{"id":"arxiv:2310.08439","kind":"preprint","title":"TensorMD: Scalable Tensor-Diagram based Machine Learning Interatomic Potential on Heterogeneous Many-Core Processors","authors":["Xin Chen","Yucheng Ouyang","Zhenchuan Chen","Rongfen Lin","Xingyu Gao","Lifang Wang","Fang Li","Yin Liu","Honghui Shang","Haifeng Song"],"author_count":10,"year":2023,"published":"2023-10-12","venue":null,"cited_by":1,"doi":null,"arxiv_id":"2310.08439","nct_id":null,"source_url":"https://arxiv.org/abs/2310.08439","archive_url":"https://space-kitty.com/papers/arxiv:2310.08439","updated":"2023-10-13","pmid":null,"pmcid":null,"openalex_id":"W4387635833","pdf_url":"https://arxiv.org/pdf/2310.08439","abstract":"Molecular dynamics simulations have emerged as a potent tool for investigating the physical properties and kinetic behaviors of materials at the atomic scale, particularly in extreme conditions. Ab initio accuracy is now achievable with machine learning based interatomic potentials. With recent advancements in high-performance computing, highly accurate and large-scale simulations become feasible. This study introduces TensorMD, a new machine learning interatomic potential (MLIP) model that integrates physical principles and tensor diagrams. The tensor formalism provides a more efficient computation and greater flexibility for use with other scientific codes. Additionally, we proposed several portable optimization strategies and developed a highly optimized version for the new Sunway supercomputer. Our optimized TensorMD can achieve unprecedented performance on the new Sunway, enabling simulations of up to 52 billion atoms with a time-to-solution of 31 ps/step/atom, setting new records for HPC + AI + MD.","abstract_license":"CC0-1.0 (arXiv descriptive metadata)","license":null,"is_open_access":false,"retracted":false,"peer_reviewed":false,"categories":["physics.comp-ph","cs.DC"],"details":null,"provenance":"arxiv","attribution":"Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.","topics":[{"slug":"ai-for-science","name":"AI for Science","domain_slug":"ai"}],"cited_in_dossiers":[]}