Random systems and replica field theory

Marc MézardarXiv 1995 · DOI · arXiv

Abstract

This paper provides a foundational overview of statistical systems characterized by quenched disorder—where randomness is frozen into the system's structure. It examines the limitations of standard perturbative methods, such as the dimensional reduction paradox, and demonstrates how the Replica Method and Replica Symmetry Breaking (RSB) provide a robust framework for analyzing complex energy landscapes dominated by metastable states. Key applications include spin glasses, manifolds in random media, and vortex lattices.

I. Introduction

In classical statistical mechanics, systems typically reach thermal equilibrium by settling into a global minimum of free energy. However, many real-world materials—ranging from "dirty" crystals and superconductors to biological protein structures—exhibit quenched disorder. In these systems, the randomness is not a thermal fluctuation but a fixed structural feature.

The presence of quenched disorder creates a rugged energy landscape filled with numerous "metastable states" (local minima). Standard theoretical tools, such as the Renormalization Group or simple Mean Field Theory, often fail in this context because they assume a single, well-defined equilibrium state. This paper explores the non-perturbative methods required to navigate these complex landscapes, specifically focusing on the Replica Method.

II. The Failure of Perturbation and Dimensional Reduction

A primary motivation for developing replica field theory is the failure of standard perturbative expansions. A classic example is the Random Field Ising Model (RFIM). Early theoretical attempts using simple perturbation predicted a "dimensional reduction" property, suggesting that a disordered system in $d$ dimensions should behave identically to a pure system in $d-2$ dimensions.

Physically, this result proved incorrect at low dimensions. The failure stems from the fact that perturbation theory effectively treats every random potential as a simple random force. It is "blind" to the existence of multiple metastable states, assuming the system remains trapped in a single basin of attraction. To correctly describe these systems, one must account for the global structure of the energy landscape.

III. The Replica Method and RSB

To calculate the observable properties of a disordered system, one must average the free energy, which requires calculating the average of the logarithm of the partition function ($\\overline{\\ln Z}$). This is mathematically intractable using standard averaging techniques.

The Replica Trick

The Replica Method bypasses this difficulty by using the identity: $$\\overline{\\ln Z} = \\lim_{n \\to 0} \\frac{\\overline{Z^n} - 1}{n}$$ This involves studying $n$ identical copies (replicas) of the system. By calculating the properties of these $n$ replicas and then taking the limit as $n$ approaches zero, one can extract the averaged properties of the original disordered system.

Replica Symmetry Breaking (RSB)

Initially, theorists assumed "Replica Symmetry" (RS), where all $n$ copies are treated as identical and interchangeable. However, in complex systems like spin glasses, this symmetry must be broken. Replica Symmetry Breaking (RSB) allows the replicas to "condense" into different metastable states. Physically, RSB represents the reality that the system's Boltzmann distribution is not a single peak, but a weighted sum of many Gaussians centered at different local minima across the energy landscape.

IV. Manifolds in Random Media: A Variational Approach

The author applies a Gaussian variational method to find the best quadratic approximation of the system's Hamiltonian. This is particularly useful for studying manifolds (such as 1D directed polymers or 2D sheets) wiggling through a medium filled with impurities.

A key metric in these studies is the "wandering exponent" $\\zeta$, which describes how fluctuations scale over a distance $l$ (fluctuations $\\sim l^{2\\zeta}$). While Replica Symmetry yields a wandering exponent of $\\zeta = (4-D)/2$, the RSB solution provides a corrected exponent: $$\\zeta = \\frac{4-D}{2(1+\\gamma)}$$ This result correctly accounts for the correlations in the random potential and provides a microscopic derivation for heuristic scaling arguments like the Imry-Ma scaling.

V. Physical Interpretation and Results

The RSB solution offers a clear physical picture of the system's state. For any given sample, the system identifies a specific set of favored configurations. The "one-step RSB" solution effectively maps out the most significant local minima. This approach becomes exact in the limit where the number of components ($N$) of the field goes to infinity, serving as a powerful "Hartree-like" approximation for high-dimensional complex systems.

VI. Limitations and Conclusion

While the Gaussian variational approach with RSB is a significant advancement, it remains a mean-field type theory. Its primary limitations include:

  • Finite N Corrections: The method is most robust as $N \\to \\infty$; systematic expansions for finite $N$ (the physical reality) remain technically challenging.
  • Statics vs. Dynamics: This framework focuses on equilibrium statics. However, many disordered systems exhibit "aging" and never truly reach equilibrium, necessitating a separate dynamical treatment.

In summary, replica field theory provides the necessary mathematical language to describe systems where randomness and complexity are intrinsic. By breaking replica symmetry, we gain a window into the hidden structure of metastable states that define the physics of quenched disorder.

Summary updated

Abstract

Contents: I. Introduction II. Manifolds in random media III. Thermal fluctuations without disorder IV. Random forces V. Random potential: variational approach VI. Physical interpretation of the solution
BibTeX
@misc{mezard1995random,
  author = {Marc Mézard},
  title = {{Random systems and replica field theory}},
  year = {1995},
  eprint = {cond-mat/9503056},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/cond-mat/9503056},
}
Figure