Vector Spaces in Physics 8/6/2015
Chapter 2. The Special Symbols ij and ijk , the Einstein
Summation Convention, and some Group Theory
Working with vector components and other numbered objects can be made easier (and
more fun) through the use of some special symbols and techniques. We will discuss two
symbols with indices, the Kronecker delta symbol and the Levi-Civita totally
antisymmetric tensor. We will also introduce the use of the Einstein summation
convention.
References. Scalars, vectors, the Kronecker delta and the Levi-Civita symbol and the
Einstein summation convention are discussed by Lea [2004], pp. 5-17. Or, search the
web. One nice discussion of the Einstein convention can be found at
http://www2.ph.ed.ac.uk/~mevans/mp2h/VTF/lecture05.pdf . You may find other of the
lectures at this site helpful, too.
A. The Kronecker delta symbol, ij .
This symbol has two indices, and is defined as follows:
0, i j
ij , i, j 1,2,3 Kronecker delta symbol (2-1)
1, i j
Here the indices i and j take on the values 1, 2, and 3, appropriate to a space of three-
component vectors. A similar definition could in fact be used in a space of any
dimensionality.
We will now introduce new notation for vector components, numbering them rather
than naming them. [This emphasizes the equivalence of the three dimensions.] We will
write vector components as
Ax
Ay Ai , i 1,3 (2-2)
A
z
We also write the unit vectors along the three axes as
iˆ, ˆj, kˆ eˆi , i 1,3 (2-3)
The definition of vector components in terms of the unit direction vectors is
Ai A eˆi , i 1,3 (2-4)
The condition that the unit vectors be orthonormal is
eˆi eˆ j ij (2-5)
This one equation is equivalent to nine separate equations: iˆ iˆ 1 , ˆj ˆj 1 , kˆ kˆ 1 ,
iˆ ˆj 0 , ˆj iˆ 0 , iˆ kˆ 0 , kˆ iˆ 0 , ˆj kˆ 0 , kˆ ˆj 0 !!! [We have now stopped
writing "i,j=1,3;" it will be understood from now on that, in a 3-dimensional space, the
"free indices" (like i and j above) can take on any value from 1 to 3.]
2-1
Vector Spaces in Physics 8/6/2015
Example: Find the value of ˆj kˆ obtained by using equation (2-5).
Solution: We substitute ê2 for ĵ and ê3 for k̂ , giving
ˆj kˆ eˆ2 eˆ3 23 0 ,
correct since ĵ and k̂ are orthogonal.
B. The Einstein summation convention.
The dot product of two vectors A and B now takes on the form
3
A B Ai Bi . (2-6)
i 1
This is the same dot product as previously defined in equation (1-40), except that AxBx
has been replaced by A1B1 and so on for the other components.
Now, when you do a lot of calculations with vector components, you find that the sum of
an index from 1 to 3 occurs over and over again. In fact, occasions where the sum would
not be carried out over all three of the directions are hard to imagine. Furthermore, when
a sum is carried out, there are almost always two indices which have the same value - the
index i in equation (2-6) above, for example. So, the following practice makes the
equations much simpler:
The Einstein Summation Convention. In expressions involving vector or tensor indices,
whenever two indices are the same (the same symbol), it will be assumed that a sum over
that index from 1 to 3 is to be carried out. This index is referred to as a paired index;
paired indices are summed. An index which only occurs once in a term of an expression is
referred to as a free index, and is not summed.
This sounds a bit risky, doesn't it? Will you always know when to sum and when not to?
It does simplify things, though. The reference to tensor indices means indices on
elements of matrices. We will see that this convention is especially well adapted to
matrix multiplication.
So, the definition of the dot product is now
A B Ai Bi , (2-7)
the same as equation (2-6) except that the summation sign is omitted. The sum is still
carried out because the index i appears twice, and we have adopted the Einstein
summation convention.
To see how this looks in practice, let's look at the calculation of the x-component of a
vector, in our new notation. We will write the vector A , referring to the diagram of
figure 1-11, as
2-2
Vector Spaces in Physics 8/6/2015
A iˆAx ˆjAy kˆAz
, (2-8)
eˆi Ai
where in the second line the summation over i = {1,2,3} is implicit. Now use the
definition of the x-component,
Ax A1 A ê1 . (2-9)
Combining (2-8) and (2-9), we have
A1 A eˆ1
(eˆi Ai ) eˆ1
Ai (eˆi eˆ1 ) (2-10)
Ai i1
A1
In the next to last step we used (2-5), the orthogonality condition for the unit direction
vectors.
Next we carried out one of the most important operations using the Kronecker delta
symbol, summing over one of its indices. This is also very confusing to someone seeing
it for the first time. In the last line of equation (2-10) there is an implied summation over
the index i. We will write out that summation term by term, just this once:
Ai i1 A1 11 A2 21 A3 31 (2-11)
Now refer to (2-1), the definition of the Kronecker delta symbol. What are the values of
the three delta symbols on the right-hand side of the equation above? Answer: 11 = 1,
21 = 0, 31 = 0. Substituting these values in gives
Ai i1 A1 11 A2 21 A3 31
A1 1 A2 0 A3 0 (2-12)
A1
What has happened? The index "1" has been transferred from the delta symbol to A.
C. The Levi-Civita totally antisymmetric tensor.
The Levi-Civita symbol is an object with three vector indices,
ijk , i 1, 2,3; j 1, 2,3; k 1, 2,3 Levi-Civita Symbol
1, i, j , k an even permutation of 1, 2,3
(2-13)
ijk 1, i, j , k an odd permutation of 1, 2,3
0 otherwise
All of its components (all 27 of them) are either equal to 0, -1, or +1. Determining which
is which involves the idea of permutations. The subscripts (i,j,k) represent three
numbers, each of which can be equal to 1, 2, or 3. A permutation of these numbers
scrambles them up, and it is a good idea to approach this process systematically. So, we
are going to discuss the permutation group.
2-3
Vector Spaces in Physics 8/6/2015
Groups. A group is a mathematical concept, a special kind of set. It is defined as
follows:
Definition: A group G is a set of objects A, B, C ,... with multiplication of one
member by another defined, closed under multiplication, and with the additional
properties:
(i) The group contains an element I called the identity, such that, for every
element A of the group,
AI IA A (2-14)
(ii) For every element A of the group there is another element B, such that
AB BA I . (2-15)
B is said to be the inverse of A:
A B 1 . (2-16)
(iii) Multiplication must be associative:
A BC AB C . (2-17)
There is an additional property which only some groups have. If multiplication is
independent of the order in the product, the group is said to be Abelian.
Otherwise, the group is non-Abelian.
AB BA Abelian group . (2-18)
This may seem fairly abstract. But the members of groups used in physics are usually
operators, operating on interesting things, such as vectors or members of some other
vector space. Right now we are going to consider permutation operators, operating on
sets of indices.
The Permutation Group. We will start by defining the objects operated on, then the
operators themselves. Consider the numbers 1, 2, and 3, in some order, just like the
indices on the Levi-Civita symbol:
( a , b, c ) . (2-19)
Here each letter represents one of the numbers, and they all three have to be represented.
It is pretty easy to convince yourself that the full set of possibilities is
(a, b, c) 1, 2,3 , 1,3, 2 , 2,1,3 , 2,3,1 , 3,1, 2 , 3, 2,1 . (2-20)
Now the permutation group of the third degree consists of operators which re-arrange the
three numbers as follows:
P a, b, c a, b, c P123 I
P a, b, c a, c, b P132
P a, b, c b, a, c P213
. (2-21)
P a, b, c b, c, a P231
P a , b, c c , a , b P
312
P a, b, c c, b, a P321
2-4
Vector Spaces in Physics 8/6/2015
The second form of the notation shows where the numbers (1,2,3) would end up under
that permutation. The first entry is the permutation which doesn't change the order,
which is evidently the identity for the group. The group consists of just these six
members
Examples of permutations operating on triplets of indices:
P123 2,3,1 2,3,1
P132 2,3,1 2,1,3
. (2-22)
P321 2,3,1 1,3, 2
P132 P321 2,3,1 1, 2,3
Do you follow the fourth line? First the permutation P321 is carried out, giving (1,3,2);
and then the permutation P132 operates on this result, giving (1,2,3). This brings us to the
subject of multiplication of group elements. This fourth line shows us that the product of
the given two permutation-group elements is itself a permutation, namely
P132 P321 P312 . (2-23)
Try this yourself, and verify that
P312 2,3,1 1, 2,3 . (2-24)
From this example, it is pretty clear that the group of six elements given above is closed
under multiplication. There is an identity, the permutation which doesn't change the
order. And it is pretty easy to identify the inverses within the group.
Example: Show that
P3121 P231 .
Proof: Try it out on the triplet (a,b,c):
p312 a, b, c c, a, b
.
P231 c, a, b a, b, c
The inverse permutation P231 just reverses the effect of P312.
There are some simpler permutation operators related to the Pijk, the binary permutation
operators which just interchange a pair of indices, while leaving the third one unchanged.
P12 a, b, c b, a, c
P13 a, b, c c, b, a . (2-25)
P a, b, c a, c, b
23
It is easy to see that the six group members given in equation (2-21) can be written in
terms of the binary permutation operators:
2-5
Vector Spaces in Physics 8/6/2015
P123 I P12 2 P132 P232
P132 P23
P213 P12
. (2-26)
P231 P12 P13
P312 P13 P12
P321 P13
(Remember, the right-hand operator in a product operates first.)
There is a special subset of permutations of a series of objects called the circular
permutations, where the last index comes to the front and the others all move over one
(see Figure 2-1).
(a,b,c,d,e,f,g) ---> (g,a,b,c,d,e,f)
Figure 2-1. A circular permutation.
For the six objects listed in eq. (2-21), three of them are circular permutations of (1,2,3),
namely
(a, b, c)circular 1, 2,3 , 3,1, 2 , 2,3,1 . (2-27)
Each of these is produced by an even number of binary permutations, and the other three
are produced by an odd number of binary permutations. So, the group divides up into
three "even permutations" and three "odd permutations:"
P123
P312 even permutations
P231
. (2-28)
P213
P odd permutations
132
P321
The Levi-Civita symbol. Now we can finally use the idea of even and odd permutations
to define the Levi-Civita symbol:
1, (ijk) an even permutation of (123)
ijk - 1, (ijk) an odd permutation of (123) Levi - Civita Symbol (2-29)
0 otherwise
Notice that there are only six non-zero symbols, three equal to +1 and three equal to -1.
And any binary permutation of the indices (interchanging two indices) changes the sign.
This is the key property in many calculations using the Levi-Civita symbol.
Example: Give the values of 312, 213, and 322,
2-6
Vector Spaces in Physics 8/6/2015
Answer: (312) is an even permutation of (123), so 312 = +1. (213) is obtained
from (312) by permuting the first and last numbers, so it must be an odd
permutation, and 213 = -1. (322) is not a permutation of (123) at all, so 322 = 0.
Question: Is the permutation group Abelian? What about the subgroup consisting of the
three circular permutations? Answering these questions will be left to the problems.
D. The cross product.
In the last chapter we found the following result for the cross product of two vectors A
and B in terms of their components:
A B iˆ( Ay Bz Az By )
ˆj ( Az Bx Ax Bz ) (1-48)
kˆ( A B A B )
x y y x
Notice that there are a lot of permutations built into this definition. In particular, each
term involves a permutation of (x,y,z), with the first letter indicating the unit vector, the
second, the component of A, and the third, the component of B. Here is an elegant way of
re-writing this expression using the Levi-Civita symbol:
A B i ijk A j Bk (2-30)
It may be less than obvious at first glance that (2-30) is the equivalent of (1-48). First
let's just examine the index structure of the expression. The left-hand side has a single
unpaired, or free, index, i. This means that it represents any single one of the components
of the vector A B - we would say that it gives the i-th component of A B . Now look
at the right-hand side. There is only one free index, and it is i, the same as on the left-
hand side. This is the way it has to be. In addition, there are two paired indices, j and k.
These have to be summed. If we were not using the Einstein summation convention, this
3 3
expression would read A B i ijk A j Bk . We have decided to follow the
j 1 k 1
Einstein convention and so we will not write the summation signs. However, for any
given value of i, there are nine terms to evaluate.
To see exactly how this works out, let's evaluate the result of (2-30) for i=2. This
should give the y-component of A B . Here it is:
A B 2 2 jk A j Bk
211 A1 B1 212 A1 B2 213 A1 B3
(2-31)
221 A2 B1 222 A2 B2 223 A2 B3
231 A3 B1 232 A3 B2 233 A3 B3
But, most of these terms are equal to zero, because two of the indices on the Levi-Civita
symbol are the same. There are only two non-zero L.-C. symbols: 213 = -1, and 231 =
+1. Using these facts, we arrive at the answer
2-7
Vector Spaces in Physics 8/6/2015
A B
A1 B3 A3 B1
2 (2-32)
This is the same as the y-component of equation (1-48), if the correct substitutions are
made for numbered instead of lettered components. So, the two versions of the cross
product agree.
Example: Use the tensor form of the cross product, equation (2-30), to prove that
A B B A .
Proof: There was a similar relation for the dot product - but with a plus sign!
Let's see
how this works in tensor notation:
A B i ijk A j Bk definition of the cross product
ikj A j Bk permuting two indices of ε gives a minus sign
ikj Bk A j the A j and the Bk can be written in any order
B A i definition of the cross product
In regard to the last step of this example, it is worth remarking that particular name
given to a summed index doesn't matter - it is sort of like the dummy variable inside a
definite integral. What matters in the definition of the cross product A B is that the
index of the components of A match with the second index of , and the index of the
components of B, with the third index of .
E. The triple scalar
product. height of
BxC A rectangular
There is a famous way of solid
area of parallelogram =
making a scalar out of three
magnitude of BxC.
vectors. It is illustrated in
figure 2-2, where the vectors C
A , B and C form the three
independent sides of a B
parallelopiped. The cross
Figure 2-2. A parallelepiped, with its sides defined by
product of B and C gives vectors A, B and C. The area of the parallelogram forming
the area of the base of the the base of this solid is equal to BC sin , where is the
parallelopiped (a angle between B and C. This is just the magnitude of the
parallelogram), and dotting cross product BxC. When BxC is dotted into A, the area of
with A gives the volume: the base is multiplied by the height of the solid, giving its
volume.
Volume A B C . (2-31)
Putting in the forms of the dot and cross product using , we have
2-8
Vector Spaces in Physics 8/6/2015
Volume A B C
A BC
Ai B C i . (2-33)
Ai ijk B j Ck
ijk Ai B j Ck
There is an identity involving the triple scalar product which is easy to demonstrate from
this form:
A B C B C A C A B . (2-34)
In the geometrical interpretation of the triple scalar product, these three forms correspond
to the three possible choices of which side of the parallelepiped to call the base (see
figure 2-2).
F. The triple vector product.
There is another way to combine three vectors, this time giving another vector:
D A B C (triple vector product) (2-35)
In tensor notation this becomes
D i A B C i
ijk A j B C k (2-36)
ijk A j klm Bl Cm
This is not very encouraging. It is not simple, and furthermore it conjures up the prospect
of more cross products. Do we have to live in dread of D A B C and all of her big
sisters?
The Epsilon Killer. Happily there is a solution. There is an identity which guarantees
that there will never be more than one ijk in an expression, by reducing a product of two
epsilons to Kronecker deltas. Here it is:
ijk ilm jl km jm kl (2-37)
This is the epsilon killer! Here are the important structural features. There are two
epsilons, with the first index of each one the same, so there is a sum over that index. The
other four indices (two from each epsilon) are all different, and so are not summed. We
will not prove eq. (2-37), but it is not too difficult, if you just consider all the possibilities
for the indices.
We will now use this identity to simplify the expression for the vector triple product:
A B C i ijk A j klm Bl Cm definition of cross product
kij klm A j Bl Cm cyclic permutation of indices
(2-38)
( il jm im jl ) A j Bl Cm use epsilon killer
Am Bi Cm Al Bl Ci sum over paired indices of deltas
2-9
Vector Spaces in Physics 8/6/2015
The last step has used the "index transferring" property of a sum over one index of a delta
symbol illustrated in equation (2-12). In the last line of (2-38) we can see two sums over
paired indices, AmCm A C and Al Bl A B. This gives
A B C i Bi A C Ci A B (2-39)
or, in pure vector form,
A B C B A C C A B (2-40)
This is sometimes referred to as the "BAC - CAB" identity. It occurs regularly in
advanced mechanics and electromagnetic theory.
PROBLEMS
In the problems below, repeated indices imply summation, according to the Einstein
summation convention. Sum from 1 to 3 unless otherwise stated.
Problem 2-1. Consider ij and ijk as defined in the text, for a three-dimensional space.
(a) How many elements does ij have? How many of them are non-zero?
(b) Give the following values:
11
23
31
(c) How many elements does ijk have? How many are equal to zero? Which elements
are equal to -1?
(d) Give the following values:
111
321
123
132
Problem 2-2. Evaluate the following sums, implied according to the
Einstein Summation Convention.
ii
12 j j 3
12 k 1k
1 jj
Problem 2-3. Consider a possible group of permutations operating on three indices, but
consisting of only the two members
I , P12 (3-25)
(a) Is this set of operators closed under multiplication? Justify your answer.
2 - 10
Vector Spaces in Physics 8/6/2015
(b) Is this set of operators Abelian? Justify your answer.
Problem 2.4. Consider a possible group of permutations operating on three indices, but
consisting of only the four members
I , P12 , P13 , P23 (3-25)
(a) Is this set of operators closed under multiplication? Explain your answer.
(b) Is this set of operators Abelian? Explain your answer.
Problem 2.5. Consider the full permutation group, operating on three indices.
(a) Is the group Abelian? Explain your answer.
(b) What about the subgroup consisting of just the two circular permutations (and the
identity)? Explain your answer.
[You might approach these questions by simply trying two successive permutations, and
then reversing the order.]
Problem 2-6. Assume that the cross product D A B is defined by the relation
Di A B ijk Aj Bk .
i
Show using tensor notation (rather than writing out all the terms) that the magnitude of
this vector agrees with the geometrical definition of the cross product. That is, show that
D has a magnitude equal to |ABsin|. [Hint: Evaluate D D using the ''epsilon-killer''
identity.]
Problem 2-7. Use tensor notation (rather than writing out all the terms) to prove the
following identity for three arbitrary vectors A , B , and C .
A B C B C A C A B
Problem 2-8. (a) Use tensor notation (rather than writing out all the terms) to prove the
following identity for two arbitrary vectors A and B .
A A B 0
[Hint: Use the symmetries of the Levi-Civita symbol to prove that
A A B A A B . This
implies that both sides of the equation are equal to zero.]
(b) Make a geometrical argument, based on the direction of A B , to show that this
identity has to be satisfied.
Problem 2-9. Let eˆi , i 1,2,3 be the usual three directional unit vectors of a 3-
dimensional Cartesian coordinate system, satisfying the orthonormality relation
eˆi eˆ j ij . In terms of components, A and B can be written as
2 - 11
Vector Spaces in Physics 8/6/2015
A Am eˆm ,
B B j eˆ j .
Using these definitions for A and B and using tensor notation, show that
A B Ai Bi .
2 - 12
