Papers
Topics
Authors
Recent
Assistant
AI Research Assistant
Well-researched responses based on relevant abstracts and paper content.
Custom Instructions Pro
Preferences or requirements that you'd like Emergent Mind to consider when generating responses.
Gemini 2.5 Flash
Gemini 2.5 Flash 177 tok/s
Gemini 2.5 Pro 43 tok/s Pro
GPT-5 Medium 26 tok/s Pro
GPT-5 High 25 tok/s Pro
GPT-4o 119 tok/s Pro
Kimi K2 202 tok/s Pro
GPT OSS 120B 439 tok/s Pro
Claude Sonnet 4.5 38 tok/s Pro
2000 character limit reached

Eigenstate Thermalization Hypothesis (ETH)

Updated 16 October 2025
  • ETH is a principle in quantum statistical mechanics where eigenstate observables match smooth microcanonical ensemble averages at corresponding energies.
  • It predicts that differences between adjacent eigenstate observables decay exponentially with system size, affirming thermalization in nonintegrable many-body systems.
  • Periodically driven systems show enhanced thermalization, as the removal of energy conservation leads to narrower distributions of ETH indicators.

The Eigenstate Thermalization Hypothesis (ETH) is a central organizing principle in the theory of quantum statistical mechanics. It posits that, for a broad class of isolated, nonintegrable quantum many-body systems, the stationary values of few-body observables in highly-excited energy eigenstates coincide with ensemble (thermal) averages at the appropriate energy density. In this strong sense, every physically relevant eigenstate is "thermal," and thus ETH provides a fundamental mechanism by which irreversibility and equilibrium arise from microscopic unitary dynamics. Conversely, the rigorous validity of ETH in nonintegrable systems, and the behavior of possible outlier states or finite-size effects, are crucial for assessing the limits of statistical mechanics in quantum systems.

1. Formal Statement and Significance of ETH

ETH asserts that, for generic nonintegrable many-body Hamiltonians HH and any few-body observable O^\hat O, the expectation value in an energy eigenstate n|n\rangle with eigenvalue EnE_n satisfies

nO^n=O(En),\langle n| \hat O | n \rangle = O(E_n),

where O(E)O(E) is a smooth function of energy that matches the prediction of the microcanonical (or canonical) ensemble at the same energy density. In the strong formulation—investigated directly in (Kim et al., 2014)—this thermalization property holds for every eigenstate in the bulk of the spectrum (excluding at most spectral edges/noise).

The physical implications are profound: it means that even a pure initial state, as long as it is concentrated near a given energy density, will thermalize, with long-time averages of observables equaling equilibrium predictions. Thus, ETH solves the apparent conflict between unitary quantum evolution and the emergence of thermodynamics by embedding statistical behavior into the structure of eigenstates themselves.

2. Numerical Methodology: Model Systems and Observables

To scrutinize ETH in the strong sense, (Kim et al., 2014) employs exact diagonalization of two paradigmatic 1D nonintegrable systems:

  • Quantum Ising Chain with Transverse and Longitudinal Fields:

H=i[gσix+hσiz+Jσizσi+1z]H = \sum_i \left[ g \sigma^x_i + h \sigma^z_i + J \sigma^z_i \sigma^z_{i+1} \right]

with (g,h,J)=(0.9045,0.8090,1)(g, h, J) = (0.9045, 0.8090, 1). The nonintegrability here is robust. Observables analyzed include single-site σ1x,σ1z\sigma^x_1, \sigma^z_1 and two-site operators σ1xσ2x\sigma^x_1 \sigma^x_2, σ1yσ2y\sigma^y_1 \sigma^y_2.

  • Hard-Core Boson Model:

H=i[t(bi+1bi+h.c.)+Vnini+1]+i[t(bi+2bi+h.c.)+Vnini+2]H = \sum_i \left[ -t (b^\dagger_{i+1} b_i + h.c.) + V n_i n_{i+1} \right] + \sum_i \left[ -t' (b^\dagger_{i+2} b_i + h.c.) + V' n_i n_{i+2} \right]

with t=V=t=V=1t = V = t' = V' = 1 at half filling (N/L=1/2N/L = 1/2). Relevant observables included two-site density correlators n1n2n_1 n_2 and real/imaginary parts of the hopping operators.

For each model, the Hamiltonians were block-diagonalized by momentum and numerically diagonalized to full precision for system sizes up to those accessible with available computational resources.

3. ETH Indicators and Outlier Analysis

To go beyond microcanonical averaging, the paper introduces the ETH indicator

rnn+1O^n+1nO^n,r_n \equiv \langle n+1 | \hat O | n+1 \rangle - \langle n | \hat O | n \rangle,

which measures the difference in observable values between adjacent eigenstates. ETH predicts that rnr_n must vanish as system size grows, since eigenvalue spacings diminish exponentially and observables become smooth functions of energy.

"Outlier" eigenstates are defined as those maximizing rn|r_n|—representing the strongest possible deviation from ETH for a given system size and observable.

Key numerical findings are:

  • Exponential reduction of deviations: Even the largest rn|r_n| (“extreme outliers”) shrink with increasing system size LL; the mean deviation r\langle |r| \rangle decays as rexp(αL)\langle |r| \rangle \sim \exp(-\alpha L), consistent with a D1/2D^{-1/2} scaling in Hilbert space dimension DD.
  • Properties of outliers: Outliers tend to have low participation ratios (i.e., less delocalized eigenvectors) and often appear in special momentum sectors (k=0k=0 or k=L/2k=L/2) due to additional symmetries. However, with increasing LL, even these outliers approach thermal values, supporting strong ETH.
  • Distribution collapse: The full distribution P(r)P(|r|) peaks more sharply at zero as LL grows, confirming that almost all eigenstates become thermal in the thermodynamic limit.

4. Effects of System Size and Distribution Properties

System size scaling is pivotal for the ETH:

System Size LL Mean deviation r\langle|r|\rangle Maximum outlier r|r| P(r)P(|r|) shape
Small O(102)O(10^{-2}) Largest Broad, non-Gaussian
Intermediate O(103)O(10^{-3}) Moderate Peaked, but finite
Large O(104)O(10^{-4}) or smaller Very small Sharply peaked at 0

With increasing LL, not only the variance but the largest outlier r|r| becomes negligible. Graphical plots in the paper show that the spread of both typical and extreme deviations collapses exponentially.

This scaling is essential—if outliers decreased only polynomially, a nonzero fraction of nonthermal states would remain at large LL, violating strong ETH.

5. Periodic Driving and Enhanced Thermalization

The investigation also includes the effect of periodically driving the nonintegrable Ising system, which removes strict energy conservation and generates a Floquet operator: U^=exp(iHxτ/2)exp(iHzτ/2)\hat U = \exp(-i H_x \tau/2) \exp(-i H_z \tau/2) with Hx=igσixH_x = \sum_i g \sigma^x_i and Hz=i[hσiz+Jσizσi+1z]H_z = \sum_i [ h \sigma^z_i + J \sigma^z_i \sigma^z_{i+1}] and period τ\tau.

In this case:

  • Floquet eigenstates thermalize to infinite-temperature predictions (i.e., σ1x0\langle \sigma^x_1 \rangle \to 0).
  • Both average and outlier deviations in σ1x| \langle \sigma^x_1 \rangle | are further reduced compared to the undriven (energy-conserving) case.
  • The removal of conserved quantities (here, energy) "improves" thermalization, reflected in a narrower distribution of ETH indicators.

This suggests that constraints from conserved quantities can inhibit thermalization, and their removal allows faster or "stronger" approach to ETH behavior.

Quantitatively, the paper presents:

  • The ETH indicator rnr_n and its full distribution.
  • Exponential fit of rexp(αL)\langle|r|\rangle \sim \exp(-\alpha L) (with α\alpha empirically extracted for each operator and model).
  • For Floquet systems, reduced fluctuations in both mean and maximal deviations.

For example, for the Ising chain at L=16L=16 versus L=20L=20, plots of nO^n\langle n | \hat O | n \rangle vs. energy density show ever-decreasing scatter around the thermal curve, and tabulated lists of outlier eigenstates verify that even the worst cases become negligible with growth in LL.

Such trends are robust across both Ising and bosonic models and for various local and few-body observables.

7. Implications for Many-Body Quantum Thermalization

These results establish numerically that, for generic nonintegrable quantum systems in one dimension:

  • Strong ETH holds: all eigenstates (in the spectral bulk) are thermal, including extreme outliers, as LL \to \infty.
  • The relaxation of observables from initial states with well-defined energy density leads to equilibrium values dictated by the microcanonical ensemble, with uniqueness guaranteed by ETH.
  • Symmetry constraints (e.g., total momentum, or energy in driven/undriven cases) create weak, finite-size subleading corrections but do not invalidate thermalization in the thermodynamic limit.

In periodically driven (Floquet) systems, the strengthened ETH further implies that energy conservation is not essential for quantum systems to thermalize; thus, dynamical protocols can enhance the universality of equilibrium properties.

In summary, through high-precision studies of both typical and extreme behavior, exact diagonalization, and analysis of both static and periodically driven systems, strong ETH is robustly validated for nonintegrable models, implying the universality of thermalization in such quantum systems (Kim et al., 2014).

Definition Search Book Streamline Icon: https://streamlinehq.com
References (1)
Forward Email Streamline Icon: https://streamlinehq.com

Follow Topic

Get notified by email when new papers are published related to Eigenstate Thermalization Hypothesis (ETH).