MASSACHUSETTS INSTITUTE OF TECHNOLOGY

Physics Department
Physics 8.07: Electromagnetism II October 11, 2012
Prof. Alan Guth
LECTURE NOTES 8
THE TRACELESS SYMMETRIC TENSOR EXPANSION
AND STANDARD SPHERICAL HARMONICS
These notes are an addendum to Lecture 14, Wednesday October 10, 2012. The
notes will describe a topic that I did not have time to include in the lecture: the relation
between the traceless symmetric tensor expansion and the standard spherical harmonics.
Using traceless symmetric tensors, we can expand any function of angle as
∞

∞


(
)
F (n̂) = Ci1 i2 ...i
n ˆ i2 . . . n̂i
≡
ˆ i1 n F
(n̂)

=1
=0 (8.1)
(1) (2) (3)
= C (0) + Ci n̂i + Cij n̂i n̂j + Cijk n̂i n̂j n̂k + . . . ,

(
)
where the Ci1 i2 ...i
are traceless symmetric tensors, the indices i1 , i2 , . . . i
are summed
from 1 to 3 as Cartesian indices, and

n̂(θ, φ) = sin θ cos φ ê1 + sin θ sin φ ê2 + cos θ ê3 , (8.2)

where ê1 , ê2 , and ê3 can also be written as êx , êy , and êz .
In the more standard approach, an arbitrary function of (θ, φ) is expanded in spher-
ical harmonics:
∞




F (n̂) = a
m Y
m (θ, φ) . (8.3)

=1 m=−

We have shown that

 
2 (
) (
)
∇θ Ci1 i2 ...i
n
ˆ i1 n̂i2 . . . n̂i
= −( + 1)Ci1 i2 ...i
n
ˆ i1 n̂i2 . . . n̂i
, (8.4)

where  
1 ∂ ∂ 1 ∂2
∇θ2 = sin θ + . (8.5)
sin θ ∂θ ∂θ sin2 θ ∂φ2
In the standard approach one would show that

∇2θ Y
m (θ, φ) = −( + 1)Y
m (θ, φ) , (8.6)

so  apparently has the same meaning in both formalisms. (I am not trying here to de-
rive the standard formalism, but instead I will simply adopt the equations from standard
textbooks, and show that we can express these functions in terms of traceless symmetric
8.07 LECTURE NOTES 8, FALL 2012 p. 2

tensors. A good example of such a standard textbook is J.D. Jackson, Classical Elec-
trodynamics, 3rd Edition (John Wiley & Sons, 1999), Sections 3.1, 3.2, 3.5, and 3.6.
That means that there must be some particular traceless symmetric tensor, which we will
(
,m)
call Ci1 ...i
which is equivalent to Y
m (θ, φ). That is,

(
,m)
Ci1 ...i
n̂i1 . . . n̂i
= Y
m (θ, φ) . (8.7)

(
,m)
Our goal is to construct Ci1 ...i
explicitly. We have already shown that the number of
linearly independent traceless symmetric tensors of rank  (i.e., with  indices) is given
by 2 + 1, which is not surprisingly equal to the number of Y
m functions for a given .
The quantity m is an integer from − to , so there are 2 + 1 possible values.
We consider first the case of azimuthal symmetry, where F (n̂) is invariant under
rotations about the z-axis, and hence independent of φ. In that case, within the standard
treatment, the most general function can be expanded in Legendre polynomials,
∞


F (n̂) = a
P
(cos θ) . (8.8)

=0

The P
functions are the same as the Y
0 functions, except that they are normalized
differently: 
2 + 1
Y
0 (θ, φ) = P
(cos θ) . (8.9)
4π
The Legendre polynomials can be written explicitly using Rodrigues’ formula:
 

1 d  
P
(x) =
(x2 − 1)
. (8.10)
2 ! dx

In the traceless symmetric tensor formalism, the azimuthal symmetry case must be de-
scribed by traceless symmetric tensors that are invariant under rotations about the z-axis.
(1)
It is easiest to begin by thinking about  = 1, where we are seeking a tensor Ci . Since
(1)
Ci has one index, it is a vector, which is the same as a tensor of rank 1. It is obvious
that the only vector that is invariant under rotations about the z-axis is a vector that
points along the z axis. I will let ẑ be a unit vector in the z-direction (which I have also
called êz and ê3 ), and then for azimuthal symmetry we have

(1)
Ci = const ẑi , (8.11)

where ẑi = δi3 is the i’th component of ẑ. The resulting function of n̂ is then

(1)
F1 (n̂) = Ci n̂i = const ẑi n̂i = const ẑ · n̂ = const cos θ , (8.12)
8.07 LECTURE NOTES 8, FALL 2012 p. 3

which certainly agrees with P1 (cos θ) = cos θ.

To generalize to arbitrary , we can construct a tensor of rank  that is invariant
under rotations about the z-axis by considering the product ẑi1 ẑi2 . . . ẑi
. This is clearly
symmetric, but it is not traceless. However, we can make it traceless by taking its traceless
part, which I denote by curly brackets.

{ ẑi1 ẑi2 . . . ẑi
} ≡ traceless symmetric part of ẑi1 ẑi2 . . . ẑi
. (8.13)

The traceless symmetric part is constructed by starting with the original expression and
then subtracting terms proportional to one or more Kronecker δ-functions, where the
subtractions are uniquely determined by the requirement that the expression be traceless.
For example,

{1} = 1
{ ẑi } = ẑi
{ ẑi ẑj } = ẑi ẑj − 31 δij

(8.14)
{ ẑi ẑj ẑk } = ẑi ẑj ẑk − 15 zˆi δjk + ẑj δik + ẑk δij

{ ẑi zˆj ẑk ẑm } = ẑi ẑj ẑk ẑm − 17 ẑi ẑj δkm + ẑi ẑk δmj + ẑi ẑm δjk + ẑj ẑk δim

1

+ ẑj ẑm δik + ẑk ẑm δij + 35 δij δkm + δik δjm + δim δjk ,

where the coefficients are all determined by the requirement of tracelessness. We will
argue later that Eq. (8.14) gives the only traceless symmetric tensors that are invariant
under rotations about the z-axis, and therefore the function

F
(n̂) = { ẑi1 . . . ẑi
} n̂i1 . . . n̂i
(8.15)

is the only function, up to a multiplicative constant, that is azimuthally symmetric and

satisfies ∇2θ F
(n̂) = −(+1)F
(n̂). Since P
(cos θ) is azimuthally symmetric and satisfies
∇2θ P
(cos θ) = −( + 1)P
(cos θ), we must have

P
(cos θ) = const{ ẑi1 . . . ẑi
} n̂i1 . . . n̂i
, (8.16)

where the constant is yet to be determined.

Both sides of Eq. (8.16) are polynomials in cos θ, where the highest power is cos
θ.
If we can find the coefficients of this highest power on each side of the equation, we
can determine the constant. On the right-hand side, the highest power comes entirely
from the ẑi1 . . . ẑi
term in { ẑi1 . . . ẑi
}, since all the other terms contain Kronecker δ-
functions which result in factors of the form n̂ · n̂ = 1, reducing the number of n̂ factors
available to give powers of cos θ. So, the leading term on the right-hand side is simply
8.07 LECTURE NOTES 8, FALL 2012 p. 4

const(ẑ · n̂)
= const cos
θ. For the left-hand side, we can use Rodrigues’ formula to
extract the highest power:
 

1 d  2 
P
(x) =
(x − 1)

2 ! dx
 

1 d  2

=
x + (lower powers)
2 ! dx
 
−1
1 d  
=
(2)x2
−1 + (lower powers)
2 ! dx (8.17)
 
−2
1 d  
= (2)(2 − 1)x2
−2 + (lower powers)
2
! dx
1 


= (2)(2 − 1) . . . ( + 1)x + (lower powers)
2
!
(2)!

=
x + (lower powers) .
2 (!)2
Matching these coefficients, we see that
(2)!
P
(cos θ) = { ẑi1 . . . ẑi
} n̂i1 . . . n̂i
. (8.18)
2
(!)2

Now we can return to the general case, in which there is no azimuthal symmetry,
and the expansion requires the spherical harmonics, Y
m . The Y
m are chosen to have a
very simple dependence on φ, namely

Y
m (θ, φ) ∝ eimφ . (8.19)

This property can be described in terms of how the functions transform under a rotation
of the coordinate system about the z-axis. Under a rotation by an angle ψ about the
z-axis, the angle φ changes by ψ, and Y
m changes by a factor eiψ . I have not been
careful here about specifying the sign of this rotation, because it will be easy to fix the
sign conventions at the end. The important point here is that if we want to match the
conventions of the spherical harmonics, we need to construct traceless symmetric tensors
that are modified by a rotation only by a multiplicative phase factor. That is, we are
looking for tensors that are complex, and that are eigenvectors of the rotation operator.
Naturally we begin by considering a vector (a tensor with one index, or a rank 1
tensor), which under a rotation about the z-axis transforms as

vx = vx cos ψ − vy sin ψ

(8.20)
vy = vx sin ψ + vy cos ψ .
8.07 LECTURE NOTES 8, FALL 2012 p. 5

We thus seek an eigenvector of the matrix

 
cos ψ − sin ψ
R= . (8.21)
sin ψ cos ψ

The eigenvalues λ of the matrix are determined by the characteristic equation

det(R − λI) = 0 , (8.22)

where I is the identity matrix, which can be expanded as

 
cos ψ − λ − sin ψ
det = 0 =⇒ λ2 − 2λ cos ψ + 1 = 0 , (8.23)
sin ψ cos ψ − λ

for which the solutions are


λ = cos ψ ± cos2 ψ − 1 = cos ψ ± i sin ψ = e±iψ . (8.24)

The eigenvectors then satisfy

  
cos ψ − e±iψ − sin ψ vx
=0, (8.25)
sin ψ cos ψ − e±iψ vy

which simplifies to   
∓i sin ψ − sin ψ vx
=0, (8.26)
sin ψ ∓i sin ψ vy
from which we see that vy = ∓ivx . Constructing normalized eigenvectors, we can define

1
û(1) ≡ û+ = √ (êx + iêy )
2
(8.27)
1
û(2) ≡ û− = √ (êx − iêy ) ,
2
which are orthonormal in the sense that

û(i)∗ · û(j) = δij . (8.28)

We can complete a basis for three-dimensional vectors by adding

û(3) ≡ ẑ = êz . (8.29)

You might ask how one should visualize a vector with imaginary components. What
direction does it point? It certainly points in a definite direction in complex three-
dimensional space, which is equivalent to a six-dimensional real-valued space, but for our
8.07 LECTURE NOTES 8, FALL 2012 p. 6

purposes we do not need to have any geometric picture of these vectors. We are simply
going to use them to form dot products to construct (complex-valued) functions of θ and
φ.
Note that complex conjugation, as used in Eq. (8.28), is essential for defining a
positive definite norm for complex vectors. The quantities û(+) · û(+) and û(−) · û(−) , by
contrast, are in fact equal to zero. This leads to the convenient fact that

û+ +
i1 . . . ûi
(8.30)

is both traceless and symmetric.

Since
1
û+ · n̂ = √ sin θeiφ , (8.31)
2
we can construct functions proportional to eimφ , for m > 0, by including m factors of
û+ , and arranging for each of them to be dotted into n̂. This can be done by considering
the function defined by

F
m (θ, φ) ≡ { û+ +

i1 . . . ûim ẑim+1 . . . ẑi
} n̂i1 . . . n̂i
. (8.32)

To see that in this expression every û+ is dotted into an n̂, recall that û+ ·ẑ = û+ ·û+ = 0.
So, when the right-hand side is expanded and all the indices are summed to give dot
products, the only terms that survive are those for which every û+ is dotted into ẑ.
Thus, the right-hand side of Eq. (8.32) is proportional to eimφ . From Eq. (8.4), we know
that the right-hand side of Eq. (8.32) is an eigenfunction of ∇2θ with eigenvalue −( + 1).
I will argue below that any such eigenvector that is proportional to eimφ is necessarily
proportional to Y
m . We will return to the question of uniqueness, but let us first assume
that uniqueness holds, so that

F
m (θ, φ) ∝ Y
m (θ, φ) . (8.33)

As in the previous derivation for Legendre polynomials, we can determine the constant
of proportionality by matching the leading term in the expansions of both sides of the
equation. F
m (θ, φ) can be written as (sin θ)m eimφ times a polynomial in cos θ, so we
can use the highest power of cos θ to determine the matching.
It is easy to extract the leading term from Eq. (8.32), because it comes from the first
term in the expansion of

{ û+ + +
ˆ+
i1 . . . ûim ẑim+1 } = ûi1 . . . u ˆim+1 . . . ẑi
+ terms ∝ δip iq .
im z (8.34)

The first term gives the highest power of cos θ, because the Kronecker δ-functions that
appear in all later terms cause one or more n̂’s to dot with other n̂’s, reducing the number
of n̂’s available to appear in the form n̂ · ẑ = cos θ. Thus,
 
F
m (θ, φ) = 2−m/2 (sin θ)m eimφ (cos θ)
−m + (lower powers of cos θ) . (8.35)
8.07 LECTURE NOTES 8, FALL 2012 p. 7

To compare Eq. (8.35) with the leading term in the expansion for the standard
function Y
m , we need a formula for Y
m (θ, φ). It is given in Jackson as Eq. (3.53),
p. 108, as 
2 + 1 ( − m)! m
Y
m (θ, φ) = P (cos θ)eimφ , (8.36)
4π ( + m)!

where P
m (cos θ) is the associated Legendre function, which can be defined by Jackson’s
Eq. (3.50),
m (−1)m 2 m/2 d

+m
P
(x) =

(1 − x )
+m
(x2 − 1)
. (8.37)
2 ! dx
Using the same technique as in Eq. (8.17), we find

d
+m 2
(x − 1)
= (2) . . . ( + 1)( − 1) . . . ( − m + 1)x
−m + (lower powers)
dx
+m
(8.38)
(2)!
−m
= x + (lower powers) .
( − m)!
Matching the coefficients of these leading terms, we find that we can write (for m ≥ 0)
(
,m)
Y
m (θ, φ) = Ci1 ...i
n̂i1 . . . n
ˆ i
, (8.39)

where
(
,m)
ˆ+
Ci1 i2 ...i
= d
m { u ˆ+
i1 . . . u ˆim+1 . . . zˆi
} ,
im z (8.40)
with 
(−1)m (2)! 2m (2 + 1)
d
m = . (8.41)
2
! 4π ( + m)! ( − m)!

For negative values of m, the calculation is identical, except that we use û− instead
+
of û . The result is
(
,m) (
,|m|)∗
Ci1 i2 ...i
= d
m { û− −
ˆi|m|+1 . . . zˆi
} = Ci1 i2 ...i
,
i1 . . . ûi|m| z (8.42)

where to allow for negative m we need to write d
m as


(−1)m (2)! 2|m| (2 + 1)
d
m = . (8.43)
2
! 4π ( + m)! ( − m)!

It is worth mentioning that the curly brackets indicating “traceless symmetric part”
can be put on either factor or both in expressions such as Eq. (8.32). That is,

{ û+ + + +
i1 . . . ûim ẑim+1 . . . ẑi
} n̂i1 . . . n̂i
= { ûi1 . . . ûim ẑim+1 . . . ẑi
} { n̂i1 . . . n̂i
} (8.44a)

= û+ ˆ+
i1 . . . u ˆim+1 . . . ẑi
{ n̂i1 . . . n̂i
} ,
im z (8.44b)
8.07 LECTURE NOTES 8, FALL 2012 p. 8

where the top line is justified because { n̂i1 . . . n̂i
} differs from n̂i1 . . . n̂i
only by terms
proportional to Kronecker δ-functions, which give no contribution when summed with
the traceless symmetric tensor { û+ +
i1 . . . ûim ẑim+1 . . . ẑi
}. Similarly, once the second
factor is written in traceless symmetric form, there is no longer a need to take the
traceless symmetric part of the first term, since { û+ ˆ+
i1 . . . u ˆim+1 . . . zˆi
} differs from
im z
+ +
ûi1 . . . u
ˆim zˆim+1 . . . ẑi
only by terms proportional to Kronecker δ-functions, which vanish
when summed with the traceless symmetric tensor { n̂i1 . . . n̂i
}.
Finally, we can return to the question of uniqueness. In asserting that F
m (θ, φ) ∝
Y
m (θ, φ), we knew that both functions are proportional to eimφ , and that both are
eigenfunctions of ∇2θ with eigenvalue −( + 1). We claimed that, up to a multiplicative
constant, there is only one function that has these properties. Assuming that the power
series representation of Eq. (8.1) always exists, the uniqueness that we need is easy to
see. We showed in lecture that the number of linearly independent traceless symmetric
(
,m)
tensors of rank  is 2+1, and now we have constructed 2+1 such tensors: the Ci1 ...i
, for
m = −, . . . , . These are clearly linearly independent, since they are each eigenfunctions
of rotations about the z-axis with different eigenvalues. Thus, any traceless symmetric
tensor of rank  must be a linear sum of the tensors in our basis. When we also specify
that the tensor being sought is an eigenvector of rotations about the z-axis, with a specific
eigenvalue, then only one of the tensors in our basis can contribute.
The above argument is solid, but one might still wonder what happens if we try,
for example, to construct a different tensor by using both û+ ’s and û− ’s in the same
expression. For example, we might consider { û+ −
i ûj } , which is invariant under rotations
about the z-axis. With a little work, however, one can show that

1
{ û+ −
i ûj } = − { ẑi ẑj } . (8.45)
2
MIT OpenCourseWare
http://ocw.mit.edu

8.07 Electromagnetism II
Fall 2012

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.