We present a novel way of constructing reduced models for systems of ordinary differential equations. The reduced models we construct depend on coefficients which measure the importance of the different terms appearing in the model and need to be estimated. The proposed approach allows the estimation of these coefficients on the fly by enforcing the equality of integral quantities of the solution as computed from the original system and the reduced model. In particular, the approach combines the concepts of renormalization and effective field theory developed in the context of high energy physics and the Mori-Zwanzig formalism of irreversible statistical mechanics. It allows to construct stable reduced models of higher order than was previously possible. The method is applied to the problem of computing reduced models for ordinary differential equation systems resulting from Fourier expansions of singular (or near-singular) time-dependent partial differential equations. Results for the 1D Burgers and the 3D incompressible Euler equations are used to illustrate the construction. We also present, for the 1D Burgers and the 3D Euler equations, a simple and efficient recursive algorithm for calculating the higher order terms.
High precision PINNs in unbounded domains: application to singularity formulation in PDEs
California Institute of Technology(加州理工学院);Massachusetts Institute of Technology(麻省理工学院);The NSF Institute for Artificial Intelligence and Fundamental Interactions(NSF人工智能与基础相互作用研究所)
A Unified Weighted-Loss Physics-Informed Neural Network for Boundary Layer Problems in Singularly Perturbed PDEs
National Central University(中央大学);National Center for Theoretical Sciences, National Taiwan University(台湾大学理论科学中心);National Chung Hsing University(中兴大学);National Yang Ming Chiao Tung University(阳明交通大学)
Adaptive feature capture method for solving partial differential equations with near singular solutions
Sichuan University(四川大学);The Chinese University of Hong Kong, Shenzhen(香港中文大学(深圳));Shenzhen International Center for Industrial and Applied Mathematics(深圳工业与应用数学中心);Shenzhen Research Institute of Big Data(深圳大数据研究院)
Weak TransNet: A Petrov-Galerkin based neural network method for solving elliptic PDEs
虽然深度学习在求解偏微分方程(PDE)方面取得了显著的成功,但它仍然面临着重大挑战,特别是当PDE解具有低正则性或奇异性时。为了解决这些问题,我们提出了基于Petrov-Galerkin公式的弱TransNet(WTN)方法,用于在本工作中求解椭圆偏微分方程,尽管其框架可以扩展到其他类型的方程。具体地,由TransNet定义的神经特征空间(Zhang et al.,2023)作为试验空间,而测试空间由径向基函数组成。由于解被表示为测试函数的线性组合,因此可以通过经由最小二乘法最小化弱PDE残差来确定系数。因此,这种方法可以帮助减轻神经网络训练中经常出现的非凸性和病态性的挑战。此外,WTN方法被扩展到处理的解决方案具有多尺度功能或具有急剧变化的问题。几个数值实验证明所提出的方法的鲁棒性和效率。