Berry’s phase and applications to periodic systems

Allan E. Sadun
[email protected]
MIT Department of Physics
(Dated: April 19, 2016)
Incomplete. This is an abstract, and will say all sorts of abstract-like things.

I. INTRODUCTION construction to hold regardless of the dimensionality of

~k-space. We can apply Stokes’ theorem to replace line
Although Berry’s phase was originally discovered in integrals along A(n) with surface integrals along B (n) ,
the context of cyclic adiabatic evolution of a quantum banishing the choice of gauge from our equations.
Hamiltonian, it can be more generally thought of as a We can derive an explicit formula for B (n) in terms
property of any differentiable cyclic family of normalized of the Hamiltonian using first-order perturbation theory.
vectors in a Hilbert space. Given two vectors |ψ1 i and Using tensor notation, A(n) is defined by
|ψ2 i, we can define a “phase shift” φ12 by
A(n)
α = i hn| ∂α |ni
hψ1 |ψ2 i and B (n) by
e−iφ12 =
| hψ1 |ψ2 i |
(n) (n) (n)
Bαβ = ∂α Aβ − ∂α Aβ
which has the property that for any set of states
|ψ1 i , |ψ2 i , . . . , |ψn i, the “total phase shift” e−iγ =
e−i(φ12 +φ23 +...+φn1 ) is invariant under phase change = i h∂α n|∂β ni − i h∂α n|∂β ni
of |ψi i. (To see this, consider that the product which, since
hψ1 |ψ2 i hψ2 |ψ3 i is invariant under phase change of |ψ2 i).
As n → ∞ and neighboring states get closer and closer X hm| ∇~k Ĥ |ni
to one another, the discrete phase shift can be replaced ∇~k |ni = |mi
En − Em
with a differential phase shift formula m6=n

dγ = i hψ| d |ψi evaluates to

(n)
X hn| ∂α Ĥ |mi hm| ∂β Ĥ |ni
where we have made use of the fact that ψ is always Bαβ = 2i
(En − Em )2
normalized to get rid of the denominator term and ensure m6=n
that γ, the Berry phase around the cycle, is real. It is
less obvious in this continuous case that γ is only gauge- This construction does not work for A(n) because pertur-
invariant up to a multiple of 2π, but it is still true - if bation theory assumes a particular gauge for the eigen-
the family of |ψi’s is parameterized by λ ranging from 0 states, and A(n) depends on a choice of gauge.
to 1 with |ψ(0)i = |ψ(1)i, multiplying |ψ(λ)i by e−2πiλ Curiously, as the Berry curvature B (n) is completely
shifts γ by 2π. gauge-invariant, it would seem that by integrating B (n)
For a quantum system with a Hamiltonian Ĥ(~k) which over a surface, we could compute a Berry phase γn ex-
depends on some external parameter or set of parameters actly, without any “up to 2π” ambiguity. This is not the
~k, the differential phase shift of the nth eigenstate |ni is case - crucially, there may be points where B (n) is not
defined, perhaps due to a degeneracy in En , and so inte-
known as the Berry connection A(n) = i hn| ∇~k |ni, a one-
grating over different surfaces will yield different results
form in ~k-space that is closely related to the adiabatic depending on whether the surface passes under or over
evolution of a system in |ni. The Berry connection is not these singularities. However, because γn is determined
gauge-invariant; if |ni is multiplied by some phase factor up to 2π, this implies the Chern theorem, which states
~
eiµ(k) , then A(n) will become A0(n) = A(n) − ∇µ. Only that the integral of the Berry curvature around a closed
the cyclic integral γn is gauge-invariant, and even that manifold must be a multiple of 2π. We can therefore as-
only up to a multiple of 2π. sign to each singularity of the Berry curvature a “Chern
We typically resolve this by using the Berry curva- number”, an integer topological charge describing the de-
ture B (n) , the anti-symmetric two-form produced by tak- gree to which the Berry curvature warps around it.
ing the exterior derivative of the Berry connection. ~k-
space must be at least 2-dimensional for this to make
sense. If ~k-space is 3-dimensional, this takes the form Example: a spin-1/2 system
B (n) = ∇ × A(n) , and the zero-curl property of gradients
makes it clear that B (n) is independent of µ, but the Consider the Hamiltonian for an electron spin in a
theory of differential forms allows this gauge-invariant magnetic field ~k. Ĥ = ~k · ~σ , where ~σ are the three
Berry’s phase and applications to periodic systems 2

Pauli matrices for a spin-1/2 system. The eigenvalues Zak’s phase and Wannier functions
are ±|k|/2, so they are nondegenerate for ~k 6= ~0. Using
the known formulas for eigenstates of this Hamiltonian, Intuitively, because k and x are conjugate variables,
the Berry curvature can be computed to come out to one might expect i∂k to be some kind of position opera-
B~+ = ~k/2|k|3 . Integrating around ~k = ~0, the flux is 2π, tor, and one might expect Zak’s phase γn to be somehow
so the Chern number for this degeneracy is 1. (This ex- equal to an “expectation value” of position (2π/a)hx̂i in
ample illustrates my motivation for the letters A and B; the nth electronic band. This is further supported by
for an electron spin, the Berry curvature is in the same the fact that just as Zak’s phase is only defined up to
direction as the magnetic field, and the Berry connection translations by 2π, it is impossible to define an “expec-
is related to the magnetic vector potential.) tation value” of position except up to translations by a.
For some quantum systems, such as 1-dimensional However, coming up with a theory of a multi-valued ex-
wave functions where the eigenvectors can be taken to pectation value of position requires a little work.
be everywhere real, the Berry phase is zero and irrele- One way of defining position is by using Wannier func-
vant. But Berry phases crop up all over condensed mat- tions, unitary transforms of the electronic bands which
ter theory due to their particular nature as multi-valued are localized to be near a particular reference position X
yet measurable quantities; they are physically observable (typically a lattice vector). They are defined by
in quantum systems, but because they are only defined a
Z
(n)
up to 2π, there is no way they can be expressed as the wX (x) = e−ikX ψn,k (x)dk
expectation value of an operator like most observables. 2π BZ
This is especially the case in periodic systems where most Z
everything is only defined up to a lattice shift. In this a
paper I will explore some manifestations and applications = eik(x−X) un,k (x)dk
2π BZ
of Berry’s phase in periodic systems.
where “BZ” refers to the Brillouin zone. For x far
away from X, the integrand varies rapidly with k, and
the Wannier function is small; for x near X where the
II. 1D CRYSTALS
electronic bands have little probability, the integrand is
small, and the Wannier function is also small. Because
2
un,k is periodic, translating X by a lattice vector trans-
p̂
When a Hamiltonian in one dimension Ĥ = 2m + V̂ is (n)
lates wX by the same lattice vector. The electronic
invariant under translations by a, Bloch’s theorem tells bands and the Wannier functions have a Fourier-like re-
us that the eigenvectors can be taken to obey boundary lationship, which implies an inverse transform
conditions of the form ψn,k (x + a) = eika ψn,k (x). For a
fixed k, we can let un,k (x) = e−ikx ψn,k (x), and get un,k 2π X −ik(x−X) (n)
un,k (x) = e wX (x)
by solving the eigenproblem of the modified Hamiltonian a
2 X
Ĥk = (p̂+~k)
2m + V̂ with the periodic boundary condi-
tion u(x + a) = u(x). So for a fixed n, we can find a where the sum is over all values of X that are a lat-
smooth family of solutions ψn,k , known as the nth elec- tice
P vector a apart. (Verifying this requires the identity
ikX
tronic band, parameterized by k. The band can either be X e = (2π/a)δ(k), which is true regardless of the
interpreted as the solution ψ under changing boundary offset of the lattice of X values.) This allows us to ex-
conditions, or as the solution u under a changing Hamil- pand Zak’s phase in the Wannier basis (neglecting factors
tonian. of 2π/a):
Because e2πi = 1, the boundary condition imposed on Z a
ψn,k is exactly the same boundary condition imposed on i hun,k | ∂k |un,k i = iu∗n,k (x)∂k un,k (x) dx
ψn,k+2π/a , and so we can constrain our allowable phase 0

gauges by requiring ψn,k+2π/a = ψn,k . It is common to Z a

consider k only up to a modulus of 2π/a, labeling the re- X 0 (n)∗ (n)
gion k ∈ [−π/a, π/a] the “first Brillouin zone”. We must = eik(x−X ) wX 0 i(−i)(x − X)e−ik(x−X) wX dx
0 X,X 0
be careful doing this, as the Hamiltonian Ĥk does not
equal the Hamiltonian Ĥk+2π/a . But because un,k+2π/a Z a
is constrained to be e−2πix/a un,k , the “open-path” Berry =
X 0
(x − X)eik(X−X ) wX 0 wX dx
(n)∗ (n)
phase from un,k to un,k+2π/a is just as gauge-invariant as 0 X,X 0
the cyclic Berry phases in the previous section. In this
application it is known as Zak’s phase: 0
Integrating eik(X−X ) over a Brillouin zone,
Z a X
Z 2π/a (n)∗ (n)
γn = (x − X)δX,X 0 wX 0 wX dx
γn = i hun,k | ∂k |un,k i dk 0 X,X 0
0
Berry’s phase and applications to periodic systems 3
Z a
(n) with a way of analyzing and describing electric polariza-
X
= (x − X)|wX |2 dx
0 X
tion of a crystal in a way which removes much of the
ill-definedness, using Berry’s phase. It’s very, very simi-
lar to the previous section and builds on it.
Z ∞
(n)
= (x − X)|wX |2 dx
−∞

= hx − XiwX
IV. CHERN NUMBERS IN 2D BRILLOUIN
where we used the translational symmetry of the Wan- ZONES
nier basis to replace an integral in one unit cell over many
functions with a sum over many unit cells of one func- Incomplete. In 2D crystals, Bloch’s theorem still
tion. This confirms the intuitive result: Zak’s phase can
holds, but ~k is a vector now, with components kx and ky .
be thought of as a representation of the “expected posi-
This allows us to define a Berry curvature on the Bril-
tion” of an electron within a band, with expected position
louin zone, and once again the Chern number becomes
defined using Wannier functions. This also demonstrates
relevant. In this section, I’ll discuss how Berry’s phase
that although the Wannier functions themselves are sym-
is just as applicable to electromagnetism as to quan-
metric only when X is translated by a lattice vector, the
tum mechanics, and discuss photonic examples of crystals
expectation value of x − X according to wX is the same
with zero Chern number and crystals with nonzero Chern
for all values of X, regardless of offset.
number. I’ll touch on how you can make very interesting
(Note that the Wannier functions are gauge-
waveguides by putting these two types of crystals next
dependent; changing the choice of phase of uk,n contin-
to each other.
uously changes the various Wannier functions. But be-
V. CONCLUSION
cause gauge transformations can only change Zak’s phase
by a multiple of 2π, they can only change the expectation
value of x − X by a multiple of a.) Incomplete. Berry’s phase is applicable to many sit-
In multi-dimensionally periodic systems, such as real uations in physics where monodromy comes up. Periodic
3D crystals, similar results hold. ∂k becomes ∇~k , and in- systems are a class of situation where monodromy tends
tegrals dk become integrals d3~k. This makes Zak’s phase to come up.
less like a Berry phase, because it is no longer the result
of the line integral of a Berry connection along a con-
tour, but the Berry-phase integral form remains useful
for computations.
Acknowledgments
III. ELECTRIC POLARIZATION
The author is grateful to Stephen Johnson and Bo
Incomplete. This section will describe the work of Zhen for a background in Bloch states of photonic crys-
David Vanderbilt. King-Smith and Vanderbilt came up tals, as well as to Michael Pretko for advice and support.

[1] Berry, M.V. ”Quantal phase factors accompanying adia- [5] Vanderbilt, D. “Berry Phases and Curvatures in
batic changes”. 1984, Proceedings of the Royal Society of Electronic-Structure Theory”. Presented March 2006,
London. Series A, Mathematical and Physical Sciences, at March APS Meeting, Baltimore. Accessible at
Vol. 392, No. 1802. pp. 45-57. http://www.physics.rutgers.edu/ dhv/talks/rahman.pdf.
[2] Lu, L., Joannopoulos, J. D., & Soljačić, M. ”Topological [6] Resta, R. “Manifestations of Berry’s phase in molecules
photonics”. 2014, Nature Photonics, 8, 821. and condensed matter”. 2000, Journal of Condensed Mat-
[3] Moore, J. “Geometry and topology in many-particle ter Physics, 12. R107-R143.
systems”. Notes for Physics 250, UC Berkeley, [7] Spaldin, N. “A beginner’s guide to the modern theory of
2009. Available at http://socrates.berkeley.edu/ je- polarization”. 2012, Journal of Solid State Chemistry, Vol-
moore/Physics 250 files/thoulessphases.pdf. ume 195. p2-10.
[4] Simon, B. “Holonomy, the quantum adiabatic theorem, [8] Zak, J. “Berry’s Phase for Energy Bands in Solids”. 1989,
and Berry’s phase”. 1983, Physical Review Letters, 51, Physical Review Letters, Vol. 62, No. 23. pp. 2747-2750.
2167