Structural constraint integration in a generative model for the discovery of quantum materials

Ryotaro Okabe; Mouyang Cheng; Abhijatmedhi Chotrattanapituk; Manasi Mandal; Kiran Mak; Denisse Córdova Carrizales; Nguyen Tuan Hung; Xiang Fu; Bowen Han; Yao Wang; Weiwei Xie; Robert J. Cava; Tommi S. Jaakkola; Yongqiang Cheng; Mingda Li

doi:10.1038/s41563-025-02355-y

Structural constraint integration in a generative model for the discovery of quantum materials

Ryotaro Okabe
, Mouyang Cheng
, Abhijatmedhi Chotrattanapituk
, Manasi Mandal
, Kiran Mak
, Denisse Córdova Carrizales
, Nguyen Tuan Hung
, Xiang Fu
, Bowen Han
, Yao Wang
, Weiwei Xie
, Robert J. Cava
, Tommi S. Jaakkola
, Yongqiang Cheng
, Mingda Li

Chemical Spectroscopy

Research output: Contribution to journal › Article › peer-review

6 Scopus citations

Abstract

Billions of organic molecules have been computationally generated, yet functional inorganic materials remain scarce due to limited data and structural complexity. Here we introduce Structural Constraint Integration in a GENerative model (SCIGEN), a framework that enforces geometric constraints, such as honeycomb and kagome lattices, within diffusion-based generative models to discover stable quantum materials candidates. SCIGEN enables conditional sampling from the original distribution, preserving output validity while guiding structural motifs. This approach generates ten million inorganic compounds with Archimedean and Lieb lattices, over 10% of which pass multistage stability screening. High-throughput density functional theory calculations on 26,000 candidates shows over 95% convergence and 53% structural stability. A graph neural network classifier detects magnetic ordering in 41% of relaxed structures. Furthermore, we synthesize and characterize two predicted materials, TiPd0.22Bi0.88 and Ti0.5Pd1.5Sb, which display paramagnetic and diamagnetic behaviour, respectively. Our results indicate that SCIGEN provides a scalable path for generating quantum materials guided by lattice geometry.

Original language	English
Pages (from-to)	223-230
Number of pages	8
Journal	Nature Materials
Volume	25
Issue number	2
DOIs	https://doi.org/10.1038/s41563-025-02355-y
State	Published - Feb 1 2026

Funding

R.O. and M.L. thank C. Batista, A. Christianson, F. Frenkel, A. May, R. Moore, B. Ortiz and F. Ronning for helpful discussions. R.O. acknowledges support from the US Department of Energy (DOE), Office of Science (SC), Basic Energy Sciences (BES), award number DE-SC0021940 and the Heiwa Nakajima Foundation. A.C. acknowledges support from National Science Foundation (NSF) Designing Materials to Revolutionize and Engineer Our Future (DMREF) Program with award number DMR-2118448. B.H. and Y.C. are partially supported by the Artificial Intelligence Initiative as part of the Laboratory Directed Research and Development (LDRD) program of Oak Ridge National Laboratory (ORNL), managed by UT-Battelle, LLC, for the US Department of Energy under contract DE-AC05-00OR22725. This research used resources of the National Energy Research Scientific Computing Center (NERSC), a Department of Energy Office of Science User Facility using NERSC award DDR-ERCAP0030758. Computing resources for a portion of the work were made available through the VirtuES project, funded by the LDRD Program and Compute and Data Environment for Science (CADES) at ORNL. Another portion of simulation results were obtained using the Frontera computing system at the Texas Advanced Computing Center. W.X. and R.C. were supported by the Department of Energy, grant DE-FG02-98ER45706. W.X. and R.C. thank G. J. Miller for offering clusters to perform LMTO calculations. M.L. acknowledges the support from NSF ITE-2345084, the Class of 1947 Career Development Chair and support from R. Wachnik.

Access to Document

10.1038/s41563-025-02355-y

Cite this

@article{07454c375c274bc08d10b9bed0e89638,

title = "Structural constraint integration in a generative model for the discovery of quantum materials",

abstract = "Billions of organic molecules have been computationally generated, yet functional inorganic materials remain scarce due to limited data and structural complexity. Here we introduce Structural Constraint Integration in a GENerative model (SCIGEN), a framework that enforces geometric constraints, such as honeycomb and kagome lattices, within diffusion-based generative models to discover stable quantum materials candidates. SCIGEN enables conditional sampling from the original distribution, preserving output validity while guiding structural motifs. This approach generates ten million inorganic compounds with Archimedean and Lieb lattices, over 10\% of which pass multistage stability screening. High-throughput density functional theory calculations on 26,000 candidates shows over 95\% convergence and 53\% structural stability. A graph neural network classifier detects magnetic ordering in 41\% of relaxed structures. Furthermore, we synthesize and characterize two predicted materials, TiPd0.22Bi0.88 and Ti0.5Pd1.5Sb, which display paramagnetic and diamagnetic behaviour, respectively. Our results indicate that SCIGEN provides a scalable path for generating quantum materials guided by lattice geometry.",

author = "Ryotaro Okabe and Mouyang Cheng and Abhijatmedhi Chotrattanapituk and Manasi Mandal and Kiran Mak and \{C{\'o}rdova Carrizales\}, Denisse and Hung, \{Nguyen Tuan\} and Xiang Fu and Bowen Han and Yao Wang and Weiwei Xie and Cava, \{Robert J.\} and Jaakkola, \{Tommi S.\} and Yongqiang Cheng and Mingda Li",

note = "Publisher Copyright: {\textcopyright} 2025. The Author(s), under exclusive licence to Springer Nature Limited.",

year = "2026",

month = feb,

day = "1",

doi = "10.1038/s41563-025-02355-y",

language = "English",

volume = "25",

pages = "223--230",

journal = "Nature Materials",

issn = "1476-1122",

publisher = "Nature Research",

number = "2",

}

TY - JOUR

T1 - Structural constraint integration in a generative model for the discovery of quantum materials

AU - Okabe, Ryotaro

AU - Cheng, Mouyang

AU - Chotrattanapituk, Abhijatmedhi

AU - Mandal, Manasi

AU - Mak, Kiran

AU - Córdova Carrizales, Denisse

AU - Hung, Nguyen Tuan

AU - Fu, Xiang

AU - Han, Bowen

AU - Wang, Yao

AU - Xie, Weiwei

AU - Cava, Robert J.

AU - Jaakkola, Tommi S.

AU - Cheng, Yongqiang

AU - Li, Mingda

PY - 2026/2/1

Y1 - 2026/2/1

N2 - Billions of organic molecules have been computationally generated, yet functional inorganic materials remain scarce due to limited data and structural complexity. Here we introduce Structural Constraint Integration in a GENerative model (SCIGEN), a framework that enforces geometric constraints, such as honeycomb and kagome lattices, within diffusion-based generative models to discover stable quantum materials candidates. SCIGEN enables conditional sampling from the original distribution, preserving output validity while guiding structural motifs. This approach generates ten million inorganic compounds with Archimedean and Lieb lattices, over 10% of which pass multistage stability screening. High-throughput density functional theory calculations on 26,000 candidates shows over 95% convergence and 53% structural stability. A graph neural network classifier detects magnetic ordering in 41% of relaxed structures. Furthermore, we synthesize and characterize two predicted materials, TiPd0.22Bi0.88 and Ti0.5Pd1.5Sb, which display paramagnetic and diamagnetic behaviour, respectively. Our results indicate that SCIGEN provides a scalable path for generating quantum materials guided by lattice geometry.

AB - Billions of organic molecules have been computationally generated, yet functional inorganic materials remain scarce due to limited data and structural complexity. Here we introduce Structural Constraint Integration in a GENerative model (SCIGEN), a framework that enforces geometric constraints, such as honeycomb and kagome lattices, within diffusion-based generative models to discover stable quantum materials candidates. SCIGEN enables conditional sampling from the original distribution, preserving output validity while guiding structural motifs. This approach generates ten million inorganic compounds with Archimedean and Lieb lattices, over 10% of which pass multistage stability screening. High-throughput density functional theory calculations on 26,000 candidates shows over 95% convergence and 53% structural stability. A graph neural network classifier detects magnetic ordering in 41% of relaxed structures. Furthermore, we synthesize and characterize two predicted materials, TiPd0.22Bi0.88 and Ti0.5Pd1.5Sb, which display paramagnetic and diamagnetic behaviour, respectively. Our results indicate that SCIGEN provides a scalable path for generating quantum materials guided by lattice geometry.

UR - https://www.scopus.com/pages/publications/105016815818

U2 - 10.1038/s41563-025-02355-y

DO - 10.1038/s41563-025-02355-y

M3 - Article

AN - SCOPUS:105016815818

SN - 1476-1122

VL - 25

SP - 223

EP - 230

JO - Nature Materials

JF - Nature Materials

IS - 2

ER -

Structural constraint integration in a generative model for the discovery of quantum materials

Abstract

Funding

Access to Document

Other files and links

Fingerprint

Cite this