again is the R-linear x ax x = 0 = independence of X . If is bijective, then any x M has a unique representation (ax ) RX with x = (ax ) = x ax x. And by (3.39.(vi)) this is equivalent to X being an R-basis of M . 345

P Proof of (3.45): In (i) the one direction: = implies (x) = (x) for any x X is trivial. Conversely consider any y M . As X generates M there are some xi X and ai R such that y = a1 x1 + + an xn . Now by the R-linearity of and we nd
n n

(y ) =
i=1 n

ai xi

ai (xi )
i=1 n

=
i=1

ai (xi ) =
i=1

ai xi

= (y )

And as y M has been chosen arbitarily we have in fact proved = . So it remains to prove statement (ii). As B is a basis of M it is clear - due to(3.39.(vi)) - that the representation y = x ax x is unique and hence is a well-dened map. It also is easy to see that this is R-linear (see below). But from this we already get the uniqueness, due to (i). As to R-linearity: given u R and y = x ax x and z = x bx x M we clearly see that (uy ) =
x B

uax x uax n(x) = u ax n(x) = u(y )

x B

=
x B

(y + z ) =
x B

(ax + bx )x ax n(x) + bx n(x) = (y ) + (z )

xB

=
x B

P Proof of (3.46): We have to prove that the push-forward and pull-back truly are homorphisms again. As (i) is contained in (iii) and likewise (ii) is contained in (iv) (as = = 1 1 we just have to omit and ) we will prove the latter two claims only. For (iii) we pick up any a S and , mhom (L, M ). Then for any x L we get (a + )(x) = (a(x) + (x)) = (a)(x) + (x) = ( (a) + )(x). And as this is true for any x L we got (a + ) = (a) + = (a) () + ( ). For (iv) the reasoning is quite similar, pick up any b T and , mhom (L, N ). Then for any x L we get (b + )(x) = (b + )((x)) = b(x) + (x) = (b + )(x). And as this is true for any x L we get (b + ) = b + = b ( ) + ( ). P Proof of (3.50): (i) The linearity can be seen with a simple computation based on the compositions of R2 : consider a, b, x and y R all elements of R, then 346 17 Proofs - Rings and Modules

((a, b)+(x, y )) = (a + x, b + y ) = (a + x)+(b + y ) = (a + b)+(x + y ) = (a, b) + (x, y ). Likewise (a(x, y )) = (ax, ay ) = ax + ay = a(x + y ) = a(x, y ). However - for example in the case R = Z - we have (2 1, 1) = (2, 1) = 2 + 1 = 3, whereas 2(1, 1) = 2(1 + 1) = 2 2 = 4. That is in the case R = Z we see that is denitely not multilinear. (ii) In this case the multilinearity follows from the denition of the compositions on R2 : (a + b, x) = (a + b)x = ax + bx = (a, x) + (b, x) and (ax, y ) = (ax)y = a(xy ) = a(x, y ). Likewise we see that (a, x + y ) = (a, x) + (a, y ) and (x, ay ) = a(x, y ) due to the commutativity of R. Hence is R-multilinear but in the case R = Z again denitely not R-linear: (2 (1, 1)) = (2, 2) = 2 2 = 4, whereas 2(1, 1) = 2(1 1) = 2 1 = 2. P Proof of (3.51): (i) We will rst prove the well-denedness of : if x + P = y + P , then y x P 1 (Q) and hence (y ) (x) = (y x) Q. This again is (y ) + Q = (x) + Q, which had to be shown. To prove that is an R-module-homorphism can be done in a straightforward computation: let a R and x, y M , then (a(x + P ) + (y + P )) = ((ax + y ) + P ) = (ax + y ) + Q = (a(x) + (y )) + Q = a ((x) + Q) + ((y ) + Q) = a(x + P ) + (y + P ) (ii) First of all, let us abbreviate K := kn. Then is well-dened, since if x + K = y + K , then k := y x K and hence - as (k ) = 0 (y + K ) = (y ) = (x + k ) = (x) + (k ) = (x) = (x + K ) That is an R-module homomorphism can be shown in complete analogy to the argument used in (i): (a(x + K ) + (y + K )) = ((ax + y ) + K ) = (ax + y ) = a(x) + (y ) = a(x + K ) + (y + K ) By construction we have im = im, such that the surjectivity of is trivial. It suces to check the injectivity: so let us regard any x, y M such that (x) = (x + K ) = (y + K ) = (y ). That is (y x) = (y ) (x) = 0 and hence y x kn = K , such that x + K = y + K. (iii) Consider the canonical projection : M M/P given by (x) = x + P . Let us denote the restricion of to Q by , that is : Q M/P : x x + P . Then we get kn = { x Q | x P } = P Q im = { x + P | x Q } = P + Q

P 347

So by the 1st isomorphism theorem induces an isomorphism : im, given to be (x + kn) = x + P and this is precisely Q/kn the isomorphism given in the claim. (iv) Consider the identity 1 1 : M M on M . As P Q = 1 11 (Q) the identity induces the map : M/P M/Q : x + P x + Q, according to (i). For this homomorphism it is clear, that kn = { x + P Q | x Q } = Q P im = { x + Q | x M } = M Q So by the 1st isomorphism theorem induces an isomorphism : im, given to be ((x + P ) + kn) = x + Q, and this (M/P )/kn is precisely the inverse of the isomorphism given in the claim. P Proof of (3.47): It is clear that (M, +) is a commutative group, as we did not change the addition on M . It remains to prove the properties (M) of an R[t]-module. Thus consider any f , g R[t] and any x, y M . As is R-linear so are the compositions k and hence we can compute 1x = t0 x = 0 (x) = 1 1(x) = x

f (x + y ) =
k=0

f [k ]k (x + y ) =
k=0

f [k ] k (x) + k (y )

=
k=0

f [k ]k (x) +
k=0

f [k ]k (y ) = f x + f y

(f + g )x =
k=0

(f [k ] + g [k ])k (x)

=
k=0

f [k ] (x) +
k=0

g [k ]k (x) = f x + gx

f [i]i

j =0

f (gx) =
i=0

g [j ]j (x)

f [i]
i=0 j =0

g [j ]i j (x) f [i]g [j ]i+j (x)

=
i=0 j =0

i+j =k

f [i]g [j ] k (x) = (f g )x P

=
k=0

Proof of (3.21): 348 17 Proofs - Rings and Modules

 We will rst verify that the exterior direct product P := i Mi is an Rmodule again (under the operations given). The well-denedness of the operations is clear. Thus consider x = (xi ), y = (yi ) and z = (zi ) P . Then x + (y + z ) = (xi + (yi + zi )) = ((xi + yi ) + zi ) = (x + y ) + z which is the associativity. Likewise x + y = (xi + yi ) = (yi + xi ) = y + x, which is the commutativity. Letting 0 := (0i ), where 0i Mi is the zero-element, we nd x + 0 = (xi + 0i ) = (xi ) = x, such that 0 P is the zero-element. And the negative of x is x = (xi ) P , as x + x = (xi + (xi )) = (0i ) = 0. Now consider any a, b R then a(x + y ) = (a(xi + yi )) = ((axi ) + (ayi )) = (axi ) + (ayi ) = (ax) + (ay ). Likewise (a + b)x = ((a + b)xi ) = ((axi ) + (bxi )) = (axi ) + (bxi ) = (ax) + (bx) and (ab)x = ((ab)xi ) = (a(bxi )) = a(bxi ) = a(bx). Finally 1x = (1xi ) = (xi ) = x, such that altogether P is an R-module again. Let us now verify that the direct sum S := i Mi is an R-submodule of the direct product P . Thus we only have to prove that S is closed under the operations of P . However for any a R and any x, y S it is clear that supp(ax) supp(x) and supp(x + y ) supp(x) supp(y ). Hence #supp(ax) #supp(x) < and #supp(x + y ) #supp(x) + #supp(y ) < . Thereby ax and x + y S again. Now consider the R-module-homomorphisms i : Mi N . By construction of S the map := i i : S N is well-dened. And it also is an R-module-homomorphism, as for any a R and for any x = (xi ), y = (yi ) S we get (ax) = i i (axi ) = i ai (xi ) = a(x) and (x + y ) = i i (xi + yi ) = i (i (xi ) + i (yi )) = (x) + (y ). (a) = (b): by denition of the arbitrary sum of the submodules Pi of M and by the assumption M = i Pi we obtain the identity M =
i

xi

I nite, xi Pi

Thus if we are given any x M there are nitely many xi Pi (i ) such that x = i xi Letting xi := 0 for i I \ we obtain an extention (xi ) i Pi satisfying x = i xi . This proves the existence of such a representation. For the uniqueness suppose i xi = i yi for some (xi ), (yi ) i Pi . Fix any j I and let xj :=
i=j

xi and yj :=
i=j

yi P j

Then we nd xj yj = yj xj , but as xj yj Pj and yj xj Pj we have xj yj Pj Pj = 0, such that xj = yj . But as j I has been arbitrary this means (xi ) = (yi ), which also is the uniqueness. (b) = (a): consider any x M , by asumption (b) there is some (xi ) i Pi such that x = i xi . And as only nitely many xi are non-zero this means x = i xi i Pi . And as x has been arbitrary this already is M = i Pi . Now consider any j I and any x Pj Pj . As x Pj we may let xj := xj Pj and xi := 0 Pi (for i = j ) to nd x = i xi . On the other hand we have x Pj , that is there are some 349

xi Pi such that x = i=j xi . Finally let xj := 0 then altogether this is i xi = i=j xi = i xi and due to the uniqueness this means xi = xi for any i I . But by choice of xi this is xi = 0 for any i = j and hence x = i=j 0 = 0, which had to be shown. P Proof of (3.22): Let us again abbreviate the exterior direct product by P := i Mi and the exterior direct sum by S := i Mi . It is immediately clear that j : P Mj is an R-module-homomorphism: j (ax + y ) = j ((axi + yi )) = axj + yj = aj (x)+ j (y ) Analogously it is clear that j is an R-module homomorphism. Also j j (xj ) = j (i,j xj ) = j,j xj = xj such that j j = 1 1. And from this it immediately follows that j is surjective and that j is injective. Note that the latter could have also been seen directly. Now consider any x = (xi ) P , then by denition i (x) = xi such that x = (xi ) = (i (x)). Conversely consider x = (xi ) S , then by denition j (xj ) = (i,j xj ) (where the index runs over all i I ). Therefore a straightforard computation yields j ( x j ) =
j I j I

i,j xj = (xi ) = x
j I

i,j xj

P Proof of (3.26): (i) It is easy to see that the map given is a homomorphism: (xi ) + (yi ) = (xi + yi ) i (xi + yi ) = ( i xi ) + ( i yi ) and a(xi ) = (axi ) ( ax ) = a ( i i xi ). And the bijectivity is just a reformulation of i property (b) of inner direct sums. (ii) Let us abbreviate the exterior direct sum by M := i Mi . That is any x M is of the form x = (xi ) for some xi Mi of which only nitely many are non-zero. Thus x = (xi ) = i i (xi ) i Pi . And as x has been arbitrary this means M = i Pi . Now consider any j I and any x Pj Pj . As x Pj = j (Mj ) there is some mj Mj such that (xi ) := x = j (mj ) = (i,j mj ). On the other hand there are pi Pi (where i = j ) such that x = i=j pi . Likewise we choose ni Mi such that pi = i (ni ). If we nally let nj := 0, then x = i=j i (ni ) = i i (ni ) = (ni ). Comparing these two expressions of x we nd ni = xi = i,j mj = 0 for any i = j . And as also nj = 0 this means x = (ni ) = (0i ) = 0 M . Therefore Pj Pj = 0, as claimed. P Proof of (3.27): Clearly the map (j i ) is well-dened: as j and i are R-module homomorphisms, so is the composition j i : Mi Nj . Conversely (i,j ) ( i i,j ) is well-dened, too - it just is the Carthesian product of the induced homomorphisms i i,j : i Mi Nj . 350 17 Proofs - Rings and Modules

 And it is also clear, that (j i ) is a homomorphism of Rmodules: as i and j are homomorphisms of R-modules, so are the push-forward i and pull-back j . Combining these we see that j i is a homomorphism of R-modules. Thus if we regard any , mhom(M, N ) and any a R, then a + (j (a + )i ) = (aj i + j i ) = a(j i ) + (j i ), as the composition in the direct product was dened to be component-wise. We will now prove that these maps are mutually inverse, we start by regarding (j i ) := ( i j i ). Thus consider an arbitrary x = (xi ) i Mi then we nd (x) =
iI

j i (xi )

j
i I

i ( x i )

as j is an R-module homomorphism. Now recall the identities x = i i (xi ) and y = (j (y )), inserting these we obtain = from (x) = j (x) = (x)

Thus we have to regard (i,j ) ( i i,j ) (i,j ) := (j ( i i,j )i ). We need to verify that for any i I and j J we get i,j = i,j . Thus x any i, j and xi Mi and compute i,j (xi ) = j
aI

a,b
b J

i (xi )

By denition we have i (xi ) = (a,i xi ), where the index runs over a I . Inserting this into the denition of a a,b we continue i,j (xi ) = j
aI

a,b (a,i xi )
b J

=
aI

a,j (a,i xi )

If a = i then a,i xi = 0 Ma such that the respective summand vanishes. It only remains the summand of a = i, which is given to be i,j (i,i xi ) = i,j (xi ). As xi has been arbitrary this proves i,j = i,j which had to be shown. We now prove the claim in the remark to the proposition: that is we consider i := (i,j ) j mhom(Mi , Nj ) and want to show that the image of the corresponding map := ( i i,j ) is contained in j Nj . Thus consider any x = (xi ) i Mi then we have to verify, that y := (x) =
iI

i,j (xi )

j J

Nj

That is we have to verify that the support of y is nite. To see this let us abbreviate the support of x by := supp(x) I , which is a nite set, as x i Mi . Then it is clear that the support of y satises the 351

following inclusions supp(y ) = = = jJ

i I

i,j (xi ) = 0

{ j J | i I : i,j (xi ) = 0 } { j J | i I : xi = 0 and i,j = 0 } { j J | i with i,j = 0 } { supp(i ) | i }

However the latter is a nite ( is nite) union of nite (any supp(i ) was assumed to be nite) subsets of J . Hence it is a nite set itself which proves y j Nj which had been claimed. P Proof of (7.15): (a) = (b): consider any element f h and pick up the homogeneous decomposition f = d fd . By assumption (a) and as f h we also nd hd h Ad such that f can be decomposed into f = d hd . As fd and hd Ad property (3) of graded algebras implies fd = hd h. (b) = (c): For any h h let us consider the homogeneous decomposition h = d hd and denote the set H := hd | h h, hd = 0 . Then it is clear that H hom(A), by denition of H . And by assumption (b) we have H h. On the other hand any h h has been decomposed as h = d hd H i . And hence h H i h. (c) = (a): Consider any f h = H i , that is there is a nite subset H and elements gh A (where h ) such that f = h gh h. For any h let us abbreviate d(h) := deg(h) and pick up homogeneous decompositions f = d fd and gh = d gh,d . Then we compute fd = f =
dD h dD

gh,d

h =
dD h

gh,d h

As gh,d and h are homogeneous elements gh,d h is (either 0 or) a homogeneous element of degree deg(gh,d h) = deg(gh,d ) + deg(h) = d + d(h). Thus we can split the sum over (d, h) D into bers (c, h) such that c + d(h) = d. Thereby we nd fd =
dD dD c+d(h)=d

gh,c h

So on both sides of this equation we have homogeneous decompositions of f . Now by property (3) of graded algebrad this means that (for any d D) the homogeneous components have to coincide, too. That is fd =
c+d(h)=d

gh,c h h Ad

352

17 Proofs - Rings and Modules

Hence we have obtained that f can be decomposed as a sum of elements fd contained in h Ad . As f has been arbitrary that is h = d h Ad . And as h Ad Ad it is also clear that (h Ad ) ( c=d h Ac ) Ad ( c=d Ac ) = 0. Altogether we have obtained claim (a). P Proof of (7.16): (i) Let us write out the homogeneous decomposition 1 = d 1d . Then for any homogeneous element h H - say c := deg(h) - we get h = h1 =
dD

h 1d

Thereby h 1d Ac+d according to property (4). But as h H is homogeneous we may compare coecients (this is property (3)) to nd h = c+d=c h 1d . Yet D has been assumed to be integral, hence c + d = c = c + 0 implies d = 0. Thus for any h H we have found h = h 10 . Thus consider any f A and decompose f = d fd into homogeneous elements. Then we compute f 10 =
dD

fd 10 =
dD

fd = 0

as the fd H { 0 } are homogeneous. Thus we have proved f = f 10 for any element f A. In particular 1 = 1 10 = 10 A0 . (ii) Clearly the kernel is an ideal a := kn() i A of A, so we only have to prove the graded property. First of all, it is clear that the submodules a Ad intersect trivially (a Ac )
d=c

a Ad

Ac
d=c

Ad = { 0 }

Thus it remains to show that a truly is the sum of the a Ad . Consider any f a and decompose f into homogeneous components f = d fd where fd Ad . Then we get 0 = (f ) =
dD

(fd )

As was assumed to be graded, we have (fd ) Bd again. Thus comparing homogeneous coecients in B we nd (fd ) = 0 for any d D. And this again translated into fd a such that fd a Ad . As f has been arbitrary this also shows
a =
dD

a Ad

(iii) As a i A is an ideal and A is an R-algebra, a a A even is an R-algebra-ideal. And hence the quotient A/a is an R-algebra again. 353

And by construction we have 0 + a hom(A/a). Let us now prove the well-denedness of deg : hom(A/a) D: suppose we are given g and h A such that g + a = h + a hom(A/a). That is f := g h a. Let us now denote b := deg(g ) and c := deg(h) and decompose f = d fd (where fd a Ad by assumption). Then we nd g 0 if d = b if d = b = g f +h =
dD

dD

fc + h fd

if d = c if d = c

Now suppose we had b = c then comparing coecients we would nd g = fb a in contradiction to g + a hom(A/a). Thus we have deg(g ) = b = c = deg(h) and hence deg : hom(A/a) D is welldened. We will now prove the identity A
a

= Ad + a a

For the inclusion we are given some h + a (A/a)d . If thereby h + a = 0 + a then h a and hence h + a (Ad + a)/a. If on the other hand h + a = 0 + a then we have deg(h) = d by assumption and hence h Ad . Altogether h + a (Ad + a)/a again. Conversely consider any h Ad and any f a. Then we have to show (h + f )+ a (A/a)d . If h = 0 then (h + f ) + a = 0 + a (A/a)d is clear. And if h = 0 then deg(h) = d by assumption such that (h + f ) + a = h + a is of degree d, as well. Thus we have established the equality and in particular (A/a)d is an R-submodule of A/a. Now consider g Ac , h Ad and a, b a. Then (g + a)(h + b) = gh + (ah + gb + ab) Ac+d + a. And as all these elements have been arbitrary this also proves property (4) of graded algebras, that is (A/a)c (A/a)d (A/a)c+d . Now A
a =

Ad + a
dD

is easy to see: if we are given any f + a A/a then we may decompose f = d fd into homogeneous components. Thereby f +a = d (fd +a). This already yields and the converse inclusion is clear. Now x any c D then it remains to prove Ac + a a
d=c

Ad + a = { 0 + a } a

Thus consider any f + a contained in the intersection. In particular that is f + a (Ac + a)/a and hence there is some fc Ac such that f + a = fc + a. And likewise there are fd Ad (where d = c) such that f + a = d=c fd + a. Thus we nd a := fc
d=c

fd a =
dD

a Ad

as a has been assumed to be a graded ideal we can decompose a into a = d ad where ad a Ad . Comparing the homogeneous coecients 354 17 Proofs - Rings and Modules

of a in A we nd fc = ac a Ac . In particular f + a = fc + a = 0 + a, which had to be shown.

(iv) As U is multiplicatively closed U 1 A is a commutative R-algebra under a(f /u) = (af /u). And from the construction of hom(U 1 A) we also have 0 = 0/1 hom(U 1 A). As D was assumed to be a group deg : hom(U 1 A) D is a well-dened map. (If D only was a monoid we would have to regard deg : hom(U 1 A) grD given by deg(h/u) := (deg(h), deg(u)) instead). Next we prove U 1 A = h u u U, h Ad+deg(u)

The inclusion is easy: the case h/u = 0/1 is clear and if h/u = 0/1 we particularly have h = 0. Hence deg(h) = deg(u) + d by assumption on h and hence deg(h/u) = deg(h) deg(u) = d. The converse is analogous: the case h/u = 0/1 is clear again, as 0 Ad+deg(u) . And if h/u = 0/1 then d = deg(h/u) = deg(h) deg(u) and hence h Ad+deg(u) again. So we have to verify that (U 1 A)d is an R-submodule. Suppose a R and g/u, h/v (U 1 A)d . Then ag Ad+deg(u) and hence a(g/u) (U 1 A)d again. Furthermore g h gv + hu + = u v uv As deg(uv ) = deg(u) + deg(v ) we have gv and hu Ad+deg(uv) thus gv + hv Ad+deg(uv) again and this means (gv + hu)/uv (U 1 A)d . So let us now verify property (4) of graded algebras. To do this consider g/u (U 1 A)c and h/v (U 1 A)d . Then we have to verify (gh)/(uv ) = (g/u)(h/v ) (U 1 A)c+d . If (gh)/(uv ) = 0 we are done, else we have deg(g ) = c + deg(u) and deg(h) = d + deg(v ). Thus deg(gh) = deg(g )+deg(h) = c +deg(u)+ d +deg(v ) = (c + d)+deg(uv ). And this already is (gh/(uv ) (U 1 A)c+d . It remains to verify property (3) of graded algebras. Given any f /u U 1 A let us pick up a homogeneous decomposition f = d fd of f in A. Then f /u = d fd /u and as fd /u (U 1 A)ddeg(u) and as f has been arbitrary this means U 1 A = d (U 1 A)d . Thus it only remains to prove U 1 A
d

U 1 A c = 0
c= d

Thus consider some h/v contained in the intersection. We have to prove h/v = 0/1 thus let us suppose h/v = 0/1 and deduce a contradiction. As h/v (U 1 A)d that is h Ad+deg(v) . And likewise there is some nite subset D \ { c } and uc U , gc Ac+deg(uc ) (where c ) such that h/v = c gc /uc . Now let u := c uc U and uc := u/uc U then we compute h = v c gc 1 = uc u gc uc
c

355

That is there is some w U such that uwh = vw c gc uc . Note that uwh and the vwgc uc are homogeneous elements of A, where deg(vwgc uc ) = deg(v ) + deg(w) + c + deg(uc ) +
b=c

deg(ub )

= deg(v ) + deg(w) + c +
c

deg(uc )

= deg(uvw) + c Hence all the vwgc uc have dierent degrees and hence form a homogeneous decomposition of uwh. This however is homogeneous too, of degree deg(uwh) = deg(uw) + d + deg(v ) = deg(uvw) + d. As c = d this is yet another homogeneous decomposition. Hence uwh = 0 which would mean h/v = 0/1, a contradiction. (v) Consider any f a that is f k a for some k N. If k = 0 then 1 a and hence a = A. This would imply a = A, too, which clearly is a graded ideal again. Thus let us consider the case k 1. Pick up the hoomogeneous decomposition f = d fd of f , then by (7.15) it suces to check fd a for any d N. Now compute fk =
Nk

f1 . . . fk =
dN ||=d

f1 . . . fk a

where || := 1 + + k N. By denition of graded algebras it is clear that f1 . . . fk A|| and thus we have found the homogeneous decomposition of f k . As a is a graded ideal, this implies d N :
| |= d

f1 . . . fk a

We will now prove that fd a by induction on d. For d = 0 we k = have { || = 0 } = { 0 } and hence f0 ||=0 f1 . . . fk a. Thus suppose d 1, then we dene the following function : { || = kd } N : min{ i | i 1 . . . k } It is clear that () d (as else || k (d +1) > kd) and that () = d is equivalent to = (d, . . . , d) (as else || > kd). Hence we get
k fd = ()=d

f1 . . . fk f1 . . . fk
||=kd ()<d

f1 . . . fk

But we have already seen from the gradedness of a that the rst term (the sum over || = kd) is contained in a again. And if () < d then the induction hypothesis yields f() a. Yet it is clear that f() | f1 . . . fk and hence f1 . . . fk a again. Thus the second term (the sum over () < d) is contained in a, too. And this implies k fd a and hence even fd a, which had to be shown. 356 17 Proofs - Rings and Modules

P Proof of (7.19): (i) If h Ad but h = 0 then the homogeneous decomposition h = c hc has precisely one non-zero entry h = hd at c = d (by comparing coecients). Thus deg(h) = max{ d } and this clearly is d. (ii) If f hom(A) with f Ad then we have just argued in (i) that deg(f ) = max{ d } = d. And likewise ord(f ) = min{ d } = d which yields ord(f ) = deg(f ). Conversely suppose d := ord(f ) = deg(f ), then we use a homogeneous decomposition f = c fc again. By construction, if c < d = ord(f ) or d = deg(d) < c then fc = 0. Thus it only remains f = fd Ad and hence f hom(A). (iii) As ord(f ) is the minimum and deg(f ) is the maximum over the common set { d D | fd = 0 } it is clear, that ord(f ) deg(f ). Now let a := deg(f ) and b := deg(g ). Then (by construction of the degree) the homogeneous decompositions of f and g are of the form f=
ca

fc and g =
db

gd

Thus the product f g is given by a sum over two indices c a and d b. Assorting the (c, d) according to the sum c + d we nd that fg =
ca

fc
db

gd =
ca db

fc gd =
eD c+d=e

fc gd

Note that thereby fc gd Ac+d = Ae . Thus we have found the homogeneous decomposition of f g again. But as D was assumed to be positive c a and d b implies e = c + d a + d a + b. Thus the homogeneous decomposition is of the following form fg =
ea+b c+d=e

fc gd

And this of course means deg(f g ) a + b = deg(f )+deg(g ). The claim for the order can be shown in complete analogy. Just let a := ord(f ) and b := ord(g ). Then f and g can be written as f=
ca

fc and g =
db

gd

Thus in complete analogy to the above one nds that the homogeneous decomposition of f g is of the following form (which already yields ord(f g ) a + b = ord(f ) + ord(g )) fg =
ea+b c+d=e

fc gd

357

(iv) We commence with the proof given in (iii), that is a = deg(f ) and b = deg(g ). In particular fa = 0 and gb = 0. But as A is now assumed to be an integral domain this yields fa gb = 0. We have already found the homogeneous decomposition of f g to be fg =
ea+b c+d=e

fc gd

So let us take a look at the homogeneous component (f g )a+b of degree a + b D. Given any c a and d b we get the implication c+d=a+b = (c, d) = (a, b)

Because if we had c = a then c < a and hence c + d < a + d a + b, as D was assumed to be strictly positive. Likewise d = b would imply c + d < a + b in contradiction to c + d = a + b. Thus the homogeneous component (f g )a+b is given to be (f g )a+b =
c+d=a+b

fc gd = fa gb = 0

In particular we nd deg(f g ) a + b = deg(f ) + deg(g ) and hence deg(f g ) = deg(f ) + deg(g ). By replacing the above implication with c a, d b and c + d = a + b implies (c, d) = (a, b) the claim for the order ord(f g ) = ord(f ) + ord(g ) can be proved in complete analogy.

(v) As D is strictly positive it already is integral, according to (7.5.(iii)). Hence by (7.16.(i)) we already know 1 A0 and hence deg(1) = 0 according to (i). And as 1 is homogeneous we also get ord(1) = deg(1) = 0, according to (ii). Now suppose f A is an invertible element of A. Then we use (iv) to compute 0 = deg(1) = deg(f f 1 ) = deg(f ) + deg(f 1 ) 0 = ord(1) = ord(f f 1 ) = ord(f ) + ord(f 1 ) And as ord(f ) deg(f ) and ord(f 1 ) deg(f 1 ) (due to (iii)) the positivity of D can be used to nd the following estimates 0 = ord(f ) + ord(f 1 ) deg(f ) + ord(f 1 ) deg(f ) + deg(f 1 ) = 0 In particular we nd ord(f ) + ord(f 1 ) = deg(f ) + ord(f 1 ). Adding ord(f ) this equation turns into ord(f ) = deg(f ) and by (ii) again this means that f hom(A) is homogeneous.

(vi) Let us suppose h = f g for some g A. As h hom(A) we have h = 0 and this also implies f , g = 0. We will assume that f was not homogeneous and derive a contradiction. Recall that by (iii) f not being homogeneous is equivalent to ord(f ) < deg(f ). But if that was 358 17 Proofs - Rings and Modules

true we could use (iv) to nd the contradiction ord(h) = ord(f g ) = ord(f ) + ord(g ) < deg(f ) + ord(g ) deg(f ) + deg(g ) = deg(f g ) = deg(h) = ord(h) P Proof of (4.14): (i) Clearly, if b im(A), then Ax = b has no solution at all, hence L(Ax = b) = . Conversely if b im(A) then we can pick up some u Rn with Au = b. Now if v kn(A) then A(u + v ) = Au + Av = b +0 = b. And as v has been arbitrary, this means u + kn(A) L(Ax = b). Conversely consider some x with Ax = b, then A(x u) = Ax Au = b b = 0 and hence v := x u kn(A). Again, as x has been arbitrary that is L(Ax = b) u + kn(A). (ii) First of all let us compute DT 1 = SAT T 1 = SA. Further note that - due to the invertibility of S - we have the equivalency b = Ax Sb = SAx = DT 1 x

Now for (a) = (b): let us dene z := T 1 x, then Dz = DT 1 x = SAx = Sb. And x = T T 1 x = T z is clear, as well. Conversely for (b) = (a): if x = T z then z = T 1 T z = T 1 x and hence also Sb = Dz = DT 1 x and hence b = Ax by the equivalency above. (iii) By denition x L(Ax = b) means that Ax = b. As we have seen in (ii) this is equivalent to x = T z and Dz = Sb for some z Rn . Again this translates into x = T z and z L(Dx = Sb). And this is nothing but x T L(Dx = Sb). (iv) Due to (i) we know that L(Dx = Sb) either is empty (if Sb im(D)) or given to be u + kn(D) for some special solution Du = Sb. Now by (iii) we have L(Ax = b) = RL(Dx = Sb) = R(u + kn(D)). And since a matrix always denes an R-linear operation, this yields L(Ax = b) = Ru + R kn(D). P Proof of (4.19): We will prove the following claim by induction on the number m of rows of the matrix A: using the Gaussian algorithm one obtains an (m m)-matrix S and an (m n)-matrix E such that E is in echelon-form. The fact, that S is a product of elementary matrices is clear from the algorithm - no other matrices are involved. In the case A = 0 the algorithm already terminates at step (2), so we are left with S = 1 1m and E = A = 0. In this case E = 0 trivially is in echelon-form and E = SA is clear. So let us now assume A = 0. We rst anchor the induction for m = 1. As A = 0 we necessarily have s = 1 = r and the pivot step is skipped. 359

Likewise the main step (5) is void, as r + 1 . . . m is empty. So again E = A is in echelon-form and E = SA. And as m = r the algorithm terminates. Let us now assume that the claim has been proved for any matrix B with less than m rows. Then we need to show, that the claim holds true for A as well. The algorithm starts with row r = 1. And as A = 0 the algorithm does not terminate in (2) but targets u and s such that u = (A) and (rows A) = u. In case s = r = 1 the rows 1 and s are interchanged E = Pr,s A = SA. So after step (4) we can be sure, that (row1 E ) = (E ) is minimal. That is the rst u 1 columns of E are zero. Let us denote p = m 1 and q := n u then E is of the form 0 0 E = . . . 0 0 x0 x1 0 y1 h1,1 . . . . . . . . . 0 yp hp,1 xq h1,q . . . hp,q

That is x = (x0 , x1 , . . . , xq ) where xj = e1,u+j and y = (y1 , . . . , yp ) where yi = ei+1,u and H = (hi,j ) where hi,j = ei+1,u+j . Now the main step (5) is initiated. For any i 2 . . . m the following row-operations are performed rowi E := er,u (rowi E ) ei,u (rowr E ) Specically for the u-th column of E (that is the u-th entry of rowi E ) this is [rowi E ]u = er,u ei,u ei,u er,u and as R is commutative this means [rowi E ]u = 0. Hence after the main step the matrix E is turned into 0 0 E = . . . 0 0 x0 x1 0 0 b1,1 . . . . . . . . . 0 0 bp,1 xq b1,q . . . bp,q

For some matrix B = (bi,j ) where i 1 . . . p = m 1 and j 1 . . . q . Now the Gaussian algorithm iterates with r = 2. As B has m 1 rows only, the induction hypothesis yields that the Gaussian algorithm will also transform B into echelon form. So after the algorithm terminates B is in echelon form i.e. there is some r 1 . . . m 1 such that (row1 B ) < < (rowr B ) and for all i > r we get rowi B = 0. As E is in the form presented in the equation above, we have (row1 E ) = u < u + (row1 B ) = (row2 E ) and hence (row1 E ) < . . . < (rowr+1 E ) And for i > r + 1 we have (rowi1 B ) = 0 and therefore (rowi E ) = 0 due to the shape of E . That is E is in echelon form, as claimed. The fact that E = SA is clear, as all the operations (multiplications with elementary matrices from the left) on A are performed on S simultaneously. P Proof of (2.4): (i) Let us abbreviate u := A, as A is non-empty there is some a A. And as a i R is an ideal we have 0 a A and hence 0 u. If now 360 17 Proofs - Rings and Modules

a, b u then - by construction - there are a, b A such that a a and b b. As A is a chain we may assume a b and hence a b, too. Therefore we get a + b b and a, b b, such that a + b, a and b u again. And if nally r R then ra a, such that ra u. (ii) Let Z := b i R | a b = R , then Z = since a Z . Now let A Z be a chain in Z , then by (i) we know that u := A i R is an ideal of R. Suppose u = R, then 1 u and hence there would be some b A with 1 b. But as b i R is an ideal this would mean b = R in contradiction to b A Z . Hence we nd u Z again. But by construction we have b u for any b A, that is u is an upper bound of A. Hence by the lemma of Zorn there is some maximal element m Z . It only remains to show that m is maximal: as m Z it is an non-full ideal m = R. And if R = b i R is another non-full ideal with m b then a m b implies b Z . But now it follows that b = m since m Z , b Z and m b. (iii) Let a R be a unit of R and m i R be a maximal ideal. Then a m, as else 1 = a1 a m and hence m = R. From this we get R R \ smax R. For the converse inclusion we go to complements and prove R \ R smax R instead. Thus let a R be a non-unit a R , then 1 aR and hence aR = R. Thus by (i) there is a maximal ideal m i R containing a aR m smax R. Thus we also have a smax R which had to be proved. (iv) If R = 0 is the zero-ring, then R = { 0 } is the one and only ideal of R. But as this ideal is full, it cannot be maximal and hence smax R = . If R = 0 then a := { 0 } is a non-full ideal of R. Hence by (ii) there is a maximal ideal m smax R containing it. In particular smax R = . P Proof of (3.12): Of course we will prove the claim using a Zornication: Let us dene the collection of all proper submodules of N , that is Z := { Z m N | P Z, Z = N } As P Z this collection is non-empty. Also it is clear that Z is partially ordered under the inclusion relation . If now (Zi ) Z is a chain in Z (that is a totally ordered subset), we take Z := i Zi . Then Z m N is a submodule again: if x, y Z then there are some i, j I such that x Zi and y Zj . As these sets are totally ordered we may assume Zi Zj without loss of generality. And hence x + y Zj Z again. Also if a R, then ax Zi Z . Now suppose Z = N , in particular for any k 1 . . . n we have xj N = Z . Therefore for any k 1 . . . n there is some i(k ) I such that xk Zi(k) . As the Zi are totally ordered we may single out the maximal element among Zi(1) , . . . , Zi(n) . That is there is some j I such that Zi(k) Zj for any k 1 . . . n. In particular xk Zi(k) Zj and as Zj is an Rmodule we nd lh { x1 , . . . , xn } Zj . Now, as also P Zj we get 361

N = P + lh { x1 , . . . , xn } Zj N , that is Zj = N . But this is in contradiction to Zj Z , such that we necessarily have Z = N . Let us subsume: we have found a submodule Z m N with Z = N and clearly also P Z . That is Z Z again. But by construction Z is an upper bound (for any i I we have Zi Z ) of the chain (Zi ). Hence we may apply Zorns lemma to nd a maximal element M Z . So by construction of M Z we have properties (1) and (2) of the claim. And if M Q N for some submodule Q, then either Q = N or Q Z and hence M = Q, due to the maximality of M , and this is (3). P Proof of (3.35): (i) Because of (D1) (that is x T = T x) we have the obvious inclusions S T T . That is we have T S . If now x M with S x, then by (D2) we also get T x. And this is just a reformulation of the claim S T . (ii) We will rst prove S = { P sub(M ) | S P }. For the inclusion we are given any x S . Then we need to show that for any P sub(M ) with S P we get x P . As P sub(M ) there is some T M with P = T . Then S P translates into S T , which is T S . Together with S x now property (D2) implies T x. And this is x T = P , which had to be shown. Conversely for we are given some x M such that for any P sub(M ) we get S P = x P . Then we may simply choose P := S , then S P is satised due to (D1). Thus by assumption on x we get x P = S , which had to be shown. Thus let us now prove P = P , where we abbreviated P := S . The inclusion is clear by (D1) again. And by what we have shown, we have P = { Q sub(M ) | P Q }. But as P sub(M ) with P P this trivially yields P P , the converse inclusion. (iii) (a) = (c): by assumption there is some s S such that S \ { s } s. Due to (D3) there is some nite subset Ss S \ { s } such that Ss s. Now let S0 := Ss { s }, then S0 is nite again, as #S0 = #Ss +1 < and S0 \{ s } = Ss . In particular S0 \{ s } s, that is S0 is depenedent. (c) = (b): consider S0 S T , by assumption there is some s S0 such that S0 \ { s } s. However S0 \ { s } T \ { s }, such that by (i) we also get T \ { s } s. That is T is dependent. (b) = (a) nally is trivial, by letting T := S . (iv) (a) = (b): suppose S was dependent, that is there was some s S , such that S \ { s } s. As S T we also have S \ { s } T \ { s }, hence we nd S \ { s } s due to (i). But as also s S T this is a contradiction to T being independent. (b) = (c) is trivial (a nite subset is a subset). (c) = (a): suppose there was some x T such that T \ { x } x. Then by (D3) there would be a nite subset Tx T \ { x } such that Tx x. Now let T0 := Tx { x }, then T0 is nite again, as #T0 = #Tx + 1 < . Also T0 \ { x } = Tx x, that is T0 is dependent in contradiction to (c). 362 17 Proofs - Rings and Modules

(v) (a) = (b): clearly S T := S { x }, such that S is independent by assumption (a) and (iv). Now suppose we had S x, then clearly T \ { x } = S x. That is T is dependent in contradiction to the assumption. (b) = (a): if we had x S , then by property (D1) also S x, in contradiction to the assumption. Now suppose that S { x } was dependent. That is there would be some s S { x } such that S := S {x} \ {s} s

We now distinguish two cases: (1) if s = x then S = S and thereby we would get S = S s = x, a contradiction to S x. (2) if s = x then s S { x } implies s S . Now let R := S \ { s }, that is S = R { s } and S = (S { x }) \ { s } = R { x }. This again means S \ {x} = R = S \ {s} As S is independent we now have S \ { x } = S \ { s } of x S , S s and S \ { x } s property (D4) implies S = S \ {s} {s} = S \ {x} {s} s. Because

This is a contradiction to S x again, thus in both cases we have derived a contradiction, that is S { x } is independent, as claimed. (vi) Let us abbreviate the chains union by T := i Si and suppose T would be dependent. Then by (iii) there would be a nite subset T0 T , such that T0 is dependent, already. Say T0 = { x1 , . . . , nn }, where for any k 1 . . . n we have xk Si(k) for some i(k ) I . As { Si | i I } is totally ordered under , so is the nite subset Si(k) | k 1 . . . n . Without loss of generality we may assume, that Si(n) is the maximal element among the Si(k) . Hence we nd xk Si(k) Si(n) for any k 1 . . . n and this again means T0 Si(n) . Yet as T0 was dependent this implies, that Si(n) is dependent (by (iii) again), a contradiction. Hence T is independent, as claimed. (vii) Not surprisingly we will prove the claim using a zornication based on Z := { P M | S P T, P is independent } Of course Z is non-empty, as S Z , and it is partially ordered under the inclusion relation . Due to (vi) any ascending chain (Si ) admits an upper bound, namely T := i Si Z . Thus by the lemma of Zorn Z contains a maximal element B . Because of B Z we have S B T and B is independent. Now suppose there was some x T with B x. However, as B is independent and B x, we nd by (v) that even B { x } is independent. But as x B we have B B { x } T . Thus we have found a contradiction to the maximality of B . This contradiction can only be solved, if T B . And using (i) and (ii) we thereby get M = T B = B M . Altogether B is independent and M = B - it is a basis. (viii) First suppose x = b, then B = B and hence there is nothing to 363

show. Thus assume x = b. The proof now consists of two parts: (1) as B \ { b } B B and B b by assumption, we have B = (B \ { b }) { b } B . And thereby M = B B = B M , according to (ii). That is M = B . (2) as B \{ b } B is a subset and B is independent, so is B \{ b }, due to (iv). Now suppose B \ { b } x, then we would get B = (B \ { b }) { x } B \ { b } . And hence b M = B B \ { b } = B \ { b } , according to (1) and (ii) once more. But this is B \ { b } b in contradiction to the independence of B . Thus we have B \ { b } x, as B \ { b } also is independent (v) now implies, that B = (B \ { b }) { x } is independent. Combining (1) and (2) we found that B is a basis. (ix) (a) = (b): M = B is clear by denition of a basis. Now consider some subset S B , that is there is some x B , with x S . As B is independent and x B we get B \ { x } x, in other words x B \{ x } . And as S B \{ x } (i) implies x S . In particular S = M , as claimed. (b) = (a): as M = B by assumption it only remains to verify the independence of B . Thus assume, there was some x B such that S := B \ { x } x. As S S and x S , we have B = S { x } S . And thereby M = B S = S M according to (ii). This is M = S even though S B , a contradiction. (ix) (a) = (c): suppose B is a basis, that by denition B already is independent. Thus suppose B S , that is we may choose some s S \ B . As s M = B we get B s. And as B S \ { s } (i) also yields S \ { s } s. That is S is dependent, as claimed. (c) = (a): as B is assumed to be independent, it remains to show M = B . Thus suppose we had B M , that is there was some s M such that B s. Then by (v) we nd that S := B { s } is independent, as well - a contradiction. (x) Let us denote m := |A| and n := |B |, then we need to show m = n. In the proof we will have to distinguish two cases: (I) m or n is nite and (II) both m and n are innite. We will deal with (I) in the following steps (1) to (4), whereas (II) will be treated in (5) (1) Without loss of generality we may assume m n (interchanging A and B if not). Then M is nite by assumption (I), let us denote the elements of A by A = { x1 , . . . , xm }. Thereby let us assume that x1 , . . . , xk B and xk+1 , . . . , xm B for some k N [note that for k = 0 this is A B and for k = m this is A B = ]. In the case k = 0 we may skip steps (2), (3) and let A(k) := A, immediately proceeding with (4). (2) As A is a basis of M , it is a minimal generating subset, by (ix). That is { x2 , . . . , xm } = M = B . Hence there is some point b B such that b { x2 , . . . , xm } [as else B { x2 , . . . , xm } and hence M = B { x2 , . . . , xm } = { x2 , . . . , x m } , according to (ii), a contradiction]. For this b B let us dene A 364 := A { b } \ { x1 } = { b, x2 , . . . , xm } 17 Proofs - Rings and Modules

Then A is independent [as A is independent, the same is true for { x2 , . . . , xm }. And due to { x2 , . . . , xm } b (v) implies, that A is independent]. Next we nd that A x1 [because if we had A x1 then - using (v) and the fact that A is independent A { x1 } would be independent, too. However this is untrue, as (A { x1 }) \ { b } = A b]. Therefore A even is a basis [as A x1 we have x1 A and clearly A \ { x1 } A , such that A = (A \ { x1 }) { x1 } A . Combined with (ii) this yields M = A A = A M , that is A = M and the independence of A has already been shown]. (3) In (2) we have replaced x1 A by some b B to obtain another basis A := (A { b }) \ { x1 } = { b, x2 , . . . , xk }. That is: venturing from A to A we have increased the number of elements in A B by 1 (with respect to A B ). Iterating this process k times we establish a basis A(k) = { b1 , . . . , bk , xk+1 , . . . , xm } B . (4) From the construction it is clear, that |A | m n = |B |. On the other hand - as B is a basis, A B and M = A - item (ix) now implies that A = B . And thereby n = |B | = |A | m, as well. Together this is m = n as claimed. (5) It remains to regard case (II), that is m and N are both innite. As B M = A we obviously have A b for any b B . Thus by property (D3) there is a subset Ab A such that #Ab < is nite and Ab b already. Now let A :=
bB

Ab

Then for any b B we have Ab A and hence A b, due to (i). That is b A for any b B and hence B A . Thus by (i) and (ii) again we get M = B A = A M , such that M = A . Now, as A is a basis, if we had A A, then (ix) would yield M = A , a contradiction. This is A = A and hence we may compute |A| =
b B

Ab

bB

|Ab | |B N| = |B |

Hereby |B N| = |B | holds true, as B was assumed to be innite. Thus we have found m n and by interchanging the roles of A and B we also nd n m, which had to be shown. (xi) As S is independent, S M and M = M we get from (vii) that there is some basis A of M , with S A M . Thus by (x) we already obtain the claim from |S | |A| = |B |. (xii) (b) = (a) has already been shown in (x), thus it only remains to verify (a) = (b): thus consider any independent subset S M with |S | = |B |. As before [S is independent, S M and M = M now use (vii)] there is some basis A of M , with S A M . Thus by (x) we nd |B | = |S | |A| = |B |. That is |A| = |S | = |B | < which was assumed to be nite. In particular S is a subset of the 365

nite set A, with the same number of elements #S = #A. Clearly this means S = A and in particular S is a basis of M . P Proof of (3.38): (i) Trivially is R-linearly independent and also generates { 0 }, as { 0 } is the one and only submodule of { 0 }. Together is an R-basis of { 0 } and in particular rank { 0 } = 0. Conversely if rankM = 0, then by denition M = m = { 0 }. (ii) If M = P + X
m

then as P Q we also get M = Q + X

m.

(iii) First suppose a = aR is principal. If a = 0, then a = 0, which is a free R-module, with basis . And if a = 0, then { a } is an R-basis of a. It generates a, as { a } m = aR = a and it is R-linearly independent, as ba = 0 = b = 0 (as a = 0 and R is an integral domain). Conversely suppose a is free, with basis B a. If B = , then a = m = 0, which is principal a = 0 = R0. And if we had #B 2, then we could choose some b1 , b2 B . But this would yield b2 b1 + (b1 )b2 = 0 (as R is commutative) and hence B would not be R-linearly independent. Thus it only remains #B = 1, say B = { a } and hence a = B m = aR is principal. (iv) We have to prove that E is an R-linearly independent set, that also R-generates RI . Thus suppose we are given any x = (xi ) RI . Let us denote := { i I | xi = 0 }, then is nite, by denition of RI . And hence we get x = i xi ei E m . Therefore E generates RI as an R-module. For the R-linear independence we have i ai ei = 0 for some nite subset I and some ai R. We now compare the coecients of 0 and i ai ei . Hereby for any i the i-th coecient of 0 RI is 0 R. And the i-th coecient of i ai ei is ai . That is ai = 0 for any i and this means that E is R-linearly independent. P Proof of (3.39): (i) (D1): consider an arbitrary subset X M and x X . Then we nd X x simply by choosing n := 1, x1 := x and a1 := 1, as in this case a1 x1 = 1x = x. (D2): consider any two subsets X , Y M such that Y X and X x. Thus by denition there are some n N, xi X and ai R such that x = a1 x1 + + an xn . But as xi X and Y X there also are some m(i) N, yi,j Y and bi,j R such that xi = bi,1 yi,1 + + bi,m(i) yi,m(i) . Combining these equations, we nd
n n m(j )

x =
i=1

ai xi =
i=1 j =1

ai bi,j yi,j

In particular this identity yields Y x, which had to be shown. (D3): consider an arbitrary subset X M and x M such that X x. 366 17 Proofs - Rings and Modules

Again this means x = a1 x1 + + an xn for sucient xi X and ai R. Now let X0 := { x1 , . . . , xn } X . Then X0 clearly is a nite set and X0 x, by construction. As (D1), (D2) and (D3) are satised, we already obtain (iii), (iv) and (v) from (3.35). (ii) The identity X = { x M | X x } = lhR (X ) is obvious from the denitions of the relation and the linear hull lhR (X ). And the identity lhR (X ) = X m has already been shown in (3.11). (vi) (b) = (a): we rst prove that B is a generating system of M : Thus consider any x M , by assumption there is some (xb ) RB such that x = b xb b. Now let := { b B | xb = 0 }. Then B is nite and we further get x =
b B

xb b =
b

xb b lhR (B )

As x has been arbitrary that is M lhR (B ) = B m . Let us now prove that B also is R-linear independent : thus consider some nite subset B and some ab R (where b ) such that b ab b = 0. Then for any b B \ we let ab := 0 and thereby obtain ab b =
bB b

ab b = 0 =
bB

0b

Due to the uniqueness of this representation (of 0 M ) we nd ab = 0 for any b B . And in particular ab = 0 for any b . By denition this means that B truly is R-linearly independent. (vi) (a) = (b): Existence : consider any x M , as M = B m = lhR (B ) there are a nite subset B and ab R (where b ) such that x = b ab b. Now let xb := ab if b and xb := 0 if b B \ . Then (xb ) RB and we also get x =
b

ab b =
b B

xb b

Uniqueness : consider any two (xb ), (yb ) RB representing the same element b xb b = b yb b. Now let := { b B | xb = 0, or yb = 0 }. Then B is nite and we clearly get 0 =
bB

(xb yb )b =
b

(xb yb )b

Yet as B is R-linearly independent, this means xb yb = 0 for any b and hence xb = yb for any b B (if b , then xb = 0 = yb ). But this already is the uniqueness of the representation. P Proof of (3.40): (i) By (3.39) already satises (D1), (D2) and (D3). Thus it only remains to prove (D4) and we will already do this in the stronger form given 367

in the claim. That is we consider x X and X y . By denition of this is y = a1 x1 + + an xn for some ai S and xi X . Of course we may drop any summand with ai = 0 and hence assume ai = 0 for any i 1 . . . n. As y = 0 we still have n 1. Now let X := (X \{ x }) { y } (that is we have to show X x) and distinguish two cases: (1) if x { x1 , . . . , xn } then { x1 , . . . , xn } X \{ x } X and in particular X x. (2) if x { x1 , . . . , xn } we may assume 1 x = x1 without loss of generality. Now let bi := a 1 ai S (for 1 any i 2 . . . n), then x = x1 = a1 a1 x1 = b2 x2 + + bn xn . And as { x2 , . . . , xn } X \ { x } X this implies X x. Thus we have proved the claim in both cases. From this property we can easily derive (D4): suppose x X , X y and X \ { x } y . Then y = 0, as even 0 and hence X \ { x } y . Thus we may apply this property to nd X x, which had to be shown. (ii) (b) = (a): suppose there was some x X such that X \ { x } x. So by denition there would be some elements xi X \ { x } and ai S (where i 1 . . . n) such that x = a1 x1 + + an xn . Now let x0 := x and a0 := 1, then we get xi X for any i 1 . . . n and also a0 x0 + a1 x1 + + an xn = 0. But as a0 = 0 this means that X is S -linear dependent, a contradiction. Thus there is no such x X and this is the independence of X . (a) = (b): suppose there were some (pairwise distinct) elements xi X and ai R (where i 1 . . . n) such that (without loss of generality) a1 = 0 and a1 x1 + + an xn = 0. As a1 = 0 we may let 1 1 bi := a 1 ai for i 2 . . . n. Then x1 = a1 a1 x1 = b2 x2 + + bn xn and hence { x2 , . . . , xn } x1 . As { x2 , . . . , xn } X \ { x1 } this yields X \ { x1 } x1 , that is X is dependent, a contradiction. Thus there are no such xi and ai and this again means, that X is S -linearly indepenendent. (iv) As B = lhS (B ) S -generation and -generation are equivalent notions. And in (ii) we have seen that dependence and S -linear dependence are eqivalent, too. In particular we nd (a) (b) according to the respective denitions of S -bases and bases. And the equivalencies (a) (c) (d) have already been proved in (3.35.(ix)). Finally the we have already shown (b) (e) in (3.39.(vi)). (iii) This is just a reformulation of (3.35.(v)) in the light of (ii). Likewise (v) resp. (vi) is just the content of (3.35.(vii)) resp. (3.35.(x)) using the equivalencies (a) (b) of (iv). Finally (vii) and (viii) are repetitions of (3.35.(xi)) and (3.35.(xii)) respectively. All we did was bringing the denition of the dimension into play. (ix) If P m M then (according to (v)) we may pick up a basis B of P . Then by assumption |B | = dim(P ) = dim(V ) and M is nitely generated. Now by (viii) we nd that B even is an S -basis of M , as |B | = dim(M ). In particular that is P = lh(B ) = M . P Proof of (3.41): 368 17 Proofs - Rings and Modules

(i) By (3.35.(vii)) and (3.40.(i)) we may choose an S -basis A of P Q. And by the same theorems we may extend this basis to a basis B of P , that is A B and B is an S -basis of P . Likewise we pick up an S -basis C of Q, that extends A, that is A C and C is an S basis of Q. Let us denote D := B C and choose any p B C then p lh(B ) lh(C ) = P Q = lh(A). That is there are some (sa ) S A such that sa a = p =
aA bB

p,b b lh(B ) = P

And as A B both sides of the equation are linear expansions of p in terms of the basis B . Due to the uniqueness of the representation this means p = a (and sa = 1) for some a A and hence p A. That is we have proved B C = A. Therefore D is the disjoint union of D = A (B \ A) (C \ A) Hence the cardinality of D is the sum of the cardinalities of A, B \ A and C \ A respectively. And as A B and A C this turns into |D| = |A| + |B \ A| + |C \ A| = |A| + |B | |A| + |C | |A| = |B | + |C | |A| We will now prove, that D = B C is an S -basis of P + Q. First of all it is clear that: P + Q = ln(B ) + lh(C ) = lh(B C ) = lh(D) That is D generates P + Q. Now by denition it remains to prove the linear independence of D, hence: suppose we have a linear dependence equation xd d = 0
dD

By the above D is the disjoint union of A, B \ A and C \ A and hence we may split this sum into three seperate parts: xa a +
aA bB \A

xb b +
cC \A

xc c = 0

The rst way to reassemble this equation is to divide it into a part contained in P = lh(B ) (the left-hand side) and a part contained in Q = lh(C ) (the right-hand side) xb b =
bB cC \A

xc c

Conseqently both - the left- and right-hand sides - are contained in both - P and Q. Now, as P Q = lh(A), there are some (uniquely

369

determined) (ya ) S A such that xb b =

b B aA

ya a

Again we rearrange this equation, splitting it into two parts as a basis representation of 0 within P = lh(C ) (xa ya )a +
aA bB \A

xb b = 0

Now, as B is linearly independent we nd xb = 0 for any b B \ A. By precisely the same arguments we can prove that xc = 0 for any c C \ A. So there only remains a xa a = 0. Again, as A is linearly independent, this means xa = 0 for any a A. Altogether we have xd = 0 for any d D and this is the linear independence of D. That is D is a S -basis of P + Q and therefore we nally got dim(P + Q) = |D| = |B | + |C | |A| = dim(P ) + dim(Q) + dim(P Q)

(ii) As in (i) above, we may choose an S -basis A P of P , as S is a skew eld. Again we may extend A to an S -basis B V of V . Let us now denote B/A := { b + P | b B \ A } V /P . It is easy to see, that B/A corresponds to B \ A, that is b b + P is bijective. Suppose we had b + P = c + P for some b, c B \ A. Then c b P = lh(A). That is either b = c or { b, c } A B is S -linearly dependent. But as B is S -linearly independent (being a basis), the latter is absurd and hence b = c. This is the injectivity and the surjectivity is clear from the deinition of B/A. We now claim that B/A is an S -basis of the quotient V /P . All we need to do is check that B/A is S -linearly independent and generates V /P . For the linear independence we suppose we had some xb S such that 0=
b+P B/A

xb (b + P ) =
b+P B/A

xb b + P =
bB \A

xb b + P

That is b xb b P = lh(A) and hence there were some xa S such that b xb b = a xa a. Altogether that is 0 =
bB \A

xb b
aA

xa a

But as B = (B \ A) A is linearly independent, this can only be if all the xb S and all the xa S are identically 0. And this is what was needed for linear inependence of B/A. Also we have to check that B/A generates V /P . To do this regard any x + P = V /P . As B is a basis of V there are some xb S (where b B ) such that x = b xb b. 370 17 Proofs - Rings and Modules

Again we use the split B = (B \ A) A and A P to get x+P =

bB

xb b

+P =
bB

xb (b + P ) xb (b + P )
b+P B/A

=
bB \A

xb (b + P ) =

As x has been arbitrary we have V /P = lh(B/A) Then as B/A is a linearly independent generating set of V /P it is a basis. And therefore dim V P = |B/A| = |B \ A| = |B | |A| = dim(V ) dim(P )

(iii) The dimension formula now is an easy application of (ii) and the rst isomorphism theorem (3.51.(ii)) which implies, that V kn() =m im()

In particular we have dim(V /kn()) = dim(im()). But by (ii) we now know that dim(V /kn()) = dim(V ) dim(kn()) such that we truly nd the claim from dim(V ) dim(kn()) = dim(im()) P

Proof of (3.31): We rst have to prove the equality aRI = aI . The elemets of aRI are of the form - where ak a and bk = (bi,k ) RI
m m

ak (bi,k ) =
k=1 k=1

ak bi,k

aI

That is aRI aI , conversely consider any a = (ai ) aI and let us denote the support of a by I0 := { i I | ai = 0 }. Note that (by denition of direct sums) I0 is a nite set. Let us also denote the canonical basis of RI by { ej | j I }, that is ej = (i,j ) RI . Then we can write (ai ) =
iI0

ai ei aRI

Altogether we have proved aRI = aI and it remains to prove the isomorphy of (R/a)I and RI /aRI . To do this regard the R-linear map : R I R a
I

: (bi ) (bi + a)

Clearly is surjective, and its kernel is given to be kn = aI RI = aI such that kn = aRI . So by the rst isomorphism theorem, this induces 371

an isomorphism of R-modules
I R : R a R I a I

: (bi ) + aRI (bi + a)

But both RI /aRI and (R/a)I are (R/a)-modules and from the explict construction of it is clear, that even is an (R/a)-module-homomorphism. P Proof of (2.5): (a) = (c) If a i R/m is an ideal, then by the correspondence theorem it is of the form a = a/m for some ideal a i R with m a. But as m is maximal this means a = m or a = R. Thus we have a = { 0 + m } (in the case a = m) or a = R/m (in the case a = R). (c) = (b) R/m is a non-zero (since m = R) commutative ring (since R is such). But it has already been shown that a commutative ring is a eld if and only if it only contains the trivial ideals (viz. seciton 1.6). (b) = (d) Let a m, this means a + m = 0 + m. And as R/m is a eld there is some b R such that b + m is inverse to a + m. That is ab + m = (a + m)(b + m) = 1 + m Therefore there is some m m such that 1 = ab + m aR + m and hence we have truly obtained aR + m = R. (d) = (a) Suppose a i R is some ideal with m a. We want to show that m is maximal hence if a = m then we are done. Else there is some a a with a m. Hence we get aR + m = R. But as a a and m a we get R = aR + m a R. P Proof of (2.6): (a) = (b): by assumption (a) there is some maximal ideal m i R such that j m. Hence we have m + jR = R such that is there is some b R (let a := b) such that 1 aj = 1 + bj = 1 R . (b) = (a): by assumption (b) there is some a R such that 1 aj R . That is (1 aj )R = R is a non-full ideal and hence there is a maximal ideal m i R containing it 1 aj (1 aj )R m by (2.4.(ii)). If we now had j jacR m then also aj m and hecne we would arrive at the contradiction 1 = (1 aj ) + aj m 372 17 Proofs - Rings and Modules

 (a) = (b): if j a jacR then we let a := 1 and as we have just proved above this yields 1 + j = 1 aj R and hence (b). (b) = (c): rst of all we have 1 1 + a since 0 a. Now consider any a = 1 + j and b = 1 + k 1 + a, that is j and k a. Then ab = 1 + (j + k + jk ) 1 + a again, as j + k + jk a. Further we get (1 + j )(1 (1 + j )1 j ) = (1 + j ) j = 1 which implies a1 = (1 + j )1 = 1 (1 + j )1 j . And hence we also have a1 1 + a, since (1 + j )1 j a due to j a. (c) = (a): x j a and consider any a R. Then we also get aj a and hence 1 aj 1 + a R by assumption. But as we have alredy proved above 1 aj R for any a R means j jacR. P Proof of (2.9): (a) = (b): consider a, b R such that 0+ p = (a + p)(b + p) = ab + p. That is ab p and as p is assumed to be prime this implies a p or b p. This again means that a + p = 0 + p or b + p = 0 + p and hence R/p is an integral domain. But R/p = 0 is clear, since p = R and hence 1 + p = 0 + p. (b) = (c): since R/p = 0 is non-zero we have 1 + p = 0 + p or in other words 1 p which already is (1). Now let u, v R \ p, then u + p = 0 + p and v + p = 0 + p. But as /p is an integral domain this then yields uv + p = (u + p)(v + p) = 0 + p. Thus we also have uv p which is (2). (c) = (a): since 1 R \ p we have p = R. Now consider a, b R with ab p. Supposed we had a p and b p then a, b R \ p and hence ab R \ p, a contradiction. Thus we truly have a p or b p. (a) = (d): consider two ideals a, b i R such that ab p. Suppose neither a p nor b p was true. That is there would be a a and b b with a, b p. Then ab ab p. But as p has is prime this yields a p or b p, a contradiciton. (d) = (a): consider a, b R with ab p. Then we let a := aR and b := bR. Then ab = abR p by (1.54), hence we nd a aR = a p or b bR = b p by assumption. As we have also assumes p = R, this means that p is prime. P Proof of (2.7): Let i = j 1 . . . k , then mi mj would imply mi = mj (since mi is maximal and mj = R). Hence there is some ai mi with ai mj . And as we have just seen in (2.5) this means ai R + mj = R. But as ai mi we nd R = ai R + mj mi + mj R. Hence the mi are pairwise coprime. And the second claim follows from this, as we have proved in (1.56.(iv)). 373

Hence it remains to prove the third statement, i.e. the strict desent of the chain m1 m1 m2 . . . m1 . . . mk . To do this we use induction on k (the foundation k = 1 is trivial), and let a := m1 . . . mk and m := mk+1 . For any i 1 . . . k there are ai mi such that ai m again. Now let a := a1 . . . ak a. Then a m, as maximal ideals are prime (see the remark to (2.9) or (2.19) for a proof) and hence a = a1 . . . ak m would imply ai m for some i 1 . . . k . In particular a am m, that is we have found the induction step a (m1 . . . mk ) \ (m1 . . . mk mk+1 ) P Proof of (2.11): (i) We will prove the statement by induction on k - the case k = 1 being trivial. So let now k 2 and let a := a1 . . . ak1 and b := ak . Then ab = a1 . . . ak1 ak p, but as p is prime this implies a p or b p. If b p then we are done (with i := k ). If not then a = a1 . . . ak1 p, so by induction hypothesis we get ai p for some i 1 . . . k 1. (ii) Analogous to (i) we prove this statement by induction on k - the case k = 1 being trivial again. For k 2 we likewise let a := a1 . . . ak1 and b := ak . Then ab p implies a p or b p due to (2.9). If b p then we are done with i := k . Else a p and hence ai p for some i 1 . . . k 1 by induction hypothesis. (iii) Let us choose a subset I 1 . . . n such that a is contained in the union of the bi (i I ) with a minimal number of elements. That is
a

i I

bi

J 1...n : a
i J

bi = #I #J

It is clear that this can be done, since J = 1 . . . n suces to cover a and the number of elements is a total order and hence allows to pick a minimal element. In the following we will prove that I contains precisely one element. And in this case I = { i } we clearly have a bi as claimed. Thus let us assume #I 2 and derive a contradiction: consider some xed j I , as I has minimally many elements we have
a

/
j =iI

bi

That is there is some aj a such that aj bi for any j = i I . In particular we nd aj bj , as the bi (i I ) cover a. Let us pick one such aj for every j I . If #I = 2 then we may assume I = { 1, 2 } without loss of generality. As a1 and a2 a we also have a1 + a2 a. Suppose we had a1 + a2 b1 then a2 = (a1 + a2 ) + (a1 ) b1 as well, a contradiction. Likewise we nd that a1 + a2 b2 . But now a1 + a2 a b1 b2 provides a contradiction. Hence #I = 2 374 17 Proofs - Rings and Modules

cannot be. Now assume #I 3, then by assumption there is some k I such that bk is prime. Without loss of generality (that is by renumbering) we may assume I = 1 . . . m and b1 to be prime. As any ai a we also have a1 + a2 . . . am a. Suppose a1 + a2 . . . am bi for some i 2 . . . m, then a1 = (a1 + a2 . . . am ) + (a2 . . . am ) bi too, a contradiction. And if we suppose a1 + a2 . . . am b1 then a2 . . . am = (a1 + a2 . . . am ) + (a1 ) b1 too. As b1 is prime we nd ai b1 for some i 2 . . . m due to (i). But this is a contradiction, altogether a1 + a2 . . . am is contained in a but none of the bi for i I , a contradiction. Hence the assumption #I 3 has to be abandoned, which only leaves #I = 1. P Proof of (2.13): We will rst show the well-denedness of spec () - it is clear that 1 (q) = { a R | (a) q } is an ideal in R. Thus it only remains to show that 1 (q) X truly is prime. First of all 1 (q) = R as else 1 1 (q) and this would mean 1 = (1) q. But this is absurd, as q is prime and hence q = S . Hence we consider ab 1 (q), i.e. (ab) = (a)(b) q. As q Y is prime, this yields (a) q or (b) q, that is a 1 (q) or b 1 (q) again. Next we will show that for any a R we have (spec )1 (Xa ) = Y(a) . But this is immediate from elementary set theory, since (spec )1 (Xa ) =
q Y | a 1 (q)

= Y(a)

Likewise (spec )(V(b)) V(1 (b)) is completely obvious, once we translate these sets back into set-theory, then we have to show 1 (q) | b q Y
p X | 1 (b) p

Thus we have to show that b q Y implies 1 (b) 1 (q) X . But b q clearly implies 1 (b) 1 (q) and we have just proved, that 1 (q) X as well. Finally it remains to prove (spec )1 (V(a)) = V( (a) i ). So let us rst translate the claim back into elementary set-theory again
q Y | a 1 (q)

= { q Y | (a) q }

But by denition we have 1 (q) = { a R | (a) q }, and hence the inclusion a 1 (q) trivially implies (a) q and vice versa. P Proof of (2.14): (i) First of all a := P is an ideal of R, due to (1.47). And if we pick any p P then a p. As 1 p we also have 1 a and hence a = R. Now let U := R \ a = {R \ p | p P } 375

Then 1 U as we have just seen. And for any u, v U there are p, q P such that u R \ p and v R \ q. But as P is a chain we may assume p q and hence v R \ q R \ p, as well. As p is a prime ideal R \ p is multimplicatively closed, so u, v R \ p implies uv R \ p and hence uv U . That is U is multiplicatively closed and hence a is a prime ideal. (ii) Clearly P is partially ordered under the inverse inclusion relation (as P = ). Now let O P be a chain in P , then by (i) we know that p := O i R is prime. Now pick any q O, then as p q and the assumption on P we nd p P . Hence p is a -upper bound of O. And by the lemma of Zorn this yields that P contains a -maximal element p . But clearly -maximal is -minimal and hence p P . (iii) Let us take Z := { p spec R | condition(p) }, then Z is partially ordered under the inverse inclusion of sets. In any case Z = is non-empty: if we imosed none then by assumption R = 0 and hence R has a maximal (in particular prime) ideal due to (2.4.(iv)). If we imposed p q then q Z . If we imposed a p then by assumption a = R and hence there is a maximal (in particular prime) ideal m containing a m due to (2.4.(ii)) again. Finally if we supposed a p q then we assumed a q and hence q Z . Now let P Z be any chain in Z and p := P . Then by (i) p is a prime ideal of R again. Also a p and p q are clear (if they have been imposed). Hence p Z is an upper bound of P (under ). And hence there is a maximal element p Z by the lemma of Zorn. But as p is maximal with respect to it is minimal with respect to . (iv) By assumption a q and (iii) there is some prime ideal p minimal among the ideals p { p spec R | a p q } and in particular a p q. It remains to prove, that p also is a minimal element of { p spec R | a p }. Thus suppose we are given any prime ideal p with a p and p p . Then in particular p q and by the minimality of p this then implies p = p . Thus p even is minimal in this larger set. (v) Let us take Z := b i R | a b R \ U , then Z is partially ordered under . Clearly Z = is non-empty, since a Z, as we have assumed a U = . Now let B Z be any chain in Z , then c := B is an ideal, due to (2.4). And if we pick any b B then a b c and hence a c. And if b c then there is some b B Z such that b b R \ U . Hence we also nd c R \ U such that c Z again. That is we have found an upper bound c of B . Hence there is a maximal element b Z by the lemma of Zorn. Now let p Z be any maximal element of Z . As p R \ U and 1 U we have 1 p such that p = R is non-full. Now consider any a, b R such that ab p but suppose a p and b p. As we have p p + aR and p Z is maximal we have (p + aR) / R \ U . That is we may choose u (p + aR) U . That is u = p + a for some p p and R. Analogously we can nd some v = q + b (p + bR) U with q p and R. Thereby we get uv = (p + a)(q + b) = pq + aq + bp + ab p 376 17 Proofs - Rings and Modules

since p, q and ab p. But as U is multiplicatively closed and u, v U we also have uv U . That is uv p U , a contradiction to p Z . Hence we have a p or b p which means that p is prime. P Proof of (2.17): (i) First of all a a : b, because if a a then also ab a, as a is an ideal. Next we will prove that a : b i R is an ideal of R. It is clear that 0 a : b since 0b = 0 a. Let now p and q a : b, that is pb and qb a. Then (p + q )b = pb + qb a, (p)b = (pb) a and for any a R we also get (ap)b = a(pb) a which means p + q , p and ap a : b respectively. (ii) First of all a a, because if a a then for k = 1 we get ak = a a. Next we will prove that a i R is an ideal. It is clear that 0 a a. Now let a and b a that is ak a and bl a for some k , l N. Then we get (a)k = (1)k ak (ab)l = al bl
k+l

(a + b)

k +l

=
i=0 k

k + l i k +l i ab i k + l i k +l i ab + i k + l i k +l i ab + i k + l i ki ab i
k+l i=k+1 l

=
i=0 k

k + l i k+li ab i

=
i=0 k

=
i=0

k + l k +j l j a b k+j j =1 l k + l j lj k bl + a b a k+j
j =1

As all these elements are contained in a again, we again found a + b, a and ab a. Thus it remains to prove that ideal. a is a radical Thus let a R be contained in the radical of a, that is ak a for kl k l some k N. And hence there is some l N such that a = (a ) a. But this already means a a and the converse inclusion is clear. (iii) If a a : b then Likewise if we ab a b and hence a b : b again. k are given a a then there is some k N such that a a b and hence a b already. (iv) If 1 R = a : b then b = 1b a. And if b a then for any a R we have ab a, since a is an ideal. But this also means a : b = R. Now let p be a prime ideal of R, if b p then we have alredy seen p : b = R. Thus assume b p. Then a p : b is equivalent to ab p for any a R. But as p is prime this implies a p or b p. The latter is untrue by assumption so we get a p. That is we have proved p : b p and the converse inclusion has been proved in (i) generally. 377

(v) If = A srad R ideal R, then the intersection A i R already is an ideal of R due to (1.47). Thus it remains to show that A is radical. Let ak A, that is ak a for any a A. But as a is radical we nd that a a. As this is true for any a A we found a A. (v) Consider any a R, then a is contained in the intersection of all ai : b i a ai : b for any i I . And this again is equivalent to ab ai for any i I . Thus we have already found the equivalence a
iI

(ai : b)

ab
i I

ai

a
iI

ai

:b P

Proof of (2.18): (c) (a) is trivial and so is (a) = (b). For the converse implication it suces to remark, that a a is true for any ideal a i R (as a a implies a = a1 a). Thus we prove (a) = (d): Let b + a be a nilpotent of R/a, that is bk + a = (b + a)k = 0 + a for some k N. This means bk a and hence b a a by assumption. And from b a we nd b + a = 0 + a. Conversely (d) = (c): consider any b R such that bk a. Then (b + a)k = bk + a = 0 + a and hence b + a is a nilpotent of R/a. By assumption this means b + a = 0 + a or in other words b a. P Proof of (2.19): The equivalencies of the respecive properties of the ideal and its quotioent ring have been shown in (2.5), (2.9) and (2.18) respectively. Hence if m i R is maximal then R/m is a eld. In particular R/m is a non-zero integral domain which means that m is prime. And if p i R is prime, then R/p is an integral domain and hence reduced which again means that p is a radical ideal. Yet we also wish to present a direct proof of this: m maximal = m prime: consider a, b m with ab m, but suppose a m and b m. This means m m + aR and hence m + aR = R, as m is maximal. Hence there are m m and R such that m + a = 1. Likeweise there are n m and R such that n + b = 1. Thereby 1 = (m + a)(n + b) = mn + an + bm + ab But as m, n and ab m we hence found 1 m which means m = R. But this contradicts m being maximal. p prime = p radical: consider a R with ak p. As p is prime we have k 0 (as else 1 = a0 p such that p = R). But p is prime and hence ak p for some k 1 implies a p due to (2.11.(i)). This means that p is radical, due to (2.18.(c)). P Proof of (2.20): 378 17 Proofs - Rings and Modules

 (i) We have to prove a = { p | a p }. Thereby the inclusion is clear, as any p contains a. For the converse inclusion we are given some a R such that a p for any prime ideal p i R with a p. Now let U := 1, a, a2 , . . . , then it is clear that U is multiplicatively closed. Suppose a U = , then by (2.14.(iv)) there is an ideal b i R maximal with a b R \ U . And this ideal b is prime. But as b U = and a U we have a b even though b is a prime ideal with a b. A contradiction. Thus we have a U = , that is there is some b U such that b a. By construction of U b is of the form k b = a for some k N. And this means a a. (i) Let us denote V(a) := { p spec R | a p } and M := V(a) . Then in (i) above we have just seen the identity
a =

V(a)

We now have to prove V(a) = M. As M V(a) the inclusion is clear. For the converse inclusion we are given any a M and any q V(a) and need to show a q. But as a q by (i) there is some p M such that a p p. And as a M we nd a p and hence a q. Thus we have also established inclusion . (ii) By denition it is clear that nilR = 0 is the radical of the zero-ideal. And the further equalities given are immediate from (ii) above. (iii) Let a R be contained in the radical of a, that is ak a for some k N. And hence there is some l N such that akl = (ak )l a. But this already means a a and the converse inclusion is clear. (v) As a b a b we have a b a b by (i). Next let a a b, k a and that is there is some k N such that ak a b. Now a ak b implies a a b. Finally consider some a a b, that is ai a and aj b for some i, j N. Then ai+j = ai aj a b and hence a a b. Altogether we have proved the following chain of inclusions
ab ab

ab

(vi) By induction on (v) we know that ak = a a (k -times). And this obviously equals a such that we get the equality claimed. (iv) In a rst step let us assume k = 1, that is a p a, then in particular we have a p and hence a p. Yet as p is prime we also have p = p due to (2.19). Thereby we get a p a by assumption. Now consider an arbitrary k N. Then by (vi) we have ak p a = ak . Thus by the case k = 1 (using ak instead of a) we get p = ak = a, the latter by (vi) again. (vii) By assumption a is nitely generated, that is there are ai R such that a = a1 , . . . , an i . And as a is contained in the radical of b we k(i) nd that for any i 1 . . . n there is some k (i) N such that ai b. 379

Now let k := k (1) + + k (n) and consider f1 , . . . , fk a. That is

n

fj =
i=1

fi,j ai

for some fi,j R. Using (1.37) we now extract the product of the fj
k k n k

fj =
j =1 j =1 i=1

fi,j ai =
iI j =1

fij ,j aij

where I = (1 . . . n)k and i = (i1 , . . . , ik ) I . For any h 1 . . . n we now let m(h) := # { j 1 . . . k | ij = h } the number of times h appears in the collection of ij . Then it is clear that m(1) + + m(n) = k = k (1) + + k (n) Hence there has to be some index h 1 . . . n such m(h) k(h) that k (h) m(h). Thereby ah divides ah b such that
k n

bi :=
j =1

a ij =
h=1

ah

m(h)

This now can be used to show f1 . . . fk b. Simply rearrange to nd

k k

fij ,j aij =
i I

k j =1

fij ,j bi b

fj =
j =1 iI j =1

Now recall that - by denition - ak consists precisely of sums of elements of the form f1 . . . fk where fj a. As we have seen any such element is contained in b and hence ak b. (viii) If a is contained in the radical of the intersection of the ai then there is some k N such that a is contained in the intersection of the ai . That is ak ai for any i I and hence a ai for any i I . (viii) Let a be contained in the sum of the radicals ai . That is a is a nite sum of the form a = a1 + + . . . an where aj is contained in the radical of ai(j ) . Thereby for any j 1 . . . n we get aj
ai(j )

k (j ) N : aj

k(j )

ai(j )

Now let k := k (1) + + k (n) N, then by the polynomial rule (1.37)

n

=
||=k j =1

aj

( j )

Suppose that for any j 1 . . . n we had (j ) = k (j ), then || = (1) + + (n) < k (1) + + k (n) = k , a contradiction. Hence there is some j 1 . . . n such that k (j ) (j ). And for this j we have (j ) aj ai(j ) . Thus ak is contained in the radical of ai(1) + + ai(n) and in particualr in the radical of the sum of all ai . 380 17 Proofs - Rings and Modules

P Proof of (2.23): Fix any commutative ring (E, +, ) and consider the polynomial ring S := E [ti | 1 i N] in countable innitely many variables over E . Then we take to the quotient where for any 1 i N we have ti i = 0. Formally R := S v where v := ti i |1iN
i

For any f S let us denote its residue class in R by f := f + v. And further let us dene the size of f to be the maximum index i such that ti appears among the variable symbols of f . Formally that is size(f ) := max{ 1 i N | : f [] = 0, i = 0 } Note that this truly is nite, as there only are nitely many such that f [] = 0 and for any = (i ) there also are nitely many i only, such that i = 0. As our exemplary ideal let us take the zero-ideal a := 0. By i construction we have (ti ) = 0 and hence ti 0. On the other hand we have (for any n N with n < i) (ti )n ( 0)n but (ti )n = 0 Thus for any n N us just take some i N with n < i. Then ti let n demonstrates that ( 0) is not contained in 0. Also 0 is not nitely generated. Because if we consider a nite collection of elements f 1 , . . . , f k 0 then just take i > max{ size(fj ) | j 1 . . . k }. Then it is clear that fj E [t1 , . . . , ti1 ] and hence ti f1 , . . . , fk i . And thereby ti f1 , . . . , fk
i

v = f 1, . . . , f k

P Proof of (2.26): (i) Consider any a R, then a b R is equivalent, to (a) b. k k That is i there is some k N such that (a ) = (a) b. This again is ak b R, which is equivalent to a b R. (ii) In order to prove aS aS we have to verify ( a) aS . Thus consider any a a, that is ak a for some k N. But from this we get (a)k = (ak ) (a) aS . And this again is ( a ) aS , which had to be shown. Next, as a a we clearly have aS aS which gives rise to the incusion
aS

aS

k Conversely consider any f S such that f That is there are gi S and ai a such that

a S for some k N.

f k = g1 (a1 ) + + gn (an ) 381

 k(i) As ai is contained in a there is some k (i) N such that ai a. Now let us abbreviate bi := (ai ), l := k (1) + + k (n) and m := k l. Then we get f m = (g1 b1 + . . . gn bn )l =
||=l

l (g1 b1 )(1) . . . (gn bn )(n)

Now x any and suppose for any i 1 . . . n we had (i) < k (i), then we had l = || = (1) + + (n) < k (1) + + k (n) = l an obvious contradiction. Hence for any there is some j 1 . . . n such that (j ) k (j ). And for this j we get bj
(j )

= bj

(j )k(j ) k(j ) bj

= bj

(j )k(j )

aj

k(j )

aS

But as this holds true for any with || = l we nd that f m a S . And this again means that f is contained in the radical of a S . (iii) The equivalence b is a i 1 (b) is a has already been proved in the correspondence theorem (1.61). Thus from now on is surjective, then (a) is a ideal of S due to (1.71). Now compute 1 (a) = { b R | (b) (a) } = { b R | a a : (b) = (a) } = { b R | a a : b a kn() } = { b R | a a : b a + kn() } = { a + kn() | a a } = a + kn() Now by the rst claim b := (a) is a i 1 (b) = 1 (a) = a +kn() is a . And this already is the second claim. P Proof of (3.67): We will rst prove the implication (a) = (b): that is we regard a non-empty family P subm (M ) of sumbodules of M . We will give a proof by contradiction, that is we suppose P had no maximal element. That is for any P P there is some Q P with P Q but P = Q. As P is non-empty we can start with some P0 P . Suppose we have already found P0 P1 . . . Pk where Pi P . As P has no maximal elements Pk is not maximal, that is Pk Pk+1 for some Pk+1 P . Continuing indenitely we would nd a non-stationary chain P0 P1 . . . Pk Pk+1 . . . in contradiction to (a). Next let us prove (b) = (c): that is we are given any submodule Q m M of M . To do this let us regard the set of nitly generated submodules of Q P := { P m Q | P nitely generated } 382 17 Proofs - Rings and Modules

Clearly 0 P , such that P is non-empty. By assumption (b) contains a maximal element P P . Now for any x Q it is clear that P := P + lh(x) is another nitely generated submodule of Q [clearly if P = lh(x1 , . . . , xn ) then P = lh(x1 , . . . , xn , x)]. That is P P again and P P . By maximality of P this is P = P and hence x P = P . As x Q has been chosen arbitarily this means Q P and hence Q = P . In particular Q P is nitely generated itself. Let us rst close the circle by proving (c) = (a): that is we regard some ascending chain P0 P1 P2 . . . of submodules Pk m M of M . As the Pk for a chain we know by (3.19.(v)) that the union P :=
kN

Pk

is a submodule of M again. And by assumption (c) it is nitely generated, that is P = lh(x1 , . . . , xr ) for some xi P . That is for any i 1 . . . r there is some k (i) N such that xi Pk(i) now take s := max{ i(1), . . . , i(r) } N. Then - as the Pk for an ascending chain we have xi Pk(i) Ps again. In particular P = lh(x1 , . . . , xr ) Ps such that for any j N we have P Ps Ps+j P which implies Ps+j = Ps . This means that the chain (Pk ) becomes stationary at the index s, which proves (a). P Proof of (3.71): (i) If M is then P clearly is as well, as any chain of submodules of P also is a chain of submodules of M . Also M/P is again: by the corrspondence theorem (3.14.(ii)) any submodule of M/P is of the form Q/P for some submodule Q m M . Hence any chain of submodules of M/P corresponds to a chain of submodules in M and hence stabilizes. Conversely suppose both P and M/P are noetherian, then we have to prove that M is noetherian. To do this we regard an ascending chain of submodules Q0 Q1 . . . Qk Qk+1 . . . Then Q0 P Q1 P Q2 P . . . is an ascending chain of submodules. As P is noetherian it stabilizes, that is there is some r N such that for any i N we have Qr+i P = Qr P . Also we obtain an ascending chain of submodules of M/P by Q0 + P Q2 + P Q1 + P P P ... P

As M/P is noetherian as well this chain stabilizes, too, that is there is some t N such that for all i N we have (Qt+i + P )/P = (Qt + P )/P . By the correspondense theorem this already is Qt+i + P = Qt + P . Now choose s := max (r, t), then for any i N we have Qs+i P = Qs P and Qs+i + P = Qs + P 383

Now by the modular rule (3.17) this implies Qs+i = Qs which means that M is noetherian, as well. In case that both P and M/P are artinian, we regard a descending chain Q0 Q1 Q2 . . . and use the same arguments as above to show that it stabilizes, as well. All we have to do is replace any by . (iv) Clearly M is isomorphic to the submodule M 0 of M N by virtue M 0 : x (x, 0). Also it is clear that N is isomorphic to of M the quotient (M N )/(M 0) under the isomorhism
M N N M 0 : y (0, y ) + (M 0)

Thus regarding M 0 as a submodule of M N we know that M N is if and only if both M 0 and (M N )/(M 0) are . But due to the isomorphisms this again is equivalent to both M and N being . (ii) It is really easy to see, that M/(P Q) can be emdedded into (M/P ) (M/Q) by virtue of the following map M M M P Q P Q : x + (P Q) (x + P, x + Q)

Because, if x + P = y + P and x + Q = y + Q then we have both x y P and x y Q. And this means x y P Q which yields x + (P Q) = y + (P Q). Hence the above map is injective. As both M/P and M/Q are so is (M/P ) (M/Q) according to (iv). But as M/(P Q) is isomorphic to a submodule of (M/P ) (M/Q) and this isomorphic copy is according to (i), so is M/(P Q). (v) As : M N is surjective, the rst isomorphism theorem (3.51.(ii)) N by virtue of x + kn() (x). yields an isomorphism M/kn() In particular N is isomprphic to a quotient module M/kn() of M . Now, as M is , we nd that M/kn() is due to (i) and hence N is because of the isomorphy. (iii) As P and Q are by (iv) we know that P Q is . But clearly there is a surjective R-module homorphism P Q P + Q by letting (x, y ) x + y . The surjectivity is trivial and the R-linearity is obvious: a(x, y ) = (ax, ay ) ax + ay = a(x + y ) and (u, v ) + (x, y ) = (u + x, v + y ) (u + x) + (v + y ) = (u + v ) + (x + y ). Hence by (v) we nd that P + Q is again, as it is an epimorphic image of P Q. (vi) Since M is nitely generated we can choose some generators xk M , that is M = lh(x1 , . . . , xn ). Then we clearly have an R-module epimorphism by letting
n

M : (a1 , . . . , an )
i=k

ak xk

As R is as a ring it also is as an R-module (in fact this even is equivalent). Using (iv) and induction on n it is clear that Rn = R R R (n-times) is again. Hence M is , by (v), as it is an epimorphic image of a module. 384 17 Proofs - Rings and Modules

(vii) Let us denote the R-module homorphisms by : L M and : M N , that is we regard the following exact chain 0 L M N 0 As is injective we nd that kn( ) = im() is isomorphic to L under L kn( ) : x (x). And as is surjective, by the rst isomorphism theorem (3.51.(ii)), we also get the isomorphism M
kn( ) N : y + kn( ) (y )

By (i) we know, that M is if and only if both kn( ) and M/kn( ) are . And due to the above isomorphisms, this now is equivalent to both L and N being . P Proof of (3.73): (a) = (b): given any x M with x = 0 we clearly know, that P = Rx m M is a submodule of M . And since x = 1 x P we have P = 0, which only leaves M = Rx, by assumption (a). (b) = (a): consider any submodule P m M of M . If P = 0, then there is nothing to prove, else pick up some x P with x = 0. Now we get M = Rx by assumption (b), but as also M = Rx P M this implies P = M . (b) = (c): choose any x M with x = 0, which is possible since M = 0 is non-zero. By assumption (b) we then have M = Rx. Now let m := ann(x) i R. Then by (3.16.(ii)) we have the following isomorphy of R-modules
M : b + ann(x) bx : R ann(x)

Now suppose a i R is an ideal of R such that m a. Then the above isomorphy transfers a/m to a submodule P = (a) m of M . But as (b) implies (a) this means P = 0 or P = M . In case P = 0 we have a = m and in case P = M we have a = R. This shows that m even is a maximal ideal of M .
M is that isomorphism of R-modules (c) = (a): suppose : R/m and consider some submodule P m M of M . Then a := 1 (P ) m R/m is an R-submodule of R/m. As R is commutative it is clear that a even is an ideal of R/m. But as m is maximal, the only ideals of R/m are 0 and R/m itself. In case a = 0 we see P = (a) = 0 and in case a = R/m we nd P = im() = M . Hence 0 and M are the only submodules of M , as has been claimed in (a).

P Proof of (3.76): For the moment being - as long as we have not proved that the length of M 385

is uniquely determined - let us denote by (M ) the minimum length of of a composition series of M (M ) := min{ r N | (M0 , M1 , . . . , Mr ) composition series of M } Also, given any two submodules U and V m M , we denote the set of all chains of submodules that start in P and end in Q by C (U, V ), that is C (U, V ) := (P0 , P1 , . . . , Pr ) i 0 . . . r : Pi m M U = P0 P1 . . . Pr = V

And we say that two chains (P0 , P1 , . . . , Pr ) and (Q0 , Q1 , . . . , Qs ) in C (U, V ) are equivalent, written as (P0 ,1 , . . . , Pr ) (Q0 , Q1 , . . . , Qs ), i r = s and there is some permutation Sr such that for any i 1 . . . r we have the following isomorphy of the quotient modules Pi Q(i) Pi1 =m Q(i)1

And in this case we say that assigns the equivalence of (P0 , P1 , . . . , Pr ) and (Q0 , Q1 , . . . , Qr ). Thereby truly is an equivanence relation: clearly (P0 , P1 , . . . , Pr ) (P0 , P1 , . . . , Pr ) is assigned by the identity 1 1 Sr . And if (P0 , P1 , . . . , Pr ) (Q0 , Q1 , . . . , Qs ) is assigned by Sr then the reverse relation (Q0 , Q1 , . . . , Qs ) (P0 , P1 , . . . , Pr ) is assigned by the inverse permutation 1 Sr . And nally, if (N0 , N1 , . . . , Nr ) (P0 , P1 , . . . , Ps ) is assigned by and (P0 , P1 , . . . , Ps ) (Q0 , Q1 , . . . , Qt ) is assigned by then we also get the equivalency (N0 , N1 , . . . , Nr ) (Q0 , Q1 , . . . , Qt ) that is assigned by the composition , since Ni P(i) Q ((i)) Ni1 =m P(i)1 =m P ((i))1

(i) Let (M0 , M1 , . . . , Mr ) be a composition series of M of minimal length r = (M ) and denote Pi := Mi P . Let us also denote i : P i Mi M : x x + Mi 1 i1 then it is clear that kn(i ) = Mi1 P = Pii . Hence we nd an R-module-monorphism (this is the rst isomorphism theorem) : x + Pi1 x + Mi1 i : Pi P Mi M i 1 i 1 That is Pi /Pi1 is isomorphic to a submodule of Mi /Mi1 . But as Mi /Mi1 is a simple module this implies Pi /Pi1 = 0 (which yields Pi = Pi1 ) or i even is an isomorphism (which yields that Pi /Pi1 is simple again). Altogether Pi = Pi1 or i is an isomorphism Let us now omit those Pi with Pi = Pi1 , that is if we take to the index set I := { i 1 . . . r | Pi = Pi1 } and q := (#I ) 1 then we may renumber the index set I = { (0), (1), . . . , (q ) } with 0 = (0) <

386

17 Proofs - Rings and Modules

(1) < < (q ) and thereby obtain a composition series of P 0 = P(0) P(1) . . . P(q) = P Thereby P(q) = Mr P = M P = P . Hence we have obtained a composition series of P of length q r. Suppose we had q = r, then we did not omit any i 0 . . . r, that is i is an isomorphsim for any i 1 . . . r. Then we can prove Pi = Mi for any i 0 . . . r by induction on r: Clearly P0 = 0 = M0 and if Pi1 = Mi1 then, as i is surjective, it is just the identity map on Mi /Mi1 . In particular we get P = M P = Mr P = Pr = Mr = M which contradicts the assumption P = M . (iii) Recall that a composition series (M0 , M1 , . . . , Mr ) of M is an element of C (0, M ) such that every quotient Mi /Mi1 is simple. We will now prove, that any two composition series (M0 , M1 , . . . , Mr ) and (N0 , N1 , . . . , Ns ) of M are equivalent. Without loss of generality let us assume r s, then we will use induction on the length r of the shorter chain. To be precise the induction hypothesis will be: if (M0 , M1 , . . . , Mr ) and (N0 , N1 , . . . , Ns ) are two composition series of some module Mr = Ns of length r n, then (M0 , M1 , . . . , Mr ) (N0 , N1 , . . . , Ns ) are equivalent already. For n = r = 0 we have 0 = M0 = Mr = M , that is M = 0 and hence s = 0, too. Thus for n = 0 there is nothing to prove. Also if n = r = 1 then M = M/0 = M1 /M0 and hence M is simple. Therefore (0, M ) is the only composition series of M , as M admits no other submodules. So from now on we only have to regard r = n 2 and can assume that the induction hypothesis is already proved for length n 1. As a rst case let us assume Mr1 Ns1 . Then we nd the inclusions Mr1 Ns1 M = Mr . But as Mr /Mr1 is simple this already means Mr1 = Ns1 as Ns1 = Ns = M = Mr . Hence (M0 , M1 , . . . , Mr1 ) (N0 , N1 , . . . , Ns1 ) by induction hypothesis. But as Mr /Mr1 = M/Mr1 = M/Ns1 = Ns /Ns1 even is an equality of R-modules it is clear that this equivalence extends to an equivalency of the same chains, appended by M (M0 , M1 , . . . , Mr1 , Mr ) (N0 , N1 , . . . , Ns1 , Ns ) Likewise, as a second case, we suppose Ns1 Mr1 then the inclusions Ns1 Mr1 M = Ns imply Mr1 = Ns1 as Ns /Ns1 is simple and Mr1 = Mr = M = Ns . Thus this case ends up in the same equivalence, as the rst case. So in the last case we may assume Mr1 / Ns1 and Ns1 / Mr1 . Then we nd the inclusions Mr1 Ms1 + Ns1 M = Mr such that we get the equality Mr1 + Ns1 = M . Now let P := Ns1 = M and Pi := Mi P as in (i) above. As we have seen there is some q < r such that (0) < (1) < < (q ) satises { (0), (1), . . . , (q ) } = I = { i 1 . . . r | Pi = Pi1 } 387

Suppose we had Pr1 = Pr = Mr P = P , then Mr1 P = Pr1 = P implies Ns1 = P Mr1 which has already been dealt with. Hence we have Pr = Pr1 that is r I and hence necessarily (q ) = r. Let us nally abbreviate Qi := P(i) = M(i) P , then Qq = P(q) = Pr = Mr P = P = Ns1 . Also, as Pr1 = Pr we nd from the construction of the (i) that Qq1 = Pr1 (even though it might well be, that Pr2 = Pr1 but this does not bother us). As we have seen in (i) the Qi then form a composition series of P = Ns1 0 = Q0 Q1 . . . Qq = Ns1 That is (Q0 , Q1 , . . . , Qq1 , Qq ) = (Q0 , Q1 , . . . , Qq1 , Ns1 ). And as as q r 1 we may use the induction hypothesis to obtain the equivalency (Q0 , Q1 , . . . , Qq1 , Ns1 ) (N0 , N1 , . . . , Ns1 ). In particular we may append these chains by Ns to nally arrive at (Q0 , Q1 , . . . , Qq1 , Ns1 , Ns ) (N0 , N1 , . . . , Ns1 , Ns ) We will now prove the equivalency of (Q0 , Q1 , . . . , Qq1 , Mr1 , Mr ) and (Q0 , Q1 , . . . , Qq1 , Ns1 , Ns ) which will be assigned by the transposition = (q 1 q ). That is we will prove the following isomorphies Mr1 Ns1
Ns Qq1 Ns1 : x + Qq1 x + Ns1 Mr Qq1 Mr1 : x + Qq1 x + Mr1

The injectivity is clear from the denition, if x Mr1 with x + Ns1 = 0 + Ns1 then x Mr1 Ns1 = Qq1 . And likewise, if x Ns1 with x + Mr1 = 0 + Mr1 then x Ns1 Mr1 = Qq1 again. Hence Mr1 /Qq1 is isomorphic to a submodule of the simple module Ns /Ns1 . But Mr1 = Qq1 = Mr1 P would mean Mr1 P = Ns1 which has been dealt with in the rst case. Hence here x + Qq1 x + Ns1 is an isomorphism. Likewise Ns1 /Qq1 is isomorphic to a submodule of the simple module Mr /Mr1 but Ns1 = Qq1 = Mr1 Ns1 would imply Ns1 Ns1 which has been dealth with in the second case. Hence x + Qq1 x + Mr1 is an isomorphism, as well. But as q r 1 we may use the induction hypothesis once more to get the equivalency (Q0 , Q1 , . . . , Qq1 , Mr1 ) (M0 , M1 , . . . , Mr2 , Mr1 ). And again this can be appended, to (Q0 , Q1 , . . . , Qq1 , Mr1 , Mr ) (M0 , M1 , . . . , Mr2 , Mr1 , Mr ). So summarizing all the results we have nally proved, that (N0 , N1 , . . . , Ns2 , Ns1 , Ns ) (Q0 , Q1 , . . . , Qq1 , Ns1 , Ns ) (Q0 , Q1 , . . . , Qq1 , Mr1 , Mr ) (M0 , M1 , . . . , Mr2 , Mr1 , Mr ) (ii) Let (P0 , P1 , . . . , Pk ) be a composition series of P and (Q0 , Q1 , . . . , Qn ) be a composition series of M/P . Due to the correspondence theorem any submodule Qi of M/P is of the form Qi = Mi /P for some 388 17 Proofs - Rings and Modules

submodule Mi m M with Pk = P M . Then we get M1 P1 M1 Q1 P0 0 = P = Q1 =m Pk =

That is: as P1 /P0 is simple, so is M1 /Pk . Also due to the third isomorphism theorem (3.51) we also nd for any i 1 . . . n Qi Mi /P Mi Qi1 = Mi1 /P Mi1

where this isomorphism is given by (x + P ) + (Mi1 /P ) x + Mi1 which is not important however. Thereby we have found, that any Mi /Mi1 is simple again, such that we can assemble a composition series of M by letting (P0 , P1 , . . . , Pk , M1 , . . . , Mn ). In particular M is of nite length and we have found the equality (M ) = k + n = (P ) + M P

It remains to prove that conversely, if M has nite length, then P and M/P have nite length, too. The case of P has already been shown in (i), it remains to prove this for M/P . Hence let (M0 , M1 , . . . , Mr ) be a composition series of M , then we dene Qi := Mi + P P m M P

It is clear that Q0 = P/P = 0 and Qr = (M + P )/P = M/P . And the third isomorphism theorem (3.51) yields the following isomorphy Mi + P Mi 1 + P =m (Mi + P )/P Qi (Mi1 + P )/P = Qi1

Explictly this isomorphism is x + (Mi1 + P ) (x + P ) + Qi1 but this doesnt matter, really. Let us now regard the following map i : M i M Mi + P M : x + Mi1 x + (Mi1 + P ) i 1 i 1 + P It is clear that i is surjective, given any x Mi and p P we see that x + Mi1 x + (Mi1 + P ) = x + p + (Mi1 + P ). And the kernel of i is obviously given to be kn(i ) = Mi (Mi1 + P ) M m Mi M i1 i 1 But as Mi /Mi1 is simple this either means kn(i ) = 0 (and hence i is injective, thereby bijective) or kn(i ) = Mi /Mi1 (and thereby Mi (Mi1 + P ) = Mi which means Mi Mi1 + P such that Mi + P = Mi1 + P which implies Qi = Qi1 ). Altogether we have Mi + P = Mi1 + P or i is an isomorphism Let us now take I := { 1 . . . r | Qi = Qi1 }, then for any i I we have

389

that i is an isomorphism, such that Mi Mi 1 =m Mi + P Mi 1 + P =m Qi Qi1

That is for any i I we nd that Qi /Qi1 is a simple module. And for any j I we have seen Qj = Qj 1 . Therefore we can pick up some indices 0 (0) < (1) < < (n) r such that I = { (0), (1), . . . , (n) } and have found a composition series of M/P by (Q(0) , Q(1) , . . . , Q(n) ). In particular M/P has nite length again and the equivalency holds true. (v) Without loss of generality we may assume P0 = 0 and Pk = M . Because if P0 = 0 or Pk = M then we may append the series (P0 , . . . , Pk ) to (0, P0 , P1 , . . . , Pk , M ) which we will rene instead. We will now prove this statement by induction on the length r = (M ) of M . If r = 0 we have M = 0 and hence the only series of sumbodules of M is (0) which is a composition series already. Likewise for r = 1 we know that M is simple, such that (0, M ) is the only series of submodules which is a composition series already. Thus let us now assume the renement-property has been established for any series of submodules in any R-module of length r 1 or less. If k 1 then any composition series of M appends (P0 , P1 ) = (0, M ). And as M was assumed to have nite length, there is some composition series. Thus we may regard the case k 2. Then 0 P1 M such that by (i) we have (P1 ) < (M ) = r and by (ii) also (M/P1 ) = (M ) (P1 ) < (M ) = r. By induction hypothesis we hence get a renement (L0 , L1 , . . . , Lp ) of (0, P1 ), that is L0 = 0 and Lp = P1 . And denoting Qi := Pi /P1 for i 1 . . . k , the induction hypothesis also yields a renement (N1 , N2 , . . . , Nq ) of (Q1 , . . . , Qk ). Thereby Ni is of the form Ni = Mi /P1 for some Mi m M due to the correspondence theorem of modules. Therefore M ( i ) Pi P1 = N(i) = Qi = P1

which yields M(i) = Pi for i 1 . . . k . We have seen in (ii) that (L0 , L1 , . . . , Lp , M2 , . . . , Mq ) is a composition series of M . And any Pi can be found inside of this composition series: P0 = L0 , P1 = Lp = M1 and Pi = M(i) for any i 2 . . . k . Hence (L0 , L1 , . . . , Lp , M2 , . . . , Mq ) is the renement of (P0 , P1 , . . . , Pk ) we sought. (iv) It is clear that is a partial order on C (0, M ), so we only have to prove the three equivalencies. (a) = (b): consider any renement (M0 , M1 , . . . , Mr ) (N0 , N1 , . . . , Ns ), then by (v) we can choose a renement of (N0 , N1 , . . . , Ns ) to a composition series (C0 , C1 , . . . , Ct ) of M . By (ii) the composition series has length t = (M ) = r. And thereby r s t = r implies r = s, which means (M0 , M1 , . . . , Mr ) = (N0 , N1 , . . . , Ns ). (b) = (c): now suppose (M0 , M1 , . . . , Mr ) was no composition series of M , that is there is some i 1 . . . r and some Q m M such that Mi1 Q Mi . Then we already had a true 390 17 Proofs - Rings and Modules

renement of (M0 , M1 , . . . , Mr ) by (M0 , . . . , Mr ) < (M0 , . . . , Mi1 , Q, Mi , . . . , Mr ) such that (M0 , M1 , . . . , Mr ) could not have been maximal. Finally the implication (c) = (a) has already been proved in (iii). (vi) Consider any i : Mi1 Mi then, by the rst isomorphism theorem, im( ) and hence by (ii) we have Mi1 /kn(i ) i (kn(i )) + (im(i )) = (kn(i )) + Mi1 kn(i ) = (Mi1 )

Now recall that the i-th homology module is dened by Hi := kn(i+1 )/kn(i ), then the claim follows by easy computation using (ii) again
n n

(1)i (Hi ) =
i=0 i=0 n

(1)i

kn(i+1 )

kn(i )

(1)i [ (kn(i+1 )) (kn(i ))]

i=0 n+1 n

=
i=1 n

(1)i1 (kn(i+1 ))
i=0

(1)i (kn(i ))

=
i=1

(1)i1 [ (kn(i )) + (kn(i ))] +(1)n (kn(n+1 )) (im(0 ))

n

=
i=1 n

(1)i1 (Mi ) + (1)n (Mn ) (0) (1)i (Mi )

i=0

P Proof of (17): Let us denote the quotient ring Q := R/ann(M ). If now M is an R-module, then M can also be turned into a Q-module under the scalar multiplication (b + ann(x)) := bx as has been shown in (3.16.(iii)). And from this construction it is clear that (for any X M ) lhR (X ) = lhQ (X ) In particular the Q-submodules of M are precisely the R-submodules of M . And therefore it is equivalent wether M is noetherian (resp. artinian) as an R-module or an an Q-module. This will be used in the proof of (i) and (ii). (i) If M is noetherian, then in particular it is nitly generated by property (c) of noetherian modules. And also, if M is noetherian, then M n is noetherian by (3.71.(iv)). But by (3.16.(vi)) we nd an embedding Q M n of R-modules (where n = rank(M )). That is, there is an isomorphic copy of Q as an R-submodule of M n . As M n is noetherian 391

this implies, that Q is a noetherian R-module, hence a noetherian Q-module and hence a noetherian ring. Conversely suppose Q is a noetherian ring and M a nitly generated Rmodule. Then Q is nitely generated as an Q-module (using the same generators) and hence a noetherian Q-module by virtue of (3.71.(vi)). But this again means that M is a noetherian R-module, as well. (ii) If M is artinian, then M n is artinian by (3.71.(iv)). But by (3.16.(vi)) we nd an embedding Q M n of R-modules (where n = rank(M ), which is nite by assumption). That is, there is an isomorphic copy of Q as an R-submodule of M n . As M n is artinian this implies, that Q is an artinian R-module, hence an artinian Q-module and hence an artinian ring. Conversely suppose Q is an artinian ring and M a nitly generated Rmodule. Then Q is nitely generated as an Q-module (using the same generators) and hence an artinian Q-module by virtue of (3.71.(vi)). But this again means that M is an artinian R-module, as well. (iii) If M is nitely generated and artinian then Q = R/ann(M ) is an artinian ring by virtue of (ii). But from (2.37.(ii)) we know that any artinian ring already is noetherian. Hence Q is noetherian and M is nitely generated, such that M is noetherian by (i). (iv) Let us regard Kn := kn(n ) m M , then it is clear that the Kn form an ascending chain of submodules of M K0 K1 K2 . . . As M is noetherian this chain has to stabilize at some point s N, that is Ks+1 = Ks . Now consider any y kn(), as is surjective so is s : M M . Hence there is some x M such that y = s (x). Now 0 = (y ) = (s (x)) = s+1 (x) that is x Ks+1 = Ks . That is y = s (x) = 0 which implies that already is injective, as well. (v) Let us regard In := im(n ) m M , then it is clear that the In form a descending chain of submodules of M I0 I1 I2 . . . As M is artinian this chain has to stabilize at some point s N, that is Is+1 = Is . Now consider any y M , as s (y ) Is = Is+1 there is some x M such that s (y ) = s+1 (x) = s ((x)). As is injective, so is s : M M and hence we nd y = (x). As y has been arbitrariy this means that already is surjective. (vi) Let us rst suppose that r = (M ) < that is M admits a composition series (M0 , M1 , . . . , Mr ). We will prove that M is noetherian and artinian by induction on r. If r = 0 then M = 0 so there is nothing to prove. And if r = 1 then M is simple and hence 0 M is the only chain of submodules of M . In particular M is noetherian and artinian. Let now r be aritrary and suppose the claim has been proved for any module of length less than r. As M1 is simple it is noetherian and artinian as we have argued for r = 1. Also by (??.(ii)) we have 392 17 Proofs - Rings and Modules

(M/M1 ) = (M ) (M1 ) < (M ) = r so by induction hypothesis M/M1 is noetherian and arininan too. But (3.71.(i)) now implies that M is noetherian and artinian, as well. Conversely suppose M is both, artinian and noetherian. Then we have to prove that M is of nite length. To do this dene the set K of all those submodules of M that have a composition series K := { K m M | (K ) < } As 0 K we see that K is non-empty and hence (as M is noetherian) has a maximal element K K. By construction K has a composition series, say (K0 , K1 , . . . , Kr ). Now suppose we had K = M , then we take to the set L of all submodules of M that strictly contain K L := { L m M | K L } As K = M we have M L such that L is non-empty. Hence (as M also is artinian) there is a minimal element L L. Now suppose there is some submodule P m M with K P L . If P = K then P = L as L is minimal with this property. That is L /K is simple and hence we nd a composition series of L by (K0 , K1 , . . . , Kr , L ). But remember, K has been maximal among those submodules of M that admit a composition series so it cannot be extended to K L . Hence the assumption K = M has to be false and in particular M = K itself has a composition series. (vii) First of all (e) = (d), as for any i 1 . . . n we can take Ni := M1 Mi 0 0 m M . To be precise also let N0 := 0 m M then for any i 1 . . . n we clearly have the isomprphy N /N Mi i i1 : xi xi + Mi1 and in particular Mi /Mi1 is simple, by assumption (e). Therefore (N0 , N1 , . . . , Nr ) is a composition series of M such that (M ) = r < is nite. The implications (d) = (a) and (d) = (b) have already been shown in (vi). Also (b) = (c) is clear, as any submodule of a noetherian module is nitely generated, in particular M itself. We will now prove (c) = (e): as M is nitely generated we may pick up some xk M such that M = lh(x1 , . . . , xn ). By assumption M is semi-simple, that is there are some simple submodules Mi m M such that M = Mi
i I

That is any xk can be written as a nite sum of some elements xk,i Mi and hence there is some nite subset (k ) I such that xk =
i(k)

xk,i

As n N is nite and any (k ) I is nite, so is the union := (1) (n) of these sets. And as the xk generate M we see that 393

M in fact is a nite direct sum, as claimed M =

i

Mi

Hence we have established the chain (e) = (d) = (b) = (c) = (e). And as we alredy know (d) = (a) as well, it suces to prove (a) = (e): as M is semi-simple we have M = i Mi as above. Thereby we may omit all those i I with Mi = 0. Now suppose I was innite, then there was some injection : N I . Then we take I (k ) := I \ { (j ) | j k } for any k N and let Nk :=
iI (k)

Mi

Then it is clear that the Nk form a strictly descending chain of submodules of M , that is N0 N1 N2 . . . in contradiction to M being artinian. Hence I could not have been innite such that (e) holds true. If now M even is a vector-space, that is R is a eld, then clearly M is nitely generated if and only if M is nite dimensional. Hence (f) is equivalent to (c) and therefore to any other of the statements, as well. (viii) Let us denote the set of all ideal-induced submodules aM of M such that M/aM is noetherian by P P :=
aM | a i R, M aM is not noetherian

Suppose M was not noetherian, then 0 P and hence P is not empty. By assumption and (2.28) P then contains a maximal element a M . Let us now denote M := M a M and A := R ann(M ) Then A is a commutative ring again and we can regard M as an Amodule. Note that by construction M is not noetherian. Now, given any ideal a i A in A, by the correspondence theorem a is of the form a = b/ann(M ) for some b i R with ann(M ) b. We will now prove a b To do this regard any a a then for any x M we get ax a M . That is a(x+a M ) = ax+a M = 0+a M and hence a ann(M ) b. As a has been arbitrary this proves a ann(M ) b. Now, by construction it is clear that aM = bM/a M . And thereby the third isomorphism theorem yields M M/a M M aM = bM/a M =m bM

But as a b and a has been maximal this implies either a = b 394 17 Proofs - Rings and Modules

(and hence a = 0) or M /aM is noetherian. Now let Q := Q m M | annA M Q = 0

Recall A = R/annR (M ) and that the scalar multiplication of M (with A) is dened by (a + ann(M ))(x + a M ) = ax + a M . Therefore we have annA (M ) = annR (M )/annR (M ) = 0 A such that 0 Q. In particular Q is non-empty. If now (Qi ) Q is a chain in Q then we take Q := i Qi to be the union of the Qi . As (Qi ) is a chain Q is a submodule of M again. Consider any a A with a annA (M /Q) = 0, that is a(M /Q) = 0 or in other words aM Q. As M is nitely generated, so is M say M = lh(x1 , . . . , xn ), then we have axk Q for any k 1 . . . n and hence axk Qi(k) for some i(k ) I . As the Qi form a chain we may take Qi to be the maximum of Qi(1) to Qi(n) . Then for any k 1 . . . n we have axk Qi . But as the xk generate M this implies aM Qi and therefore a ann(M /Qi ) = 0 such that a = 0. That is annA (M /Q) = 0 and hence Q Q again. Hence any chain (Qi ) of Q has an upper bound Q and hence the lemma of Zorn yields a maximal element Q of Q. Now take to M := M Q

If a i A is any ideal with a = 0 then by (3.30.(v)) we have aM = (aM + Q )/Q . Therefore the third isomorphism theorem again yields M
aM
= M /Q (aM + Q )/Q =m M aM + Q

As we have shown before we have a b for a = b/ann(M ). But as a = 0 and a has been maximal we nd that M /aM is noetherian. But as we also have an epimorphism M
aM

aM + Q

This implies that M /(aM + Q ) is noetherian. By the above isomorphism, this makes M /aM noetherian. So at last we have arrived at an A-module N := M that satises the following two properites a i A : a = 0 = N aN is noetherian Q m N : Q = 0 = ann N Q = 0 We will now prove that N = M is noetherian by showing, that any submodule of N is nitely generated: So suppose Q m N is any submodule of N . If Q = 0 then there is nothing to show, else there is some a A with a = 0 such that a(N/Q) = 0, that is aN Q. But as aA = 0 we know that N/aN is noetherian and hence Q/aN m N/aN is nitely generated. And as M is nitely generated, so are N = M and aN . But as both Q/aN and aN are nitely generated (3.59) tells us that Q is nitely generated. 395

By now we have established that N = M = M /Q is noetherian. Therefore (i) implies that A/ann(N ) is a noetherian ring. Yet by construction we have ann(N ) = ann(M /Q ) = 0 such that A/ann(N ) = A. That is A is a noetherian ring. Yet M is a nitely generated Amodule such that (3.71.(vi)) implies that M = M/a M is noetherian. But we have constructed M is such a way, as to be not noetherian. This contradiction can only be solved, if M has been noetherian from the start.

Proof of (3.48):

(i) Let B := adj(t1 1n A) matn (R[t]) be the adjoint matrix of t1 1n A, then it has been shown (as R[t] is a commutative ring) that (t1 1n A)B = det(t1 1n A)1 1n = c(t)1 1n That is we have n2 equations in the polynomial ring R[t] encoded in a single equation for n n matrices over R[t]. And by construction the degree of all these polynomial equations is n as well. Let us now write
n1

B = adj(t1 1n A) =
k=0

tk Bk

for some Bk matn (R) then the above equation takes the following form
n1 n1 n1

c(t)1 1n = (t1 1n A)
k=0 n1

t Bk =
k=0

k+1

Bk
k=0

tk ABk

= tn Bn1 +
k=1

tk (Bk1 ABk ) AB0

Comparing coecients yields c[n]1 1n = Bn1 and hence Bn1 = 1 1n as c(t) is monic c[n] = 1. Also AB0 = c[0]1 1n = (1)n det(A)1 1n . Also this is no wonder, as B0 = adj(A). Also for any k 1 . . . (n 1) c[k ]1 1k = Bk1 ABk Let us now multiply all these equations with Ak from the left, that is Bn1 An = An and for any k 1 . . . n 1 we get c[k ]Ak = Ak Bk1 Ak+1 Bk Then c(A) = 396
k

c[k ]Ak can be replaced in every coecient c[k ]Ak 17 Proofs - Rings and Modules

seperately to nd a telescopic sum, that vanishes identically

n n1

c(A) =
k=0

c[k ]Ak = An +
k=1 n1

c[k ]Ak + c[0]1 1n

= An + = An +
k=1

(Ak Bk1 Ak+1 Bk ) + c[0]1 1n

k=1 n1 n

Ak B k 1
k=2

Ak Bk1 + c[0]1 1n

= An + AB0 An Bn1 + c[0]1 1n = An c[0]A An + c[0]1 1n = 0 (ii) As M is nitely generated we nd some xk M such that M = lh(x1 , . . . , xn ) where n = rank(M ). Now, as (xj ) M again, for any j 1 . . . n we also nd some ai,j R such that
n

(xj ) =
i=1

ai,j xi

Let A := (ai,j ) matn (R) be the n n matrix of those coecients. Then it is clear that for any x = j bj M we get
n n n

(x) =
j =1 n

bj (xj ) =
j =1 n

bj
i=1 n

ai,j xi

=
i=1 i=1

ai,j bj

xi =
i=1

[Ab]i xi

where we denoted b = (b1 , . . . , bn ) Rn the coecient-vector of x. 1n A) be And likewise k (x) = i [Ak b]i xi . Now let f (t) := det(t1 the characteristic polynomial of A again. Then we have shown in (i) that f (A) = 0. And hence for any x = j bj M we nd
n n n

f ()(x) =
k=0 n

f [k ]k (x) =
k=0 n k

f [k ]

[Ak b]i xi
i=1 n n

=
i=1 n k=0

f [k ][A b]i
n

xi =
i=1 k=0

f [k ]Ak b
i

xi

=
i=1

[f (A)b]i xi =
i=1

0 xi = 0 P

Proof of (3.32): (i) Let us denote J := { i I | mi aM }, then we have to prove that { mj + aM | j J } is an (R/a)-basis of M/aM . So we rst have to check, that this set (R/a)-linearly generates M/aM . Thus consider 397

any x + aM M/aM , that is x M . As the mi generate M there are some ai R (where i I and only nitely many ai = 0 are non-zero), such that x = i ai mi . And thereby we already get x + aM =
iI

ai mi + aM =
i I

(ai + a)(mi + aM )

=
i J

(ai + a)(mi + aM )

where the latter equality holds, as any i I \J yields xi +aM = 0+aM . It remains to prove the (R/a)-linear independence of this set. Thus suppose we are given any bj + a R/a (of which only nitely many are non-zero) such that 0 + aM =
j J

(bj + a)(mj + aM ) =
j J

bj mj + aM

In other terms that is j bj mj aM . But as { mi | i I } generates M we may use (3.30.(ii)) to nd some ai a, of which only nitely many are non-zero, such that bj mj =
j J i I

ai mi

As the mi form an R-basis of M we may compare coecients in these two representations, which yields ai = 0 for any i I \ J and bj = aj a for any j J . Hence we have bj + a = 0 for any j J and this is the (R/a)-linear independence of { mj + aM | j J }. (i) It remains to prove that the mi + aM are pairwise distinct, too. First, if a = R, then aM = M and hence mi aM for any i I . In this case the basis given is empty and there is nothing to prove. So let us now assume a = 0. If mi + aM = mj + aM then we have mi mj aM . Yet as { mi | i I } generates M we may use (3.30.(ii)) again, to nd some ak a, of which only nitely many are non-zero, such that mi mj =
k I

ak m k

If we had i = j then this would yield an equation of R-linear dependence, simply by moving mi mj on the left hand side and decomposing (ai 1)mi + (aj + 1)mj + ak mk = 0
kI, k=i,j

As the mi are R-linearly independant this implies all coecients in this equation to be 0, in particular 1 = ai a. That is a = R which cannot occur. Hence we necesserily have i = j . (ii) (a) = (b): let us choose a generating set { m1 , . . . , mn } of M with a minimal number n of elements. Suppose n 1 then, as m1 M = aM we may invoke (3.30.(ii)) to nd some a1 , . . . , an a such that 398 17 Proofs - Rings and Modules

m1 = a1 m1 + + an mn . Let us now write this equation in the form (1 a1 )m1 = a2 m2 + + an mn As a1 a jacR we know by (2.6) that 1 a1 R is invertible. And hence m1 is contained in the linear hull of { m2 , . . . , mn } by
n

m1 =
i=2

(1 a1 )1 ai mi

That is even { m2 , . . . , mn } generates M in contradiction to the minimality of the number n of generators. So our assumption n 1 has been false, that is n = 0 and hence M = 0. (a) = (b): if a / jacR then - by denition of the Jacobson radical - there is a maximal ideal m i R of R such that a / m. And this again means that there is some u a with u m. Now, as M := R/m is a eld, and u + m = 0 M there is some v + m M such that uv + m = (u + m)(v + m) = 1 + m. Clearly M also is nitely generated, by { 1 + m } as an R-module. Also
aM

= { ab + m | a a, b R } = {c + m | c R} = M

Hereby the central equality is due to the following reasoning: given c R we may take a := u and b := vc, that ab + m = uvc + m = c + m, and the converse inclusion is clear. Thus we have found a nitely generated R-module M with aM = M , but M = 0. (ii) Alternative Version: the more advanced lemma of Dedekind allows to prove the implication (a) = (b) in a straightforward way: as M is nitely generated we nd some a a such that (1 a)M = 0, due to (v). Yet as a a jacR proposition (2.6) yields that 1 a R is invertible. Hence M = (1 a)M = 0 is proved already. (iii) By (3.30.(v)) we have know that a(M/P ) equals (aM + P )/P and hence we have the following identities (by the assumptions on P )
a M P

= aM + P P = M P

Yet M/P is nitely generated and a jacR so by the lemma of Nakayama (ii) we get M/P = 0 and this is nothing but M = P . (iv) If is surjective, then /a clearly is surjective as well - see (3.30.(vi)) for details. So conversely assume /a is surjective, that is for any y N there is some x M that y + aN = a(x + aM ) = (x) + aN That is v := y (x) aN , such that y = (x) + v . As y has been arbitrary this means N = im + aN . Yet as N/im is nitely generated and a jacR this implies im = N by (iii), which is nothing, but the surjectivity of . 399

(v) () If is bijective, then /m is bijective for any ideal a R, as we have seen in (3.30.(vii)). Conversely: as M is nitely generated, the zero-module 0 m M is conite and hence also im m M is conite. So by (iv) we already know, that is surjective, as /a is surjective (even bijective). But a surjective endomorphism of a nitely generated R-module already is bijective by (3.48). (vi) () Consider the identity-map 1 1 : M M : x x on M . Clearly this is an R-linear map satisfying im(1 1) M = aM . And as M is nitely generated the Cayley-Hamilton theorem (3.48) yields some polynomial f = tn + a1 tn1 + + an R[t] such that f (1 1) = 0 and k ak a for any k 1 . . . n. Now compute f (1 1) = 1 1n + a1 1 1n1 + + an1 1 1 + an 1 1 = (1 + a1 + + an )1 1 = (1 a)1 1 where we let a := (a1 + an ) a. Now 0 = im(0) = im(f (1 1)) = im((1 a)1 1) = (1 a)M . That is the a a we constructed using Cayley-Hamilton truly satises (1 a)M = 0. P Proof of (3.52): (ii) In case of a skew-eld S things are simple now: according to the dimension formula (3.41.(iii)) we know that dim(V ) = dim(kn()) + dim(im()) This if is injective then we have kn() = 0 and hence dim(kn()) = 0 which leaves dim(V ) = dim(im()). Thus (as V is nite dimensional) by (3.40.(ix)) we nd that im() = V such that is surjective. Conversely if is surjective, then im() = V and hence dim(im()) = dim(V ), which leaves dim(kn() = 0. Therefore kn() = 0 [if there was some 0 = x kn(), then { x } would be a linearly independent set in kn() and hence dim(kn( )) 1]. But of course this means that is injective. So in any case we have seen surjective implies bijective and injective implies bijective. And the converse implications are trivial. (i) Dedekind-Proof: we will rst present a rather short proof using the lemma of Dedekind: let us regard M as an R[t]-module by virtue of (3.47). Then by construction we get tx = (x) for any x M hence, as is surjective, for any y M there is some x M with tx = y . Hence we get tM = M , in particular aM = M for a := tR[t] i R[t]. Now by then lemma of Dedekind (3.32.(vi)) there is some g a such that (1 g )M = 0. As g a there is some f R[t] with g (t) = tf (t). Let us dene := g () end(M ). Then gx = g ()(x) = (x) for any x M and hence (1 f )M = 0 implies 0 = (1 tg (t))x = x t (x) = x ( (x)) 400 17 Proofs - Rings and Modules

0 = (1 gt(t))x = x g(x) = x ((x)) That is ( (x)) = x and ((x)) = x for any x M . That is is just the inverse map of and in particular is bijective.

(i) Noether-Proof: as the proof above requires the somewhat abstract Dedekind lemma, we would like to supply another proof that requires the theory of noetherian modules only. As M is nitely generated there are some xi M such that M = lh(x1 , . . . , xn ). Then, as (xj ) M we can pick up some pi,j R such that
n

(xj ) =
i=1

pi,j xi

Step (A): we will now prove, that M = lh((x1 ), . . . , (xn )). That is we choose some y M arbitarily. As is surjective there is some x M such that y = (x). And hence there are some aj R such that x = j aj xj . Now compute
n

y = (x) =
j =1

aj (xj ) lh (x1 ), . . . , (xn )

Step (B): as the (xj ) generate M we can conversely choose some qj,i R such that the xi M can be written as
n

xi =
j =1

qj,i (xj )

Remember all we have to prove is the injectivity of . And to do this it suces to prove kn() = 0. Hence regard some k M with (k ) = 0. All we have to prove is k = 0, then the injectivity is established. Step (C): write k as k = i ki xi for some ki R and let A denote the Z-algebra generated by all the pi,j qi,j and ki A := Z
i=1 j =1 n n

{ pi,j , qi,j , ki }

Then A is nitely generated and Z is notherian (even a PID) such that A is a noetherian subring of R, by the Hilbert basis theorem (2.33). Also let N be the A-sub-module of M that is generated by the xi . As N is a nitely generated module over the noetherian ring N is a noetherian A-module itself N := lhA (x1 , . . . , xn ) Step (D): we will now prove that : N M restricts to : N N . That is we will prove that (x) N for any x N . But this is easy 401

to see: if x N then x =
n

aj xj for some aj N . Hence

n n n n

(x) =
j =1

aj (xj ) =
j =1

aj
i=1

pi,j xi =
i=1 j =1

pi,j aj xi

and this is contained in N again, as pi,j A as well. Also we will prove that : N N is surjective, that is for any y N there is some x N such that y = (x). If y = i bi xi then let x := i bi j qi,j A
n n n

(x) =
i=1

bi
j =1

qi,j (xj ) =
i=1

bi xi = y

Step (E): as : N N is surjective, so is k : N N . Now let Ki := kn(k ) m N . Then it is clear, that the Ki form an ascending chain of submodules of N K0 K1 K2 . . . But as N is noetherian this chain has to stabilize, that is there is some s N such that Ks+1 = Ks . As s is surjective and k N , there is some z N such that k = s (z ). But now 0 = (k ) = (s (z )) = s+1 (z ) implies z Ks+1 = Ks . That is k = s (z ) = 0 already. Hence we have nally arrived at k = 0 which is the injectivity of : M M . P Proof of (3.33): (i) Suppose M = 0, as M is nitely generated, the submodule 0 m M is conite. Hence by (3.12) there is a maximal submodule Q m M such that P Q M . As Q is maximal, M/Q is a non-zero, simple R-module [M/Q = 0, as Q = M and by the correspondence theorem (3.14.(ii)): consider a submodule U/Q m M/Q, then U m M satises Q U M and hence Q = U or U = M due to the maximality of Q]. Therefore
m := annR M Q

is a maximal ideal m smax R, according to (3.16.(ix)). By denition of m we have ax + Q = a(x + Q) = 0 + Q for any a m and x M . That is ax Q and hence mM Q. Now by assumption we get Q M = mM Q. But that is Q = M in contradiction to Q = M , such that the assumption M = 0 must have been false. (ii) By (3.30.(v)) we know that m(M/P ) equals (mM + P )/P and hence we have the following identities (by the assumptions on P )
m M P

= mM + P P = M P

for any maximal ideal m i R. Yet as P is conite the quotient module 402 17 Proofs - Rings and Modules

M/P is nitely generated. Hence we may use (i) to nd M/P = 0 and this is nothing but M = P . (iii) If is surjective, then /m is surjective for any ideal m R, as we have seen in (3.30.(vii)). Conversely if /m is surjective, then for any y N there is some x M such that (x) + mN = y + MN . That is there is some z mN such that y = (x) + z im() + mN . And as y N has been arbitrary, this means N = im() + mN . As im is conite and this is tru for any maximal ideal m, we may invoke (ii) to nd im = N , but this is nothing but the surjectivity of . (iv) If is bijective, then /m is bijective for any ideal m R, as we have seen in (3.30.(vii)). Conversely: as M is nitely generated, the zeromodule 0 m M is conite and hence also im m M is conite. So by (iii) we already know, that is surjective, as the /m are surjective (even bijective). But a surjective endomorphism of a nitely generated R-module is bijective by (3.48). P Proof of (3.56): (i) Suppose I and J have the same cardinality, by denition this means, that there is a bijection : I J . And by construction M := RI has the canoncial basis ek = (i,k ) M , where k I . That is the k -th component of ek is 1, all others (i = k ) are 0. Likewise we denote the canonical basis of N := RJ by fl = (j,l ) N , where l J . This clearly gives rise to a well-dened homorphism of R-modules: : M N :
iI

ai ei
i I

ai f(i)

As is surjective and { fj | j J } generates N we see that is surjective. Suppose (ai ) = 0, as is injective and { fj | j J } is R-linearly independent this means that ai = 0 for any i I . This is kn() = 0 and hence also is injective. Altogether is an isomorphism. It remains to prove the converse implication, that is we suppose that M := RI and N := RJ are isomorphic. As R is a non-zero, commutative ring we may use (2.4.(ii)) to pick up a maximal ideal m i R of R (containing 0 R). Then, as m is maximal, F := R/m is a eld. As we have seen in (3.31) the quotient M/mM is isomorphic to F I , by virtue of the following isomorphism
R F I I I mRI : (bi + m) (bi ) + mR

In complete analogy we nd an isomorphism of F -vector-spaces between F J and the quotient N/mN . Now, as M =m N are isomorphic (by assumption), so are the quotient modules M/mM =m N/mN , by virtue of (3.30.(vii)). And therefore we nd a chain of isomorphies of F -vector-spaces connecting
I J J F I =m R mRI =m R mRJ =m F

403

Clearly isomorphic modules have the same dimension (the isomorphism transfers bases). And as F I has dimension |I | (using the canonical basis) altogether we nd |I | = dim(F I ) = dim(F J ) = |J | (ii) Let : R/a R/b be the isomorphism of R-modules, that we assumed to exist. Further let v + b := (1 + a) and u + a := 1 (1 + b). Then - as is R-linear - we get (uv + a) = v (u + a) = v (1 + b) = v + b = (1 + a) As is invertible this implies uv + a = 1 + a and hence u + a = (v + a)1 (R/a) is invertible. Now consider any a R, then a a is equivalent to a + a = 0 + a. And as u + a is a unit this is equivalent to ua + a = 0 + a. Now compute 0 + b = (0 + a) = (ua + a) = a(u + a) = a(1 + b) = a + b That is a + a = 0 + a is equivalent to a + b = 0 + b. Therefore a a is equivalent to a b and this is just the equality a = b of subsets of R. P Proof of (3.58): (i) It has already been shown in section 2.6 that Z is a PID. It only remains to verify, that P is innite. Hence we suppose, that P is nite, say P = { p1 , . . . , pn }. Then we let b := p1 pn + 1 Z and choose an arbitrary prime divisor p 0 of b, that is b = ap for some a Z. As P encompasses all (positive) prime elements of Z there is some i 1 . . . n with p = pi . Without loss of generality we may take p = p1 . Now p1 pn + 1 = b = ap = ap1 Hence 1 = p1 (a p2 pn ) which means that p1 is a unit of Z - an obvious contradiction to being a prime element. Hence P is not nite. (ii) Suppose Q = lh(x1 , . . . , xn ) was nitely generated, where xi = ai /bi . Then we choose a prime number p P such that for any i 1 . . . n we have p | bi . Such a prime exists, as by (i) there are innitely many primes and all the bi are only divisible by nitley many primes. Now suppose 1/p = k1 x1 + + kn xn for some ki Z. Let b := b1 bn and bi := b/bi Z, then 1 = p
n

ki
i=1

ai 1 = bi b

ki ai bi
i=1

b = p
i=1

ki ai bi

In particular p would divide b and hence some of the bi (as p is prime) in contradiction to the choice of p. Hence there can be no nite generating set of Q as an Z-module. 404 17 Proofs - Rings and Modules

It remains to show, that the set P truly generates Q. To do this consider any x = a/b Q and write down a primary decomposition b = sp1
k(1) (n) pk n

where s = 1 or s = 1 and pi P. We will now use induction on the number n of distinct prime factors of b. If n = 0 then we readily have x = a/s = as Z. And if n = 1 then x = as 1
k(1) p1

1
k(1) p1

lh(P )
k(1) k(n1)

For the induction step n 1 n let us abbreviate q := p1

k(n)

pn1

and p := pn . Then x = a/(sqp) = (as)/(qp). As p and q are relatively prime gcd(p, q ) = 1 and Z is a PID there are some u and v Z such that uq + vp = as. Hence we nd (note that u/p lh(P) as in the case n = 1 and v/q lh(P) by induction hypothesis) x = as uq + vp u v = = + lh(P) + lh(P) = lh(P) qp qp p q P Proof of (3.59): Suppose X is a generating set of M with |X | = rank(M ). As : M N is surjective, for any y N there is some x M such that y = (x). Also as X generates M we can choose some ai R and xi X such that x = a1 x1 + + an xn . And clearly this yields
n

y = (x) =
i=1

ai (xi ) lh (X )

As y N has been chosen arbitarily we have just seen, that N = lh((X )). And from this we nd the rst inequality rank(N ) |(X )| |X | = rank(M ) For the second inequality choose some generating sets X P of P and V M/P of M/P such that rank(P ) = |X | and rank(M/P ) = |V |. By construction of M/P any v V is of the form u + P for some u M . Thus for any v V we may pick up some u M such that v = u + P . We now take those u to build up a set U M , that is we have V = { u + P | u U } and |U | = |V |. We will now show, that X U generates M . To do this consider any x M : as x + P M/P = lh(V ) there are some ai R and ui U such that
m n

x+P =
i=1

ai (ui + P ) =
i=1

ai ui

+P

That is x (a1 u1 + + am um ) P . From this again we nd some bj R and pj X such that x (a1 u1 + + am um ) = b1 p1 + + bn pn since X 405

generates P . Altogether this yields

m n

x =
i=1

ai ui +
j =1

bj pj lh(U X )

Again, as x M has been chosen arbitarily, this proves M = lh(X U ), that is X U generates M and from this we get rank(M ) |P U | |P | + |U | = |P | + |V | = rank(P ) + rank M P P Proof of (3.62): (i) If n = 0 then X := { x1 , . . . , xn } degenerates to X = and as X generates M this implies M = 0. Thus the only choice of an element y1 M is y1 = 0 and this obviously is R-linear dependent. Thus we may assume 1 n. As M is the R-linear hull of X , we can choose some ai,j R (where i 1 . . . n and j 1 . . . n + 1) such that
n

yj =
i=1

ai,j xi

Using these elements we rst dene the Z-subalgebra S of R generated by the ai,j . That is, we let S := Z [ai,j | i 1 . . . n, j 1 . . . n + 1] Note that Z is a noetherian ring (even a PID, by (2.59)) and S is a nitely generated Z-algebra (by construction). Hence by the Hilbert basis theorem (2.33) S is a noetherian ring again. We now take to the S -module generated by the xi MS := lhS { x1 , . . . , xn } Now, as MS is a nitely generated S -module and S is a noetherian ring, MS is a noetherian S -module, according to (??). And clearly
n

yj =
i=1

ai,j xi MS

We will now recursively dene the elements yj [m] MS for any index j 1 . . . n + 1 and any m N. We start by taking yj [0] := yj , then we continue
n

yj [m + 1] :=
i=1

ai,j yi [m]

To streamline our notation let us abbreviate zm := yn+1 [m]. Now suppose that Y := { y1 , . . . , yn , yn+1 } was R-linearly independent. As S r R is a subring of R, then in particular Y would be S -linearly independent. In an intermediate claim we will prove, that in this case 406 17 Proofs - Rings and Modules

the elements y1 [m], . . . , yn [m], zm = yn+1 [m], z0 , . . . , zm1 are S -linearly independent for any m N. The case m = 0 is trivial, as yj [0] = yj our assumption on the yj grants their S -linear independence. Let us use induction on m, then for the induction step we have to regard an S -linear relation of the form
n+1 m

aj yj [m + 1] +
j =1 l=0

bl zl = 0
(m+1)

Let us insert the recursive deniton for the yj then we nd a relation of the form
n+1 m

into this relation,

0 =
j =1 n+1

aj yj [m+] +
l=0 n

bl zl
m

=
j =1 n

aj
i=1

ai,j yi [m] +
l=0 n+1 j =1

bl zl
m1

aj ai,j yi [m] +

=
i=1

bl zl + bm yn+1 [m]
l=0

By the induction hypothesis these elements are S -linearly independent, such that we nd bl = 0 for any l 1 . . . m and for any i 1 . . . n also
n+1

aj ai,j = 0
j =1

But this allows to give a linear relation between the yj , just use the denition of the ai,j and compute (in the same manner as before)
n+1 n

n+1 j =1

aj ai,j xi = 0

a j yj =
j =1 i=1

But by assumption the yj are S -linearly independent, such that we necessarily have aj = 0 for any j 1 . . . n + 1 as well. Altogether we have proved the intermediate claim of linear independance. In particular the elements z0 , z1 , z2 and so on are S -linearly independent. Thus - if we dene Mm := lhS { zl | l 0 . . . m } We get an ascending chain M0 M1 M2 . . . of S -submodules of MS . Yet this chain is strictly increasing: suppose we had zm+1 Mm

407

for some m N then there would be some s0 , . . . , sm S such that

m

zm+1 =
l=0

s l zl

which contradicts the S -linear independence of the zl and hence we have M0 M1 M2 . . . . But a strictly increasing chain within the noetherian module MS is impossible. The only way to solve this contradiction is to drop the assumption of R-linear independence of the yj . That is Y is R-linear dependent, as we had claimed. (ii) Since the basis B by denition generates M as an R-module and the rank of M is the minimal cardinality of a generating set, trivially rank(M ) |B | It remains to prove the equality. Let us choose some generating set G M of minimal cardinality, that is |G| = rank(M ). In a rst step we assume that n := |G| N is nite. Suppose we had n < |B |, then we could choose some { b1 , . . . , bn , bn+1 } B . But as G generates M this would mean, that { b1 , . . . , bn , bn+1 } is Rlinearly dependent, according to (i). But this contradicts the R-linear independence of B and hence we have rank(M ) = |G| = n = |B | In the second step, we suppose that |G| is innite. Then necessarily B is innite as well - else we would have |B | < |G| = rank(M ). As G generates M , for any b B M , we can choose a nite subset Fb G such that b Fb m . Now dene F :=
bB

Fb

Then as B generates M and any b B is generated by Fb F it is clear, that F also is a generating set of the entire module M : F
m

=
bB

Fb

bB

Rb = lhR (B ) = M

Since any Fb has been nite, but B is innite the cardinality of F is precisely the cardinality of B . But as F G this already yields the converse inequality |B | = |F | |G| (iii) Choose some generating set X M of M with |X | = rank(M ). Further, as F is free, we can choose some basis B F of F . Let us rst suppose X is a nite set, say X = { x1 , . . . , xn }. The basis B M is an R-linearly independent set, and hence B cannot contain more than n elements, as we have seen in (i). Hence by (ii) we have rank(F ) = |B | n = |X | = rank(M ) 408 17 Proofs - Rings and Modules

So let us now consider the case that X is innite - this will require some slight tricks of cardinal arithmetic: As X generates M = lh(X ) for any y B there is a nite set (y ) X such that y lh( (y )). Hence there is a map from B to the set (X ) := { Z X | #Z < } of nite subsets of X , given by : B (X ) : y (y ) But as X is innite we have |X | = |(X )|. Now consider any nite subset Z (X ), say Z = { z1 , . . . , zn } X . Then the bre of Z is given to be 1 (Z ) = { y B | (y ) = Z }. Thereby (y ) = Z yields y lh( (y )) = lh(Z ). Note that since 1 (Z ) B is contained in the basis B of F it is R-linearly independent. And as lh(Z ) is generated by at most n elements (i) implies, that 1 (Z ) cannot contain more that n elements: # 1 (Z ) #Z . Thus any bre of is nite. This now means, that the cardinality of B cannot exceed the cardinality of (X ). And altogether that is rank(F ) = |B | |(X )| = |X | = rank(M ) P Proof of (3.60): If : M =m N is an isomorphism of R-modules, then in particular : M N is an epimorphism. Hence by (3.59) we nd that rank(N ) rank(M ). But as the inverse 1 : N M is surjective as well, we also have rank(M ) rank(N ). Altogether we have found rank(M ) = rank(N ) Conversely suppose that rank(M ) = rank(N ). As both M and N are free we may choose bases A M of M and B N of N . As R is commutative and non-zero we may invoke (3.62.(ii)) to see |A| = rank(M ) = rank(N ) = |B | That is there is some bijection : A B N . As A is a basis of M we may now expand this map linearly. That is we use (3.45.(ii)) to dene an R-module homorphism :M N :
xA

ax x
x A

ax (x) =
y B

a1 (y) y

Likewise, as B is a basis as well, we can linearly expand 1 : B A M , as well. Then it is immediately clear, that = 1 1M and = 1 1N . That is : M N is an isomorphism with inverse 1 = . In particular we have : M =m N as we had claimed. P Proof of (3.65): 409

(i) Of course the proof will be an application of the lemma of Zorn. So let us regard the collection of all R-lineary independent subsets L M containing L, formally that is L := L M | L is R-lineary independent, L L

It is clear that L L and hence L = is non-empty. Also L is partially ordered under the inclusion . Now consider some chain (Li ) L (where i I ) in L and let L :=
iI

Li

In order to establish L L we have to show that L is R-linearly independent again. Thus consider any xk L and any ak R (where k 1 . . . n) such that a1 x1 + + an xn = 0 By construction of L for any i 1 . . . k there is some i(k ) I such that xk Li(k) . Since the Li form a chain we can assume (by renumbering the xk ), that Li(1) Li(2) . . . Li(n) In particular xk Li(n) for any k 1 . . . n. But as Li(n) is linearly independent (it is contained in L) the above equality implies a1 = = an = 0. Hence L is linearly independent, as well, such that L L. Hence every chain in L has an upper bound so that we may invoke the lemma of Zorn, to nd a maximal element of L. (ii) If x L, then x lh(L) is clear, such that we may take a = 1. Also if there is some a = 0 with ax = 0, then there is nothing to prove. So from now on we assume x L and ax = 0 = a = 0 for any a R. Then, as L is a maximal linear indepenedent set, L { x } is no longer linear independent. That is there are some y1 , . . . , yn L and some a, b1 , . . . , bn R that are not identically zero, such that ax + b1 y1 + + bn yn = 0 Thereby ax = 0 cannot be, as else { y1 , . . . , yn } L would be linearly dependent. And by assumption this already implies a = 0. Now we see, that ax = (b1 y1 + + bn yn ) lh(L) holds true as well. (iii) If L is a linearly independent set of M , then clearly F := lh(L) is generated by L (by construction) and L is linearly independent (by assumption). That is L is a basis of F and F is free. If now R = 0 is a non-zero commutative ring, then lemma (3.62.(iii)) yields |L| = rank(F ) rank(M ) (iv) We rst prove that M/F where F = lh(L) is a torsion module. Hence we consider any z = x+F M/F . As L is maximal linear independent 410 17 Proofs - Rings and Modules

(ii) yields some a R with a = 0 such that ax F . Hence az = ax + F = 0 + F = 0 M F And if F m M is a free submodule of M with basis B , then B M clearly is linearly independent. It remains to prove the maximality of B . Thus suppose B X . If there is some x X with x B , then (as M/F is a torsion module) we can choose some a R with a = 0 such that a(x + F ) = 0 + F . As above that is ax F = lh(L). That is there are some b1 , . . . , bn R and y1 , . . . , yn B such that ax = b1 y1 + + bn yn Thus b1 y1 + + bn yn ax = 0 is a linear relation with a = 0 within the set X . That is X is not linearly independent and hence B M is maximally linear indepenent. (v) We will rst prove that |L| = |L |. To do this we regard the map : L L : x ax x. By denition of L it is clear that is surjective. Thus it remains to show the injectivity: suppose there were some x, y L such that x = y and ax x = ay y . Then ax x ay y = 0 As { x, y } L is linear independent, this would imply ax = ay = 0 in contradiction to the choice of ax and ay . Hence also is injective, that is we have established the bijecton : L L . It remains to prove the linear independence of L . Suppose we are given some x(k ) L and b(k ) R (where k 1 . . . n) such that 0 = b(1)a(1)x(1) + + b(n)a(n)x(n) Hereby we denoted a(k ) := ax(k) R. As { x(1), . . . , x(n) } L is linear independent this implies b(k )a(k ) = 0 for any k 1 . . . n. But as R is an integral domain and a(k ) = 0 this already is b(k ) = 0 for any k 1 . . . n. That is L also is linearly independent. (vi) As L is a maximal linear independent set, we have seen in (ii) that for any x K there is some ax R with ax = 0 and ax x Q := lh(L). Let us now denote K := { ax x | x K } and P := lh(K ). By (iv) we know, that K is linearly independent again and hence P is a free submodule of M with basis K . In fact, as K Q we even have P Q. Now by (3.62) again we nd |K | = |K | = rank(P ) rank(Q) = |L| Reversing the roles of K and L in the above argument we also see that |L| |K | and hence |K | = |L|. (vii) According to (i) we may pick up maximal linear independent sets K P and { y + P | y L } M/P for some sucient L M . First of all it is clear, that K L = , because if we had x K L 411

then x K P and hence x + P = 0 + P such that { x + P } would be linearly dependent in contrast to x L. Hence we have |K L| = |K | + |L|. It remains to prove that K L is a maximal linear independent set in M , as we then have free(M ) = |K L| = |K | + |L| = free(P ) + free M P For the linear independence of K L suppose we have some xi K (where i 1 . . . m) and yj L (where j 1 . . . n) such that z := a1 x1 + + am xm + b1 y1 + + bn yn = 0 for some ai and bj R. As xi K P we then nd that 0 + P = z+P = j bj yj + P = j bj (yj + P ). But as yj L we know that the yj + P are linearly independent and hence necessarily b1 = = bn = 0. That is 0 = z = i ai xi , but as the xi are linearly independent as well, this implies a1 = = am = 0 once more. Hence there only are trivial linear relations in K L, that is K L is linearly independent. It remains to prove the maximality of K L. Hence suppose we are given any z M with z K L. As z +P M/P and { y + P | y L } is a maximal linear independent set (ii) implies that there is some c R with c = 0 such that cz + P lh(y + P | y L). That is there are some yj L and bj R (where j 1 . . . n again) such that cz + P = j bj (yj + P ) = j bj yj + P . That is cz j bj yj P . And as K M is a maximal linear set as well (ii) once more implies that there is some d R with d = 0 such that d(cz j bj yj ) lh(K ). Altogether that is d cz
j =1 n

bj yj =

ai xi
i=1

for some xi K and ai R (where i 1 . . . m again). Rearranging this we found dcz ( i ai xi + j dbj yj ) = 0. Thereby dc = 0, as c = 0 and d = 0 and R is an integral domain. That is we have found a non-trivial linear relation in K L { z } which means that this set is linearly dependent. So at last K L is maximal linear independent. (viii) P

412

17 Proofs - Rings and Modules

Chapter 18

Proofs - Commutative Algebra

Proof of (2.28): The implication (a) = (b) is quite easy to see: given an ascending chain (xk ) X (where k N), that is x0 x1 x2 . . . , let us take A := { xk | k N }. By assumption A has a maximal element a A. Pick up some s N such that a = xs . If now i N, then (as (xk ) is an ascending chain) xs xs+i and due to the maximality of xs this already is xs+i = xs . The implication (b) = (a) will be proved as (a) = (b), which is only slightly more dicult: as (a) is untrue there is some non-empty subset A X without maximal element. Negating the statement that A contains a maximal element yields: a A b A : a < b That is (by the axiom of choice) we may pick up a map : A A such that (a) := b for some b A with a < b. And as A is non-empty we may start with some x0 A. Now recursively dene xk+1 := (xk ) = k (x0 ). Then by construction we have found an innitely ascending chain of elements x0 < x1 < x2 < . . . which is the negation of (a). P Proof of (2.27): (a) = (b): we have the assumption of (ACC) and need to show, that there is some maximal element a A . That is there has to be some a A such that (a a) = (a = a) for any a A. Yet this statement is logically equivalent, to saying a A a A : (a a) (a = a) Suppose this was not the case - that is the negation of this statement was true. This is a A a A such that (a a) (a = a). And clearly this means nothing but a A a A : a a 413

Thus we could start by picking up any one a1 A as a . And by the above property there would be some a2 A such that a1 a2 . Continuing this way (using induction with a := ak and ak+1 := a) we could construct an innitely ascending chain if ideals a1 a2 . . . ak ak+1 . . . in contradiction to (ACC). Hence the assumption has to be false and this means, that (b) holds true. (b) = (c): Consider any ideal b i R. Then we denote the set of all ideals generated by a nite subset B b by A, that is A := B
i

| B b, #B <

As 0 b we have { 0 } A, in particular A = is non-empty. Hence by assumption (b) - there is a maximal element a A of A. And as a A there is some nite set B = { a1 , . . . , ak } b generating a . In particular a b. Now suppose b = a , that is there is some b b such that b a . Then we would also have
a

a1 , . . . , ak , b

in contradiction to the maximality of a . Thus a equals b and in particular we conclude that b = a = a1 , . . . , ak i is nitely generated. (c) = (a): consider an ascending chain a0 a1 . . . ak . . . of ideals in R. Then we dene the set
b :=
k N

ak

As the ak have been a chain, we nd that b i R is an ideal due to (2.4.(i)). And hence by assumption (c) there are nitely many elements a1 , . . . , an R generating b = a1 , . . . , an i . So by construction of b for any i 1 . . . n (as ai b) there is some s(i) N such that ai as(i) . Now let s := max{ s(1), . . . , s(n) }. Then for any i 1 . . . n and any t s we get ai as(i) as at . In particular
b = a1 , . . . , an
i

at b

Of course this means b = at for any t s. And this is just another way of saying that the chain of the ak stabilized at level s. By now we have proved the equivalence of (a), (b) and (c) in the denition of noetherian rings. And trivially we also have (c) = (d). These are the important equivalencies, it merely is a neat fact that (d) also implies (c). Though we will not use it we wish to present a proof (note that this is substantially more complicated than the important implications above). We will now prove (d) = (c) in several steps: we want to prove that any ideal of R is nitely generated. Thus let us take the set of all those ideals of R not being nitely generated: Z := { a i R | a is not nitely generated } The idea of the proof will then be the following: suppose Z = , then 414 18 Proofs - Commutative Algebra

by an application of Zorns lemma, there is some maxmal element p Z . In a second step we will then prove that any p Z is prime. So by assumption (d) p is nitely generated. But because of p Z it also is not nitely generated. This can only mean Z = and hence any ideal of R is nitely generated. Thus it remains to prove: Z = = Z = : thus consider a chain (ai ) (where i I ) in Z . Then we denote the union of this chain by
b =
i I

ai

By (2.4.(i)) b i R is an ideal of R again. Now suppose b Z , then b would be nitely generated, say b = b, . . . , bn i . By construction of b for any k 1 . . . n (as bk b) there is some i(k ) I such that bk ai(k) . And as the ai form a chain we may choose i I such that ai := max{ ai(1) , . . . , ai(n) }. Thereby b = b1 , . . . , bn i ai b. That is ai = b1 , . . . , bn i would be nitely generated in contradiction to ai Z . Thus we have b Z again and hence b is an upper bound of (ai ). Now - as any chain has an upper bound, by the lemma of Zorn - there is some maximal element p Z . p Z = p prime: First note thet p = R, as R = 1R is nitely generated. Now suppose p was not prime, then there would be some f , g R such that f p, g p but f g p. Now let a := p + f R, as f p we nd p a. And as p is a maximal element of Z this means that a is nitely generated, say a = a1 , . . . , ak i . As the ai a = p + f R there are some pi p and bi R such that ai = pi + f bi . Therefore
a = p1 , . . . , pk , f
i

= p1 R + + pk R + f R

The inclusion is clear, as pi p a and f a. Conversely consider x a, that is there are some xi R such that x = i xi ai = i xi pi + f i xi bi p1 R + + pk R + f R. Next we note that
p = f (p : f ) + p1 R + + pk R

where f (p : f ) = (f R) (p : f ) = { f b | b p : f } i R. The inclusion is easy: pi p is true by denition. Thus consider any element x f (p : f ). That is x = f b for some b p : f = { b R | f b p }. In particular x = bf p, too. For the converse inclusion we are given any q p a. That is there are some xi R and y R such that q = x1 p1 + + xk pk + yf . As q and all the pi are contained in p this implies yf p and thereby y p : f . Thus yf f (p : f ) and hence q p1 R + + pk R + f (p : f ). Finally we prove
p:a Z

Clearly we get p p : a and b p : a (because of ab p). But b p and hence p p : a. But by the maximality of p in Z this means p : a Z . Thus p : a is nitely generated and hence a(p : a) is nitely generated, too. But as the sum of nitely generated ideals is nitely 415

generated by (1.54) this means that p = a(p : a) + p1 R + + pk is nitely generated, too, in contradiction to p Z . Thus the assumption of p not being prime is false. P Proof of (2.31): (i) We will only prove the noetherian case - the arinian case is an analogous argument involving descending (instead of ascending) chains of ideals. Thus consider an ascending chain of ideals in R/a
u0 u1 . . . uk . . . i R a

By the correspondence theorem (1.61) the ideals uk are of the form uk = bk /a for sucient ideals bk := 1 (uk ) i R. Now suppose b bk then b + a uk uk+1 = bk+1 /a and hence b bk+1 again. Thus we have found an ascending chain of ideals in R
b0 b1 . . . bk . . . i R

As R is noetherian this chain has to be eventually constant, that is there is some s N such that for any i N we get bs+i = bs . But this clearly implies us+i = us as well and hence R/a is noetherian again. (ii) If : R S is a surjective homomorphism, then by the rst isomorphism theorem (1.77) we get R/kn() =r S . But by assumption R is and hence R/kn() is , too, due to (i). And because of this isomorphy we nd that S thereby is , as well (this is clear by transfering a chain of ideals from S to R/kn() and returning to S ). (iii) First suppose that R S is . Trivially we have the following surjective ring homomorphisms R S R : (a, b) a and R S R : (a, b) b. Thus both R and S are due to (ii). Conversely suppose both R and S are noetherian. We will prove that this implies R S to be noetherian again (the proof in the artinian case is completely analogous). Thus consider an ascending chain of ideals of R S
u0 u1 . . . uk . . . i R S

Clearly R 0 and 0 S i R S are ideals of R S , too. Thus we obtain ideals ak := uk (R 0) and bk := uk (0 S ) i R S be intersection. And for these we get
uk = ak + bk

The inclusion is clear, as ak , bk uk . For the converse inclusion we are given some (a, b) uk . As uk is an ideal we nd (a, 0) = (1, 0)(a, b) and (0, b) = (0, 1)(a, b) uk . And as also (a, 0) R 0 and (0, b) 0 S this yields (a, b) = (a, 0) + (0, b) ak + bk . But as ak is an ideal of R S contained in R 0 it in particular is an ideal of ak i R 0. But clearly R and R 0 are isomorphic under R =r R 0 : a (a, 0). Thus ak corresponds to the following 416 18 Proofs - Commutative Algebra

ideal a k := { a R | (a, 0) ak } i R. And thus we have found an ascending chain of ideals in R

a 0 a1 . . . ak . . . i R

Yet as R was assumed to be noetherian there is some p N such that for any i N we get a p+i = ap . And returning to R 0 (via ak = { (a, 0) | a a k }) we nd that ap+i = ap as well. With the same argument for the bk we nd some q N such that for any i N we get bq+i = bq . Now let s := max{ p, q } then it is clear that for any i N we get
us+i = as+i + bs+i = as + bs = us

that is the chain of the uk has been eventually constant. And this means nothin but R S being noetherian again. P Proof of (2.33): This proof requires some working knowledge of polynomials as it is presented in sections 7.3 and 7.4. Yet as the techniques used in this proof are fairly easy to see we chose to place the proof here already. In case you encounter problems with the arguements herein refer to these sections rst. (1) In a rst step let us consider the case where S = R[t] is the polynomial ring over R. Thus consider an ideal u i S , then we want to verify that u is nitely generated. The case u = 0 is clear, thus we assume u = 0. Then for any k N we denote
ak := { lc (f ) | f u, deg(f ) = k } { 0 }

where lc (f ) := f [deg(f )] denotes the leading coecient of f . We will rst prove that ak i R is an ideal of R: 0 ak is clear by construction. Thus suppose a, b ak say a = lc (f ) and b = lc (g ). As both f and g are of degree k we get (f + g )[k ] = f [k ] + g [k ] = lc (f ) + lc (g ) = a + b. Thus if a + b = 0 then a + b ak is clear. And else, if a + b = 0 then f + g is of degree k , too and lc (f + g ) = (f + g )[k ] = a + b. And as f + g u again we nd a + b ak in both cases. Now let r R be any element, if ra = 0 then ra ak is trivial again. And else rf : k rf [k ] is of degree k again and satises lc (rf ) = (rf )[k ] = r(f [k ]) = rlc (f ) = ra. And as also rf u we again found ra ak in both cases. Altogether ak i R is an ideal. Next we will also prove the containment ak ak+1 . That is we consider some a = lc (f ) ak again. As f u we also have tf : k f [k 1] u. Obviously deg(tf ) = deg(f ) + 1 = k + 1 and lc (tf ) = f [k + 1] = f [k ] = a. Thus we have found a = lc (tf ) ak+1 . Altogether we have found the following ascending chain of ideals of R
a0 a1 . . . ak . . . i R

As R is noetherian this chain has to be eventually constant, that is there is some s N such that for any i N we get as+i = as . And 417

furthermore any ak is nitely generated (as R is noetherian). As we may choose generators of ak we do so

ak =

ak,1 , . . . , ak,n(k)

and pick up polynomials fk,i u of degree deg(fk,i ) = k such that lc (fk,i ) = ak,i . Then we dene the following ideal of S
w :=

fk,1 | k 0 . . . s, i 1 . . . n(k )

For the rst step it remains to prove that u = scriptw. Then u is nitely generated, as we have given a list of generators explictly. The inclusion w u is clear, as any fi,k u. For the converse inclusion we start with some f u and need to show f w. This will be done by induction on the degree k := deg(f ) of f . The case f = 0 is clear. If deg(f ) = k = 0 then f R is constant, that is f = lc (f ) a0 = a0,1 , . . . , a0,n(0) i w. Thus for the induction step we suppose k 1 and let a := lc (f ) = f [k ]. If k s then a ak = ak,1 , . . . , ak,n(k) i . That is there are some bi R such that a = i ak,i bi . Now let us dene the polynomial
n(k)

g :=
i=1

bi fk,i w

Then it is clear that lc (g ) = i bi lc (fi,k ) = a = lc (f ) and hence deg(f g ) < k . Thus by the induction hypothesis we get d := f g w again and hence f = d + g w, too. It remains to check the case k > s. Yet this can be dealt with using a similar argument. As k s we have a ak = as . That is a = i ak,i bi for sucient bi R. This time
n(s)

g :=
i=1

bi tks fk,i w

Then lc (g ) = i bi lc (fi,k ) = a = lc (f ) again and as before this is deg(f g ) < k . By induction hypothesis again d := f g w and hence f = d + g w. Thus we have nished the case S = R[t]. (2) As a second case let us regard S = R[t1 , . . . , n] the polynomial ring in nitely many variables. Again we want to prove that S in noetherian, and this time we use induction on the number n of variables. The case n = 1 has just been treated in (1). Thus consider R[t1 , . . . , tn , tn+1 ]. Then we use the isomorphy R[t1 , . . . , tn , tn+1 ] =r R[t1 , . . . , tn ][tn+1 ]

n+1 =i

n i 1 f []t tn+1 1 . . . tn

f []t
i=0

By induction hypothesis R[t1 , . . . , tn ] is noetherian already. And by (1) this implies R[t1 , . . . , tn ][tn+1 ] to be noetherian, too. And using the 418 18 Proofs - Commutative Algebra

above isomorphy we nally nd that R[t1 , . . . , tn , tn+1 ] is noetherian. (3) We now easily derive the the general claim: let S be nitely generated over R. That is there are nitely many e1 , . . . , en S such that S = R[e1 , . . . , en ]. Then we trivially have a surjective homomoprphism R[t1 , . . . , tn ] S : f f (e1 , . . . , en ) form the polynomial ring onto S . And by (2.31.(iii)) this implies that S is noetherian, too. P Proof of (2.38): Let us denote the prime ring of R by P := r . This is the image of the uniquely determined ring-homomorphism from Z to R (induced by 1 1). That is P = im(Z R) and in particular P is noetherian by (2.31.(ii)). Let us now denote the set of nite, non-empty subsets of R by I I := { i R | i = , #i < } And for any i I let us denote Ri := P [i] the P -subalgebra of R generated by i. Then by the Hilberts basis theorem Ri is noetherian as well. Given any a R we clearly have i := {a} I and a Ri . In particular the Ri cover R. And given any two i, j I we have k := i j I , too. We now claim that Ri Rj Ri : thus suppose i = {i1 , . . . , in } and let a = f (i1 , . . . , in ) for some polynomial f P [t1 , . . . , tn ]. Then it is clear that a = g (i1 , . . . , in , j1 , . . . , jm ) P [i j ] = Rk if we only let g (t1 , . . . , tn , tn+1 , . . . , tn+m ) := f (t1 , . . . , tn ) P [t1 , . . . , tn+m ]. Thus we have seen Ri Rk and Rj Rk can be proved in complete analogy. P Proof of (2.36): (i) This has been proved in (2.27) as implication (a) = (b) already. The reader is asked to refer to page 413 for the proof. (iii) Just let P := b spec R | p b q . Then P = is non-empty, as p P . But as we have seen (i) this already implies that there is some maximal element q P . (iv) For any ideal R = a i R let us denote V(a) := { p spec R | a p } and n(a) := #V(a) N {}. That is n(a) is the number of minimal prime ideals lying over a. By (2.14.(iii)) we know 1 n(a). Now let A := { a i R | n(a) = } be the set of all ideals having innitely many minimal primes lying over it. We want to show that A = . Hence assume A = , then - by (i), as R is noetherian - we may choose some maximal ideal u A in A. Clearly u cannot be prime: if it was then u would be contained in precisely one minimal prime - namely u itself, that is V(u) = {u}. In particular we had n(u) = 1 < which contradicts u A. Now, as u 419

is not prime, there are some a, b R such that ab u but a, b u. Then we dene the respective ideals
a := u + aR and b := u + bR

As a u we have u a and hence a A due to the maximality of u. Likewise b A and this means that n(a), n(b) < are nite. But we will now prove V(u) V(a) V(b) To see this let p V(u) be a minimal prime ideal over u. If a p then a p. Else suppose a p then ab u p implies b p, as p is prime. And thereby we get b p. In any case p lies over one of the ideals a or b, say a p (w.l.o.g.). If now q V(a) then in particular we get q V(u). Thus if q p then the minimality of p over u implies q = p. That is p is a minimal element of V(a) and hence the inclusion. As both V(a) and V(b) < are nite, V(u) < would have to be nite, too. But this contradicts u A again, so the assumption A = has to be abandoned. That is any ideal a of R only has nitely many prime ideals lying over it. (ii) Consider a set of prime ideals P = of the noetherian ring R. It has already been shown in (2.27) that P = contains maximal elements. We will now prove that P = contains a minimal element, as well. Suppose this was untrue, then we start with any p = p0 P . As p0 is not minimal in P , there is some p1 such that p0 p1 . Again p1 is not minimal in P , such that we may choose some p2 P such that p1 p2 . Continueing that way we nd an innite, strictly descending chain of prime ideals
p = p0 p1 p2 . . . pk . . .

This means that p is a prime ideal of innite height height(p) = . But in a noetherian ring we have Krulls Principal Ideal Theorem (??). And this states that height(p) rank(p) < . Hereby the rank is nite, as any ideal p in a noetherian ring is nitely generated. P Proof of (2.39): We start with the equivalencies for a | b. Hereby (a) = (c) is clear: b aR = { ah | h R } yields b = ah for some h R. (c) = (b): consider f bR, that is f = bg for some g R. But as b = ah this yields f = ahg aR. (b) = (a): as b bR aR we have b aR. We also have to check the equivalencies for a b - recall that now R is an integral domain. (a) = (b): if aR = bR then in particular aR bR and hence b | a. Likewise we get a | b. (b) = (c): Suppose b = a and a = b. If b = 0 then a = b = 0 = 0, as well. And hence we may choose := 1. Thus we continue with b = 0. Then b = a = b. And as R is an integral domain and b = 0 we may divide by b and thereby obtain 1 = . That is R and b = a as claimed. (c) = (a): As b = a we get bR aR. And from a = 1 b we get aR bR. 420 18 Proofs - Commutative Algebra

P Proof of (2.40): (i) By denition of 1 we have 1 b = b and hence 1 | b, likewise we have a 0 = 0 and hence a | 0 by (1.37). Now ah = ab for h := b and hence a | ab is trivial again. (i) Next suppose b = 0, then b = 0 1 and hence 0 | b. Conversely suppose 0 | b, that is there is some h R such that 0 = 0 h = b. Then we have just proved 0 = b, as well. If a R then 1 = aa1 and hence a | 1. Conversely suppose a | 1, that is there is some h R such that ah = 1. As R is commutative this also is ha = 1 and hence a R . Finally let a | b, that is ah = b for some h R again. In particular this yields (ac)h = (ah)c = bc and hence ac | bc. (ii) Now suppose u R is a non-zero divisor, by (i) we only have to show the implication = . Thus let (au)h = bu for some h R, then (ah b)u = (au)h bu = 0. But as u is a non-zero divisor this can only happen if ah b = 0 and this already is a | b. (iii) First of all a | a is clear by a 1 = a. If now a | b and b | c (say ag = b and bh = c) then we get a | c from a(gh) = (ag )h = bh = c. And that a | b and b | a implies a b is true by denition (2.39). (iv) As we have seen in (2.39) a b is equivalent, to aR = bR. And from this it is clear that is reexive, symmetric and transitive (i.e. an equivalence relation). Thus it only remains to observe that [a] = { b R | a b } = { a | R } = aR P Proof of (2.48): (i) The statement is clear by induction on the number k of elements involved: if k = 1 then p | a1 is just the assumption. Thus consider k + 1 elements and let a := a1 . . . ak and b := ak+1 . Then by assumption we have p | ab and (as p is prime) hence p | a or p | b. If p | b then we are done (with i = k + 1). And if p | a then by induction hypothesis we nd some i 1 . . . k such that p | ai . Altogether p | ai for some i 1 . . . k + 1. (ii) If p R is prime then p = 0 by denition. And as also p R we have pR = R, as well. Now ab pR translates into p | ab and hence p | a or p | b. But this again is a pR or b pR. That is we have proved that pR is prime. Conversely suppose p = 0 and that pR is a prime ideal. As pR = R we have p R . And as we also assumed p = 0 this means p R . If we are now given a, b R with p | ab, then ab pR and hence a pR or b pR, as pR is prime. Thus we nd p | a or p | b again, and altogether p is prime. 421

(iii) p irreducible = p irreducible: suppose p = ab for some a, b R. As is a unit this yields p = (1 a)b and as p is irreducible hence 1 a R or b R . But 1 a R already implies a R (with inverse a1 = 1 (1 a)1 ). Conversely if p is irreducible then p = 1 p is irreducible, too by what we have just proved. (iii) p prime = p prime: suppose p | ab for some a, b R. As is a unit this yields p | (1 a)b and as p is prime we hence get p | 1 a or p | b. From p | 1 a we get p | a and p | b implies p | 1 p which also implies p | b. Thus we have proved, that p is prime too. Conversely if p is prime then p = 1 p is prime, too by what we have just proved. (iv) Let f := ab, then it is clear that a | f and hence f R aR. And as both a and b = 0 are non-zero, so is f (as R is an integral domain). Now suppose we had a f R, that is a = f g for some g R. Then we compute a = f g = abg . And as a = 0 this means 1 = bg . In particular b R is a unit, in contradiction to b R . Hence a aR \ f R and this yields the claim f R aR. (v) Now let R be an integral domain, p R be prime and suppose p = ab for some a, b R. In particular p | ab and hence p | a or p | b. W.l.o.g. let us assume p | b, that ph = b for some h R. Then p = ab = aph and hence p(1 ah) = 0. Now as p = 0 this implies ah = 1, that is a R is a unit or R. Thereby we have veried, that p truly is irreducible. (vi) Clearly we have R D, as k = 0 is allowed. And if c = p1 . . . pk and d = q1 . . . ql D it is clear that cd = ( )(p1 . . . pk q1 . . . ql ) D again. Thus it remains to show the implication cd D = c D (due to the symmetry of the statement it also follows that d D). Thus consider any c, d R such that cd = p1 . . . pk D. We will prove the claim by induction on k - that is we will verify c, d R : cd = p1 . . . pk D = cD

for any k N. The case k = 0 is trivial, since cd = R implies c R D. Thus regard k 1. We let I := {i 1 . . . k | pi | d}. (1) If there is some i I then pi | b. W.l.o.g. we may assume i = k . That is there is some h R such that d = hpk . Then p1 . . . pk = cd = chpk As R is an integral domain and pk = 0 we may divide by pk to nd ch = p1 . . . pk1 D. By induction hypothesis that is c D already. (2) If I = then for any i 1 . . . k weve got pi | d. But as pk is prime and pk | cd this yields pk | c, say c = gpk . Then p1 . . . pk = cd = gdpk We divide by pk again to nd gd = p1 . . . pk1 D. And by induction hypothesis this means g D. But pk D is clear, as well and therefore c = gpk D as claimed. 422 18 Proofs - Commutative Algebra

(vii) First consider the case = 1 = . Without loss of generality we may assume k l and we will use induction k to prove the statement. If k = 0 we have to prove l = 0. Thus assume l 1 then 1 = q1 . . . ql = q1 (q2 . . . ql ). That is q1 R is a unit. But q1 has been prime and hence q1 R a contradiction. Thus we necessarily have k = l = 0. Now consider the induction step k 1, then pk | p1 . . . pk = q1 . . . ql and as pk is prime this means pk | qj for some j 1 . . . l. Without loss of generality we may assume j = l. Hence we found ql = pk for some R. But ql has been prime and hence is irreducible by (iv). This means R (as pk R ) and hence pk qk . Further we have (p1 . . . pk1 )pk = p1 . . . pk = q1 . . . ql = (q1 . . . ql1 )pk As pk = 0 and R is an integral domain we may divide by pk and thereby get p1 . . . pk1 = (q1 )q2 . . . ql1 . (As q1 is prime again) the induction hypothesis now yields that k 1 = l 1 and there is some Sk1 such that for any i 1 . . . k 1 we have pi q(i) . But this also is k = l and we may extend to Sk by letting (k ) := k = l. (As q1 q1 ) we hence found pi q(i) for any i 1 . . . k . Thus we have completed the case = 1 = . If now and R are arbitrary, then we let p1 := p1 , pi := pi (for i 2 . . . k ) and q1 := q1 , qj := qj (for j 2 . . . l). By (iii) pi and qj are prime again and by construciton it is clear that pi pi and qj qj for any i 1 . . . k and j 1 . . . l. As p1 . . . pk = q1 . . . ql we may invoke the rst case to nd k = l and the required permutation Sk . But now for any i 1 . . . k we get pi p1 q(i) q(i) and hence pi q(i) as claimed. (viii) We only prove the equivalence for the least common multiple (the like statement for the greatest common divisor can be proved in complete analogy). Clearly m mR = lcm(A) implies m lcm(A). Thus let us assume m lcm(A), then we need to prove lcm(A) = mR . For we are given any R . As A | m it is clear that A | m, as well. And if A | n then m | n (by assumption) and hence also m | n (by (m)(1 h) = n for mh = n). Concersely for consider n lcm(A). As A | m and n lcm(A) we get n | m. Analogously A | n and m lcm(A) impliy m | n. Together that is m n and hence n = n for some R . P Proof of (2.49): (i) Suppose a | b = that is for any k N there is some hk R such that ak hk = b. First suppose there was some k N such that hk = 0. Then b = ak 0 = 0, too and then a | 0 also implies a = 0, a contradiction. Thus hk = 0 for any k N. Now for any k N we nd hk+1 ak+1 = b = hk ak . And as a = 0 and R is an integral domain hk+1 ak+1 = hk ak also yields hk+1 a = hk . That is hk+1 | hk and hence we have found the ascending chain of ideals h0 R h1 R . . . hk R hk+1 R . . . 423

Now suppose we had hk R = hk+1 R at any one stage k N. That is there is some R such that hk+1 = hk . Thereby we compute hk = ahk+1 = ahk . And as hk = 0 this implies a = 1. That is a R a contradiction to a R . And this only leaves a | b < . (ii) As p is prime we have p R and hence p | a < due to (i) (likewise for b and ab). That is we let k := p | a , l := p | b and n := p | ab N. Then by assumption there are some , R such that a = pk and b = pl . In particular ab = pk+1 such that pk+l | ab and hence k + l n, as n is maximal among the integers with pn | ab. Now suppose n > k + l, say pn h = ab = pk+l . As R is an integral domain we may divide by pk+l and thereby obtain pn(k+l) h = . And as n (k + l) 1 this yields p | . Now recall that p has been prime, then we may assome (w.l.o.g.) that p | , say pg = . Then a = pk = gpk+1 and hence pk+1 | a. But this is a contradiction to k being maximal and hence the assumption n > k + l has to be dropped. Altogether we have found n = k + l. () In a rst step let us prove, that for any a R there is an irreducible p R such that p | a. To do this we recursively dene a sequence of elements, starting with a0 := a. Now suppose we had already constructed a0 , . . . , ak R such that a0 R a1 R . . . ak R. Of course ak = 0 cannot be, as 0 = a = a0 ak R. Then we iterate as follows: if ak is not irreducible then there are f , g R such that ak = f g . Then by (2.48.(iv)) we get ak R = (f g )R f R. Thus if we now let ak+1 := f then we have appended our chain of ideals to a0 R a1 R . . . ak R ak+1 R As R is noetherian we cannot construct a strictly ascending chain of innite length. That is our construction has to terminate at some nite level s N. And this can only be if p := as is irreducible. Thus we have a = a0 a0 R as R = pR and thereby p | a. (iii) If a R is a unit, then the claim is clear by choosing := a and k := 0. Thus let us assume a R . Then by () above there is some irreducible element p1 R such that p1 | a, say p1 a1 = a. Thus suppose we have already constructed the elements a1 , . . . , ak R and p1 , . . . , pk R such that the pi are irreducible, a = ak p1 . . . pk and aR a1 R . . . ak R. Suppose that ak R is no unit, then we iterate as follows: Note that ak = 0 as well (else a = 0) and hence ak R . Thus by () there is some irreducible element pk+1 R such that pk+1 | ak , say ak = pk+1 ak+1 . We note that by (2.48.(iv)) this implies ak R = (pk+1 ak+1 )R ak+1 R. So we have found ak+1 and pk+1 irreducible such that ak+1 p1 . . . pk+1 = ak p1 . . . pk = a and aR a1 R . . . ak R ak+1 R As R is noetherian we cannot construct a strictly ascending chain of innite length. That is our construction has to terminate at some nite level s N. And this can only be if as R is a unit. Thus we let := as and thereby get a = p1 . . . ps as claimed. 424 18 Proofs - Commutative Algebra

(iv) As R is noetherian and p i R is an ideal there are nitely many ai R such that p = a1 , . . . , ak i . Without loss of generality we may assume ai = 0 (else we might just as well drop it). Then we use (iii) to nd a decomposition of ai into irreducible elements pi,j R
n(i)

ai = i
j =1

pi,j

As ai p and p is prime this means that for any i 1 . . . k there is some j (i) 1 . . . n(i) such that pi,j p (note that i = p as else 1 1 = i i p). Now let pi := pi,j (i) p. Then it is clear that ai pi R p1 , . . . , pk
i

p = a1 , . . . , ak

i i

for any i . . . k . And thereby p = a1 , . . . , ak i p1 , . . . , pk well such that we truly nd p = p1 , . . . , pk i as claimed.

as

(vi) Let us prove the rst statement a | b by induction over k . If k = 0 then a = 1 and hence a | b is clear. Let us now x some some abbreviations: let n(i) := qi | b N due to (i). And by denition n(i) n(i) of qi | b it is clear that qi | b, say b = bi qi for any i 1 . . . k . n(1) Hence in case k = 1 we are done already, as a = q1 | b. Thus we may assume k 2. Then the induction hypothesis reads as
k 1

h :=
i=1

qi

n(i)

Assume that qk | h, as qk is prime (by construction of h) this would mean qk | qi for some i 1 . . . k 1. That is qi = i qk for some i R. But as R is an integral domain and qi is prime it also is irreducible by (2.48.(v)). This yields i R and hence qi qk . But by assumption this is i = k in contradiction to i 1 . . . k 1. Thus we have seen qk | h, or in other words qk | h = 0. Now recall that h | b, that is there is some g R such that b = gh. then we may easily compute (using (ii)) n(k ) = qk | b = qk | gh = qk | g + qk | h = qk | g That is qk = qk k | g such that g = qk f for some f R. Multiplying this identity with h we nally nd a | b from b = gh = qk
n(k) n(k) q |g n(k)

f h = af

It remains to verify the second claim in (vi). That is consider any prime element p R such that p | b = af . As p is prime this means p | a or p | f . In the rst case p | a (by construction of a) we nd some i 1 . . . k such that p | qi . That is p = qi for some R. Again we use that qi is prime and hence irreducible to deduce that R and hence p qi . In the second case p | f we are done already, as f = b/a. Thus we have proved p | (b/a) or p pi for 425

some i 1 . . . k . Now suppose both statements would be true, that is p | f and (w.l.o.g.) p | qk . That is f = pr and qk = p for some r R and R . Then we compute b = af =
n(k)+1

hqk

n(k)

1 qk r

= qk

n(k)+1

1 hr

In particular qk | b. But this clearly is a contradiction to the maximality of n(k ) = qk | b . Thus we have proved the truth of either p | (b/a) or p qi (and not both ) as claimed. (v) Suppose we had a pR for innitely many p P . In particular we may choose a countable subset (pi ) P such that a pi R for any i N. By (vi) we get
k

ak :=
i=0

pi

p i |a

a
p |a

From the construction it is clear that for fk := pk k we get ak1 fk = ak . And as pk | a we have pk | a 1 such that fk = 1. In fact fk R as pk R and R is an integral domain. Next let us choose some gk R such that a = ak gk . Then for any k N we get ak gk = a = ak+1 gk+1 = ak fk+1 gk+1 . It is clear that ak = 0, as else a = 0 gk = 0. Thus we may divide ak from ak gk = ak fk+1 gk+1 and thereby nd gk = fk+1 gk+1 . Now using (2.48.(iv)) with gk+1 = 0 and fk+1 R we nd gk R gk+1 R. As this is true for any k N we have found an innitely ascending chain g0 R g1 R g2 R . . . of ideals in R. But this is a contradiction to R being noetherian. Thus the assumption has to be false and that is: there are only nitely many p P with a pR. P Proof of (2.50): In a rst step we will prove the equivalence of (a), (b) and (c). Using these as a denition we will prove some more porperties of UFDs rst, before we continue with the proof of the equivalencies of (d), (e) and (f) on page (469). (a) = (c): By (2.48.(v)) we already know that prime elements are irreducible (as R is an integral domain). From this (1) and the one part of (2) are trivial by assumption (a). We only have to prove the second part of (2): q R : q irreducible = q prime. By assumption (a) q admits a prime decomposition q = p1 . . . pk . It is clear that k = 0 as else q R . Now suppose k 2 then we would write q as q = (q1 )(q2 . . . qk ). And as q is irreducible this would mean pi R or p2 . . . pk R . But as p1 is prime, so is p1 - by (2.48.(iii)) - and as R is an integral domain R is closed under multiplication such that p2 . . . pk R . Hence none of this can be and this only leaves k = 1. That is q = p1 and hence q is prime. 426 18 Proofs - Commutative Algebra

 (c) = (b): consider any 0 = a R, property (1) in (b) and (c) are identical so we only need to verify (2) in (b). Thus let (, p1 , . . . , pk ) and (, q1 , . . . , ql ) be two irreducible decompositions of a. By assumption (2) in (c) all the pi and pj are prime already. Therefore p1 . . . pk = a = q1 . . . ql implies k = l and pi = q(i) (for some Sk and any i 1 . . . k ) by (2.48.(vii)). And by denition this already is (, p1 , . . . , pk ) (, q1 , . . . , ql ). (b) = (a): By assumption (b) any 0 = a R has an irreducible decomposition. Thus it suces to show that any irreducible element p R already is prime. Thus consider an irreducible p R and arbitrary a, b R. Then we need to show p | ab = p | a or p | b. To do this we write down irreducible decompositions of a and b a = p1 . . . pk and b = pk+1 . . . pk+l As p | ab there is some h R with ab = ph. Again we use an irreducible decomposition of h = q1 . . . qm . Finally let qm+1 := p q1 . . . qm qm+1 = ph = ab = p1 . . . pk+l As all the pi and qj are irreducible assumption (2) now yields that n := m + 1 = k + l and that there is some Sn such that for any i 1 . . . n we get qi p(i) . Now let j := (m + 1), that is p = qm+1 pj . If j 1 . . . k then p pj | a and hence p | a. Likewise if j k + 1 . . . k + l then p pj | b and hence p | b. Together this means that p is truly prime. P Proof of (2.54): Until now we only have proved the equivalence of (a), (b) and (c) in the denition of an UFD. That is we understand by an UFD an integral domain satisfying (a) (equivalently (b) or (c)). And using this denition only we are able to prove the theorem given: (i) Let I := {i 1 . . . k | p pi } and n := #I . Then for any i I there is some i R such that pi = i p. Then it is immediately clear that a = bpn where :=
i I

i and b :=
j I

pj

In particular pn | a and this means n p | a . Conversely suppose pm | a for some m > n, say pm h = a = bpn . Then pmn 1 h = b and hence p | b as p dim pmn . By denition of b this would imply p | pj for some j I . Say pj = p for some R. But as pj is irreducible (and p R this yields R and hence p pj in contradiction to j I . Thus we have proved the maximality of n and hence n = p | a . (ii) Using (i) this boils down to an easy computation: pick up two prime decompositions a = p1 . . . pk and b = q1 . . . ql . For j 1 . . . l we let 427

pk+j := qj . Then it is clear that ab = p1 . . . pk+l and thereby p | ab = # { i 1 . . . (k + l) | p pi } = # { i 1 . . . k | p pi } + # { j 1 . . . l | p qj } = p|a + p|b

(iv) Existence: consider any 0 = a R and pick up a prime decomposition a = q1 . . . qk of a. As the qi is prime there is a uniquely determined pi P such that qi pi , say qi = i pi for some i R . Then a = p1 . . . pk where := 1 . . . k For any p P now let n(p) := # { i 1 . . . k | pi = p } N. Then it is clear that only nitely many n(p) are non-zero, as they sum up to n(p) = k
pP

That is n p N And further it is clear that a has the representation

k

a =
i=1

pi =
p P

pn(p)

Uniqueness: Consider any a = 0 represented as a = p pn(p) as presented in the theorem. Then we x any q P, it is clear that a = hq n(q) where h :=
p=q

pn(p)

It is clear that q | q n = n, as q n | q n and q n+1 | q would imply q | 1 and hence q R . Further it is easy to see that q | h = 0. Because if q | h then q | h (as R ) and hence there has to be some p = q such that q | p. That would be q p due to the irreducibility of p in contradiction to p = q (as P was a representing system). Hence we may use (i) again to nd q | a = q | hq n(q) = q | h + q | q n(q) = n(q ) In particular the exponent n(q ) in this representation of a is uniquely determined to be q | a . And as q has been arbitrary this means that n is uniquely determined. And as R is an integral domain this nally yields that is uniquely determined, as well. (iii) Suppose b = ah for some h R. Then for any p R prime we get p | b = p | ah = p | a + p | h p | a . Conversely suppose p | a p | b for any p R prime. Pick up a pepresenting system P and use it to write a and b in the form a =
pP

pm(p) and b =
pP

pn(p)

428

18 Proofs - Commutative Algebra

By assumption we have m(p) = p | a p | b = n(p). That is k (p) := n(p) m(p) 0. Hence we explitly nd some h R such that b = ah where h := 1
p P

pk(p)

(v) Let us abbreviate d := p pm(p) then we need to verify that d is a greatest common divisor of A. By denition of m(p) we have p | d = m(p) p | a for any p P and any a A. And by (iii) that is d | a and for any a A and hence d | A. Conversely consider any c R with c | A. In particular we have c = 0 as 0 A = . And hence c | a implies p | c p | a for any p P (and a A) by (iii) again. This is p | c m(p) = p | d for any p P and hence c | d by (iii). The statement for the least common multiple can be proved in complete analogy. We only have to note that the assumption #A < guarantees n p N again. (vi) Once more we choose a representing system P of the primes of R. And for any p P let us abbreviate m(p) := min { p | a , p | b } and likewise n(p) := min { p | a , p | b }. Then it is a standard fact that m(p) + n(p) = p | a + p | b . By (ii) and (v) that is p | ab = = p | a + p | b = m(p) + n(p) p | a + p | b = p | dm

And as this holds true for any p P (iv) immediately yields the claim. (vii) We commence with the notation of (vi) and choose any e gcd{a, b, c}. Then by (v) the order of p P in c is given to be the following p|e = min{ p | a , p | b , p | c } = min{ min{ p | a , p | b }, p | c } = min{ p | d , p | c } = p | e And the latter is the order of any greatest common divisor of e gcd{c, d}. As this is true for any p P we thereby nd that e and e are associated. The claim for the greatest common divisors follows immediately. The analogous argumentation also rests the case for the least common multiples. P Proof of (2.55): As R is an UFD, it is an integral domain, by denition. Thus choose any 0 = x = b/a quotR that is integral over R, i.e. a1 , . . . , an R with xn + a1 xn1 + + an = 0 Then we need to show that a | b already. But as R is an UFD we may let d := gcd{a, b} and write a = d and b = d. Then and are relatively 429

prime and a | b can be put as R (as then b = (1 )a). By multiplying the above equation with n we nd (note x = /) n + (a1 n1 + + n1 an ) = 0 And hence divides n - but as and were relatively prime, so are and n . But together this implies that truly is a unit of R. P Proof of (2.56): (ii) We rst prove the irreducibility of 2 Z[ 3]. Thus let := i 3 and suppose 2 = (a + b )(c + d ). Then by complex conjugation we nd 2 = (a b )(c d ). Multiplying these two equations we get 4 = 2 2 = (a2 + 3b2 )(c2 + 3d2 ) 9(bd)2 Thus if both b = 0 and d = 0 this formula ends up with the contradicion 4 9(bd)2 9. Thus without loss of generality (interchanging a, c and b, d if necessary) we may assume d = 0. Then we have 4 = (a2 + 3b2 )c2 Of course c = 0 cannot be (else 4 = 0). If |c| = 1 then c + d = 1 which of course is a unit in Z[ 3]. Thus suppose |c| 2 then c2 4 such that necessarily a2 + 3b2 = 1. This of course can only be if a = 1 and b = 0 such that now a + b is a unit of Z[ 3]. Altogether 2 is irreducible. On the other hand 2 | 4 = (1 + )(1 ). Now suppose we had 2 | 1 + , say 2(a + b ) = 1 + . Then 2a = 1 and hence 2 = Z which is false. Likewise we see that 2 does not divide 1 and this means that 2 is not prime. (iii) We rst prove that p = s2 + t2 1 is irreducible in R[s, t]. To do this we x the graded lexicographic order on R[s, t], that is R[s, t] is ordered by (total) degree and by s < t. Now consider any two polynomials f , g R[s, t] such that f g = p. Then lt (f )lt (g ) = lt (f g ) = lt (p) = lt (1 + s2 + t2 ) = t2 By prime decomposition of the leading coecient in R[s, t] we see that either lt(f ) = 1, lt (f ) = t or lt (f ) = t2 . If lt (f ) = 1 then f is a unit in R[s, t], likewise if lt (f ) = t2 then lt (g ) = 1 and hence g is a unit in R[s, t]. thus we only have to exclude the case lt (f ) = t. But in this case lt (g ) = t as well such that f = a + bs + ct

g = u + vs + wt for some coecients a, b, c, u, v and w R. From this we may compute f g and compare its coecients with those of p. Doing this an easy computation yields 430 18 Proofs - Commutative Algebra

(1) (2) (3) (4) (5) (6)

au av + bu aw + cu bv bw + cv cw

= = = = = =

-1 0 0 1 0 1

constant term s-term t-term s2 -term st-term t2 -term

In particular a = 0 and hence we may regard (f /a)(ag ) = f g = p instead. That is we may assume a = 1, then u = 1 by (1). And thus (2) turns into b = v and (3) becomes c = w. From this and (5) we get 2bc = 0. That is b = 0 or c = 0. But b = 0 would be a contradiction to (4) and c = 0 would be a contradiction to (6). Thus lc (f ) = t cannot occur and this is the irreducibility of p. We now prove that R = R[s, t]/p is no UFD. To do this let us denote the set X := (x, y ) R2 | x2 + y 2 = 1 . And thereby we dene the following multiplicatively closed subset of R U := { f + p R | (x, y ) X : f (x, y ) = 0 } By construction it is clear that U is truly well dened (if f g = q p then for any (x, y ) X we get f (x, y ) = g (x, y )+ q (x, y ) = g (x, y )) and multiplicatively closed (as R is an integral domain). Let us now denote the element r := s + p R. Then we will prove that r/1 U 1 R is irreducible but not prime. Thus U 1 R is no UFD and hence R has not been an UFD by virtue of (2.110). Let us rst prove that r/1 is not prime. To do this let us dene the following elements a := 2(t + 1) + p and b := 2(t 1) + p R. Then ab = 4(t2 1) + p = 4s2 + p = 4r2 . In particular r | ab and hence r/1 | (ab)/1 = (a/1)(b/1). Now assume r/1 | a/1 that is there are some h + p R and u + p U such that a/1 = (r/1)(h + p/u + p). As R[s, t] is an integral domain that is 2(t + 1)u + p = sh + p or in other words 2(t + 1)u sh =: q p. Now regard (0, 1) X then 0 = q (0, 1) = 4u(0, 1) = 0, a contradiction. That is r/1 | a/1 and likewise we nd r/1 | b/1 by regarding (0, 1) X . So it only remains to prove that r/1 is irreducible. Thus suppose there are some f , g R[s, t] and u + p, v + p U such that s+p r f +pg+p fg + p = = = 1 1 u+pv+p uv + p That is suv + p = f g + p or in other words suv f g =: q p. As q p = pR[s, t] and p vanishes on X identically, so does q . And this means that for any (x, y ) X we have the following identity xu(x, y )v (x, y ) = f (x, y )g (x, y ) In particular f g and suv and have the same roots. And as u, v do not vanish on X these are precisely the points (0, 1) and (0, 1) X . If f does not vanish on X then f + p U and hence f + p/u + p is a unit of U 1 R. Likewise if g does not vanish on X then g + p/v + p is a unit 431

of U 1 R. Thus in order to prove the irreducibility of r/1 it suces to check that at f or g does not vanish on X . Thus assume this was false, that is f and g vanish on at least one point, say f (0, 1) = 0 and g (0, 1) = 0 (else interchange f and g ). We now parametrisize X using the following curve : [0, 2 ] X : cos( ) sin( )

First suppose neither f (0, 1) = 0 nor g (0, 1) = 0, that is f and g both have precisely one root on X . Then f cannot change signs on X (else [0, 2 ] R : f ( ) would have to have at least two roots because of f (0) = f (2 ) = 0). And the same is true for g . Thus (x, y ) f (x, y )g (x, y ) does not change sign on X . Likewise uv does not have a single root on X and hence its sign on X is constant. Hence (x, y ) xu(x, y )v (x, y ) does change sign in contradiction to the equality of these two functions. Thus at least one of the two f or g even has to have two roots on X . We only regard the case f (0, 1) = 0 and g (0, 1) = 0 = g (0, 1). The other case (where f has both roots) is completely analogous. The basic ideas is the following: as both f and g vanish in (0, 1) this is a root of order 2 for f g . But xuv only has a root of order 1 in (0, 1). Thus let us do same basic calculus - we have (/2) = (0, 1) and in this point we get (/2) = (1, 0). Thus in this case we nd another contradiction to suv = f g by evaluating (suv ) (/2) = uv + svs u + sus v svt u + sut v = u(0, 1)v (0, 1) = 0 0 1 | 1 0

(f g ) (/2) = = Proof of (2.59):

gs f + gs g 0 | gt f + gt f 1 (0, 0) | (1, 0) = 0

1 0

(ii) First consider a > 0, then it is clear that q := b div a Z exists, as the set {q Z | aq b} = ] , b/a] Z contains the maximal element q = a/b rounded down. And from the construction of r := b mod a it is clear that b = aq + r. From the denition of q it is also clear that 0 r. Now assume r a then a(q +1) = aq + a aq + r = b and hence q would not have been maximal - a contradiction. Thus we also have 0 r < a and hence (r) = r < a = (a). Thus we have established the division with remainder for a > 0. If conversely a < 0 then a > 0 and hence we may let q := (b) div (a) and r := ((b) div (a)). Then we have just proved b = q (a) r which yields b = qa + r and (r) = (r) < (a) = (a). (iii) Consider any g E [t], 0 = f E [t] and let := lc (f ). Then f := 1 f E [t] is normed. Further let g := 1 g . Then by (??) 432 18 Proofs - Commutative Algebra

there are q , r E [t] such that g = qf + r. Multiplying with we nd g = qf + r and deg(r) = deg(r) < deg(f ) (as E ). Thus we have established the division with remainder on E [t]. (iv) It is straightforward is a subring of C. to check that Z[ d] truly Clearly 0 = 0 + 0 d Z[ d] and 1 = 1 + 0 d Z[ d]. Now consider x = a + b d and y = f + g d Z[ d]. Then it is clear that x+y = Z[ d] (a + f ) + (b + g ) d xy = (af + dbg ) + (ag + bf ) d Z[ d]

Thus Z[ d] is a subring of C and in particular an integral domain. Now assume x = y , then we wish to verify (a, b) = (f, g ) (the converse implication is clear). Assume we had b = g then we would be able to divide by b g = 0 and hence get d = f a Q bg

a contradiction to the d Q. Thus we have b = g and assumption hence (substracting b d = g d) also a = f . Thus x x is truly well dened and the bijectivity is clear. We also see 0 = 0 0 d = 0 and 1 = 1 0 d = 1 immediately. It remains to prove that x x also is additive and multiplicative, by computing x + y = (a b d) + (f g d) = (a + f ) (b + g ) d) = x + y xy = (a b d)(f g d) = (af + dbg ) (ag + bf ) d = xy But from this it is clear that also is multiplicative. Just compute note that (xy ) = |xyxy | = |xxyy | = |xx| |yy | = (x) (y ). We nally 2 for any 0 = x Z[ d] we get (x) = 0, since (x) = |a db2 | = 0 is equivalent to a2 = db2 . Thus if b = 0 then a = 0 and hence x = 0 as 2 = d. In claimed. And if b = 0 then we may divide by b to nd (a/b) other words d = a/b Q, a contradiction to the assumption d Q. Thus the case b = 0 cannot occur in the rst place. Next we prove that the units of Z [ d] are precisely the elements x such that (x) = 1. If x Z[ d] is a unit, then we have 1 = (1) = (xx1 ) = (x) (x1 ) and hence (x) Z . But (x) 0 this as also yields (x) = 1. Thus consider any x = a + b d Z[ d] such that (x) = 1. Then xx = a2 db2 = 1 and hence x1 = x is invertible. Now assume that d 3 then we want to prove that 2 is an irreducible element of Z[ d] that is not prime. We begin by proving that 2 is not prime: as either d or d 1 is even we have 2 | d(d 1) = d2 d = (d + d)(d d)

but we have 2 | d d , because if we assume 2( a + b d ) = d d then (comparing the d-coecient) we had 2b = 1 Z in contradiction to 2 Z . However 2 is irreducible, assume 2 = xy for some x, y Z[ d]. Then 4 = (2) = (xy ) = (x) (y ). Hence we have (x) {1, 2, 4}. If (x) = 1 then x Z[ d] is a unit. And if (x) = 4 then (y ) = 1 and 433

 hence y Z[ d] is a unit. It suces to exclude the case (x) = 2, that is assume (x) = 2. As d 3 we have 2 = (x) = a2 db2 . Thus in the case b = 0 we obtain 2 = a2 db2 0 + 3 1 = 3 a contradiction. And in case b = 0 we get 2 = a2 for some a Z, in contradiction to the irreducibility of 2 in Z. Thus (x) = 2 cannot occur and this means that 2 Z[ d] is irreducible. It remains to prove that (Z[ d], ) even is an Euclidean domain for d {2, 1, 2, 3} . We consider the elements 0 = x = a + b d and y = f + g d Z[ d], then (as (y ) = 0) we may dene p := q := af dbg a2 db2 ag bf a2 db2

That is p + q d := (1/(a2 db2 ))yx. And therefore we obtain the 2 dg 2 ))yxx = y . In other words we have identity (p + q d)x = (1/(f found y/x to be p + q d Q[ d]. Now choose r, s Z such that |p r| and |q s| 1/2 (which is possible, as the intervals [r 1/2, r + 1/2] (for r Z) cover Q) and dene u := r + s d v := y ux Then it is clear that u, v Z[ d] and y = ux + v . But further we nd y x (p r ) + ( q s ) d = x u = y ux = v x Due to the multiplicativity of (which even remains true for coecients (here pr and q s) in Q ) we nd (v ) = (x)|(pr)2 d(q s)2 |. But as |d| 3 we can estimate |(p r)2 d(q s)2 | (p r)2 + |d|(q s)2 (1/2)2 + 3(1/2)2 = 1 The case |(p r)2 d(q s)2 | = 1 can only occur if |p r| = |p s| = 1/2 and d = 3. But in this case we have |(p r)2 d(q s)2 | = |1/4 3/4| = 1/2 < 1. Thus we have even proved the strict estimate (v ) = (x)|(p r)2 d(q s)2 | < (x) Altogether we have domain and that seen that Z[ d] is an integral given 0 = x, y Z[ d] we can nd u, v Z[ d] such that y = ux + v and (v ) < (x). But this is the condition to be an Euclidean domain. P Proof of (2.60): Without loss of generality we assume (a) (b) and let f1 := b and f2 := a. That is if f and g are determined using the initialisation of the algorithm, then we get f1 = g and f2 = f . Suppose we have 434 18 Proofs - Commutative Algebra

already computed f1 , . . . , fk R and fk = 0, then in the next step the algorithm computes qk := q and fk+1 := r R such that fk1 = qk fk + fk+1 and (fk+1 ) < (fk ). Thus by induction on k we nd the following strictly decreasing sequence of natural numbers (b) = (f1 ) (f2 ) > (f3 ) > . . . > (fk ) > (fk+1 ) > . . . As (b) is nite this chain cannot decrease indenitely but has to terminate at some level n. That is f = fn+1 = 0. And therefore fn1 = qn g where g = fn . We will now prove that for any 1 k n we have g | fk . For k = n we trivially have g | g = fn . And for k = n 1 we have fn1 = qn g by construction. Thus by induction we can assume k n 1 and g | fn to g | fk . Then from fk1 = qk fk + fk+1 it also is clear that g | fk1 completing the induction. In particular we have g | f2 = a and g | f1 = b. Now suppose we are given some d R with d | a = f2 and d | b = f1 . Then we will prove that for any k 1 . . . n we get d | fn by induction on k . Thus assume k 2 and d | f1 to d | fk . Then from fk+1 = fk1 + qk fk we also get d | fk+1 completing the induction. In particular we get d | fn = g . And hence g truly is the greatest common divisor of a and b. We will now prove the correctness of the rened Euclidean algorithm. The rst part (the computation of g ) is just a reformulation of the standard Euclidean algorithm. And as the second part is a nite computation, it is clear that the algorithm terminates. Next we will rst prove that for any 2 k N we have fk = rk f2 + sk f1 . For k = 2 this is trivial f2 = 1 f2 + 0 f1 = r2 f2 + s2 f1 . And for k 3 we have computed qk R such that fk1 = qk fk + fk+1 for some fk+1 R. In particular f3 = f1 q2 f2 = (q2 )f2 + 1 f1 = r3 f2 + s3 f1 . Now we will use induction on k . That is we assume fk1 = rk1 f2 + sk1 f1 and fk = rk f2 + sk f1 . Then a computation immediately yields rk+1 f2 + sk+1 f1 = (rk1 qk rk )f2 + (sk1 qk sk )f1 = (rk1 f2 + sk1 f1 ) qk (rk f2 + sk f1 ) = fk1 qk fk = fk+1 In particular for k = n we nd g = fn = rn f2 + sn f1 . Thus let us distinguish the two cases. If (a) (b) we have f1 = b, f2 = a, r = rn and s = sn . Thereby we get g = rn f2 + sn f1 = ra + sb. And conversely if (a) > (b) then f1 = a, f2 = b, r = sn and s = rn . Thereby we get g = rn f2 + sn f1 = sb + ra. Thus in any case we have obtained r and s R such that g = ra + sb. P Proof of (2.62): 435

(i) We use induction on n to prove the statement. For n = 0 and any a R0 = R it is clear that a R1 , as for any b R we may choose r = 0 and thereby get a(ba1 ) = b such that a | b r. Thus for n n + 1 we are given a Rn and any b R. That is we have a | b r for some r Rn1 . But by induction hypothesis we know Rn1 Rn and hence r Rn . But this again means a Rn+1 (as b has been arbitrary) which had to be proved. (ii) We use induction on n again. For n = 0 we are given any a R such that (a) 0. If (a) < 0 then (a) = and hence a = 0. Otherwise (a) = 0 and we use division with remainder to nd some q , r R such that 1 = qa + r and (r) < (a) = 0. But this can only be if (r) = which is r = 0. Hence we have 1 = qa, that is a R = R0 is a unit of R. In the induction step n n + 1 we are given some a R with (a) n + 1. Given b R we use division with remainder to nd q , r R such that b = qa + r and (r) < (a) n + 1. Thus we have (r) n and by induction hypothesis this yields r Rn {0}. That is a | b r for some r Rn {0}. And as b has been arbitrary this means a Rn+1 , which had to be shown. (iii) Clearly is well dened, as by assumption any 0 = a R is contained in some Rn . We need to prove that truly allows division with remainder: thus consider any 0 = a and b R, then we need to nd q , r R such that b = qa + r and (r) < (a). If (a) = 0 then a R0 = R and hence we may choose q := ba1 and r = 0. If (a) 1 then we let n := (a) 1 N. As a Rn+1 and b R we have a | b r for some r Rn {0}. This again yields (r) n < n + 1 = (a) and b = qa + r for some q R, just what we wanted. (iii) Next we want to prove (a) (ab). If ab = 0 then (as b = 0 and R is an itegral domain) we have a = 0. Thereby (a) = = (ab). Thus assume ab = 0, and let n := (ab) N. If n = 0 then ab R0 = R and hence a R = R0 (with a1 = b(ab)1 ). Thus we likewise have (a) = 0 = (ab). If now n 1 then for any c R there is some r Rn1 {0} such that ab | c r. But as a | ab this yields a | c r, too and hence a Rn again. Thereby (a) n = (ab). (iii) If a R then by (iii) above we have (b) (ab) (a1 ab) = (b) and hence (b) = (ab). Conversely suppose (b) = (ab). As b = 0 we also have ab = 0 (else (ab) = = (b)). Thus we may use division with remainder to nd some q , r R such that b = q (ab) + r and (r) < (b). Rearranging this equation we nd (1 qa)b = r. Now suppose we had 1 qa = 0, then by (iii) above again we had (b) ((1 qa)b) = (r) < (b), a contradiction. Thus we have qa = 1 and this means a R as claimed. (iv) If R \ {0} = n Rn then (R, ) is an Euclidean domain by (iii). Conversely suppose (R, ) is an Euclidean domain for some Euclidean function . Now consider any 0 = a R, and let n := (a). Then we have already seen in (ii) that a Rn and hence we have R \ {0} = n Rn . (v) As (R, ) is an Euclidean domain we have R \ {0} = n Rn by (iv). But by (iii) this implies that (R, ) is an Euclidean domain, too. Thus 436 18 Proofs - Commutative Algebra

consider any a R, if a = 0 then (a) = = (a) by convention. Else let n := (a) N. Then a Rn by (ii) and hence (a) n = (a) by construction of . P Proof of (2.65): We will only prove parts (i), (ii) and (iii) here. Part (iv) requires the theory of Dedekind domains an will be postponed until then - see page (487). (i) Clearly 0 i R is a prime ideal, as R is an integral domain (if ab 0 then ab = 0 and hence a = 0 or b = 0 such that a 0 or b 0 again). And further any maximal ideal is prime, because of (2.19). Conversely consider any non-zero prime ideal 0 = p i R. As R is a PID there is some p R such that p = pR. By (2.48.(ii)) this means that p is prime and due to (2.48.(v)) this implies that p is irreducible (as R is an integral domain). Now consider any ideal a such that p a i R. As R is a PID there is some a R with a = aR again. Now p pR aR means that there is some b R such that p = ab. But p has been irreducible, that is a R or b R . In the case a R we have a = aR = R. And in the case b R we have a b and hence a = aR = pR = p. As a has been arbitrary this means that p already is maximal. (ii) As R is a PID every ideal a of R is generated by one element a = aR. In particular any ideal a is generated by nitely many elements. Thus R is notherian by property (c) in (2.27). We will now prove property (d) of UFDs. As R is noetherian property (1) in (2.50.(d)) is trivially satised by (ACC) of noetherian rings. Hence it suces to check property (2) of (2.50.(d)). But as R is an integral domain any prime element p R already is irreducible, due to (2.48.(ii)). This leaves p R irreducible = p R prime

Thus suppose p | ab for some p, a and b R, say pq = ab. As R is a PID and aR + pR i R is an ideal we nd some d R such that dR = aR + pR. That is there are r, s R such that d = ar + ps. And as a aR + pR = dR there is some f R with a = df . Likewise we nd some g R with p = dg . But as p is irreducible we have g R or d R . If g R then we get p | a from p(f g 1 ) = (g 1 p)f = df = a And if d R we get p | b by the equalities below. Thus we have found p | a or p | b which means that p is prime p(d1 qr + d1 bs) = d1 (pqr + pbs) = d1 (abr + pbs) = bd1 (ar + ps) = bd1 d = b (iii) If a = 0 then a = 0R is a principal ideal (generated by 0 R). And if a = 0 there is some 0 = b b. Therefore we may choose a as given in the claim of (iii). Now consider any b a and choose q , r R such 437

that b = qa + r and (r) < (a). Then r = b qa a (as a, b a). But a has minimal value (a) among all those b a with b = 0. Therefore r a and (r) < (a) implies r = 0. Thus b = qa aR, and as b has been arbitrary this means a aR. And as a R we have aR a. Together we nd that a = aR is a principal ideal. And as a has been arbitrary this means that R is a PID. P Proof of (2.67): (i) As R is an UFD by (2.65.(ii)) the length (a) N of 0 = a R is well dened, due to (2.50.(b)). Therefore : R N is a well-dened function and it is clear that it satises properties (1) and (2). Thus consider any 0 = a, b R, as R is a PID we may choose some r R such that rR = aR + bR. It is clear that r = 0, as 0 = a rR. Now suppose b aR, Then it is clear that r aR + bR and it remains to prove (r) < (a). As a aR + bR = rR there is some s R such that a = rs. Suppose s R , then a r and hence aR = rR. But b aR + bR = rR = aR is a contradiction to b aR. Thus s R which means (s) 1. Hence we have (s) 2 and as (r) > 0 (recall r = 0) this nally is (r) < 2 (r) (r) (s) = (rs) = (a) (ii) As a = 0 there is some 0 = b a. And hence we may choose a a as given in the claim of (ii). We now want to verify a = aR, as a a the inclusion aR a is clear. Conversely choose any 0 = b a (0 aR is clear). If we had b aR, then by property (3) there would be some r aR + bR such that (r) < (a). But as a, b a we have aR + bR a and hence r a. As a has been chosen to have minimal value (a) this only leaves r = 0 and hence 0 = a aR + bR = rR = 0, a contradiction. Thus we have b aR and hence a aR as well. (iii) By construction of it is clear that : R N and that satises (2). It remains to verify (3), thus consider any 0 = a, b R then we choose q , r R such that b = qa + r and (r) < (a). If r = 0 then b = qa aR. And if r = 0 we have r = qa + b aR + bR and (r) = (r) + 1 < (a) + 1 = (a). P Proof of (2.68): (i) First suppose mR = a aR, in particular mR aR for any a R. But this is a | m for any a A and hence A | m. Now consider any n R such that A | n, that is a | n and hence n aR for any a A. Therfore n a aR = mR and hence m | n. Thus we have proved m lcm(A). Conversely suppose m lcm(A), then A | m and hence m a aR again. In particular we have mR a aR. Now consider any n a aR, then A | n again and hence m | n by assumption on m. Thus we have n mR and as n has been arbitrary this is the converse inclusion a aR mR. 438 18 Proofs - Commutative Algebra

(ii) Let dR = a aR then for any a A we have aR dR and hence d | a. Thus we found d | A, as a has been arbitrary. Now consider some c R such that c | A. That is for any a A we get c | a and hence aR cR. But from this we get dR = a aR cR which also is c | d. Together we have proved d gcd(A) again. (iii) Now assume that R is a PID, then we will also prove the converse implication. By assumption on R there is some g R such that gR = a aR and by (ii) this means g gcd(A). Now from (2.48.(viii)) and d, g gcd(A) we get d g and hence dR = gR = a aR. (iv) For n = 1 the statement is trivial for b1 := b, thus let us assume n 2. Then we denote a := a1 . . . an and ai := a/ai R. Now consider d gcd{a1 , . . . , an } and suppose we had d R . Then (as R is an UFD) there would be a prime element p R dividing p | d. Thus we had p | d | a1 and hence (as p is prime) there would be some i 2 . . . n such that p | ai . Likewise for i we have p | d | ai and hence p | aj for some j = i. But ai and aj are relatively prime by assupmtion, a contradiction. Thus we have d R , that is 1 gcd{a1 , . . . , an }. And by (iii) this means that there are some h1 , . . . , hn R such that 1 = h1 a1 + + hn an . Now let bi := bhi , then the claim is immediate from
n i=1

bi = ai

n i=1

b bhi ai = a a

hi ai =
i=1

b a P

Proof of (2.69): (a) = (b): consider a, b R, by assumption there are r, s R such that d = ra + sb gcd(a, b). By construction we have d aR + bR and hence dR aR + bR. On the other hand we have d gcd(a, b) and hence d | a and d | b. But that is aR dR and bR dR such that aR + bR dR as well. Together that is dR = aR + bR. (b) = (c): a being nitely generated means that there are some a1 , . . . , an R such that a = a1 R + + an R. We will now use induction on n to prove that a is principal. In the case n = 0 we have a = 0 = 0R and in the case n = 1 trivially a = a1 R is principal. Thus we may assume n 2. Then by induction hypothesis u := a1 R + + an1 R i R is principal, that is there is some u R such that u = uR. Now a = a1 R + + an R = u + an R = uR + an R. Thus by assumption (b) there is some d R such that a = dR is principal. (c) = (a): consider any a, b R and let a := aR + bR. By construction a is nitely generated and hence there is some d R such that a = dR. Now d aR + bR means that there are some r, s R such that d = ra + sb. And by (2.68.(ii)) dR = aR + bR yields that ra + sb = d gcd(a, b) as well. P 439

Proof of (2.70): (a) = (b) and (c): if R is a PID then R is a noetherian UFD due to (2.65.(ii)). And it is clear that R also is a Bezout domain (e.g. by property (c) in (2.69)). (c) = (a): consider any ideal a i R, as R is noetherian a is nitely generated, due to property (c) of noetherian rings (2.27). And due to poroperty (c) of Bezout domains (2.69) this means that a is principal. As a Bezout domain also is an integral domain this means that R is a PID. (b) = (a): consider any ideal a i R. If a = 0 then a = 0R trivially is principal. Thus we assume a = 0 and choose any 0 = a R with (a) = min{ (b) | 0 = b a } where (a) is the number of prime factors of a, that is (a) = k , where a = p1 . . . pk for some prime elements pi R. As a a we have aR a. Conversely consider any 0 = b a. As R is a Bezout domain the ideal aR + bR is principal aR + bR = dR for some d R, due to property (b) in (2.69). As a, b a we have d aR + bR a and clearly d = 0 (as a = 0). Thus by construction of a we have (a) (d). But on the other hand we have a aR + bR = dR and hence d | a which implies (d) (a) by (2.54.(iii)). But from d | a and (a) = (d) we get a d by choosing prime decompositions of a and d as in (2.54.(iv)). Therefore we have b aR + bR = dR = aR. This proves b aR and hence a aR, as b has been arbitrary (b = 0 is clear). Altogether we have a = aR is principal and hence R is a PID, as a has been arbitrary. P Proof of (2.73): (a) = (b): consider a + a zdR/a, that is a zdR R/a, then by assumption we have a a. That is ak a for some k N and hence (a + a)k = ak + a = 0 + a such that a + a is a nilpotent of R/a. (b) = (c): the nil-radical of R/a is trivially included in the set of zero-divisors of R/a due to (1.43.(ii)), which implies equality. (c) = (a): consider a zdR R/a, that is a + a zdR/a. Then by k = 0+a assumption a + a is a nilpotent of R/a, that is ak + a = (a + a) k for some k N. But this again means a a and hence a a. (c) = (d): consider a, b R such that ab a but b a. That is (a + a)(b + a) = ab + a = 0 + a and as b + a = 0 + a this means that a + a zdR/a is a zero-divisor of R/a. By assumption this means k that a + a is a nilpotent of R/a and hence ak + a = (a + a) = 0 + a k for some k N. But this again is a a and hence a a. (d) = (b): consider a + a zdR/a that is there is some 0 + a = b + a R/a such that ab + a = (a + a)(b + a) = 0 + a. But this means 440 18 Proofs - Commutative Algebra

 ab a and as also b a we nd a a by assumption. That is ak a for some k N and hence (a + a)k = ak + a = 0 + a such that a + a is a nilpotent of R/a. The equivalency (d) (e) is nally true by elementary logical operations (note that we may commute the quantiers a and b) (d) a, b R : (ab a) (b a) (a a) a, b R : (ab a) (b a) (a a) b, a R : (ba a) (a a) (b a) a, b R : (ab a) (b a) (a a) a, b R : (ab a) (b a) (a a) (e) P Proof of (2.75): We will only prove statements (i) to (vii) here. Statement (viii) is concerned with localisations and requires the correspondence theorem of ideals in localisations. Its proof is hence postponed until we have shown this theorem. Hence it can be found on page (470). (i) If p i R is prime and we assume ab p for some a, b R, then a p or b p. Thus if also b p then necessarily a p p. Thus p is primary, by property (d) in (2.73). (ii) Let p := a, then p is an ideal of R, due to (2.17.(ii)). Now assume 1 p, then there would be some k N such that 1 = 1k a. That is 1 a and hence a = R in contradiction to a being a primary ideal. Now assume ab p for some a, b R. That is there is some k N k k k k such that (ab) = a b a. If we have b a = p then b p, as p is a radical ideal. Thus assume bk a. Then by property (e) we get ak a and hence a p. Thus p is a prime ideal. (ii) It remains to prove that p = a is the uniquely determined minimal prime ideal containing a. Thus assume a q i R is any prime ideal containing a. Then we have p = a q = q. Thus if q p, then p = q, such that p is a minimal prime ideal containing a. And if p is minimal then p p again and hence p = p, due to minimality. Thus p also is uniquely determined. (iii) Let m := a and assume that m is maximal. Now consider any a, b R such that ab a but b a. Then we have to prove a m (in order to satisfy property (d) of primary ideals). Thus suppose a m, then (because of the maximality of m) we had R = m + aR. That is there would be some h R such that 1 ah m. And by denition of m this is (1 ah)k a for some k N. Now dene
k

f :=
i=1

(1)i

k (ah)i R i a 441

Then using the binomial rule we nd 1 af = (1 ah)k a by a straughtforward computation. That is af + a = 1+ a and hence 0+ a = (f + a)(0+ a) = (f + a)(ab + a) = (f a + a)(b + a) = (1+ a)(b + a) = b + a. That is b a in contradiction to the assumption. Thus we have a m, which we had to prove. (iv) From the assumption we get m = m = mk a m = m due to (2.17.(vi)) (and as any maximal ideal is prime and hence radical). Thus we have a = m and hence a is a primary ideal by (iii). (v) Denote a := a1 ak , then we rst note that according to (2.20.(v))
k k

a =
i=1

ai =
i=1

ai =
i=1

p = p

Now consider any a, b R such that b a. That is there is some i 1 . . . k such that b ai . But as ab a ai and ai is primary we nd a ai = p = a. That is a satises property (d) in (2.73). (vi) Consider any a a : u, that is there is some ak u a. k N such that k As u a and a is primary, we nd a a and hence a a = p, as a is a radical ideal. That is we have proved a : u p, but as a a : u the converse incluison is clear from p = a a : u. Thus we have a : u = p. Now consider any a, b R such that ab a : u but b a : u. This means abu a but bu a. As a is primary we again nd a a = p = a : u and hence a : u is primary, as well. (vii) First observe that the radical of a is truly given to be p, as we compute
a =

aR aR aR

k N : ak a = 1 (b) k N : (a)k = (ak ) p (a)

b=q

= =

= 1 (q) = p

Thus consider a, b R sucht that ab a but b a, that is (a)(b) = (ab) b but (b) b. As b is primary this yields (a) b = q and hence a 1 (q) = a. That is a is primary again. P Proof of (2.76): (i) Let us denote p := pR, then p is a prime ideal, as p is a prime element. Also we have pn = pn R = a and hence a = pn = p = p. That is a is associated to p. We will now prove, that a is primary, by using induction on n. For n = 1 we have a = pR = p, that is a is prime and hence primary by (2.75.(i)). Now consider any n 2, and a, b R such that ab a but b a = p. That is there is some h R such that ab = pn b but p | b. As p is prime, p | hpn = ab and p | b we have p | a, that is there is some R such that a = p. As R is an integral domain we get b = (ab)/p = (pn h)/p = pn1 h. Now 442 18 Proofs - Commutative Algebra

from ab pn1 R and b pn1 R = p the induction hypothesis yields pn1 R and therefore a = p pn R = a, that is a is primary. (ii) If a = 0 then a is prime (as R is an integral domain) and hence primary (by (2.75.(i))). And if a = pn R for some prime element p R and 1 n N, then a is primary, as we have seen in (i). Conversely consider any primary ideal a i R. As R is a PID it is generated by a single element a = aR. Clearly a R (as else a = R would not be primary). If a = 0 then a = 0 and hence there is nothing to prove. Thus assume a = 0, recall that R is an UFD by (2.65) and assume there would be two non-associate prime elements p, q R such that p | a and q | a. Then we let m := a | p and n := a | q (clearly 1 m, n N by (2.54.(iv))) and thereby get pm q m | a. That is there m n is some b R such that p q b = a and p | b and q | b. It is clear that m p a as else there would be some k N such that q | a | pmk . On the other hand we have q n b a, as else p | a | q n b. This contradicts a being a primary ideal and hence there is only one prime element p R (up to associateness) dividing a. Thus a = pm for some R and this nally means a = aR = pn R. (iii) First of all m = s, t i is a maximal ideal, as R/m is a eld (in fact we nd R/m =r E : f + m f (0, 0)). Further we have a = s, t2 i s, t i = m. It is even true that t a and hence a m (suppose t = f s + gt2 for some f , g R. Then letting t = 0 we would nd 0 = f s and hence f = 0. But this yields 1 = gt a contradiction to t R ). Next we have m = s2 , st, t2 i s, t2 i = a. And in analogy to the above we nd s m2 , altogether this means
m2

As m is maximal (2.75.(iv)) yields that a is primary and associated to m. Now suppose there would be some k N and p spec R k 2 k such that a = p . Then from m a = p m we would get 2 k m = m p = p m = m. That is p = m and hence m2 a = mk m. If k 1 then we have mk m a contradiction. And if k 2 then m2 mk a contradiction, too. Thus there is no prime ideal p i R and k N with a = pk . (iv) First of all it is clear that u = u2 st i t, u i . And as t, u i is a prime ideal of E [s, t, u] we nd that p = t, u i /u is a prime ideal of R, due to the corresopndence theorem (1.61). Now regard
a = b, c
2 i

= b2 , bc, c2

= b2 , bc, ba

= (bR) a, b, c

Then it is clear that a = p2 = p and ab = c2 a. However a (a bit cumbersome) computation yields b a and a p = a. This means that a is not primary. P Proof of (??): 443

(i) Suppose p = a b for some ideals a, b i R with a = p and p = b. As p = a b a this means p a, that is there is some a a with a p. Likewise there is some b b with b p. Yet ab a b = p, and as p is prime this yields a p or b p, a contradiction. Thus there are no such a, b and this means that p is irreducible. (ii) Consider a, b R such that ab p but b p. We will prove a p which is property (d) of primary ideals. Thus let b := p + bR. Then we regard the following ascending chain of ideals
p p : a p : a2 . . . p : ak p : ak+1 . . .

As R is noetherian this chain has to stabilize. That is there is some s N such that for all i N we get p : as = p : as+i . Now let a := p + as R. Then we will prove
p = ab

The inclusion is clear, as p a and p b by construction. Thus consider any h a b. As h b = p + bR there are some q p and g R such that h = q + bg . Therefore ah = aq + abg p as we have ab p. And as h a = p + as R as well, there are some p p and f R such that h = p + as f . Hence ah = ap + as+1 f such that as+1 f = ah ap. Now recall that ah p, then as+1 f = ah ap p and this means f p : as+1 = p : as . Therefore we have as f p and this nally means h = p + as f p. Thus we have established p = a b. However as b p we have p b. And hence the irreducibility of p yields p = a = p + as R. In particular as p and that is a p which had to be shown. (iii) Let us denote the set of all proper ideals that do not satisfy the claim (i.e. that are no nite intersection of irreducible ideals) by A := / p1 , . . . , pk i R : (1) and (2) a i R | a = R,

We have to prove A = , thus assume A = , as R is a noetherian ring this would mean that A contains a maximal element a A . As a A we in particular have that a is not irreducible (else a would be its own irreducible decomposition). And as also a = R this means that there are ideals a, b i R such that a = a b and a a and a b. By maximality of a this means a, b A and hence there are irreducible decompositions (p1 , . . . , pk ) of a and (q1 , . . . , ql ) of b. Hence
k l

= ab =
i=1

pi

j =1

qi

This means that a admits (p1 , . . . , pk , q1 , . . . , ql ) as an irreducible decomposition and hence a A, a contradiction. Thus we have A = , that is every proper ideal of R admits an irreducible decomposition. P

444

18 Proofs - Commutative Algebra

Proof of (2.81): By assumption (2) we have a = j aj for J := 1 . . . k . And hence we may choose a subset I of minimal cardinality with this property. Now it is clear that is an equivalency relation on I , as it belongs to the function p : I spec R : i
ai

Hence A = I/ is a well dened partition of I (the equivalency classes A are just the bers of p). And for any = [i] A we may dene p := ai . That is for any A the set {ai | i } is a nite collection of primary ideals associated to p . Hence by (2.75.(v)) a is a primary ideal associated to p as well. In particular (a ) (where A) satises property (1) of primary decompositions. Further we have
a =
A A i

ai =
i I

ai = a

as the partition I and construction of I . Thus also property (2) of primary decompositions is satised. And property (4) is clear from the construction of A precisely by this relation. It remains to verify property (3). That is x any = [j ] A and assume a = = a . As j we in particular nd
a =
iI

ai
j = i I

ai
=A

a = a

But this would mean a = i=j ai in contradiction to the minimality of I . Altogether we have proved properties (1) to (4), that is (a ) (where A) is a minimal primary decomposition of a. P Proof of (2.82): (i) By denition #ass(a1 , . . . , ak ) = #{p1 , . . . , pk } k . And if we suppose pi = pj then by minimality of (a1 , . . . , ak ) we have i = j (this is property (4)). Hence we even nd #{p1 , . . . , pk } = k as claimed. (ii) As a =
i ai

and using (2.17.(vi)) and (2.20.(v)) we can easily compute

k k k

a:u =
i=1

ai

:u =

(ai : u) =
i=1 i=1

ai : u

If u ai then we have ai : u = R according to (2.17.(iv)). And if u ai then ai : u is primary again with ai : u = ai = pi by virtue of (2.75.(vi)). Thus we have obtained
k

a:u =
i=1

R pi

if u ai if u ai

Thus we may ignore any i 1 . . . k with u ai , as they do not con445

tribute to the intersection. What remains is precisely the claim

a:u =
uai

pi

(iv) If p ass(a) then by denition of ass(a) there is some u R such that p = a : u. Thus by (ii) we have the identity
p =
uai

pi

And hence by (2.11.(ii)) there is some i 1 . . . k satisfying (u ai and) pi p. And as p pi is clear this is p = pi ass(a1 , . . . , ak ). Conversely choose any j 1 . . . k . As (a1 , . . . , ak ) is minimal we have a i=j ai and hence there is some ui ai for any i = j but uj uj . Thus for this uj by (ii) we get
pi = pj =
uj ai

a : uj ass(a)

(iii) Without loss of generality we consider p1 ass(a1 , . . . , ak ) . That is for any i 2 . . . k we have pi / p1 . And this again means that there is some a pi with a p1 . Now let U := R \ p1 , in particular a U . Then because of a pi = ai there is some n N such that an ai . And as an U , too, we get 1/1 = (an )/(an ) U 1 ai . Hence we get U 1 ai = U 1 R and this again means R = U 1 ai R = ai : U

for any i 2 . . . k . This now can be used to prove a : U = a1 by virtue of the following computation (see (2.105) - for convenience we even gave an elementary proof of (a b) : U = (a : U ) (b : U ) there)
k k

a:U =
i=1

ai

:U =

(ai : U ) = a1 : U = a1
i=1

Hereby a1 : U = a1 holds true because a1 is primary and U = R \ p1 . To be precise consider a ai : U , that is there is some u U such that au ai . But by denition of U we have u p1 = a1 , and as a1 is primary this yields a a1 . Thus we have a1 : U a1 and the converse inclusion is trivial. (v) If p ass(a) then by virtue of (iv) we have p = pi for some i 1 . . . k . And as pi is minimal (iii) implies a : (R \ p) = ai iso(a1 , . . . , ak ). Conversely consider any ai iso(a1 , . . . , ak ). Then by denition pi ass(a1 , . . . , ak ) and by (iii) this means ai = a : (R \ p). And by (iv) we also have pi ass(a) and thereby ai iso(a). P Proof of (2.84): 446 18 Proofs - Commutative Algebra

(1) By (2.78.(iii)) there are some irreducible ideals p1 , . . . , pk i R such that a = p1 pk . And by (2.78.(ii)) these irreducible ideals already are primary. Thus (p1 , . . . , pk ) is a primary decomposition of a. And by (2.81) we nd that a then already has a minimal primary decomposition. (2) Now assume that (a1 , . . . , ak ) and (b1 , . . . , bl ) are minimal primary decompositions of a. Then by (2.82.(iii)) we get the identity ass(a1 , . . . , ak ) = ass(a) = ass(b1 , . . . , bl ) (Note that by denition ass(a) and iso(a) only depend on a, not on the decomposition). And by (2.82.(i)) this in particular implies k = l. The third identity we nally get from (2.82.(iv)) iso(a1 , . . . , ak ) = iso(a) = iso(b1 , . . . , bl ) (3) Let us denote the minimal prime ideals over a by pi , that is we let {p1 , . . . , pk } := {p spec R | a p} (note that there are only nitely many pi , due to (2.36.(iii))). Then by (2.20.(i)) we have
a =

a =

{ p spec R | a p } =
i=1

pi

And as the pi are prime they in particular are primary according to (2.78.(i)). That is (p1 , . . . , pk ) is a primary decomposition of a. And as pi = pi = pj = pj (for i = j ) it even satises property (4) of minimal primary decompositions. Now assume that for some index j 1 . . . k we had the equality
pi = a
i=j

pj

Then by (2.11.(ii)) there would be some i = j such that a pi pj . And by minimality of pj this would mean pi = pj , a contradiction. Thus (p1 , . . . , pk ) also satises property (3), that is it is a minimal primary decomposition of a. In fact this even is ass(p1 , . . . , pk ) = iso(p1 , . . . , pk ), that is every associated ideal is an isolated component. Now assume that (a1 , . . . , al ) is any minimal primary decomposition of a, then by (2) we have l = k and { p1 , . . . , pk } = ass(p1 , . . . , pk ) = iso(p1 , . . . , pk ) = iso(a1 , . . . , ak ) That is for any i 1 . . . k we have ai iso(a1 , . . . , ak ) = {p1 , . . . , pk } and this again means that for any i 1 . . . k there is some (i) 1 . . . k such that ai = p(i) . Clearly is injective, because if we assume (i) = (j ) then ai = p(i) = p(j ) = aj in contradiction to the minimality of (a1 , . . . , ak ). But as k is nite this already means that is bijective, that is Sk . Altogether (a1 , . . . , ak ) is just a rearranged 447

version of (p1 , . . . , pk ). P Proof of (2.86): (i) Let (a1 , . . . , ak ) be a minimal primary decomposition of a (which exists, due to (2.84)) and pi := ai . Then we have seen in (2.82.(iv)) that ass(a) = {p1 , . . . , pk }. Thus from i ai = a q we get ai q for some i 1 . . . k , according to (2.11.(ii)). Therefore pi = ai q = q and pi ass(a). Now let pj q for some pj ass(a). Then we may choose some minimal pi ass(a) with pi pj (as ass(a) is nite). And hence we get ai pi pj q, that is ai q. Finally consider ai q then a ai q is clear. (ii) We stick to the notation used for (i) that is (a1 , . . . , ak ) is a minimal primary decomposition and pi := ai . if a q then by (i) there exists some ai iso(a) such that ai q. Then a ai pi = ai q = q. By minimality of q this yields q = pi ass(a) . conversely, if pi ass then a ai pi . Thus if we consider any prime ideal p i R such that a p pi , then by (i) there is some pj ass(a) such that pj p pi . By minimality of pi in ass(a) this means pj = pi and hence p = pi . Hence pi is a minimal prime ideal containing a. (iii) This follows immediately from (ii) and (2.20.(i)): by the propositions cited, we have a = {q spec R | a q} = ass(a) . And it is clear that ass(a) ass(a) , as ass(a) ass(a). But on the other hand, if i ass(a) there some i ass(a) such that i i (as ass(a) is nite). Thus we also have ass(a) ass(a). P Proof of (2.87): We will rst prove the truly is a homomorphism of rings. By denition we have (0) = 0R and (1) = 1R . If now k N then (k ) = k (1R ) = (k 1R ) = (k ) by general distributivity. Hence we have (k ) = (k ) for any k Z (because if k < 0 then k N and hence (k ) = ( k ) = (k ) = (k )). This will be used to prove the additivity of . First of all it is clear that for i, j N we get (i + j ) = (i + j )1R = (i1R ) + (j 1R ) = (i) + (j ) And likewise ((i) + (j )) = ((i + j )) = (i + j ) = ( (i) + (j )) = (i)+ (j ) = (i)+ (j ). Now assume i j then by general associativity it is clear that (i + (j )) = (i j ) = (i j )1R = (i1R ) (j 1R ) = (i) (j ) = (i)+ (j ). And if i j then (i +(j )) = ((j i)) = (j i) = (j ) + (i) = (j ) + (i) = (i) + (j ). Thus in any case we have (i+(j )) = (i)+ (j ) and likewise we see ((i)+j ) = (i)+ (j ). Thus we have nally arrived at (i + j ) = (i)+ (j ) for any i, j Z. For the multiplicativity things are considerably simpler: let i, j N again, then by the general rule of distributivity we get (ij ) = (ij )1R = (i1R )(j 1R ) = (i) (j ). 448 18 Proofs - Commutative Algebra

And as also (k ) = (k ) this clearly yields (ij ) = (i) (j ) for any i, j Z. Thus is a homomorphism of rings. And if : Z R is any homomorphism, then (1) = 1R . By induction on k N we see (k ) = (1 + + 1) = (1) + + (1) = 1R + + 1R = k 1R = (k ). And of course (k ) = (k ) = (k ) = (k ) such that = . This also is the uniqueness of . Thus im( ) is a subring of R by virtue of (1.71.(v)). Let us denote the intersection of all subrings of R by P - note that this is a subring again, due to (1.47). From the construction of P it is clear, that P im( ) r R. But as 1R P we have and hence (k ) = k 1R P f0r any k N. Likewise 1R P and hence (k ) = k (1R ) P . This is the converse inclusion im( ) P and hence im( ) = P . Now kn( ) i Z is an ideal in Z due to (1.71.(v)) again. But as (Z, ) is an Euclidean domain Z is a PID due to (2.65.(iii)). Hence there is some n Z such that kn( ) = nZ. Thereby n is determined up to associateness. And as Z = {1, 1} this is n is uniquely determined up to sign n. Thus by xing n 0 it is uniquely determined. Now kn( ) = { k Z | (k ) = 0R } = { k | k N, (k ) = 0R } = { k | k N, k 1R = 0R } Thus if n = 0 then kn( ) = 0 and hence k 1R = 0R implies k kn( ) = 0 which is k = 0. And if n = 0 then kn( ) = 0. And hence we have seen in (2.65.(iii)) that (n) = |n| = n is minimal in kn( ). And due to the explict realisation of the kernel above this is n = min{ 1 k N | k 1R = 0 }. P Proof of (2.89): (i) As R is an integral domain, so are its subrings P r R. In particular im( ) r R is an integral domain. But by denition and (1.77.(ii)) weve got the isomorphy (where n := charR) Z Z nZ = kn( ) =r im( ) = prr R r R

Hence Z/nZ is an integral domain, too and by (2.9) this implies that nZ = Z or nZ is a prime ideal in Z. If we had nZ = Z, then n = 1, which would mean 1R = 0R in contradiction to R = 0. This only leaves, that nZ is a prime ideal. But this again means that n = 0 or that n is prime because of (2.48.(ii)). (ii) Suppose n := charR = 0 and kR = 0R , then k kn( ) = nZ = 0 which is k = 0. Conversely suppose n nZ = kn( ), that is nR = (n) = 0R and by assumption this implies n = 0. We now prove the second equivalency. That is we assume n = 0 and want to show :Z =r prr R. But as in (i) we nd the following isomorphies Z =r Z 0Z = Z kn( ) =r im( ) = prr R where k k + 0Z = k + kn( ) (k ) - which precisely is again. Conversely suppose : Z =r prr R for some isomorphism . If we 449

had n = 0, then Zn = Z/nZ would be nite. But as we have the isomorphy Z =r prr R =r Zn once more this cannot be (as we thereby nd the contradiction = #Z = #Zn = n < ). (iii) If n := charR = 0, then we have already proved the minimality of n, that is n = min{ 1 k N | kR = 0R } in (2.87). Conversely suppoese n is this minimum and denote c := charR. As cR = 0R and n is minimal with this property we have n c. On the other hand we have (n) = nR = 0R and hence n kn( ) = cZ. That is c | n and hence c n, altogether n = c = charR. And n = 0 is trivial (as n is contained in a subset of {k | 1 k N}). The second equivalency results from the following isomorphy again (refer to (1.77.(ii)) Zn = Z nZ = Z kn( ) =r im( ) = prr R where k + nZ = k + kn( ) (k ). Thus if charR = n = 0 we are done already. Conversely suppose there was some isomorphism such that : Zn =r prr Z. Then we had (where c := charR again) Zc = Z kn( ) =r im( ) = prr R =r Zn In particular we nd c = #Zc = #Zn = n (as n = 0 and hence c = 0). That is we have found n = c = charR and n = 0 has been assumed. (iv) If n := charR = 0, then by (ii) we have : Z =r prr R. And as prf R is the quotient eld of prr R this implies Q =r prf R under the isomorphy a/b (a) (b)1 . Likewise if n := charR = 0 then by (iii) we have Zn =r prr R. As R is a eld it is a non-zero integral domain. Thus by (i) n is prime. Hence nZ is a non-zero prime ideal and hence maximal (due to (2.65.(i))), such that Zn is a eld. As prf R is the quotient eld of (the eld) prr R this implies Zn =r prr R = prf R. Conversely if Q =r prf R, then charR = 0 as else Zn =r prf R =r Q (where n = charR = 0) by what we have just proved. And if : Zn =r prf R for some isomorphism , then we have nR = n1R = n(1 + nZ) = (n + nZ) = (0 + nZ) = 0R . Thus if c := charR, then n kn( ) = cZ and hence c | n. On the other hand we have 0R = cR = c1R = c(1 + nZ) = (c + nZ). From the injectivity of we get c + nZ = 0 + nZ and hence n | c. As both n and c are positive this only leaves n = c. (v) First suppose we had c := charR = 0, then : Z R would be injective (as kn( ) = cZ = 0), which contradicts R being nite. Thus we have c = 0, now let P := prr R, then by (iii) we have P =r Zc and hence #P = #Zc = c. But - being a subring - P g R is a subgroup (under the addition + on R). Thus by the theorem of Lagrange (1.7) we have c = #P | #R. P Proof of (2.90): 450 18 Proofs - Commutative Algebra

(i) Consider a unit a R of R and suppose ab = 0, then b = a1 0 = 0 and hence a is no zero-divisor of R. Conversely suppose that a zdR is a non-zero-divisor of R. Then the mapping R R : b ab is injective (as ab = ac implies a(b c) = 0 and as a zdR this yields b c = 0 such that b = c). But as R is nite any injective map on R also is surjective (and vice versa). Hence there is some b R such that ab = 1. But this means that a R is a unit of R. (ii) If f = pa then we easily have (f + pn S )(pn1 + pn S ) = pn a + pn S = 0 + pn S . And as S is an integral domain we have pn | pn1 or in other words pn1 + pn S = 0 + pn S . And hence f + pn S is a zerodivisor of S/pn S . Conversely suppose (f + pn S )(g + pn S ) = 0 + pn S for some g + pn S = 0 + pn S . This means pn | g and hence we obtain a well-dened l 1 . . . n 1 by letting l := g [p] := max{ k N | pk | g } If we now write g = pl b then by assumption we have pn | f g = f pl b and hence p | pnl | f b. As p is prime we get p | f or p | b. But as l has been chosen maximally p | b is absurd, which only leaves p | f . (iii) As R = 0 (because of pn S pS = S ) we have already seen the inclusions nilR zdR R \ R in (1.43.(ii)). And by (ii) we already know zdR = (p + pn S )R. And if pa + pn S zdR then (pa + pn S )n = pn an + pn S = 0 + pn S which also proves zdR nil(R). Thus it only remains to prove the inclusion R \ R (p + pn S )R. Thus consider x = f + pn S R , in particular we have f S (as else x1 = f 1 + pn S ). Thus f has a true (i.e. r 1) prime factorisation
r

f = q where q :=
i=1

qi

with S and qi S prime. Now suppose that p is not associated to any of the qi (formally that is (p qi ) for any i 1 . . . r) then we had gcd(pn , q ) = 1. The lemma of Bezout (2.68.(iii)) then implies pn S + qS = gcd(pn , q )S = S Therefore there are a and b S such that apn + bq = 1 or in other words bq = 1 apn . Now let g := apn + 1 b, then it is clear that f g = q (apn + 1 b) = (aq a)pn + 1 And hence (f + pn S )(g + pn S ) = 1 + pn S which implies x1 = g + pn S and in particular x R , a contradiction. Hence there has to be some i 1 . . . r such that p qi . And therefore we nd p qi | f such 451

that p | f , or in other words again f (p + pn S )R. Thus we got nilR = zdR = R \ R = (p + pn S ) R Now consider any prime ideal p i R, then it is clear that nilR p R \ R . But by the equality nilR = R \ R this means p = nilR. Therefore nilR is the one and only prime ideal of R. But nilR already is maximal - if nilR a i R, then there is some a R and hence a = R. Thus we have also proved the second equality spec R = smax R = { nilR } (iv) We will poove the claim in several steps: (1) given any commutative ring R and any ideal a i R let us rst denote the following set Xn (R, a) := { p spec R | a p, #R/p n } Then it is clear that a b = Xn (R, b) Xn (R, a) (because if we are given p Xn (R, b), then a b p and hence a p, such that p Xn (R, a)). Further Xn (R, a b) Xn (R, a) Xn (R, b) (because if we are given p Xn (R, a b) then a b p and as p is prime this implies a p or b p such that p Xn (R, a) or p Xn (R, b)). Now suppose a b i R, then we will prove #Xn (R, b) = #Xn R a, b a This is clear once we prove that p p/a is a bijection from Xn (R, b) to Xn (R/a, b/a). But by the correspondence theorem (1.61) (and the remarks following it) p i R/a is prime if and only if p = p/a for some p i R prime, with a p. And due to the correspondence theorem we also have b p if and only if b/a p/a. Finally weve got the third isomorphism theorem (1.77.(iv)) which provides the isomorphy R/p =r (R/a)/(p/a). In particular we have #R/p = #(R/a)/(p/a). Thus p p/a is well-dened and surjective. But the injectivity is clear (by the correspondence theorem) and hence it is bijective. In a second step (2) let us now denote the set A := { a i R | #Xn (R, a) = } Suppose A = was non-empty, then (as R is noetherian) there would be a maximal element a A . We now claim that this maximal element a is prime. It is clear, that a = R, as R/R = 0 has not a single prime ideal, in particular #Xn (R, R) = 0. Thus suppose there would be some a, b R such that ab a but a, b a . Then we let a := a + aR and b := a + bR and thereby get a a, b and a b a . Thus by the inclusions in (1) we would nd Xn (R, a ) Xn (R, a b) Xn (R, a) Xn (R, b) But as a is maximal in A we have a, b A. That is both Xn (R, a) and Xn (R, b) are nite. Thus by the above inclusion Xn (R, a ) is 452 18 Proofs - Commutative Algebra

nite, too, in contradiction to a A - thus a is prime. For the third step (3) let Q := R/a , then by (2) Q is an integral domain. And if 0 = u i Q is a non-zero ideal, then by the correspondence therorem there is some ideal b i R with a b such that u = b/a . But as a has been maximal the equality in (1) yields #Xn (Q, u) = Xn R a , b a = #Xn (R, b) <

(4) the same reasonin yields #Xn (Q, 0) = #Xn (R, a ) = , such that Q is innite (if it was nite, that so was its spectrum and hence the subset Xn (Q, 0) was nite as well). Thus we may choose pairwise distince elements f1 , . . . , fn+1 Q and let f :=
i=j

(fi fj ) Q

as fi fj = 0 for any i = j and Q is an integral domain we have f = 0 as well. Now consider any prime ideal q i Q, if we have f q then (as q is an ideal) for any i = j 1 . . . n + 1 we have fi fj q. Thus for any i = j 1 . . . n + 1 we have fi + q = fj + q. That is Q/q contains (at least) n + 1 distinct elements and hence q Xn (Q, 0). Thus for any q Xn (Q, 0) we necessarily have f q and hence Xn (Q, 0) = Xn (Q, f Q) But as f = 0 we have f Q = 0 such that (3) yields #Xn (Q, f Q) < . By the equalities above this means = #Xn (R, a ) = #Xn (Q, 0) = Xn (Q, f Q) < a contradiction. This means that our assumption A = has to be false. In other words A = and in particular 0 A. That is #Xn (R, 0) < which is a trivial reformulation of the claim. P Proof of (2.92): The implications (a) = (b) and (a) = (c) are trivial. Thus it only remains to check the converse implications. The proof of (c) = (a) is elementary and has been given in (2.90.(i)). Yet (b) = (a) requires knowledge of group actions (which will be presented in book 2, or any other standard text on algebra) and cyclotomic elds (which also will be presented in book 2, or any standard text on Galois theory). Never the less we want to give a proof of (b) = (a) and the reader is asked to refer to the sources given for the statements not contained in this text: Let us regard the multiplicative group F = F \ { 0 } (this truly is a group, as F is a skew-eld) of F . And let us denote the center of the ring F by C := { a F | x F : ax = xa } It is obvious that C r F is a commutative subring of F and hence it even is a eld. Thus (F, +) is a C -vectorspace under its own multiplication. And as F is nite it clearly has to be nite-dimensional n := dimC F < . If we 453

denote q := #C then this implies #F = #F 1 = #C n 1 = q n 1 For any x E let us now denote the conjugacy class, the centralizer and the centralizer appended by 0, using the notations con(x) := yxy 1 | y E

cen(x) := { y E | xy = yx } C (x) := { y E | xy = yx } It is well-known from group theory that the conjugacy classes con(x) form a partition of F and that con(a) = { a } for any a C . Thus let us denote by X F \ C a representing system of the conjugacy classes, that is X { con(x) | x F \ C } under x con(x). As the conjugacy classes form a partition of F we nd the equality q n 1 = #F =
0=aC

#con(a) +
x X

#con(x)

= #C +
xX

#con(x) #con(x)
x X

= (q 1) +

It is clear that (C (x), +) is a subgroup of the additive group (F, +). But it even is a C -subspace of F , we denote its (nite) dimension by n(x) := dimC C (x), then we get #cen(x) = #C (x) 1 = #C n(x) 1 = q n(x) 1 We now apply the orbit formula #F = #con(x) #cen(x). Yet as #con(x) is an iteger we nd the divisibility #cen(x) | #F . Using division with remainder we obtain q n 1 : q n(x) 1 = q nn(x) + q n2n(x) + + q nr(x)n(x) As this division leaves no remainder we necessarily nd n = r(x)n(x), i.e. n(x) divides n. On the other hand the orbit formula yields #con(x) = qn 1 q n(x) 1

Substituting this into our partition formula for the cardinality of F we nd q n 1 = (q 1) +

xX

qn 1 q n(x) 1

This is the point where the cyclotomic polynomials m Z[t] come into play. Let m N and m := exp(2i/m) C, further denote k m : k 1 . . . m and gcd(k, m) = 1. Then the m-th cyclotomic polynomial is

454

18 Proofs - Commutative Algebra

dened to be m :=
k (t m ) k m

In book 2 it will be shown that m Z[t] has integer coecients in fact and (if we denote d | m : d 1 . . . m and d | m) satises d = tm 1
d|m

We now assume F = C (i.e. F is not commutative) and regard any x X . It is clear that we nd n (q ) | q n 1. But as x C we get #con(x) = 1 and hence n = n(x) on the other hand we have seen n(x) | n and hence (tn(x) 1)n =
d | n(x)

d n
d|n

d = tn 1

Thus we get n (q ) | (q n 1)/(q n(x) 1), too and substituting this and n (q ) | q n 1 into the equation of the partition we obtain n (q ) (q n 1)
xX

qn 1 = q1 q n(x) 1

In particular we get n (q ) q 1. But this clearly cannot be - as q 2 (at k| least 0 and 1 C ) we see by the triange inequality that |q 1| < |q wn for any k 1 . . . n 1. Hence we arrived at a contradiction (to C = F ) q 1 |n (q )| =
k n k | q n | > k n

|q 1| q 1 P

Proof of (2.94): (i) R and nzdR are multiplicatively closed, due to (1.43). And if conversely uv R , then there is some a R such that 1 = a(uv ) = u(av ). Hence we also have u R . Likewise, if uv nzdR and a R then au = 0 implies auv = 0 and hence a = 0 such that u nzdR. (iii) 1 = 1 1 U V is clear and if u, u U and v , v V , then for any uv and u v U V we get (uv )(u v ) = (uu )(vv ) U V again. (iv) Immediate: if a and b a then (1 + a)(1 + b) = 1 + (a + b + ab) 1 + a. (v) Clearly 1 (U ), as 1 U and (1) = 1. And if p, q (U ) then there are u, v U such that p = (u) and q = (v ). Hence pq = (u)(v ) = (uv ) (U ) since uv U . Likewise 1 1 (V ) is clear, as (1) = 1 V . And if u, v 1 (V ) then (u), (v ) V . thereby we get uv 1 (V ) from (uv ) = (u)(v ) V . So let us nally suppose V is saturated and uv 1 (V ). Then (uv ) = (u)(v ) V implies (u) V and hence u 1 (V ), as well. (vi) As U = there is some 1 U U and hence 1 U . And if now uv U then there are some U , V U such that u U and v V . 455

As U is a chain we may assume U V and hence u V . This yields uv V and hence uv U . (vii) Since 1 U for any U U we also have 1 U . And if u, v U , then u, v U for any U U . Thereby uv U , as U is multiplicatively closed and hence uv U . Likewise if any U U is saturated, then uv U implies uv U and hence u U for any U U . And the latter translates into u U again. (viii) Clearly 1 = u0 U and if ui and uj U then also ui uj = ui+j U . That is U is a multiplicatively closed set with u U . And if conversely V R is any multiplicatively closed set with u V , then by induction on k we also nd uk V and hence U V . Together this proves, that U is the smallest multiplicatively closed set containing u. (ix) Let us dene U to be the set given, then we need to show that U is the intersection of all saturated multiplicatively closed subsets V R containing U . : Thus consider any saturated multiplicatively closed V R with U V . If now a U then there are b R and u, v U such that uab = uv U V . In particular a(ub) = uab V and as V is saturated this implies a V . Hence U V is contained in any such V . : If a U that 1a1 = 1a and hence a U . Hence we have U U so it remains to prove that U is saturated multiplicatively closed. Thus consider a and a U , that is uab = uv and u a b = u v for sucient b, b R and u, v , u , v U . Then (uu )aa (bb ) = (uu )(vv ) and hence aa U , as uu , vv U . Conversely suppose aa U , that is u(aa )b = uv for sucient b R, u, v U . Then ua(a b) = u(aa )b = uv and hence a U . P Proof of (2.95): We rst prove (b) = (a): consider any prime ideal p i R, then 1 R \ p, as 1 p would imply p = R which is disallowed. And as p is both, prime and an ideal we get the following equivalence for all u, v R uv p u p or v p

Hence if we go to complements (i.e. if we negate this equivelency) then we nd that R \ p is a saturated multiplicative set, since uv R \ p u R \ p or v R \ p

And from this we nd that R \ P is a saturated multiplicative set, as it is the intersection of such sets R \ p for p P , formally R\ P = R\
pP

p =
pP

R\p

Next we prove (a) = (a): x a R \ U then also aR R \ U : consider b R, then ab U - because of the saturation of U - would imply a U , in contradiction to a U . And as U is multiplicatively closed we max choose 456 18 Proofs - Commutative Algebra

an ideal p(a) maximal among the ideals b with aR b R \ U . And this p(a) is prime (for all this refer to (2.14.(iv))). Thus we get R\U
aU

aR
aU

p(a) R \ U

U = R\

{ p(a) | a U } P

Proof of (2.96): We will rst prove that is an equivalence relation: the reexivity is due to 1 U , as u1a = u1a implies (u, a) (u, a). For the symmetry suppose (u, a) (v, b), i.e. there is some w U such that vwa = uwb. As = is symmetric we get uwb = vwa and this is (v, b) (u, a) again. To nally prove the transitivity we regard (u, a) (v, b) and (v, b) (w, c). That is there are p, q U such that pva = pub and qwb = qvc. Since p, q and v U we also have pqv U . Now compute (pqv )wa = qw(pva) = qw(pub) = pu(qwb) = pu(qvc) = (pqv )uc, this yields that truly (u, a) (w, c) are equivalent, too. Next we need to check that the operations + and are well-dened. Suppose a/u = a /u and b/v = b /v that is there are p and q U such that pu a = pua and qv b = qvb. Then we let w := pq U and compute u v w(av + bu) = vv q (pu a) + uu p(qv b) = vv q (pua ) + uu p(qvb ) = uvw(a v + u b ) u v wab = (pu a)(qv b) = (pua )(qvb ) = uvwa b That is (a/u) + (b/v ) = (av + bu)/uv = (a v + b u )/u v = (a /u ) + (b /v ) and (a/u)(b/v ) = ab/uv = a b /u v = (a /u )(b /v ) respectively. And this is just the well-denedness of sum and product. Next it would be due to verify that U 1 R thereby becomes a commutative ring. That is we have to check that + and are associative, commutative and satisfy the distributivity law. Further that for any a/u U 1 R we have a/u + 0/1 = a/u and a/u + (a)/u = 0/1 and nally that a/u 1/1 = a/u. But this can all be done in straightforward computations, that we wish to omit here. It is obvious that thereby becomes a homomorphism of rings, (0) = 0/1 is the zero-element and (1) = 1/1 is the unit element. Further (a + b) = (a + b)/1 = (a1 + b1)/1 1 = a/1 + b/1 = (a) + (b) and (ab) = ab/1 = ab/1 1 = (a/1)(b/1) = (a)(b). Next we wish to prove that is injective, if and only if U nzdR. 457

To do this we take a look at the kernel of kn() = { a R | a/1 = 0/1 } = { a R | u U : au = 0 } By (1.74.(i)) is injective i kn() = { 0 }. That is i for any u U we get au = 0 = a = 0. And this is just a reformulation of U nzdR. So it remains to prove that is bijective if and only if U R . First suppose U R nzdR by (1.43). Then as we have just seen is injective. Now consider any a/u U 1 R, then a au1 u au1 = = = (au1 ) im() u u 1 Hence we found that also is surjective. Conversely suppose that is bijective. In particular is injective and hence U nzdR. But also is surjective, that is for any u U there is some b R such that 1/u = (b) = b/1. That is there is some w U such that w = uwb. Equivalently w(ub 1) = 0, but as w nzdR this implies ub = 1 and hence u R . As u has been arbitrary we found U R . P Proof of (2.99): Two of the three implications are clear, thus we are only concerned with: if for any maximal ideal m i R we have a/1 = b/1 Rm then already a = b R. But since a/1 = b/1 is equivalent to (a b)/1 = 0/1 it suces m : a 0 = Rm = a = 0 1 1

By denition a/1 = 0/1 means that there is some u m such that ua = 0. And the latter can again be formulated as u ann(a). Thus we have m : ann(a) / m Hence there is no maximal ideal m cntaining ann(a). But as the annulator ann(a) is an ideal this is only possible if ann(a) = R and hence a = 1 a = 0. In the second claim it is clear that for any prime ideal p i R we get R Rp F := quotR. Hence it is clear that R p Rp . And as smax R spec R it also is clear that p Rp m Rm such that we only have to verify m Rm R. Thus consider any a/b F with a/b m Rm . That is for any maximal ideal m i R ther are some a(m) R and b(m) m auch that a/b = a(m)/b(m). Now dene the ideal
b :=

b(m) | m smax R

Suppose we had b = R, then there would be some maximal ideal m such that m m. But clearly b(m) b although b(m) m, a contradiction. That is b = R and hence 1 b. That is there is a nite subset M smax R and there are r(m) R (where m M ) such that 1 = m r(m)b(m). Now recall 458 18 Proofs - Commutative Algebra

that a/b = a(m)/b(m), then we nd a/b R by virtue of a a = 1 = b b r(m)b(m)

mM

a = b

r(m)a(m) R
mM

P Proof of (2.100): In the rst part we will prove the existance of the induced homomorphism . And of course we start with the well-denedness of as a mapping. Thus let us regard a/u = b/v U 1 R. That is there is some w U such that vwa = uwb. And thereby we get (a)(u)1 = (b)(v )1 from (v )(w)(a) = (vwa) = (uwb) = (u)(w)(b) Next we have to check that truly is a homomorphism of rings. To do this consider any a/u and b/v U 1 R and compute a b u v = = a b + u v ab uv = (a)(b)(u)1 (v )1 (b)(v )1 = = a u b v

(a)(u)1 av + bu uv

(a)(v ) + (b)(u) (u)1 (v )1 a + u b v

= (a)(u)1 + (b)(v )1 =

So we have proved the existence of the homomorphism . It remains to verify that this is the unique homomorphism such that = . That is we consider some homomorphism : U 1 R S such that for any a R we get (a/1) = (a). Then for any u U R we get 1 u = u 1
1

u 1

= (u)1

And thereby we nd for any a/u U 1 R that (a/u) = (a/1 1/u) = (a/1) (1/u) = (a)(u)1 = . That is = , that is is unique. P Proof of (2.102): By denition 1 (u) is the set of all a R such that a/1 = (a) u, and this has already been the rst claim. And 1 (a) is an ideal of R by virtue of (1.71.(vii)). If a a then a/1 = (a) (a) U 1 a. And for any u U we hence also get a/u = (1/u)(a/1) U 1 a. Conversely consider any x (a) i . I.e. there are ai a and xi = bi /vi U 1 R such that
n n

x =
i=1

xi (ai ) =
i=1

ai bi vi 459

We now let v := v1 . . . vn U and vi := v/vi U then a straightforward computation shows x = a/v for some a := i ai bi vi a
n

x =
i=1

ai bi = vi

n i=1

ai bi vi a = v v P

Proof of (2.104): First of all a : U is an ideal - 0 a : U since 0 = 1 0 a. Now consider a, b a : U . That is there are some u, v V such that ua and vb a. This yields uv (a + b) = v (ua) + u(vb) a and hence a + b a : U . Likewise for any r R we get u(ra) = r(ua) a and hence ra a : U . Next it is clear that a a : U , because if a a then a = 1 a a (due to 1 U ) and hence a a : U . We nally wish to prove (U 1 a) R = a : U . Consider any b a : U , that is vb a for some v U . Then b/1 = vb/v U 1 a and hence b (U 1 a) R. Conversely let b (U 1 a) R, by denition this means b/1 U 1 a. That is there is some a a, u U such that b/1 = a/u. That is there is some v U such that uvb = va a. And as uv U this means b a : U . P Proof of (2.105): U 1 (a b) = (U 1 a) (U 1 b): if x U 1 (a b) then by denition there are some a a b and u U such that x = a/u. But as a a and a b this already means x U 1 a and x U 1 b. If conversely x = a/u U 1 a and x = b/v U 1 b then a/u = b/v means that there is some w U such that vwa = uwb a b. Hence x = (vwa)/(uvw) = (uwb)/(uvw) U 1 (a b). U 1 (a + b) = (U 1 a) + (U 1 b): if x U 1 (a + b) then there are some a a, b b and u U such that x = (a + b)/u = (a/u) + (b/u) (U 1 a) + (U 1 b). And if we are conversely given any a/u U 1 a and b/v U 1 b then a/u + b/v = (au + bv )/(uv ) U 1 (a + b). U 1 (a b) = (U 1 a) (U 1 b): if x U 1 (a b) then there are some f a b and u U such that x = f /u. Again f is of the form f = i ai bi for some ai a and bi b. Altogether we get f x = = u
n i=1

ai bi = u

n i=1

ai bi 1 u

U 1 a

U 1 b

Conversely consider ai /ui U 1 a and bi /vi U 1 b. As always we let u := u1 . . . un and ui := u/ui U again, likewise v := v1 . . . vn and vi := v/vi U . Then ai /ui = (ai ui )/u, bi /vi = (bi vi )/v and
n i=1

ai bi = ui vi

n i=1

ai ui bi vi 1 = u v uv

ui ai vi ai bi U 1 a b
i=1

460

18 Proofs - Commutative Algebra

 U 1 a = U 1 a: if x U 1 a then x = b/v for some b a and v U . That is there is some k N such that bk a. And therefore xk = bk /v k U 1 a. Hence x is contained in the radical of U 1 a. Conversely consider any x = b/v such that there is some k N with xk = bk /v k U 1 a. Then there is some a a and u U such that bk /v k = a/u. That is uwbk = v k wa a for some w U . Hence k k 1 k (uwb) = (uw) (uwb ) a, that is uwb a. And this agian yields x = b/v = (uwb)/(uvw) U 1 a. Next we prove (a : U ) : U = a : U . Clearly, as 1 U we have a : U (a : U ) : U (as for any a a : U we have a = a 1 (a : U ) : U ). Conversely if a (a : U ) : U then there is some u U such that au a : U . Again this means that there is some v U such that a(uv ) = (au)v a. But as uv U this again yields a a : U . For (a b) : U = (a : U ) (b : U ) we use a straightforward reasoning again: If a (a b) : U then there is some u U such that au a b. And this again means a a : U and a b : U . Conversely consider a (a : U ) (b : U ). That is there are v , w U such that av a and aw b. Now let u := vw U then au = (av )w = (aw)v a b such that a (a b) : U . We turn our attention to a : U = a : U . If a : U = R then we get a : U = R = R. And as a a we also have R = a : U a : U R. And hence we get a : U = R = a : U . Thus assume a : U = R. If x a : U then by denition there is some k N such that xk a : U . And this means xk u a for some u U . Clearly we have k 1 as else u a such that a : U = R. And thereby we k k k 1 get ( xu) = (x u)u a, too. But this means xu a and hence x a:U . Conversely if x a : U then there is some u U such k k k that xu a. Thus there is some k N such that x u = (xu) a. k k But as u U this yields x a : U and hence x a : U . (u w) R = 1 (u w) = 1 (u) 1 (w) = (u R) (w R) = a b. (u + w) R = (a + b) : U : if c (a + b) : U then there are some a a, b b and u U such that uc = a + b. That is (uc)/1 = (a + b)/1 = (a/1) + (b/1) u + w. And thereby also c/1 = (1/u)((uc)/1) u + w. But this nally is c (u + w) R. Conversely let c (u + w) R, that is c/1 u + w. Thereby we nd a/u u and b/v w such that c/1 = a/u + b/v = (va + ub)/uv . Hence there is some w U such that uvwc = vwa + uwb a + b (as a/u u implies a/1 = (u/1)(a/u) u and hence a a, likewise b b). And as uvw U this is c (a + b) : U . (u w) R = (a b) : U : if c (a b) : U then there are some ai a, bi b and u U such that uc = i ai bi a b. Thereby (uc)/1 = i (ai /1)(bi /1) u w and hence c/1 = (1/u)((uc)/1) u w which means c (u w) R. And if conversely c (u w) R then there are ai /ui u and bi /vi w such that c = 1
n i=1

ai bi = ui vi

n i=1

ai bi ui vi uw uv 461

where as usual u := u1 . . . un , v := v1 . . . vn and ui := u/ui , vi := v/vi U . From this equality we again get uvwc = i ai bi ui vi a b for some w U (as ai /ui u implies ai /1 = (ui /1)(ai /ui ) u and hence ai a, likewise bi b). And as uvw U this is c (a b) : U . u R = a: if b u R then b/1 u, that is there is some k k k N such this is bk u R = a and that (b/1) = b /1 u. Again k hence b a. Conversely if b a then b a for some k N. Hence k k (b/1) = b /1 u which yields b/1 u and thereby b u R. P Proof of (2.107): Nearly all of the statements are purely denitional. The two exceptions to this are the equalities claimed for U 1 spec R and U 1 smax R: Consider any prime ideal p i R with p U = . If now ua p then u p or a p, as p is prime. But u p is absurd, due to p U = . This only leaves a p and hence p U 1 ideal R. Conversely let p i R be a prime ideal with p = p : U . If now u 1 = u p U then 1 p : U = p in contradiction to p = R. Hence we get p U = . Now recall that any maximal ideal is prime. Then repeating the arguments above we also nd m = m : U m U = for any maximal ideal m i R. P Proof of (2.108): First of all it is clear that the mapping u u R = 1 (u) maps ideals to ideals, as it is just the pull back along the homomorphism . Now suppose that ua u R - this is ua/1 u and hence a/1 = (1/u)(ua/1) u. But this again implies a u R and hence u R U 1 ideal R. Conversely for any ideal a i R the set U 1 a is an ideal of U 1 R by (2.102). Thus both of the mappings are well-dened. Next we remark that for any a U 1 ideal R and u i U 1 R we obtain (U 1 a) R = a : U = a U 1 (u R) = { a/u | a/1 u, u U } = u Thereby the rst equality follows from (2.104) and the denition of U 1 ideal R. And for the second equality consider a/u where a/1 u. Then we nd that a/u = (1/u)(a/1) u as well. And if conversely a/u u then we likewise get a/1 = (u/1)(a/u) u again. Altogether these equalities yield that the maps given are mutually inverse. If a b and we consider any a/u U 1 a then a a b and hence a/u U 1 b. And if conversely u w and a u R then a/1 u w such that a w R. That is both of the mappings are order preserving. Next we want to prove that this correspondence preserves radical ideals. Thus if a U 1 ideal R is a radical ideal then we have a = a and 462 18 Proofs - Commutative Algebra

thereby U 1 a is a radical ideal due to (2.105) (see below). Likewise we see that if u ideal U 1 R is a radical ideal, then u R is radical, too: U 1 a = U 1 a = U 1 a uR = uR = uR Thus it only remains to verify that prime (resp. maximal) ideals are truly correspond to prime (resp. maximal) ideals. For prime ideals this is an easy, straightforward computation: ab/uv U 1 p implies ab p thus a p or b p which again is a/u U 1 p or b/v U 1 p. Conversely ab r R implies ab/1 r and hence a/1 r R or b/1 r R. And for maximal ideals this is clear, as the mappings given obviously preserve inclusions. P Proof of (2.112): (i) We will rst prove the injectivity of the map given: thus suppose f /u and g/v are mapped to the same point in (U 1 R)[t]. Comparing the coecients for these polynomials we nd that f []/u = g []/v U 1 R for any N. That is there is some w() U such that vw()f [] = uw()g []. Without loss of generality assume deg f deg g and let w := w(0)w(1) . . . w(deg g ) U . Then clearly also vwf [] = uwg [] for any 0 . . . deg g and hence even for any N. This now yields

vwf (t) =
=0

vwf []t =
=0

uwg []t = uwg (t)

And hence we have just seen that f /u = g/v as elements of U 1 (R[t]). For the surjectivety we are given a polynomial f in (U 1 R)[t]

f =
=0

f [] t u()

Let now u := u(0)u(1) . . . u(deg(f )) U and u() := u/u() R, then it is clear that we get f /u f for the polynomial

f :=
=0

u()f []t

(ii) We regard the homomorphism R[t] Ru : f f (1/u). It is clear that this is surjective, since atk a/uk reaches all the elements of Ru . We will now show that the kernel is given to be kn(f f (1/u)) = (ut 1)R[t] The inclusion is clear, as ut 1 u/u 1/1 = 0/1. Thus suppose f 0/1, then we need to show that ut 1 | f in R[t]. Let f /1 := (f ) (U 1 R)[t] then it is clear that 1/u is a root (f /1)(1/u) = 463

f (1/u) = 0/1 of f /1 and hence t 1/u | f /1 in (U 1 R)[t]. Since u/1 is a unit in Ru multiplication with u/1 changes nothing and hence (ut 1)/1 | f /1 (U 1 R)[t]. Thus f 1 (U R)[t] 1 ut 1 1 (U R)[t] 1

But it is clear that the ideal f /1(U 1 R)[t] corresponds to f R[t] formally (f /1(U 1 R)[t]) R[t] = f R[t]. And likewise the other ideal (ut 1)/1f /1(U 1 R)[t] corresponds to (ut 1)R[t]. As the correspondence respects inclusions we have nally found f R[t] (ut 1)R[t] This means f (ut 1)R[t] and hence we have also proved . Thus f f (1/u) induces the isomorphism given its inverse is obviously given by a/uk atk + (ut 1)R[t].

(iii) First of all U 1 R is an integral domain by (2.110) and hence the quotient eld quotU 1 R is well-dened again. We will now prove the well-denedness and injectivity of the map a/b (a/1)/(b/1). As R is an integral domain (and 0 U ) we get (for any a, b, c and d R) a/1 c/1 = b/1 d/1 ad a = 1 1 u U ad = bc c a = b d d b c bc = = 1 1 1 1 : uad = ubv

The homomorphism property of a/b (a/1)/(b/1) is clear by definition of the addition and multiplication in localisations. Thus it only remains to check the surjectivity: consider any (a/u)/(b/v ) quotU 1 R, then we claim av/bu (a/u)/(b/v ). To see this we have to show (av/1)/(bu/1) = (a/u)/(b/v ) and this follows from: a bu abu ab abv b av = = = = u 1 u 1 v v 1

(iv) We rst check the well-denedness and injectivity of the map given. This can be done simultaneously by th following computation (recall 464 18 Proofs - Commutative Algebra

that R was assumed to be an integral domain and a, b = 0) x/ah y/aj = (b/an )i (b/an )k b k x b i x = an an aj ah xbk ybi = ain+j akn+h in+j k xa b = yakn+h bi ai+k bh+j xain+j bk yakn+h bi = 0 xain+i+j +k bh+j +k = yakn+h+i+k bh+i+j (ab)j +k xa(n+1)i bh = (ab)h+i ya(n+1)k bj xa(n+1)i bh ya(n+1)k bj = (ab)h+i (ab)j +k

Hence we have a well-dened injective mapping and it is straightforward to check that it even is a homomorphism of rings. Thus it only remains to check the surjectivity. That is we are given any x/(ab)k Rab , then we obtain a preimage by x/a(n+1)k xa(n+1)k b(n+1)k x = (b/an )k (ab)k (ab)n+1)k+k (v) Let us regard the map b/v (b + a)/(v + a). We rst check its welldenedness: suppose a/u = b/v U 1 R, that is there is some w U such that vwa = uwb. Of course this yields (v + a)(w + a)(a + a) = (u + a)(w + a)(b + a). But as w U we also have w + a U/a and hence (a + a)/(u + a) = (b + a)/(v + a). And it is immediately clear that this map even is a homomorphism of rings. Next we will prove its surjectivity, that is we are given any (b + a)/(v + a) (U/a)1 (R/a). As v + a U/a there is some u U such that v + a = u + a. But from this we nd b/u U 1 R and b/u (b + a)/(u + a) = (b + a)/(v + a). Thus it remains to prove kn b b+a v v+a = U 1 a = a u a a, u U

The inclusion is clear, given a/u where a a we immediately nd that (a + a)/(u + a) = (0 u + a)/(1 u + a) = (0 + a)/(1 + a) = 0. Conversely consider some b/v U 1 R such that (b + a)/(v + a) = 0 = (0 + a)/(1 + a). That is there is some x + a U/a such that xb + a = (x + a)(b + a) = 0 + a. And this means xb a. Further since x + a U/a - there is some w U such that x + a = w + a. That is a := w x a. Now compute bw = bw bx + bx = ba + bx a. Hence we found b/v = (bw)/(vw) U 1 a. Thus according to the isomorphism theorem b/v (b + a)/(v + a) induces the isomorphy as given in the claim. (vi) Let us denote U := R \ p, as a p we get U a = and ap = U 1 a is just the denition. Hence the claim is already included in (v). (vii) We rst show that this mapping is well dened, i.e. u/1 p for u p. But this is clear, since p = p R. Thus we concern ourselves with the 465

surjectivety: given (r/ak )/(u/al ) (Ra )p we see that ral /1 ral 1/ak+l ral /1 r/ak = = uak uak /1 1/ak+l uak /1 u/al And for the injectivety we have to consider (r/1)/(u/1) = (s/1)/(v/1). Hence there is some w/am p such that vwr/am = uws/am Ra . Thus there is some n N such that am+n vwr = am+n uws R. But w/am p means that w p and as also a p we get that z := am+n w p. But as vrz = usz this means that r/u = s/v Rp . P Proof of (2.110): R integral domain = U 1 R integral domain Suppose (ab)/(uv ) = (a/u)(b/v ) = 0/1 U 1 R. By denition of the localisation this means that there is some w U such that wab = 0. But w = 0 (as 0 U ) and R is an itegral domain, thus we get ab = 0. And this again yields a = 0 or b = 0. Of course this yields a/u = 0/1 or b/v = 0/1, that is U 1 R is an itegral domain. R noetherian/artinian = U 1 R noetherian/artinian Suppose that R is noetherian and consider an ascending chain of ideals u1 u2 u3 . . . in U 1 R. Let us denote the corresponding ideals of R by ak := uk R. Then the ak form an ascending chain a1 a2 a3 . . . of idelas of R, too. And as R is noetherian this means that there is some s N such that the chain stabilizes
a1 a2 a3 . . . as = as+1 = as+2 = . . .

Yet the ideals correspond to one and another, that is uk = U 1 ak . So if we transfer this stabilized chain back to U 1 R we nd that
u1 u2 u3 . . . us = us+1 = us+2 = . . .

the corresponding chain stabilized, too. And as the chain has been arbitrary this means, that U 1 R is noetherian, as well. The fact that R artinian implies U 1 R artinian can be proved in complete analogy. R PID = U 1 R PID Consider any ideal u i U 1 R, by the correspondence theorem (2.108) u corresponds to the ideal a := U R i R of R. But as R is a PID there is some a R such that a = aR. And the following computation yields that u = (a/1)U 1 R is a principal ideal, too
u = U 1 (aR) =

ab u

b R, u U

a 1 U R 1

R UFD = U 1 R UFD If 0 U then U 1 R = 0 and hence U 1 R is a UFD. Thus suppose 0 U . By denition any UFD R is an integral domain, and hence 466 18 Proofs - Commutative Algebra

U 1 R is an integral domain, too. Thus it suces to check that any a/u U 1 R admits a decomposition into prime factors. To see this we will rst prove the following assertion p R prime = p U 1 R 1

or

p U 1 R prime 1

Thus let p R be prime and assume p/1 | (a/u)(b/v ) = (ab)/(uv ). That is there is some h/w U 1 R such that ph/w = ab/uv . And as R is an integral domain this implies uvph = wab. In particular p | wab and as p is prime this means p | w or p | a or p | b. First suppose p | w, i.e. there is some R such that p = w. Then (p/1)(/w) = w/w = 1 and hence p/1 (U 1 R) is a unit. Otherwise, if p | a - say p = a - then (p/1)(/u) = a/u. That is p/1 | a/u. Analogously we nd p | b = p/1 | b/v . Thus either p/1 is a unit or prime, as claimed. Now we may easily verify that U 1 R is an UFD: consider any 0 = a/u U 1 R. As R is an UFD we nd a decomposition of a = 0 into prime elements a = p1 . . . pk (where R and pi R prime). Now let I := i I | pi /1 (U 1 R) be the set of indices i for which pi /1 is a unit. Then a p1 pk = ... = u u 1 1 u pi 1 pi 1

iI

i I

That is we have found a decomposition of a/u into (a unit and) prime factors. And as a/u has been an arbitrary non-zero element, this proves that U 1 R truly is an UFD. R normal = U 1 R normal We have already seen that U 1 R is an integral domain, since R is an integral domain. To prove that U 1 R is integrally closed in its quotient eld we will make use of the isomorphy (2.112.(iii))
quotU 1 R : : quotR

r r/1 s s/1

Recall that R is embedded canonically into quotR by a a/1. Hence quotR is considered to be an R-algebra under the scalar multiplication a(r/s) := (a/1) (r/s) = ar/s. Analogously, for any a/w U 1 R and any r/u, s/v U 1 R we have a/w r/u ar/uw a r/u = = w s/v 1/1 s/v s/v Thus if we consider any a R and any r/s quotR, then the isomorphism satises (ax) = (a/1)(x), which can be seen in a straightforward computation (ax) = = ar ar/1 a/1 r/1 = = s s/1 1/1 s/1 a r/1 a r a = = (x) 1 s/1 1 s 1 467

Now consider any x/y quotU 1 R, that is integral over U 1 R. That is there are some n N and p0 , . . . , pn1 U 1 R such that x y
n n1

+
k=0

pk

x y

= 0

Then we need to show that x/y U 1 R. To do this let us pick up some r/s quotR such that x/y = (r/s). Further note that pk U 1 R, that is pk = ak /uk for some ak R and uk U and let u := u0 . . . un1 U . Then multiplication with un /1 U 1 R yields u r 1 s
n n1

+
k=0

ak unk uk

u r 1 s

= 0

Now recall that (u/1)(r/s) = (u(r/s)) = (ur/s). Further note that due to k < n we have uk | u | unk . Therefore we may dene bk := ak unk /uk R. Inserting this in the above equation yields ur s
n n1

+
k=0

ur bk 1 s

= 0

Again we use (bk /1)(ur/s) = (bk (ur/s)) and use the homorphism properties of to place in front, obtaining an equation of the form (. . . ) = 0. But as has been an isomorphism it can be eliminated from this equation ultimatly yielding ur s
n n1

+
k=0

bk

ur s

= 0

Thus we have found that ur/s quotR is integral over R. Hence by assumption on R we have ur/s R, that is there is some a R such that ur/s = a/1. And this is ur = as such that r/s = a/u U 1 R. Thus we have nally found x r = y s = a u = a/1 a/u = U 1 R u/1 1/1

R Dedekind domain = U 1 R Dedekind domain If R is a Dedekind domain, then R is noetherian, integrally closed and spec R = { 0 } smax R. As we have seen this implies, that U 1 R is noetherian and integrally closed, too. Thus consider a non-zero prime ideal 0 = w i U 1 R. Then m := w R i R is a nonzero prime ideal of R, too (if we had m = 0 then w = U 1 m = 0, a contradiction). Thus m is maximal and hence w = mR is maximal, too. That is spec U 1 R = { 0 } smax U 1 R. Altogether U 1 R is a Dedekind domain, too. P

468

18 Proofs - Commutative Algebra

Proof of (2.111): We have already proved (a) = (b) in (2.110) and (b) = (c) is trivial. Thus it only remains to verify (c) = (a): we denote F := quotR the quotient eld of R and regard R and any localisation Rm as a subring of R (also refer to (2.99) for this). Now regard any x F integral over R, that is f (x) = 0 for some normed polynomial f R[t]. If we interpret f Rm [t], then f (x) = 0 again and hence x is integral over Rm (where m i R is any maximal ideal of R). By assumption Rm is normal and hence x Rm again. And as m has been arbitrary the local global principle (2.99) yields: x
msmaxR

Rm = R P

Proof of (2.50): (continued) (c) = (d): property (2) of (c) and (d) are identical so we only need to verify (1) in (d). If all ai = 0 are zero then there is nothing to prove (s := 0). And if at least one ar = 0 is non-zero, then so are all ar+i for i N. Omitting the a0 to ar1 we may assume ai = 0 for any i N. By assumption we have ai R ai+1 R for any i N. And this translates into ai+1 | ai . By (2.54.(iii)) this yields a descending chain of natural numbers for any prime element p R p | a0 p | a1 . . . p | ai . . . 0 Of course this means that the sequence p | ai has to become startionary. That is there is some s(p) N such that for all i N we get p | as(p)+i = p | as(p) . We now choose a representing system P of the prime elements of R modulo associateness. Then there only are nitely many p P such that p | a0 = 0. And for those p P with p | a0 = 0 we may clearly take s(p) = 0. Thereby we may dene s := max{ s(p) | p P } N Then it is clear that for any p P and any i N we get p | as+i = p | as . And by (2.54.(iv)) this is nothing but as+i as which again translates into as+i R = as R. And this is just what had to be proved. (d) = (c): property (2) of (c) and (d) are identical so we only need to verify (1) in (c). But the proof of this would be a literal repetition of (2.49.(iii)). There we have proved the existence of an irreducible decomposition using the ascending chain condition of noetherian rings. But in fact this property has only been used for principal ideals and hence is covered by the assumption (1) in (d). Hence the same proof can be applied here, too. (a) = (e): As p = 0 is non-zero, there is some 0 = a R with a p. Clearly a R as else p = R. We now use a prime decomposition of a, that is p1 . . . pk = a p. Thereby k 1 as else a R . And as p is prime we either get p or pi p for some i 1 . . . k . But again p would yield p = R which cannot be. Thus p := pi p where p is prime by construction. 469

 (e) = (f): We assume R = 0 (else the stetements herin do not make sense!). It is clear that 0 D (as R is an integral domain R and pi = 0). Therefore D R \ {0} such that we obtain the canonical homomorphism a a D1 R quotR : d d Due to 0 D we also see that D1 R = 0 is not the zero-ring. Therefore D1 R contains a maximal ideal u i D1 R - by (2.4.(iv)) - such that p := u R D1 smax R is maximal too - by (2.108) - in particular prime. Suppose p = 0 then by assumption (e) there is some prime element p p. And thereby p p D in contradiction to p D1 smax R. Thus p = 0 and therefore u = D1 p = 0, too. But 0 D1 R being maximal yields that D1 R is a eld. In particular the canonical homomorphism is injective (as it is non-zero). Now choose any a/b quotR then a, b R and hence a/1, b/1 D1 R. Yet b = 0 and hence (1/b)1 = 1/b D1 R, as D1 R is a eld. Therefore a/b D1 R such that the canonical homomorphism also is surjective. By construction this even is an equality of sets. (f) = (a): Consider any 0 = a R. Then 1/a quotR = D1 R, that is there are b R and d D such that 1/a = b/d. And as R is an integral domain this is ab = d D. Yet D is saturated by (2.48.(vi)) and therefore a D. But this just means that a admits a prime decomposition. P Proof of (2.75): We have already proved statements (i) to (vii) on page (441), so it only remains to prove statement (viii). We have already seen in (vii) that the map b b R = 1 (b) is well-dened (as 1 (q) = q R = p again). Thus we next check the well-denedness of a U 1 a. First of all we have U 1 a = U 1 a = U 1 p = q due to (2.105). Now consider a/u and b/v U 1 R such that (ab)/(uv ) = (a/u)(b/v ) U 1 a but b/v U 1 R. Then ab (U 1 a) R = a : U , that is there is some w U such that 1 a. As a is primary this abw a. But /(vw) U bw a, as else b/v = (bw) 1 implies a a = p and hence a/u U p = q = U 1 a. This means that U 1 a is primary again and hence a U 1 a is well-dened. P Proof of (2.15): (i) As the union ranges over all prime ideals p with p zdR it is clear that the union is contained in zdR again. That is the incluison is trivial. For the converse inclusion let us denote U := nzdR, then U is multiplicatively closed. Let now a zdR be a zero-divisor. That is we may choose some 0 = b R such that ab = 0. This implies aR zdR (because given any x R we get (ax)b = x(ab) = x0 = 0) or in other words aR U = . Thus by (2.4.(iii)) we may choose an ideal p maximal among the ideals satisfying aR b R \ U = zdR and this is prime. In particular a p such that a is also contained in the union of prime ideals contained in zdR. 470 18 Proofs - Commutative Algebra

(ii) Let p be a minimal prime ideal of R, then by the correspondence theorem (2.108) the spectrum of the localised ring Rp corresponds to spec Rp { q spec R | q p } { p } Hence Rp has precisely one prime ideal, namely the ideal p = (R \ p)1 p corresponding to p. And by (2.17.(v)) this yields nilRp = spec Rp = p

Now consider any a p, then a/1 p and hence there is some k N such that ak /1 = (a/1)k = 0/1. That is there is some u R \ p such that uak = 0. But k = 0 as else u = 0 p. Hence we may choose 1 k minimally such that uak = 0. Then we let b := uak1 and thereby get b = 0 (as k has been minimal) and ab = 0. That is a zdR and hence p zdR (as a has been arbitrary). P Proof of (2.113): (b) = (a): of course R = 0 cannot occur, as else R = R and hence R \ R = was no ideal of R. Thus R = 0 and hence R contains a maximal ideal, by (2.4.(iv)). And if m i R is any maximal ideal of R then we have m = R and hence m R \ R = R. By maximality of m we get m = R \ R which also is the uniqueness of the maximal ideal. (a) = (b): let m be the one and only maximal ideal of R. As m = R we again get m R \ R . Conversely, if a R \ R then aR = R. Hence there is a maximal ideal containing aR, by (2.4.(ii)). But the only maximal ideal is m such that a aR m. And as a has been arbitrary this proves R \ R m and hence m = R \ R . In particular R \ R is an ideal of R. P Proof of (2.114): (a) = (b): consider any a = R, then by (2.4.(ii)) a is contained in some maximal ideal, and the only maximal ideal available is m, such that a m. (b) = (a): rst of all m is a maximal ideal, as (due to (b)) m a i R implies m = a or a = R. And if n i R is any maximal ideal of R, then we have n = R and hence (due to (b)) n m. But by maximality of n and m = R we now conclude n = m. Altogether we have smax R = {m} as claimed. (a) = (c): has already been proved in (2.113) - in the form R \ R = m. (c) = (d) is trivial, so it only remains to prove the implication (d) = (b): consider any ideal a = R again, then we have a R \ R and due to (d) this yields a R \ R m. P Proof of (2.116): By the correspondence theorem (2.108) it is clear that mp is in fact a maximal ideal of Rp . Now suppose mp = (R \ p)1 n i Rp is any maximal ideal of Rp . In particular mp is a prime ideal and hence (by the correspondence theorem again) n (R \ p) = . In other words that is n p and hence np mp . 471

But by the maximality of mp this is np = mp . Thus mp is the one and only maximal ideal of Rp - that is Rp is a local ring. Now regard Rp quotR/p : a/u (a + p)/(u + p). This is welldened and surjective, since u + p = 0 + p if and only if u p. And since R/p is an integral domain we have (a + p)/(u + p) = (0 + p)/(1 + p) if and only if a + p = 0 + p which is equivalent to a p again. Thus the kernel of this epimorphism is just mp and hence it induces the isomorphism given. P Proof of (2.119): (U) As R = 0 we have 1 = 0 and hence by property (1) we nd (1) N. Thus by (2) we get (1) = (1 1) = (1) + (1) Z. And together this implies (1) = 0. Now consider any R then by (2) again 0 = (1) = (1 ) = () + (1 ) () 0 hence () = 0. (I) If we are given a, b R with ab = 0 then by (1) and (2) we have = (ab) = (a) = (b). Thus (a) and (b) may not both be nte and by (1) this means a = 0 or b = 0. (P) First of all 0 [ ] as by (1) we have (0) = 1 and 1 [ ] as (1) = 0 < 1 by (U). If now a, b [ ] and c R then we compute (a + b) min{ (a), (b) } 1 (ac) = (a) + (c) (a) 1 And hence a + b and ac [ ] again. Hence [ ] truly is an ideal of R and [ ] = R as 1 [ ]. Now suppose ab [ ] for some a, b R, that is (ab) = (a) + (b) 1. Then it cannot be that (a) = 0 = (b) and hence a [ ] or b [ ], which means [ ] is prime. (iii) It is clear that 1 | a for any a R and hence 0 {k N | pk | a}. Therefore a[p] N { } is well-dened. As any pk | 0 it is clear that (0) = . Conversely suppose that (a) = for some a R. Then for any k N we may choose some ak R such that a = pk ak . Then pk+1 ak+1 = a = pk ak and as R is an integral domain this implies pak+1 = ak , in particular ak+1 | ak . Going to ideals this yields an ascending chain a0 S a1 S . . . ak S . . . As R is noetherian there is some s N such that as R = as+1 R, i.e. there is some R such that as+1 = as = pas+1 . If as+1 = 0 this would mean p = 1 (as R is an integral domain) which is absurd, since p is prime. Thus we have as+1 = 0 and hence a = ps+1 as+1 = 0. If g = 0 then (f g ) = = (f ) + (g ), thus we only need to consider 0 = f, g R. Now write f = pk a and g = pl b where k = f [p] and l = g [p], then by maximality we have p | a and p | b and hence p | ab, since p is prime. Hence (f g ) = (pk+l ab) = k + l = (f ) + (g ). 472 18 Proofs - Commutative Algebra

 If g = 0 then (f + g ) = (f ) = min{ (f ), (g ) }, thus we only need to consider 0 = f, g R. As above let us write f = pk a and g = pl b where k = f [p] and l = g [p]. Without loss of generality we may assume k l then f + g = pk (a + plk b) and hence (f + g ) = (pk (a + plk b)) = k + (a + plk b) k = min{ (f ), (g ) }. Thus by now is a valuation on R and it is normed, since (p) = 1. Continuing with the previous item now prove the fth property: i.e. we even have k < l. If we now had p | a + plk b, then there would be some h S such that a = ph plk b. But this would imply p | a which is absurd, by maximality of k . Hence (f + g ) = k = min{ (f ), (g ) }. (iv) [ ] is well-dened: As is normed, we have [ ] = 0. Hence [ ] is a non-zero-prime ideal, and as R is a PID this even implies maximality. m p is well-dened: If we have pR = m = qR then there is some R with p = q and as (p)k | a pk | a we then also get q = p = p . m p [p ] = m: It is clear that [p ] = pR and as also m = pR by construction [p ] = m. [ ] p = : By construction we have [ ] = pR, thus if a R with p | a then a pR = [ ] and hence (a) = 0. And as is normed, there is some q S such that (q ) = 1. Hence q [ ] = pR, which means p | q . Therefore 1 (p) (as p pR = [ ]) and (p) (q ) = 1 (as p | q ) imply (p) = 1. If now f R is given aribtarily then we write f = pk a where p | a, then (f ) = k (p) + (a) = k = p (f ). And hence = p . P Proof of (2.121): (i) The properties (1), (2) and (3) of valued rings and discrete valuation rings are identical. Hence it suces to check that R is non-zero. But by (N) there is some m R with (m) = 1 = and by (1) this is m = 0. In particular R = 0. (ii) We have already seen in (??.(i)) that [ ] is a prime ideal of R. Thus by (2.114) it suces to check R \ [ ] R . Thus consider any a R \ [ ], that is (a) = 0 = (1) and hence (1) (a) and (a) (1). By (4) this implies 1 | a and a | 1 which again is a 1 or in other words a R . (iii) By (i) and (??.(i)) we know that R is an integral domain. Thus consider a, b R with a = 0, then we need to nd some q , r R such that b = qa + r and (r = 0 or (r) < (a)). To do this we distinguish three cases: if b = 0, then we may choose q := 0 and r := 0. If b = 0 and (b) < (a), then we may let q := 0 and r := b. Thus if b = 0 and (a) (b), then by (4) we get a | b. That is there is some q R such that b = qa. Hence we are done by letting r := 0. 473

(iv) By induction on k N it is clear that (mk ) = k (m) = k . Thus for (a) = k = (mk ) we obtain a | mk and mk | a. That is a mk are associates and hence there is some R such that a = mk . And the uniqueness of is clear, as R is an integral domain. (v) Let a i R be any ideal of R, combining (iii) and (2.65.(iii)) we know that R is a PID. That is there is some a R such that a = aR. In case a = 0 we have a = 0, else using (iv) we may write a as a = mk for some R . Thus we have a = aR = mk R = mk R as claimed. (vi) As (0) = 0 and (1) = 0 0 we have 0, 1 R. Also we have (1) = ((1)(1)) = 2 (1) such that (1) = 0 and hence 1 R, too. If now a, b R then by (3) we get a + b R and by (2) also ab R. Altogether R is a subring of F . We will now prove that is a discrete valuation on R again. Properties (1), (2), (3) and (N) are inherited trivially. And for (4) we are given a, b R with (a) (b). In case b = 0 we have a 0 = 0 = b and hence a | b trivially. Thus assume b = 0, then we have 0 (a) (b) < and hence a = 0, as well. Therefore we may let x := ba1 F . Then by (2) we get (b) = (xa) = (x) + (a), such that (x) = (b) (a) 0. This again means x R and hence a | b. It only remains to verify, that F truly is the quotient eld of R. Thereby the inclusion is trivial. Hence consider any x F , if (x) 0 then x = x 11 and we are done. Else we have (x1 ) = (x) > 0 and hence x1 R. Thus we likewise obtain x = 1 (x1 )1 . P Proof of (2.123): (a) = (c): If (R, ) is a discrete valuation ring, then R already is a local ring (with maximal ideal [ ]) and an Euclidean ring, due to (2.121). But the latter implies that R is a PID, by (2.65.(iii)). And of course R cannot be a eld, as there is some m R with (m) = 1 (and hence we have both m = 0 and m R as else (m) = 0). (c) = (d): As R is a PID it is a UFD according to (2.65.(ii)). Now consider two prime elements p, q R, that is the ideals pR and qR are non-zero prime ideals. Hence - as R is a PID - pR and qR already are maximal by (2.65.(i)). But as R also is local this means pR = qR. (d) = (b): Choose any 0 = n R \ R which is possible, as R is not a eld. As R is an UFD we may choose some prime factor m of n and let m := mR. Now consider any 0 = a R, as R is an UFD a admits a primary decomposition a = p1 . . . pk . But by assumption any prime factor pi is associated to m, that is a admits the representation a = mk for some R and k N. If we had a R then we necessarily have k 1 and hence a mR = m. Thus we have proved R \ R m. And as m R we have m = R such that R is a local ring with maximal ideal m, according to (2.114). And this again yields R \ m = R as claimed. Now suppose 0 = a k mk R, as we have seen we may write a = mk for some R and k N. But by assumption on a we also have a = bmk+1 for some b R. Dividing 474 18 Proofs - Commutative Algebra

by mk (R is an integral domain) we get = bm and hence m R a contradiction. Thus we also got k mk R = 0. (b) = (a): Let : R N {} : a sup{k N | mk | a}, then we will prove that is a discrete valuation on R: (1) if a = 0 then for any k N we clearly have mk | a and hence (0) = . Conversely if (a) = , then we have mk | a for any k N and hence a k mk R = 0. (2) and (3) are easy by denition of - refer to the proof of (2.119.(ii)) for this. (4) Consider a and b R with (a) (b). If b = 0 then a | b is clear, otherwise we write a = mi and b = mj for i = (a) and j = (b). By maximality of i we get m | and hence R \mR = R . Thus we may dene h := 1 mj i and thereby obtain ha = b such that a | b. (N) suppose m2 | m, then there was some h R such that hm2 = m and hence hm = 1, which is m R . But in this case we had R = R \ mR = - a contradiction. Thus we have (m) 1 which clearly is (m) = 1. (c) = (f): Any PID is a noetherian UFD and by (2.55) an UFD is a normal ring. And the maximal ideal m is non-zero, as we would else nd R = R \ m = R \ {0}, which would mean that R was a eld. Thus let p i R be any prime ideal of R, if p = 0 is non-zero, then p already is maximal, due to (2.65.(i)). But in this case we necessarily have p = m, as R was assumed to be local. (f) = (e): First of all m is a nitely generated R-module, as R is a noetherian ring. Thus if we had m = m2 = m m then (as jacR = m) Nakayamas lemma (3.32) would imply m = 0 - a contradiction. Thus we have m2 m, that is there is some m m with m m2 . In what follows we will prove m = mR. (1) First note that m = 0 (as else m m2 ), thus by assumption and (2.20.(i)) we get mR = { p spec R | mR p } = m

In particular m mR and hence there is some k N such that mk mR, by (2.20.(vii)). And of course we may choose k minimal with this property (that is mk1 / mR). Clearly k = 0 cannot occur (as m m and hence m is not a unit of R). And if k = 1 we are done, as in this case we get m mR m. Thus in the following we will assume k 2 and derive a contradiction. (2) by minimality of k there is some a mk1 with a mR. Let am := {ab | b m}, as a mk1 we have am mk1 m = mk mR. Now let x := a/m E , where E := quotR is the quotient eld of R. Then clearly x R (that is m | a) as else a mR in contradiction to the choice of a. (3) Now it is easy to see that xm := {xb | b m} is a subset xm R of R: for any b m we have xb = ab/m (1/m) mR = R. And in fact xm even is an ideal xm i R of R (0 xm is clear, and obviously xm is closed under +, and multiplication with elements of R). Now by construction we even get xm = R (as else there would be some b m such that xb = 1. But this would imply m = ab am mk m2 (as we assumed k 2) in contradiction to the choice of m). Thus xm is a proper ideal of R and as R is a local ring with maximal ideal m this implies xm m. (4) Let us now choose generators of m, say 475

m = b1 R + + bn R. As xm m there are coecients ai,j R (where j 1 . . . n) such that for any i 1 . . . n

xbi = ai,1 b1 + + ai,n bn Let us now dene b := (b1 , . . . , bn ) and the n n matrix A := (ai,j ) over R. Then the above equations can be reformulated as Ab = xb, that is b kn(x1 1 A). Thereby (x1 1 A) is a n n matrix over the eld E , and as it has a kernel we have det(x1 1 A) = 0 E . Thus if f := det(t1 1 A) R[t] is the characteristic polynomial of A then we have f (x) = det(x1 1 A) = 0. But as R was assumed to be normal and as f is a normed polynomial over R, this implies x = a/m R or in other words a mR, a contradiction. This contradiction can only be solved in case k = 1 and then we have already seen m = mR. (e) = (c): First of all R is not a eld, as m is a proper non-trivial ideal of R (that is 0 = m = R). It remains to verify, that R is a PID, where m = mR was assumed to be a principal ideal. To do this we rst let z := k mk i R, as R is noetherian z is a nitely generated R-module. And further we have m z = z ( is clear and conversely if a z then for any k N there is some hk R such that a = hk mk . Therefore a/m = hk+1 mk for any k N and hence a/m z which is a m z again). As m = jacR the lemma of Nakayama (3.32) implies that z = 0. We now dene the following map : i R N { } : a sup{ k N | a mk } Obviously is well dened, as a R = m0 for any a i R. And if a = R, then a is contained in some maximal ideal, which is a m and hence (a) 1. Now suppose (a) = , this means a k mk = 0 and hence a = 0. Of course both 0 and R are principal ideals. Thus consider any ideal a i R with 0 = a = R. As we have just seen, this yields 1 k := (a) N. And as a / mk+1 we may choose some k +1 k a a with a m . As a m we in particular have a mk , that is a = mk for some R. If we had m then a mk+1 would yield a contradiction. Hence R \ m = R is a unit. Therefore we found mk = mk R = mk R = aR a mk and hence a = mk R is a principal ideal. Altogether R is a PID. P Proof of (2.126): (i) By denition of a local ring it suces to check R \ R i R. As R = 0 we have 0 R \ R . Now consider any a, b R \ R and r R. Then ra R \ R , as else a1 = r(ar)1 and in particular a R \ R . It remains to show that a + b R \ R . Without loss of generality we may assume a | b (else we may interchange a and b). That is b = ab for some h R and hence a + b = a(1 + h). Thus we have a + b R \ R , as else a1 = (1 + h)(a + b)1 . (i) By assumption R is an integral domain. Thus regard any nitely generated ideal a i R. That is there are a1 , . . . , ak R (where 476 18 Proofs - Commutative Algebra

we may choose k minimally) such that a = a1 R + + ak R. We need to show that a is a principal ideal, that is k = 1. Thus assume k = 1, then a = a1 R (as k is minimal) and hence there is some b a such that b a1 R. Now write b = a1 b1 + + ak bk and let d := a2 b2 + + ak bk . If we had a1 | d then d a1 R and hence b = a1 b1 + d a1 R, a contradiction. Thus we have d | a1 as R is a valuation ring. But this means a1 dR a2 R + + ak R and hence a = a1 R + + ak R = a2 R + + ak R in contradiction to the minimality of k . This only leaves k = 1 which had to be shown. (ii) (c) = (b) is true by (2.121.(iii)) and (b) = (a) is clear. Thus it remains to verify (a) = (c): as R is a valuation ring, it is a Bezout domain, by (i). Thus if R also is noetherian it clearly is a PID. But R also is local by (i) and a local PID is a eld or a DVR by (2.123). P Proof of (2.129): It is clear that f + g, f g and f g are R-submodules again. If now rf R and sg R then we can also check property (2) of fraction ideals (rs)(f + g) = s(rf) + r(sg) r ( f g) (rs)(f g) = rf (rf) (sg) R R R

Further f : g is an R-submodule of F since: if x, y f : g and a R then (ax + y )g af + f = f. Thus it remains to prove that f : g f R truly is a fraction ideal. To do this let rf R and sg R. As g = 0 we may choose some 0 = g g, then clearly sg g R and sg = 0, as R is an integral domain. But now (rsg )(f : g) R as for any x f : g we get(rsg )x = r(xsg ) rf R. P Proof of (2.131): (b) = (a) is trivial, by choosing g := R : f. Thus consider (a) = (b): that is we assume f g = R for some g f R. Then by construction we have g {x F | x f R} = R : f. Hence R = f g f (R : f) R (the latter inclusion has been shown in the remark above) which is R = f (R : f). It remains to verify the uniqueness of the inverse g - but this is simply due to the associativity of the multiplication. We start with f g = R = f (R : f). Multiplying (from the left) with g yields g = R g = (f g) g = (g f) g = g (f g) = g (f (R : f)) = (g f) (R : f) = (f g) (R : f) = R (R : f) = R : f. P Proof of (2.132): (a) = (b): by assumption (a) weve got 1 R = a g. Hence there are some ai a and xi g such that 1 = a1 x1 + . . . an xn a g. And as the xi are contained in xi g = a1 = R : a we have xi a R as claimed. (b) = (a): conversely suppose we are given such ai and xi then by denition ai a and xi R : a. And as a1 x1 + . . . an xn = 1 we also we have 477

R a(R : a) R and hence R = a (R : a). Thus we next concern ourselves with the implication (b) = (c): suppose we have chosen some ai a and xi F as noted under (b), then we dene two R-module homomorphisms : Rn a : (r1 , . . . , rn ) a1 r1 + . . . an rn : a Rn : a (ax1 , . . . , axn ) Note that the latter is well-dened, since xi a R by assumption. But if now a a is chosen arbitarily then we get (a) = aa1 x1 + + aan xn = a as a1 x1 + + an xn = 1. This is = 1 1 on a. Now regard the following (obviously exact) chain of R-modules 0 kn Rn a 0 As we have just seen, this chain splits, as = 1 1 and hence a is a direct summand of the free R-module Rn (which in particular means that a is projective) by virtue of Rn =m a kn : r (r), r (r) Conversely suppose (c), that is M = a P , then we need to prove (b). To do this we dene the R-module homomorphisms : M a : (a, p) a and : a M : a (a, 0). Then clearly = 1 1. As M is free by assumption it has an R-basis { mi | i I }. And hence any a a has a representation in the form a = (a) =
iI

ai mi

=
iI

ai (mi )

Note that only nitely many of the ai are non-zero, as these are the coefcients of the basis representation of (a). This representation allows us to dene another R-module homomorphism i : a R : a ai Now consider any two non-zero elements 0 = a, b a then we see that bi (a) = i (ba) = i (ab) = ai (b). And hence we may dene an element xi F independently of the choice of 0 = a a by letting xi := i (a) a

Note that only nitely many ai were non-zero and hence only nitely many xi are non-zero. The one property is easy to see: if 0 = a a is any element then we get xi a = i (a) R and hence xi a R. The other property

478

18 Proofs - Commutative Algebra

requires only slightly more eort: just compute a =

iI

i (a) (mi ) =
i I

axi (mi ) = a
iI

xi (mi )

Recall that only nitely many xi were non-zero and for these let us denote hi := (mi ). As a was non-zero we may divide by a to see that 1 = i xi hi which was all that remained to prove. P Proof of (2.133): (a) = (b): as a = 0 is non-zero a = 0 is non-zero and hence we may choose g := (1/a)R. Then we clearly get a g = (aR)(1/aR) = R. (b) = (a): as a is invertible, it is a projective R-module. But R is a local ring and hence any projective R-module already is free. Now a being free implies that a is a principal ideal (if the basis contained more than one element - say b1 and b2 - then b2 b1 + (b1 )b2 = 0 would be a non-trivial linear combination). P Proof of (2.135): We will now prove the equivalencies in the denition of Dedekind domains. Amongother useful statements we will thereby proof (2.137.(i)) already. Note that the order of proofs is somewhat unusual: (d) (e) followed by (e) = (a) = (b) = (c) = (b) = (d). (d) = (e): consider any prime ideal q i R and let U := R \ q, that is Rq = U 1 R. Then Rq is a local ring by (2.116). And as R was assumed to be normal, so is Rq , by (2.110). And by (2.108) the spectrum of Rq corresponds to { p spec R | p q }. Yet if p = 0, then p already is maximal by assumption and hence p = q. Thus we have { p spec R | p q } = { 0, q }. Thereby spec Rq has precisely the prime ideals 0 (corresponding to 0 in R) and its maximal ideal (corresponding to q) in R. Altogether it satises condition (f) of DVRs. (e) = (d): by assumption R is a noetherian integral domain. Thus it remains to prove, that it is normal and satises spec R = smax R {0}. By assumption any localisation Rp is a DVR, hence a PID, hence an UFD and hence normal by (2.55). And as R is an integral domain this implies that R itself is normal, too by (2.111). Thus consider any prime ideal 0 = p i R. It remains to prove that p is maximal. Thus we choose any maximal ideal ideal m with p m. By assumption Rm is a DVR, and hence has precisely the prime ideals 0 and mm (2.123). But as 0 = p m by (2.108) the localisation pm is a non-zero prime ideal of Rm and hence pm = mm . Invoking the correspondence again we nd p = m and hence p truly is maximal. R DVR = (a): let m be the maximal ideal of R and m a uniformizing parameter, that is m = mR. (1) If now 0 = f f R is a non-zero fraction 479

ideal, then a := f R i R is an ideal of R. And by denition there is some 0 = r R such that rf R. Recall that by (2.121.(iv)) in a DVR r R is uniquely represented, as r = mk with R and k = (r) N. Among those r R with rf R we now choose one with minimal k = (r). (2) We rst note that a = 0, just choose some 0 = x f, then rx a and as r, x = 0 we also have rx = 0. Thus by (2.121.(v)) a is of the form a = mn R for some n N. (3) Now
f = mnk R

if x f then by (2) we have rx a = mn R, say rx = mn p. Therefore x = mn p/r = 1 pmnk mnk R. If k = 0, then r = R is a unit and hence f = rf R. Thus we get f = n a = m R = mnk R. Now suppose k 1, as k has been chosen minimally, there is some x = a/b f such that mk1 x R. That is b | mk1 a and hence (b) > (mk1 a) = k 1 + (a) by the properties of discrete valuations. However rx R and hence b | mk a. This yields (b) (amk ) = k + (a) altogether (b) = k + (a). Thus if we represent a = mi with R and i = (a) then b = mj with R and j = (b) = k + i. Therefore x = a/b = 1 mk such that rx = 1 R . On the other hand rx a = mn R and hence a = R, that is n = 0. Thus mk = 1 x f such that mnk R = mk R f. This settles the proof of the equality (3). But now we have seen, that f = mnk R is a principal ideal, in particular 1 it is invertible, with inverse f = mkn R. (e) = (a): if R is a eld, then f = R is the one and only non-zero fraction ideal of R. And f = R trivially is invertible, with inverse 1 f = R. Thus in the following we assume that R is not a eld, then we choose any non-zero prime ideal 0 = p i R in R and let U := R \ p. That is U 1 R = Rp and by assumption Rp is a DVR. Thus consider a non-zero fraction ideal 0 = f f R, then 0 = U 1 f f Rp is a fraction ideal of Rp . But as we have just seen this implies, that U 1 R is invertible (regarding Rp ). But by the remarks in section 2.11 this already yields that f is invertible (regarding R). (a) = (b) is trivial, so we will now prove (b) = (c): by assumption (b) any non-zero ideal is invertible and hence nitely generated (by the remarks in section 2.11). And as 0 trivially is nitely generated R is a noetherian ring. Now consider any non-trivial ideal a i R, that is a {0, R} and let a0 := a. As a = R we may choose some prime ideal p1 spec R containing a p1 , by (2.4.(ii)). Now let
a1 := a0 (p1 )1 = a0 (R : p1 )

a0 (R : a0 ) = R

Here the inclusion R : p1 R : a0 holds because of a0 p1 and a0 (R : a0 ) = R is true because a0 is invertible by assumption (a). Hence a1 i R is an ideal of R and by construction
a = a0 = p1 a1

We will now construct an ascending chain a = a0 a1 . . . an . . . 480 18 Proofs - Commutative Algebra

of ideals of R. Suppose we have already constructed ak i R and the prime ideals p1 , . . . , pk spec R such that a = p1 . . . pk ak . If ak = R then we are done as we have already decomposed a = p1 . . . pk in this case. Thus assume ak = R. Then we may choose any prime ideal pk+1 containing ak pk+1 again. As for the case k = 1 above, we let ak+1 := ak (pk+1 )1 and nd that ak+1 i R is an ideal of R. Then by construction we get
a = p1 . . . pk ak = p1 . . . pk pk+1 ak+1

As R is noetherian this chain stabilizes at some point an = an+1 and by construction this means an+1 = an = pn+1 an+1 . As an+1 is a nitely generated R-module the lemma of Dedekind implies that there is some p pn+1 such that (1 p)an+1 = 0. But as R is an integral domain and an+1 = 0 this means that p = 1, which is absurd, since pn+1 = R is prime. Thus the construction fails at this stage, which can only be if an = R. And as we have mentioned already this is a = p1 . . . pn . (b) = (2.137.(i)): the existence of the factorisation of non-trivial ideals into prime non-zero ideals has been shown in (b) = (c) already. Hence it only remains to show that the factorisation is unique up to permutations. Thus assume p1 . . . pm = q1 . . . qn where 0 = pi and qj i R are non-zero prime ideals. Whithout loss of generality we assume m n and use induction on n. The case m = 1 = n is trivial so we are only concerned with the induction step: by renumbering we may assume that q1 is minimal among { q1 , . . . , qn }. And as
m

p1 . . . pm

i=1

pi

q1

and q1 is prime (2.11.(ii)) implies that there is some i 1 . . . m such that pi q1 . Now conversely as pi is prime and q1 . . . qn pi there is some j 1 . . . n such that qj pi q1 . By minimality of q1 this means qj = q1 and hence pi = q1 . By renumbering we may assume 1 1 i = 1. Then multiplying by p 1 = q1 yields p2 . . . pm = q2 . . . qn so we are done by the induction hypothesis. (c) = any invertible prime ideal is maximal. Thus consider some invertible prime ideal p i R, in particular p = 0. We will show that for any u R \ p we get p + uR = R. Suppose we had p + uR = R then by (b) we could decompose p + uR = p1 . . . pm for some prime ideals pi i R. Now take to the quotient ring R/p, then by (1.61) 0 = (u + p)R p = p + uR p = p1 p . . . pm p But (u + p)(R/p), being a principal ideal, is invertible. In particular any of its factors pi /p is invertible with inverse given to be
pi
1

1 R p u+p

pj
i= j

p 481

Now we open up a second line of argumentation: we decompose the ideal p + u2 R = q1 . . . qn and take to the quotient ring again
2 0 = (u2 + p)R p = p + u R p = q1 p . . . qn p

As ((u + p)R/p)2 = (u2 + p)R/p we have found two decompositions of (u2 + p)R/p into prime ideals - to be precise we found
p1
2

...

pm

= q1 p . . . qn p

R/p also satises the assumption (c) because of the correspondence theorem and as any pi /p is invertible the decomposition is unique up to permutations - note that these truly are the only assumptions that were used in the proof of (iii). Hence (by the uniqueness) we get 2m = n and (p1 /p, p1 /p, . . . , pm /p, pm /p) (q1 /p, . . . , qn /p). But by the correspondence theorem (1.61) again we can return to the original ring R getting (p1 , p1 . . . , pm , pm ) (q1 , . . . , qn ). Therefore
p + u2 R = q1 . . . qn

= p1 p1 . . . pm pm = (p + uR)2 = p2 + up + u2 R Hence p p + u2 R = p2 + u(p + uR) p2 + uR. Thus any p p has a representation as p = a + ub where a p2 and b R. Thus ub = p a p such that b p. This is p = a + ub p2 + up and as p has been arbitrary p p2 + up = p(p + uR). Multiplying with p1 this is R p + uR and hence nally p + uR = R. (c) = (b): we will rst show that non-zero prime ideals 0 = p i R are invertible. To do this choose some 0 = a p and decompose aR = p1 . . . pn . As aR is invertible - with inverse (aR)1 = (1/a)R any pi is invertible, as its inverse is just
pi
1

1 R a

pj
i=j

As p is prime and p1 . . . pn = aR p there is some i 1 . . . n such that pi p. But pi is invertible and hence maximal (see above) such that pi = p. Thus p is invertible. If now 0 = a i R is any non-zero ideal then we decompose a = p1 . . . pn and as any pi is now known to 1 1 be irreducible we have a1 = p 1 . . . pn invertible. (b) (c) and (b) = spec R = smax R { 0 }. We have just established the equivalence (b) (c). And we have seen above that (because of (c)) any invertible prime ideal is maximal. But because of (b) any non-zero prime ideal already is invertible and hence maximal. (b) = (d): we have already seen - in (b) = (c) - that R is noetherian and that spec R = smax R { 0 }. Thus it only remains to show that R is normal. If R is a eld already then there is nothing to 482 18 Proofs - Commutative Algebra

prove, else let F := quotR and regard x = a/b F . If x is integral over R then R[x] is an R-module of nite rank k + 1 := rankR[x]. Let now r := bk - as any element of R[x] is of the form ak xk + + a1 x + a0 (for some ai R) we clearly have rR[x] R (in particular R[x] f R is a fraction ideal of R). And hence for any n N we have sn := rxn R. For any prime ideal p i R let us now denote by p (a) the multiplicity of p in the primary decomposition of a, i.e. we let p (p1 . . . pm ) := # { i 1 . . . m | p = pi } N Then clearly p turns the multiplication of ideals into an addition of multiplicities and hence ran = sn bn turns into p (rR) + np (aR) = p (sn R) + np (bR) Hence for any n N and any prime ideal p i R we get the estimate p (rR) + n(p (aR) p (bR)) p (sn R) 0 By choosing n > p (rR) this implies that for any prime ideal p we have p (bR) p (aR). Writing down decompositions of aR and bR we hence nd that aR bR. But this is b | a or in other words x R which had to be shown. P Proof of (2.139): The proof of the equivalencies in the denition (2.138) and the properties (2.139) of valuations of ideals in Dedekind domains is interwoven tightly. Thus we do not attempt to give seperate proofs, but verify the properties and equivalencies simultaneously. b | a a b: if a = b c, then b a is clear. Conversely suppose we had a b. If additionally b = 0 then a = 0, too and hence we 1 may already take c := R. Thus suppose b = 0 and let c := b a f R. 1 Then c b b = R and hence c = c R i R. But on the other hand 1 b c = b b a = a is clear, such that b | a. (i) In case m M we may pick up m0 := m and k (0) := 0 reducing this to to the case m M. And if m M then by renumbering we may k(1) k(n) assume that m = m1 . Thus regard a := m1 . . . mn , then a mk(1) is clear and hence m (a) k (1), by denition of m . Thus assume m > k (1), this means a mk(1)+1 and by what we have just shown there is some c i R such that a = mk(1)+1 c. Therefore
m1
k(2) (n) . . . mk = mk(1) a = m c m n

But as m is prime this means that there is some i 2 . . . n such that k(i) mi m. Clearly k (i) = 0 as else R m and hence mi m by the same argument. But mi has been maximal and hence mi = m = m1 in contradiction to the mi being pairwise distinct. 483

(ii) Decompose a = m1 . . . mn and b = m1 . . . mn as in the formulation of (iii). Also as in the proof of (i) we may also assume that m = m1 (by picking up m0 := m, k (0) := 0 and l(0) := 0 if necessary k(1)+l() k(n)+l(n) and renumbering). Then it is clear that a b = m1 . . . mn . And using (i) the claim follows readily: m (a b) = k (1) + l(1) = m (a) + m (b). a | b m : m (b) m (a). Clearly if a = b c then by (ii) we get m (a) = m (b) + m (b) m (b). Conversely decompose a = k(1) k(n) l(1) l(n) m1 . . . mn and b = m1 . . . mn once more. By (i) that is k (i) = mi (a) and l(i) = mi (b). Thus by assumption we have l(i) k (i) k(1)l(1) k(n)l(1) and may hence dene c = m1 . . . mn i R. Then by construction we immediately get b c = a. (iii) First let m(m) := min{ m (a), m (b) } for any m smax R. Note that for m = mi we in particular get m(m) = m(i) and if m is not contained in the mi then m(m) = 0. And therefore we get
m1
m(1) m(n) . . . mn =

k(1)

k(n)

l(1)

l(n)

mm(m)
m

Now consider any ideal c i R, then by the equivalencies we have already proved for a | b it is evident that we obtain the following chain of equivalent statements m : m (c) m (a + b)
a+b c a c and b c

m : m (c) m (a) and m (c) m (b) m : m (c) m(m)

Thus for a xed m smax R we may regard c := mj . And by (i) this satises m (c) = j , that is we may take m (c) to be any j N of our liking. And therefore the above equivalency immediately yields m (a + b) = m(m) and hence the equality for a + b claimed. (iv) Consider any two non-zero ideals a, b of R, then by (iii) a and b are coprime (i.e. a + b = R) if and only if for any m smax R we get m (a) = 0 or m (b) = 0. In particular for any m = m smax R and any k , l N we see that mk and nl are coprime. Thus the claim follows immediately from the original chinese remainder theorem (1.92). (v) If a = aR is a principal ideal, then we may take b := R, then a + b = R and a b = a = aR are satised trivially. Thus suppose a is no principal k(1) k(n) ideal (in particular a = 0), then we may decompose a = m1 . . . mn as usual. Now for any i 1 . . . n pick up some bi R such that k(i) k(i)+1 k(i) k(i)+1 bi mi but bi mi (thereby mi \ mi = is non-empty, as k ( i) we else had R = mi by multiplication with mi ). Hence by (iv) there k(i)+1 is some b R that is mapped to (bi mi ) under the isomorphism of the chinese remainder theorem. That is for any i 1 . . . n we have k(i)+1 k(i)+1 k(i) k(i)+1 b+mi = bi +mi . In particular we nd b mi and b mi 484 18 Proofs - Commutative Algebra

again. This is to say that mi (b) = ki for any i 1 . . . n. That is if we decompose bR then there are some n N N, mi smax R and 1 k (i) N (where i n + 1 . . . N ) such that
N

bR =
m

mk(m) =
i=1

mi

k(i)

Note that n < N , if we had n = N then bR = a already would be a principal ideal, which had been dealt with already. Thus if we let k(n+1) k(N ) b := mn+1 . . . mN , then it is evident, that a b = bR is principal. But also by construction a and b satisfy m (a) = 0 or m (b) = 0 for any m smax R. And as we have already argued in (iv) this is a + b = R. P Proof of (2.137): Note that (i) has already been shown while proving the equivalencies in the denition of Dedekind domains on page 479. Thus it only remains to verify (ii) to (v), which we will do now: (v) Let us denote the mapping (of course we still have to check the welldenedness) given by , that is (a + mk ) := a/1 + mk Rm . Now consider any a, b R and u, v m, then in a rst step, we will prove a b + mk Rm + + mk Rm u v av bu mk

If av bu mk then (as uv m) it is clear, that (a/u) (b/v ) = (av bu)/uv mk Rm . And thereby also a/u + mk Rm + b/v + mk Rm . Conversely let us assume (av bu)/uv = (a/u) (b/v ) mk Rm . That is there are some m mk and w m such that (av bu)/uv = m/w and hence w(av bu) = muv mk . Therefore w(av bu)R mk such that by (2.138) and (2.139) k = m (mk ) m (w(av bu)) = m (w) + m (av bu) Yet w m and hence m (w) = 0. Therefore we get m (av bu) = k and this translates into av bu mk as claimed. And this immediately yields the equivalencies: a + mk = b + mk i a b mk i (a + mk ) = (b + mk ). And this has been the well-denedness and injectivity of . Thus it remains to prove the surjectivity of : we are given any a/u + mk Rm and need to nd some b R such that a/u + mk Rm = b/1 + mk Rm . Yet u m and m is a maximal ideal, that is R/m is a eld and hence there is some v m such that v + m = (u + m)1 . That is 1 uv m and in particular (1 uv )k mk . But by the binomial rule we may compute
k

(1 uv )k = (1)k (uv 1)k = (1)k

i=0 k

k (uv )i (1)ki i
k

=
i=0

(1)i

k (uv )i = 1 + u i

(1)i
i=1

k i 1 i u v i 485

That is we get (1 uv )k = 1 + uq for some adequately chosen q R. Then we dene b := aq , now an easy computation yields a bu = a(1 + uq ) = a(1 uv )k mk . Thus by the above equivalency we nd (b + mk ) = a/u + mk Rm . (ii) (1) Let us rst assume, that a = mk is a power of a maximal ideal m i R. If k = 0, then a = R such that R/a = 0 such that we have nothing to prove. Thus assume k 1, then by (v) R/a is isomorphic to Rm /aRm . But as R is a Dedekind domain Rm is a DVR (by (2.135.(e))) and in particular a PID (by (2.123.(c))). Therefore the quotient Rm /aRm is a principal ring (see section 2.6). Thus the isomorphy shows, that R/a is a principal ring. (2) Now consider any 0 = a i R. We have already dealt with the case a = R in k(1) k(n) (1).(k = 0). Thus suppose a = R and decompose a = m1 . . . mn (with n(i) 1). Then by the chinese remainder theorem in (2.139) R
k(i)

=r

m1

k(1)

(n) R mk n

Now every R/mi is a principal ring due to case (1). But the direct sum of principal rings is a principal again, due to the some remark in section 2.6. Thus the isomorphy shows, that R/a is a principal ring. (iii) Consider any a i R and 0 = a R, then by (ii) R/aR is a principal ring and hence a/aR is a principal ideal - say a/aR = (b + aR)(R/aR) for some b R. As b + aR a/aR it is clear that b a, in particular aR + bR a. Now consider any c a, then c + aR a/aR = (b + aR)(R/aR). That is there is some v R such that c + aR = (b + aR)(v + aR) = bv + aR. Hence c bv aR which means that c bv = au for some u R. And as c = au + bv a has been arbitrary this also is a aR + bR (iv) Let us denote the set of all prime ideals that are not principal by / p R : p = pR . (1) clearly n := P - formally P := p spec R | #P < , as P spec R \ {0} = smax R and the maximal spectrum of R is nite by assumption. If we have n = 0 then any prime ideal is principal. Thus if a i R is any ideal decompose a = p1 . . . pk where pi i R is prime (using (c)). Then pi = pi R is principal and hence a = aR is principal, too by letting a := p1 . . . pk . Thus it suces to assume n 1 and to derive a contradiction. (2) Let us denote b := p1 . . . pn and a := p1 b = p2 1 p2 . . . pn . Then a = b, as else 1 1 R = b b = a b = p1 - a contradiction (recall that b = 0, as R is an integral domain and hence we may apply (b)). (3) Next we claim that there is some u p1 such that u p2 1 , u pi for any i 2 . . . n and p1 = a + uR. In fact a = 0, as R is an integral domain and hence R/a is a principal ring by (ii). And as p1 /a is an ideal of R/a, there is some u R such that
p1

R a + uR a a = ( u + a) a =

By the correspondence theorem we nd p1 = a + uR as a p1 . Thus 2 2 suppose u p2 1 , then p = a + uR p1 p1 and hence p1 = p1 . Dividing by p1 we would nd R = p1 , a contradiction. And if we 486 18 Proofs - Commutative Algebra

suppose u pi then likewise p1 = a + uR pi . But as p1 already is maximal this would imply p1 = pi a contradiction (to i = 1). (4) now we prove that uR = p1 q1 . . . qm for some prime ideals qj P . First of all uR admits a decomposition uR = q0 . . . qm according to (b). Then q0 . . . qm = uR p1 and hence q0 p1 for some j 0 . . . m, as p1 is prime. Without loss of generality we may assume j = 0. Then q0 p1 and this means q0 = p1 , as q0 is maximal (it is non-zero, prime). Thus we have established the decomposition uR = p1 q1 . . . qm , now assume qj P for some j 1 . . . m. If we had qj = pi for some i 2 . . . n then u uR p1 in contradiction to (3). And if we had qj = p1 then u uR p2 1 in contradiction to (3) again. (5) Thus the qj P are principal ideals, say qj = qj R. Then we let q := q1 . . . qm and thereby nd uR = p1 q1 . . . qm = p1 (qR). Thus uR qR, say u = qv for some v R. Then dividing by qR we nd p1 = (u/q )R = vR such that p1 P - a contradiciton at last. P Proof of (2.65): It remains to prove statement (iv) of (2.65), as parts (i), (ii) and (iii) have already been proved on page 437. So by assumption every prime ideal p of R is generated by one element p = pR. In particular any prime ideal p is generated by nitely many elements. Thus R is notherian by property (d) in (2.27). Further 0 = p = pR implies p to be prime by (2.48.(ii)) (and as R is an integral domain this also means that p is irreducible due to (2.48.(v))). In particular any nonzero prime ideal 0 = p i R contains a prime element p p (namely p with p = pR). And thus R is an UFD according to property (e) in (2.50). And by (??) this implies that R even is a normal domain. Now regard a non-zero prime ideal 0 = p = pR i R again and suppose p a i R for some ideal a of R. If a = R then a is contained in some maximal ideal a m i R of R. And as any maximal ideal m is prime, by assumption there is some m R such that m = mR. Now pR = p a m = mR implies m | p. That is there is some q R such that qm = p. But as p is irreducible, this implies q R (as m R would imply m = R). That is p m and hence p = m such that p = a, as well. This mean that p already is a maximal ideal of R. Altogether we have proved, that R is a normal, noetherian domain in which any non-zero prime ideal is maximal. But this means that R is a Dedekind domain according to (c) in (2.135). Now consider any ideal a i R. Then due to (b) (2.135) there are prime ideals p1 , . . . , pn i R such that a = p1 . . . pn . By assumption there are pi R such that pi = pi R and hence a = (p1 R) . . . (pn R) = (p1 . . . pn )R is a principal ideal, too. As a has been arbitrary, this nally means that R is a PID. P Proof of (2.117): as R is noetherian we may nd nitely many m1 , . . . , mk m such that m = Rm1 + + Rmk . We now choose k minimal with this property, that is k = rankR (m). Then it is clear that {mi + m2 | i 1 . . . k } is a generating set of m/m2 (given any n + m2 m/m2 we may choose a1 , . . . , ak R such that 487

n = a1 m1 + . . . ak nk and thereby n + m2 = a1 (m1 + m2 ) + + ak (mk + m2 )). And thereby we have dimE m m2 # mi + m2 | i 1 . . . k k = rankR (m)

choose any E -basis {mi + m2 | i 1 . . . k } of m/m2 , then we let a := Rm1 + + Rmk . Then by construction we have m = a + m2 (it is clear that m2 m and as any mi m we also have a m, together a + m2 m. Conversely if we are given any n m then we may choose ai + m E such that n + m2 = i (ai + m)(mi + m2 ) = ( i ai mi ) + m2 . Hence we have n i ai mi m2 which means n a + m2 ). Now remark that jacR = m, as R is a local ring. Further we have m2 = m m and m is a nitely generated R-module, as R is noetherian. Thus by the lemma of Nakayama (3.32.(??)) we nd m = a and in particular rankR (m) k = dimE (m/m2 ). P

488

18 Proofs - Commutative Algebra