跳转至

从实验与模拟的不一致中发现物理机制

文章背景与核心概要

科学突破往往源自观测与预测之间的矛盾。随着计算与机器学习对化学空间的广泛探索，实验与模拟之间的不一致性在大规模范围内显现，而追溯这些不一致背后的物理机制传统上仍高度依赖专家的主导。本文介绍了 可解释密度泛函理论（XDFT），这是一个能够将这一发现过程转化为可执行搜索的自进化智能体。XDFT 将候选机制形式化为可执行假设，对照实验数据检验其结果，并提炼其内部先验。

在针对 112 个标准计算错误地将半导体预测为金属的案例测试中，XDFT 在单 GPU 环境下成功解析了 105 个案例，并给出了有证据支持的机制。值得注意的是，在评估了仅 60 个案例后，正确解析机制在留出测试集的前三个假设中的排名达到了 80%——与初始 7% 的专家先验基线相比有了巨大的提升。这些发现证明，模拟不一致性可以作为自动化科学智能体的强大起点。

Discovering Physical Mechanisms from Experiment-Simulation Mismatches

Discovering Physical Mechanisms from Experiment-Simulation Mismatches

Summary

Summary

Scientific breakthroughs often emerge from discrepancies between observation and prediction. To address how experiment-simulation mismatches are traditionally handled via expert-led tracing, the authors introduce eXplainable DFT (XDFT). XDFT is a self-evolving agent that transforms the discovery process into an executable search by formalizing candidate mechanisms as hypotheses, testing their outcomes against experimental data, and refining its internal priors.

Tested on 112 cases where standard calculations incorrectly predicted a metal instead of a semiconductor, XDFT successfully resolved 105 cases with evidence-supported mechanisms within a single-GPU setup. Notably, after evaluating just 60 cases, the correct resolving mechanism ranked within the top three hypotheses for 80% of held-out cases—a massive improvement over the initial 7% baseline expert prior. These findings demonstrate that simulation mismatches can serve as powerful starting points for automated scientific agents.

Article Metadata

Article Metadata

  • arXiv Identifier: arXiv:2604.26703 [cond-mat.mtrl-sci]
  • Subject Categories:
  • Materials Science (cond-mat.mtrl-sci)
  • Artificial Intelligence (cs.AI)
  • Computational Physics (physics.comp-ph)
  • Authors: Yue Li, Penghui Yang, Yushan Xiao, Zhonghan Zhang, Jianguo Huang, Yuhao Lu, Cuntai Guan, Bo An, Bijun Tang, Zheng Liu
  • Timeline:
  • Submitted: 29 April 2026
  • Revised: 18 Aug 2026 (v2)
  • Format: 6 pages, 4 figures

Abstract

Abstract

Scientific discovery often begins where observation and prediction disagree. As computation and machine learning survey chemical space, experiment-simulation mismatches are exposed at scale, while tracing them to physical mechanisms remains expert-led. Here we present eXplainable DFT (XDFT), a self-evolving agent that turns this process into an executable search. XDFT formalizes candidate mechanisms as executable hypotheses, adjudicates their consequences against experiment and distils trajectories into priors for later searches. This couples a solving loop from mismatch to mechanism with a learning loop through which solving changes the solver. Across 112 source-audited cases in which standard calculations predict a metal whereas experiments find a semiconductor, XDFT resolved 105 with evidence-supported mechanisms within a single-GPU envelope. After 60 cases, the resolving mechanism ranked among the first three hypotheses for 80% of held-out cases, up from 7% under the initial expert prior. XDFT also returned evidence-graded mechanisms for seven expert-curated questions about physical mechanisms. These results establish experiment-simulation mismatches as tractable starting points for scientific agents that discover physical mechanisms while learning how to find the next.

Access & Resources

Access & Resources