Aharonov–Anandan Phase and Non-Adiabatic Generalizations

The article Geometric (Berry) Phase attaches its phase to a slow round trip of external parameters. That framing invites a suspicion: is the geometry an artifact of adiabaticity? In 1987 Aharonov and Anandan answered with a clean no. Take any evolution — fast, driven, wildly non-adiabatic — that is cyclic: the physical state returns to itself, $|\psi(T)\rangle = e^{i\phi}|\psi(0)\rangle$ . Subtract from the total phase $\phi$ the dynamical bookkeeping $-\tfrac{1}{\hbar}\int_0^T\langle H\rangle\,dt$ , and what remains is a geometric phase $\beta$ that depends on exactly one thing: the closed curve the state's ray traces in projective Hilbert space. Not the speed, not the Hamiltonian, not any parameter loop — the curve. This article builds that statement and stress-tests it by machine. We derive the phase decomposition and prove $\beta = \oint i\langle\tilde\psi|d\tilde\psi\rangle$ is gauge- and reparametrization-invariant; we prove the Hamiltonian-independence theorem by driving the same Bloch-sphere circle with two entirely different Hamiltonians — a static field and a rotating field — and finding identical $\beta$ ( $4\times10^{-16}$ apart) while the dynamical phases differ by radians. For spin-½ we solve the rotating-field problem exactly and watch a cancellation delete every trace of the sweep rate, leaving $\beta = -\tfrac12\Omega$ with $\Omega$ the solid angle of the circle the spin actually traces — verified for a wobbly non-circular loop to $4\times10^{-13}$ . In the slow-drive limit the actual path snaps onto the parameter path and Berry's phase re-emerges as a special case, with the convergence measured.

Key takeaways

A cyclic evolution returns the ray to itself: $|\psi(T)\rangle = e^{i\phi}|\psi(0)\rangle$ . No adiabaticity is required — only closure of the loop in ray space.
The total phase splits uniquely: $\phi = \phi_{\text{dyn}} + \beta$ , with $\phi_{\text{dyn}} = -\tfrac{1}{\hbar}\int_0^T \langle\psi|H|\psi\rangle\,dt$ and $\beta$ the Aharonov–Anandan (AA) geometric phase.
$\beta = \oint i\langle\tilde\psi|d\tilde\psi\rangle$ over the closed curve in projective Hilbert space — invariant under gauge choices, reparametrization, and even changes of the Hamiltonian that preserve the curve.
Spin-½: $\beta = -\tfrac12\tilde\Omega$ , half the solid angle of the actual Bloch-sphere trajectory — exact at any speed, not just adiabatically.
Berry's phase is the adiabatic limit: as the drive slows, the actual path converges to the parameter path and $\beta \to \gamma_{\text{Berry}}$ (deviation measured linear in sweep rate).
Key idea: strip the unphysical overall phase and watch the ray — whenever it closes a loop, the leftover phase after removing the dynamical part is pure geometry of that loop.

The core insight

Strip the unphysical overall phase from the state and watch the ray. Whenever the ray returns to where it started — a cyclic evolution — the leftover phase, after removing the dynamical phase $-\tfrac1\hbar\int\langle H\rangle\,dt$ , is a geometric quantity fixed entirely by the loop in projective Hilbert space: no adiabaticity, no parameters, no Hamiltonian details.

Where this lives today

The AA formulation is the working language wherever geometric phases are engineered rather than merely observed. Non-adiabatic holonomic quantum gates — realized in trapped ions, NV centers and superconducting circuits — deliberately drive qubits around fast closed Bloch-sphere loops so that the gate angle is a solid angle: rate-independent by Derivation B, hence insensitive to timing noise that would scramble a dynamical phase, and fast enough to beat decoherence precisely because adiabaticity is not required. The design rule is this article's Step 3 cancellation, run in reverse: pick the loop for the phase you want, then pick any convenient Hamiltonian that drives it. The same phase is routine in interferometry. NMR experiments read $\beta$ by spin-echo sequences that cancel $\phi_{\text{dyn}}$ ; polarization optics implements Pancharatnam's version in geometric-phase lenses and metasurface holograms, where a spatially varying loop on the Poincaré sphere prints a phase profile on a beam with no optical path difference at all. In all of these the operational content is the one proved above: the phase survives anything that preserves the loop.

The physical picture

Berry's construction needs a supporting cast: a Hamiltonian $H(\mathbf{R})$ , a slowly closed parameter loop, an eigenstate faithfully tracking it. Aharonov and Anandan noticed that the phase never cared about the cast — only about the trajectory the state itself performs. The natural stage is ray space: the set of physical states, with overall phase discarded. For a two-level system, ray space is exactly the Bloch sphere. Any evolution whose Bloch vector comes back to its starting point is cyclic, whether it got there by gentle steering or violent driving. Picture a spin in a magnetic field whose direction sweeps a cone of half-angle $\theta$ at rate $\omega$ . Adiabatically ( $\omega \to 0$ ) the spin hugs the field and traces that same cone. At finite $\omega$ it cannot keep up: the exact cyclic state traces a different cone, of half-angle $\tilde\theta$ , tilted away from the field's. The AA phase reads the solid angle of the cone the spin actually traces — $\beta = -\tfrac12\Omega(\tilde\theta)$ — and the Berry value $-\tfrac12\Omega(\theta)$ is recovered only in the limit $\tilde\theta \to \theta$ .

Figure

The loop that matters: the actual Bloch path, not the parameter path

The deepest property is what the phase ignores. Two Hamiltonians that push a state around the same ray-space loop — however different their energies, matrix elements, or time profiles — imprint the same $\beta$ . The punchline figure shows the numerical experiment: a static field and a rotating field drive the identical equatorial circle; their dynamical phases differ by $5.44$ rad, their geometric phases agree to $4\times10^{-16}$ . On the right, the measured convergence of $\beta$ to Berry's value as the drive slows.

Figure

Punchline: same loop, different Hamiltonians, same phase — and the measured Berry limit

The phase anatomy of a cyclic evolution

Fix a solution of the time-dependent Schrödinger equation over one period. Cyclic means the physical state closes, which constrains the vector only up to phase. That single phase $\phi$ is the raw observable-in-waiting:

|\psi(T)\rangle \;=\; e^{i\phi}\,|\psi(0)\rangle \qquad (1)

where: $T$ — the return time of the ray, not of any external dial; $\phi$ — the total phase, defined mod $2\pi$ . A stationary state satisfies (1) trivially with $\phi = -E T/\hbar$ (the standing phase factor of Stationary States and Energy Eigenstates); the interesting cases are loops that genuinely move. Part of $\phi$ is mere clockwork: every instant, the state accrues phase at rate $-\langle H\rangle/\hbar$ , exactly as an energy eigenstate would at rate $-E/\hbar$ . Integrating that rate defines the dynamical phase,

\phi_{\text{dyn}} \;=\; -\,\frac{1}{\hbar}\int_0^T \langle\psi(t)|\,H(t)\,|\psi(t)\rangle\,dt \qquad (2)

where: the expectation value is taken in the actual evolving state — for non-adiabatic motion this is not $-E_nT/\hbar$ , because $|\psi\rangle$ need not be an instantaneous eigenstate. Whatever (2) fails to explain is the Aharonov–Anandan phase, $\beta = \phi - \phi_{\text{dyn}}$ , and the derivation below shows $\beta$ is a property of the closed ray-space curve alone.

Derivation ASplitting the total phase into dynamical and geometric parts

We are deriving the AA decomposition: for any cyclic evolution, $\phi - \phi_{\text{dyn}}$ equals a loop integral $\oint i\langle\tilde\psi|d\tilde\psi\rangle$ built from a single-valued lift $|\tilde\psi\rangle$ of the ray curve, and that integral is unchanged by rephasing the lift or reparametrizing time. The strategy follows Aharonov–Anandan's 1987 paper: peel the offending phase off by hand, then let the Schrödinger equation tell us what the peeled state's phase velocity is. Watch the energy integral cancel — that cancellation is the whole theorem.

Step 1. Define the lift $|\tilde\psi(t)\rangle = e^{-i f(t)}\,|\psi(t)\rangle$ with any smooth $f$ obeying $f(T) - f(0) = \phi$ . Then $|\tilde\psi(T)\rangle = e^{-if(T)}e^{i\phi}|\psi(0)\rangle = |\tilde\psi(0)\rangle$ : the lift is single-valued around the loop — a closed curve in Hilbert space projecting onto the closed ray curve.

Step 2. Differentiate the lift and use the Schrödinger equation $i\hbar\,\partial_t|\psi\rangle = H|\psi\rangle$ :

i\langle\tilde\psi|\partial_t\tilde\psi\rangle \;=\; \dot f \;+\; \frac{1}{\hbar}\langle\psi|H|\psi\rangle .

Why this is the pivot: the left side is a purely kinematic object (a connection evaluated along the curve, the same $i\langle\cdot|d\cdot\rangle$ structure as the Berry connection); the right side splits it into the phase bookkeeping $\dot f$ and the energy rate. No adiabatic assumption was available to be used.

Step 3. Integrate over the period. $\int_0^T \dot f\,dt = \phi$ and the energy term integrates to $-\phi_{\text{dyn}}$ by definition (2):

\beta \;\equiv\; \oint i\langle\tilde\psi|\,d\tilde\psi\rangle \;=\; \phi - \phi_{\text{dyn}} .

Why the energy integral cancels out of the geometry: it has been subtracted explicitly — what remains is the integral of a one-form along the closed lift, with the Hamiltonian appearing nowhere.

Step 4. Invariance. Rephasing the lift, $|\tilde\psi\rangle \to e^{i\chi(t)}|\tilde\psi\rangle$ with $\chi$ single-valued around the loop, shifts the integral by $-\oint d\chi = 0$ . Reparametrizing time changes $dt$ but not the differential $d\tilde\psi$ — the final expression contains no clock. So $\beta$ is fixed by the closed curve in ray space together with its orientation, mod $2\pi$ . Verified: the decomposition is checked end-to-end in the article's verification script — an independent TDSE integration of the rotating-field problem reproduces the exact total phase to $2\times10^{-13}$ with return overlap $|\langle\chi|\psi(T)\rangle| = 1$ , and $\phi - \phi_{\text{dyn}}$ lands on the geometric prediction to $10^{-9}$ .

The phase belongs to the curve, not the Hamiltonian

Derivation A leaves a loose thread: many Hamiltonians drive a state around the same ray-space loop. If $\beta$ deserves the name geometric, all of them must produce the same value. They do, and the proof is two lines.

Go deeper · Derivation B · 1 minHamiltonian-independence of the geometric phase

Derivation BHamiltonian-independence of the geometric phase

We are proving that two different evolutions tracing the same closed curve in ray space, in the same orientation, carry identical $\beta$ — the theorem that divorces the phase from dynamics. The idea: two state histories with the same rays differ only by a time-dependent phase, and Derivation A already showed such phases drop out of the loop integral. We then confirm by machine with the least similar pair of drives we can build.

Step 1. Let $|\psi_1(t)\rangle$ and $|\psi_2(t)\rangle$ be solutions (of possibly different Hamiltonians $H_1$ , $H_2$ ) whose rays coincide at every reparametrized instant. Equal rays means $|\psi_2\rangle = e^{i\alpha(t)}|\psi_1\rangle$ for some smooth $\alpha(t)$ .

Step 2. Their single-valued lifts (Step 1 of Derivation A) then differ by a phase $e^{i\chi(t)}$ that is itself single-valued around the loop (both lifts close). By Derivation A Step 4, the loop integrals agree: $\beta_2 = \beta_1$ . The total and dynamical phases can differ arbitrarily — only their difference is pinned.

Step 3. Machine exhibit (the punchline figure). Drive A: static $H_A = \tfrac{\hbar\omega}{2}\sigma_z$ precesses a state tilted at $90°$ around the equator; since $\langle\sigma_z\rangle = 0$ on the equator, $\phi_{\text{dyn}}^A = 0$ — the total phase is the geometric phase. Drive B: the rotating field of the next section with $\theta = 60°$ , $\omega = 0.5\,\omega_0$ , whose exact cyclic state traces the same equator with $\phi_{\text{dyn}}^B = -5.44$ rad. Computed: $\beta_A = \beta_B = -\pi$ with difference $4\times10^{-16}$ . Verified: the twin-drive test in the verification script confirms equal circles ( $\tilde\theta$ equal to $10^{-12}$ ), equal geometric phases to $4\times10^{-16}$ , and dynamical phases differing by $5.44$ rad.

The right arena for all of this is projective Hilbert space $\mathbb{CP}^{N-1}$ : states modulo phase. For a two-level system that space is the Bloch sphere — the half-angle dictionary of Spinors and Two-Component Wavefunctions maps every ray to a unit vector, and cyclic evolutions to closed circuits. The AA phase is a functional on closed circuits; the Berry connection and curvature of the neighboring article organize the same geometry field-theoretically, and Berry Connection and Berry Curvature takes that formal story up.

Spin-½ exactly: the solid angle of the path actually traced

Now the flagship solvable case. Let a field of fixed magnitude sweep a cone: $H(t) = \tfrac{\hbar\omega_0}{2}\,\hat{\mathbf{B}}(t)\cdot\boldsymbol\sigma$ with $\hat{\mathbf{B}} = (\sin\theta\cos\omega t,\ \sin\theta\sin\omega t,\ \cos\theta)$ . The rotating-frame construction that solves the driven two-level problem exactly (the machinery of Transition Probability for a Periodic Two-Level System) turns this into a static problem: in the frame rotating at $\omega$ about $z$ ,

H_{\text{eff}} \;=\; \frac{\hbar}{2}\Big[\,\omega_0\sin\theta\,\sigma_x \;+\; (\omega_0\cos\theta - \omega)\,\sigma_z\Big] \qquad (3)

where: the $-\omega\sigma_z/2$ piece is the frame's fictitious field. The eigenstates of (3) are the cyclic states: dragged back to the lab frame they trace perfect cones about $z$ and return exactly after $T = 2\pi/\omega$ . Their cone half-angle is

\tan\tilde\theta \;=\; \frac{\omega_0\sin\theta}{\omega_0\cos\theta - \omega} \qquad (4)

where: $\tilde\theta$ — the tilt of the actual trajectory. Slow drive ( $\omega \ll \omega_0$ ): $\tilde\theta \to \theta$ , the spin hugs the field. Fast drive: $\tilde\theta \to \pi$ , the state pins to the (inverted) rotation axis. And when $\omega = \omega_0\cos\theta$ the $\sigma_z$ term vanishes: $\tilde\theta = 90°$ — the spin rides the equator, a great circle, even though the field still leans at $\theta$ .

Go deeper · Derivation C · 1 minExact AA phase for a spin in a rotating field

Derivation CExact AA phase for a spin in a rotating field

We are computing $\beta$ for the cyclic states of the rotating-field problem exactly — no adiabatic approximation — and watching every trace of the drive details cancel, leaving $-\tfrac12\times$ (solid angle of the actual cone). This is the calculation the article's claims stand on: the total phase from the exact propagator, the dynamical phase from a time-independent expectation value, and a cancellation of the effective splitting $\lambda$ between them.

Step 1. Total phase. With $a = \omega_0\sin\theta$ , $b = \omega_0\cos\theta - \omega$ , $\lambda = \sqrt{a^2 + b^2}$ , the lab propagator over one period factorizes as $U(T) = e^{-i\omega T\sigma_z/2}\,e^{-iH_{\text{eff}}T/\hbar}$ . Acting on the upper cyclic state $|\chi\rangle$ (eigenvalue $+\hbar\lambda/2$ ), and using $\omega T = 2\pi$ so $e^{-i\pi\sigma_z} = -\mathbb{1}$ :

U(T)|\chi\rangle = -\,e^{-i\lambda T/2}\,|\chi\rangle \;\;\Longrightarrow\;\; \phi = \pi - \frac{\lambda T}{2} \pmod{2\pi}.

Step 2. Dynamical phase. $\langle H\rangle$ is time-independent on a cyclic state (checked to $10^{-12}$ in the script). At $t=0$ the Bloch vector sits at tilt $\tilde\theta$ and the field at $\theta$ , so $\langle H\rangle = \tfrac{\hbar\omega_0}{2}\cos(\theta - \tilde\theta)$ . Expanding with $a = \lambda\sin\tilde\theta$ , $b = \lambda\cos\tilde\theta$ :

\omega_0\cos(\theta - \tilde\theta) = (b+\omega)\cos\tilde\theta + a\sin\tilde\theta = \lambda + \omega\cos\tilde\theta \;\;\Longrightarrow\;\; \phi_{\text{dyn}} = -\frac{T}{2}\big(\lambda + \omega\cos\tilde\theta\big).

Why this expansion: it re-expresses the field direction in terms of the trajectory's own geometry — the step that lets the dynamics cancel.

Step 3. Subtract. The $\lambda T/2$ terms — the only place the effective splitting, hence the drive strength and detuning, appear — cancel identically:

\beta = \phi - \phi_{\text{dyn}} = \pi + \frac{\omega T}{2}\cos\tilde\theta = \pi + \pi\cos\tilde\theta \;\equiv\; -\,\pi\,(1 - \cos\tilde\theta) \pmod{2\pi}.

That is $-\tfrac12\,\Omega(\tilde\theta)$ , half the solid angle $\Omega = 2\pi(1-\cos\tilde\theta)$ of the cone actually traced. The sweep rate $\omega$ survives only inside $\tilde\theta$ — only through the shape of the path, exactly as the general theorem demands.

Step 4. Beyond circles. The law is not about cones. Drive the spin with the parallel-transport Hamiltonian $H = \tfrac{\hbar}{2}(\mathbf{r}\times\dot{\mathbf{r}})\cdot\boldsymbol\sigma$ along a wobbly closed path $\theta(t) = 1.05 + 0.22\sin(3\varphi)$ : then $\langle H\rangle = 0$ identically (the field is perpendicular to the Bloch vector), so the total phase is the geometric phase. The machine returns $\beta = -1.5972843$ , matching $-\tfrac12\oint(1-\cos\theta)\,d\varphi$ to $4\times10^{-13}$ . Verified: solid-angle law confirmed for the cone family at three drive rates ( $10^{-9}$ ), by independent TDSE integration ( $2\times10^{-13}$ on the total phase), and on the non-circular path above ( $4\times10^{-13}$ ); the great-circle case gives $\beta = -\pi$ exactly.

The adiabatic limit: recovering Berry

Send $\omega/\omega_0 \to 0$ in (4): $\tilde\theta \to \theta$ , the actual cone collapses onto the field cone, and $\beta \to -\pi(1-\cos\theta)$ — the solid-angle result carried by Geometric (Berry) Phase for exactly this setup. Berry's phase is thus the adiabatic edge of the AA phase: the special case in which the ray-space loop is inherited from a parameter loop. The convergence is measurable and linear. At $\theta = 60°$ , where the Berry value is $-\pi/2$ , the computed deviations $|\beta - \gamma_{\text{Berry}}|$ are

\omega/\omega_0:\ 0.5 \to 1.571,\quad 0.2 \to 0.542,\quad 0.1 \to 0.254,\quad 0.05 \to 0.122,\quad 0.02 \to 0.048\ \text{rad} \qquad (5)

where: the first entry is the great-circle case — at half the Larmor rate the trajectory is already $30°$ away from the field cone, and the 'correction' is as large as the Berry phase itself. A five-fold slowdown ( $0.1 \to 0.02$ ) shrinks the deviation five-fold: the leading error is first order in $\omega/\omega_0$ . Adiabaticity here is a convergence statement about paths, not a precondition for geometry.

Limiting cases

case	result	meaning
$\omega/\omega_0 \to 0$ (adiabatic)	$\tilde\theta \to \theta$ , $\beta \to -\pi(1-\cos\theta)$	actual path → parameter path; Berry's phase recovered as the slow limit
$\omega = \omega_0\cos\theta$ (great circle)	$\tilde\theta = 90°$ , $\beta = -\pi$	maximal geometric phase; with a static drive this loop has $\phi_{\text{dyn}} = 0$ — the phase is pure geometry
$\omega/\omega_0 \to \infty$ (fast drive)	$\tilde\theta \to \pi$ , $\beta \to -2\pi \equiv 0$	the state pins to the rotation axis; the loop degenerates and the phase trivializes
stationary eigenstate	ray never moves; $\beta = 0$ , $\phi = -ET/\hbar$	the trivial cyclic case — all phase is dynamical clockwork
$\langle H\rangle = 0$ along the path	$\phi = \beta$	parallel-transport drives: the total phase is directly geometric (the wobbly-path exhibit)
non-cyclic evolution	$\beta$ undefined as stated	the ray must close; the Samuel–Bhandari geodesic-closure extension handles open paths (one line, not owned here)

Worked examples

Example 1 — a great-circle spin, end to end. A two-level system has Larmor frequency $\omega_0/2\pi = 100$ MHz. Its field sweeps a cone of half-angle $\theta = 60°$ at the great-circle rate. Find the drive frequency, the period, and both phases of the upper cyclic state.

Step 1. Great-circle condition: $\omega = \omega_0\cos 60° = \omega_0/2$ , so $\omega/2\pi = 50$ MHz and $T = 2\pi/\omega = \mathbf{20\ ns}$ .

Step 2. Geometric phase: $\tilde\theta = 90°$ , solid angle $\Omega = 2\pi$ , so $\beta = -\tfrac12\Omega = \boldsymbol{-\pi}$ — independent of $\omega_0$ , $\omega$ , and $T$ individually (rate-independence checked in the script to $10^{-9}$ ).

Step 3. Dynamical phase: from Derivation C Step 2, $\phi_{\text{dyn}} = -\tfrac{T}{2}(\lambda + \omega\cos\tilde\theta) = -\lambda T/2$ on the great circle. Numerically $|\phi_{\text{dyn}}| = 5.44$ rad — comparable to $|\beta| = \pi$ and carrying all the Hamiltonian details that $\beta$ forgot.

Step 4. Reading: an interferometric comparison of this run against a reference that cancels the $5.44$ rad of clockwork isolates a clean $-\pi$ : the state comes back minus itself geometrically — the AA face of the spinor sign flip.

Example 2 — one field cone, three trajectories. Keep $\theta = 60°$ , $\omega_0 = 1$ (units of the Larmor rate), and drive at $\omega = 0.2,\ 0.5,\ 0.8$ . Find each actual cone and its AA phase.

Step 1. From (4), $\tan\tilde\theta = \sin 60°/(\cos 60° - \omega)$ , giving $\tilde\theta = 70.9°,\ 90.0°,\ 109.1°$ — the trajectory leans further from the axis as the drive speeds up, crossing the equator exactly at $\omega = 0.5$ .

Step 2. Solid angles $\Omega = 2\pi(1-\cos\tilde\theta) = 4.227,\ 6.283,\ 8.340$ sr.

Step 3. Phases $\beta = -\Omega/2 = \mathbf{-2.113,\ -3.142,\ -4.170}$ rad (the last $\equiv +2.113$ mod $2\pi$ — only $e^{i\beta}$ is physical). All three verified against the exact propagator to $10^{-9}$ ; the Berry value $-\pi/2 = -1.571$ is approached only far to the left of this table, per (5).

Origins · 1 minYakir Aharonov and how this was discovered

Yakir Aharonov

1932– · Haifa, Bristol, Columbia (South Carolina), Tel Aviv

Yakir Aharonov has spent a career locating the physics that hides in phases. As a doctoral student with David Bohm in Bristol he showed in 1959 that a charged particle records the vector potential of a solenoid it never touches — the Aharonov–Bohm effect, the archetype every geometric-phase argument leans on (its story is told with Geometric (Berry) Phase). Twenty-eight years later, working with Jeeva Anandan at the University of South Carolina, he pulled Berry's discovery off its adiabatic scaffolding: the 1987 paper reframed the geometric phase as a property of any cyclic evolution in projective Hilbert space, with Berry's phase as its slow limit and the solid-angle rule promoted to the actual trajectory. The pair went further in 1988, showing ray space carries a natural metric that turns quantum evolution itself into geometry. Aharonov's other signature — weak values and weak measurement (1988, with Albert and Vaidman) — extends the same instinct: interrogate the quantum phase gently and it answers with structure no textbook expectation value contains.

History

1956

— Pancharatnam formulates a phase for cyclic changes of light polarization, the optical ancestor of the geometric phase

1959

— Aharonov and Bohm predict the solenoid phase shift; the potential acquires loop-level physical meaning

1984

— Berry publishes the adiabatic geometric phase for slowly cycled Hamiltonians

1987

— Aharonov and Anandan remove adiabaticity: the phase belongs to any closed loop in projective Hilbert space

1988

— Samuel and Bhandari extend the construction to non-cyclic evolutions via geodesic closure

1988

— Suter, Mueller and Pines measure the AA phase directly by NMR interferometry

Going further

References. Aharonov & Anandan, Phys. Rev. Lett. 58, 1593 (1987) · Anandan & Aharonov, Phys. Rev. D 38, 1863 (1988) · Berry, Proc. R. Soc. Lond. A 392, 45 (1984) · Sakurai & Napolitano, Modern Quantum Mechanics, geometric-phase section · Griffiths & Schroeter, 3rd ed., Ch. 11 · Cohen-Tannoudji, Diu & Laloë, geometric-phase complements · Bohm, Mostafazadeh, Koizumi, Niu & Zwanziger, The Geometric Phase in Quantum Systems (Springer, 2003)

Provenance. Phase-decomposition derivation reproduces Aharonov–Anandan, Phys. Rev. Lett. 58, 1593 (1987), Eqs. (3)–(5); spin-1/2 solid-angle result $\beta=-\pi(1-\cos\theta)$ checkable against Sakurai & Napolitano Modern QM geometric-phase section and Griffiths Ch. 11. Verifier should: (1) confirm the cancellation of the energy integral in the lift derivation; (2) check the rotating-frame spin solution and that $\omega$ cancels from $\beta$ ; (3) verify $\Omega=2\pi(1-\cos\theta)$ as the solid angle of a cone of half-angle $\theta$ ; (4) confirm the gauge term $\oint\mathrm{d}\chi=0$ for single-valued $\chi$ . No nonstandard numerical constants are asserted.

Builds on: The Interaction Picture and Time-Dependent Hamiltonians · Dyson Series and the Time-Ordered Evolution Operator · Transition Probability for a Periodic Two-Level System Leads to: Semiclassical Theory of Atom–Radiation Interaction · Holonomic Quantum Computation · Pancharatnam Phase in Optics

Do I need a Hamiltonian at all to define the AA phase?

Only to produce the loop and to define the dynamical phase you subtract. The value of $\beta$ itself is a functional of the closed curve in ray space — Derivation B shows any Hamiltonian driving the same curve gives the same answer. In the extreme case of a parallel-transport drive with $\langle H\rangle = 0$ , nothing is subtracted at all and the total phase is directly geometric. That is why the modern view treats $\beta$ as belonging to the geometry of projective Hilbert space, with the Hamiltonian demoted to a means of transportation.

If the overall phase of a state is unphysical, what exactly is measured?

A relative phase. Split a system into two arms (interferometry, or two spin components), let one undergo the cyclic evolution, recombine, and the fringe shift reads $\phi = \phi_{\text{dyn}} + \beta$ . Isolating $\beta$ is an engineering problem — spin-echo and phase-cycling sequences cancel the dynamical part (that is precisely how the 1988 NMR experiment measured the AA phase). The geometric piece survives such cancellation exactly because it does not care about the energies being echoed away.

The rotating field never stops leaning at 60° — why does the spin trace the equator instead?

Because the cyclic state is an eigenstate of the rotating-frame Hamiltonian (3), and that operator includes the fictitious $-\omega\sigma_z/2$ field of the frame transformation. At $\omega = \omega_0\cos\theta$ the fictitious field exactly cancels the $z$ -component of the real one, leaving an effective field in the equatorial plane; the cyclic state aligns with it, not with the lab field. The lab-frame trajectory is that tilted state carried around $z$ once per period — the equator. The AA phase reads this actual circle, which is why it is $-\pi$ and not the adiabatic $-\pi/2$ .

Is the evolution's period T doing any work in the answer?

None beyond defining the loop. $\beta = \oint i\langle\tilde\psi|d\tilde\psi\rangle$ contains no $dt$ — reparametrize the traversal arbitrarily and the integral is untouched (Derivation A, Step 4). The verification script exhibits this concretely: the great-circle phase is $-\pi$ at any drive rate satisfying the great-circle condition, while the period and the dynamical phase change underneath it.

What happens if the state only approximately returns to its initial ray?

Then $\phi$ in equation (1) is not defined and neither, strictly, is $\beta$ . Two standard responses: close the gap explicitly — follow the evolution with a short geodesic in ray space connecting endpoint to start (the Samuel–Bhandari construction, 1988), which reduces to the AA phase when the gap vanishes; or accept a small systematic error of order the closure defect. In experiments one usually engineers exact cyclicity instead — as with the rotating-frame eigenstates here, which return exactly by construction.

Is the minus sign in β = −Ω/2 a convention?

Half of it. The sign of $\beta$ flips with the orientation of the loop and with which spin state (upper or lower) you transport — the lower cyclic state acquires $+\tilde\Omega/2$ . What is not conventional: the relative sign between the two states, the magnitude $\tfrac12\tilde\Omega$ , and the mod- $2\pi$ identification (Example 2's $-4.170 \equiv +2.113$ ). The verification script pins the stated conventions: upper state, loop traversed with increasing $\varphi$ , $\beta = -\pi(1-\cos\tilde\theta)$ .