A HYBRID SPECTRAL DEFERRED CORRECTION METHOD WITH PHYSICS-INFORMED NEURAL NETWORK INITIALIZATION
Litsenziya: cc-by-nc
Annotatsiya
Spectral Deferred Correction (SDC) methods are iterative methods used to solve ordinary differential equations. The SDC method can achieve high-order convergence by adjusting the initial guess, quadrature nodes, and the number of iterations. Therefore, its convergence strongly depends on the initial approximation. In this work, we introduce a hybrid method in which a Physics-Informed Neural Network (PINN) provides the initial guess for SDC iterations. We present a rigorous formulation of both SDC and PINN, along with a convergence analysis of the proposed method. Finally, we validate our results using numerical examples. The proposed method significantly reduces the number of SDC sweeps required to achieve high-order accuracy.
Hali tarjima qilinmagan