A database of vacancy formation enthalpies for materials discovery

<strong>A database of vacancy formation enthalpies for materials discovery</strong> Matthew Witman^a, Anuj Goyal^b, Tadashi Ogitsu^c, Anthony McDaniel^a, Stephan Lany^b ^a Sandia National Laboratories, ^b National Renewable Energy Laboratory, ^c Lawrence Livermore National Laboratories <strong>Abstract</strong> This dataset provides DFT calculations of cation and oxygen vacancy defects in oxides which can be used to derive efficient data-driven models for vacancy formation enthalpy. DFT calculations were performed as described in &lt;DOI: 10.26434/chemrxiv-2022-frcns&gt;, where a graph neural network surrogate model was trained and used to screen the Materials Project for promising solar thermochemical water splitting materials. The data, models, scripts and code needed to reproduce the results in &lt;DOI: 10.26434/chemrxiv-2022-frcns&gt; are described below. <strong>Data &amp; Models</strong> 1) data_01_03_22/* corresponds to oxide compounds used in model training 2) known_cmpds/* corresponds to known STCH compounds 3) screeningMP/* corresponds to the screening related data screening_inelements/* stores only Materials Project oxides whose composition is a subset of the training elements and contains all the vacancy defect predictions MP_O_PDs/* stores offline PDs from Materials Project so that adjusting oxide stability metrics can be done somewhat rapidly MP_O_Compounds/* stores possible MP oxide compounds to screen In general, the above folders contain: DFT data/structures are included in sub-directories: poscars, magnetic moments, oxidation states, and csvs (containing the vacancy enthalpy for each unique site) cgcnn/* contains the processed DFT data for use in the CGCNN code (see <strong>Scripts</strong> for how to prepare this) id_prop.csv.* contains [cif name, defect formation enthalpy] pairs Different id_prop.csv.* files correspond to different K-fold stratifications in the screening directory, defect formation enthalpy is omitted since it has not been computed with DFT model-(X1)k(X2)_(X3)_(X4) corresponds to different CV models for X1 different training set sizes (i.e., try to train with only 10%, 40%, or 100% of the data) X2 different k folds X3 = "struct" or "" for "structure-wise validation" or "defect-wise validation", respectively X4 for different encoding strategies structure X-Yz.cif indicates structure X, defect element Y, symmetry site z, where one instance of that site has been re-ordered to be the first atom in the cif file *.locals contains a one-hot encoding of oxidation states of all sites in that crystal *.locals_continuous contains a continuous encoding of oxidation state in that crystal *.globals contains global properties of the host structure <strong>Scripts</strong> scripts/*.sh scripts to rerun the screenings for different k-folds, encodings, etc. scripts/*.ipynb to analyze results scripts/prepare_cgcnn.py for translating the data in (poscars/*, csvs/*, oxstate/*, mags/*) to the ML input needed in cgcnn/* <strong>Code</strong> Install CGCNN and its defect modifications from https://github.com/mwitman1/cgcnndefect <strong>Questions/Collaborations</strong> Please contact [e-mail] <strong>Acknowledgements</strong> This material is based upon work supported by the U.S. Department of Energy (DOE), Office of Energy Efficiency and Renewable Energy (EERE), specifically the Hydrogen and Fuel Cell Technologies Office. Sandia National Laboratories is a multi-mission laboratory managed and operated by National Technology and Engineering Solutions of Sandia, LLC., a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525. Part of the work was performed under the auspices of the US Department of Energy by Lawrence Livermore National Laboratory under contract No.~DE- AC52-07NA27344. The National Renewable Energy Laboratory (NREL) is operated by the Alliance for Sustainable Energy, LLC, for the DOE under Contract No.~DE-AC36-08GO28308. This work used High-Performance Computing resources at NREL, sponsored by DOE-EERE. The views expressed in this article do not necessarily represent the views of the U.S. Department of Energy or the United States Government.<strong>用于材料发现的空位形成焓数据库</strong> Matthew Witman^a, Anuj Goyal^b, Tadashi Ogitsu^c, Anthony McDaniel^a, Stephan Lany^b ^a 桑迪亚国家实验室，^b 美国国家可再生能源实验室，^c 劳伦斯利弗莫尔国家实验室 <strong>摘要</strong> 本数据集收录了氧化物中阳离子与氧空位缺陷的密度泛函理论（Density Functional Theory, DFT）计算结果，可用于构建高效的数据驱动型空位形成焓预测模型。本研究的DFT计算按照<DOI: 10.26434/chemrxiv-2022-frcns>中所述方法完成，该研究中训练了图神经网络替代模型，并利用该模型对材料项目（Materials Project）数据库进行筛选，以寻找具有应用潜力的太阳能热化学水分解材料。下文将详述复现<DOI: 10.26434/chemrxiv-2022-frcns>中研究结果所需的数据集、模型、脚本与代码。 <strong>数据与模型</strong> 1) data_01_03_22/* 对应模型训练所用的氧化物化合物 2) known_cmpds/* 对应已报道的太阳能热化学水分解（STCH）化合物 3) screeningMP/* 对应筛选相关数据 screening_inelements/* 仅存储元素组成属于训练集元素子集的材料项目氧化物，并包含全部空位缺陷预测结果 MP_O_PDs/* 存储来自材料项目的离线相图（Phase Diagrams, PDs），以便能够快速调整氧化物稳定性指标 MP_O_Compounds/* 存储待筛选的材料项目氧化物候选化合物 一般而言，上述文件夹包含以下内容： DFT数据与结构包含于子目录：poscars、磁矩数据、氧化态数据以及记录各独特位点空位焓的CSV文件 cgcnn/* 存储经处理的DFT数据，用于CGCNN代码（详见“脚本”部分了解数据预处理方法） id_prop.csv.* 文件包含[cif文件名，缺陷形成焓]键值对 不同的id_prop.csv.* 文件对应不同的K折分层策略 在筛选目录中，未包含缺陷形成焓数据，因为尚未通过DFT完成相关计算 model-(X1)k(X2)_(X3)_(X4) 代表不同的交叉验证（Cross Validation, CV）模型：其中X1代表不同的训练集规模（即仅使用10%、40%或100%的数据进行训练），X2代表不同的K折划分，X3分别为"struct"或空字符串，对应“结构维度验证”或“缺陷维度验证”，X4代表不同的编码策略 structure X-Yz.cif 表示结构X、缺陷元素Y、对称位点z，该文件中该位点的一个实例已被重排为cif文件中的第一个原子 *.locals 文件包含该晶体中所有位点氧化态的独热编码 *.locals_continuous 文件包含该晶体中氧化态的连续编码 *.globals 文件包含主体结构的全局属性 <strong>脚本</strong> scripts/*.sh 为用于重新运行不同K折划分、编码策略等条件下筛选任务的脚本 scripts/*.ipynb 为用于分析实验结果的脚本 scripts/prepare_cgcnn.py 用于将poscars/*、csvs/*、oxstate/*、mags/*中的数据转换为cgcnn/*所需的机器学习输入格式 <strong>代码</strong> 可从https://github.com/mwitman1/cgcnndefect 安装CGCNN及其缺陷修正版本 <strong>问题咨询与合作</strong> 如有疑问或合作意向，请联系[e-mail] <strong>致谢</strong> 本研究工作得到美国能源部（DOE）能源效率与可再生能源办公室（EERE）下属氢能与燃料电池技术办公室的资助。 桑迪亚国家实验室为由霍尼韦尔国际全资子公司桑迪亚国家技术与工程解决方案有限责任公司为美国能源部国家核安全管理局运营的多任务实验室，相关合同编号为DE-NA0003525。 本研究部分工作由劳伦斯利弗莫尔国家实验室在美国能源部支持下完成，合同编号为DE-AC52-07NA27344。 美国国家可再生能源实验室（NREL）由可持续能源联盟有限责任公司为美国能源部运营，合同编号为DE-AC36-08GO28308。 本研究使用了由DOE-EERE资助的NREL高性能计算资源。 本文所表达的观点不一定代表美国能源部或美国联邦政府的官方立场。