A quantum chemistry dataset containing ground-state and conical-intersection structures of 260k molecules
Jiahui Zhang, Yifei Zhu, Chuqiao Feng, Yingjin Ma, Chao Xu, Zhenggang Lan · 2026
Read A quantum chemistry dataset containing ground-state and conical-intersection structures of 260k molecules by Jiahui Zhang, Yifei Zhu, Chuqiao Feng, Yingjin Ma, Chao Xu, Zhenggang Lan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-14287.txt.
Introduction
Zhang and five co-authors build QCDGE-CI, a quantum chemistry dataset of ground-state and minimum-energy conical-intersection forms for over 259k small molecules. Conical intersections matter greatly for photoinduced reactions, but comprehensive sets remain rare. Minimum-energy conical-intersection geometries were done with OM2/MRCI and checked through systematic benchmarks with a larger active space and CASSCF calculations. Benchmarks support the geometries' reliability, and cheminformatics checks reveal broad chemical diversity. The dataset supports machine-learning studies of excited-state chemistry, including nonadiabatic events.
Catalog record
| CREATOR | Jiahui Zhang, Yifei Zhu, Chuqiao Feng, Yingjin Ma, Chao Xu, Zhenggang Lan |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28849 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.14287
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A quantum chemistry dataset containing ground-state and conical-intersection structures of 260k molecules Jiahui Zhang, Yifei Zhu, Chuqiao Feng, Yingjin Ma, Chao Xu, Zhenggang Lan Abstract Conical intersections play central roles in photoinduced reactions. However, comprehensive conical-intersection datasets that could advance our understanding of excited-state reaction processes remain scarce. To address this gap, we constructed a quantum chemistry dataset containing ground-state and conical-intersection structures of small molecules (up to ten heavy atoms: C, N, O, F). Ground-state geometries were optimized at the semi-empirical OM2 level, with single-point energies calculated at the OM2/MRCI level. Conical-intersection geometries and energies were also computed at the OM2/MRCI level. This dataset is designed to enable a deep integration of photochemistry with machine learning, bridging the gap between photochemical insight and data-driven approaches. † † affiliation: MOE Key Laboratory of Environmental Theoretical Chemistry, Guangdong Provincial Key Laboratory of Chemical Pollution and Environmental Safety, SCNU Environmental Research Institute, School of Environment, South China Normal University, Guangzhou 510006, P. R. China. † † affiliation: MOE Key Laboratory of Environmental Theoretical Chemistry, Guangdong Provincial Key Laboratory of Chemical Pollution and Environmental Safety, SCNU Environmental Research Institute, School of Environment, South China Normal University, Guangzhou 510006, P. R. China. † † affiliation: MOE Key Laboratory of Environmental Theoretical Chemistry, Guangdong Provincial Key Laboratory of Chemical Pollution and Environmental Safety, SCNU Environmental Research Institute, School of Environment, South China Normal University, Guangzhou 510006, P. R. China. † † affiliation: Computer Network Information Center, Chinese Academy of Sciences, Beijing 100190, China † † affiliation: MOE Key Laboratory of Environmental Theoretical Chemistry, Guangdong Provincial Key Laboratory of Chemical Pollution and Environmental Safety, SCNU Environmental Research Institute, School of Environment, South China Normal University, Guangzhou 510006, P. R. China. † † email: [email protected] † † affiliation: MOE Key Laboratory of Environmental Theoretical Chemistry, Guangdong Provincial Key Laboratory of Chemical Pollution and Environmental Safety, SCNU Environmental Research Institute, School of Environment, South China Normal University, Guangzhou 510006, P. R. China. † † email: [email protected] 1 Background & Summary Photophysics and photochemistry play vital roles in daily life, as solar radiation initiates many critical processes on Earth 14 ; 15 ; 16 ; 12 ; 4 ; 13 ; 36 ; 52 . Upon photoabsorption, molecules often undergo electronic transitions to excited states. On the excited-state potential energy surface (PES), nuclear motion tends to drive the system into regions where two electronic states of the same spin become degenerate. In such regions, the topology of the two coupled PESs exhibits a characteristic double-cone shape. These crossing regions are therefore termed “conical intersections” (CIs) 14 ; 15 ; 16 ; 36 ; 12 ; 13 . In the vicinity of a conical intersection, radiationless electronic transitions from one state to the other can occur as the nuclei move, giving rise to so-called nonadiabatic dynamics. Such nonadiabatic processes play a crucial role in photoinduced reactions and govern many reaction mechanisms in polyatomic molecular systems. Consequently, conical intersections are often regarded as the “funnels” of photochemistry. 14 ; 15 ; 6 Given the widespread occurrence of nonadiabatic dynamics at CIs, a comprehensive study of these intersections is of fundamental importance for understanding photochemical reaction pathways. However, a major limitation of current research on CIs and nonadiabatic dynamics is its highly system-specific nature. Typically, researchers locate CIs using suitable electronic structure theories and study nonadiabatic dynamics using various dynamics-based approaches 14 ; 15 ; 16 ; 36 ; 12 ; 13 ; 35 , while all studies are performed individually for each molecular system at the atomic level. Although such research provides many mechanistic insights and deepens our understanding of photophysics and photochemistry, broader exploration remains severely limited because this system-specific workflow cannot be easily scaled up. Artificial intelligence may provide an effective way to address these challenges. In recent decades, machine learning (ML) becomes increasingly important in chemical research 1 ; 53 ; 20 ; 18 ; 61 ; 29 ; 31 ; 8 ; 10 ; 25 ; 33 . Nevertheless, ML applications in photochemistry and photophysics are still confined to a few topics, such as predicting excitation energies of molecular systems 45 ; 55 ; 63 ; 11 and accelerating excited-state dynamical evolution 25 ; 9 ; 17 ; 30 ; 33 ; 57 ; 3 ; 62 . Critically, ML studies focusing on CI properties across many compounds remain rare. The key reason is the lack of a high-quality quantum chemistry CI dataset. While several ground-state datasets 43 ; 64 ; 59 ; 46 ; 49 ; 37 ; 2 ; 54 ; 7 ; 44 ; 65 exist and some datasets provide preliminary excited-state information 66 ; 47 ; 67 ; 32 ; 42 ; 38 ; 34 , comprehensive datasets containing CI geometries and energies are still lacking. As a direct consequence, ML models for CI prediction cannot be developed at scale. To fill this gap, we present the QCDGE-CI dataset, a quantum chemistry dataset containing ground-state and CI properties. The molecular structures are sourced from the QCDGE dataset 66 . Ground-state geometry optimizations are performed at the semi-empirical OM2 level 60 , and single-point calculations are performed at the OM2/MRCI 60 ; 27 ; 26 level. CI geometries are also obtained at the OM2/MRCI level. 26 The OM2/MRCI method is chosen because it is computationally efficient while providing a reasonable description of CIs, 26 ; 39 ; 15 allowing us to process over 260k molecules. The QCDGE-CI dataset comprises 260,541 molecules, all containing up to ten heavy atoms limited to carbon (C), nitrogen (N), oxygen (O), and fluorine (F). This dataset is intended for applications such as predicting CI structures and energies, accelerating the understanding of photochemical reaction mechanisms, and aiding molecule design. Among all types of CIs, those between the ground state ( S_{0} ) and the first excited state ( S_{1} ) are particularly important, as they are involved in many nonadiabatic processes, such as photoisomerization of molecular switches, photostability of biological molecules 14 ; 15 ; 16 ; 6 ; 12 ; 4 ; 13 ; 36 . Moreover, S_{0} - S_{1} CIs are computationally more manageable than those involving higher excited states, and the OM2/MRCI method offers better accuracy for this lowest-lying crossing. Therefore, the present dataset focuses exclusively on S_{0} - S_{1} CIs. We wish to emphasize that the QCDGE-CI dataset is intended for qualitative to semi-quantitative applications, such as ML model training, large-scale pattern analysis, and preliminary screening of conical intersection properties. It does not aim to replace high-level multi-reference 56 ; 14 calculations for individual systems. Therefore, although the inherent approximations of the semi-empirical OM2/MRCI method should not be neglected, we believe this dataset will open new opportunities for large-scale CI searches, shedding light on a wide range of applications relevant to photochemistry and photophysics. 2 Methods As shown in Figure 1, the workflow for constructing the QCDGE-CI dataset consists of five steps, which are described in the following subsections. Here all electronic-structure calculations were performed at OM2 and OM2/MRCI levels using MNDO software 58 . For simplicity, we only outline a few key computational setups here and more details are given in Supporting Information (SI). Several filtering steps were also applied during dataset construction. Figure 1: