Qingyu Shao | arXiv Science

arXiv2026

Model Structures Arising from Extendable Cotorsion Pairs

The aim of this paper is to construct exact model structures from so called extendable cotorsion pairs. Given a hereditary Hovey triple $(\mathcal{C}, \mathcal{W}, \mathcal{F})$ in a weakly idempotent complete exact category with enough projectives and injectives. If one of the cotorsion pairs $(\mathcal{C}\cap\mathcal{W}, \mathcal{F})$ and $(\mathcal{C}, \mathcal{W}\cap\mathcal{F})$ is extendable, then there is a chain of hereditary Hovey triples whose corresponding homotopy categories coincide. As applications, we obtain a new description of the $Q$-shaped derived categories introduced by Holm and Jørgensen. We can also interpret the Krause's recollement in terms of ``$n$-dimensional'' homotopy categories. Finally, we have two approaches to get ``$n$-dimensional'' hereditary Hovey triples, which are proved to coincide, in the category Rep$(Q,\mathcal{A})$ of all representations of a rooted quiver $Q$ with values in an abelian category $\mathcal{A}$.

math.CT↗

arXiv2026

ResMAS: Resilience Optimization in LLM-based Multi-agent Systems

Large Language Model-based Multi-Agent Systems (LLM-based MAS), where multiple LLM agents collaborate to solve complex tasks, have shown impressive performance in many areas. However, MAS are typically distributed across different devices or environments, making them vulnerable to perturbations such as agent failures. While existing works have studied the adversarial attacks and corresponding defense strategies, they mainly focus on reactively detecting and mitigating attacks after they occur rather than proactively designing inherently resilient systems. In this work, we study the resilience of LLM-based MAS under perturbations and find that both the communication topology and prompt design significantly influence system resilience. Motivated by these findings, we propose ResMAS: a two-stage framework for enhancing MAS resilience. First, we train a reward model to predict the MAS's resilience, based on which we train a topology generator to automatically design resilient topology for specific tasks through reinforcement learning. Second, we introduce a topology-aware prompt optimization method that refines each agent's prompt based on its connections and interactions with other agents. Extensive experiments across a range of tasks show that our approach substantially improves MAS resilience under various constraints. Moreover, our framework demonstrates strong generalization ability to new tasks and models, highlighting its potential for building resilient MASs.

cs.AI↗