848 words
4 minutes
Discrete Symmetry Groups: Rotations, Mirrors, and Vanishing Matrix Elements

When a Hamiltonian has a symmetry, one can often show that whole blocks of its matrix vanish without doing a single integral. In my opinion this, more than the aesthetics, is the reason symmetry is worth the effort in quantum mechanics. In this post we work through a small discrete example (rotations by multiples of π/2\pi/2, plus a mirror) and use it to show which matrix elements of a perturbation are zero.


Setup: the group G\mathcal{G}#

Take the set of rotations about the zz-axis by multiples of π/2\pi/2,

G={Dz(0), Dz(π/2), Dz(π), Dz(3π/2)}\mathcal{G} = \{D_z(0),\, D_z(\pi/2),\, D_z(\pi),\, D_z(3\pi/2)\}

with Dz(a)Dz(b)=Dz(a+b)D_z(a)D_z(b) = D_z(a+b), i.e. angles add under composition (modulo 2π2\pi).

Checking that G\mathcal{G} is a group#

Writing Dz(n)≡Dz(nπ/2)D_z(n) \equiv D_z(n\pi/2) for short, the multiplication table is

∗*0123
0Dz(0)D_z(0)Dz(1)D_z(1)Dz(2)D_z(2)Dz(3)D_z(3)
1Dz(1)D_z(1)Dz(2)D_z(2)Dz(3)D_z(3)Dz(0)D_z(0)
2Dz(2)D_z(2)Dz(3)D_z(3)Dz(0)D_z(0)Dz(1)D_z(1)
3Dz(3)D_z(3)Dz(0)D_z(0)Dz(1)D_z(1)Dz(2)D_z(2)

and we can go through the four group axioms one by one.

  • Identity. Dz(0)D_z(0), as one sees from the first row and column.
  • Closure. Every entry of the table is in G\mathcal{G} ✓
  • Inverses. Dz(N)−1=Dz(∣N−4∣)D_z(N)^{-1} = D_z(|N-4|), i.e. Dz(1)−1=Dz(3)D_z(1)^{-1} = D_z(3), Dz(2)−1=Dz(2)D_z(2)^{-1} = D_z(2).
  • Associativity. This follows from angle addition, D(F)D(G)D(H)=D(F+G+H)D(F)D(G)D(H) = D(F+G+H), which is associative because addition is.

So G≅Z4\mathcal{G} \cong \mathbb{Z}_4, the cyclic group of order 4.

It is abelian#

The table is symmetric about the diagonal, so Dz(a)Dz(b)=Dz(b)Dz(a)D_z(a)D_z(b) = D_z(b)D_z(a) for all elements. Again this is just angle addition, a+b=b+aa+b = b+a.

Subgroups#

G\mathcal{G} contains the subgroup Z2={Dz(0),Dz(2)}={Dz(0),Dz(π)}\mathbb{Z}_2 = \{D_z(0), D_z(2)\} = \{D_z(0), D_z(\pi)\}, with table

∗*02
0Dz(0)D_z(0)Dz(2)D_z(2)
2Dz(2)D_z(2)Dz(0)D_z(0)

It is closed, and Dz(2)D_z(2) is its own inverse, since applying Dz(π)D_z(\pi) twice gives Dz(2π)=Dz(0)D_z(2\pi) = D_z(0).


Eigenvalues of symmetry operators#

Mirror operator MxM_x#

The mirror MxM_x sends (x,y)→(−x,y)(x,y) \to (-x, y). Doing it twice gives back the original state,

Mx2∣x,y⟩=Mx∣−x,y⟩=∣x,y⟩  ⟹  Mx2=1M_x^2 |x,y\rangle = M_x|-x,y\rangle = |x,y\rangle \implies M_x^2 = \mathbf{1}

So if Mx∣ψ⟩=λ∣ψ⟩M_x|\psi\rangle = \lambda|\psi\rangle, then λ2=1\lambda^2 = 1 and

λ(Mx)=±1\lambda(M_x) = \pm 1

States with λ=+1\lambda = +1 are even (symmetric) under the reflection, and states with λ=−1\lambda = -1 are odd (antisymmetric).

Rotation operators Dz(nπ/2)D_z(n\pi/2)#

For the rotations, the possible eigenvalues are set by the order of the element, i.e. how many times it has to be applied to get back to the identity. If Dk=1D^k = \mathbf 1, then λk=1\lambda^k = 1.

  • Dz(0)D_z(0): order 1, so λ1=1  ⟹  λ=+1\lambda^1 = 1 \implies \lambda = +1
  • Dz(π/2)D_z(\pi/2): order 4, so λ4=1  ⟹  λ=±1,±i\lambda^4 = 1 \implies \lambda = \pm 1, \pm i
  • Dz(π)D_z(\pi): order 2, so λ2=1  ⟹  λ=±1\lambda^2 = 1 \implies \lambda = \pm 1
  • Dz(3π/2)D_z(3\pi/2): order 4, so λ4=1  ⟹  λ=±1,±i\lambda^4 = 1 \implies \lambda = \pm 1, \pm i

Do rotations and mirrors commute?#

In general, no. To see it, put four atoms in the four quadrants and label the states by the quadrant, ∣1⟩,∣2⟩,∣3⟩,∣4⟩|1\rangle, |2\rangle, |3\rangle, |4\rangle. Applying the two operators in both orders to ∣1⟩|1\rangle,

MxDz(1)∣1⟩=Mx∣2⟩=∣1⟩M_x D_z(1)|1\rangle = M_x|2\rangle = |1\rangleDz(1)Mx∣1⟩=Dz(1)∣2⟩=∣3⟩D_z(1) M_x|1\rangle = D_z(1)|2\rangle = |3\rangle

Since ∣1⟩≠∣3⟩|1\rangle \neq |3\rangle, the operators do not commute, [Mx,Dz(π/2)]≠0[M_x, D_z(\pi/2)] \neq 0.

But MxM_x and Dz(π)D_z(\pi) do commute#

We can check this by following how each operator permutes the four quadrant states. Let an ordered tuple (a,b,c,d)(a,b,c,d) record which atom sits in quadrants 1, 2, 3, 4. Then the two operators act as

Mx{a,b,c,d}={b,a,d,c}Dz(2){a,b,c,d}={c,d,a,b}M_x\{a,b,c,d\} = \{b,a,d,c\} \qquad D_z(2)\{a,b,c,d\} = \{c,d,a,b\}

and applying them in both orders,

Dz(2) Mx{1,2,3,4}=Dz(2){2,1,4,3}={4,3,2,1}D_z(2)\,M_x\{1,2,3,4\} = D_z(2)\{2,1,4,3\} = \{4,3,2,1\}Mx Dz(2){1,2,3,4}=Mx{3,4,1,2}={4,3,2,1}M_x\,D_z(2)\{1,2,3,4\} = M_x\{3,4,1,2\} = \{4,3,2,1\}

The results agree, so [Mx,Dz(π)]=0[M_x, D_z(\pi)] = 0 and the two operators have simultaneous eigenstates. This is what allows us to label states by two parities in the next section.


Using symmetry to kill matrix elements#

Now for the actual application. Suppose a potential V=aXYV = aXY acts on the system. Under the mirror and under the rotation by π\pi it transforms as

MxV=Mx(aXY)=a(−X)Y=−aXY=−V⇒{Mx,V}=0 (anticommutes)M_x V = M_x(aXY) = a(-X)Y = -aXY = -V \quad \Rightarrow \quad \{M_x, V\} = 0 \text{ (anticommutes)}Dz(2)V=Dz(2)(aXY)=a(−X)(−Y)=aXY=V⇒[Dz(2),V]=0 (commutes)D_z(2) V = D_z(2)(aXY) = a(-X)(-Y) = aXY = V \quad \Rightarrow \quad [D_z(2), V] = 0 \text{ (commutes)}

We label the eigenstates by their two parities, ∣ϵD,ϵM⟩|\epsilon_{D}, \epsilon_{M}\rangle, where ϵD=±1\epsilon_D = \pm 1 is the Dz(2)D_z(2) eigenvalue and ϵM=±1\epsilon_M = \pm 1 is the MxM_x eigenvalue. Take the matrix element ⟨+,+∣V∣+,+⟩\langle +,+|V|+,+\rangle and insert the symmetry operators (each one acting on its own eigenstate just gives +1+1),

⟨+,+∣V∣+,+⟩=⟨+,+∣Dz(2)Mx V MxDz(2)∣+,+⟩\langle +,+|V|+,+\rangle = \langle +,+|D_z(2)M_x\,V\,M_x D_z(2)|+,+\rangle

Since MxM_x anticommutes with VV and commutes with Dz(2)D_z(2), we can move one MxM_x through VV at the price of a minus sign,

=(−1)⟨+,+∣Dz(2) V MxMx⏟1 Dz(2)∣+,+⟩=−⟨+,+∣V∣+,+⟩= (-1)\langle +,+|D_z(2)\,V\,\underbrace{M_x M_x}_{1}\,D_z(2)|+,+\rangle = -\langle +,+|V|+,+\rangle

and a number equal to minus itself is zero,

⟨+,+∣V∣+,+⟩=0\boxed{\langle +,+|V|+,+\rangle = 0}

Notice that the argument only used the MxM_x label. Whatever ϵM\epsilon_M is, it appears twice and squares to 1, while the anticommutation supplies the minus sign. So every diagonal element vanishes, including the mixed-parity ones such as ⟨+,−∣V∣+,−⟩\langle +,-|V|+,-\rangle, i.e. the expectation value of VV in any MxM_x eigenstate is zero.

The same trick applied to an off-diagonal element ⟨ϵD,ϵM∣V∣ϵD′,ϵM′⟩\langle \epsilon_D, \epsilon_M|V|\epsilon_D', \epsilon_M'\rangle gives the general rule.

  • MxVMx=−VM_x V M_x = -V gives ϵMϵM′⟨⋯ ⟩=−⟨⋯ ⟩\epsilon_M\epsilon_M' \langle\cdots\rangle = -\langle\cdots\rangle, so the element vanishes unless ϵM′=−ϵM\epsilon_M' = -\epsilon_M.
  • Dz(2)VDz(2)=VD_z(2) V D_z(2) = V gives ϵDϵD′⟨⋯ ⟩=⟨⋯ ⟩\epsilon_D\epsilon_D' \langle\cdots\rangle = \langle\cdots\rangle, so the element vanishes unless ϵD′=ϵD\epsilon_D' = \epsilon_D.

So V=aXYV = aXY only connects states with the same Dz(π)D_z(\pi) parity and opposite mirror parity, e.g. ⟨+,+∣V∣+,−⟩\langle +,+|V|+,-\rangle. All other elements are zero.

In short, knowing how a perturbation transforms under the symmetry group of the Hamiltonian tells you which matrix elements vanish, with no integrals at all. Selection rules in atomic physics work in exactly this way, and the Wigner–Eckart theorem is the general version of the same idea.

Discrete Symmetry Groups: Rotations, Mirrors, and Vanishing Matrix Elements
https://rohankulkarni.me/posts/notes/discrete-symmetry-groups/
Author
Rohan Kulkarni
Published at
2022-10-17
License
CC BY-NC-SA 4.0
Share Post
Comments