Level correlations in disordered metals: The replica<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mi>σ</mml:mi></mml:math>model

Alex Kamenev, Marc MézardPhysical 1999 · DOI · arXiv

Abstract: We compute energy-level correlations in weakly disordered metallic grains using the fermionic replica method. We use the standard $\sigma$-model approach and show that nontrivial saddle points, which break replica symmetry, must be included in the calculation to reproduce the oscillatory behavior of the correlations. We calculate the correlation functions in all three classical ensembles Gaussian-orthogonal, -unitary, and -sympletic, including the finite-dimensional gradient corrections. Our results coincide with those obtained by the supersymmetric $\sigma$ model and the semiclassical trace formula.

Introduction

In the theoretical physics of disordered systems, the 1999 work of Alex Kamenev and Marc Mézard stands as a significant milestone. It resolved a long-standing debate regarding the "replica trick"—a mathematical technique used to study systems with random impurities. While the replica method was often more intuitive than its alternatives, it historically struggled to capture the fine-scale, oscillatory energy-level statistics of disordered metals. Kamenev and Mézard demonstrated that by accounting for subtle "replica symmetry breaking" (RSB) effects, the replica method could fully describe these complex correlations, matching the accuracy of more mathematically dense supersymmetric theories.

The Problem of Level Correlations

In quantum chaos and the study of disordered metals, electron energy levels are not distributed randomly. Instead, they exhibit "level repulsion," a phenomenon where levels prefer to maintain a specific distance from one another, governed by the Wigner-Dyson statistics of Random Matrix Theory (RMT).

To calculate these correlations, physicists typically employ the non-linear $\sigma$-model. For decades, a divide existed between two primary field-theoretic tools:

  1. The Supersymmetry (SUSY) Method: Highly successful and accurate, but mathematically complex and often physically opaque.
  2. The Replica Method: Conceptually simpler—treating the system as $n$ identical copies and taking the limit $n \to 0$—but notoriously unable to capture the "oscillatory" part of correlation functions. These "wiggles" define the fine-scale statistics of energy levels, and their absence led many to believe the replica method was fundamentally incomplete for this application.

Methodology: Breaking Replica Symmetry

Kamenev and Mézard addressed this gap using a fermionic replica $\sigma$-model combined with a sophisticated saddle-point analysis. Their innovation lay in looking beyond the standard, obvious solutions of the functional integral.

Saddle-Point Innovation

In the standard approach, researchers typically seek the "replica symmetric" saddle point, where all $n$ copies of the system behave identically. This approach only yields the smooth, non-oscillatory component of the correlations. The authors identified non-trivial saddle points that spontaneously break the symmetry between replicas. These solutions correspond to different ways of partitioning the $n$ replicas into sub-blocks.

Analytical Continuation and Geometry

A major technical hurdle in replica theory is the $n \to 0$ limit. The authors utilized group theory to calculate the volumes of the manifolds (coset spaces) associated with these saddle points. They demonstrated that only specific configurations survive the limit to contribute to the final correlation function. Furthermore, they included spatial variations (gradient corrections) to account for finite-dimensional effects, allowing the theory to describe how correlations evolve at the Thouless energy scale ($E_c$).

Key Insights and Results

The paper provides several critical insights into the nature of disordered systems:

  • Physicality of RSB: The oscillations in energy-level correlations are shown to be the physical manifestation of replica symmetry breaking. Without RSB, the theory only captures average behavior, missing the quantum interference effects that cause the characteristic "wiggles."
  • Geometric Determinism: The final results are heavily dependent on the geometry of abstract replica spaces (such as $U(n)/U(p)U(n-p)$). The authors showed that these geometries dictate the physics of electron levels within a physical metal grain.
  • Universality: The framework elegantly separates universal RMT behavior—which depends solely on the symmetry class (Orthogonal, Unitary, or Symplectic)—from non-universal features determined by the specific shape and conductance of the metal.

Validation

The authors successfully derived the two-point correlation function $R(\omega)$ for all three classical ensembles: Gaussian-orthogonal (GOE), Gaussian-unitary (GUE), and Gaussian-symplectic (GSE). Their results for the oscillatory terms perfectly match the landmark results obtained by Andreev and Altshuler via the SUSY method and align with the semi-classical trace formula. Additionally, the formulas correctly describe the crossover from universal RMT behavior at low energies ($\omega \ll E_c$) to the power-law or exponential decay observed at high energies ($\omega > E_c$).

Limitations and Future Directions

While the paper represents a breakthrough, it operates within certain theoretical bounds. The saddle-point approximation used is strictly valid when the energy difference $\omega$ is much larger than the mean level spacing $\Delta$. Although the results are suspected to hold for all $\omega$, the current derivation is technically an asymptotic expansion.

Furthermore, the theory assumes a "good" metal with weak disorder ($g \gg 1$) and does not directly address the Anderson transition to an insulating state. Finally, the study treats non-interacting electrons; extending this replica approach to interacting electron systems remains a primary challenge for the field.

Summary updated

Abstract

We compute energy-level correlations in weakly disordered metallic grains using the fermionic replica method. We use the standard $\ensuremath{\sigma}$-model approach and show that nontrivial saddle points, which break replica symmetry, must be included in the calculation to reproduce the oscillatory behavior of the correlations. We calculate the correlation functions in all three classical ensembles Gaussian-orthogonal, -unitary, and -sympletic, including the finite-dimensional gradient corrections. Our results coincide with those obtained by the supersymmetric $\ensuremath{\sigma}$ model and the semiclassical trace formula.
BibTeX
@inproceedings{kamenev1999level,
  author = {Alex Kamenev and Marc Mézard},
  title = {{Level correlations in disordered metals: The replica<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mi>σ</mml:mi></mml:math>model}},
  booktitle = {Physical},
  year = {1999},
  doi = {10.1103/physrevb.60.3944},
}
Figure