Theory of Eigenstate Thermalisation

Tobias Helbig,Tobias Hofmann,Ronny Thomale,Martin Greiter

2024-06-03

Abstract:If we prepare an isolated, interacting quantum system in an eigenstate and perturb a local observable at an initial time, its expectation value will relax towards a thermal expectation value, even though the time evolution of the system is deterministic. The eigenstate thermalization hypothesis (ETH) of Deutsch and Srednicki suggests that this is possible because each eigenstate of the full quantum system acts as a thermal bath to its subsystems, such that the reduced density matrices of the subsystems resemble thermal density matrices. Here, we use the observation that the eigenvalue distribution of interacting quantum systems is a Gaussian under very general circumstances, and Dyson Brownian motion random matrix theory, to derive the ETH and thereby elevate it from hypothesis to theory. Our analysis provides a derivation of statistical mechanics which neither requires the concepts of ergodicity or typicality, nor that of entropy. Thermodynamic equilibrium follows solely from the applicability of quantum mechanics to large systems and the absence of integrability.

Quantum Physics,Statistical Mechanics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the theoretical basis of the thermalization phenomenon in quantum systems. Specifically, the authors explored how the expected value of a local observable in an isolated and interacting quantum system tends to the thermal equilibrium value after being perturbed, even though the evolution of the system is deterministic. The core of this problem lies in the **Eigenstate Thermalization Hypothesis (ETH)**, that is, each eigenstate can serve as a thermal bath for the subsystem, so that the reduced density matrix of the subsystem is similar to the thermal density matrix.

### Core Problems of the Paper
1. **Theoretical Basis of the Eigenstate Thermalization Hypothesis (ETH)**:
- The authors used **Gaussian Random Matrix Theory (RMT)** and **Dyson Brownian Motion** to derive ETH, thus elevating it from a hypothesis to a theory.
- They proved that for interacting quantum systems, ETH holds without relying on concepts such as ergodicity or typicality.

2. **Origin of Thermodynamic Equilibrium**:
- Thermodynamic equilibrium is explained only by the applicability of quantum mechanics to large systems and non - integrability, without the need to introduce the concept of entropy.

### Main Contributions
1. **Derivation of ETH**:
- The authors derived ETH by analyzing the eigenvalue distribution (Gaussian distribution) of interacting quantum systems and Dyson Brownian Motion.
- This derivation not only verifies the validity of ETH but also provides a new statistical mechanics derivation method that does not rely on ergodicity or typicality.

2. **Numerical Results**:
- The authors carried out detailed numerical simulations to verify the results of the theoretical derivation. These simulations include the overlap curves of eigenstates, the width and displacement of the Lorentzian distribution, etc.
- The numerical results are highly consistent with the theoretical predictions, further supporting the correctness of ETH.

3. **Consistency Conditions**:
- The authors studied the consistency conditions of ETH, especially whether the ETH of each eigenstate is consistent when the entire system is composed of multiple eigenstates.
- They proved the general validity of ETH through nested conditions and the self - averaging property in the large - system limit.

### Formula Summary
- **Reduced Density Matrix of Eigenstates**:
\[
\hat{\rho}_S^n \equiv \text{Tr}_B |\psi_n\rangle \langle \psi_n|
\]
\[
\hat{\rho}_S(\beta) \equiv \text{Tr}_B \hat{\rho}(\beta) \quad \text{with} \quad \hat{\rho}(\beta) = \frac{1}{Z(\beta)} e^{-\beta \hat{H}}
\]
\[
\langle \hat{H} \rangle_\beta = -\frac{\partial}{\partial \beta} \ln Z(\beta) = \lambda_n
\]
\[
Z(\beta) = \text{Tr} \, e^{-\beta \hat{H}}
\]

- **Eigenvalue Distribution**:
\[
\rho(\lambda) = \sqrt{\frac{\alpha}{2\pi}} e^{-\frac{1}{2\alpha} \lambda^2}
\]
\[
\alpha = \Delta_{\text{tot}}^2
\]

- **Lorentzian Distribution**:
\[
\chi_{\mu i, n} \approx \frac{1}{\pi \gamma_\mu} \frac{1}{(\epsilon_\mu + E_i - \lambda_n)}