arXiv:2606.30773v1 Announce Type: cross Abstract: We introduce a novel technique for scalable sampling of spin-system states with continuous symmetries using diffusion models. By applying our approach to the XY model, a fundamental continuous-spin model in condensed matter physics, we show that our technique addresses the shortfalls of the Markov chain Monte Carlo (MCMC) in generalization to varying system sizes. More specifically, we show that training a temperature-conditioned diffusion model on smaller-size XY model lattices enables the generation of accurate samples in larger lattice sizes
This research is emerging now as diffusion models mature and researchers are exploring their applicability to fundamental physics problems, leveraging the advancements in generative AI techniques.
This development can significantly accelerate a foundational area of scientific computing (physics simulations) by enabling more efficient and scalable sampling methods, which are crucial for understanding complex systems and developing new materials.
The ability to accurately model spin-system states at larger scales and with higher efficiency than traditional MCMC methods means that computational bottlenecks for certain physics simulations may be substantially reduced, opening new avenues for research and discovery.
- · Condensed Matter Physicists
- · Materials Science Researchers
- · Computational Scientists
- · AI/ML Research Institutions
- · Providers of less efficient simulation software
Faster and more accurate simulations of complex physical systems will become possible.
This could lead to accelerated discovery of new materials with desired properties or a deeper understanding of phase transitions.
The methodology might generalize to other scientific domains where efficient sampling and state generation are critical, impacting areas beyond condensed matter physics.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG