Holonomy in Cyclic Quantum Transport

Holonomy is the transformation left after a state, vector, or internal degree of freedom is transported around a closed loop. In quantum mechanics, the most common holonomy is a phase:

$$\lvert\psi\rangle \longmapsto e^{i\gamma}\lvert\psi\rangle.$$

The central example in this chapter is Berry phase. When a nondegenerate eigenstate is transported adiabatically around a closed loop $C$ in parameter space, the geometric part of the phase is the holonomy of the Berry connection:

$$e^{i\gamma_n[C]} = \exp \left( i\oint_C A_n \right), \qquad A_n=i\langle n(R)|d n(R)\rangle.$$

The displayed integral uses a periodic eigenvector gauge. For a nonperiodic representative, add $\arg\langle n(0)|n(T)\rangle$; Berry phase gives that endpoint formula. Holonomy in the mathematical toolkit develops the general transport construction.

Closed-Loop Transport

Along an open path, transport compares different fibers: the state line or eigenspace at the initial parameter value with the one at the final parameter value. Along a closed loop, the starting and ending base point are the same, so the result can be compared directly with the identity.

For a nondegenerate adiabatic eigenstate, the transported line returns to itself:

$$R(T)=R(0), \qquad \mathbb C\lvert n(T)\rangle = \mathbb C\lvert n(0)\rangle.$$

The vector representative can still return with a phase. After removing the dynamical phase,

$$\lvert n(0)\rangle \longmapsto e^{i\gamma_n[C]} \lvert n(0)\rangle.$$

That phase factor is the Berry holonomy.

Berry Phase as Holonomy

The Berry connection is the local one-form

$$A_n = i\langle n(R)|d n(R)\rangle.$$

Pull it back to a time-parametrized loop $R(t)$:

$$A_t(t) = i\left\langle n(R(t)) \middle| \frac{d}{dt}n(R(t)) \right\rangle.$$

Then the Berry phase is

$$\gamma_n[C] = \int_0^T A_t(t)\,dt = \oint_C A_n.$$

The total adiabatic phase is

$$\alpha_n = -\frac{1}{\hbar} \int_0^T E_n(t)\,dt + \gamma_n[C].$$

Holonomy names the second term’s geometric origin. The first term records energy over time; the second records closed-loop transport of the eigenline.

Gauge Invariance of the Phase Factor

The local eigenvector may be rephased:

$$\lvert n(R)\rangle \mapsto e^{i\chi(R)} \lvert n(R)\rangle.$$

With the convention used here,

$$A_n \mapsto A_n-d\chi.$$

Thus

$$\gamma_n[C] \mapsto \gamma_n[C] - \oint_C d\chi.$$

For a smooth single-valued real $\chi$, $\oint_Cd\chi=0$. If only $e^{i\chi}$ is single valued, this integral can be $2\pi k$. More generally the phase itself is defined modulo $2\pi$, while the phase factor $e^{i\gamma_n[C]}$ is the invariant object. This is why a Berry connection can be gauge dependent while its closed-loop holonomy is physically meaningful.

Curvature and Holonomy

Berry curvature is the local field

$$F_n=dA_n.$$

If the loop $C$ bounds an oriented surface $\Sigma$ and a smooth gauge is available on that surface, Stokes’ theorem gives

$$\gamma_n[C] = \int_\Sigma F_n \quad \operatorname{mod}2\pi.$$

This is the curvature-flux formula for Berry phase. It is useful, but it has assumptions. If the surface passes through a degeneracy, if no single smooth gauge covers it, or if the accessible parameter space has holes, the holonomy may be well-defined even though the naive one-patch flux calculation fails.

The practical rule is:

• small contractible loops measure local curvature;
• large or noncontractible loops can also measure global topology and gauge patching;
• a flat local connection can still have nontrivial holonomy on a space with noncontractible loops.

Parallel-Transport Gauge

Along an open path, one can choose a Parallel Transport phase convention in which

$$A_t(t) = i\langle n(t)|\dot n(t)\rangle = 0.$$

This is a parallel-transport gauge: the eigenvector representative is chosen so that it has no local phase rotation along the path. For a closed loop, this gauge may not return to the original vector phase. Instead,

$$\lvert n(T)\rangle_{\mathrm{pt}} = e^{i\gamma_n[C]} \lvert n(0)\rangle_{\mathrm{pt}}.$$

The holonomy is precisely the obstruction to choosing a single phase convention that is locally parallel and globally returns unchanged.

Sphere Analogy

Parallel transport of tangent vectors on a sphere is the classical picture to keep nearby. A vector carried around a closed curve on a curved surface can return rotated even though it was kept locally parallel at every point.

Berry holonomy is analogous, but the transported object is not usually a tangent vector in physical space. It is a phase line or eigenspace over parameter space. The loop might be a loop of magnetic-field directions, molecular coordinates, crystal momenta, or other control parameters.

The analogy is structural:

$$\text{closed-loop transport} \quad\longrightarrow\quad \text{final mismatch}.$$

The type of mismatch depends on the connection: a rotation for tangent vectors, a phase for a $U(1)$ quantum eigenline, and a matrix for a degenerate eigenspace.

Spin Example

For a spin-$1/2$ adiabatically following a magnetic-field direction, the parameter space at fixed field magnitude is a sphere. In the convention used in Berry Phase for Spin-1/2, the aligned state has

$$\gamma_+[C] = -\frac{\Omega[C]}{2},$$

where $\Omega[C]$ is the oriented solid angle enclosed by the loop of field directions.

The corresponding holonomy is

$$e^{i\gamma_+[C]} = \exp \left( -\frac{i\Omega[C]}{2} \right).$$

This is not a statement that the spin is literally being carried along the surface like a tangent arrow. It is a statement that the eigenline over the field-direction sphere has a Berry connection whose closed-loop holonomy is controlled by solid angle.

Aharonov–Bohm Holonomy

The Aharonov–Bohm phase is a close cousin of Berry holonomy, but the base space and connection are different. A charged particle encircling magnetic flux $\Phi$ acquires

$$\exp \left( \frac{iq}{\hbar} \oint_C \mathbf A\cdot d\mathbf r \right) = \exp \left( \frac{iq\Phi}{\hbar} \right)$$

in the common minimal-coupling convention. Here $\mathbf A$ is the electromagnetic vector potential over physical configuration space, not the Berry connection over Hamiltonian parameter space.

The family resemblance is important: both are loop phases of a connection. The distinction is equally important: Berry phase comes from adiabatic transport of eigenstates, while the Aharonov–Bohm phase comes from electromagnetic gauge holonomy around inaccessible flux. See Aharonov–Bohm Effect for the physical setup.

Nonabelian Holonomy

If an exactly degenerate eigenspace of dimension $r>1$ remains isolated from other energies, adiabatic transport can rotate states within it. Its common dynamical phase can be factored out. Choose an orthonormal local basis $\lvert n,a(R)\rangle$, $a=1,\ldots,r$, and define the matrix-valued Berry connection

$$\mathcal A_{ab} = i\langle n,a(R)|d n,b(R)\rangle.$$

The holonomy around a loop is a unitary matrix

$$\mathcal U_C = \mathcal P \exp \left( i\oint_C \mathcal A \right),$$

with path ordering $\mathcal P$ because matrices at different points need not commute. This formula uses a basis that matches at the loop endpoints; otherwise an endpoint basis-matching matrix is also required. Under a change of basis inside the degenerate subspace, $\mathcal U_C$ changes by conjugation at the base point. Its eigenvalues and trace-type data are the gauge-invariant content.

This is the Wilczek–Zee or nonabelian Berry phase setting; see Non-Abelian Berry Phase Preview.

What Holonomy Is Not

Holonomy is not just “phase caused by going in a circle.” The loop must be accompanied by a connection that tells what it means to transport the state. A cyclic time-dependent Hamiltonian can produce ordinary dynamical phase, nonadiabatic transitions, or resonant rotations without being well described by Berry holonomy.

Holonomy is not automatically topological. A spin Berry phase changes continuously with the loop’s solid angle. Flatness makes holonomy insensitive to deformations within a fixed homotopy class, but the value can still vary between different flat connections. Quantization requires additional structure, such as a symmetry constraint; a Chern number instead quantizes a curvature integral over a closed surface. Neither a gap nor noncontractibility alone quantizes a particular loop phase.

Common Mistakes

• Calling any cyclic phase a holonomy without specifying the connection and transported object.
• Confusing the Berry connection with the electromagnetic vector potential.
• Assuming a Berry phase is always topological.
• Applying the curvature-flux formula across a degeneracy or gauge singularity.
• Treating the phase $\gamma$ rather than the phase factor $e^{i\gamma}$ as absolutely defined.
• Forgetting that nonabelian holonomy matrices are basis covariant, not individually gauge invariant.
• Dropping the dynamical phase when predicting a measured total phase.

Exercises

1. Show that $e^{i\gamma[C]}$ is unchanged under $A\mapsto A-d\chi$ for a smooth single-valued $\chi$ around a closed loop.

Solution

The transformed phase is

$$\gamma'[C] = \oint_C(A-d\chi) = \gamma[C]-\oint_Cd\chi.$$

For a closed loop and single-valued $\chi$,

$$\oint_Cd\chi = \chi(\text{end})-\chi(\text{start}) = 0.$$

Therefore $\gamma'[C]=\gamma[C]$, and the phase factor is unchanged. If the phase changes by $2\pi k$ under a large convention change, $e^{i\gamma}$ is still unchanged.

2. A spin-$1/2$ aligned eigenstate follows a loop enclosing oriented solid angle $\Omega=\pi$. In the convention $\gamma=-\Omega/2$, what is the holonomy phase factor?

Solution

The Berry phase is

$$\gamma = -\frac{\pi}{2}.$$

The phase factor is

$$e^{i\gamma} = e^{-i\pi/2} = -i.$$

3. A charged particle of charge $q$ encircles flux $\Phi=h/q$ once. What is the Aharonov–Bohm phase factor in the convention above?

Solution

The phase is

$$\Delta\varphi = \frac{q\Phi}{\hbar} = \frac{q(h/q)}{\hbar} = \frac{h}{\hbar} = 2\pi.$$

Therefore the phase factor is

$$e^{i\Delta\varphi} = e^{i2\pi} = 1.$$

4. Why does $F=0$ not always imply trivial holonomy?

Solution

Curvature is local. On a space with noncontractible loops, a flat connection can still have a nonzero integral around a loop. For example, on a circle,

$$A=\alpha\,d\theta$$

has

$$F=dA=0,$$

but

$$\oint A = \int_0^{2\pi}\alpha\,d\theta = 2\pi\alpha.$$

The holonomy is $e^{i2\pi\alpha}$, which need not be $1$.

5. In the nonabelian case, why is path ordering needed?

Solution

The connection is matrix-valued. Matrices evaluated at different points along the loop need not commute:

$$[\mathcal A(t_1),\mathcal A(t_2)] \ne 0.$$

The ordered exponential records the sequence in which infinitesimal transports are applied. Without path ordering, the exponential would treat noncommuting factors as though their order did not matter.

References

• Y. Aharonov and D. Bohm, “Significance of electromagnetic potentials in the quantum theory,” Physical Review 115, 485–491, 1959.
• M. V. Berry, “Quantal phase factors accompanying adiabatic changes,” Proceedings of the Royal Society A 392, 45–57, 1984.
• M. Nakahara, Geometry, Topology and Physics, 2nd ed., CRC Press, 2003.
• A. Shapere and F. Wilczek, eds., Geometric Phases in Physics, World Scientific, 1989.
• B. Simon, “Holonomy, the quantum adiabatic theorem, and Berry’s phase,” Physical Review Letters 51, 2167–2170, 1983.
• F. Wilczek and A. Zee, “Appearance of gauge structure in simple dynamical systems,” Physical Review Letters 52, 2111–2114, 1984.