arXiv subjects

Mingze Zhao

Publications and source records attributed to Mingze Zhao.

2 recordsLinked to original sources

arXiv2026

Topological origin of the Gutenberg-Richter law: Dynamic decoupling of fault cascades and energy dissipation

The Gutenberg-Richter (G-R) law is seismology's most robust empirical regularity, yet a first-principles derivation of its slope-the b-value-has remained elusive. Here, we present an analytical theory based on a finite hierarchical fault network composed of velocity-weakening asperities separated by velocity-strengthening creeping zones. By coupling network topology with fracture mechanics, we derive a generalized closed-form expression of b-value,$$b=\frac{3}{3+2γ}\frac{\log_{10}k}{\log_{10}q}$$, where k is the threshold branching ratio and q is the area amplification factor. Crucially, we reveal a dynamic decoupling between k and q governed by energy dissipation. The empirical universality of b=1 naturally emerges as the fault network self-organizes into a critical attractor (k=q) that maximizes two-dimensional spatial filling. Furthermore, grounded in Rate-and-State Friction rheology, we demonstrate that foreshock acceleration (inverse-Omori) and aftershock decay (Omori-Utsu) are dual manifestations of the similar hierarchical cascade. In the mean-field limit, these temporal exponents are decoupled from static fracture energy, revealing a profound orthogonality between time-dependent friction and static energy release. By unifying the G-R law, Bath's law, and the foreshock-aftershock asymmetry, this framework might provide a physics-based pathway for operational earthquake forecasting.

cond-mat.stat-mech↗

arXiv2026

Unifying Temporal and Structural Credit Assignment in LLM-Based Multi-Agent Prompt Optimization

While Multi-Agent Systems (MAS) empower Large Language Models to tackle complex reasoning tasks through collaborative interaction, optimizing their dynamics remains a formidable challenge due to the discrete, non-differentiable nature of the computation graph and the sparsity of global supervisory signals. Existing black-box optimizers struggle to attribute trajectory-level failure to specific local components, resulting in inefficient, high-variance exploration. We argue that tractable MAS optimization needs structural inductive biases to disentangle error signals. We propose temporal and structural credit assignment, which decomposes the objective along two axes: (i) temporal credit, using state-space bottlenecks to identify critical rounds, and (ii) structural credit, using stationary role policies to isolate agent contributions. Leveraging these decomposed signals, we introduce a discrete, verbalized block coordinate descent algorithm for iterative refinement. Rather than indiscriminate global updates, it alternates between optimizing role prompts and aggregation protocols, using LLM-generated "proxy gradients" to target only the identified weak links. Across diverse reasoning benchmarks, our approach substantially reduces query complexity while improving performance, providing a principled and interpretable path toward self-improving MAS.

cs.MA↗