Aligning incentives and resilience: joint node selection and resource allocation in the lightning network
提出一个基于深度强化学习的框架，解决闪电网络中节点选择与资源分配的联合优化问题，发现个体收益最大化可与去中心化目标一致。
Abstract The lightning network (LN) addresses Bitcoin’s scalability as a second-layer solution. While payment channel networks (PCNs) incentivize participation through profit opportunities, revenue-driven behaviour risks centralization, creating hub nodes that undermine decentralization and privacy. Current research inadequately models resource allocation and lacks realistic simulations of LN’s routing dynamics, limiting practical insights. This paper introduces a deep reinforcement learning (DRL) framework, enhanced by transformer-based architectures, to solve the Joint Combinatorial Node Selection and Resource Allocation (JCNSRA) problem. We refine an existing simulation environment by integrating enhanced routing modules, better aligning it with real-world LN behaviour and JCNSRA requirements. Our model outperforms baseline methods and heuristics across diverse network settings. To evaluate decentralization, we deploy revenue-driven agents in localized simulations and analyze network evolution using betweenness and closeness centrality, entropy, inequality measures, and modularity analysis. Results demonstrate that individual revenue-maximization incentives can align with LN’s decentralization goals. Our rational RL agents promote a more equitable and decentralized network structure. Overall, this work highlights the feasibility of incentive-compatible solutions to balance profitability and decentralization in PCNs, offering actionable insights for sustainable network growth.