Multi-property Feynman-Kac Steering of Crystal Diffusion Models
Abstract
Pretrained diffusion models provide powerful priors for generating new materials, including inorganic crystals, but generating useful materials requires satisfying specific chemical and physical properties, from mechanical stiffness to bandgaps. Existing approaches either retrain models for new target properties or steer only the atom coordinates and lattice during generation, leaving composition unchanged despite its fundamental role in determining physical properties. We present FK-ReMask, an inference-time, gradient-free method that steers both the continuous and discrete channels of a pretrained crystal diffusion model. During generation, FK-ReMask maintains multiple denoising trajectories and replicates promising candidates using Feynman-Kac particle selection. However, selection alone is insufficient: in the atom-type channel, committed elements cannot be revised, so cloning propagates existing compositions and limits the population's ability to explore new ones. We address this with an atom re-masking mechanism, which reopens committed sites during the generation process and allows alternative compositions to emerge. Across four electronic, photovoltaic, and mechanical objectives, our method improves the yield of SUN (stable, unique, novel) and property-on-target materials. We further demonstrate joint steering of three photovoltaic properties, showing that multiple physical objectives can be specified entirely at inference time without retraining.
Related
- DiffGA: Steering Frozen Diffusion Models for Materials Design via Evolutionary Search
- Symmetry- and Property-Aware Crystal Generation with Reinforcement Learning for Inverse Materials Design
- GEODE: Symmetry-Preserving Cartesian Diffusion for Crystal Generation
- Vacancy-mediated nitrogen diffusion and aggregation via high-fluence electron beam irradiation in HPHT synthesized diamond crystal
- Quantification of the crystallographic anisotropy and thermal dependence of Tb diffusion coefficients in the Nd2Fe14B phase through single-crystal diffusion couples
BibTeX
@misc{mousai2026multiproperty,
title = {{Multi-property Feynman-Kac Steering of Crystal Diffusion Models}},
author = {Oded Mousai and Sapir Esther Yiflach and Barak Itzchak Davidovitch and Yuval Mualem and Sarit Kraus and Hannah-Noa Barad and Karsten Kreis and Gal Chechik},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.24828},
eprint = {2610.24828},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.24828},
}
RIS
TY - UNPB
TI - Multi-property Feynman-Kac Steering of Crystal Diffusion Models
AU - Oded Mousai
AU - Sapir Esther Yiflach
AU - Barak Itzchak Davidovitch
AU - Yuval Mualem
AU - Sarit Kraus
AU - Hannah-Noa Barad
AU - Karsten Kreis
AU - Gal Chechik
PY - 2026
DA - 2026/10/17
PB - arXiv
DO - 10.48550/arxiv.2610.24828
UR - https://arxiv.org/abs/2610.24828
AB - Pretrained diffusion models provide powerful priors for generating new materials, including inorganic crystals, but generating useful materials requires satisfying specific chemical and physical properties, from mechanical stiffness to bandgaps. Existing approaches either retrain models for new target properties or steer only the atom coordinates and lattice during generation, leaving composition unchanged despite its fundamental role in determining physical properties. We present FK-ReMask, an inference-time, gradient-free method that steers both the continuous and discrete channels of a pretrained crystal diffusion model. During generation, FK-ReMask maintains multiple denoising trajectories and replicates promising candidates using Feynman-Kac particle selection. However, selection alone is insufficient: in the atom-type channel, committed elements cannot be revised, so cloning propagates existing compositions and limits the population's ability to explore new ones. We address this with an atom re-masking mechanism, which reopens committed sites during the generation process and allows alternative compositions to emerge. Across four electronic, photovoltaic, and mechanical objectives, our method improves the yield of SUN (stable, unique, novel) and property-on-target materials. We further demonstrate joint steering of three photovoltaic properties, showing that multiple physical objectives can be specified entirely at inference time without retraining.
ER -