---
格式版本: 2
标题: "Coupled-cluster molecular properties across the main group that extrapolate beyond training size"
原文链接: "https://arxiv.org/abs/2608.18346"
发布日期: "2026-08-18"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:local:strict_original_body"
发布时间证据: "**\\[v1\\]** Tue, 18 Aug 2026 21:57:02 UTC (1,415 KB)"
发布时间校准原因: "规则确认唯一严格发布时间，来源 local:strict_original_body"
发布时间校准置信度: "high"
发布时间候选数量: 18
发布时间严格候选数量: 6
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-08-20T15:56:34+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-08-20T15:54:04+08:00"
入库时间: "2026-08-20T07:56:34.435Z"
来源平台: "arXiv 学术论文搜索"
搜索渠道: "source_template"
搜索词: "https://arxiv.org/search/?query=inference%20networking&searchtype=all"
匹配关键词:
  - "AI"
相关厂家:
  []
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 5
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "资料为arXiv化学物理论文，主题为耦合簇分子性质与机器学习，与超节点/AI Rack/机柜级AI基础设施无关，属于明显无关内容。"
AI质检模型: "ali-deepseek-v4-flash"
AI质检时间: "2026-08-20T16:02:49+08:00"
AI主题相关性: 0
AI来源权威性: 5
AI新颖性: 0
AI技术细节: 0
AI商业部署信号: 0
AI完整性: 0
AI摘要: "MEHnet-MG 用单个等变网络从一次 B3LYP/def2-SVP 计算预测有效单电子哈密顿量，进而在九种主族元素上得到耦合簇 CCSD(T) 精度的能量、光学带隙、偶极矩、极化率等性质，比各类 DFT 误差降低 3.8–230 倍。"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-07T03:21:22.051Z"
采集批次: "2026年8月20日14点19分32秒"
采集批次ID: "20260820-141932-079"
去重键: "https://arxiv.org/abs/2608.18346"
---

## Physics > Chemical Physics

## Title:Coupled-cluster molecular properties across the main group that extrapolate beyond training size

Authors:[Wenhao He](https://arxiv.org/search/physics?searchtype=author&query=He,+W), [Xu Chen](https://arxiv.org/search/physics?searchtype=author&query=Chen,+X), [Noah Song](https://arxiv.org/search/physics?searchtype=author&query=Song,+N), [Haowei Xu](https://arxiv.org/search/physics?searchtype=author&query=Xu,+H), [Tim S. Hindges](https://arxiv.org/search/physics?searchtype=author&query=Hindges,+T+S), [Bohan Li](https://arxiv.org/search/physics?searchtype=author&query=Li,+B), [Zihan Lin](https://arxiv.org/search/physics?searchtype=author&query=Lin,+Z), [Yu Yao](https://arxiv.org/search/physics?searchtype=author&query=Yao,+Y), [Avetik R. Harutyunyan](https://arxiv.org/search/physics?searchtype=author&query=Harutyunyan,+A+R), [Fang Liu](https://arxiv.org/search/physics?searchtype=author&query=Liu,+F), [Yao Wang](https://arxiv.org/search/physics?searchtype=author&query=Wang,+Y), [Hao Tang](https://arxiv.org/search/physics?searchtype=author&query=Tang,+H), [Ju Li](https://arxiv.org/search/physics?searchtype=author&query=Li,+J)

[View PDF](https://arxiv.org/pdf/2608.18346) [HTML (experimental)](https://arxiv.org/html/2608.18346v1)

> Abstract:Coupled-cluster theory defines the accuracy standard for molecular electronic-structure properties but scales too steeply for routine application, whereas density-functional theory is affordable yet systematically biased. We resolve this trade-off with a single equivariant network, MEHnet-MG, that predicts an effective one-electron Hamiltonian from one inexpensive B3LYP/def2-SVP calculation and derives a broad suite of properties from it (energy, optical gap, dipole, quadrupole, polarizability, Mulliken atomic charges, and Mayer bond orders) at coupled-cluster accuracy across nine main-group elements, including the under-served phosphorus, sulfur, and chlorine chemistries. The model is trained on a new in-house dataset of multi-property labels computed at the CCSD(T) level for all nine elements. On a held-out test set, it reduces the error of every property by a factor of 3.8 to 230 relative to semi-local, hybrid, and double-hybrid DFT (referenced to composite CCSD(T)/cc-pVTZ; Methods), while adding only ~25 ms wall time per molecule, delivering coupled-cluster-quality predictions at the cost of a single DFT calculation. Critically, deriving every property from a predicted Hamiltonian rather than pooling per-atom features builds the correct size-scaling into the model architecture: on pi-conjugated oligothiophenes it matches finite-field CCSD polarizability and the EOM-CCSD optical gap to ~2% at the largest sizes where those references remain affordable (44 and 37 atoms, where a single CCSD field point already costs ~500x the model's entire inference) and extrapolates the corrected trends to 58-atom chains, a regime where pooling-based architectures fail by construction. Accurate extrapolation is therefore set by the model's inductive bias rather than by the training data.

| Comments: |  |
| --- | --- |
| Subjects: | Chemical Physics (physics.chem-ph); Materials Science (cond-mat.mtrl-sci); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Computational Physics (physics.comp-ph) |
| Cite as: | [arXiv:2608.18346](https://arxiv.org/abs/2608.18346) \[physics.chem-ph\] |
|  | (or [arXiv:2608.18346v1](https://arxiv.org/abs/2608.18346v1) \[physics.chem-ph\] for this version) |
|  | [https://doi.org/10.48550/arXiv.2608.18346](https://doi.org/10.48550/arXiv.2608.18346) |

## Submission history

From: Ju Li \[[view email](https://arxiv.org/show-email/59aecaac/2608.18346)\]  
**\[v1\]** Tue, 18 Aug 2026 21:57:02 UTC (1,415 KB)

[Which authors of this paper are endorsers?](https://arxiv.org/auth/show-endorsers/2608.18346) | Disable MathJax ([What is MathJax?](https://info.arxiv.org/help/mathjax.html))