Magnetic crystal generation with site-resolved magnetic moments via Gaussian-mixture retargeting and finite-budget yield prediction
Abstract
We introduce GMMag, a workflow for generating crystal structures together with their site-resolved magnetic moments, built on a conditional variational autoencoder with a graph neural network encoder and a transformer decoder. Gaussian-mixture sampling redirects generation towards selected chemical and magnetic targets without retraining. We systematically investigate how encoder and decoder design, the representation of magnetic-moment magnitudes and relative signs, and sampling choices affect reconstruction, generation quality and diversity. Because a higher screening pass rate does not necessarily yield more qualifying candidates with distinct chemical formulas absent from the reference dataset, we model how repeated generation of the same chemical formulas limits their accumulation and test the model in two separate ten-million-draw runs. As an application, the workflow identifies Nd-containing candidates with chemical formulas absent from the reference dataset, negative formation energies and high estimated saturation magnetization.
Related
- Momentum-resolved magnetic noise spectroscopy using ensembles of diamond quantum sensors
- High-resolution angle-resolved photoemission spectroscopy with tunable magnetic field
- Enhancement of axial magnetic field generation during relativistic self-channeling of laser radiation propagating along thin films
- Magnetic thresholds and spin-resolved spectral moments of ϕ and K*0 in a hadronic model
- Automated Generation of Commensurate Magnetic Structures based on Spin Space Groups and Graph Theory
BibTeX
@misc{shang2026magnetic,
title = {{Magnetic crystal generation with site-resolved magnetic moments via Gaussian-mixture retargeting and finite-budget yield prediction}},
author = {Wen-Chuang Shang and Chenyu Shen and Mikito Tsujimoto and Tomonori Tanaka and Yoshihiro Gohda},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.37222},
eprint = {2610.37222},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.37222},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Magnetic crystal generation with site-resolved magnetic moments via Gaussian-mixture retargeting and finite-budget yield prediction
AU - Wen-Chuang Shang
AU - Chenyu Shen
AU - Mikito Tsujimoto
AU - Tomonori Tanaka
AU - Yoshihiro Gohda
PY - 2026
DA - 2026/10/27
PB - arXiv
DO - 10.48550/arxiv.2610.37222
UR - https://arxiv.org/abs/2610.37222
AB - We introduce GMMag, a workflow for generating crystal structures together with their site-resolved magnetic moments, built on a conditional variational autoencoder with a graph neural network encoder and a transformer decoder. Gaussian-mixture sampling redirects generation towards selected chemical and magnetic targets without retraining. We systematically investigate how encoder and decoder design, the representation of magnetic-moment magnitudes and relative signs, and sampling choices affect reconstruction, generation quality and diversity. Because a higher screening pass rate does not necessarily yield more qualifying candidates with distinct chemical formulas absent from the reference dataset, we model how repeated generation of the same chemical formulas limits their accumulation and test the model in two separate ten-million-draw runs. As an application, the workflow identifies Nd-containing candidates with chemical formulas absent from the reference dataset, negative formation energies and high estimated saturation magnetization.
ER -