Index notation in 3D

1 Why index notation?

Vectors are objects that have properties that are independent of the coordinate system that they are written in. Vector notation is advantageous since it is elegant and deals with vectors as single objects rather than collections of components. This helps to make the physical and geometric meaning of equations manifest. However vector notation has some diculties, a major one being that there is a whole heap of vector algebraic and dierential identities that are hard to remember and hard to derive using vector notation. For example: a (b c) = b (c a) = c (a b) a (b c) = b(c a) c(a b) (a b) (c d) = (a c)(b d) (a d)(b c) (a b) (c d) = (a c d)b (b c d)a ( ( f =0 a) b+b bb )b + ( b b) b(
2

(1.1) (1.2) (1.3) (1.4) (1.5) (1.6)

a) = 0 a) = (b) = (b) = (a b) = (a ( a

(1.7) (1.8) )a + a ( b) )a (a )b . b) + b ( a) (1.9) (1.10) (1.11) (1.12)

(a b) = b ( (a b) = a(

a) a (

a) + (b

Most of these identities you should have seen before and all of these identities are used throughout physics, e.g. in classical mechanics, quantum mechanics, electrodynamics etc. (A similar notation is also used in general relativity and other more advanced topics). When using index notation to represent the above identities the proofs become very simple. Thus you dont even need to remember the identities, you can just rederive whichever result you happen to need as you go. To move from vector notation to an indexed notation we just need to choose an orthonormal basis in which to expand the vectors. Then any vector can be written as a triple in that basis, for example in Cartesian coordinates a vector can be represented as v = (vx , vy , vz ) = vx x + vy y + vz z . In general we choose the basis {ei , i = 1, 2, 3} and write v = (v1 , v2 , v3 ) = v1 e1 + v2 e2 + v3 e3 . Orthonormal means that e1 e1 = e2 e2 = e3 e3 = 1 and e1 e2 = e2 e3 = e3 e1 = 0 . 1

The trick to index notation is to write any vector as an indexed symbol: so the vector v R3 is represented by vi R where i takes the values 1, 2 or 3. Then all of the operations of vector calculus can be written using indexed objects. 2 2.1 Algebraic relations Free and dummy indices

Any index that occurs precisely once on each side of an equation is a free index. An equation with a free index represents a vector equation, as it has to be satised for each value the index can take. So for example if A and B are vectors then Ai = Bi A = B . For any index that is repeated on one side of the equation, we normally use the Einstein summation convention. That is we sum over all possible values of that index (in 3D that is 1,2 and 3). For example
3

Ai Bij =
i=1

Ai Bij .

These indices are then called dummy indices because they just stand in for the summation index and can be replaced with any other index without changing the meaning (just like integration variables are dummy varibles). 2.2 Dot product

To write the dot product in index notation we need the Kronecker symbol. It is just the identity matrix, 1, written in an indexed form: 1 0 . ij = 0 1 . . . (2.13) . .. .. . . . . Then we see that the orthonormality condition on the basis vectors can be written as ei ej = ij . Using the linearity of the dot product we can now see how it works in index notation: a b = (ai ei ) (bj ej ) = ai bj (ei ej ) = ai ij bj . (2.15) (2.14)

Since ij j = i (which can be seen from the denition above or from the vector equivalent 1b = b) we see the dot product take the normal form of the sum of the components: a b = ai ij bj = ai bi . 2 (2.16)

The symmetry of the dot product can be seen either from the symmetry of ij combined with the fact that the components ai and bj are just numbers and therefore commute. The magnitude of a vector is just |a|2 = a a = ai ai . 2.3 Cross product

Once again we expand two vectors wrt the basis and use linearity to get a b = ai bj (ei ej ) . we know that the cross product of a vector times itself is zero, and using the standard denition of the cross product (right hand rule etc) we see that e1 e2 = e3 , e2 e3 = e1 , e3 e1 = e2 ,

the others being obtained via antisymmetry of the cross product. This can be summarised by dening a new symbol ijk such that ei ej = ijk ek . This symbol is known as the Levi-Civita symbol, it is also called the totally antisymmetric tensor or the permutation symbol. It is dened by: 0 if any of i, j or k are the same 123 ijk = 1 if ijk is an even permutation of 123 = sgn . (2.17) i j k 1 if ijk is an odd permutation of 123. An equivalent denition is ijk = jki = jik , 123 = 1 . (2.18)

We can illustrate this by looking at, for example, the rst component. So setting i = 1 and expanding the implied summation (a b)1 = 1jk aj bk = 123 a2 b3 + 132 a3 b2 + . . . = a2 b 3 a3 b 2 , where the dots are terms with repeated indices on the Levi-Civita symbol and thus zero. The other two components can be similarly found. The basic antisymmetry of the cross product can be thought of in terms of the antisymmetry of ijk (a b)i = ijk aj bk = ikj aj bk = ikj bk aj relabelling the dummy indices to get (a b)i = ijk bj ak = (b a)i . 3

2.4

Relating and

The basic identity that relates these two symbols is il jl kl = im jm km , in jn kn

ijk lmn

(2.19)

but this is over-kill. The only identity that is normally needed is obtained by setting k = l above to get ijk kmn = im jn in jm . (2.20) This identity can be derived/checked via brute force (try a few cases) or by arguments based on invariant tensors under spatial rotations. It is also good to note that it respects all of the symmetries that it should, i.e. both sides are antisymmetric under exchange of i j or equivalently m n. 2.5 Scalar triple product and determinants

The scalar triple product is easily seen to be a (b c) = ai (ijk bj ck ) = ijk ai bj ck and by cycling the indices ijk ai bj ck = kij ck ai bj = jki bj ck ai , we get the identities a (b c) = c (a b) = b (c a). (2.22) The determinant of a 3 3 matrix can also be written in indicial form and can be shown to be a1 a2 a3 b1 b2 b3 = a (b c) = ijk ai bj ck . c1 c2 c3 2.6 Vector triple product (2.23) (2.21)

The identity we want to prove is a (b c) = b(a c) c(a b) . Diving in (a (b c))i = ijk aj kmn bm cn = (im jn in jm )aj bm cn = bi aj cj ci aj bj = (b(a c) c(a b))i , we see that its just a simple consequence of the identity (2.20). 2.7 More vector products (2.24)

Products of more vectors can always be reduced down to repeated applications of the two triple products above, but they can still be computed directly (and eciently) using index notation. 4

Vector Calculus

The basic element of vector calculus is the operator. It is a vector operator and is sometimes written as . Its dening property is that for any unit vector s and function f (x, y, z) = f (r), s f is the partial derivative of f in the direction of s . Its components, with r = (x, y, z), = i i = ri take partial derivatives in the ri direction. Since each component of i is just a rst order partial derivative, then each component must obey the Leibniz (product) rule. This implies that itself obeys the Leibniz rule. For any scalar functions f = f (r) and g = g(r),
i (f g)

=(

if )

g+f

i g.

(3.25)

The familiar gradient, divergence and curl are written as grad(f ) = div(v) = curl(v) = f=
i vi if

(3.26) (3.27) (3.28)

v =

and
j vk

v = ijk

where we introduced the notational convenience v(r) = v = vi . An important (and obvious) result is rj = ij . (3.29) i rj = ri For any constant vector b this immediately leads to (r b) =
i rj bj

= ij bj = bi = b ,

(3.30)

it also implies that curl(r) = 0 (prove me!). 3.1 Five Leibniz type identities

All of these proofs can be done using only the denitions and properties of ij and ijk and the identities (3.25) and (2.20). The rst three are trivial: (ab) = a (a b) = b (ab) = a To prove (a b) = (b we write LHS = ijk = bj
j (klm al bm ) j ai

b+b aa bb b) (a

a b a. )b b( a) ,

(3.31) (3.32) (3.33) (3.34)

)a + a(

= (il jm im jl )(bm
j bi

j al

+ al

j bm )

+ ai

j bj

aj

bi

j aj

= RHS .

The proof of the last identity is left as an exercise: (a b) = a ( b) + (a )b + b ( a) + (b )a (3.35)

Note that this identity has (3.30) as a special case. 3.2 Repeated derivatives

Assuming we are working in the space of smooth functions, all partial derivatives commute: i jf = j i f . Using this, the standard identities are easy to prove: curl gradf = div curl v = curl curl v = ( f) = 0 , ( v) = 0 , v =
2

(3.36) (3.37) v) .
2

( =

v = grad div v v .

(3.38)

where the Laplacian is dened as = 4 Exercises

Anything that is stated but not derived in these notes! Also you could try: div(r) = grad(|r|) = r = ... |r| = . . . (4.39) (4.40) (4.41) etc . (4.42)

f (|r|) = . . . a(|r|) = . . .

In the Kepler problem of classical mechanics there is one conserved scalar, the energy, and two conserved vectors, the angular momentum (L = r p) and the Laplace-Runge-Lenz vector. An important component of the LRL vector is p L. Show that it is equal to |p|2 r (p r)p. In quantum mechanics the momentum operator is p = i L = r p = ir . Prove that (r a) = a + r( a) + i(L a) . so the angular momentum is