Robust Variational Physics-Informed Neural Networks
share
We introduce a Robust version of the Variational Physics-Informed Neural Networks (RVPINNs) to approximate the Partial Differential Equations (PDEs) solution. We start from a weak Petrov-Galerkin formulation of the problem, select a discrete test space, and define a quadratic loss functional as in VPINNs. Whereas in VPINNs the loss depends upon the selected basis functions of a given test space, herein we minimize a loss based on the residual in the discrete dual norm, which is independent of the test space's choice of test basis functions. We demonstrate that this loss is a reliable and efficient estimator of the true error in the energy norm. The proposed loss function requires computation of the Gram matrix inverse, similar to what occurs in traditional residual minimization methods. To validate our theoretical findings, we test the performance and robustness of our algorithm in several advection-dominated-diffusion problems in one spatial dimension. We conclude that RVPINNs is a robust method.
READ FULL TEXT
