| 267 | 0 | 348 |
| 下载次数 | 被引频次 | 阅读次数 |
晶体材料的逆向设计一直是材料科学面临的重大挑战，近年来生成式人工智能技术的发展为解决这一难题带来了新的契机。本文围绕晶体表示方法与生成模型架构两大核心问题，系统综述了该领域的最新研究进展。在晶体表示方法方面，重点评述了文本符号化表示、几何图周期性编码以及像素的实验成像与计算像素化等方法，深入分析了这些方法在可逆性、不变性和对称性编码等关键技术特性上的优劣。在生成模型架构方面，按照技术演进路线，详细介绍了基于潜空间的变分自编码器和生成对抗网络、扩散或流匹配的联合生成模型，以及Transformer自回归模型的工作原理与应用现状。最后，针对当前面临的表示方法优化、缺陷材料建模和评估标准统一等关键科学问题，本文展望了面向工业应用的未来发展方向，旨在为晶体材料智能设计提供系统性的理论参考。
Abstract：The inverse design of crystalline materials aims to generate novel structures with targeted properties. However, achieving this objective remains a long-standing challenge in materials science. Unlike molecular systems that benefit from mature representations such as SMILES, crystalline materials impose stringent physical constraints, requiring representations that are simultaneously periodicity-aware, invertible, and invariant under symmetry operations. Emerging generative artificial intelligence(AI) technologies present transformative opportunities to address these complexities. This mini-review systematically examines the latest advancements in this field, centering on two critical pillars: crystal representation methods and generative model architectures. In terms of crystal representations, we critically evaluate text-based symbolic encodings(e.g., SLICES), geometry-based periodicity-aware graphs, and pixel-based imaging approaches. These methods are analyzed based on their ability to balance invertibility with symmetry preservation. With respect to generative model architectures, we trace the technological evolution from latent space-based models, such as Variational Autoencoders(VAE) and Generative Adversarial Networks(GAN), to state-of-the-art frameworks including diffusion models, flow-matching, and Transformer-based autoregressive models(e.g., MatterGen, MatterGPT). The review highlights how these architectures integrate physical constraints to navigate the vast chemical space of crystalline materials efficiently. Summary and Prospects Despite significant progress, the field of AI-driven crystal design faces critical bottlenecks. A primary challenge lies in the absence of a universally accepted representation framework that simultaneously satisfies invertibility, invariance, and periodicity encoding capabilities. Furthermore, current generative models predominantly focus on ideal crystal structures, often neglecting real-world complexities such as defects, impurities, and service environments. To bridge this gap and facilitate the transition from theoretical exploration to practical application, future research should focus on three key directions: 1) Physics-informed representation design: Developing representations that intrinsically embed crystallographic constraints, such as Wyckoff positions and space groups, to ensure symmetry-aware and invertible encodings. 2) High-fidelity modeling of real-world materials: Extending beyond idealized lattice structures to model realistic defects and environmental responses(e.g., temperature, pressure) by integrating symmetry-constrained data augmentation with molecular dynamics simulations. 3) Standardization of benchmarks: Establishing community-wide benchmarks that rigorously evaluate generation stability, novelty, diversity, and computational cost, ensuring objective comparisons across different chemical systems. Addressing these challenges will be pivotal in realizing the automated, intelligent design of functional crystalline materials.
[1]WALTERS W P, MURCKO M. Assessing the impact of generative AI on medicinal chemistry[J]. Nat Biotechnol, 2020, 38(2):143–145.
[2]DU Y Q, JAMASB A R, GUO J, et al. Machine learning-aided generative molecular design[J]. Nat Mach Intell, 2024, 6(6):589–604.
[3]WEININGER D. SMILES, a chemical language and information system. 1. Introduction to methodology and encoding rules[J]. J Chem Inf Comput Sci, 1988, 28(1):31–36.
[4]HELLER S R, MCNAUGHT A, PLETNEV I, et al. InChI, the IUPAC international chemical identifier[J]. J Cheminform, 2015, 7:23.
[5]KRENN M, HÄSE F, NIGAM A, et al. Self-referencing embedded strings(SELFIES):A 100%robust molecular string representation[J].Mach Learn:Sci Technol, 2020, 1(4):045024.
[6]ZHUNG W, KIM H, KIM W Y. 3D molecular generative framework for interaction-guided drug design[J]. Nat Commun, 2024, 15(1):2688.
[7]JIANG Y Y, ZHANG G, YOU J, et al. PocketFlow is a data-and-knowledge-driven structure-based molecular generative model[J]. Nat Mach Intell, 2024, 6(3):326–337.
[8]CHANG J, YE J C. Bidirectional generation of structure and properties through a single molecular foundation model[J]. Nat Commun, 2024,15(1):2323.
[9]WESTERMAYR J, GILKES J, BARRETT R, et al. High-throughput property-driven generative design of functional organic molecules[J].Nat Comput Sci, 2023, 3(2):139–148.
[10]PARK H, LI Z Z, WALSH A. Has generative artificial intelligence solved inverse materials design?[J]. Matter, 2024, 7(7):2355–2367.
[11]BUTLER K T, DAVIES D W, CARTWRIGHT H, et al. Machine learning for molecular and materials science[J]. Nature, 2018,559(7715):547–555.
[12]WANG Z L, YOU F Q. Leveraging generative models with periodicity-aware, invertible and invariant representations for crystalline materials design[J]. Nat Comput Sci, 2025, 5(5):365–376.
[13]ZENI C, PINSLER R, ZÜGNER D, et al. A generative model for inorganic materials design[J]. Nature, 2025, 639(8055):624–632.
[14]TANIAI T, IGARASHI R, SUZUKI Y, et al. Crystalformer:Infinitely connected attention for periodic structure encoding[J]. arXiv preprint arXiv, 2024,2403.11686.
[15]CHEN Y, WANG X, DENG X, et al. MatterGPT:A generative transformer for multi-property inverse design of solid-state materials[J].arXiv preprint arXiv, 2024,2408.07608
[16]XIAO H, LI R, SHI X Y, et al. An invertible, invariant crystal representation for inverse design of solid-state materials using generative deep learning[J]. Nat Commun, 2023, 14(1):7027.
[17]JIANG S L, DIENG A B, WEBB M A. Property-guided generation of complex polymer topologies using variational autoencoders[J]. NPJ Comput Mater, 2024, 10:139.
[18]ANSTINE D M, ISAYEV O. Generative models as an emerging paradigm in the chemical sciences[J]. J Am Chem Soc, 2023, 145(16):8736–8750.
[19]WANG K, GOU C, DUAN Y, et al. Generative adversarial networks:introduction and outlook[J]. IEEE/CAA J Automatic, 2017, 4(4):588–598.
[20]KRENN M, POLLICE R, GUO S Y, et al. On scientific understanding with artificial intelligence[J]. Nat Rev Phys, 2022, 4(12):761–769.
[21]HALL S R, ALLEN F H, BROWN I D. The crystallographic information file(CIF):A new standard archive file for crystallography[J]. Acta Crystallogr Sect A, 1991, 47(6):655–685.
[22]CAO Z D, LUO X S, LV J, et al. Space group informed transformer for crystalline materials generation[J]. Sci Bull, 2025, 70(21):3522–3533.
[23]GRUVER N, SRIRAM A, MADOTTO A, et al. Fine-Tuned language models generate stable inorganic materials as text[J]. arXiv E Prints,2024:arXiv:2402.04379.
[24]ANTUNES L M, BUTLER K T, GRAU-CRESPO R. Crystal structure generation with autoregressive large language modeling[J]. Nat Commun, 2024, 15(1):10570.
[25]CHUNG S J, HAHN T, KLEE W E. Nomenclature and generation of three-periodic nets:The vector method[J]. Acta Crystallogr A Found Crystallogr, 1984, 40(1):42–50.
[26]WANG B N, XU Z Y, HAN Z Y, et al. SLICES-PLUS:A crystal representation leveraging spatial symmetry[J]. Mater Des, 2025, 253:113856.
[27]ALAMPARA N, MIRET S, JABLONKA K M. MatText:Do language models need more than text&scale for materials modeling?[J]. arXiv preprint arXiv:2406.17295, 2024.
[28]FLAM-SHEPHERD D, ASPURU-GUZIK A. Language models can generate molecules, materials, and protein binding sites directly in three dimensions as XYZ, CIF, and PDB files[J]. arXiv E Prints, 2023:arXiv:2305.05708.
[29]LIU S, LI Y, LI Z, et al. Symmetry-informed geometric representation for molecules, proteins, and crystalline materials[C]//Advances in Neural Information Processing Systems, 2023, 36:66084–66101.
[30]DAMEWOOD J, KARAGUESIAN J, LUNGER J R, et al.Representations of materials for machine learning[J]. Annu Rev Mater Res, 2023, 53(1):399–426.
[31]PETERSEN M H, ZHU R M, DAI H W, et al. Dis-GEN:Disordered crystal structure generation[J]. arXiv:2507.18275.
[32]YANG Z D, LIU X Q, ZHANG X Y, et al. Modeling crystal defects using defect informed neural networks[J]. NPJ Comput Mater, 2025,11:229.
[33]LIU F Z, CHEN Z T, LIU T Y, et al. Self-supervised generative models for crystal structures[J]. iScience, 2024, 27(9):110672.
[34]YAMASHITA R, NISHIO M, DO R K G, et al. Convolutional neural networks:An overview and application in radiology[J]. Insights Imaging, 2018, 9(4):611–629.
[35]ZILETTI A, KUMAR D, SCHEFFLER M, et al. Insightful classification of crystal structures using deep learning[J]. Nat Commun,2018, 9(1):2775.
[36]YANG S H, CHOI W, CHO B W, et al. Deep learning-assisted quantification of atomic dopants and defects in 2D materials[J]. Adv Sci, 2021, 8(16):2101099.
[37]CHOUDHARY K, DECOST B, CHEN C, et al. Recent advances and applications of deep learning methods in materials science[J]. NPJ Comput Mater, 2022, 8:59.
[38]JANGID D K, BRODNIK N R, ECHLIN M P, et al. Q-RBSA:High-resolution 3D EBSD map generation using an efficient quaternion transformer network[J]. NPJ Comput Mater, 2024, 10:27.
[39]NOH J, KIM J, STEIN H S, et al. Inverse design of solid-state materials via a continuous representation[J]. Matter, 2019, 1(5):1370–1384.
[40]COURT C J, YILDIRIM B, JAIN A, et al. 3-D inorganic crystal structure generation and property prediction via representation learning[J]. J Chem Inf Model, 2020, 60(10):4518–4535.
[41]CASTELLI I E, LANDIS D D, THYGESEN K S, et al. New cubic perovskites for one-and two-photon water splitting using the computational materials repository[J]. Energy Environ Sci, 2012, 5(10):9034–9043.
[42]CASTELLI I E, OLSEN T, DATTA S, et al. Computational screening of perovskite metal oxides for optimal solar light capture[J]. Energy Environ Sci, 2012, 5(2):5814–5819.
[43]XIE T, FU X, GANEA O E, et al. Crystal diffusion variational autoencoder for periodic material generation[J]. arXiv E Prints, 2021:arXiv:2110.06197.
[44]PEI H, WEI B, CHANG K C C, et al. Geom-GCN:Geometric graph convolutional networks[J]. arXiv E Prints, 2020:arXiv:2002.05287.
[45]ZHAO W X, ZHOU K, LI J, et al. A survey of large language models[J]. arXiv E Prints, 2023:arXiv:2303.18223, 1(2).
[46]CROITORU F A, HONDRU V, IONESCU R T, et al. Diffusion models in vision:A survey[J]. IEEE Trans Pattern Anal Mach Intell,2023, 45(9):10850–10869.
[47]YAN D, SMITH A D, CHEN C C. Structure prediction and materials design with generative neural networks[J]. Nat Comput Sci, 2023, 3(7):572–574.
[48]KINGMA D P, WELLING M. Auto-encoding variational bayes[J].arXiv E Prints, 2013:arXiv:1312.6114.
[49]JAIN A, ONG S P, HAUTIER G, et al. Commentary:The materials project:A materials genome approach to accelerating materials innovation[J]. APL Mater, 2013, 1:011002.
[50]LIU K, GAO S D, YANG K F, et al. PCVAE:A physics-informed neural network for determining the symmetry and geometry of crystals[C]//2023 International Joint Conference on Neural Networks(IJCNN). Gold Coast, Australia. IEEE, 2023:1–8.
[51]MAL S, SEAL G, SEN P. MagGen:A graph-aided deep generative model for inverse design of permanent magnets[J]. J Phys Chem Lett,2024, 15(12):3221–3228.
[52]ZHU R M, NONG W, YAMAZAKI S, et al. WyCryst:Wyckoff inorganic crystal generator framework[J]. Matter, 2024, 7(10):3469–3488.
[53]GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al.Generative adversarial networks[J]. Commun ACM, 2020, 63(11):139–144.
[54]NOUIRA A, SOKOLOVSKA N, CRIVELLO J C. CrystalGAN:Learning to discover crystallographic structures with generative adversarial networks[J]. arXiv preprint arXiv. 2018, 1810.11203.
[55]KIM B, LEE S, KIM J. Inverse design of porous materials using artificial neural networks[J]. Sci Adv, 2020, 6(1):eaax9324.
[56]ZHAO Y, SIRIWARDANE E M D, WU Z Y, et al. Physics guided deep learning for generative design of crystal materials with symmetry constraints[J]. NPJ Comput Mater, 2023, 9:38.
[57]SU T H, CAO B, HU S B, et al. CGWGAN:Crystal generative framework based on Wyckoff generative adversarial network[J]. J Mater Inf, 2024, 4(4):20.
[58]YE Z, WANG N N, ZHOU J T, et al. Organic crystal structure prediction via coupled generative adversarial networks and graph convolutional networks[J]. Innovation, 2024, 5(2):100562.
[59]LI Z, BIRBILIS N. NSGAN:A non-dominant sorting optimisationbased generative adversarial design framework for alloy discovery[J].NPJ Comput Mater, 2024, 10:112.
[60]CHEN Z A, LI H C, ZHANG C, et al. Crystal structure prediction using generative adversarial network with data-driven latent space fusion strategy[J]. J Chem Theory Comput, 2024, 20(21):9627–9641.
[61]ARJOVSKY M, CHINTALA S, BOTTOU L. Wasserstein generative adversarial networks[C]//Proceedings of the 34th International Conference on Machine Learning–Volume 70. Sydney, NSW,Australia. ACM, 2017:214–223.
[62]LONG T, FORTUNATO N M, OPAHLE I, et al. Constrained crystals deep convolutional generative adversarial network for the inverse design of crystal structures[J]. NPJ Comput Mater, 2021, 7:66.
[63]CHEN L T, ZHANG W T, NIE Z W, et al. Generative models for inverse design of inorganic solid materials[J]. J Mater Inform, 2021, 1:4.
[64]SOHL-DICKSTEIN J, WEISS E A, MAHESWARANATHAN N, et al.Deep unsupervised learning using nonequilibrium thermodynamics[C]//International Conference on Machine Learning, France, 2015:2256–2265.
[65]HO J, JAIN A, ABBEEL P. Denoising Diffusion Probabilistic Models[C]//Advances in Neural Information Processing Systems, 2020,33:6840–6851.
[66]JANMOHAMED H, WOLINSKA M, SURANA S, et al.Multi-objective quality-diversity for crystal structure prediction[C]//Proceedings of the Genetic and Evolutionary Computation Conference,2024:1273–1281.
[67]Towards symmetry-aware generation of periodic materials[C]//Proceedings of the 37th International Conference on Neural Information Processing Systems. New Orleans, LA, USA. ACM, 2023:53308–53329.
[68]KLIPFEL A, FREGIER Y, SAYEDE A, et al. Vector field oriented diffusion model for crystal material generation[J]. Proc AAAI Conf Artif Intell, 2024, 38(20):22193–22201.
[69]PAKORNCHOTE T, CHOOMPHON-ANOMAKHUN N, ARRERUT S, et al. Diffusion probabilistic models enhance variational autoencoder for crystal structure generative modeling[J]. Sci Rep, 2024,14(1):1275.
[70]CHEN Z, YUAN Y, ZHENG S, et al. Transformer-enhanced variational autoencoder for crystal structure prediction[J]. ar Xiv E Prints, 2025:arXiv:2502.09423.
[71]LUO X S, WANG Z Y, GAO P Y, et al. Deep learning generative model for crystal structure prediction[J]. NPJ Comput Mater, 2024, 10:254.
[72]Crystal structure prediction by joint equivariant diffusion[C]//Proceedings of the 37th International Conference on Neural Information Processing Systems. ACM, 2023:17464–17497.
[73]JIAO R, HUANG W, LIU Y, et al. Space group constrained crystal generation[J]. arXiv E Prints, 2024:arXiv:2402.03992.
[74]Equivariant diffusion for crystal structure prediction[C]//Proceedings of the 41st International Conference on Machine Learning. ACM, 2024:29890–29913.
[75]LEVY D, PANIGRAHI S S, KABA S O, et al. SymmCD:Symmetry-preserving crystal generation with diffusion models[J].arXiv E Prints, 2025:arXiv:2502.03638.
[76]LIU Y, ZHOU C, ZHANG S, et al. Equivariant hypergraph diffusion for crystal structure prediction[J]. arXiv E Prints, 2025:arXiv:2501.18850.
[77]FU X, XIE T, ROSEN A S, et al. MOFDiff:Coarse-grained diffusion for metal-organic framework design[J]. arXiv E Prints, 2023:ar Xiv:2310.10732.
[78]SONG Y, SOHL-DICKSTEIN J, KINGMA D P, et al. Score-based generative modeling through stochastic differential equations.[J]. arXiv preprint arXiv, 2020, 2011.13456.
[79]MILLER B K, CHEN R T Q, SRIRAM A, et al. FlowMM:Generating materials with Riemannian flow matching[J]. arXiv preprint arXiv,2024, 2406.04731.
[80]LUO X S, WANG Z Y, WANG Q C, et al. CrystalFlow:A flow-based generative model for crystalline materials[J]. Nat Commun, 2025,16(1):9267.
[81]DEVLIN J, CHANG M W, LEE K, et al. BERT:Pre-training of deep bidirectional transformers for language understanding[C]//North American Chapter of the Association for Computational Linguistics.,2019
[82]VASWANI A, SHAZEER N, PARMAR N, et al. Attention is All you Need[C]//Advances in neural information processing systems, 2017:30.
[83]HU Y, BUEHLER M J. Deep language models for interpretative and predictive materials science[J]. APL Machine Learning, 2023, 1(1).
[84]PITIKE K C, MACIAS A, EISENBACH M, et al. Computationally accelerated discovery of high entropy pyrochlore oxides[J]. Chem Mater, 2022, 34(4):1459–1472.
[85]BREUCK P P D, PIRACHA H A, RIGNANESE G M, et al. A generative material transformer using Wyckoff representation[J]. arXiv preprint arXiv, 2025, 2501.16051.
[86]BERGERHOFF G, HUNDT R, SIEVERS R, et al. The inorganic crystal structure data base[J]. J Chem Inf Comput Sci, 1983, 23(2):66–69.
[87]CHOUDHARY K, DECOST B. Atomistic Line Graph Neural Network for improved materials property predictions[J]. NPJ Comput Mater, 2021, 7:185.
[88]ELMAN J L. Finding structure in time[J]. Cognitive science. 1990,14(2):179–211.
[89]ZHANG C, LV S Y, GONG H J, et al. Inverse design of high-performance piezoelectric semiconductors via advanced crystal representation and large language models[J]. Appl Phys Lett, 2025,126(11):111901.
[90]DENG Z, LI S, ZHANG T, et al. An Property-prompted multi-scale data augmentation approach for crystal representation[C]//ICLR 2025Workshop on Machine Learning Multiscale Processes, 2025.
[91]AYKOL M, BATZNER S, CUBUK E, et al. Generative hierarchical materials search[C]//Advances in Neural Information Processing Systems 37. Vancouver, BC, Canada. Neural Information Processing Systems Foundation, Inc.(NeurIPS), 2024:38799–38819.
[92]GAN J, ZHU Y, SCHWALBE-KODA D, et al. Large language models are innate crystal structure generators[C]//AI for Accelerated Materials Design-ICLR 2025, 2025.
[93]DING Q, MIRET S, LIU B, et al. Matexpert:Decomposing materials discovery by mimicking human experts[J]. arXiv preprint arXiv. 2024,2410.21317.
[94]ZHONG K, BU R, JIAO F, et al. Toward the defect engineering of energetic materials:A review of the effect of crystal defects on the sensitivity[J]. Chem Eng J, 2022, 429:132310.
[95]GU Q Q, ZHOUYIN Z, PANDEY S K, et al. Deep learning tight-binding approach for large-scale electronic simulations at finite temperatures with ab initio accuracy[J]. Nat Commun, 2024, 15(1):6772.
[96]YANG M Y, RAUCCI U, PARRINELLO M. Reactant-induced dynamics of lithium imide surfaces during the ammonia decomposition process[J]. Nat Catal, 2023, 6(9):829–836.
[97]DE BREUCK P P, WANG H C, RIGNANESE G M, et al. Generative AI for crystal structures:A review[J]. NPJ Comput Mater, 2025, 11:370.
基本信息:
DOI：10.14062/j.issn.0454-5648.20250845
中图分类号:TP18;TB30
引用信息:
[1]吴克登,陈炎,邓晓彬,等.AI驱动的材料发现：生成式模型与晶体逆向设计进展[J].硅酸盐学报,2026,54(01):90-105.DOI:10.14062/j.issn.0454-5648.20250845.
基金信息:
国家重点研发计划（2024YFA1209801）; 国家自然科学基金（22203066,12302140）; 中国博士后科学基金（2023M732794,2025T180517）,中国博士后科学基金博士后创新人才支持计划B档（GZB20230575）; 中央高校基本科研业务费专项资金（sxzy012023213）; 陕西省科学技术厅人工智能赋能科研专项经费（2025YXYC012）; 西安市科学技术协会青年人才托举计划（959202413069）
2025-11-15
2025
2025-12-16
2025
1
2026-01-05
2026-01-05
2026-01-05