$d$-dimensional spherical ferromagnets in random fields: Metastates, continuous symmetry breaking, and spin-glass features

Technical Summary: $d$-Dimensional Spherical Ferromagnets in Random Fields

Problem Statement
This paper investigates the large-volume asymptotic behavior of the vector-valued ferromagnetic spherical model in the presence of quenched $d$-dimensional random fields, specifically for spin dimensions $d \ge 2$. The study focuses on the phenomenon of "random symmetry breaking," where the system possesses multiple infinite-volume Gibbs measures for almost every realization of the disorder, yet no natural boundary condition preselects a specific state. The authors analyze two distinct models: one with standard (non-scaled) random fields and another where the random fields are scaled by a factor of $1/\sqrt{n}$ (volume-dependent scaling). The primary objective is to characterize the asymptotic volume dependence of these systems using Aizenman-Wehr (AW) and Newman-Stein (NS) metastates, as well as overlap distributions, and to compare these behaviors across different spin dimensions ($d=1, 2, \ge 3$).

Methodology
The analysis relies on a rigorous asymptotic treatment of finite-volume Gibbs states. The core methodology involves:

1. Mixture Representation: The finite-volume Gibbs states are represented as integral mixtures of shifted microcanonical probability measures, weighted by exponential tilting functions defined on a $2d$-dimensional unit ball.
2. Concentration Analysis: The authors employ Laplace method arguments and concentration estimates to determine the limiting behavior of the mixing measures. This involves analyzing the global maximizing points of the limiting exponential tilting functions.
3. Random Walk Asymptotics: The behavior of the system is linked to the asymptotics of $d$-dimensional random walks (specifically the sums of the random fields $S_n = \sum h(i)$). The authors utilize functional central limit theorems and properties of recurrence (for $d=2$) versus transience (for $d \ge 3$) to describe the limiting distributions of the empirical measures.
4. Metastate Construction:
   • AW-metastate: Derived via a conditioning procedure, representing the probability distribution of infinite-volume states for a typical large volume independent of the specific volume sequence.
   • NS-metastate: Constructed as the limit of empirical measures along volume sequences, capturing the chaotic size dependence.
5. Overlap Analysis: The overlap distribution between two replicas is analyzed using pushforward measures, distinguishing between non-scaled and scaled random field regimes.

Key Contributions and Results

• Cluster Points and Pure States:

  • For $d \ge 3$, the set of cluster points of the Gibbs measures consists solely of a continuum of pure product states $\nu^h_\Omega$, indexed by the direction $\Omega \in S^{d-1}$.
  • For $d=2$, the set of cluster points is strictly larger, containing both the pure product states and non-trivial mixtures of exponentially tilted states ($\bar{\nu}^h_z$). This distinction arises from the recurrence of the underlying random walk in two dimensions.
  • In the non-scaled model, the set of cluster points depends on whether the random field distribution is lattice or continuous (affecting the set of recurrent points of the random walk). Specifically, for $d=2$, the cluster points include non-trivial mixtures of tilted measures, whereas for $d \ge 3$ they do not.

• Metastate Structures:

  • Aizenman-Wehr Metastate: For $d \ge 2$, the AW-metastate is fully supported on the continuum of pure product states $\nu^h_\Omega$. The weight distribution $\rho_P(\Omega)$ depends on the covariance matrix of the random fields. Notably, mixtures (which appear in the cluster points for $d=2$) have zero mass in the AW-metastate, implying that a typical large volume is asymptotically close to a pure state.
  • Newman-Stein Metastate: The NS-metastate exhibits "chaotic size dependence." Its limit is described by an integral over a Brownian motion projected onto the sphere. For $d \ge 2$, the set of cluster points of the NS-metastate includes all possible mixtures of the pure states (i.e., any probability measure on the sphere can be realized as a limit point). This contrasts with the $d=1$ case, where the structure is simpler.

• Overlap Distributions:

  • Non-Scaled Fields: The overlap distribution converges to a trivial (deterministic) value $\delta_{1-d/\beta}$ in law for all $d$. Despite the existence of a continuum of pure states and non-trivial mixtures in the set of cluster points for $d=2$, the limit of the overlap distribution itself is trivial.
  • Scaled Fields ($1/\sqrt{n}$): This regime exhibits spin-glass characteristics. The overlap distribution is non-self-averaging and displays replica symmetry breaking (RSB).
    • For $d=1$, the distribution is discrete (supported on $\pm (r^*)^2$).
    • For $d \ge 2$, the distribution is continuous (atomless) and supported on the whole interval $(r^*)^2[-1, 1]$.
  • Ultrametricity: The paper proves that ultrametricity holds for the overlap distribution if and only if $d=1$. For $d \ge 2$, the overlap distributions do not satisfy ultrametricity.

• Technical Improvements: The authors note that their proofs allow for weaker moment assumptions compared to previous work on $d=1$. Specifically, they require only finite second moments for $d \ge 3$ and finite absolute third moments for $d=2$, relaxing the previous $4+\epsilon$ moment condition.

Significance and Claims
The paper claims to provide a comprehensive comparison of random symmetry breaking in spherical models across dimensions, highlighting the critical role of dimensionality ($d=2$ vs. $d \ge 3$) and the nature of the random field scaling.

• It establishes that while the set of cluster points for Gibbs measures can be complex (containing mixtures in $d=2$), the AW-metastate "washes out" these mixtures, focusing only on pure states.
• It demonstrates that the NS-metastate captures the full richness of the system's chaotic size dependence, revealing that the system can visit any mixture of pure states depending on the volume sequence.
• It clarifies the conditions under which spin-glass features (non-self-averaging, RSB) emerge in ferromagnetic systems with random fields, showing that these features are strictly tied to the scaling of the disorder and the dimensionality of the spin space.
• The work extends the understanding of the random field Ising model (RFIM) and spherical models by providing explicit descriptions of metastates and overlap distributions in higher dimensions, bridging the gap between deterministic coupling studies and random field phenomena.

The authors emphasize that their results are derived from rigorous concentration estimates and asymptotic analysis of random walks, offering a mathematically precise description of the "chaotic size dependence" inherent in these disordered systems.