Prepared for submission to JHEP

Group cohomology and applications in

physics

Y. X. Zhao
Department of Physics, The University of Hong Kong, Hong Kong, China
E-mail: [email protected]

Abstract: In this notes, we introduce some basics of projective representations of

groups.
Contents

1 Group cohomology 1
1.1 The vertex viewpoint 1
1.2 The bond viewpoint 2
1.3 The long exact sequence 3

2 Space groups 4
2.1 H 1 (P, Rd /L) 4
2.2 H 2 (P, L) 5
2.3 An example with mirror symmetry 5

3 Projective representations in quantum mechanics 6

3.1 Twofold anti-unitary symmetries and the tenfold symmetry classes 6
3.2 Semidirect product groups 9
3.2.1 Direct product groups 10
3.2.2 The Heisenberg algebra 11
3.2.3 Translation groups 12
3.2.4 Euclidean groups 12
3.2.5 Quantum mechanical interpretations 13
3.2.6 Galilean group 14
3.3 Clifford algebras 16

4 The projective representations of crystal symmetries 16

4.1 Realization on flux lattices 16
4.2 The classification with time-reversal invariance 19
4.3 The canonical form of multipliers 20

1 Group cohomology

1.1 The vertex viewpoint

Consider a group G and an Abelian group A. G acts on A with the action ρ satisfying

ρg (a + b) = ρg (a) + ρg (b), ρgh (a) = ρg (ρh (a)), (1.1)

for all g, h ∈ G and a, b ∈ A.

–1–
 Let C n (G, A) be the set of equivariant functions from Gn+1 to A. A function
f : Gn+1 → A is called equivariant if

ρh f (g0 , g1 , · · · , gn ) = f (hg0 , hg1 , · · · , hgn ), (1.2)

for all h, g0 , · · · , gn ∈ G. C n (G, A) is an Abelian group under the addition of func-

tions, i.e., for all f1 , f2 ∈ C n (G, A),

(f1 + f2 )(g0 , · · · , gn ) := f1 (g0 , · · · , gn ) + f2 (g0 , · · · , gn ). (1.3)

The homomorphism δ n : C n (G, A) → C n+1 (G, A) is defined by

n+1
X
n
δ f (g0 , g1 , · · · , gn+1 ) := (−1)i f (g0 , · · · , gi−1 , gi+1 , · · · , gn+1 ) (1.4)
i=0

for all f ∈ C n (G, A) and all g0 , · · · , gn+1 ∈ G. Obviously, we have

δ n+1 ◦ δ n = 0. (1.5)

for all non-negative integers n.

Hence, we have actually constructed the cochain complex,
δ0 δ1 δ2 δ3
0 → C 0 (G, A) −
→ C 1 (G, A) −
→ C 2 (G, A) −
→ C 3 (G, A) −
→ ··· . (1.6)

Accordingly, the cohomology groups are defined by

H n (G, A) := Z n (G, A)/B n (G, A), (1.7)

where
Z n (G, A) := ker δ n , B n (G, A) := imδ n−1 . (1.8)
An immediate result is that

H 0 (G, A) = AG . (1.9)

Here, AG is the subset of A, where each element is invariant under the action of all
g ∈ G.

1.2 The bond viewpoint

Let C n (G, A) be the Abelian group consisting of all functions from Gn to A. Then,
it is easy to observe that there is an isomorphism from C n (G, A) to C n (G, A) given
by
f¯(h1 , · · · , hn ) := f (1, h1 , h1 h2 , · · · , h1 h2 · · · hn ), (1.10)
where for each f ∈ C n (G, A) the image is f¯. Then, we can define the coboundary
map dn : C n (G, A) → C n+1 (G, A) as

dn f¯(h1 , · · · , hn+1 ) = δ n f (1, h1 , h1 h2 , · · · , h1 · · · hn+1 ). (1.11)

–2–
Then,
δf (1, h1 , h1 h2 , · · · , h1 · · · hn+1 )
=f (h1 , h1 h2 , · · · , h1 h2 · · · hn+1 ) − f (1, h1 h2 , · · · , h1 h2 · · · hn+1 ) + · · ·
(1.12)
=ρh1 f (1, h2 , · · · , h2 h3 · · · hn+1 ) − f (1, h1 h2 , · · · , h1 h2 · · · hn+1 ) + · · ·
=ρh f¯(h2 , h3 , · · · , hn+1 ) − f¯(h1 h2 , h3 , · · · , hn+1 ) + · · · .
1

Hence, we obtain
dn f¯(h1 , · · · , hn+1 ) =ρh1 f¯(h2 , h3 , · · · , hn+1 ) + (−1)n+1 f¯(h1 , h2 , · · · , hn )
X n
(1.13)
+ f¯(h1 , · · · , hi−1 , hi hi+1 , hi+2 , · · · , hn+1 )
i=1

The relation d ◦ d = 0 is inherited from δ n+1 ◦ δ n = 0.

n+1 n

Then, we can construct the cochain complex

d0 d1 d2
→ C 1 (G, A) −
0→A− → C 2 (G, A) −
→ ··· . (1.14)
Here, d0 a(g) = ρg a − a for all a ∈ A and g ∈ G. Accordingly, the cohomology groups
are defined by
Hn (G, A) := Z n (G, A)/B n (G, A), (1.15)
where
Z n (G, A) := ker dn , B n (G, A) := imdn−1 . (1.16)
The two definitions of cohomology groups are equivalent, i.e., we have
H n (G, A) = Hn (G, A). (1.17)
For instance, when n = 0, a ∈ ker d0 iff ρg a − a for all g ∈ G, i.e., a ∈ AG . Hence,
H0 (G, A) = AG .

1.3 The long exact sequence

Consider the short exact sequence of coefficient Abelian groups,
1 → N → A → A/N → 1. (1.18)
G acts on each of them, and every homomorphism in the sequence commutes with
the G-actions. Then, we have the long exact sequence of cohomology groups,
· · · → H n (G, N ) → H n (G, A) → H n (G, A/N ) → H n+1 (G, N ) → · · · (1.19)
A useful example is
1 → Z → R → U (1) → 1. (1.20)
Because H n (G, R) = 0 for all n > 0, we have
0 → H n (G, U (1)) → H n+1 (G, Z) → 0, (1.21)
which is equivalent to
H n (G, Z) ∼
= H n+1 (G, U (1)). (1.22)

–3–
2 Space groups
Given a point group P and a lattice L, all possible space groups can be classified by
H 2 (P, L) ∼ = H 1 (P, Rd /L), which is explained in the following.
Here, L is spanned by a set of primitive lattice vectors ei ∈ Rd with i =
1, 2, · · · , d, i.e.,
L = {ni ei , ni ∈ Z}. (2.1)
Clearly, L is an Abelian group under the addition of vectors. The action of P on L
is inherited from the natural action of O(d) on Rd , since P ⊂ O(d) and L ⊂ Rd . For
each R ∈ P , the action may be explicitly written by
d
X
Rei = ej R̂ji (2.2)
j=1

with R̂ji ∈ Z.
For a space group Γ associated with the lattice L and the point group P , L is a
normal subgroup of Γ, and the quotient group Γ/L is isomorphic to P , i.e., we have
the short exact sequence,
0 → L → Γ → P → 1. (2.3)

2.1 H 1 (P, Rd /L)

One may lift P into Γ. That is, each R ∈ P is associated with a translation τR ∈ Rd ,
and (τR , R) is the assigned element of Γ for R, noting that Γ is a subgroup of
the Euclidean group E d = Rd ⋊ O(d). We then consider the multiplication of two
arbitrary lifted elements,

(τR1 , R1 )(τR2 , R2 ) = (τR1 + R1 τR2 , R1 R2 ). (2.4)

Compared with (τR1 R2 , R1 R2 ), the difference of τR1 R2 from τR1 + R1 τR2 should be
an element of L, as written by

τR1 + R1 τR2 = τR1 R2 mod L. (2.5)

This is just the 1-cocycle equation for H 1 (P, Rd /L), or alternatively τ ∈ Z 1 (P, Rd /L).
There are two centers, the coordinate origin and the center respected by O(d). If
we shift the displacement of the two centers by r, then τR for each R ∈ P is changed
by
drR = Rr − r. (2.6)
It is noticed that dr satisfies the 1-cocycle equation above, and dr ∈ B 1 (P, Rd /L). It
is clear that changing the displacement of two centers does not change the isomorphic
class of the space group or the space group for short, and hence each space group
corresponds to an element of H 1 (P, Rd /L).

–4–
2.2 H 2 (P, L)
H 1 (P, Rd /L) is isomorphic to H 2 (P, L), which can be explained as follows in the
context of space groups. Each τ corresponds to a function ω : P × P → L by

ω(R1 , R2 ) = τR1 + R1 τR2 − τR1 R2 ∈ L. (2.7)

Then, it is straightforward to check that ω satisfies the 2-cocycle equation for H 2 (P, L),
as given by

ω(R1 , R2 ) + ω(R1 R2 , R3 ) = R1 ω(R2 , R3 ) + ω(R1 , R2 R3 ), (2.8)

and therefore we have ω ∈ Z 2 (P, L). For the lifting of τR , τR + ℓ(R) is equally well
for an arbitrary ℓ(R) ∈ L, since both (τR , R) and (τR + ℓ(R), R) are mapped to R
in the short exact sequence for Γ. Accordingly, ω(R1 , R2 ) is changed by

dℓ(R1 , R2 ) = R1 ℓ(R2 ) + ℓ(R1 ) − ℓ(R1 R2 ). (2.9)

Here, we observe that dℓ ∈ B 2 (P, L). Hence, each space group corresponds to an
element of H 2 (P, L).

2.3 An example with mirror symmetry

For an example, consider P = ZM
2 = {E, M } with
   
10 −1 0
E= , M= , (2.10)
01 0 1

and L is spanned by
e1 = [1, 0]T , e2 = [0, 1]T . (2.11)
We only need to find all possible τM = [a, b]T . There is only one nontrivial equation,
τM + M τM = 0 mod L. The left-hand side is simplified by
      
−1 0 a a 0
+ = . (2.12)
0 1 b b 2b

Hence, b = 0 or b = 1/2. Elements of B 1 (P, R2 /L) only change the a-component of

τM , since       
−1 0 x x 2x
− = . (2.13)
0 1 y y 0
Thus, H 1 (ZM ∼
2 , R /L) = Z2 , which is represented by
2

   
0 0
τM = ′
, τM = , (2.14)
0 1/2

τM and τM
′
correspond to wallpaper groups P m and P g, respectively.

–5–
3 Projective representations in quantum mechanics
Let H be a Hilbert space representing a symmetry group G. Each group element
g ∈ G is represented by a unitary or anti-unitary operator Lg . Thus, in general we
assume a G-action c on U (1), i.e.,

cg (u) = u or u∗ , (3.1)

depending on whether Lg is unitary or anti-unitary. Here, u ∈ U (1), and u∗ is the

complex conjugate of u.
Consider the successive action of two group elements g1 and g2 on an arbitrary
quantum state |ψ⟩ ∈ H. There are two natural ways to implement the action,
namely Lg1 Lg2 |ψ⟩ and Lg1 g2 |ψ⟩. The resulting quantum states should be the same.
One might require Lg1 Lg2 |ψ⟩ = Lg1 g2 |ψ⟩, but which is too strict. Multiplying an
arbitrary U (1) phase u to a normalized vector |ψ ′ ⟩ does not change the quantum
state that it represents, since ρ = |ψ ′ ⟩⟨ψ ′ | = u|ψ ′ ⟩⟨ψ ′ |u∗ . In other words, quantum
states form the projective space P (H), where two vectors v, w ∈ H are identified if
v = cw for some nonzero c ∈ C. Hence, Lg1 Lg2 |ψ⟩ may differ from Lg1 g2 |ψ⟩ by a
phase factor ν(g1 , g2 ). Since |ψ⟩ is arbitrary, we require the more general relation,

Lg1 Lg2 = ν(g1 , g2 )Lg1 g2 , (3.2)

for all g1 , g2 ∈ G.
Since for all g ∈ G Lg are unitary or anti-unitary operators, the multiplication
of Lg satisfies the associativity, i.e.,

(Lg1 Lg2 )Lg3 = Lg1 (Lg2 Lg3 ), (3.3)

for all g1 , g2 , g3 ∈ G. The associativity leads to the 2-cocycle equation,

cg1 (ν(g2 , g3 ))ν(g1 , g2 g3 ) = ν(g1 , g2 )ν(g1 g2 , g3 ), (3.4)

which implies ν ∈ Z 2 (G, U (1)). If we multiply an arbitrary phase χ(g) ∈ U (1) to

Lg , ν(g1 , g2 ) is modified to be ν(g1 , g2 )/dχ(g1 , g2 ) with
cg1 (χ(g2 ))χ(g1 )
dχ(g1 , g2 ) = . (3.5)
χ(g1 g2 )
It is noticed that dχ ∈ B 2 (G, U (1)). Thus, the phase system ν is classified by
H 2 (G, U (1)).

3.1 Twofold anti-unitary symmetries and the tenfold symmetry classes

Consider the two-element group Z2 ∼
= {E, R} with R2 = E. Let us consider the
projective representations of the group with R being an anti-unitary operator AR .
Then, there is only one nontrivial phase factor ν(T, T ), which appears in

AR AR = ν(R, R)1. (3.6)

–6–
Considering the associativity of three AR ,

(AR AR )AR = AR (AR AR ), (3.7)

we find that
ν(R, R) = [ν(R, R)]∗ . (3.8)
Hence, there are two solutions,

ν(R, R) = ±1, (3.9)

and accordingly,
Z 2,c (Z2 , U (1)) ∼
= Z2 . (3.10)
In fact, B 2,c (Z2 , U (1)) = 0. This is because A2R = ν(R, R) is invariant under any
modification of the phase of AR , i.e.,

(uAR )(uAR ) = uu∗ (AR AR ) = AR AR = ν(R, R)1. (3.11)

for any u ∈ U (1). Thus, we have established that

H 2,c (Z2 , U (1)) ∼

= Z2 . (3.12)

With this mathematical result, we proceed to consider twofold anti-unitary sym-

metries. Such symmetries can be categorized into two classes according to whether
they inverse the time coordinate t. We denote operators in the two classes by
T = UT KIˆt , and C = UC K, respectively. Here, UT and UC are unitary operators,
K is the complex conjugation, and Iˆt simply inverses t. Then, the two multipliers
correspond to
UT UT∗ = ηT , UC UC∗ = ηC (3.13)
for T and C, respectively, with ηT /C ∈ {±1}. For a single particle, the time-reversal
operator AT is usually written as T , with

T 2 = (−1)2s 1. (3.14)

Here, s is the spin of the particle. For instance,

T = KIˆt or iσ2 KIˆt (3.15)

for spinless and spin- 21 particles.

A symmetry should preserve the time-evolution operator:
 
i
U (t) = exp − Ht . (3.16)
ℏ
Hence, for a time-reversal invariant system, we have

UT H ∗ UT† = H. (3.17)

–7–
But if C is preserved by the system, then the Hamiltonian satisfies the following
identity:
UC H ∗ UC† = −H. (3.18)

C is called the particle-hole symmetry because the above equation requires that the
spectrum of H be symmetric with respect to the zero energy.
In the presence of both C and T , we may consider the projective representations
of ZT2 × ZC2 with both C and T represented by anti-unitary operators AT /C . Then,
for a given multiplier ν, the commutation relation of AT and AC is modified as

AT AC = eiα AC AT (3.19)

with eiα = ν(C, T )/ν(T, C). We can always modify the phases of the operators as

ÃC := eiα/2 AC , ÃT := AT . (3.20)

Then, for the equivalent representation ÃC , ÃT and ÃC commute with each other:

ÃC ÃT = ÃT ÃC . (3.21)

Moreover, we always modify the phase of AT C as

√ √
ÃT C := ηT ηC eiα/2 AC AT = ηT ηC ÃT ÃC . (3.22)

Then,
(ÃT C )2 = 1 (3.23)

An upshot of the above arguments is that we have obtained the cohomology

group:
H 2,c (Z2 × Z2 ) ∼
= Z2 × Z2 . (3.24)

The four elements of H 2,c (Z2 × Z2 ) correspond to ηT = ±1 and ηC = ±1.

In fact, we have established the so-called tenfold Altland-Zirnbauer symmetry
classes. Consider the symmetry operators:
√
T, C, i, S := ηT ηC T C, (3.25)

with the algebraic relations:

T 2 = ηT , C 2 =ηC , S 2 = 1, i2 = −1,
(3.26)
[T , C] = 0, {T , i} = 0, {C, i} = 0.

Then, the ten symmetry classes are tabulated as Tab.1.

–8–
 Non-chiral classes Chiral classes
A AI D AII C AIII BDI DIII CII CI
T 0 +1 0 −1 0 0 +1 −1 −1 +1
C 0 0 +1 0 −1 0 +1 +1 −1 −1
S 0 0 0 0 0 1 1 1 1 1
Table 1. Tenfold symmetry classes. The signs ±1 denote the square of the symmetry
operators, and 0 denotes the absence of the symmetry.

3.2 Semidirect product groups

If group G is a semidirect product of N and P , i.e.,

G = N ⋊ P, (3.27)

both N and P can be identified as subgroups of G, and N ∩ P = {e}. Then, each

group element g can be uniquely written as

g = np (3.28)

with n ∈ N and p ∈ P . On the other hand, varying n over N and p over P , the
product np exhausts G. Since N is a normal subgroup of G, the conjugation by
elements of P is a P -action on N , which is denoted as

cp (n) := pnp−1 , (3.29)

for all n ∈ N and p ∈ P .

We proceed to discuss the multipliers of N ⋊ P . We assume that the multipliers
are valued in an Abelian group A, e.g., A = U (1) or Z2 = {±1} of especial interest.
From Mackey’s paper, any multiplier ν of G is similar to a multiplier of the following
factorized form,

ν(n1 p1 , n2 p2 ) = σ(n1 , cp1 (n2 ))g(n2 , p1 )α(p1 , p2 ). (3.30)

Here, σ and α are just the restrictions of ν on N and P , respectively, namely the
restricted multipliers. Moreover, for µ is a multiplier, σ and g satisfy the following
consistency relations:

σ(cp (n1 ), cp (n2 )) g(n1 n2 , p)

= , (3.31)
σ(n1 , n2 ) g(n1 , p)g(n2 , p)

g(n, p1 p2 ) = g(cp2 (n), p1 )g(n, p2 ). (3.32)

The form of multipliers is not difficult to prove. It is interesting to explain the
meanings of the two consistency relations.

–9–
 • There is the canonical action of P on the set of multipliers Z 2 (N, A) of N ,
which is induced from the conjugation action of P on N . Explicitly, under
the action, p sends σ to σ ◦ (cp × cp ). The P -action on Z 2 (N, A) preserves
cohomological classes, i.e., it is trivial on H 2 (N, A). This is ensured by the
first consistency relation, which implies that σ and σ ◦ (cp × cp ) are similar to
each other through g(∗, p).

• To make the second point, we first recast g as

g̃(n, p) := g(cp−1 (n), p). (3.33)

For a given p, g̃(∗, p) can be regarded as a function from N to A, namely

g̃(∗, p) ∈ C 1 (N, A). Canonically, P acts on C 1 (N, A) by the conjugation action
of P on N . Particularly, p2 sends g̃(∗, p1 ) to g̃(cp−1
2
(∗), p1 ) for all p1 , p2 ∈ P .
Through the P -action on C (N, A), the second consistency relation just means
1

that g̃ as a map from P to C 1 (N, A) is just a twisted 1-cocyle, namely g̃ ∈

Z 1,c (P, C 1 (N, A)).

• The twisted cohomology for g

λ(cp n)/λ(n) (3.34)

for g̃
λ(n)/λ(cp−1 n) (3.35)
χ(np) = λ(n)
λ(n1 Cp1 (n2 )) λ(n1 cp1 (n2 )) λ(cp1 (n2 ))
= (3.36)
λ(n1 )λ(n2 ) λ(n1 )λ(cp1 n2 ) λ(n2 )

We can learn much from the factorized form about the possibilities of multipliers
of some important groups in physics. In the following subsections, let discuss some
of them.

3.2.1 Direct product groups

In the special case of G = N × P , the conjugation of P on N is trivial. Then, the
two consistency relations are simplified as

g(n1 n2 , p) = g(n1 , p)g(n2 , p), g(n, p1 p2 ) = g(n, p1 )g(n, p2 ), (3.37)

for all n1 , n2 , n ∈ N and p, p1 , p2 ∈ P . It is illuminating to introduce

χp (n) := g(n, p), (3.38)

for all n ∈ N and p ∈ P . Then, for each p ∈ P , χp is a group homomorphism

from N to A, and therefore is 1-cocyle of N , namely χp ∈ Z 1 (N, A). Moreover, χ

– 10 –
is a group homomorphism from P to Z 1 (N, A). Formally, this can be expressed as
χ ∈ Z 1 (P, Z 1 (N, A)).
Then, the factorization form becomes

ν(n1 p1 , n2 p2 ) = σ(n1 , n2 )χp1 (n2 )α(p1 , p2 ), (3.39)

with σ ∈ Z 2 (N, A), α ∈ Z 2 (P, A) and χ ∈ Z 1 (P, Z 1 (N, A)). It is significant to

observe that the difference of any two different χ’s cannot be equal to a cobound-
ary, since χp1 (n2 ) is not symmetric in n1 p1 and n2 p2 . Hence, there is a one-to-one
correspondence f between H 2 (N, A)×Z 1 (P, Z 1 (N, A))×H 2 (P, A) and H 2 (N ×P, A):

f : H 2 (N, A) × Z 1 (P, Z 1 (N, A)) × H 2 (P, A) → H 2 (N × P, A), (3.40)

which is given by the factorization form above.

3.2.2 The Heisenberg algebra

Specializing to an Abelian group N , Z 1 (N, U (1)) is just the Pontryagin dual of N ,
i.e., N
b = Z 1 (N, U (1)). Then, Z 1 (P, N
b ) = Z 1 (P, Z 1 (N, A)). A particularly special
case is that P = N b , and χ is the identity map in Z 1 (N b ). Then, a canonical
b, N
multiplier of N × N b is given by

α(n1 φ1 , n2 φ2 ) = φ1 (n2 ), (3.41)

for all n1 , n2 ∈ N and φ1 , φ2 ∈ N

b . This multiplier corresponds to a short exact
sequence
1 → N → Heis(N, N b) → N b → 1. (3.42)
b ) is just the α-twisted direct product of N and N
Heis(N, N b with twist given by the
canonical multiplier α. Heis(N, N b ) is called the Heisenberg group constructed from
the Abelian group N .

• N =R Recall that if N = R, R
b = R. (q, p/ℏ) ∈ R × R b can be interpreted
as a point in the phase space R × R
b of a particle on a line. Accordingly, the an
α-representation W of R × R satisfies
W (q)W (p) =eipq/ℏ W (p)W (q),
(3.43)
W (q1 )W (q2 ) = W (q2 )W (q1 ), W (p1 )W (p2 ) = W (p2 )W (p1 ).

for all q, q1 , q2 ∈ R and p, p1 , p2 ∈ R.

b They are just the integral forms of the
more familiar Heisenberg algebras:

[q̂, p̂] = iℏ, [q̂1 , q̂2 ] = 0, [p̂1 , p̂2 ] = 0. (3.44)

It is a fundamentally important that N ×Nb has a unique irreducible α-representation,

which is called the Stone-von Neumann-Mackey theorem. We will give a proof for
the result later by Mackey’s representation theory of group extensions.

– 11 –
3.2.3 Translation groups
Let us start with considering the translation group Rd of a d-dimensional flat space.
We denote all elements of Rd as x = (x1 , x2 , · · · , xd )T with every entry xi ∈ R. Here,
g satisfies We note that every multiplier of R is trivial. Then, every multiplier of
R2 = R × R is equivalent to

σ(x1 , x2 ) = g(x11 , x22 ). (3.45)

and
g(x + y, z) = g(x, z)g(y, z), g(z, x + y) = g(z, x)g(z, y). (3.46)
for all x, y, z ∈ R. Hence,
1 2
σ(x1 , x2 ) = eiF12 x1 x2 . (3.47)
Inspired from the simple case of d = 2, we claim the following form of multipliers
Fij xi1 xj2
Pd
σ(x1 , x2 ) = ei i<j , (3.48)

which can be proved by induction. Here, F is an upper-triangular matrix with

diagonal entries being zero. It is sometimes more convenient to transform σ by
i T
e− 2 x F x , which leads to the equivalent form
1 T
σA (x1 , x2 ) = ei 2 x1 Ax2 , (3.49)

where A is the skew-symmetric real matrix:

A = F − FT. (3.50)

Since every coboundary is symmetric, the d(d − 1)/2 independent entries of A rep-
resent distinct cohomological classes, i.e.,

H 2 (Rd , U (1)) = Rd(d−1)/2 . (3.51)

3.2.4 Euclidean groups

For the Euclidean group E d for d dimensions, E d is of the semiproduct form E d =
Rd ⋊ SO(d). Hence, every element can be represented as (x, R) with x ∈ Rd and
R ∈ SO(d). Note that in this convention

cR (x) = Rx. (3.52)

Using the skew-symmetric form of multipliers of the translation subgroup Rd , we

have
T g(x1 + x2 , R)
eix1 (R AR−A)x2 = . (3.53)
g(x1 , R)g(x2 , R)
Since the left-hand side of the equation is skew-symmetric while the right-hand side
is symmetric in x1 and x2 , both sides should be equal to 1.

– 12 –
 The constraint for the restricted multiplier ν is analyzed as follows. For d = 2,
the left-hand side is always equal to 1. But for d ≥ 3, an immediate consequence is
that A = 0, since RT AR = A for all R ∈ SO(d).
For all d ≥ 2, we see ln g(x, R) is additive in x, and therefore g(x, R) takes the
form
g(x, R) = eibR ·x , (3.54)
where bR is a vector determined by R. Then, the identity g(x, R1 R2 ) = g(R2 x, R1 )g(x, R2 )
leads to
bR1 R2 = R2T bR1 + bR2 . (3.55)
Let us introduce Z
b̄ = dµ(R) bR (3.56)
SO(d)

using the Haar measure of SO(d). Then,

bR = b̄ − RT b̄. (3.57)

The corresponding factor g of the multiplier is given by

eib̄·x2
g(x2 , R1 ) = . (3.58)
eib̄·R1 x2
It is significant to observe that g can always be eliminated by the transformation
e−ib̄·x .
Thus, the equivalence classes of multipliers of E 2 is determined by νR for the
translation subgroup and ω for SO(2). Since H 2 (SO(2), U (1)) = 0, the second
cohomological group is given by

H 2 (E 2 , U (1)) = H 2 (R2 , U (1)) = R. (3.59)

For E d with d ≥ 3, the equivalence classes of multipliers are completely determined

by the restricted multiplier ω for SO(d), i.e.,

H 2 (E d , U (1)) = H 2 (SO(d), U (1)) = Z2 (3.60)

3.2.5 Quantum mechanical interpretations

In quantum mechanics, H 2 (Rd , U (1)) = Rd(d−1)/2 represented by ν(x1 , x2 ) = eix1 F x2
T

has a natural interpretation. The entry Fij can be interpreted as the gauge flux
through the unit area specified by the ith and jth directions, which can be seen from

Li Lj L−1 −1
i Lj = e
iFij
(3.61)

with Li the translation for a unit length along the ith direction. A uniform magnetic
field Bij obviously respect the translations invariance. However, the gauge connection

– 13 –
Ai adopted to describe B is not invariant under translations. Lb transforms Ai (x)
to Ai (x − b). Since Ai (x) and Ai (x − b) corresponds to the same magnetic field Bij ,
they are related by a gauge transformation, i.e.,

Ai (x + b) = Ai (x) + ∂i ϕb (x), (3.62)

or equivalently Z x+b
ϕb (x) = Ai (y)dy i . (3.63)
x
Then, the translation operator is defined by
R x+b
Ai (y)dy i
Lb ψ(x) = eiϕb (x) ψ(x + b) = ei x ψ(x + b), (3.64)

i.e., a physical translation should be the corresponding spatial translation combined

with a gauge transformation. The relation between the uniform magnetic field Bij
and the forms of multipliers are given by

B = F − F T = 2A. (3.65)

It is significant to note that for translation groups the gauge invariance in quantum
mechanics is now consistent with the coboundary invariance of cohomological classes.
Then, we further consider rotations R ∈ SO(d). In two dimensions, no rotation
can change B12 , i.e., any uniform magnetic field can preserve the rotation symmetry
group SO(2). But in more than two dimensions, Bij can be rotated to −Bij . For
instance, B12 can be rotated for π through the x direction to −B12 . Hence, any
nonzero Bij breaks the symmetry group SO(d) with d ≥ 3. Consistently, the compo-
nent of translation subgroup is present in H 2 (E 2 , U (1)), but not in H 2 (E d , U (1)) for
d ≥ 3. Moreover, the trivial and nontrivial classes of H 2 (SO(d), U (1)) with d ≥ 3
can be realized by integral spins and half-integral spins, respectively.

3.2.6 Galilean group

The Galilean group G is also a semidirect product group. The normal subgroup
R3 × R3 consists of translations x and velocities v relative to the coordinate frames
of two observers, which is denoted as

(x, v) ∈ R3 × R3 . (3.66)

The quotient group G/N = SO(3) × R consists of rotations R and time translations
t, which denoted by
(R, t) ∈ SO(3) × R. (3.67)
The action of SO(3) × R on R3 × R3 is given by

c(R,t) (x, v) = (Rx + Rvt, Rv). (3.68)

– 14 –
Hence, the group multiplication is given by

((x1 , v1 ), (R1 , t1 ))((x2 , v2 ), (R2 , t2 ))

= ((x1 + R1 x2 + R1 v2 t1 , v1 + R1 v2 ), (R1 R2 , t1 + t2 )). (3.69)
The inverse of ((x, v), (R, t)) is
((R−1 vt − R−1 x, −R−1 v), (R−1 , −t)). (3.70)
We now proceed to analyze the multipliers of G using the factorization form.
• The first factor σ, the multiplier of N = R3 × R3 , has to take the form
λ
σ((x1 , v1 ), (x2 , v2 )) = exp[i (x1 · v2 − x2 · v1 )], (3.71)
2
with λ ∈ R, because the action of SO(3) the other entries of the skew-symmetric
matrix A. It is straightforward to check that such a multiplier σ is invariant
under the action of G/N = SO(3) × R.

• Similar to the Euclidean group, for the Galilean group, g can also be trivialized.
The justification can be obtained from generalizing the argument for E d . Since
g(∗, (R, t)) corresponds to a homomorphism from SO(3) × R to R c3 × R c3 ∼=
R × R . Hence,
3 3

g((x, v), (R, t)) = exp i(kR,t · x + qR,t · v). (3.72)

Then, the 1-cocycle equation leads to
kR1 R2 ,t1 +t2 = kR2 ,t2 + R2T kR1 ,t1 , (3.73)
qR1 R2 ,t1 +t2 = R2T kR1 ,t1 t2 + R2T qR1 ,t1 + qR2 ,t2 . (3.74)
For the first equation for k, integrating over both R1 and R2 gives
k̄t1 +t2 = k̄t2 , (3.75)
which simply implies k̄t is independent of t. Hence, we denote k̄t = k̄. Then,
integrating only over R1 leads to
kR,t = k̄ − RT k̄, (3.76)
from which we see kR,t is independent of t as well. Analogous analysis can be
performed for the second equation for qR,t , which leads to
qR,t = q̄ − RT q̄ − RT k̄t. (3.77)
Hence, g is given by
g((x2 , v2 ), (R1 , t1 )) = exp i[k̄ · x2 − k̄ · R1 (x2 + v2 t1 ) + q̄ · v2 − q̄ · R1 v2 ]. (3.78)
It is equal to the coboundary derived from e−i(k̄·x+q̄·v) . Hence, we can always
set g = 1.

– 15 –
 • For the multiplier α of SO(3)×R, we observe that SO(3)×R is a direct product
group. Hence, we can apply the factorization form to α. It turns out that α is
always similar to a multiplier of SO(3). Hence, we set

α((R1 , t1 ), (R2 , t2 )) = α′ (R1 , R2 ), (3.79)

with α′ a multiplier of SO(3).

Finally, we arrive at the factorization form of multipliers of the Galilean group G =

(R3 × R3 ) ⋊ (SO(3) × R),

λ
ν(((x1 , v1 ), (R1 , t1 )), ((x2 , v2 ), (R2 , t2 ))) = exp[i (x1 ·v2 −x2 ·v1 )]ω(R1 , R2 ). (3.80)
2
Later when discussing the σ-representations of the Galilean group, we shall see ℏλ is
the mass of a particle, and [ω] ∈ Z2 = {±1} corresponds to integral and half-integral
spins, respectively.

3.3 Clifford algebras

Consider N generators γi with i = 1, 2 · · · , N . The Clifford algebra is given by

{γi , γj } = 2δij 1. (3.81)

The Clifford algebra can be understood as a projective representation of the group:

G = Z2 × Z2 × · · · × Z2 , (3.82)

namely a direct product of N two-element groups. The multiplier of the projective

representation is specified by
(
1 j≤i
ν(ei , ej ) = (3.83)
−1 j > i
Two irreps if N is even, and a unique irrep if N is odd. For any Abelian group,
all irreps for a given multiplier have the same dimensionality.

4 The projective representations of crystal symmetries

4.1 Realization on flux lattices
Let us consider a set of lattice sites and hopping amplitudes among them, which give
rise to a tight-binding model,
X
Ĥ = Hij a†i aj . (4.1)
ij

– 16 –
Here, a†i and aj are the particle creation and annihilation operators at sites i and
j, respectively. Hij represents the hopping amplitudes tij from site j to i if i ̸= j
and the onsite energy ϵi at site i if i = j. H is a Hermitian matrix and called the
one-particle Hamiltonian of the tight-binding model.
Each hopping amplitude tij may have a phase eiϕij (such that tij = |tij |eiϕij ),
which is called the gauge connection of the lattice model. For each closed loop C
formed by successive hoppings, one can compute the product WC of the phases of all
the hopping amplitudes involved,
Y
WC = eiϕij (4.2)
⟨ij⟩∈C

Here, WC is called the Wilson loop operator of the loop C, and the gauge flux ΦC
through the loop C is given by
WC = e−iΦC . (4.3)
For each site i, we may change the phase of a†i for each i by an arbitrary eiθ .
i

Accordingly, the hopping amplitudes are transformed as tij 7→ eiθ tij e−iθ , which lead
i j

to the gauge transformation:

i j
eiϕij 7→ eiθ eiϕij e−iθ (4.4)

An immediate result is that WC = e−iΦC is invariant under any gauge transformation.

This can be seen from that the ending site of a hopping is the starting site of the
next hopping in a loop C, and therefore all phase changes involved are cancelled out.
To summarize, the gauge fluxes are gauge-invariant quantities, whereas the gauge
connections are not.
Only gauge-invariant quantities are physical. In the current case, the gauge
flux configuration completely determines the physics of the model. Hence, a spatial
transformation R that leaves the crystal and the gauge flux configuration invariant
is regarded as a symmetry of the system. However, R does not necessarily preserve
the gauge-connection configuration A with Aij = eiϕij . After the action of R, A is
generally changed to another one,

A′ = RART . (4.5)

It is significant to note that the two gauge-connection configurations A and A′ de-

scribe the same flux configuration. Hence, they are related by a gauge transformation
GR . Here, GR is a diagonal matrix indexed by the lattice sites, i.e., [GR ]ii = eiθR ,
i

where the phase assigned to the ith site is eiθR . Then, the physical symmetry operator
i

in this case should be the combination

ρ(R) = GR R. (4.6)

– 17 –
That is, after the spatial transformation R, the gauge transformation GR is needed
to recover the original gauge connection configuration A. Notably, it is ρ(R) = GR R
that commutes with the Hamiltonian H, i.e.,

[ρ(R), H] = 0. (4.7)

The commutation relation is equivalent to the requirement,

tij = GR (i)tR−1 (i)R−1 (j) G∗R (j), (4.8)

where GR (i) = eiθR , namely the phase assigned to site i, and R(i) is the site trans-
i

formed from i by R.
Then, we consider the successive action of two spatial symmetries, ρ(R1 ) =
GR1 R1 and ρ(R2 ) = GR2 R2 . Their product is given by

ρ(R1 )ρ(R2 ) = ∆G (R1 , R2 )ρ(R1 R2 ), (4.9)

with
∆G (R1 , R2 ) = GR1 R1 GR2 R1−1 /GR1 R2 . (4.10)

∆G (R1 , R2 ) is a diagonal matrix with ith diagonal entry being

[∆G (R1 , R2 )]ii = GR1 (i)GR2 (R1−1 (i))/GR1 R2 (i) (4.11)

and therefore represents a gauge transformation.

It is clear that ∆G (R1 , R2 ) commutes with all possible symmetry-preserving
Hamiltonians. Particularly, let us presume the usual case that H is a connected
lattice model, i.e., any two sites are connected by hoppings. The presumption suf-
ficiently leads to the fact that ∆G (R1 , R2 ) is proportional to the identity matrix,
namely
[∆G (R1 , R2 )]ij = ν(R1 , R2 )δij (4.12)

with ν(R1 , R2 ) ∈ U (1). Thus, we have the projective multiplication relation:

ρ(R1 )ρ(R2 ) = ν(R1 , R2 )ρ(R1 R2 ). (4.13)

If ν and ν ′ are related by transforming ρ(R) to ρ′ (R) = χ(R)ρ(R) with χ(R) ∈

U (1), the two multipliers belong to the same cohomology class. It is significant to
note that the cohomology class of such a projective representation is independent of
the choice of gauge connections, and is solely determined by the flux configuration. In
other words, the cohomology class of the projective representation is gauge invariant.

– 18 –
4.2 The classification with time-reversal invariance
Let us consider the symmetry group:

G × ZT2 . (4.14)

Here, G is a symmetry group whose representations are unitary, and ZT2 is the two-
element group generated by time reversal T . T should be represented by an anti-
unitary operator.
Then, each group element can be written as gT a with g ∈ G and a = 0, 1.
Suppose that under the projective representation ρ, the phase factor λ arises through

ρ(g1 T a1 )ρ(g2 T a2 ) = λ(g1 T a1 , g2 T a2 )ρ(g1 g2 T a1 +a2 ). (4.15)

We shall prove that by appropriately modifying the phase of each operator ρ(gT a ),
we can always transform λ(g1 T a1 , g2 T a2 ) into the form,

λ̃(g1 T a1 , g2 T a2 ) = ν(g1 , g2 )ω(T a1 , T a2 ), (4.16)

where ν(g1 , g2 ), ω(T a1 , T a2 ) ∈ Z2 . As a result, we have the simplified expression of

the twisted cohomology group:

H 2,c (G × ZT2 ) ∼
= H 2 (G, Z2 ) × Z2 . (4.17)

Hence, the computation is reduced to computing the second cohomology group of G

with coefficients in Z2 = {±1}.
We start the proof with observing that for all g ∈ G,
λ(g, T )
ρ(g)ρ(T ) = λ(g, T )ρ(gT ) = ρ(T )ρ(g), (4.18)
λ(T, g)

which motivates us to modify the phase of each ρ(g) as

s
λ(T, g)
ρ̃(g) := ρ(g), (4.19)
λ(g, T )

Note that ρ(T ) is an anti-unitary operator, i.e., ρ(T )c = c∗ ρ(T ) for c ∈ C. Hence,

ρ̃(g)ρ(T ) = ρ(T )ρ̃(g). (4.20)

We further modify the operators for the other half of group elements as

(4.21)
p
ρ̃(gT ) := λ(g, T )λ(T, g)ρ(gT ),

for all g ∈ G. Note that ρ̃(T ) = ρ(T ). Then, one observes that

ρ̃(g)ρ̃(T ) = ρ̃(gT ). (4.22)

– 19 –
 Let λ̃ denote the phase factor for ρ̃. Restricting on G, λ̃ satisfies

ρ̃(g1 )ρ̃(g2 ) = λ̃(g1 , g2 )ρ̃(g1 g2 ) (4.23)

for all g1 , g2 ∈ G. The left-hand side commutes with ρ̃(T ), so does the right-hand
side. Hence, ν := λ̃|G×G ∈ Z2 = {±1}. On the other hand, λ̃(T, T ) appears in

ρ̃(T )ρ̃(T ) = λ̃(T, T )1. (4.24)

Clearly, λ̃(T, T ) commutes with ρ̃(T ), and therefore ω(T, T ) := λ̃(T, T ) ∈ Z2 .

Finally, it is straightforward to check that

ρ̃(g1 T a1 )ρ̃(g2 T a2 ) = ρ̃(g1 )ρ̃(T a1 )ρ̃(g2 )ρ̃(T a2 )

= ρ̃(g1 )ρ̃(g2 )ρ̃(T a1 )ρ̃(T a2 )
(4.25)
= ν(g1 , g2 )ρ̃(g1 g2 )ω(T a1 , T a2 )ρ̃(T a1 +a2 )
= ν(g1 , g2 )ω(T a1 , T a2 )ρ̃(g1 g2 T a1 +a2 ).

Above, we have repeatedly used the relations: ρ̃(g)ρ̃(T ) = ρ̃(gT ) and ρ̃(g)ρ̃(T ) =
ρ̃(T )ρ̃(g).

4.3 The canonical form of multipliers

Each symmorphic group is a semidirect product of the translation group L and the
point group P . More specifically, a symmorphic group ΓP may be written as

ΓP ∼
= L ⋊D P. (4.26)

Here, D is the crystal class that specifies how P acts on L.

As a semidirect product, the multipliers of a symmorphic group satisfy Mackey’s
canonical form, and therefore can be decomposed as

ν((t1 , R1 ), (t2 , R2 )) = σ(t1 , R1 t2 )γ(R1 t2 , R1 )α(R1 , R2 ). (4.27)

With this decomposition, the 2-cocycle equation requires σ, γ, α satisfy

σ(t1 , t2 )σ(t1 + t2 , t3 ) = σ(t1 , t2 + t3 )σ(t2 , t3 ),

(4.28)
α(R1 , R2 )α(R1 R2 , R3 ) = α(R1 , R2 R3 )α(R2 , R3 ),

and
σ(RT t1 , RT t2 ) γ(t1 + t2 , R)
= , (4.29)
σ(t1 , t2 ) γ(t1 , R)γ(t2 , R)
γ(t, R1 R2 ) = γ(t, R1 )γ(R1T t, R2 ). (4.30)
Let us consider a particular case with both σ and α trivial. That is, the only
nontrivial components of the multiplier correspond to the phase factors modifying the

– 20 –
algebraic relations between translations and point-group elements. Then, equation
(4.29) is simplified to be

γ(t1 + t2 , R) = γ(t1 , R)γ(t2 , R). (4.31)

Hence, γ can be written in the form,

γ(t, R) = e−iκR ·t . (4.32)

Substituting it into Eq. (4.30), we obtain

e−iκR1 R2 ·t = e−iκR1 ·t e−iR1 κR2 ·t , (4.33)

which is equivalent to

κR1 R2 = κR1 + R1 κR2 mod L.

b (4.34)

Here, Lb is the reciprocal lattice dual to L. The similarity between (2.5) and (4.37)
motivates us to interpret κR as momentum-space fractional translations on the re-
ciprocal lattice.
The momentum k corresponds to the representation ρ(t, 1) = eik·t of transla-
tions. Then, the operation of (t′ , R) on k is given by

ρ(t′ , R)ρ(t, 1)[ρ(t′ , R)]†

=γ(Rt, R)ρ(t′ + Rt, R)ρ(−RT t′ , RT )/γ(−t′ , R)
, (4.35)
=γ(Rt, R)g(−t′ , R)ρ(Rt, 1)/γ(−t′ , R)
=γ(Rt, R)ρ(Rt, 1)

i.e.,
T k−RT κ )·t
eik·t 7→ ei(R R
(4.36)
Since
κR + RκRT = 0 mod L,
b (4.37)
we obtain
T k+κ
eik·t 7→ ei(R RT
)·t
. (4.38)
Thus, the operation of R on momentum space is given by

R : k 7→ Rk + κR . (4.39)

Hence, κR is indeed the fractional translation associated to R on the reciprocal

lattice.

– 21 –