跳转至

文章背景与核心概要

本文介绍了名为“The Station”的开放世界多智能体环境,旨在实现自主数学 discovery(发现)。在缺乏中央协调器或硬编码脚本化流程的情况下,来自不同模型家族的AI智能体能够在该环境中自主追求共享的研究目标:自主选择研究方向、运行实验、开展合作并维护共享的科学文献库。

通过对AlphaEvolve目录中的12个构造问题以及另外2个案例研究进行测试,该多智能体系统在5个不同领域成功取得了超越先前文献的创新成果,包括:有限域Kakeya集的全新无限族、11维中全新的精确604点亲吻数配置(kissing configuration)、离散化Kakeya针问题与符号不确定性问题的新性能记录、大幅改进的埃尔德什(Erdős)最小重叠问题下界,以及Book Ramsey数字的全新无限族。

尤为重要的是,智能体产出的不仅是原始的数值构造,还生成了严格的定理及解释这些构造运作原理的分析报告,从而产出了人类数学家能够轻松在其基础上继续推进的可解释性输出。作者已公开了所有的原始智能体对话记录、证明以及验证代码。


Autonomous Mathematical Discovery in an Open-World Multi-Agent Environment

Authors: Stephen Chung, Wenyu Du, William J. Wesley
Submitted on: August 24, 2026
arXiv: 2608.23691 [cs.AI]
Subjects: Artificial Intelligence (cs.AI); Discrete Mathematics (cs.DM); Multiagent Systems (cs.MA)


📌 Summary

本文介绍了 the Station,这是一个专为自主数学发现而设计的开放世界多智能体环境。在没有中央协调器或脚本化流程的情况下,来自不同模型家族的 AI 智能体可以自主追求共享的研究目标:选择研究方向、运行实验、进行协作并策划共享的科学文献。

跨越 AlphaEvolve 目录中的 12 个构造问题和两个额外的案例研究,该多智能体系统在五个不同领域取得了相较于先前文献具有创新性的结果: 1. 有限域 Kakeya 集的一个新的无限族。 2. 11 维中新的精确 604 点亲吻数配置。 3. 离散化 Kakeya 针问题和符号不确定性问题的新纪录。 4. 埃尔德什(Erdős)最小重叠问题的大幅改进下界。 5. Book Ramsey 数的全新无限族。

至关重要的是,智能体不仅交付了原始的数值构造,还生成了严格的定理和分析,解释了这些构造是如何运作的,从而产生了人类数学家可以轻松在其基础上进行构建的可解释输出。作者已公开了所有原始智能体对话、证明和验证代码。

This paper introduces the Station, an open-world multi-agent environment designed for autonomous mathematical discovery. Operating without a central coordinator or scripted pipeline, AI agents from diverse model families autonomously pursue shared research goals: choosing directions, running experiments, collaborating, and curating a shared scientific literature.

Across 12 construction problems from the AlphaEvolve catalogue and two additional case studies, the multi-agent system successfully achieved novel results superior to prior literature in five distinct areas: 1. A new infinite family of finite-field Kakeya sets. 2. New exact 604-point kissing configurations in dimension 11. 3. New performance records for the discretized Kakeya needle and sign uncertainty problems. 4. A substantially improved lower bound for Erdős's minimum-overlap problem. 5. Novel infinite families for Book Ramsey numbers.

Crucially, the agents delivered more than raw numerical constructions—they generated rigorous theorems and analytical proofs explaining how those constructions function, yielding interpretable outputs that human mathematicians can easily build upon. The authors have publicly released all raw agent dialogues, proofs, and verification code.


📋 Abstract

我们在 Station 中研究了自主数学发现,这是一个开放世界的多智能体环境,其中来自不同模型家族的 AI 智能体在没有中央协调器或脚本化管道的情况下追求共同的研究目标。智能体选择自己的研究方向、进行实验、进行协作并构建共享的科学文献。

跨越 AlphaEvolve 目录中的 12 个构造问题和两个额外的案例研究,Station 在五个问题上获得了相对于先前文献具有创新性的结果:有限域 Kakeya 集的新无限族、11 维中的新精确 604 点亲吻数配置、离散化 Kakeya 针和符号不确定性问题的新记录,以及埃尔德什最小重叠问题的大幅改进下界。智能体还发现了 Book Ramsey 数的新无限族。

重要的是,智能体不仅产生了数值构造,还产生了定理和分析,解释了这些构造是如何工作的,这使得结果更具可解释性,并使数学家更容易在其上进行构建。我们发布了所有原始智能体对话、证明和验证代码,为这些发现是如何出现的提供了透明的记录。

We study autonomous mathematical discovery in the Station, an open-world multi-agent environment in which AI agents from different model families pursue a shared research goal without a central coordinator or scripted pipeline. Agents choose their own research directions, conduct experiments, collaborate, and build a shared scientific literature.

Across 12 construction problems from the AlphaEvolve catalogue and two additional case studies, the Station obtained results novel relative to the prior literature on five problems: a new infinite family of finite-field Kakeya sets, new exact 604-point kissing configurations in dimension 11, new records for the discretized Kakeya needle and sign uncertainty problems, and a substantially improved lower bound for Erdős's minimum-overlap problem. Agents also discovered novel infinite families for Book Ramsey numbers.

Importantly, the agents produced not only numerical constructions but also theorems and analyses explaining how those constructions work, making the results more interpretable and easier for mathematicians to build upon. We release all raw agent dialogues, proofs, and verification code, providing a transparent record of how these discoveries emerged.


📊 Metadata & Classifications

  • Comments: 38 页,12 张图表,3 个表格。
  • MSC Classes: 68T42(主要);68T20, 05C55, 52C17(次要)
  • ACM Classes: I.2.11; I.2.3
  • DOI: 10.48550/arXiv.2608.23691
  • Comments: 38 pages, 12 figures, 3 tables.
  • MSC Classes: 68T42 (Primary); 68T20, 05C55, 52C17 (Secondary)
  • ACM Classes: I.2.11; I.2.3
  • DOI: 10.48550/arXiv.2608.23691


license icon