Sunteți pe pagina 1din 9

Any primitive Pythagorean Triple can be used to generate an infinite number of solvable

negative Pell equations.


Paul David Jackson October 2009

INTRODUCTION
In general it is hard to solve x2 – Dy2 = -1, for D, that is find non- square D >1, such that the equation
has solutions, this is in contrast to x2 – Dy2 = 1, for which solutions always exist in non-square D >1.
We shall call the former equation Pell(D, -1) for brevity. Now it is standard that if the Infinite
continued fraction(ICF) representation of D has odd cycle length then Pell(D, -1) is solvable,
otherwise it is not. If D contains no primes of the form 4k+3, then Pell(D, -1), may be solvable, the case
D = 221 = 1317, being a counter example, otherwise if D is a prime of 4k+1 then we always have
solutions.
It is also known that for certain forms of D such as n2 +1, then the cycle length is 1, and so odd.
But in general it seems hard to predict without computing the ICF whether the cycle length will be odd
or even, and so hard to know whether for a particular D not of certain forms, that Pell(D, -1) is in fact
solvable.
So we give an elementary method of generating an infinite number of D such that for all D's so
generated, Pell(D, -1) is solvable. In fact we show

D  ( Rn  a )2  ( Sn  b) 2
where the pair (a, b) are solutions to Rb – Sa = 1, and (R, S, T), are a primitive Pythagorean Triple.
Unfortunately generally this doesn't answer the question given; a particular D is Pell(D, -1) solvable,
but it may be possible to answer a question like does there exist a D such that Pell(D, -1) is solvable in
a certain interval. However we give a simple test that can show for a candidate D, Pell(D, -1) has no
solutions.
First we show that for each primitive Pythagorean triple we can generate an infinite number of
solutions to x2 + y2 = z2 + 1, and then we use this set up to generate an infinite number of D's so
Pell(D, -1) is solvable.

Consider the Diophantine equation x 2 + y 2 – 1 = z2, we can think of this as this as defining the rational
points on the surface Hyperboloid of one sheet, call these solutions H. Then say we wished to find an
infinite number of solutions, then we use the equation thus;
put; 1 – y 2 = x 2 – z2 thus we get; ( 1 + y )(1 – y) = ( x + z )( x – z) , which gives the system of
generators for parameter t; 1 + y = t( x + z ) , and t(1 – y) = x – z, now if we put t = d/c, and solve for
x, and y, we find;
x = (Ry+T)/S , where R = c2 – d2 , S = 2cd, and T = c2 + d2, and c > d.
Now if d, and c are coprime R, S, T, form a Pythagorean Triplet.
Then we can write x = (Ry+T)/S , as a linear Diophantine equation, that is Sx – Ry = T, and solve in the
standard manner to get the general solutions; x = Rk + Tx0 . y = Sk + Ty0 , k an integer, where x0 ,and
y0 , are the solutions to Sx – Ry = 1.

1
Actually we do not need d, and c , to be coprime, as any common factors would cancel, in the
expression, x = (Ry+T)/S , and we would end up with primitive Pythagorean Triplets anyway.

By eliminating x, from the system of generators and substituting for t, we can find a general relation
between z, and y, viz; z = (Ty+R)/S.

Now say X, and Y satisfy x = (Ry+T)/S, then squaring this and adding to Y2, gives;
X2 + Y2 = (R2Y2 + 2RTY + T2 +S2Y2)/S2 = (T2Y2 + 2RTY + R2 +S2 )/S2 = [(TY +R)/S]2 +1 = z2 + 1,
using; z = (Ty+R)/S, and R2 + S2 = T2.
This shows that any solutions in x and y derived from solving x = (Ry+T)/S , satisfies the relation,
x 2 + y 2 = z2 + 1, and so using (Ty+R)/S, with the value y gives the correct value for z, and therefore the
following is valid.

Using the general solutions for x and y, viz;


x = Rk + Tx0 ., and
y = Sk + Ty0 , if we let the last terms in each of the right hand sides of these equations be x' and y'
respectively we have;
x = Rk + x'
y = Sk + y'
we can derive a similar expression for z by using z = (Ty+R)/S, that is we have;
z = (T(Sk + y') +R)/S = Tk + (Ty'+R)/S, which is just Tk + z'.
So we have;
x = Rk + x'
y = Sk + y'
z = Tk + z'.
Thus we have parametric equations in k , for solutions H, and hence we see there will be an infinite set
of triples (x, y, z), that satisfy;
x 2 + y 2 – 1 = z2 , as (R, S, T), are mutually coprime.

We can also write the solution set as the vector equation of a straight line in 3D space; that is;
X = Pk + x', where X = (x, y, z), P = (R, S, T), and x' = (x', y', z') . Now all the points on this vector
equation not only for integer values of the parameter k, but also for all real values of k,
gives solutions for all real (x, y, z), and all these lie in the surface. So our vector equation must
represent a generator of the hyperboloid x 2+ y 2 – 1 = z2.

T divides z2 +1 for all z

Next we show that any z in the solution on H, has the property that z2 +1, divides by T.
This is obvious, as in solving the system, x = (Ry+T)/S, z = (Ty+R)/S, to obtain X, means we have one
z', that satisfies H or x2 + y2 = z2 +1.

2
Say (x, y, z) is such a solution then;

z2 +1= ((Ty+R)/S)2 +1 = (T 2 y2 + 2TRy + R2 +S2 )/S2 = T/S(y(Ty+R)/S + (Ry+T)/S) = T(yz +x)/S .


Now the LHS is a positive integer and so must the first term on the RHS be, then the numerator of the
second term on the right divides by S2 , and so has the form S2M, where M is an integer. But by the third
term on the right S2M = TK, with K an integer which is a multiple of S, so we have SM = TK', but S and
T are coprime and the last equation has been arrived at by valid manipulations therefore S must divide
K' as we require, thus we see that S divides yz + x, and T always divides z2 +1 as claimed.

Now we are ready to show that any primitive Pythagorean Triple can be used to generate an
infinite number of solvable Pell(D,-1).
Proof:
1)
So given the Pythagorean Triple (R,S,T) and X = Pk + x', with z = Tk + z', we know that T divides
all z2 +1 . Let r = (z2 +1 )/T.
We show that (z2 +1)/T2 = D, for values of k that are T apart, viz for k of the form; k = Tn + v, with n
and v integers, with D the sum of two squares.

2)
We need to show that (T, 2z') = 1, now we know that T divides z'2 +1 => z'2 +1 = rT, for some integer r,
then this means that Tr – z'2 = 1, and from this we see that T and 2z' cannot share a common factor
greater than unity.
3)
Using z = Tk + z', the expression (z2 +1)/T2 simplifies to k2 + (2z'k +r)/T, where Tr = z'2 +1 , which we
know to always hold for all z'.
Now for the moment assume that T doesn't divide r, we also know that (T, 2z') = 1, so (2z'k +r)/T = m
always has a solution in integers. Which means we can write this as;
Tm – 2z'k = r, and we can solve this in the usual way with m0 , and k0 , one integer solution to
Tm – 2z'k = 1, so the general solution is;

m = 2z'n + m0 r
k = Tn + k0 r, with integer n, showing that the solutions occur as a subset of the k, at intervals T apart.

Thus we can write (z2 +1)/T2 as k2 + m = T2n2 + 2An +B, (*) where
A = Tk0 r + z' , and B = k02r2 + m0r.
Remembering that T2 = R2+ S2 of course, means that the first term in (*) can be written as the sum of
two squares. Also because we have found D = (*), and we have z and T that satisfy;
z2 – DT2 = -1, this has no solutions if D is divisible by a prime of the form 4c + 3,
so D contains no prime factors of this form. Thus the only primes that could divide D are those of the
form 4c + 1, or powers of 2. Also D cannot be a square, and we know that odd or even products of
primes of the form 4c +1, are sums of two squares. Evaluating z2 – DT2 = -1, in modulo 4 shows that
if D divides by 4, then z2 , is congruent to 3, which is impossible for a square, and thus by contradiction
we see that D does not divide by 4, and thus by no higher power of 2, and is at most twice an odd
number or equal to 2.
Thus (*) can be expressed as the sum of two squares.
3
4)

We express (*) as the sum of two squares.


Now comparing (*) with the expression (Rn+a)2 + (Sn+b)2 means that;

a2 +b2 = B, and Ra +Sb = A, which we can solve in the obvious way, to get;

a = (AR +S)/T2 , and b = (AS – R)/T2 , where the signs may be the opposite way round.

If T does divide r then in the above put, m0 r = r' = Tr, and k0 = 0.

To solve for a and b we need to use; Tm0 – 2z'k0 = 1, and Tr = z'2 +1 , we do the obvious and form
a = (A – Sb)/R substitute this into a2 + b2 = B, and solve the quadratic for b, we find that the
discriminant simplifies to 1, viz T2B – A2 , there seems no obvious way to chose the sign when taking
the square root. But in principal this shows there is an algorithm to compute a and b and thus find what
amounts to the D's and express them as the sum of two squares. Therefore;

D  ( Rn  a )2  ( Sn  b) 2

(this actually equals (Rb – Sa)2),


Now what guarantees we always get at least one set of integer solutions for a and b ?
Perhaps this is easy to see as (*) is an integer and what we have said above about D coming from the
solutions in Z and T means that D and hence (*) is expressible as the sum of two squares, which must
be in integers as everything else is, crucially B is a positive integer.

Now we show that the pair (a,b) given in the expression for D is obtained by solving Sa – Rb = 1.

The author noticed that in, the expression for D, the terms Rn+a, and Sn+b, were similar to;
x = Rk + x'
y = Sk + y',
the equations we found when solving Sx – Ry = T, in which x', and y', are one solution to Sx – Ry =1,
scaled by T. And in fact the general solution to this last equation is just;
x = Rn + x0
y = Sn + y0 ,
where we can see that k = Tn. So we guess that the pair (a,b) given in the expression for D is obtained
by solving Sa – Rb = 1.

4
We need to show that;
(Sa  Rb)2  1

But we know T 2 B 2  A2  1, so we show T 2 B 2  A2  (Sa  Rb) 2 .


Now T 2 B  ( R 2  S 2 )(a 2  b 2 )  R 2 a 2  R 2 b 2  S 2 a 2  S 2 b 2 , and A2  R 2 a 2  2 RaSb 2  S 2 b 2
 T 2 B 2  A2  S 2 a 2  2 SaRb  R 2 b 2  (Sa  Rb) 2 =1, as required.
We can show the same thing using the expressions we derived for a, and b, that is;
AR  S AS  R
a 2
, and b  , then forming Sa  Rb, gives,  1
T T2
Next we show that solving Sx  Ry  1, with x  a, y  b, is sufficient to satisfy, Sa  Rb  1. Say we
have for some R, and S , part of a Primitive Pythagorean Triple (R, S , T ), Sx  Ry  1, then the general
solution, with the pair (a, b), being one solution, is;
x  Rn  a,
y  Sn  b, which implies we have D  ( Rn  a ) 2  ( Sn  b) 2 . Say on the other hand we have;
Sx  Ry  1, then this has the same pair as the first equation but with negative signs, that is ( a, b),
and we have the general solution;
x  Rn  a,
y  Sn  b, which implies we have D  ( Rn  a) 2  ( Sn  b) 2 . But D will take the same values, as the
version with the pair (a, b), that is ( Rn  a ) 2  ( Sn  b) 2 , but with n, negative, so all we need do is solve
Sx  Ry  1.
Also it seems obvious if we can find one solution to z2 – DT2 = -1, for a particular D then we can find
others and of course there are an infinite number of solutions.

Therefore we have shown that an infinite number of D's such that Pell(D, -1) is solvable are generated
by every primitive Pythagorean triple, with the pair (a,b) in the expression for the D's given by solving,
Sa – Rb = 1.

5
We show that all solvable Pell(D, -1) in D lie on H, that is all appear as solutions to the above get
up.

This follows if for solvable Pell(D, -1) the y's are T's, that is if at least one y is a T, that is part of a
Primitive Pythagorean Triple, for a particular D.

for y >1 we show;


(i) all y are odd.

z and y are coprime so they are both not even, the assumption that y is even is easily seen to be absurd.
Just let y = 2u, and z = 2v+1, say and substituting into z2 – Dz2 = -1, and rearranging we find;
4v2 + 4v +2 = 4Du2, which on canceling through by 2 gives LHS odd = RHS even.

(ii) y only has prime factors of the form 4k+1.


Consider z2 = Dy2 – 1 , in modulo some odd prime, we have, z2 = – 1, which only has solutions for
primes of the form 4k+1, thus not only D, but y cannot contain any factors prime of the form 4k+ 3.

(iii) y is the sum of two squares of opposite parities.

y only contains primes of the form 4k+1, thus as these are the sums of two squares their product is by
the sum of squares identity. Furthermore it is odd so the squares have opposite parities.

(iv) if y > 1, then it is the sum of two coprime squares, that is all y's are T's, thus are primitive.
We show this below.

If fact we prove the following;


if N is a positive integer that only contains primes of form 4k+1 then it is always expressible as the
sum of two coprime squares in at least one way.

proof;
consider even powers of an arbitrary prime of the above form, we first show all these have at least one
primitive representation, then we use these to show the same is true for odd powers. Then we show that
any product of the sums of two primitive elements, is itself primitive taken two at a time, and hence we
prove the claim.

We begin with p2 = a2 + b2 , this is primitive because there are no possible squares to factor out other
than the trivial one, viz 1, to form a non primitive, as p is prime. Or another way to look at this is the
only square we could scale the RHS by is 1. Also this is the sum of two squares in only one way.

Next p4 , this has two representations as the sum of squares both distinct, from above we know
p2 , is primitive so one non – primitive is this scaled by the other square in p4 , viz p2 , viz to form;
the pair (pa, pb), but what form must the other representation take? It can't be the same as the scaled
version, also we have used up the possible squares to do the scaling, so it must be primitive.
And next we look at p6 , this has 3 representations and 2 are non- primitive as we can scale the above
primitive representations of the last two cases to form non – primitives but then we have again used up
all possible squares for the scaling and as all 3 representations are distinct we must have a primitive
representation.
6
Now it should seem obvious that this argument can be continued indefinitely by induction, and thus it
follows that all even powers have a primitive representation.
The argument for the odd powers is the same.

p3 , has two distinct representations as the sum of two squares, and as p is primitive then we can form a
non – primitive by scaling this by p2, but this is the only nontrivial square factor so we conclude that the
other possibility is primitive.

p5 , has three distinct representations as the sum of two squares, and as p is primitive then we can form a
non – primitive by scaling this by p4 , and we have the same with p3, which can be scaled by p2. But
now we have used the possible squares as scalars and so these are the only non-primitives and so we
conclude that the other is primitive.

Again as with the even case which we can use we see by induction that all odd powers have one
primitive representation, and we see that this is so for all powers.

Next we note that if all the primes in N are just to the first power viz N is square free, then N has only
primitive representations, because there are no squares to factor out or act as scalars of primitives.

Now say N = (a2 + b2)(c2 + d2), where both representations are primitive, that is no prime p of the above
form divides any of a, b, c, or d.
Then N = (ac + bd )2 + ( ad – bc )2 = (ac – bd )2 + (ad + bc )2 .
We wish to show at least one of the sums of squares is primitive.
We assume p divides one of the sums, say p divides (ac + bd )2 + ( ad – bc )2 , then;
ac + bd = pm, say and
ad – bc = pn, then this means the pair in the other sum of squares can be written as;

ac – bd = pm – 2bd, and
ad + bc = pn + 2bc.
Then the assumption that the second pair also are divisible by p and so both are non primitive means
that;
pm – 2bd = ps, say and
pn + 2bc = pt, so then we have;

pm – ps = 2bd or p(m – s) = 2bd, and


pt – pn = 2bc or p(t – n ) = 2bc.

But p > 2 and so doesn't divide 2, and thus must divide at least one of b, d, and at least one of b, c, but
this contradicts the assumption that the imput representations are primitive therefore we see the
assumption that p divides both the output representations is false, and thus at least one output is a
primitive representation as required.

So to conclude, any N that is a product of primes p of the above form must have at least one primitive
representation as a sum of squares as each prime raised to a power is primitive and we have just shown
that taking products of primitives two at a time results in at least one primitive representation, hence
this implies and proves the claim!
7
EXAMPLE
Say we have T = 65, then as this is a sum of two squares in two different ways we have two primitive
Pythagorean triples, viz;
{i} (33,56,65), which using Sa – Rb = 1, gives (a,b) = (10, 17), so Rn+a = 33n+10, Sn+b = 56n+17.
{ii}(63,16,65), which using Sa – Rb = 1, gives (a,b) = (4, 1), so Rn+a = 63n+4, Sn+b = 16n+1.

We tabulate the results;


n 0 -1 1 -2 2 -3 3 -4 4 -5 5
D {i} 389 2050 7178 12161 22417 30722 46106 57733 78245 93194 118834
D {ii} 17 3706 4778 15845 17989 36434 39650 65473 69761 102962 108322

So T = 65, generates the set of D's = {17, 389, 2050, 3706, 4778, 7178, .......}.

Conclusion
Hence we have shown that an infinite number of D's such that Pell(D, -1) is solvable are generated by
every primitive Pythagorean triple, with the pair (a,b) in the expression for the D's given by solving,
Sa – Rb = 1, and further that all solvable Pell(D,-1) appear as solutions on H. That is as D is the sum of
two squares, we have the solution (x, y, z) = (Ta, Tb, Z), on H.

Whilst completing this article the author became aware of the two articles listed in the references
below, which show that the connection between the Pell equation and Pythagorean triples had been
noticed a long time ago by firstly Euler, [2], and then by many others. Euler apparently was trying to
simplify the amount of calculation needed to find the fundamental solutions to the positive Pell
equation. In the present article the motivation to look into the solvability of Pell(D, -1) arose from
playing around with the simple parametric method of finding integer solutions on H, and noticing first
that if D = 2, we have the solution on H, (x, x, z), that is solutions to Pell(2, -1), and then wondering
about whether this generalised. In [1], the essentially the same results are given but by using continued
fractions, showing that all Y's are T's. Also shown is the squares that equal D are unique, or what is the
same, the expressions we have, viz; Rn+a, Sn+b. Now we have shown above that for any solvable
Pell(D, -1), we have for all Y>1, are T's, and therefore there must be an infinite number of ways to
generate a particular D, and so the uniqueness given in [2] cannot be in this sense, in fact this is also
implied by Theorem 2.1. We assume what is meant is uniqueness in in the sense of if D is composite
then only one of the representations as a sum of two coprime squares satisfies Sa – Rb = +1.

Thinking about the uniqueness of (a, b), with b odd, leads us to give a simple test for non membership
of the set of D's such that Pell(D, -1) is solvable. The idea being to use each pair (a, b) in the
representation of the candidate D in turn, and assume the existence of a primitive Pythagorean triple
(R,S,T), that along with the given values of (a, b) satisfy Sa – Rb = +1, and then to use the fact that
R = m2 + n2 , and S = 2mn, which on substitution in Sa – Rb = +1, leads to a quadratic in n, or m
dependent on sign whose discriminant must be the square of a natural number which then gives us the
equation; x 2 – Dy 2 = b, to solve.
Now there must be a prime p = 4k +1, that divides D, hence working in modulo p, b, must be a
quadratic residue, if it is not then there are no solutions and hence no solution to Sa – Rb = +1.
8
The drawback to this idea is if D is large and composite then the number of representations to test may
afford an answer no quicker that calculating the period of the continued fraction representation. But
using this idea it is easy to show 221 is not a D. Working with (a, b) = {(10, 11), (14, 5)} in turn we see
that neither 11, nor 5 are quadratic residues modulo 13, and so 221 is not a D.

REFERENCES

[1] Keith Matthews, Primitive Pythagorean triples and the negative Pell equation, November 16,
2007, www.numbertheory.org/pdfs/negative_pell.pdf

[2] FRANZ LEMMERMEYER, HIGHER DESCENT ON PELL CONICS.II. TWO CENTURIES


OF MISSED OPPORTUNITIES, www.arXiv:Math/0311296v.I [math.NT] 18 Nov 2003

S-ar putea să vă placă și