{"title":"用深度学习方法求解 PDE 的自适应轨迹采样","authors":"Xingyu Chen , Jianhuan Cen , Qingsong Zou","doi":"10.1016/j.amc.2024.128928","DOIUrl":null,"url":null,"abstract":"<div><p>In this paper, we propose a novel adaptive technique, named <em>adaptive trajectories sampling</em> (ATS), to select training points for learning the solution of partial differential equations (PDEs). By the ATS, the training points are selected adaptively according to an empirical-value-type instead of residual-type error indicator from trajectories which are generated by a PDE-related stochastic process. We incorporate the ATS into three known deep learning solvers for PDEs, namely, the adaptive physics-informed neural network method (ATS-PINN), the adaptive derivative-free-loss method (ATS-DFLM), and the adaptive temporal-difference method for forward-backward stochastic differential equations (ATS-FBSTD). Our numerical experiments show that the ATS remarkably improves the computational accuracy and efficiency of the original deep solvers. In particular, for a high-dimensional peak problem, the relative errors by the ATS-PINN can achieve the order of <span><math><mi>O</mi><mo>(</mo><msup><mrow><mn>10</mn></mrow><mrow><mo>−</mo><mn>3</mn></mrow></msup><mo>)</mo></math></span>, even when the vanilla PINN fails.</p></div>","PeriodicalId":55496,"journal":{"name":"Applied Mathematics and Computation","volume":"481 ","pages":"Article 128928"},"PeriodicalIF":3.5000,"publicationDate":"2024-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Adaptive trajectories sampling for solving PDEs with deep learning methods\",\"authors\":\"Xingyu Chen , Jianhuan Cen , Qingsong Zou\",\"doi\":\"10.1016/j.amc.2024.128928\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><p>In this paper, we propose a novel adaptive technique, named <em>adaptive trajectories sampling</em> (ATS), to select training points for learning the solution of partial differential equations (PDEs). By the ATS, the training points are selected adaptively according to an empirical-value-type instead of residual-type error indicator from trajectories which are generated by a PDE-related stochastic process. We incorporate the ATS into three known deep learning solvers for PDEs, namely, the adaptive physics-informed neural network method (ATS-PINN), the adaptive derivative-free-loss method (ATS-DFLM), and the adaptive temporal-difference method for forward-backward stochastic differential equations (ATS-FBSTD). Our numerical experiments show that the ATS remarkably improves the computational accuracy and efficiency of the original deep solvers. In particular, for a high-dimensional peak problem, the relative errors by the ATS-PINN can achieve the order of <span><math><mi>O</mi><mo>(</mo><msup><mrow><mn>10</mn></mrow><mrow><mo>−</mo><mn>3</mn></mrow></msup><mo>)</mo></math></span>, even when the vanilla PINN fails.</p></div>\",\"PeriodicalId\":55496,\"journal\":{\"name\":\"Applied Mathematics and Computation\",\"volume\":\"481 \",\"pages\":\"Article 128928\"},\"PeriodicalIF\":3.5000,\"publicationDate\":\"2024-07-15\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Applied Mathematics and Computation\",\"FirstCategoryId\":\"100\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S0096300324003898\",\"RegionNum\":2,\"RegionCategory\":\"数学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"MATHEMATICS, APPLIED\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Applied Mathematics and Computation","FirstCategoryId":"100","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0096300324003898","RegionNum":2,"RegionCategory":"数学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"MATHEMATICS, APPLIED","Score":null,"Total":0}
Adaptive trajectories sampling for solving PDEs with deep learning methods
In this paper, we propose a novel adaptive technique, named adaptive trajectories sampling (ATS), to select training points for learning the solution of partial differential equations (PDEs). By the ATS, the training points are selected adaptively according to an empirical-value-type instead of residual-type error indicator from trajectories which are generated by a PDE-related stochastic process. We incorporate the ATS into three known deep learning solvers for PDEs, namely, the adaptive physics-informed neural network method (ATS-PINN), the adaptive derivative-free-loss method (ATS-DFLM), and the adaptive temporal-difference method for forward-backward stochastic differential equations (ATS-FBSTD). Our numerical experiments show that the ATS remarkably improves the computational accuracy and efficiency of the original deep solvers. In particular, for a high-dimensional peak problem, the relative errors by the ATS-PINN can achieve the order of , even when the vanilla PINN fails.
期刊介绍:
Applied Mathematics and Computation addresses work at the interface between applied mathematics, numerical computation, and applications of systems – oriented ideas to the physical, biological, social, and behavioral sciences, and emphasizes papers of a computational nature focusing on new algorithms, their analysis and numerical results.
In addition to presenting research papers, Applied Mathematics and Computation publishes review articles and single–topics issues.