Tag
The paper evaluates whether LLM agents can reliably construct complete executable JSON specifications for automated PDE simulation setup (geometry, meshing, solving, export), showing a full workflow with scientific checks and clarification achieves 98.3% correct specification outcomes versus 92.5% schema-assisted single-shot and 82.5% model-only baselines.
The paper proposes AgenticTuner, an agentic pipeline using an LLM as the optimizer to calibrate cosmological simulations by leveraging human-readable priors like codebases and documentation, achieving competitive calibration performance at a fraction of the CPU cost of Bayesian optimization.
Keplearn 是一个用 Rust 编写的确定性符号回归引擎，可从表格数据中通过最佳优先递归归约在数秒内恢复闭式方程（如牛顿引力定律），并输出带 r² 评分的 JSON 候选方程，支持零依赖单二进制跨平台分发。
Lanyon AI 从 11 月 1 日起向早期客户和测试者推出 Lanyon 的预发布版本，并在旧金山和芝加哥同时举办公开演示，致力于构建可形式化验证的科学模型。
The paper introduces Phronix, a capability-centric learning paradigm that lets LLM-based agent harnesses acquire, validate, and reuse capabilities from their own executions via a predictor-validator-corrector loop, achieving 93-95% first-attempt success on recurring tasks and boosting cheaper models from ~28% to ~90% on SWE-bench.
NVIDIA cuDSS, a CUDA-X library for GPU-accelerated sparse linear system solving, is now generally available, with adoption by Applied Materials, COMSOL, Cadence, Dassault Systèmes, Siemens, and Synopsys to speed up engineering and scientific simulations by up to 10x.
Fortran 语言迎来 70 周年，剑桥大学计算气候科学研究所（ICCS）将于 2026 年 12 月 1 日在曼彻斯特科学与工业博物馆举办纪念活动，包含主题演讲、闪电演讲和小组讨论，庆祝其在核工程、天气气候预报和材料科学等领域的贡献。
Open-Science 是 AIPOCH 推出的开源、本地优先 AI 科研工作台，结合智能体工作流、Python 与 R 执行以及可追溯的研究产物，所有结果都保留输入、代码与执行证据，便于复现与对比工作流。
NumPy 2.5 introduces a vectorized implementation of np.searchsorted that achieves up to 25× speedup using batching and cache-friendly techniques, benefiting the broader Python scientific ecosystem including SciPy and scikit-learn.
该论文提出 Gen-PINNs，一种将生成对抗网络与物理信息神经网络结合的确定性残差-对抗框架，用于求解具有尖锐波前和刚性特征的偏微分方程，在 Burgers、Allen-Cahn 和 Kuramoto-Sivashinsky 方程上相较标准 PINNs 显著提升了精度和收敛性。
FlashSinkhorn 2 introduces a two-stage block-sparse GPU solver for entropic optimal transport that scales to 1.34e8-particle point clouds, solving them on a single A100 to a marginal residual below 0.01 in under 2.5 hours — claimed to be the largest discrete EOT problem solved at this accuracy within hours. An open-source implementation is released.
A paper showing an LLM agent rewriting the Sella molecular geometry optimizer to cut force-evaluation counts, producing AutoSella, a family of optimizers that reduces r2SCAN-3c DFT force calls to 40.2–77.2% of Sella's while matching energy reduction on held-out molecules.
Rice University researchers introduce CompMat-Bench, a 94-task benchmark derived from published computational materials studies that evaluates AI agents on preparing inputs and analyzing outputs of expensive simulations, avoiding costly reruns during grading. Agents based on three LLMs achieve 66.0–90.4% pass rates with full guidance, but performance drops under longer workflows or reduced methodological hints, with most failures stemming from scientific errors rather than software usage.
Rules to Tools (R2T) studies whether giving LLM coding agents ready-to-run executable checks of written scientific requirements improves program repair on SciCode-style tasks. Results are mixed and task-dependent: the tool group slightly outperforms text-based checks in some cohorts (29/30 vs 26/30 complete repair), while others tie, and checks can lower agent-side model output costs but raise CPU use.
The paper introduces Auto-Diagnosis and Skill Discovery (ADSD), a framework that teaches AI agents to diagnose why numerical solvers perform poorly and then discover reusable solver skills, achieving up to 71x error reduction on power flow benchmarks and transferring to unseen grid topologies.
提出软件在环重构（software-in-the-loop reconstruction）的自监督框架，从现有科学软件工作流中自动生成可执行参考行为与验证目标，用于训练终端智能体；使用 Qwen3.8-Max 生成训练数据并对 Qwen3.8-27B 微调，使 Terminal-Bench2 平均性能从 47.94% 提升至 53.56%。
Harvard physicist Matthew Schwartz, in an Anthropic Science Blog guest post, argues that current LLMs face an "impedance mismatch" with scientific research and describes BootLoops, a Claude-built toolkit for exact quantitative calculations applied across fields like ecology and population genetics.
A historical paper examining how Los Alamos National Laboratory adopted the ENIAC computer for nuclear Monte Carlo simulations in 1947–1948, analyzing the decision-making and early computational practices that shaped modern scientific computing.
OSWorld-Science introduces a benchmark and evaluation environment for computer-using VLM agents on scientific software, covering 146 expert-designed tasks across domains like molecular drawing, pathology imaging, statistics, and physical simulation with artifact-based evaluators and a special agent harness.
论文提出 Rules to Tools (R2T)，为科学计算中的 LLM 代码修复代理提供可执行的公开需求检查工具，并在 SciCode 基准上对比纯文本说明与可调用检查的修复效果与成本。