Free Energy Subadditivity for Symmetric Random Hamiltonians

Mark Sellke

Abstract

We consider a random Hamiltonian $H:\Sigma\to\mathbb R$ defined on a compact space $\Sigma$ that admits a transitive action by a compact group $\mathcal G$. When the law of $H$ is $\mathcal G$-invariant, we show its expected free energy relative to the unique $\mathcal G$-invariant probability measure on $\Sigma$ obeys a subadditivity property in the law of $H$ itself. The bound is often tight for weak disorder and relates free energies at different temperatures when $H$ is a Gaussian process. Many examples are discussed including branching random walk, several spin glasses, random constraint satisfaction problems, and the random field Ising model. We also provide a generalization to quantum Hamiltonians with applications to the quantum SK and SYK models.

Introduction

In the study of statistical physics and disordered systems, the calculation of the "free energy" is a fundamental objective. For systems where the energy function—the Hamiltonian—is itself a random variable, determining the thermodynamic limit as the system size grows typically requires proving a property known as subadditivity. Historically, proving subadditivity has relied on model-specific interpolation techniques or complex convexity arguments, such as those developed for mean-field spin glasses.

This paper introduces a more universal source of subadditivity: symmetry. By leveraging transitive group actions on the state space, the author demonstrates that the expected free energy of combined random systems obeys a simple upper bound. This framework provides a unified approach that applies to a wide range of classical and quantum models without requiring the system size itself to vary.

The Symmetry Framework

The core theoretical contribution is centered on two mathematical requirements for the state space $\Sigma$ and the random Hamiltonian $H$:

  1. Transitive Symmetry: The state space must allow a group of transformations $\mathcal{G}$ (such as rotations, bit-flips, or permutations) that can map any state to any other state.
  2. Invariance: The probability distribution (the "law") of the random Hamiltonian must remain invariant under these transformations.

Using Jensen’s Inequality, the author proves that if a Hamiltonian $H_1$ satisfies these symmetry conditions, the expected free energy of the sum of $H_1$ and an independent Hamiltonian $H_2$ is less than or equal to the sum of their individual expected free energies:

$$F(L_1 + L_2) \le F(L_1) + F(L_2)$$

This result suggests that the disorder in a combined system is, in a sense, "less than the sum of its parts" regarding its free energy contribution.

Applications to Classical and Quantum Models

The utility of this subadditivity property is demonstrated across a diverse set of theoretical models. Because the proof relies on symmetry rather than specific structural details, it provides a streamlined path to bounds that were previously difficult to obtain.

Classical Systems

  • Spin Glasses: The framework applies to the Sherrington-Kirkpatrick (SK) model and mixed p-spin glasses.
  • Branching Random Walks: It covers directed polymers on trees.
  • Constraint Satisfaction: The results extend to random k-SAT and other random constraint satisfaction problems (CSPs).
  • Random Field Ising Model (RFIM): The paper provides a significantly shorter and more intuitive proof for known free energy bounds in the RFIM.

Quantum Systems

Extending these results to the quantum regime requires addressing non-commuting operators. By utilizing the Golden-Thompson inequality, the author proves that the subadditivity property holds for quantum Hamiltonians. This is particularly relevant for the Sachdev-Ye-Kitaev (SYK) model and the Quantum SK model, both of which are central to modern research in holography and black hole physics.

Key Theoretical Insights

Symmetry as a Proxy for Convexity

While previous breakthroughs in spin glass theory (notably by Guerra) utilized the specific covariance structures of Gaussian processes, this work shows that geometric symmetry alone is often sufficient to establish useful bounds. This broadens the scope of subadditivity to non-Gaussian and non-mean-field models.

High-Temperature Tightness

The subadditivity bound is found to be "asymptotically tight" in the high-temperature (weak disorder) regime. In this state, the "quenched" free energy—the value typically sought in physics—behaves similarly to the "annealed" free energy, which is much easier to compute.

Temperature Scaling in Gaussian Processes

For Hamiltonians that are Gaussian processes, the paper demonstrates that the free energy is subadditive in the square of the inverse temperature ($\beta^2$). This provides a concrete mathematical relationship between the free energies of a system at different temperatures.

Limitations and Future Directions

Despite its generality, the subadditivity bound has limitations in the low-temperature (strong disorder) regime. In these "frozen" or highly frustrated states, the specific landscape of the energy function becomes more influential than the broad symmetries of the system, leading to increased "slack" in the bound. Furthermore, at absolute zero ($T=0$), the inequality simplifies to a standard statement about maximum values ($\max(A+B) \le \max(A) + \max(B)$), losing the probabilistic nuance found at finite temperatures.

Future research may look toward extending these results to higher-order tensors (p-spin models for $p \ge 3$), which may require integrating tools from free probability theory to maintain the precision found in the matrix-based (2-spin) models.

Summary updated

Abstract

We consider a random Hamiltonian $H:Σ\to\mathbb R$ defined on a compact space $Σ$ that admits a transitive action by a compact group $\mathcal G$. When the law of $H$ is $\mathcal G$-invariant, we show its expected free energy relative to the unique $\mathcal G$-invariant probability measure on $Σ$ obeys a subadditivity property in the law of $H$ itself. The bound is often tight for weak disorder and relates free energies at different temperatures when $H$ is a Gaussian process. Many examples are discussed including branching random walk, several spin glasses, random constraint satisfaction problems, and the random field Ising model. We also provide a generalization to quantum Hamiltonians with applications to the quantum SK and SYK models.
BibTeX
@article{sellke2023free,
  author = {Mark Sellke},
  title = {{Free Energy Subadditivity for Symmetric Random Hamiltonians}},
  journal = {Journal of Mathematical Physics},
  year = {2023},
  doi = {10.1063/5.0124718},
}
Figure