SerpApi.Org
In recent years, research using artificial intelligence to predict material properties has advanced rapidly. Neural network quantum Monte Carlo methods have attracted attention as highly accurate simulation techniques. However, their extremely high computational cost has limited their application to small molecular systems. This study introduces a new computational method that overcomes this limitation.
Subscribe to our newsletter for the latest sci-tech news updates.
Methods that simulate electron-level mechanisms on supercomputers are widely used to explore novel materials and understand biological phenomena. There is strong demand for new approaches that can deliver faster predictions while maintaining high accuracy.
Lowering the cost of precision
In a recent breakthrough, researchers at Japan Advanced Institute of Science and Technology (JAIST), in collaboration with ByteDance Seed in China, combined neural network techniques with a newly developed "Bayesian localization of pseudo Hamiltonian" approach, achieving accurate predictions at reduced computational cost.
The research team included associate professor Tom Ichibha and doctoral student Ryunosuke Fujimaru (one of the co-first authors) from JAIST, together with researchers from ByteDance Seed. The findings were published online in Nature Computational Science.
"By integrating AI techniques into research fields that have traditionally been advanced through physics and chemistry, significant progress has been achieved. This method is expected to contribute to the discovery of novel materials and the understanding of biological phenomena, and the present results will greatly advance research in these areas," Ichibha said.
Broader reach for quantum simulations
This work has yielded significant results for the advancement of next-generation simulation technologies that integrate AI techniques with quantum chemical calculations. The developed method opens possibilities for the high-precision analysis of large-scale materials and complex chemical reaction systems—tasks that were previously difficult because of computational resource constraints.
Future efforts to expand the scope to a wider range of elements are expected to facilitate applications across diverse fields, such as the discovery of novel materials, the design of high-performance catalysts and the elucidation of biomolecular functions.
Applications in solid-state physics and excited-state calculations are also anticipated, promising to contribute to resolving unsolved problems in quantum science and materials science.
More information: Weizhong Fu et al, Empowering neural network-based quantum Monte Carlo with local pseudopotentials, Nature Computational Science (2026). DOI: 10.1038/s43588-026-01008-7
Provided by Japan Advanced Institute of Science and Technology
This story was originally published on Phys.org.
