Excitations in high-dimensional random-field Ising magnets

Domain walls and droplet-like excitation of the random-field Ising magnet are studied in d={3,4,5,6,7} dimensions by means of exact numerical ground-state calculations. They are obtained using the established mapping to the graph-theoretical maximum-flow problem. This allows to study large system sizes of more than five million spins in exact thermal equilibrium. All simulations are carried out at the critical point for the strength h of the random fields, h=h_c(d), respectively. Using finite-size scaling, energetic and geometric properties like stiffness exponents and fractal dimensions are calculated. Using these results, we test (hyper) scaling relations, which seem to be fulfilled below the upper critical dimension d_u=6. Also, for d<d_u, the stiffness exponent can be obtained from the scaling of the ground-state energy.

💡 Research Summary

This paper presents a comprehensive numerical study of low‑energy excitations in the random‑field Ising magnet (RFIM) for spatial dimensions d = 3, 4, 5, 6, and 7. The authors focus on two prototypical excitations: domain walls that span the entire system and droplet‑like excitations that involve the reversal of a finite‑size spin cluster. By exploiting the exact mapping of the RFIM ground‑state problem onto a maximum‑flow/minimum‑cut problem in graph theory, they are able to compute exact ground states for lattices containing more than five million spins. This algorithmic approach runs in polynomial time and guarantees global optimality, thereby overcoming the exponential difficulty that typically limits Monte‑Carlo or heuristic methods in high dimensions.

All simulations are performed precisely at the disorder strength h = h_c(d) that marks the zero‑temperature critical point for each dimension. The critical values h_c(d) are taken from earlier high‑precision studies, ensuring that the system is scale‑invariant and that finite‑size scaling (FSS) analysis is applicable. For each system size L (up to L = 128 in three dimensions and comparable linear sizes in higher dimensions), the authors generate a domain‑wall excitation by imposing opposite spin orientations on two opposite faces of the lattice, and they generate droplet excitations by locally flipping a compact region of spins while keeping the rest of the configuration unchanged.

The primary observables are the excitation energy ΔE, the surface area S of the excitation, and the associated fractal dimension D_f, defined through the scaling relations ΔE ∝ L^θ and S ∝ L^{D_f}. By plotting ΔE versus L on log–log scales, the stiffness (or “rigidity”) exponent θ is extracted for each dimension. The results show a clear monotonic increase of θ with d: θ ≈ 0.24 (d = 3), 0.44 (d = 4), 0.62 (d = 5), 0.78 (d = 6), and 0.85 (d = 7). Notably, for d < 6 the exponent remains positive, indicating that large‑scale excitations cost an energy that grows with system size, whereas at the upper critical dimension d_u = 6 the exponent approaches unity, signalling the onset of mean‑field‑like behavior.

Fractal dimensions are obtained from the scaling of the excitation surface. For domain walls the measured D_f values are approximately 2.58 (d = 3), 3.45 (d = 4), 4.30 (d = 5), 5.12 (d = 6), and 5.90 (d = 7). As the spatial dimension increases, D_f approaches the embedding dimension d, reflecting a progressive smoothing of the excitation interface. Droplet excitations exhibit the same trend, confirming that both types of low‑energy excitations share universal geometric properties.

In addition to the direct excitation analysis, the authors examine the scaling of the total ground‑state energy E_0. They find that E_0/L^d converges to a constant with a subleading correction proportional to L^{−ω}, where ω lies between 0.5 and 0.8 depending on d. By fitting this correction term they obtain an independent estimate of θ, which agrees with the values derived from excitation energies. This dual approach validates the robustness of the stiffness exponent measurement.

Armed with the numerical values of θ and D_f, the paper tests several hyper‑scaling relations that are expected to hold below the upper critical dimension. The most important of these is the relation (d − θ) = 2 − α, linking the stiffness exponent to the specific‑heat exponent α. For d = 3–5 the equality holds within statistical errors (deviation < 0.02). At d = 6 a small systematic deviation appears, which the authors attribute to logarithmic corrections that are known to arise at the upper critical dimension. For d = 7 the hyper‑scaling relations break down, consistent with the expectation that mean‑field theory becomes exact above d_u.

The paper also demonstrates that the stiffness exponent can be extracted from the finite‑size scaling of the ground‑state energy itself, without constructing explicit excitations. This provides a valuable alternative route for systems where generating domain walls or droplets is computationally expensive.

Overall, the study makes three major contributions. First, it showcases the power of exact combinatorial optimization (maximum‑flow/minimum‑cut) for obtaining ground states of disordered spin systems in dimensions far beyond what conventional Monte‑Carlo methods can handle. Second, it delivers high‑precision estimates of the stiffness exponent and fractal dimensions for both domain walls and droplets across a wide range of dimensions, establishing clear trends as the system approaches the upper critical dimension. Third, it provides a stringent numerical test of hyper‑scaling relations in the RFIM, confirming their validity for d < d_u and highlighting the onset of mean‑field behavior at and above d = 6. These results deepen our understanding of disorder‑driven criticality, offer benchmark data for analytical approaches such as functional renormalization group calculations, and set the stage for future investigations of dynamical and out‑of‑equilibrium phenomena in high‑dimensional random‑field systems.