CrystalREPA: Transferring Physical Priors from Universal MLIPs to Crystal Generative Models
Abstract
Crystal generative models mainly learn what stable crystals look like, with little explicit supervision for what makes them stable. We reveal a substantial representation gap between state-of-the-art crystal generative models and pretrained universal machine learning interatomic potentials (MLIPs) via energy probing, and show this gap can be closed by a simple training-time alignment. We propose Crystal REPresentation Alignment (CrystalREPA), a plug-and-play framework that aligns the atom-wise hidden states of generative encoders with frozen MLIP representations through an element-aware contrastive objective, transferring stability-aware atomistic priors with marginal training overhead and no additional inference cost. Across three generative frameworks, ten MLIP teachers, and two benchmark datasets, CrystalREPA consistently improves the thermodynamic stability, structural validity, and structural fidelity of generated crystals. Equally important, we find that an MLIP's transfer effectiveness is poorly predicted by its accuracy on standard leaderboards (e.g., Matbench Discovery) but strongly predicted by the distinguishability of its atom-wise representation space, yielding a practical, accuracy-independent criterion for selecting MLIP teachers for generative transfer. Code is available at https://github.com/Chengqian-Zhang/CrystalREPA.
Related
- CrystalGRW: generative modeling of crystal structures with targeted crystallographic properties via geodesic random walks
- Crystal-to-crystal solar energy storage in a hydrogen bonded dianthracene lattice retrocyclized by heat or e-beam
- Limits of geometric packing descriptors in molecular crystal energy ranking
- The CPOSS Database of Computed Crystal Structures
- A database-wide atlas of single-crystal filters for thermal neutron beams
BibTeX
@misc{zhang2026crystalrepa,
title = {{CrystalREPA: Transferring Physical Priors from Universal MLIPs to Crystal Generative Models}},
author = {Chengqian Zhang and Yucheng Jin and Duo Zhang and Tiejun Li and Han Wang},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2605.08960},
eprint = {2605.08960},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2605.08960},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - CrystalREPA: Transferring Physical Priors from Universal MLIPs to Crystal Generative Models
AU - Chengqian Zhang
AU - Yucheng Jin
AU - Duo Zhang
AU - Tiejun Li
AU - Han Wang
PY - 2026
DA - 2026/10/26
PB - arXiv
DO - 10.48550/arxiv.2605.08960
UR - https://arxiv.org/abs/2605.08960
AB - Crystal generative models mainly learn what stable crystals look like, with little explicit supervision for what makes them stable. We reveal a substantial representation gap between state-of-the-art crystal generative models and pretrained universal machine learning interatomic potentials (MLIPs) via energy probing, and show this gap can be closed by a simple training-time alignment. We propose Crystal REPresentation Alignment (CrystalREPA), a plug-and-play framework that aligns the atom-wise hidden states of generative encoders with frozen MLIP representations through an element-aware contrastive objective, transferring stability-aware atomistic priors with marginal training overhead and no additional inference cost. Across three generative frameworks, ten MLIP teachers, and two benchmark datasets, CrystalREPA consistently improves the thermodynamic stability, structural validity, and structural fidelity of generated crystals. Equally important, we find that an MLIP's transfer effectiveness is poorly predicted by its accuracy on standard leaderboards (e.g., Matbench Discovery) but strongly predicted by the distinguishability of its atom-wise representation space, yielding a practical, accuracy-independent criterion for selecting MLIP teachers for generative transfer. Code is available at https://github.com/Chengqian-Zhang/CrystalREPA.
ER -