ETH and Thermalization

The proof requires careful treatment of the degeneracy structure of the rotor on \(S^2\):

Step 1 (Action variables and degeneracy): The unperturbed system has two degrees of freedom \((\theta, \phi)\) with action-angle variables \((I_{1}, I_{2}, \varphi_{1}, \varphi_{2})\) where \(I_{1} = L\) (total angular momentum magnitude) and \(I_{2} = L_{z}\) (z-component). The Hamiltonian

Step 2 (Degeneracy lifting): Because \(H_{0}\) is degenerate, the standard KAM non-degeneracy condition (\(\det(\partial^{2}H_{0}/\partial I_{i}\partial I_{j}) \neq 0\)) is not directly satisfied. However, for perturbations \(\epsilon V(\theta,\phi)\) that break the SO(3) symmetry to at most SO(2), first-order secular averaging over the fast angle \(\varphi_{1}\) generically lifts the \(L_{z}\)-degeneracy. The effective secular Hamiltonian \(\bar{H}(I_{1},I_{2}) = H_{0}(I_{1}) + \epsilon\langle V\rangle_{\varphi_{1}}(I_{1},I_{2})\) then satisfies non-degeneracy for \(\epsilon \neq 0\), allowing KAM theory to be applied to the averaged system. Alternatively, one can bypass KAM altogether and apply the Chirikov resonance-overlap criterion directly, which does not require non-degeneracy of the unperturbed Hamiltonian.

Step 3: By either route, tori with sufficiently irrational frequency ratios survive under small perturbation. Resonant tori where \(\omega_{1}/\omega_{2} = p/q\) (rational) are destroyed, and their neighborhoods become chaotic.

Step 4: The Chirikov overlap criterion: chaos becomes global when resonance widths exceed the spacing between resonances. For the standard map, \(K_{\text{crit}} \approx 0.9716\); on \(S^2\) with monopole perturbations, the effective threshold is geometry-dependent but of order unity. In practice, global chaos sets in when \(\epsilon/H_{0} \gtrsim O(1)\). \(\blacksquare\) □

The Lyapunov exponent has dimensions of inverse time. The only relevant scales are the velocity \(v \sim \sqrt{E/m}\) and the system size \(R\). Dimensional analysis gives \(\lambda_{L} \sim v/R\). This scaling is confirmed by numerical studies of chaotic billiards on curved surfaces. \(\blacksquare\) □

Step 1: Start with the Schrödinger equation on \(S^2\) with monopole:

Step 2: Substitute the ansatz \(\psi = A \exp(iS/\hbar + i\gamma)\) and expand in powers of \(\hbar\):

At \(O(\hbar^{0})\) (Hamilton-Jacobi equation):

At \(O(\hbar^{1})\) (transport equation):

Step 3: The Hamilton-Jacobi equation is solved by classical trajectories. For chaotic systems, multiple trajectories reach each point \(\Omega\), labeled by \(\alpha\).

Step 4: The Berry phase arises from the monopole:

The full proof (see Zelditch, 1987) proceeds as follows:

Step 1: Define the quantum observable \(\hat{A}\) corresponding to classical observable \(A(\Omega)\):

Step 2: The Egorov theorem relates quantum and classical evolution:

Step 3: For eigenstates, \(\langle E_{n}|\hat{A}|E_{n}\rangle\) is time-independent. Taking time averages:

Step 4: By classical ergodicity, the time average equals the phase space average:

Step 5: Combining steps 3-4 with careful control of the \(O(\hbar)\) error terms proves the theorem. The density-1 statement follows from a variance estimate showing that atypical eigenstates have measure zero. \(\blacksquare\) □

The proof has five main steps.

Step 1: Quantum ergodicity implies coefficient uniformity.

By Theorem thm:60v-quantum-ergodicity, the Husimi distribution \(\mathcal{H}_{n}(\Omega) \to 1/(4\pi)\) for almost all eigenstates. In the semiclassical limit (\(j \to \infty\)), this implies that the position-space probability density also becomes uniform:

Uniformity of \(|\psi_{n}|^{2}\) requires that the sum \(\sum_{m}|c_{nm}|^{2}|Y_{jm}|^{2}\) plus cross-terms averages to \(1/(4\pi)\).

Step 2: Phase randomization from chaos.

Write \(c_{nm} = |c_{nm}|e^{i\phi_{nm}}\). We claim the phases \(\phi_{nm}\) are uniformly distributed on \([0, 2\pi)\).

This follows from Theorem thm:60v-quantum-ergodicity (quantum ergodicity): Husimi uniformity requires that cross-terms \(\sum_{m \neq m'} c_{nm}^{*}c_{nm'} Y_{jm}^{*}Y_{jm'}\) average to zero over \(S^2\), which demands uncorrelated phases. More concretely, each coefficient \(c_{nm}\) corresponds to a projection onto a specific angular momentum state. Under chaotic evolution, the accumulated phase is:

where \(\omega_{m}\) is the instantaneous frequency and \(\gamma_{m}\) is the Berry phase. The mixing property ensures that \(\phi_{nm}\) explores \([0, 2\pi)\) uniformly over long times.

Step 3: Central limit theorem for amplitudes.

In the WKB representation (Theorem thm:60v-wkb-monopole), the wavefunction is a sum over classical trajectories:

The number of contributing trajectories scales as \(j\) (by the density of states). Each trajectory contributes with random phase (due to chaos).

By the central limit theorem for sums of random phasors, the coefficient \(c_{nm}\) (which involves an integral over \(\Omega\)) is a sum of \(O(j)\) random contributions, giving:

where \(N_{\text{eff}} \sim j\) is the effective number of contributing terms.

Step 4: Normalization determines variance.

The wavefunction normalization \(\sum_{m}|c_{nm}|^{2} = 1\) constrains the variance. Since there are \((2j+1)\) terms and each \(|c_{nm}|^{2}\) has expectation \(\sigma^{2}\):

Step 5: Independence from mixing.

The mixing property of classical chaos ensures that correlations between different modes decay on the timescale \(\tau_{L} = 1/\lambda_{L}\). Specifically:

The off-diagonal terms arise from residual correlations that vanish in the semiclassical limit.

Error estimate: The corrections come from:

• Finite-\(j\) corrections to Husimi uniformity: \(O(1/\sqrt{j})\)
• Residual mode correlations: \(O(1/j)\)
• WKB corrections: \(O(\hbar) = O(1/j)\)

The dominant error is \(O(1/\sqrt{2j+1}) = O(e^{-S/2})\) where \(S = \ln(2j+1)\) is the Boltzmann entropy (in natural units \(k_{B} = 1\)) of the \((2j+1)\)-dimensional Hilbert space. \(\blacksquare\) □

Step 1: Express the diagonal matrix element using the wavefunction:

Step 2: Expand \(|\psi_{n}|^{2}\) using the random wave representation:

Step 3: Take the ensemble average over random wave coefficients:

using the addition theorem \(\sum_{m}|Y_{jm}|^{2} = (2j+1)/(4\pi)\).

Step 4: The diagonal element becomes:

where \(\delta\rho_{n} = |\psi_{n}|^{2} - 1/(4\pi)\) is the density fluctuation.

Step 5: Compute the variance of \(\delta A_{n} = \int \delta\rho_{n} A \, d\Omega\):

Step 6: The density-density correlation for random waves is:

Using \(\sum_{m}Y_{jm}^{*}(\Omega)Y_{jm}(\Omega') = \frac{2j+1}{4\pi}P_{j}(\cos\gamma)\) where \(\gamma\) is the angle between \(\Omega\) and \(\Omega'\):

Step 7: Integrate to get the variance:

where \(A_{\ell}\) are the multipole moments of \(A\). For smooth observables, this gives:

The standard deviation is \(O(1/\sqrt{2j+1}) = O(e^{-S/2})\). \(\blacksquare\) □

Step 1: Write the off-diagonal element using wavefunctions:

Step 2: Expand in monopole harmonics:

where \(A_{k\ell} = \langle jk|\hat{A}|j\ell\rangle\) are the matrix elements of \(A\) in the monopole harmonic basis.

Step 3: For independent random wave eigenstates \(m \neq n\), the coefficients \(\{c_{mk}\}\) and \(\{c_{n\ell}\}\) are independent random variables. Therefore:

so \(\langle\langle E_{m}|\hat{A}|E_{n}\rangle\rangle = 0\) (zero mean).

Step 4: Compute the variance:

where \(\|A\|^{2} = \sum_{k,\ell}|A_{k\ell}|^{2}\) is the squared Frobenius norm.

Step 5: For local observables, \(\|A\|^{2} \sim (2j+1)\) (one factor from the trace). Therefore:

Step 6: The function \(f_{A}(\bar{E}, \omega)\) encodes the frequency dependence:

where \(\Delta m\) labels the off-diagonal band of \(\hat{A}\) in the angular momentum basis, and the mapping \(\Delta m \leftrightarrow \omega\) is determined by the spectral structure of the chaotic Hamiltonian. For physical (local) observables, \(f_{A}\) is smooth in \(\omega\) since selection rules restrict contributions to \(|\Delta m| \leq L_{A}\) where \(L_{A}\) is the multipole order of \(A\). \(\blacksquare\) □

Properties 1-2 follow immediately from the Berry random wave analysis in Theorem thm:60v-eth-off-diagonal.

Property 3 (Gaussianity): By Theorem thm:60v-berry-random-wave, each coefficient \(c_{nm}\) is Gaussian. The matrix element \(\langle E_{m}|\hat{A}|E_{n}\rangle\) is a sum of products \(c_{mk}^{*}c_{n\ell}\). By the central limit theorem for products of Gaussians, the result is Gaussian in the limit of large \(j\).

More precisely, let \(X = \text{Re}(R_{mn})\) and \(Y = \text{Im}(R_{mn})\). Then:

Each sum has \(O((2j+1)^{2})\) terms. For large \(j\), the CLT applies.

Property 4 (Independence): Consider two different pairs \((m,n)\) and \((k,l)\) with \(\{m,n\} \cap \{k,l\} = \emptyset\). Then \(R_{mn}\) depends on coefficients \(\{c_{m\bullet}, c_{n\bullet}\}\) while \(R_{kl}\) depends on \(\{c_{k\bullet}, c_{l\bullet}\}\).

By Theorem thm:60v-berry-random-wave, coefficients from different eigenstates are independent. Therefore \(R_{mn}\) and \(R_{kl}\) are independent.

For overlapping pairs (e.g., \(n = k\)), residual correlations exist but are suppressed by \(O(1/j)\):

This vanishes in the semiclassical limit. \(\blacksquare\) □

Step 1: Consider a perturbation initially localized to a single mode on \(S^2\). In the coherent state basis, this corresponds to a region of area \(\Delta\Omega \sim 1/j\) (the coherent state width).

Step 2: Under chaotic evolution, the perturbation spreads exponentially:

Step 3: Scrambling is complete when the perturbation has spread over the entire sphere:

Step 4: Solving for \(t_{*}\):

Step 5: In quantum terms, scrambling corresponds to the perturbation affecting all \(\dim\mathcal{H} = 2j+1\) degrees of freedom. The OTOC decays to its thermal value when:

This confirms \(t_{*} \sim \ln(\dim\mathcal{H})/\lambda_{L}\). \(\blacksquare\) □

Step 1: Expand the initial state in energy eigenbasis:

The time-evolved expectation value is:

Step 2: Separate diagonal and off-diagonal contributions:

where \(\omega_{mn} = (E_{m}-E_{n})/\hbar\).

Step 3: By ETH (Theorems thm:60v-eth-diagonal and thm:60v-eth-off-diagonal):

Step 4: The off-diagonal term dephases. The dephasing time is set by the spread in frequencies:

After time \(t \gg \hbar/\Delta E\), the oscillating terms average to zero.

Step 5: More precisely, the off-diagonal contribution decays as:

where \(\tilde{\rho}\) is the Fourier transform of the energy distribution \(|c_{n}|^{2}\) and \(N_{\text{eff}}\) is the effective number of participating states.

For a Gaussian energy distribution with width \(\Delta E\):

giving thermalization time \(t_{\text{th}} \sim \hbar/\Delta E\).

Step 6: For thermal initial states, \(\Delta E \sim k_{B}T\), so:

This is the fundamental quantum thermal timescale. \(\blacksquare\) □