Eigenstate Thermalization and Simple Observables

The eigenstate thermalization hypothesis (ETH) is a claim about matrix elements of a declared class of observables within a narrow energy window and a single symmetry sector. It is not the statement that every operator looks thermal in every eigenstate. For a chaotic many-body Hamiltonian, few-body or otherwise simple observables often have smoothly varying diagonal elements and entropy-suppressed off-diagonal elements. That structure can explain equilibration of those observables in an individual pure state. Whether the relevant black-hole microstates and gravitational observables satisfy it is a frontier question with important counterexamples.

Required background. Quantum Information supplies density matrices, entropy, and distinguishability; Gauge Theory and String-Theory Preparation supplies constrained sectors and the large-system setting for black-hole microstates.

Helpful background. Semiclassical Gravity supplies the thermodynamic quantities matched by simple observables; Quantum Fields in Curved Spacetime supplies the exterior operator algebra whose correlators are compared.

The ETH ansatz and its labels

Let $H|n\rangle=E_n|n\rangle$ after resolving every exact symmetry used in the analysis. For an observable $O$ in a specified simple class, the standard ETH ansatz is

$$O_{mn} =\overline O(\bar E)\delta_{mn} +e^{-S(\bar E)/2}f_O(\bar E,\omega)R_{mn},$$

where $\bar E=(E_m+E_n)/2$, $\omega=E_m-E_n$, $S(\bar E)$ is the microcanonical entropy, and $R_{mn}$ is an irregular order-one quantity with the Hermiticity correlations required by $O$. The formula is assessed inside an energy window narrow on macroscopic scales but containing many levels.

The diagonal part says that neighboring eigenstates give nearly the same expectation value for $O$. The off-diagonal part sets fluctuation and dephasing scales. The ansatz does not apply to the projector $|n\rangle\langle n|$, which distinguishes one exact state by construction, or automatically to long nonlocal strings, exponentially complex decoders, and operators connecting different charge sectors.

Srednicki gives this energy-basis ansatz and its entropy normalization in Srednicki 1999, § II, equations (4)–(6). D’Alessio and collaborators organize its numerical tests, observable dependence, and known exceptions in D’Alessio et al. 2016, §§ 2–4 and 6.

How simple observables equilibrate

For an initial state $|\psi(0)\rangle=\sum_n c_n|n\rangle$,

$$\langle O(t)\rangle =\sum_n |c_n|^2O_{nn} +\sum_{m\ne n}c_m^*c_n e^{i(E_m-E_n)t}O_{mn}.$$

If the energy distribution is narrow, smooth $O_{nn}$ makes the diagonal term approximately microcanonical. Nondegenerate energy gaps and entropy-suppressed off-diagonal elements make temporal fluctuations small after dephasing. This argument uses both the ETH matrix elements and dynamical phase cancellation; canonical typicality alone supplies neither.

To see the role of the gap assumption, write the infinite-time mean as $\overline{\langle O\rangle}=\sum_n|c_n|^2O_{nn}$. Distinct nonzero energy gaps make all unmatched phases average to zero, leaving

$$\overline{\left|\langle O(t)\rangle-\overline{\langle O\rangle}\right|^2} =\sum_{m\ne n}|c_m|^2|c_n|^2|O_{mn}|^2.$$

If the off-diagonal elements in the occupied window obey a uniform bound $|O_{mn}|^2\le K e^{-S}$, the variance is at most $K e^{-S}$, because the probability weights sum to at most one. A typical-size ETH estimate alone needs control of exceptional elements before making that uniform claim. This is a long-time fluctuation statement, not a bound on the equilibration time; the corresponding distinction appears in Srednicki 1999, § III, equations (12)–(17).

For a black-hole sector with $S\gg1$, the factor $e^{-S/2}$ is nonperturbatively small in a semiclassical expansion. Simple exterior probes can then agree across many exact microstates to all accessible perturbative orders, while a finely tuned global measurement still distinguishes orthogonal states. That is restricted indistinguishability, not exact equality of states or interiors.

Results and open questions through 2026-08-19

This discussion reflects results available through 2026-08-19. Competing positions differ mainly on how far finite many-body evidence and holographic intuition can be extended to the gravitational observable algebra.

Model or evidence	Result	Scope or limitation
Chaotic matrix-element ansatz in Srednicki 1999, §§ II–III, equations (4)–(17)	Smooth diagonal data and entropy-scaled off-diagonal fluctuations yield microcanonical behavior for suitable operators.	The ansatz and observable class are hypotheses, not a theorem for arbitrary Hamiltonians.
Exact-diagonalization evidence summarized in D’Alessio et al. 2016, §§ 2–4 and 6	Many nonintegrable lattice models show ETH scaling after resolving symmetries and controlling finite size.	Accessible sizes are finite; integrability, localization, spectrum edges, and hidden charges produce failures.
Quench comparison in Rigol, Dunjko, and Olshanii 2008, pp. 2–4	Nonintegrable dynamics approaches predictions tied to eigenstate thermalization, unlike the integrable comparison.	The result concerns a particular lattice model and observable set, not gravity or every initial state.
Black-hole extrapolation from $e^{-S/2}$ suppression	Simple exterior correlators may be nearly identical across an entropy-sized sector.	A gravitational factorization, exact Hamiltonian, operator class, and single-theory finite-size test are not yet generally available.

The permissive interpretation treats black holes as maximally chaotic systems and expects ETH for all low-complexity exterior observables in an appropriate microcanonical sector. The conservative interpretation accepts that semiclassical correlators are coarse but regards ETH as unestablished until a single microscopic theory, symmetry sector, energy window, and gravitationally dressed operator set are specified. The evidence does not settle this extrapolation, which remains unresolved.

Adverse checks and formulation dependence

Take the integrable Hamiltonian $H=\sum_i h_i Z_i$. Its energy eigenstates are product states, and the simple observable $Z_1$ can retain values $+1$ or $-1$ among nearby energies. The smooth-diagonal ETH statement fails even though the Hilbert space is exponentially large. Adding an unaccounted conserved charge can create the same apparent failure by mixing sectors that should have been separated.

Many-body localized systems and quantum-scar subspaces provide further adverse cases: a Hamiltonian can be nonintegrable in a broad sense yet contain special eigenstates that violate strong ETH. At a spectral edge, the energy window may contain too few states for smooth scaling. A credible black-hole claim must therefore name weak versus strong ETH, the state fraction allowed to fail, the observable norm, the entropy range, and the finite-size residual.

Generic quantum-information theory, stable thermodynamics, and explicit string or holographic constructions lie outside this page. The bounded claim is that ETH can make individual microstates appear thermal to simple probes only after its observable, energy-window, symmetry, and dynamical hypotheses are made explicit.

The conclusions could change after a controlled microscopic black-hole Hamiltonian shows stable ETH scaling across increasing entropy, or a gravitationally dressed simple operator supplies a persistent counterexample, or a theorem changes the allowed scar fraction, symmetry resolution, or off-diagonal error bound in the relevant single-theory regime.

Exercises

In an energy window with entropy $S=100$, assume $f_O R_{mn}$ has magnitude one. Estimate the ETH off-diagonal scale and explain why this does not make two energy eigenstates identical.

Solution

The scale is $e^{-S/2}=e^{-50}\simeq1.93\times10^{-22}$. It suppresses the off-diagonal matrix element of this selected observable, so ordinary probes can have extremely small temporal fluctuations. The exact eigenstates remain orthogonal and can be distinguished by a sufficiently state-specific projector or a more complex observable. ETH constrains a declared observable class; it does not collapse the Hilbert-space dimension.

References

• D’Alessio, Luca, Yariv Kafri, Anatoli Polkovnikov, and Marcos Rigol. 2016. “From Quantum Chaos and Eigenstate Thermalization to Statistical Mechanics and Thermodynamics.” Advances in Physics 65, 239–362. arXiv:1509.06411.
• Rigol, Marcos, Vanja Dunjko, and Maxim Olshanii. 2008. “Thermalization and Its Mechanism for Generic Isolated Quantum Systems.” Nature 452, 854–858. arXiv:0708.1324.
• Srednicki, Mark. 1999. “The Approach to Thermal Equilibrium in Quantized Chaotic Systems.” Journal of Physics A: Mathematical and General 32, 1163–1175. DOI. arXiv:cond-mat/9809360.