Abstract
Crystal graph convolutional neural networks (CGCNNs) have revolutionized materials research by eliminating the need for manual feature engineering. However, their lack of interpretability and sensitivity to structural distortions hinders their application in substitution engineering. Therefore, we propose an innovative adversarial example method for guiding feature construction and enhancing the interpretability of CGCNNs. In this study, our focus lies on identifying CsSnBr3−xIx configurations with high-energy conversion efficiency. Initially, we train a CGCNN classifier as a benchmark. Subsequently, we perturb input data to generate a low-performance classifier and identify adversarial examples based on incorrect predictions. Upon comparing these examples with normal examples, we observe substantial structural distortions in adversarial cases, serving as inspiration for the creation of disorder-related features. Consequently, an interpretable model is developed, which surpasses CGCNNs with atomic position perturbations and a gradient-boosting classifier using general features. Notably, the previously overlooked feature “number of unequal atoms” plays an important role in offering crucial insights. Further analysis reveals that configurations with pronounced disorder can exhibit increased power density, thereby enhancing the energy conversion efficiency. Our work not only elucidates the impact of atom substitution on energy conversion efficiency but also provides a roadmap for constructing interpretable machine learning models.
摘要
晶体图卷积神经网络(CGCNN)不需要手动创建描述符, 因而改变了材料研究方式. 然而, 它们在可解释性以及对结构畸变的敏感性方面的不足, 阻碍了该模型在材料掺杂工程中的应用. 因此, 我们提出了一种具有创新性的对抗样本方法, 用于引导特征构建并增强CGCNN的可解释性. 在这项研究中, 重点是寻找具有高能量转换效率的CsSnBr3−xIx构型. 首先, 训练了一个CGCNN分类器作为基准. 随后, 对输入数据进行扰动以生成低性能分类器, 并基于错误预测结果来识别对抗样本. 将对抗样本与正常样本进行比较, 发现对抗样本中存在着明显的结构畸变, 这为创建描述符提供了思路. 基于此, 建立了一个可解释模型, 该模型超越了使用原子位置扰动的CGCNN模型以及使用通用特征的梯度提升分类器模型. 值得注意的是, 之前被忽视的描述符, 即“不等价原子数”, 在提供关键见解方面发挥了重要作用. 进一步研究发现, 具有明显畸变的结构可以表现出增强的功率密度, 从而提高能量转换效率. 本工作不仅阐明了原子替代对能源转换效率的影响, 还为构建可解释的机器学习模型提供了可行方案.
Article PDF
Similar content being viewed by others
References
Noel NK, Abate A, Stranks SD, et al. Enhanced photoluminescence and solar cell performance via Lewis base passivation of organic-inorganic lead halide perovskites. ACS Nano, 2014, 8: 9815–9821
Hao J, Li W, Zhai J, et al. Progress in high-strain perovskite piezoelectric ceramics. Mater Sci Eng-R-Rep, 2019, 135: 1–57
Fechner M, Maznichenko IV, Ostanin S, et al. Magnetic phase transition in two-phase multiferroics predicted from first principles. Phys Rev B, 2008, 78: 212406
Gao LK, Zhao XH, Diao XF, et al. First-principles study of photoelectric properties of CsSnBr3 under hydrostatic pressure. Acta Phys Sin, 2021, 70: 158801
Mahmood Q, Yaseen M, Hassan M, et al. The first-principle study of mechanical, optoelectronic and thermoelectric properties of CsGeBr3 and CsSnBr3 perovskites. Mater Res Express, 2019, 6: 045901
Liang Y, Cui X, Li F, et al. Electrode-induced impurities in tin halide perovskite solar cell material CsSnBr3 from first principles. npj Comput Mater, 2021, 7: 63
Luo T, Zhang Y, Chang X, et al. Dual interfacial engineering for efficient Cs2AgBiBr6 based solar cells. J Energy Chem, 2021, 53: 372–378
Li T, Wang J, Gao Z, et al. Local stress enhanced photocurrent of visible light photo-detection in Cs2AgBiBr6 single crystal. Appl Phys Lett, 2019, 115: 131103
Sabba D, Mulmudi HK, Prabhakar RR, et al. Impact of anionic Br− substitution on open circuit voltage in lead free perovskite (CsSnI3−xBrx) solar cells. J Phys Chem C, 2015, 119: 1763–1767
Iqbal A, Sun Z, Wang G, et al. Optimizing band gap of inorganic halide perovskites by donor-acceptor pair codoping. Inorg Chem, 2020, 59: 6053–6059
Shi R, Fang Q, Vasenko AS, et al. Structural disorder in higher-temperature phases increases charge carrier lifetimes in metal halide perovskites. J Am Chem Soc, 2022, 144: 19137–19149
Liu Y, Long R, Fang WH. Great influence of organic cation motion on charge carrier dynamics in metal halide perovskite unraveled by un-supervised machine learning. J Phys Chem Lett, 2022, 13: 8537–8545
Lu S, Zhou Q, Ouyang Y, et al. Accelerated discovery of stable lead-free hybrid organic-inorganic perovskites via machine learning. Nat Commun, 2018, 9: 3405
Gladkikh V, Kim DY, Hajibabaei A, et al. Machine learning for predicting the band gaps of ABX3 perovskites from elemental properties. J Phys Chem C, 2020, 124: 8905–8918
Yuan R, Liu Z, Balachandran PV, et al. Accelerated discovery of large electrostrains in BaTiO3-based piezoelectrics using active learning. Adv Mater, 2018, 30: 1702884
Jin H, Zhang H, Li J, et al. Data-driven systematic search of promising photocatalysts for water splitting under visible light. J Phys Chem Lett, 2019, 10: 5211–5218
Davies DW, Butler KT, Walsh A. Data-driven discovery of photoactive quaternary oxides using first-principles machine learning. Chem Mater, 2019, 31: 7221–7230
Wang T, Tan X, Wei Y, et al. Unveiling the layer-dependent electronic properties in transition-metal dichalcogenide heterostructures assisted by machine learning. Nanoscale, 2022, 14: 2511–2520
Zhuo Y, Mansouri Tehrani A, Brgoch J. Predicting the band gaps of inorganic solids by machine learning. J Phys Chem Lett, 2018, 9: 1668–1673
Ouyang R, Curtarolo S, Ahmetcik E, et al. SISSO: A compressed-sensing method for identifying the best low-dimensional descriptor in an immensity of offered candidates. Phys Rev Mater, 2018, 2: 083802
Wang T, Tan X, Wei Y, et al. Accurate bandgap predictions of solids assisted by machine learning. Mater Today Commun, 2021, 29: 102932
Oftelie LB, Rajak P, Kalia RK, et al. Active learning for accelerated design of layered materials. npj Comput Mater, 2018, 4: 74
Hibat-Allah M, Ganahl M, Hayward LE, et al. Recurrent neural network wave functions. Phys Rev Res, 2020, 2: 023358
Jin H, Zhang H, Li J, et al. Discovery of novel two-dimensional photovoltaic materials accelerated by machine learning. J Phys Chem Lett, 2020, 11: 3075–3081
L. Agiorgousis M, Sun Y, Choe D, et al. Machine learning augmented discovery of chalcogenide double perovskites for photovoltaics. Advcd Theor Sims, 2019, 2: 1800173
Pfau D, Spencer JS, Matthews AGDG, et al. Ab initio solution of the many-electron Schrödinger equation with deep neural networks. Phys Rev Res, 2020, 2: 033429
Carleo G, Choo K, Hofmann D, et al. NetKet: A machine learning toolkit for many-body quantum systems. SoftwareX, 2019, 10: 100311
Wu Z, Pan S, Chen F, et al. A comprehensive survey on graph neural networks. IEEE Trans Neural Netw Learn Syst, 2021, 32: 4–24
Xie T, Grossman JC. Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties. Phys Rev Lett, 2018, 120: 145301
Chen C, Ye W, Zuo Y, et al. Graph networks as a universal machine learning framework for molecules and crystals. Chem Mater, 2019, 31: 3564–3572
Wang Z, Han Y, Cai J, et al. DeepTMC: A deep learning platform to targeted design doped transition metal compounds. Energy Storage Mater, 2022, 45: 1201–1211
Choudhary K, DeCost B. Atomistic line graph neural network for improved materials property predictions. npj Comput Mater, 2021, 7: 185
Park CW, Wolverton C. Developing an improved crystal graph convolutional neural network framework for accelerated materials discovery. Phys Rev Mater, 2020, 4: 063801
Kailkhura B, Gallagher B, Kim S, et al. Reliable and explainable machine-learning methods for accelerated material discovery. npj Comput Mater, 2019, 5: 108
Deng J, Jia G. An interpretable hybrid machine learning prediction of dielectric constant of alkali halide crystals. Chem Phys, 2022, 555: 111457
Anker AS, Kjær ETS, Juelsholt M, et al. Extracting structural motifs from pair distribution function data of nanostructures using explainable machine learning. npj Comput Mater, 2022, 8: 213
Takehara Y, Sekimoto A, Okano Y, et al. Explainable machine learning for the analysis of transport phenomena in top-seeded solution growth of SiC single crystal. J Ther Sci Technol, 2021, 16: JTST0009
Bartök AP, Kondor R, Csányi G. On representing chemical environments. Phys Rev B, 2013, 87: 184115
Gastegger M, Schwiedrzik L, Bittermann M, et al. wACSF—Weighted atom-centered symmetry functions as descriptors in machine learning potentials. J Chem Phys, 2018, 148: 241709
Ghiringhelli LM, Vybiral J, Levchenko SV, et al. Big data of materials science: Critical role of the descriptor. Phys Rev Lett, 2015, 114: 105503
Ward L, Agrawal A, Choudhary A, et al. A general-purpose machine learning framework for predicting properties of inorganic materials. npj Comput Mater, 2016, 2: 16028
Ward L, Liu R, Krishna A, et al. Including crystal structure attributes in machine learning models of formation energies via Voronoi tessellations. Phys Rev B, 2017, 96: 024104
De S, Bartök AP, Csányi G, et al. Comparing molecules and solids across structural and alchemical space. Phys Chem Chem Phys, 2016, 18: 13754–13769
Seko A, Maekawa T, Tsuda K, et al. Machine learning with systematic density-functional theory calculations: Application to melting temperatures of single- and binary-component solids. Phys Rev B, 2014, 89: 054303
Liang J, Zhu X. Phillips-inspired machine learning for band gap and exciton binding energy prediction. J Phys Chem Lett, 2019, 10: 5640–5646
M. Dieb T, Hou Z, Tsuda K. Structure prediction of boron-doped graphene by machine learning. J Chem Phys, 2018, 148: 241716
Lu S, Zhou Q, Ma L, et al. Rapid discovery of ferroelectric photovoltaic perovskites and material descriptors via machine learning. Small Methods, 2019, 3: 1900360
Choudhary K, Garrity KF, Reid ACE, et al. The joint automated repository for various integrated simulations (JARVIS) for data-driven materials design. npj Comput Mater, 2020, 6: 173
Hicks D, Toher C, Ford DC, et al. AFLOW-XtalFinder: A reliable choice to identify crystalline prototypes. npj Comput Mater, 2021, 7: 30
Ward L, Dunn A, Faghaninia A, et al. Matminer: An open source toolkit for materials data mining. Comput Mater Sci, 2018, 152: 60–69
Schutt KT, Glawe H, Brockherde F, et al. How to represent crystal structures for machine learning: Towards fast prediction of electronic properties. Phys Rev B, 2014, 89: 205118
Mathew K, Singh AK, Gabriel JJ, et al. MPInterfaces: A materials project based Python tool for high-throughput computational screening of interfacial systems. Comput Mater Sci, 2016, 122: 183–190
Morales-García Á, Valero R, Illas F. An empirical, yet practical way to predict the band gap in solids by using density functional band structure calculations. J Phys Chem C, 2017, 121: 18862–18866
Szegedy C, Zaremba W, Sutskever I, et al. Intriguing properties of neural networks. 2013, http://arxiv.org/abs/1312.6199
Liu X, Zhang J, Lin Y, et al. ATMPA: attacking machine learning-based malware visualization detection methods via adversarial examples. In: Proceedings of the International Symposium on Quality of Service. New York: ACM, 2019. 1–10
Kwon H, Yoon H, Park KW. Robust CAPTCHA image generation enhanced with adversarial example methods. IEICE Trans Inf Syst, 2020, E103.D: 879–882
Goodfellow I, Pouget-Abadie J, Mirza M, et al. Generative adversarial nets. 2014, http://arxiv.org/abs/1406.2661
Okhotnikov K, Charpentier T, Cadars S. Supercell program: A combinatorial structure-generation approach for the local-level modeling of atomic substitutions and partial occupancies in crystals. J Cheminform, 2016, 8: 17
Acknowledgements
This work was supported by Guangdong Basic and Applied Basic Research Foundation (2022A1515012006), Shenzhen Natural Science Fund (the Stable Support Plan Program 20231121110218001), the National Natural Science Foundation of China (12175150), and China Postdoctoral Science Foundation (2023M742403).
Author information
Authors and Affiliations
Contributions
Author contributions Jin H conceived the project. Wang T and Lai X carried out the DFT calculations. Wang T and Jin H wrote the manuscript. Jin H, Wei Y, and Guo H supervised the project. All authors analyzed the data and contributed to the discussions of the results.
Corresponding author
Ethics declarations
Conflict of interest The authors declare that they have no conflict of interest.
Additional information
Supplementary information Supporting data are available in the online version of the paper.
Tao Wang received his PhD degree from Shanxi University in 2022. His research focuses on the optical properties of materials and applying machine learning in the field of materials science.
Xiaolong Lai is an undergraduate student at the College of Physics and Optoelectronic Engineering, Shenzhen University, currently conducting research in the field of machine learning under the guidance of Professor Hao Jin.
Yadong Wei is a professor at the College of Physics and Optoelectronic Engineering, Shenzhen University. His research centers on tackling essential aspects of functional materials, encompassing electronic properties, optical characteristics, and energy conversion efficiencies, through the application of first-principles and machine-learning methodologies.
Hong Guo is the James McGill Professor of Physics at McGill University, and is a fellow of the Royal Society of Canada. His current research interests involve modeling in nanoelectronics, and materials physics of nanotechnology.
Hao Jin is an associate professor at the College of Physics and Optoelectronic Engineering, Shenzhen University. He obtained his PhD degree from the University of British Columbia in 2014. His current research interests focus on the development of machine-learning models for optoelectronic and photovoltaic applications.
Supporting Information for
40843_2023_2800_MOESM1_ESM.pdf
Enhancing Interpretability in the exploration of high-energy conversion efficiency in CsSnBr3−xIx configurations using crystal graph convolutional neural networks and adversarial example methods
Rights and permissions
About this article
Cite this article
Wang, T., Lai, X., Wei, Y. et al. Enhancing interpretability in the exploration of high-energy conversion efficiency in CsSnBr3−xIx configurations using crystal graph convolutional neural networks and adversarial example methods. Sci. China Mater. 67, 1183–1191 (2024). https://doi.org/10.1007/s40843-023-2800-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s40843-023-2800-x