Predicting the synthesizability of inorganic crystals by bridging crystal graphs and phonon dynamics

doi:10.1088/1674-1056/ae0161

中国物理B ›› 2026, Vol. 35 ›› Issue (1): 16101-016101.doi: 10.1088/1674-1056/ae0161

Predicting the synthesizability of inorganic crystals by bridging crystal graphs and phonon dynamics

Mei Ma(马梅)¹, Wei Ma(马薇)², Le Gao(高乐)³, Zong-Guo Wang(王宗国)⁴, and Hao Liu(刘昊)^5,†

1 School of Physics, Ningxia University, Yinchuan 750021, China;
2 School of Materials and New Energy, Ningxia University, Yinchuan 750021, China;
3 School of Information Engineering, Ningxia University, Yinchuan 750021, China;
4 Computer Network Information Center, Chinese Academy of Sciences, Beijing 100083, China;
5 School of Artificial Intelligence, Ningxia University, Yinchuan 750021, China

收稿日期:2025-06-17 修回日期:2025-08-20 接受日期:2025-09-02 发布日期:2025-12-30
通讯作者: Hao Liu E-mail:liuhao@nxu.edu.cn

Predicting the synthesizability of inorganic crystals by bridging crystal graphs and phonon dynamics

Mei Ma(马梅)¹, Wei Ma(马薇)², Le Gao(高乐)³, Zong-Guo Wang(王宗国)⁴, and Hao Liu(刘昊)^5,†

1 School of Physics, Ningxia University, Yinchuan 750021, China;
2 School of Materials and New Energy, Ningxia University, Yinchuan 750021, China;
3 School of Information Engineering, Ningxia University, Yinchuan 750021, China;
4 Computer Network Information Center, Chinese Academy of Sciences, Beijing 100083, China;
5 School of Artificial Intelligence, Ningxia University, Yinchuan 750021, China

Received:2025-06-17 Revised:2025-08-20 Accepted:2025-09-02 Published:2025-12-30
Contact: Hao Liu E-mail:liuhao@nxu.edu.cn

摘要/Abstract

摘要： Accurately predicting the synthesizability of inorganic crystal materials serves as a pivotal tool for the efficient screening of viable candidates, substantially reducing the costs associated with extensive experimental trial-and-error processes. However, existing methods, limited by static structural descriptors such as chemical composition and lattice parameters, fail to account for atomic vibrations, which may introduce spurious correlations and undermine predictive reliability. Here, we propose a deep learning model termed integrating graph and dynamical stability (IGDS) for predicting the synthesizability of inorganic crystals. IGDS employs graph representation learning to construct crystal graphs that precisely capture the static structures of crystals and integrates phonon spectral features extracted from pre-trained machine learning interatomic potentials to represent their dynamic properties. Our model exhibits outstanding performance in predicting the synthesizability of low-energy unsynthesizable crystals across 41 material systems, achieving precision and recall values of 0.916/0.863 for ternary compounds. By capturing both static structural descriptors and dynamic features, IGDS provides a physics-informed method for predicting the synthesizability of inorganic crystals. This approach bridges the gap between theoretical design concepts and their practical implementation, thereby streamlining the development cycle of new materials and enhancing overall research efficiency

关键词: crystal synthesizability prediction, deep learning, graph learning, AI for science

Abstract: Accurately predicting the synthesizability of inorganic crystal materials serves as a pivotal tool for the efficient screening of viable candidates, substantially reducing the costs associated with extensive experimental trial-and-error processes. However, existing methods, limited by static structural descriptors such as chemical composition and lattice parameters, fail to account for atomic vibrations, which may introduce spurious correlations and undermine predictive reliability. Here, we propose a deep learning model termed integrating graph and dynamical stability (IGDS) for predicting the synthesizability of inorganic crystals. IGDS employs graph representation learning to construct crystal graphs that precisely capture the static structures of crystals and integrates phonon spectral features extracted from pre-trained machine learning interatomic potentials to represent their dynamic properties. Our model exhibits outstanding performance in predicting the synthesizability of low-energy unsynthesizable crystals across 41 material systems, achieving precision and recall values of 0.916/0.863 for ternary compounds. By capturing both static structural descriptors and dynamic features, IGDS provides a physics-informed method for predicting the synthesizability of inorganic crystals. This approach bridges the gap between theoretical design concepts and their practical implementation, thereby streamlining the development cycle of new materials and enhancing overall research efficiency

Key words: crystal synthesizability prediction, deep learning, graph learning, AI for science

中图分类号: (Theory of crystal structure, crystal symmetry; calculations and modeling)

61.50.Ah

63.20.D- (Phonon states and bands, normal modes, and phonon dispersion) 81.05.Zx (New materials: theory, design, and fabrication) 07.05.Mh (Neural networks, fuzzy logic, artificial intelligence)

Mei Ma(马梅), Wei Ma(马薇), Le Gao(高乐), Zong-Guo Wang(王宗国), and Hao Liu(刘昊). Predicting the synthesizability of inorganic crystals by bridging crystal graphs and phonon dynamics[J]. 中国物理B, 2026, 35(1): 16101-016101.

参考文献

[1] Nzereogu P U, Omah A D, Ezema F I, Iwuoha E I and Nwanya A C 2022 Appl. Surf. Sci. Adv. 9 100233
[2] Li X Y, Weng S T and Lin G 2020 Chin. Phys. B 29 028801
[3] Ding K, Yang T, Leung M T, Yang K, Cheng H, Zeng M, Li B and Yang M 2023 Curr. Opin. Electrochem. 42 101404
[4] Jo B, Chen W and Jung H S 2025 J. Energy Chem. 101 298
[5] Luo Y, Bag S, Zaremba O, Cierpka A, Andreo J, Wuttke S, Friederich P and Tsotsalas M 2022 Angew. Chem. Int. Ed. 61 e202200242
[6] Therrien F, Jones E B and Stevanovic V 2021 Appl. Phys. Rev. 8 031310
[7] Bartel C J, Millican S L, Deml A M, Rumptz J R, Tumas W, Weimer A W, Lany S, Stevanovic V, Musgrave C B and Holder A M 2018 Nat. Commun. 9 4168
[8] Chen J, Calvin J, Asplund M, King S W, Woodfield B F and Navrotsky A 2019 J. Chem. Thermodyn. 128 320
[9] Sun W, Dacek S T, Ong S P, Hautier G, Jain A, Richards W D, Gamst A C, Persson K A and Ceder G 2016 Sci. Adv. 2 e1600225
[10] Aykol M, Dwaraknath S S, Sun W and Persson K A 2018 Sci. Adv. 4 eaaq0148
[11] Antoniuk E R, Cheon G, Wang G, Bernstein D, Cai W and Reed E J 2023 npj Comput. Mater. 9 155
[12] Davariashtiyani A, Kadkhodaie Z and Kadkhodaei S 2021 Commun. Mater. 2 115
[13] Zhu R, Tian S I P, Ren Z, Li J, Buonassisi T and Hippalgaonkar K 2023 ACS Omega 8 8210
[14] Xie T and Grossman J C 2018 Phys. Rev. Lett. 120 145301
[15] Jang J, Gu G H, Noh J, Kim J and Jung Y 2020 J. Am. Chem. Soc. 142 18836
[16] Kittel C and McEuen P 2018 Introduction to Solid State Physics (8th edn.) (Hoboken: John Wiley & Sons)
[17] Tolborg K, Klarbring J, Ganose A M and Walsh A 2022 Digit. Discov. 1 586
[18] Jain A, Ong S P, Hautier G, Chen W, Richards W D, Dacek S, Cholia S, Gunter D, Skinner D, Ceder G and Persson K A 2013 APL Mater. 1 011002
[19] Schutt K T, Arbabzadah F, Chmiela S, M uller K R and Tkatchenko A 2017 Nat. Commun. 8 13890
[20] Rupp M, Tkatchenko A, Muller K R and von Lilienfeld O A 2012 Phys. Rev. Lett. 108 058301
[21] Gorfman S 2020 Acta Crystallogr. A: Found. Adv. 76 713
[22] Choudhary K and DeCost B 2021 npj Comput. Mater. 7 185
[23] Stoffel R P, Wessel C, Lumey M W and Dronskowski R 2010 Angew. Chem. Int. Ed. 49 5242
[24] Wallace D C and Callen H 1972 Am. J. Phys. 40 1718
[25] Giustino F 2017 Rev. Mod. Phys. 89 015003
[26] Batatia I, Benner P, Chiang Y, et al. 2023 arXiv:2401.00096 [physics.chem-ph]
[27] Batatia I, Kovacs D P, Simm G, Ortner C and Csanyi G 2022 Advances in Neural Information Processing Systems, November 28–December 9, 2022 New Orleans, USA, pp. 11423
[28] Deng B, Zhong P, Jun K, Riebesell J, Han K, Bartel C J and Ceder G 2023 Nat. Mach. Intell. 5 1031
[29] Togo A, Chaput L, Tadano T and Tanaka I 2023 J. Phys.: Condens. Matter 35 353001
[30] Togo A 2022 J. Phys. Soc. Jpn. 92 012001
[31] Kingma D P and Ba J 2015 International Conference on Learning Representations, May 7–9, 2015 San Diego, USA, pp. 13
[32] Loshchilov I and Hutter F 2019 International Conference on Learning Representations, May 6–9, 2019 New Orleans, USA, pp. 11254–11263
[33] Smith L N and Topin N 2018 International Conference on Learning Representations Workshop, April 30–May 3, 2018 Vancouver, Canada, (Workshop paper)
[34] Chen C, Ye W, Zuo Y, Zheng C and Ong S P 2019 Chem. Mater. 31 3564
[35] Hellenbrandt M 2004 Crystallogr. Rev. 10 17
[36] Scheirer W J, Jain L P and Boult T E 2014 IEEE Trans. Pattern Anal. Mach. Intell. 36 2317
[37] Yan K, Liu Y, Lin Y and Ji S 2022 Advances in Neural Information Processing Systems, November 28–December 9, 2022 New Orleans, USA, p. 15066
[38] Choudhary K, Garrity K F, Reid A C E, et al. 2020 npj Comput. Mater. 6 173
[39] Tshitoyan V, Dagdelen J, Weston L, Dunn A, Rong Z, Kononova O, Persson K A, Ceder G and Jain A 2019 Nature 571 95
[40] Wei F, Deng Z, Sun S, Hartono N T, Seng H, Buonassisi T, Bristowe P and Cheetham A 2019 Chem. Commun. 55 3721
[41] Volonakis G, Haghighirad A A, Milot R L, Sio W H, Filip M R, Wenger B, Johnston M B, Herz L M, Snaith H J and Giustino F 2017 J. Phys. Chem. Lett. 8 772
[42] Kegler V D and Fiscehr S C 2018 Int. J. Adv. Eng. Res. Sci. 5 349
[43] Qi J, Ko T W, Wood B C, Pham T A and Ong S P 2024 npj Comput. Mater. 10 43
[44] Setyawan W and Curtarolo S 2010 Comput. Mater. Sci. 49 299

Predicting the synthesizability of inorganic crystals by bridging crystal graphs and phonon dynamics

Predicting the synthesizability of inorganic crystals by bridging crystal graphs and phonon dynamics

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

Metrics

本文评价

推荐阅读 0

[1]	Yan Zou(邹艳), Lang Yang(杨浪), Yanhui Liu(刘艳辉), and Yuyu Feng(冯玉宇). RLsite: Integrating 3D-CNN and BiLSTM for RNA-ligand binding site prediction[J]. 中国物理B, 2025, 34(8): 88709-088709.
[2]	Zengxuan Zhao(赵曾轩), Xiuying Zhang(张秀颖), Pengchen Zhao(赵鹏琛), Chunyang Wang(王春阳), Chunlei Xia(夏春雷), Mushtaq Rana Imran, and Joelous Malamula Nyasulu. Trajectory tracking on the optimal path of two-dimensional quadratic barrier escaping[J]. 中国物理B, 2025, 34(5): 50205-050205.
[3]	Haolin Yang(杨昊霖), Bo Ni(倪波), Junhong Guo(郭俊宏), Hua Zhou(周华), and Jianhua Chang(常建华). Deep learning-enabled inverse design of polarization-selective structural color based on coding metasurface[J]. 中国物理B, 2025, 34(5): 50702-050702.
[4]	Wenchao Weng(翁文超), Hanyu Jiang(蒋涵羽), Xiangjie Kong(孔祥杰), and Giovanni Pau. Text-guided diverse-expression diffusion model for molecule generation[J]. 中国物理B, 2025, 34(5): 50701-050701.
[5]	Yanan Guo(郭亚楠), Xiaoqun Cao(曹小群), Hongze Leng(冷洪泽), and Junqiang Song(宋君强). Learning complex nonlinear physical systems using wavelet neural operators[J]. 中国物理B, 2025, 34(3): 34702-034702.
[6]	Dezhi Lu(鲁德志), Hao Wu(吴淏), Yutong Hou(侯俞彤), Yuncheng Wu(吴云成), Yuanyuan Liu(刘媛媛), and Jinwu Wang(王金武). Accurate prediction of essential proteins using ensemble machine learning[J]. 中国物理B, 2025, 34(1): 18901-018901.
[7]	Wenhao Yuan(袁文浩), Cheng Peng(彭程), and Qian He(何迁). A large language model-powered literature review for high-angle annular dark field imaging[J]. 中国物理B, 2024, 33(9): 98703-098703.
[8]	Kang Liu(刘炕), Cheng Zhou(周成), Jipeng Huang(黄继鹏), Hongwu Qin(秦宏伍), Xuan Liu(刘轩), Xinwei Li(李鑫伟), and Lijun Song(宋立军). High-quality ghost imaging based on undersampled natural-order Hadamard source[J]. 中国物理B, 2024, 33(9): 94204-094204.
[9]	Xi Wang(王玺), Meng Tang(唐孟), Ming-Xuan Jiang(蒋明璇), Yang-Chun Chen(陈阳春), Zhi-Xiao Liu(刘智骁), and Hui-Qiu Deng(邓辉球). Properties of radiation defects and threshold energy of displacement in zirconium hydride obtained by new deep-learning potential[J]. 中国物理B, 2024, 33(7): 76103-076103.
[10]	Xiaoyu Cheng(程晓昱), Chenxue Xie(解晨雪), Yulun Liu(刘宇伦), Ruixue Bai(白瑞雪), Nanhai Xiao(肖南海), Yanbo Ren(任琰博), Xilin Zhang(张喜林), Hui Ma(马惠), and Chongyun Jiang(蒋崇云). Image segmentation of exfoliated two-dimensional materials by generative adversarial network-based data augmentation[J]. 中国物理B, 2024, 33(3): 30703-030703.
[11]	Zhi-Gang Zheng(郑志刚), Fei-Fei Han(韩菲菲), Le Wang(王乐), and Sheng-Mei Zhao(赵生妹). Generation of orbital angular momentum hologram using a modified U-net[J]. 中国物理B, 2024, 33(3): 34207-034207.
[12]	Haowen Xiao(肖皓文) and Zhiguang Han(韩枝光). Quantum state estimation based on deep learning[J]. 中国物理B, 2024, 33(12): 120307-120307.
[13]	Denghui Peng(彭登辉), Shenlong Wang(王神龙), and Yuanchen Huang(黄元辰). A deep learning method based on prior knowledge with dual training for solving FPK equation[J]. 中国物理B, 2024, 33(1): 10202-10202.
[14]	Jiamei Cui(崔佳梅), Yunjie Li(李韵洁), Cai Zhao(赵偲), and Wen Zheng(郑文). Classification and structural characteristics of amorphous materials based on interpretable deep learning[J]. 中国物理B, 2023, 32(9): 96101-096101.
[15]	Tian Wang(王田), Jiahui Chen(陈家辉), Jing Teng(滕婧), Jingang Shi(史金钢),Xinhua Zeng(曾新华), and Hichem Snoussi. Crysformer: An attention-based graph neural network for properties prediction of crystals[J]. 中国物理B, 2023, 32(9): 90703-090703.