谢 丹  教授学术报告

Title: Leveraging AI Agents for Research in Mathematics and Physics

Speaker: Dan Xie (谢丹)

Affiliation: Tsinghua University（清华大学）

Time: 16:00-17:00, Thursday, 25th June, 2026 (UTC+8, Beijing Time)

Venue: Room 1502, Shing-Tung Yau Center, Sipailou Campus of Southeast University（东南大学四牌楼校区丘成桐中心1502室）

Inviter: Qiang Wen（文强）、Zhijin Li（李志金）

Abstract

This talk will explore the foundational mechanisms of coding agents, focusing on their three core components: large language models (LLMs), tools, and user interaction. Using the Pi coding agent as a primary example, I will illustrate how these elements interact to enable automated problem-solving and code generation. Building on this foundation, I will discuss strategies for applying AI agents to research in mathematics and physics, highlighting the role of specialized skills, prompt engineering, and the use of subagents to tackle complex scientific challenges. Finally, I will address the current limitations of AI agents in this context—including issues of reasoning reliability, hallucination, symbolic manipulation, interpretability, and the difficulty of verifying novel mathematical or physical insights—and outline open challenges that must be overcome before these systems can serve as truly autonomous research partners.

Speaker

谢丹，现为清华大学数学科学系长聘教授。2005年于浙江大学获学士学位，2011年于美国德克萨斯农工大学获博士学位，2011至2018年先后在美国普林斯顿高等研究院和哈佛大学从事博士后研究工作，2018年9月入职清华大学。

他的研究方向为数学物理，主要研究超对称场论和超弦理论中各种物理和数学问题，近年来主要关注超对称场论和代数几何，低维拓扑，以及表示论的关系。另一研究方向为机器学习的数学原理，以及物理方法在机器学习中的应用。