Berry Connection and Berry Curvature

The article Geometric (Berry) Phase ends with a loop integral: transport an eigenstate slowly around a closed circuit in parameter space and it returns wearing a phase $\gamma_n$ that remembers the path. This article opens that integral up. The integrand, $\mathbf{A}_n = i\langle n|\nabla_{\mathbf{R}}n\rangle$ , is the Berry connection — a vector field behaving, line for line, like the electromagnetic vector potential: real, shifted by a gradient under rephasing, unmeasurable at a point. Its curl, the Berry curvature $\boldsymbol{\Omega}_n$ , is the gauge-invariant field strength, and by Stokes' theorem the geometric phase is the curvature flux through the loop. Here we prove the gauge law and the reality of the connection, derive the sum-over-states formula that makes the curvature manifestly gauge-invariant and exposes degeneracies as its point sources, and work spin-½ on both gauge patches to exhibit a monopole of charge $-\tfrac12$ at $\mathbf{B}=0$ . Integrating the curvature over a closed surface then forces a quantized answer — the Chern number — proved analytically ( $-2\pi$ flux, $C=-1$ ) and numerically, where a plaquette algorithm on a crude $8\times8$ grid still returns $-1.000000000000$ . That is the deepest lesson here: the curvature is a local field you can compute, but its total flux is topology — the reason Hall conductances quantize.

Key takeaways

The Berry connection $\mathbf{A}_n = i\langle n|\nabla_{\mathbf{R}}\,n\rangle$ is a real vector field on parameter space; rephasing $|n\rangle \to e^{i\chi}|n\rangle$ shifts it by $-\nabla\chi$ , like the electromagnetic vector potential.
The Berry curvature $\boldsymbol{\Omega}_n = \nabla\times\mathbf{A}_n$ is gauge-invariant; the geometric phase around a loop equals the curvature flux through it (Stokes).
The sum form $\boldsymbol{\Omega}_n = i\sum_{m\neq n}\frac{\langle n|\nabla H|m\rangle\times\langle m|\nabla H|n\rangle}{(E_m-E_n)^2}$ needs no phase choice and diverges at degeneracies — they are the sources.
Spin-½: the curvature is a monopole $\boldsymbol{\Omega}_\pm = \mp\hat{\mathbf{B}}/2B^2$ ; flux through any enclosing surface $= \mp2\pi$ , so the Chern number is $C = \mp1$ .
A plaquette computation returns that integer exactly even on an 8×8 grid — totals of Berry flux are topology, not arithmetic.
Key idea: the geometric phase has a field theory — a gauge potential $\mathbf{A}_n$ whose gauge-invariant curl $\boldsymbol{\Omega}_n$ acts as a magnetic field on parameter space, with degeneracies as monopoles and integer total charge.

The core insight

The geometric phase is governed by a gauge field on parameter space: the Berry connection $\mathbf{A}_n = i\langle n|\nabla_{\mathbf{R}}n\rangle$ is the potential, its curl $\boldsymbol{\Omega}_n$ the gauge-invariant curvature, and the phase around a loop is the curvature flux — mirroring how magnetic flux sets the Aharonov–Bohm phase.

Where this lives today

Berry curvature is now a standard column in band-structure tables. In crystals the parameter space is the Brillouin zone, and the curvature enters the semiclassical equations of motion as an anomalous velocity $\dot{\mathbf{r}} \supset -\dot{\mathbf{k}}\times\boldsymbol{\Omega}(\mathbf{k})$ — electrons drift sideways with no magnetic field present. This produces the intrinsic anomalous Hall effect, shapes transport in graphene's valleys, and underlies the theory of electric polarization as a Berry phase of filled bands; the Chern number classifies quantum Hall systems and Chern insulators, and its time-reversal-symmetric descendants ( $\mathbb{Z}_2$ invariants) define topological insulators. The objects are measured, too: ultracold atoms realized the Haldane model (2014), matter-wave interferometry has mapped Berry curvature across a band, and the plaquette algorithm above is essentially how first-principles codes assign Chern numbers to real materials.

The physical picture

Carry a tangent vector around a closed loop on a sphere, keeping it as parallel as you can, and it comes back rotated by the solid angle enclosed. Nothing rotated it locally; the twist is the surface's curvature, revealed only by the round trip. A quantum state's phase behaves the same way: the connection is the rule for staying 'parallel' at each step, the curvature the failure of those adjustments to commute. The electromagnetic analogy is an exact structural match, not decoration. Parameter space is threaded by an effective magnetic field $\boldsymbol{\Omega}_n$ ; a dragged state behaves like a charge encircling flux, and as in the Aharonov–Bohm setup the potential is unobservable pointwise while its loop integral is physical. Gauge transformations, field strength, monopoles, quantized charge — every row of the dictionary has a Berry-side entry.

Figure

Degeneracy as a monopole: curvature flux through a loop in parameter space

Why does the curvature 'know about' other eigenstates? Because moving $|n(\mathbf{R})\rangle$ through parameter space borrows amplitude, at first order, from the neighbors with weight $\langle m|\nabla H|n\rangle/(E_n-E_m)$ . The curvature is the interference of these virtual excursions — hence its divergence near a level crossing: a degeneracy is a source, radiating Berry flux like a magnetic charge. The payoff is quantization — a monopole's total flux cannot change continuously. The punchline figure shows the experiment: a plaquette sum on an $8\times8$ grid returns the Chern number to machine precision, and a two-band model's integer jumps only where the gap closes.

Figure

Punchline: the Chern integer survives a crude grid — and jumps only when the gap closes

The connection and its gauge law

Start where Geometric (Berry) Phase leaves off: the geometric phase is the loop integral of $i\langle n|\nabla_{\mathbf{R}}n\rangle$ . Give the integrand its own life. For each non-degenerate level of $H(\mathbf{R})$ , define

\mathbf{A}_n(\mathbf{R}) \;=\; i\,\langle n(\mathbf{R})|\,\nabla_{\mathbf{R}}\,n(\mathbf{R})\rangle \qquad (1)

where: $\nabla_{\mathbf{R}}$ — gradient in the external parameters (field components, flux, nuclear coordinates…); $|n(\mathbf{R})\rangle$ — a smooth local choice of instantaneous eigenstate. Differentiating $\langle n|n\rangle = 1$ shows $\langle n|\nabla n\rangle$ is purely imaginary, so the $i$ makes $\mathbf{A}_n$ real. But $|n(\mathbf{R})\rangle$ is only defined up to a phase at each point — choosing the phases is choosing a gauge, and the connection responds the way a potential must:

|n\rangle \;\to\; e^{i\chi(\mathbf{R})}|n\rangle \quad\Longrightarrow\quad \mathbf{A}_n \;\to\; \mathbf{A}_n - \nabla_{\mathbf{R}}\chi \qquad (2)

where: $\chi(\mathbf{R})$ — any smooth single-valued function. This is verbatim the electromagnetic gauge transformation: the connection at a point is meaningless, while around a closed loop the shift integrates to $2\pi\times$ integer, leaving $e^{i\gamma_n}$ gauge-invariant. What we need is the local field whose flux reproduces those loop-level facts.

Derivation AGauge law and reality of the Berry connection

We are proving the two structural facts that make equation (1) a genuine gauge potential: $\mathbf{A}_n$ is real, and a smooth rephasing of the eigenbasis shifts it by an exact gradient. The strategy is direct computation from the definition — no dynamics enters, which is the point: these are kinematic properties of a phase-ambiguous basis. Normalization alone forces reality; the product rule alone forces the gauge law.

Step 1. Differentiate the normalization $\langle n(\mathbf{R})|n(\mathbf{R})\rangle = 1$ along any parameter direction:

\langle \nabla n|n\rangle + \langle n|\nabla n\rangle = 0 \;\;\Longrightarrow\;\; \langle n|\nabla n\rangle^{*} = -\,\langle n|\nabla n\rangle .

Why this matters: a number equal to minus its own conjugate is purely imaginary, so $i\langle n|\nabla n\rangle$ is real — a legitimate potential.

Step 2. Rephase the basis, $|n'\rangle = e^{i\chi(\mathbf{R})}|n\rangle$ , and apply the product rule inside the definition:

\mathbf{A}_n' = i\langle n'|\nabla n'\rangle = i\langle n|e^{-i\chi}\left( i(\nabla\chi)e^{i\chi}|n\rangle + e^{i\chi}|\nabla n\rangle \right) = \mathbf{A}_n - \nabla\chi .

Why this matters: the inhomogeneous term is an exact gradient — precisely electromagnetism's freedom in $\mathbf{A}$ . Everything built from $\mathbf{A}_n$ must survive this shift before being called physical.

Step 3. Integrate the shift around a closed loop: $\oint_C \nabla\chi\cdot d\mathbf{R} = 2\pi\times$ integer, since $e^{i\chi}$ must return to itself. So the geometric phase is gauge-invariant modulo $2\pi$ , and the curl of $\mathbf{A}_n$ is gauge-invariant exactly (the curl of a gradient vanishes). Verified: in the verification script the spin-½ connection components compute real, and a random smooth gauge function leaves every discretized-curvature plaquette phase unchanged to machine precision.

Curvature: the gauge-invariant field

Since only gradient-shifts are allowed, take the curl and the ambiguity dies. For a three-dimensional parameter space,

\boldsymbol{\Omega}_n(\mathbf{R}) \;=\; \nabla_{\mathbf{R}} \times \mathbf{A}_n(\mathbf{R}) \qquad (3)

where: $\boldsymbol{\Omega}_n$ — the Berry curvature, the analog of $\mathbf{B} = \nabla\times\mathbf{A}$ . In any dimension the honest object is the two-form $\Omega^n_{\mu\nu} = \partial_\mu A^n_\nu - \partial_\nu A^n_\mu$ (in 3D, $\Omega_{\mu\nu} = \epsilon_{\mu\nu\lambda}\Omega_\lambda$ ). Gauge invariance is one line: under (2) it shifts by $\partial_\mu\partial_\nu\chi - \partial_\nu\partial_\mu\chi = 0$ . Stokes' theorem converts the loop integral into a flux — the working form for everything that follows:

\gamma_n \;=\; \oint_{C}\mathbf{A}_n\cdot d\mathbf{R} \;=\; \iint_{S}\boldsymbol{\Omega}_n\cdot d\mathbf{S} \qquad (4)

where: $S$ — any surface with boundary $C$ . The phase is the Berry flux through the circuit. Notice the quiet condition 'any surface': if two spanning surfaces gave different fluxes, their difference — a closed surface — would enclose net Berry charge. That is the door to monopoles and Chern numbers.

The sum-over-states form

Definition (3) still differentiates basis states. There is a form with no basis derivatives at all:

\boldsymbol{\Omega}_n \;=\; i\sum_{m\neq n} \frac{\langle n|\nabla H|m\rangle \times \langle m|\nabla H|n\rangle}{(E_m - E_n)^{2}} \qquad (5)

where: $\nabla H$ — the parameter-gradient of the Hamiltonian, written without ever choosing eigenstate phases; $(E_m-E_n)^2$ — the squared gap to each neighbor. Equation (5) says three things at once: the curvature is manifestly gauge-invariant; it is built from virtual transitions (the matrix elements first-order perturbation theory mixes in); and it diverges as the gap closes.

Go deeper · Derivation B · 1 minSum-over-states formula for the curvature

Derivation BSum-over-states formula for the curvature

We are deriving equation (5) from the curl definition, because this form makes gauge invariance manifest, exposes degeneracies as curvature sources, and is the numerically stable route used in every band-structure code. The plan: expand the curl, kill the diagonal term with the reality lemma, then trade basis-state derivatives for Hamiltonian matrix elements via the off-diagonal Hellmann–Feynman relation. Watch which steps use only completeness and which use the eigenvalue equation.

Step 1. Expand the curl of $\mathbf{A}_n = i\langle n|\nabla n\rangle$ ; the symmetric second-derivative piece is killed by the cross product, leaving

\boldsymbol{\Omega}_n = i\,\langle\nabla n| \times |\nabla n\rangle \equiv i\sum_{\mu\nu}\hat{e}_\mu\epsilon_{\mu\nu\lambda}\langle\partial_\nu n|\partial_\lambda n\rangle .

Step 2. Insert $\sum_m |m\rangle\langle m| = 1$ between the derivative states. The $m=n$ term is $\langle\nabla n|n\rangle\times\langle n|\nabla n\rangle$ — a cross product of a vector with its own negative (reality lemma), zero. Only $m\neq n$ survives:

\boldsymbol{\Omega}_n = i\sum_{m\neq n}\langle\nabla n|m\rangle \times \langle m|\nabla n\rangle .

Why the diagonal term dies: $\langle n|\nabla n\rangle = -i\mathbf{A}_n$ with $\mathbf{A}_n$ real — bra and ket factors are anti-parallel, and their cross product vanishes.

Step 3. Differentiate $H|n\rangle = E_n|n\rangle$ with respect to a parameter and project onto $\langle m|$ , $m \neq n$ :

\langle m|\nabla H|n\rangle + \langle m|H|\nabla n\rangle = E_n\langle m|\nabla n\rangle \;\Longrightarrow\; \langle m|\nabla n\rangle = \frac{\langle m|\nabla H|n\rangle}{E_n - E_m} .

Why this is familiar: it is the first-order perturbation-theory mixing coefficient — moving in parameter space is a perturbation.

Step 4. Substitute Step 3 (and its conjugate) into Step 2 — which is equation (5). No basis-phase derivative remains, and the $(E_m-E_n)^{-2}$ divergence at level crossings is explicit. Verified: equation (5) on a randomly generated two-level $H(\mathbf{R})$ agrees with an independent gauge-invariant plaquette computation to 9.5×10⁻¹² relative error (verification script).

The spin-½ monopole and the Chern integer

Now the flagship geometry. Take $H = \hat{\mathbf{B}}\cdot\boldsymbol{\sigma}$ (units absorbed); the two levels $\pm|B|$ touch only at $\mathbf{B} = 0$ . The half-angle eigenspinors constructed in Spinors and Two-Component Wavefunctions are $\chi_+ = (\cos\tfrac{\theta}{2},\, e^{i\phi}\sin\tfrac{\theta}{2})^{\mathsf T}$ and $\chi_- = (-e^{-i\phi}\sin\tfrac{\theta}{2},\, \cos\tfrac{\theta}{2})^{\mathsf T}$ . Derivation C computes the connection and finds a startlingly clean field:

\boldsymbol{\Omega}_\pm \;=\; \mp\,\frac{1}{2}\,\frac{\hat{\mathbf{B}}}{B^{2}} \qquad (6)

where: the $\mp$ tracks the upper/lower level. This is the field of a magnetic monopole of charge $\mp\tfrac12$ at the degeneracy; its flux through any closed surface around the origin is $\mp 2\pi$ , surface-independent because the field falls off as $1/B^2$ exactly. Dividing by $2\pi$ defines the Chern number

C \;=\; \frac{1}{2\pi}\oint_{\text{closed}}\boldsymbol{\Omega}_n\cdot d\mathbf{S} \;\in\; \mathbb{Z} \qquad (7)

where: the integral runs over a closed surface (any sphere around $\mathbf{B}=0$ ; in a crystal, the whole Brillouin zone). For the spin-½ upper band, $C = -1$ . The integer is forced, and the cleanest way to see why is that no single smooth gauge covers the whole sphere.

Go deeper · Derivation C · 1 minMonopole curvature and the integer flux for spin one-half

Derivation CMonopole curvature and the integer flux for spin one-half

The structural centerpiece: from the explicit eigenspinors we compute the connection on two gauge patches, extract the curvature, integrate over the sphere to get $-2\pi$ , then show why the answer had to be $2\pi\times$ integer — the Wu–Yang patching argument. Both patches matter: the single-patch connection is singular at one pole, and the way the patches disagree is the topology.

Step 1. On the 'north' patch use $\chi_+^{N} = (\cos\tfrac{\theta}{2},\, e^{i\phi}\sin\tfrac{\theta}{2})^{\mathsf T}$ , smooth everywhere except $\theta = \pi$ (there $e^{i\phi}$ multiplies a non-vanishing component — singular gauge). Compute both components:

A^{N}_\theta = i\,\chi_+^{\dagger}\partial_\theta\chi_+ = 0, \qquad A^{N}_\phi = i\,\chi_+^{\dagger}\partial_\phi\chi_+ = -\sin^{2}\tfrac{\theta}{2} = -\tfrac12(1-\cos\theta).

Step 2. On the 'south' patch use $\chi_+^{S} = e^{-i\phi}\chi_+^{N}$ , smooth everywhere except $\theta = 0$ ; the same computation gives $A^{S}_\phi = +\tfrac12(1+\cos\theta)$ . On the equatorial overlap the connections differ by exactly the gauge shift of the transition function $e^{i\phi}$ : $A^{N}_\phi - A^{S}_\phi = -1 = -\partial_\phi(\phi)$ . Why two patches: a globally smooth gauge would force the total flux to vanish (shrink the loop around either pole). The obstruction to one smooth gauge is the monopole.

Step 3. The curvature is patch-independent. From either patch, $\Omega_{\theta\phi} = \partial_\theta A_\phi - \partial_\phi A_\theta = -\tfrac12\sin\theta$ . Integrate over the whole sphere:

\oint \Omega_{\theta\phi}\, d\theta\, d\phi = \int_0^{2\pi}\!\!\int_0^{\pi} \left(-\tfrac{\sin\theta}{2}\right) d\theta\, d\phi = -2\pi \;\Longrightarrow\; C = -1 .

In vector language this is $\boldsymbol{\Omega}_+ = -\hat{\mathbf{B}}/2B^2$ : flux magnitude $(1/2B^2)\times 4\pi B^2 = 2\pi$ , inward.

Step 4. Why an integer, always: split the surface along the equator. Stokes on each cap (opposite boundary orientations) gives total flux $= \oint_{\text{eq}} (A^{N} - A^{S})\cdot d\mathbf{R} = 2\pi\times$ (winding of the transition function $e^{i\phi}$ ) $= 2\pi\times(-1)$ here. The total flux of any Berry curvature over any closed surface is $2\pi$ times the winding of a phase — an integer by topology, not by calculation. Verified: every line checked symbolically in the verification script (eigenspinors, both connection components, $\Omega_{\theta\phi}=-\sin\theta/2$ , cap and full-sphere fluxes), and confirmed numerically: $C = -1.000000000000$ on an 8×8 plaquette grid.

The integer you cannot break: a numerical experiment

Formula (7) invites a lattice test. Discretize the sphere (or Brillouin zone) into plaquettes; on each, multiply the four corner-state overlaps $\langle u_1|u_2\rangle\langle u_2|u_3\rangle\langle u_3|u_4\rangle\langle u_4|u_1\rangle$ and take minus the complex argument. Each plaquette phase is a tiny gauge-invariant flux (basis phases cancel between adjacent factors), and the Fukui–Hatsugai–Suzuki observation is that their sum over a closed surface is exactly $2\pi C$ while no plaquette phase exceeds $\pi$ . For the spin-½ sphere an $8\times8$ grid returns $C = -1.000000000000$ ; $16\times16$ , the same; a wild smooth rephasing of every state moves nothing. That is 'topological', operationally: the local field shifts with gauge and grid, the total does not. The same algorithm on the two-band model $H(\mathbf{k}) = \sin k_x\,\sigma_x + \sin k_y\,\sigma_y + (m + \cos k_x + \cos k_y)\,\sigma_z$ over the Brillouin-zone torus gives lower-band Chern numbers $0, +1, -1, 0$ at $m = -3, -1, +1, +3$ , jumps pinned to the gap closings at $m = 0, \pm 2$ (Dirac points located in the script). Between closings $C$ cannot move: integers do not flow.

Limiting cases

case	result	meaning
real Hamiltonian (time-reversal symmetric, spinless)	$\boldsymbol{\Omega}_n = 0$ pointwise; $\gamma_n \in \{0, \pi\}$	real eigenvectors gauge the connection away; only a sign flip survives (conical intersections)
spin- $j$ , level $m$ in a rotated field	$\gamma_m = -m\,\Omega_{\text{solid}}$ ; curvature $-m\hat{\mathbf{B}}/B^2$	monopole charge scales with the magnetic quantum number
far from all degeneracies	$\boldsymbol{\Omega}_n \sim \langle\nabla H\rangle^2 / \Delta^2$ , small	large gaps suppress curvature quadratically
approaching a degeneracy $E_m \to E_n$	$\boldsymbol{\Omega}_n \to \infty$ as $(E_m - E_n)^{-2}$	the crossing is a flux source; non-Abelian (Wilczek–Zee) description takes over
closed-surface integral, any $H$	$\oint \boldsymbol{\Omega}_n \cdot d\mathbf{S} = 2\pi C$ , $C \in \mathbb{Z}$	totals are topological: immune to deformation, discretization, disorder

Worked examples

Example 1 — cone loop by two routes. A spin-½ rides a field tracing a cone of half-angle $\theta_0 = 60°$ . Find the upper state's geometric phase from the curvature flux; check the line-integral route.

Step 1. Solid angle of the cone: $\Omega_{\text{solid}} = 2\pi(1 - \cos 60°) = \pi$ steradians.

Step 2. Flux route (equation 4 with 6): $\gamma_+ = \int_0^{2\pi}\!\!\int_0^{\pi/3}(-\tfrac12\sin\theta)\,d\theta\,d\phi = -\pi(1-\cos 60°) = \boldsymbol{-\pi/2}$ — confirmed by quadrature in the script ( $-1.5707963\ldots$ ).

Step 3. Line-integral route: $\gamma_+ = 2\pi A_\phi(\theta_0) = -\pi/2$ . The routes agree — Stokes at work — reproducing the $-\tfrac12\Omega_{\text{solid}}$ law quoted in Geometric (Berry) Phase.

Step 4. Limits: $\theta_0 \to 0$ gives zero; $\theta_0 = 90°$ (great circle) gives $\gamma_+ = -\pi$ — a sign flip, the Berry-phase face of spinor $4\pi$ periodicity.

Example 2 — reading a topological phase diagram. For $\mathbf{d}(\mathbf{k}) = (\sin k_x,\ \sin k_y,\ m + \cos k_x + \cos k_y)$ , find where the lower-band Chern number can change, and evaluate it per region.

Step 1. $C$ can jump only where $|\mathbf{d}| = 0$ . The in-plane components vanish at $\mathbf{k} \in \{(0,0), (\pi,0), (0,\pi), (\pi,\pi)\}$ ; $d_z = 0$ there gives $m = -2$ (at $(0,0)$ ), $m = 0$ (at $(\pi,0)$ , $(0,\pi)$ ), $m = +2$ (at $(\pi,\pi)$ ).

Step 2. Compute $C$ once per region with the plaquette algorithm: $C(m=-3) = 0$ , $C(m=-1) = +1$ , $C(m=+1) = -1$ , $C(m=+3) = 0$ . The phase diagram is $0 \to +1 \to -1 \to 0$ with walls exactly at the gap closings.

Step 3. TKNN reading: were this a 2D insulator with the lower band filled, $\sigma_{xy} = C\,e^2/h = \pm 3.874\times10^{-5}\ \mathrm{S}$ in the topological regions, zero outside.

Step 4. Note what was not needed: no gauge-fixing, no fine grid, no transport simulation. Counting the flux of $\boldsymbol{\Omega}$ is a bookkeeping of integers.

Origins · 1 minBerry Curvature and how this was discovered

Berry Curvature

1931–1984 · Cambridge, Seattle, Bristol

The curvature came before the phase. In 1931 Dirac, asking whether a magnetic monopole could coexist with quantum mechanics, found that a wavefunction around one cannot be smoothly phased everywhere — and the obstruction forces charge to quantize. Half a century later the same mathematics resurfaced with no monopole in sight: in 1982 Thouless, Kohmoto, Nightingale and den Nijs found the Hall conductance of a 2D electron gas equal to $e^2/h$ times an integer that was secretly the total flux of a curvature over the Brillouin zone. When Berry isolated the adiabatic phase in 1984, Barry Simon had already recognized (1983, published first) the loop phase as the holonomy of this connection and the TKNN integer as its Chern class. The article's object is thus a meeting point — Dirac's quantization, TKNN's integer, and Berry's phase are one field seen three ways. The discoverer's biography belongs to the phase's own article; the curvature's story is that of an integer that refused to be anything else.

History

1931

— Dirac shows a quantum theory with magnetic monopoles forces charge quantization — the first Chern number in physics

1982

— Thouless, Kohmoto, Nightingale and den Nijs (TKNN) express the quantized Hall conductance as a Brillouin-zone integral equal to an integer

1983

— Barry Simon identifies the adiabatic phase as the holonomy of a line bundle and the TKNN integer as its Chern number

1984

— Berry's paper on quantal phase factors gives the connection and curvature their physical names and the spin-½ monopole example

2005

— Fukui, Hatsugai and Suzuki publish the plaquette algorithm that returns exact Chern integers on coarse discretized grids

2014

— cold-atom experiments realize topological band models and measure Berry curvature effects directly in the lab

Going further

References. Berry, Proc. R. Soc. Lond. A 392, 45 (1984) · Griffiths & Schroeter, 3rd ed., §11.5 · Sakurai & Napolitano, 2nd ed., §5.6 · TKNN, Phys. Rev. Lett. 49, 405 (1982) · Simon, Phys. Rev. Lett. 51, 2167 (1983) · Fukui, Hatsugai & Suzuki, J. Phys. Soc. Jpn. 74, 1674 (2005) · Xiao, Chang & Niu, Rev. Mod. Phys. 82, 1959 (2010) · Wilczek & Shapere (eds.), Geometric Phases in Physics (1989)

Provenance. Definitions of $\mathbf{A}_n=i\langle n|\nabla n\rangle$ , curvature $\mathbf{\Omega}_n=\nabla\times\mathbf{A}_n$ , and the spin-½ result $\gamma_\pm=\mp\tfrac12\Omega_{\text{solid}}$ are standard (Berry, Proc. R. Soc. A 392, 1984; Griffiths §11.5; Sakurai §5.6). A verifier should reconfirm: (1) the reality of $\mathbf{A}_n$ from $\nabla\langle n|n\rangle=0$ ; (2) the spin-½ connection $A_\phi=-\sin^2(\theta/2)$ from the explicit eigenspinor; (3) the sum-over-states curvature via the off-diagonal Hellmann–Feynman relation $\langle m|\nabla n\rangle=\langle m|\nabla H|n\rangle/(E_n-E_m)$ ; (4) the Chern-number/Hall-conductance relation $\sigma_{xy}=Ce^2/h$ (TKNN, PRL 49, 1982). All equations cross-checked against Xiao–Chang–Niu, Rev. Mod. Phys. 82, 1959 (2010). No non-standard numerical values are asserted.

Builds on: The Interaction Picture and Time-Dependent Hamiltonians · Time-Dependent Perturbation Theory: General Formalism · First-Order Transition Amplitudes and Probabilities Leads to: Quantum Hall Effect and Chern Numbers · Born–Oppenheimer Approximation and Molecular Dynamics · Aharonov–Bohm Effect

Is the Berry connection a real physical field like the vector potential, or just notation?

It has exactly the status of the electromagnetic vector potential. Neither is measurable at a point; both produce measurable loop integrals (Aharonov–Bohm phase, Berry phase) and measurable field strengths (magnetic field, Berry curvature). ' $\mathbf{A}$ is physical only through $\oint\mathbf{A}\cdot d\mathbf{l}$ and $\mathbf{B}$ ' — the same sentence holds for $\mathbf{A}_n$ , $\gamma_n$ , $\boldsymbol{\Omega}_n$ .

Why does the curvature involve other eigenstates when the adiabatic theorem says the system never leaves state n?

Adiabatic transport keeps the population in level $n$ , but $|n(\mathbf{R})\rangle$ itself changes as $\mathbf{R}$ moves, and its rate of change points into the neighbors, with first-order weights $\langle m|\nabla H|n\rangle/(E_n-E_m)$ . No probability flows to level $m$ ; the geometry of how $|n\rangle$ tilts through Hilbert space is encoded in those overlaps, and the curvature is assembled from them.

What exactly is a Chern number, in one sentence?

The total Berry-curvature flux through a closed surface, divided by $2\pi$ — always an integer, equal to the winding of the gauge-patch transition function; physically, the enclosed monopole charge (spin-½ sphere: $-1$ upper band) or the Hall conductance in units of $e^2/h$ (Brillouin zone).

Why is the curvature zero for real Hamiltonians?

If $H(\mathbf{R})$ is real symmetric, its eigenvectors can be chosen real, making $\langle n|\nabla n\rangle$ real; the reality lemma says it is also purely imaginary — so it vanishes, and with it the curvature in that gauge. What survives is a discrete obstruction: around a conical intersection a real eigenvector flips sign, giving $\gamma = \pi$ . Complex Hamiltonians (magnetic fields, spin–orbit coupling) generate continuous curvature.

How does the plaquette (FHS) method get an exact integer from a coarse grid — surely discretization has errors?

Individual plaquette phases do carry $O(h^2)$ errors as local-curvature estimates. But their sum is exactly $2\pi\times$ (integer) by construction: each overlap's arbitrary phase cancels between the two plaquettes sharing an edge, so the total is a winding — a topological count. Discretization could only shift the answer by whole units of $2\pi$ , which requires a plaquette phase to slip past $\pi$ ; with any reasonable grid it cannot. The script shows the same $-1.000000000000$ on 8×8 and 16×16, before and after a violent gauge change.

What changes when two levels are exactly degenerate over a whole region?

The single-level connection is undefined (which state do you follow?); the right object is the matrix-valued, non-Abelian Wilczek–Zee connection $A_{ab} = i\langle a|\nabla b\rangle$ over the subspace. Loop transport then produces a unitary matrix rather than a phase; this article's Abelian story is the one-dimensional special case.