Notes: Tensor Operators

Ben Baragiola

I.

VECTORS

We are already familiar with the concept of a vector, but lets review vector properties to refresh ourselves. A 3-component Cartesian vector is v = vx ex + vy ey + vz ez . (1)

Operationally, what makes this a vector is the way it transforms under rotations. General rotations can be specied by Euler angles, but for the sake of simplicity we will consider a rotation by around the z-axis, which is represented in the Cartesian basis by the matrix cos sin 0 Rz () = sin cos 0 . (2) 0 0 1 We use this matrix to rotate v into a the new, rotated vector denoted by v = vx ex + vy ey + vz ez . We will use the explicit for of the rotation matrix to write the elements of v in terms of those of v and the angle . vx cos vx + sin vy vx cos sin 0 (3) v = Rz ()v = sin cos 0 vy = sin vx + cos vy = vy . vz vz 0 0 1 vz By denition, the components of a vector transform like vi =
j

Rij vj .

(4)

Lets verify that according to this denition our v is a vector: vx =

Rxj vj = Rxx vx + Rxy vy + Rxz vz

(5) (6)

= cos vx + sin vy

and one can similarly check the other components to verify that indeed v is a vector. In quantum mechanics, the rotation operator is D(n, ) = eiJn

(7)

where n is the unit vector about which the rotation occurs, and J is the angular momentum vector operator, which has operator components that transform according to Eq. (4). A vector operator V is rotated in the following way D(n, ) VD(n, ) =
j

Rij Vj

(8)

To evaluate this, we make use of a property of vector operators [Vi , Ji ] = i and the Hadamard lemma eX Y eX = Y + [X, Y ] + 1 [X, [X, Y ]] + ... 2! (10)
ijk

Vk

(9)

Writing out a couple terms in the series for the instructive case of n J = Jz , we recognize quickly that
eiJz Vx eiJz = cos Vx + sin Vy

(11)

with similar, expected transformations for the other components of V.

2
II. TENSORS

Classically, tensors are dened by the way they transform under rotations. We are primarily concerned with tensors of rank 0 (scalars), rank 1 (vectors), and of rank 2 and we will restrict the discussion to this subset. Scalars do not transform under rotations, vectors transform according to Eq. (4), and rank 2 tensors transform according to Tij =
ij

Rii Rjj Ti j .

(12)

Rank 2 tensors have two sets of indices each which runs from 1 to 3, so there are nine components. For this reason, they are often represented as matrices. Given a rank 2 tensor T , lets compute an element of a rotated tensor T where the rotation matrix is around the z-axis as above. Txx =
ij

Rxi Rxj Ti j

(13)

=Rxx Rxx Txx + Rxx Rxy Txy + Rxx Rxz Txz + Rxy Rxx Tyx + Rxy Rxy Tyy + Rxy Rxz Tyz + Rxz Rxx Tzx + Rxz Rxy Tzy + Rxz Rxz Tzz = cos2 Txx + cos sin (Txy + Tyx ) + sin2 Tyy Clearly, calculating all 9 elements would be exhaustive but exhausting.
A. Cartesian Tensors

(14)

Two vectors, v and w, expressed in the Cartesian basis (ex , ey , ez ) can be used to create a rank 2 Cartesian tensor T . It is formed as the dyad of the two vectors Tij vi wj and can be expressed as a 3 3 matrix

(15)

vx wx vx wy vx wz T = vy wx vy wy vy wz . vz wx vz wy vz wz

(16)

Dyadic Cartesian tensors, such as T , can be decomposed into irreducible representations in the following way Tij = Tij + Tij + Tij =
(0) (1) (2)

(17) vi wj + vj wi vw ij . 2 3 (18)

(vi wj vj wi ) (v w) ij + + 3 2

Each of these irreducible representations has particular properties. T (0) is a rank 0 tensor and transforms under rotations like a scalar. In our matrix representation, it can also be written as the trace of the full, reducible tensor T 0 1 1 vw 0 0 vw 0 T (0) = Tr(T ) 1 = (19) 3 3 0 0 vw where v w = vx wx + vy wy + vz wz . T (0) has only one independent component. T (1) is a rank 1 tensor and transforms under rotations like a vector. It can be represented as a vector (cross) (1) product Tij = 1 ijk (v w)k and has a matrix representation 2 0 (vx wy vy wx ) (vx wz vz wx ) 1 (vx wy vy wx ) 0 (vy wz vz wy ) . T (1) = (20) 2 (v w v w ) (v w v w ) 0
x z z x y z z y

3
(1)

T has three independent components. T (2) is a rank 2 tensor and transforms according to Eq. (12). It has the following form in our matrix representation (vx wy +vy wx ) (vx wz +vz wx ) vx wx vw 3 2 2 (21) T (2) = (vx wy +vy wx ) vy wy vw (vy wz +vz wy ) . 2 3 2 (vy wz +vz wy ) (vx wz +vz wx ) vw vz wz 3 2 2

Due to the fact that they are antisymmetric and traceless, Tr(T (2) ) = 0, irreducible rank 2 tensors have 5 independent entries. Notice that the number of independent components of T is equal to the number of independent components of T (0) + T (1) + T (2) : 3 3 = 1 + 3 + 5. In addition, each of the irreducible representations transforms like angular momentum according to its number of independent components.

B.

Spherical Tensors

The fact that Cartesian tensors are reducible prompts us to seek out an irreducible set of tensors. A useful set of these are the spherical tensors.

1.

Spherical Basis

Spherical tensors are dened on a set of basis vectors dened as follows e = (ex + iey ) , 2 e0 = ez . (22)

and we use the letter q to designate an arbitrary spherical basis element. The fact that these are complex will lead to some denitions that may seem strange at rst, but arise only to maintain the familiar properties of Cartesian space. The components of a vector A in the spherical basis are Aq = eq A so that A may be decomposed in the spherical basis as A=
q

(23)

Aq e = q
q

(1)q Aq eq = A+ e + A0 e0 + A e . +

(24)

The dot product of two vectors has a form that seems unfamiliar, but preserves the norm of a vector |A|2 : A B = A+ B + A0 B0 A B+ =
q

(1)q Aq Bq

(25)

2.

Spherical Harmonics

An example of irreducible spherical tensors that will prove quite useful are the spherical harmonics. Recalling the denition of spherical harmonics as the angular representation of the |l, m angular momentum eigenstates: Ylm (, ) = , |l, m = (2l + 1)(l m)! m Pl (cos )eim 4(l + m)! (26)

Spherical harmonics are irreducible and transform like tensors of rank k=l, where m indexes the number of unique elements.

4
C. Formal Denition of Spherical Tensor Operators

Motivated by the above discussion, we dene a spherical tensor operator of rank k as a set of 2k+1 operators (k) Tq q = k, k 1... k + 1, k (27)

which transform among themselves like 2j+1 angular momentum eigenstates |j = k, m = q (and thus like spherical harmonics) according to
k

(k) D(n, ) Tq D(n, ) =

Dq,q ()Tq

(k)

(k)

(28)

where Dq,q () = k, q |einJk |k, q .

(29)

Now, we are prepared to represent a dyad formed from two vectors v and w (written in the spherical basis) in terms of irreducible spherical tensors: T0
(0)

(1) Tq (2)

1 v w = (v1 w1 + v0 w0 + v1 w1 ) 2 3 (v w)q = i 2 =

(30) (31) (32) (33) (34)

T2 = v1 w1 1 (2) T1 = (v1 w0 + v0 w1 ) 2 1 (2) T0 = (v1 w1 + 2v0 w0 + v1 w1 ) 6

Wigner Eckhart Theorem

Given a spherical tensor operator Tq , the Wigner-Eckhart theorem states:

(35)

Essentially, the theorem says that it is possible to factor the matrix element into a reduced matrix element which is independent of m (and thus of any specic geometry) and a Clebsch-Gordan coecient. The CG coecients determine the selection rules m =m+q and |j k| j j + k. (36)

III.

THE DIPOLE OPERATOR

The motivation for the previous sections has been to arrive at a place where we are prepared to discuss the dipole operator d which shows up in the dipole Hamiltonian for the interaction between an atom and an electric eld. HAF = d E The dipole operator is related to the position operator through the charge of the electron: r d = e. r (38) (37)

5 we write down those corresponding to l = {0, 1, 2} as they are irreducible representations of corresponding rank: Y00 = Y10 = Y11 = Y20 = Y21 = Y22 = Now, we see that the position vector r = rx ex + ry ey + rz ez = r sin cos ex + r sin sin ey + r cos ez can be written in the spherical basis (and thus as spherical harmonics): r r r = sin ei e + r cos e0 + sin ei e + 2 2 =r with individual components rq = r 4 q Y . 3 1 (49) 4 1 (Y e + Y10 e0 + Y11 e ) 3 1 + (47) (48) (45) (46) 1 4 3 cos 4 3 sin ei 8 5 (3 cos2 1) 16 5 sin cos ei 16 5 sin2 e2i . 32 (39) (40) (41) (42) (43) (44)