Predictive Advantage, Regime Dependence, and Direction-Specific Causal Control in Kinetic Ising Dynamics
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
We study how the choice of description affects predictive and causal information about future evolution in the two-dimensional kinetic Ising model. We report four empirical results together with a predictor–handle dissociation result. Fine descriptions based on locally exposed and junction-type defects outperform a block-variance coarse rival for predicting future domain-wall change in the ordered phase; the best organism is E≥3 + J, with predictive advantage ΔR² = +0.415 (95% CI [+0.389, +0.443]). (Axis 2, regime crossover) This advantage reverses sign in the disordered phase (+0.183 ordered, −0.064 disordered), showing that the fine-over-coarse advantage is regime-dependent. (Axis 3, target specificity) Under a fine-over-coarse ΔR² criterion, future domain-wall change ΔW is the target most specifically favoured by the fine description (ΔR² = +0.415); this is not merely the most predictable target in absolute terms. (Axis 4, causal control) Biasing the Glauber generator toward wall-reducing flips at exposed sites reduces future wall loss, while biasing against the same flips increases it; this direction-specific sign flip holds across all tested perturbation strengths (4/4). (Dissociation) The best predictor of ΔW—the coarse macro-state [W₀, Bmag] (R² = 0.378)—differs from the causal handle family (exposed-site structure, R² = 0.181), with gap +0.197. The best predictor and the best causal handle are empirically distinct within the same system at the same temperature.
Fields of Study & Subjects
Theoretical and Computational Physics
Cellular Automata and Applications
Complex Systems and Dynamics
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 26 March 2026
- Language
- EN
- Document type
- Preprint