[Paper Review] Scaling, domains, and states in the four-dimensional random field Ising magnet

A. Alan Middleton|arXiv (Cornell University)|Aug 9, 2002

Theoretical and Computational Physics5 citations

TL;DR

This study investigates the zero-temperature four-dimensional Gaussian random field Ising model using exact ground state computations on systems up to $64^4$ spins. It confirms conventional scaling relations, finds a single ferromagnetic-to-paramagnetic transition, and reveals that frozen spins of both signs percolate below the critical disorder, fundamentally altering domain wall interpretation compared to three dimensions.

ABSTRACT

The four dimensional Gaussian random field Ising magnet is investigated numerically at zero temperature, using samples up to size $64^4$, to test scaling theories and to investigate the nature of domain walls and the thermodynamic limit. As the magnetization exponent $β$ is more easily distinguishable from zero in four dimensions than in three dimensions, these results provide a useful test of conventional scaling theories. Results are presented for the critical behavior of the heat capacity, magnetization, and stiffness. The fractal dimensions of the domain walls at criticality are estimated. A notable difference from three dimensions is the structure of the spin domains: frozen spins of both signs percolate at a disorder magnitude less than the value at the ferromagnetic to paramagnetic transition. Hence, in the vicinity of the transition, there are two percolating clusters of opposite spins that are fixed under any boundary conditions. This structure changes the interpretation of the domain walls for the four dimensional case. The scaling of the effect of boundary conditions on the interior spin configuration is found to be consistent with the domain wall dimension. There is no evidence of a glassy phase: there appears to be a single transition from two ferromagnetic states to a single paramagnetic state, as in three dimensions. The slowing down of the ground state algorithm is also used to study this model and the links between combinatorial optimization and critical behavior.

Motivation & Objective

• To test conventional scaling theories in the four-dimensional random field Ising model (RFIM) at zero temperature.
• To investigate the nature of domain walls and spin configurations in the ferromagnetic phase, especially the role of frozen spins.
• To determine critical exponents such as $\beta/\nu$, $\theta$, and $\alpha$ using finite-size scaling and ground state energy analysis.
• To examine the impact of boundary conditions on spin configurations and their relation to domain wall dimension.
• To assess the presence or absence of a glassy phase and verify the transition type via algorithmic slowing down near criticality.

Proposed method

• Numerical simulations of the four-dimensional RFIM with Gaussian-distributed random fields ($h_i$) and $J=1$, using zero-temperature exact ground state calculations.
• Employment of a max-flow algorithm to compute ground states, enabling precise determination of energy, magnetization, and stiffness.
• Finite-size scaling analysis of magnetization, energy, and boundary condition effects to extract critical exponents $\beta/\nu$, $\theta$, and $\alpha$.
• Use of the Binder parameter and peak analysis of algorithmic operations (e.g., relabel operations) to locate the critical disorder $h_c$ with high precision.
• Analysis of spin domain structure via percolation of frozen spins (invariant under all boundary conditions) and minority spin clusters.
• Scaling collapse of probability distributions of domain wall-related observables to estimate fractal dimension $d_I$ and confirm consistency with $\theta$ and $\nu$.

Experimental results

Research questions

• RQ1 Does the four-dimensional RFIM exhibit a single second-order phase transition from a ferromagnetic to a paramagnetic state, consistent with conventional scaling?
• RQ2 How do the critical exponents $\beta/\nu$, $\theta$, and $\alpha$ in four dimensions compare to theoretical predictions and three-dimensional results?
• RQ3 What is the nature of spin domain structure in the ferromagnetic phase, particularly regarding percolation of frozen spins of both signs?
• RQ4 How does the domain wall structure in four dimensions differ from that in three dimensions, especially in terms of connectivity and fractal dimension?
• RQ5 To what extent does algorithmic slowing down near the critical point reflect the physical correlation length and support the scaling hypothesis?

Key findings

• The critical disorder for the ferromagnetic-to-paramagnetic transition is located at $h_c = 4.179(2)$, with $\beta/\nu = 0.19(3)$, indicating a measurable magnetization exponent in four dimensions.
• The stiffness exponent is found to be $\theta = 1.82 \pm 0.07$, significantly different from $d/2 = 2$ and consistent with conventional bounds.
• The heat capacity exponent is estimated as $\alpha = 0.26 \pm 0.05$, distinct from $\alpha = 0$, supporting the disorder variant of Widom scaling.
• Frozen spins of both signs percolate in the ferromagnetic phase at $h_p^f = 3.680(5)$, with minority spins percolating at $h_p^m$, indicating two coexisting percolating clusters.
• The fractal dimension of domain walls at criticality is estimated as $d_I = 3.2$, consistent with scaling collapse of boundary condition effects.
• Algorithmic slowing down, measured via peak in relabel operations per spin, scales linearly with system size $L$, supporting the physical correlation length scaling hypothesis near $h_c$.