The Adiabatic Theorem

The adiabatic theorem is quietly load-bearing across physics. The Born-Oppenheimer approximation — all of quantum chemistry — is the theorem with nuclear coordinates as the slow parameters: electrons ride instantaneous eigenstates while nuclei crawl, and potential-energy surfaces are just $E_n(\mathbf R)$ landscapes. Adiabatic-passage techniques in NMR and atomic physics sweep fields to transfer population with near-unit fidelity, robust against pulse imperfections because following is gap-protected. And adiabatic quantum computation stakes a computing paradigm on the theorem: encode the answer in a complicated Hamiltonian's ground state, start from an easy one, interpolate slowly — runtime governed by the minimum gap, which is why gap engineering, not qubit count, is the field's hard currency. Where protocols must outrun the theorem, "shortcuts to adiabaticity" add counterdiabatic terms that cancel the leak — engineering aimed squarely at the off-diagonal identity derived above.

Think of a compass needle in a field you are rotating. Turn the field slowly compared with the needle's precession and the needle glides along, always pointing with the field — it follows. Jerk the field and the needle is left behind, precessing about the new axis. The quantum spin-½ does exactly this, and the criterion is a race between two clocks: the external sweep clock ($\omega$) and the internal Larmor clock of the gap ($\omega_0$). Adiabaticity is $\omega \ll \omega_0$ — nothing about the total angle turned, everything about the rate.

Why does slow win? Because the coupling between levels is not just small — it is fast-oscillating. In the moving basis, the term feeding level $m$ from level $n$ beats at the gap frequency $(E_m-E_n)/\hbar$; a slowly varying prefactor times a fast oscillation integrates to almost nothing, leaving boundary terms of order $\hbar\dot H/(\Delta E)^2$. The gap does double duty — it divides the coupling and drives the protective oscillation — which is why the criterion involves the gap squared, and why everything dies at a level crossing. The theorem is quantitative, and one system lets us watch it with no approximations. For the spin on the cone the exact leak probability is known in closed form, and its ceiling collapses like the square of the sweep rate — the curve below spans six decades, with both worked-example points on it.

The same physics runs in a box. Widen an infinite well slowly and a ground-state particle stays in the instantaneous ground state, its energy sliding down $E_1(t) = \pi^2\hbar^2/2mL(t)^2$ as the walls recede — not lost, but paid out as work by quantum pressure on the retreating wall. Yank the wall instead and the wavefunction has no time to adapt: it stays frozen and must be re-expanded in the new box — the sudden regime, owned by The Sudden Approximation later in this chapter.

Freeze the Hamiltonian at each instant and diagonalize it. This defines the instantaneous eigenproblem — a family of snapshots, with no dynamics in it yet:

where: $|n(t)\rangle$ — the $n$-th eigenstate of the frozen $H(t)$, assumed non-degenerate; $E_n(t)$ — its eigenvalue. These are not solutions of the Schrödinger equation; they are a moving coordinate system the true solution will cling to in the slow regime. Expand the true state over this moving basis, peeling off the phase each snapshot would accumulate if it were genuinely stationary — the choice that cancels the $E_n$ term identically:

where: $c_n(t)$ — the amplitude actually carrying the dynamics; $\theta_n$ — the dynamical phase, the running generalization of $-E_nt/\hbar$. Were $H$ constant, every $c_n$ would freeze — the statistics-frozen stationary states of Stationary States and Energy Eigenstates; everything beyond that lives in the motion of the basis itself.

Suppose the leak terms are negligible. The exact equation of Step 2 collapses to $\dot c_n = -c_n\langle n|\dot n\rangle$, which integrates to a pure phase — $\langle n|\dot n\rangle$ is purely imaginary since $\partial_t\langle n|n\rangle = 0$. The adiabatic solution is

where: $\theta_n$ — the dynamical phase of (2), clocking the running energy; $\gamma_n$ — the geometric phase, fed not by energy but by the motion of the eigenvector itself. For real eigenvector families $\gamma_n$ vanishes identically (checked symbolically), which is why it went unnoticed for decades. When the parameter path closes into a loop, $\gamma_n$ becomes gauge-invariant and measurable — that loop physics, the spin-½ solid-angle result, and the curvature picture are owned by Geometric (Berry) Phase; here we establish only that the phase exists and where it comes from. "Stays in the eigenstate" thus has fine print: the system stays in the eigenstate ray, and the two phases riding on it are physics — interference experiments read them.

Claims about slowness deserve a system where nothing is approximated. Take an electron spin (Larmor frequency $\omega_0 = \gamma B$) in a field of fixed magnitude whose direction sweeps a cone of half-angle $\alpha$ at rate $\omega$:

where: $\omega_0$ — the internal gap clock ($E_+ - E_- = \hbar\omega_0$); $\omega$ — the external sweep clock. The instantaneous eigenstates are the spin-along-$\hat{\mathbf n}$ spinors $(\cos\frac{\alpha}{2},\, e^{i\omega t}\sin\frac{\alpha}{2})$ — verified as exact eigenvectors in the script. This is the compass needle of the opening picture, exactly solvable because the time dependence is a rigid rotation.

The theorem has a classical ancestor, and honoring it sharpens the quantum statement. The mechanics track's Adiabatic Invariants shows a pendulum with a slowly shortened string conserving not its energy but $J = E/\omega$, the energy growing as $E \propto \ell^{-1/2}$ and the amplitude as $\theta_{\max} \propto \ell^{-3/4}$; its Adiabatic Theorem for Slowly Varying Hamiltonians is honest about where invariance must fail — a pendulum pumped toward the top of its arc violates adiabaticity at the swing-to-spin crossover no matter how slow the pump. The quantum heir of the invariant action is the quantum number: since $E_n/\omega \propto (n+\frac12)\hbar$ for an oscillator, "conserve $E/\omega$" and "stay in level $n$" are one sentence in two languages — and the classical failure mode translates too: a diverging period is a collapsing gap. The infinite square well makes the energy bookkeeping vivid. Stay in $n=1$ while the wall recedes from $L$ to $L'$ and the energy must slide from $E_1(L)$ to $E_1(L')$ — where does the difference go? Into mechanical work: the confined state pushes on the wall with the quantum pressure force $F = -\mathrm dE_1/\mathrm dL = 2E_1/L$ (derived symbolically in the script), and the wall, retreating slowly, is paid exactly the energy the level loses. For an electron in a 1 nm box, $E_1 = 0.376$ eV and $F = 1.2\times10^{-10}$ N; widening to 2 nm delivers 0.282 eV to the wall. Adiabatic following conserves the quantum number, never the energy.

Example 1 — how slow is slow: the cone numbers. An electron spin sits in a $B = 0.1$ T field ($\omega_0/2\pi = 28.02\ \text{GHz/T} \times 0.1\ \text{T} = 2.80$ GHz) whose direction sweeps a cone of half-angle $\alpha = 45°$. Compute the worst-case leak probability for sweep rates $\omega/2\pi = 1$ MHz and 1 GHz.

Step 1. Form the exact ceiling $P_{\text{max}} = (\omega\sin\alpha/\lambda)^2$ with $\lambda = \sqrt{\omega_0^2 + \omega^2 - 2\omega_0\omega\cos\alpha}$.

Step 2. Slow sweep, $\omega/2\pi = 1$ MHz: $\omega/\omega_0 = 3.57\times10^{-4}$, $\lambda \approx \omega_0$, so $P_{\text{max}} \approx (3.57\times10^{-4}\times0.707)^2 = \textbf{6.4}\times\textbf{10}^{-8}$ — following is near-perfect through every turn.

Step 3. Fast sweep, $\omega/2\pi = 1$ GHz: $\omega/\omega_0 = 0.357$, $\lambda/\omega_0 = 0.789$, so $P_{\text{max}} = (0.357\times0.707/0.789)^2 = \textbf{0.102}$ — a ten-percent leak; the intermediate ladder runs $6.4\times10^{-6}$ at 10 MHz, $6.7\times10^{-4}$ at 100 MHz, riding the slope-2 law.

Step 4. Cross-check: direct numerical integration of the lab-frame Schrödinger equation reproduces the exact $P(t)$ at both rates to $5\times10^{-15}$ — the "theorem" here is not asymptotic hand-waving but a measured property of the exact solution.

Example 2 — the quantum piston's energy ledger. An electron occupies the ground state of a 1 nm infinite well whose right wall recedes slowly to 2 nm. Track the energy.

Step 1. Initial energy: $E_1(1\ \text{nm}) = \pi^2\hbar^2/2mL^2 = \textbf{0.376 eV}$.

Step 2. Adiabatic following keeps $n = 1$, so the final energy is $E_1(2\ \text{nm}) = 0.376/4 = \textbf{0.094 eV}$.

Step 3. The difference, $\textbf{0.282 eV}$, is work done on the wall: the level exerts $F = 2E_1/L$ ($1.2\times10^{-10}$ N at 1 nm — a tenth of a nanonewton of pure confinement pressure), and $\int F\,\mathrm dL$ reproduces the energy drop exactly, since $F = -\mathrm dE_1/\mathrm dL$.

Step 4. Contrast the yank: pull the wall instantly and the energy expectation is unchanged (no time to do work) — the old $\psi_1$ is now a superposition over the new box's levels. Which superposition is The Sudden Approximation's story.

References. Griffiths & Schroeter, 3rd ed., Ch. 11 · Sakurai & Napolitano, 2nd ed., §5.6 · Messiah, Vol. II, Ch. XVII · Born & Fock, Z. Phys. 51, 165 (1928) · Kato, J. Phys. Soc. Japan 5, 435 (1950) · Griffiths Example 10.1 (rotating-field spin, 2nd ed. numbering).

Provenance. Equations derived from the time-dependent Schrödinger equation in the instantaneous eigenbasis. A verifier should: (1) reproduce the coupled amplitude equations and the identity $\langle m|\dot n\rangle=\langle m|\dot H|n\rangle/(E_n-E_m)$ by differentiating the eigenproblem (Griffiths §11.1, Sakurai §5.6); (2) confirm the $(E_m-E_n)^2$ scaling via integration by parts (Messiah Vol. II Ch. XVII); (3) recompute the spin-½ Berry connection and the $\mp\tfrac12\Omega$ result against Berry (1984) and Griffiths §11.4; (4) check the Landau–Zener exponent against standard references. All numerical factors (½ solid angle, $2\pi$ in $\Omega$) are standard textbook values.

Builds on: The Interaction Picture and Time-Dependent Hamiltonians · Time-Dependent Perturbation Theory: General Formalism · First-Order Transition Amplitudes and Probabilities. Leads to: Berry Phase and Geometric Phases · Landau–Zener Transitions · The Born–Oppenheimer Approximation. (Scope note: the base export's Berry-connection/curvature formulas and spin-½ solid-angle derivation are owned by Geometric (Berry) Phase per the chapter ownership map; the Landau-Zener exponent by Landau–Zener Transitions.)