CbLDM: A diffusion model for recovering nanostructure from atomic pair distribution functions

The nanostructure inverse problem is a compelling challenge that helps researchers to understand the structure–property relationships of nanomaterials. This work focuses on recovering a model system of monometallic nanoparticles from their pair distribution function (PDF) data, framing the task as a highly ill-posed conditional generation problem. We introduce a Condition-based Latent Diffusion Model as a practical solution. The model incorporates an accelerated sampling strategy that initializes generation from a conditional-prior sample rather than from pure Gaussian noise, allowing the sampler to skip a contiguous block of intermediate denoising steps, reducing inference time by 8% under the default conservative setting, with a maximum speedup of 27% achievable at the cost of marginal accuracy loss. In addition, we employ a Laplacian matrix in place of a conventional distance matrix to represent nanostructures, which enhances recovery stability. Our results demonstrate that a latent diffusion model with a conditional prior can generate physically meaningful nanostructures consistent with PDF observations, laying the groundwork for solving more complex inverse problems in the future work.

Authors

Institutions

Publication Details

Journal
APL Machine Learning
Published
2026-11-18
DOI
https://doi.org/10.1063/5.0332941
Primary Topic
Machine Learning in Materials Science
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

CbLDM: A diffusion model for recovering nanostructure from atomic pair distribution functions

Jian Cao, Simon J. L. Billinge, Heming Wang, Zhiyang Zhang et al.
2 links
APL Machine Learning
Machine Learning in Materials Science
article

CbLDM: A diffusion model for recovering nanostructure from atomic pair distribution functions

Jian Cao, Simon J. L. Billinge, Heming Wang, Zhiyang Zhang, Gu Ran, Ling Lan
article en

Abstract

The nanostructure inverse problem is a compelling challenge that helps researchers to understand the structure–property relationships of nanomaterials. This work focuses on recovering a model system of monometallic nanoparticles from their pair distribution function (PDF) data, framing the task as a highly ill-posed conditional generation problem. We introduce a Condition-based Latent Diffusion Model as a practical solution. The model incorporates an accelerated sampling strategy that initializes generation from a conditional-prior sample rather than from pure Gaussian noise, allowing the sampler to skip a contiguous block of intermediate denoising steps, reducing inference time by 8% under the default conservative setting, with a maximum speedup of 27% achievable at the cost of marginal accuracy loss. In addition, we employ a Laplacian matrix in place of a conventional distance matrix to represent nanostructures, which enhances recovery stability. Our results demonstrate that a latent diffusion model with a conditional prior can generate physically meaningful nanostructures consistent with PDF observations, laying the groundwork for solving more complex inverse problems in the future work.

APL Machine LearningVol. 4(4)
University of California, Santa Barbara (US), Nankai University (CN), Columbia University (US)
U.S. Department of Energy, National Natural Science Foundation of China, Natural Science Foundation of Tianjin City, Office of Science, National Key Research and Development Program of China, Fundamental Research Funds for the Central Universities, Basic Energy Sciences
Machine Learning in Materials Science
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon