机器学习和密度泛函理论加速二维材料开 发：基于单层 HfSe2 的气体传感器

新材料的发现能够推动现代科学技术的不断进步。传统的实验方法往 往成本高、周期长、重复性差，而气敏材料的筛选也不例外。使用密度泛 函理论计算吸附能的计算成本变得非常昂贵，无法满足高效的要求，因此 需要更便宜且足够准确的替代方法。基于机器学习的方法可以通过从已有 数据中学习模式，快速预测新材料的性质和行为。因此，机器学习在气敏 材料筛选的应用已经成为一个热门的研究领域，能更快速、更有效地发现 具有重要应用价值的二维材料。为了研究二硒化铪(HfSe 2 )单层作为气敏材 料的潜力以及对其气敏特性的预测，本研究采用机器学习技术和密度泛函 理论计算。研究首先采用八种不同的机器学习模型，通过利用从密度泛函 理论计算得到的 HfSe2 单层与 10 种气体之间的吸附数据，建立了气体吸附 分类预测模型。这些模型被用来预测气体在 HfSe2 单层上的吸附情况，包 括 C2H6、CH4、H2O、O2、N2、NO、NO2、NH3、HCHO 和 O3。之后扩 充了 HfSe2 单层与 10 种气体吸附的计算数据，采用基于图卷积神经网络的 机器学习方法，建立了气体吸附回归预测模型。该模型可以仅仅从原子的 位置信息中预测气体的吸附能和电荷转移情况，而无需使用更复杂的密度 泛函理论计算。本研究成功地利用基于分类和回归的机器学习模型，预测 了 HfSe2 单层的气体吸附结果、吸附能和转移电荷。同时，通过机器学习 获得了一种较好的搜索策略，可以更快速、高效地探索新材料，从而加速 材料发现的进程。这项研究结果证明了机器学习在加速材料发现方面的巨 大潜力，特别是在气体传感材料筛选中的应用。

[1] Correa-Baena J P, Hippalgaonkar K, van Duren J, et al. Accelerating materials development via automation, machine learning, and high-performance computing[J]. Joule, 2018, 2(8): 1410-1420.
[2] Li Y, Li X, Xu Y. The sensing mechanism of pristine and transition metals doped Zn12 O12, Sn12 O12 and Ni12 O12 nanocages towards NH3 and PH3: a DFT study[J]. Journal of Materials Chemistry C, 2021, 9(48): 17382-17391.
[3] Liang X Y, Ding N, Ng S P, et al. Adsorption of gas molecules on Ga-doped graphene and effect of applied electric field: A DFT study[J]. Applied Surface Science, 2017, 411: 11-17.
[4] Zala V B, Shukla R S, Bhuyan P D, et al. Highly selective and reversible 2D PtX2 (X= P, As) hazardous gas sensors: Ab-initio study[J]. Applied Surface Science, 2021, 563: 150391.
[5] Wu Y, Chen X, Weng K, et al. Highly sensitive and selective gas sensor using heteroatom doping graphdiyne: a DFT study[J]. Advanced Electronic Materials, 2021, 7(7): 2001244.
[6] Xia S Y, Tao L Q, Jiang T, et al. Rh-doped h-BN monolayer as a high sensitivity SF6 decomposed gases sensor: A DFT study[J]. Applied Surface Science, 2021, 536: 147965.
[7] Lu S, Zhou Q, Guo Y, et al. Coupling a crystal graph multilayer descriptor to active learning for rapid discovery of 2D ferromagnetic semiconductors/half‐metals/metals[J]. Advanced Materials, 2020, 32(29): 2002658.
[8] Lu S, Zhou Q, Ouyang Y, et al. Accelerated discovery of stable lead-free hybrid organic-inorganic perovskites via machine learning[J]. Nature communications, 2018, 9(1): 3405.
[9] Panapitiya G, Avendaño-Franco G, Ren P, et al. Machine-learning prediction of CO adsorption in thiolated, Ag-alloyed Au nanoclusters[J]. Journal of the American Chemical Society, 2018, 140(50): 17508-17514.
[10] Borboudakis G, Stergiannakos T, Frysali M, et al. Chemically intuited, large-scale screening of MOFs by machine learning techniques[J]. npj Computational Materials, 2017, 3(1): 40.
[11] Hansen K, Biegler F, Ramakrishnan R, et al. Machine learning predictions of molecular properties: Accurate many-body potentials and nonlocality in chemical space[J]. The journal of physical chemistry letters, 2015, 6(12): 2326-2331.
[12] Rupp M, Tkatchenko A, Müller K R, et al. Fast and accurate modeling of molecular atomization energies with machine learning[J]. Physical review letters, 2012, 108(5): 058301.
[13] Pereira F, Xiao K, Latino D A R S, et al. Machine learning methods to predict density functional theory B3LYP energies of HOMO and LUMO orbitals[J]. Journal of chemical information and modeling, 2017, 57(1): 11-21.
[14] Gui Y, Shi J, Yang P, et al. Platinum modified MoS2 monolayer for adsorption and gas sensing of SF6 decomposition products: A DFT study[J]. High Voltage, 2020, 5(4): 454-462.
[15] Gui Y, Shi J, Xu L, et al. Aun (n= 1–4) cluster doped MoSe2 nanosheet as a promising gas-sensing material for C2H4 gas in oil-immersed transformer[J]. Applied Surface Science, 2021, 541: 148356.
[16] Xu L, Gui Y, Li W, et al. Gas-sensing properties of Ptn-doped WSe2 to SF6 decomposition products[J]. Journal of Industrial and Engineering Chemistry, 2021, 97: 452-459.
[17] Zeng Y, Lin S, Gu D, et al. Two-dimensional nanomaterials for gas sensing applications: The role of theoretical calculations[J]. Nanomaterials, 2018, 8(10): 851.
[18] Cui H, Zhu H, Jia P. Adsorption and sensing of SO2 and SOF2 molecule by Pt-doped HfSe2 monolayer: A first-principles study[J]. Applied Surface Science, 2020, 530: 147242.
[19] Cui H, Jia P, Peng X. Adsorption of SO2 and NO2 molecule on intrinsic and Pd-doped HfSe2 monolayer: A first-principles study[J]. Applied Surface Science, 2020, 513: 145863.
[20] Lin L, Feng Z, Dong Z, et al. Adsorption of small gas molecules of transition metal (Pt and Au) modified HfSe2 monolayer[J]. Materials Today Communications, 2022, 32: 103885.
[21] Wang Y, Liu B, Fang R, et al. Adsorption and Sensing of CO2, CH4 and N2O Molecules by Ti-Doped HfSe2 Monolayer Based on the First-Principle[J]. Chemosensors, 2022, 10(10): 414.
[22] Li F, Shi C. NO-sensing performance of vacancy defective monolayer MoS2 predicted by density function theory[J]. Applied Surface Science, 2018, 434: 294-306.
[23] Gui Y, Lin Y, Ji C, et al. Density Functional Theory Calculations for the Adsorption Property of Hazardous Industrial Gasses on Transition-Metal-Modified MoS2 Nanosheets[J]. ACS Applied Nano Materials, 2022, 5(8): 11111-11118.
[24] Ni J, Wang W, Quintana M, et al. Adsorption of small gas molecules on strained monolayer WSe2 doped with Pd, Ag, Au, and Pt: A computational investigation[J]. Applied Surface Science, 2020, 514: 145911.
[25] Cui H, Jiang J, Gao C, et al. DFT study of Cu-modified and Cu-embedded WSe2 monolayers for cohesive adsorption of NO2, SO2, CO2, and H2S[J]. Applied Surface Science, 2022, 583: 152522.
[26] Li B, Chen X, Su C, et al. Enhanced dimethyl methylphosphonate detection based on two-dimensional WSe 2 nanosheets at room temperature[J]. Analyst, 2020, 145(24): 8059-8067.
[27] Pilania, G., Mannodi-Kanakkithodi, A., Uberuaga, B. et al. Machine learning bandgaps of double perovskites. Sci Rep 6, 19375 (2016).
[28] Tran K, Ulissi Z W. Active learning across intermetallics to guide discovery of electrocatalysts for CO2 reduction and H2 evolution[J]. Nature Catalysis, 2018, 1(9): 696-703.
[29] Jie J, Hu Z, Qian G, et al. Discovering unusual structures from exception using big data and machine learning techniques[J]. Science Bulletin, 2019, 64(9): 612-616.
[30] Chen C, Ye W, Zuo Y, et al. Graph networks as a universal machine learning framework for molecules and crystals[J]. Chemistry of Materials, 2019, 31(9): 3564-3572.
[31] Xie T, Grossman J C. Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties[J]. Physical review letters, 2018, 120(14): 145301.
[32] Louis S Y, Zhao Y, Nasiri A, et al. Graph convolutional neural networks with global attention for improved materials property prediction[J]. Physical Chemistry Chemical Physics, 2020, 22(32): 18141-18148.
[33] Butler K T, Davies D W, Cartwright H, et al. Machine learning for molecular and materials science[J]. Nature, 2018, 559(7715): 547-555.
[34] Hohenberg P, Kohn W. Inhomogeneous electron gas[J]. Physical review, 1964, 136(3B): B864.Srgs
[35] Kohn W, Sham L J. Self-consistent equations including exchange and correlation effects[J]. Physical review, 1965, 140(4A): A1133.Sda
[36] Ceperley D M, Alder B J. Ground state of the electron gas by a stochastic method[J]. Physical review letters, 1980, 45(7): 566.Gbrtj
[37] Langreth D C, Perdew J P. Theory of nonuniform electronic systems. I. Analysis of the gradient approximation and a generalization that works[J]. Physical Review B, 1980, 21(12): 5469.FddjDfdg
[38] Perdew J P, Burke K, Ernzerhof M. Generalized gradient approximation made simple[J]. Physical review letters, 1996, 77(18): 3865.
[39] Côté D. Using machine learning in communication networks[J]. Journal of Optical Communications and Networking, 2018, 10(10): D100-D109.
[40] McCluskey W J, Deddis W G, Lamont I G, et al. The application of surface generated interpolation models for the prediction of residential property values[J]. Journal of Property Investment & Finance, 2000.
[41] Zabel J, Dalton M. The impact of minimum lot size regulations on house prices in Eastern Massachusetts[J]. Regional Science and Urban Economics, 2011, 41(6): 571-583.
[42] Kruschke J K. Bayesian data analysis[J]. Wiley Interdisciplinary Reviews: Cognitive Science, 2010, 1(5): 658-676.
[43] Breiman L. Random forests[J]. Machine learning, 2001, 45: 5-32.
[44] Mangalathu S, Jang H, Hwang S H, et al. Data-driven machine-learning-based seismic failure mode identification of reinforced concrete shear walls[J]. Engineering Structures, 2020, 208: 110331.
[45] Morgan J. Classification and regression tree analysis[J]. Boston: Boston University, 2014, 298.
[46] Berry D A. Teaching elementary Bayesian statistics with real applications in science[J]. The American Statistician, 1997, 51(3): 241-246.
[47] Chen T, Guestrin C. Xgboost: A scalable tree boosting system[C]//Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining. 2016: 785-794.
[48] Freund Y, Schapire R E. A decision-theoretic generalization of on-line learning and an application to boosting[J]. Journal of computer and system sciences, 1997, 55(1): 119-139.
[49] Ke G, Meng Q, Finley T, et al. Lightgbm: A highly efficient gradient boosting decision tree[J]. Advances in neural information processing systems, 2017, 30.
[50] Prokhorenkova L, Gusev G, Vorobev A, et al. CatBoost: unbiased boosting with categorical features[J]. Advances in neural information processing systems, 2018, 31.
[51] Dorogush A V, Ershov V, Gulin A. CatBoost: gradient boosting with categorical features support[J]. arXiv preprint arXiv:1810.11363, 2018.
[52] Zhang S, Tong H, Xu J, et al. Graph convolutional networks: a comprehensive review[J]. Computational Social Networks, 2019, 6(1): 1-23.
[53] Kipf T N, Welling M. Semi-supervised classification with graph convolutional networks[J]. arXiv preprint arXiv:1609.02907, 2016.
[54] Kingma D P, Ba J. Adam: A method for stochastic optimization[J]. arXiv preprint arXiv:1412.6980, 2014.
[55] Reddi S J, Kale S, Kumar S. On the convergence of adam and beyond[J]. arXiv preprint arXiv:1904.09237, 2019.
[56] Pedregosa F, Varoquaux G, Gramfort A, et al. Scikit-learn: Machine Learn-ing in Python, Journal of Machine Learning Re-search, 12[J]. 2011.
[57] Liu Z H, Shi T T, Chen Z X. Machine learning prediction of monatomic adsorption energies with non-first-principles calculated quantities[J]. Chemical Physics Letters, 2020, 755: 137772.
[58] Hao J, Ho T K. Machine learning made easy: a review of scikit-learn package in python programming language[J]. Journal of Educational and Behavioral Statistics, 2019, 44(3): 348-361.
[59] Yuan J, Yu N, Wang J, et al. Design lateral heterostructure of monolayer ZrS2 and HfS2 from first principles calculations[J]. Applied Surface Science, 2018, 436: 919-926.
[60] Yang H, Li J, Shao Z, et al. Strain-engineered S-HfSe2 monolayer as a promising gas sensor for detecting NH3: A first-principles study[J]. Surfaces and Interfaces, 2022, 34: 102317.
[61] Song H Y, Sun J J, Li M. Enhancement of monolayer HfSe2 thermoelectric performance by strain engineering: A DFT calculation[J]. Chemical Physics Letters, 2021, 784: 139109.
[62] Hafner J. Ab‐initio simulations of materials using VASP: Density‐functional theory and beyond[J]. Journal of computational chemistry, 2008, 29(13): 2044-2078.
[63] Ong S P, Richards W D, Jain A, et al. Python Materials Genomics (pymatgen): A robust, open-source python library for materials analysis[J]. Computational Materials Science, 2013, 68: 314-319.
[64] Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90.
[65] Wallach I, Dzamba M, Heifets A. AtomNet: a deep convolutional neural network for bioactivity prediction in structure-based drug discovery[J]. arXiv preprint arXiv:1510.02855, 2015.
[66] Gao C, Zhang Y, Yang H, et al. A DFT study of In doped Tl2O: a superior NO2 gas sensor with selective adsorption and distinct optical response[J]. Applied Surface Science, 2019, 494: 162-169.