基于机器学习方法的二维材料带隙预测

doi:10.12066/j.issn.1007-2861.2089

上海大学学报(自然科学版) ›› 2020, Vol. 26 ›› Issue (5): 824-833.doi: 10.12066/j.issn.1007-2861.2089

基于机器学习方法的二维材料带隙预测

游洋¹^,², 杜婉³, 李惟驹¹, 陈竞哲¹^,²()

1.上海大学理学院, 上海 200444
2.上海大学量子与分子结构国际研究中心, 上海 200444
3.上海大学材料基因组工程研究院, 上海 200444

收稿日期:2018-09-12 出版日期:2020-10-30 发布日期:2020-11-06
通讯作者: 陈竞哲 E-mail:jingzhe@shu.edu.cn
基金资助:
国家自然科学基金青年科学基金资助项目(11404206)

Two-dimensional material band gap prediction based on machine learning method

YOU Yang¹^,², DU Wan³, LI Weiju¹, CHEN Jingzhe¹^,²()

1. College of Sciences, Shanghai University, Shanghai 200444, China
2. International Centre for Quantum and Molecular Structures, Shanghai University, Shanghai 200444, China
3. Materials Genome Institute, Shanghai University, Shanghai 200444, China

Received:2018-09-12 Online:2020-10-30 Published:2020-11-06
Contact: CHEN Jingzhe E-mail:jingzhe@shu.edu.cn

摘要/Abstract

摘要：

利用密度泛函理论与机器学习相结合的方法, 对二维金属化合物的带隙进行研究, 得到比传统理论计算成本更低且更有效的带隙预测方法. 以广义梯度近似(general gradient approximation, GGA)-Perdew-Burke-Ernzerhof (PBE) (GGA-PBE) 和 G₀W₀ 的带隙计算结果为参考, 考察了化学通式为 MX₂ 的二维材料数据集. 利用套索回归, 即最小绝对值收敛和选择算子(least absolute shrinkage and selection operator, LASSO)、支持向量回归(support vector regression, SVR)和梯度树提升回归(gradient boosting regression, GBR)等机器学习方法建立带隙的预测模型. 测试结果表明, 对于大多数二维材料, 基于线性核函数的 SVR 与 LASSO 模型的预测性能相对较好, 训练模型的平均绝对误差为 0.34 eV, 测试集误差为 0.5 eV. 这说明对于二维材料带隙采取的特征参数集具有一定的完备性和合理性, 对新材料带隙的初步预测有一定的参考价值.

关键词: 二维材料, 第一性原理, 机器学习, 带隙

Abstract:

Machine learning (ML) algorithm and the traditional density functional theory (DFT) are combined to study the band gap of two-dimensional metal compounds, and as a result, a simple and effective model which is more cost-effective than the traditional quantum calculation method is established. Results of general gradient approximation-Perdew-Burke-Ernzerhof (GGA-PBE) and G₀W₀are taken as reference and a two-dimensional material data set with chemical formula MX₂ is investigated. Least absolute shrinkage and selection operator (LASSO), support vector machine regression (SVR) and gradient boosting regressor (GBR) and other machine learning methods are used to build a band gap prediction model. Among these models, it is found that the SVR model based on linear kernel function and LASSO model both can give a good prediction result, the mean absolute error (MAE) of training model is 0.34 eV and MAE of testing set is 0.5 eV. Thus, for the prediction of two-dimensional material band gap, the feature parameter set adopted by us has a certain completeness and rationality, which has a certain reference value for the preliminary prediction for the band gap of new materials.

Key words: two-dimensional materials, first principles, machine learning, band gap

中图分类号:

TB39
O469

游洋, 杜婉, 李惟驹, 陈竞哲. 基于机器学习方法的二维材料带隙预测[J]. 上海大学学报(自然科学版), 2020, 26(5): 824-833.

YOU Yang, DU Wan, LI Weiju, CHEN Jingzhe. Two-dimensional material band gap prediction based on machine learning method[J]. Journal of Shanghai University（Natural Science Edition）, 2020, 26(5): 824-833.

图/表 8

图1

图2

图3

表1

表2

图4

表3

表4

参考文献 21

[1]	Chhowalla M, Shin H S, Eda G, et al. The chemistry of two-dimensional layered transition metal dichalcogenide nanosheets[J]. Nature Chemistry, 2013,5(4):263-275. pmid: 23511414
[2]	Miro P, Audiffred M, Heine T. An atlas of two-dimensional materials[J]. Chemical Society Reviews, 2014,43(18):6537-6554. pmid: 24825454
[3]	Liu G B, Xiao D, Yao Y, et al. Electronic structures and theoretical modelling of two-dimensional group-VIB transition metal dichalcogenides[J]. Chemical Society Reviews, 2015,44(9):2643-2663. pmid: 25474725
[4]	Segall M, Lindan P J, Probert M A, et al. First-principles simulation: ideas, illustrations and the CASTEP code[J]. Journal of Physics: Condensed Matter, 2002,14(11):2717.
[5]	Bickelhaupt F M, Baerends E J. Kohn-Sham density functional theory: predicting and understanding chemistry[J]. Reviews in Computational Chemistry, 2007. DOI: 10.1002/9780470125922.ch1 doi: 10.1016/S1574-1400(05)01011-X pmid: 32288696
[6]	Fuchs F, Furthmller J, Bechstedt F, et al. Quasiparticle band structure based on a generalized Kohn-Sham scheme[J]. Physical Review B, 2007,76(11):115109.
[7]	Landis D D, Hummelsh J J S, Nestorov S, et al. The computational materials repository[J]. Computing in Science & Engineering, 2012,14(6):51-57.
[8]	Tibshirani R. Regression shrinkage and selection via the lasso[J]. Journal of the Royal Statistical Society Series B, 1996,58(1):267-288.
[9]	Breiman L. Better subset regression using the nonnegative garrote[J]. Technometrics, 1995,37(4):373-384.
[10]	Cortes C, Vapnik V. Support-vector networks[J]. Machine Learning, 1995,20(3):273-297.
[11]	Smola A J, Schlkopf B. A tutorial on support vector regression[J]. Statistics and Computing, 2004,14(3):199-222.
[12]	Ben-Hur A, Horn D, Siegelmann H T, et al. Support vector clustering[J]. Journal of Machine Learning Research, 2001,2:125-137.
[13]	Friedman J H. Stochastic gradient boosting[J]. Computational Statistics & Data Analysis, 2002,38(4):367-378.
[14]	Friedman J H. Greedy function approximation: a gradient boosting machine[J]. Annals of Statistics, 2001,29(5):1189-1232.
[15]	Samuel A L. Some studies in machine learning using the game of checkers[J]. IBM Journal of Research and Development, 1959,3(3):210-229.
[16]	Wernick M N, Yang Y, Brankov J G, et al. Machine learning in medical imaging[J]. IEEE Signal Processing Magazine, 2010,27(4):25-38. doi: 10.1109/MSP.2010.936730 pmid: 25382956
[17]	Armstrong J. Illusions in regression analysis[J]. International Journal of Forecasting, 2012,28(3):689-698.
[18]	Hofmann T, Schlkopf B, Smola A J. Kernel methods in machine learning[J]. Annals of Statistics, 2008,36(3):1171-1220.
[19]	Suykens J A, De Brabanter J, Lukas L, et al. Weighted least squares support vector machines: robustness and sparse approximation[J]. Neurocomputing, 2002,48:85-105.
[20]	Wilson J A, Yoffe A D. The transition metal dichalcogenides discussion and interpretation of the observed optical, electrical and structural properties[J]. Advances in Physics, 1969,18(73):193-335.
[21]	Kohavi R. A study of cross-validation and bootstrap for accuracy estimation and model selection[C]// IJCAI. 1995: 1137-1145.

基于机器学习方法的二维材料带隙预测

Two-dimensional material band gap prediction based on machine learning method

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 8

参考文献 21

相关文章 5

编辑推荐

Metrics

本文评价

[1]	李晖, 辛子华. 硼氮掺杂C₆₄-石墨炔材料几何及电子结构第一性原理[J]. 上海大学学报(自然科学版), 2020, 26(5): 816-823.
[2]	马帅, 李拥华, 高裕博. Ca 对氧化铝晶界处氧空位扩散的活化机理[J]. 上海大学学报(自然科学版), 2020, 26(4): 562-569.
[3]	金山, 靳锡联, 焦正, 孟醒. Ca$_{\textbf{0.5}}$Ba$_{\textbf{0.5}}$MnO$_{\textbf{3}}$多铁性的第一性原理[J]. 上海大学学报(自然科学版), 2019, 25(4): 590-596.
[4]	严俊1, 刘培钧2, 张旭2, 俞丹燕2, 胡仙超3, 王小祥1. 海水养殖金色珍珠独特的吸收光谱及其微结构[J]. 上海大学学报(自然科学版), 2014, 20(6): 707-714.
[5]	余斌,徐飞,马忠权,周平华,石建伟,郑玲玲,李拥华，洪峰. 铜铟硒与铜铟硫太阳能电池中有序缺陷化合物的性质及对能带偏移的影响[J]. 上海大学学报(自然科学版), 2012, 18(3): 271-276.