A piecewise neural network method for solving large interval solution to initial value problem of ordinary differential equations

TL;DR

This paper introduces a piecewise neural network method that divides the interval of an initial value problem for differential equations into smaller parts, enabling large interval solutions with improved accuracy and efficiency.

Contribution

The paper proposes a novel piecewise neural network approach that constructs large interval solutions without increasing network size or training data, improving over existing neural methods.

Findings

The method guarantees continuous differentiability over the entire interval.

It enhances approximation accuracy through parameter transfer and pre-training.

Numerical experiments demonstrate the method's efficiency and effectiveness.

Abstract

Various traditional numerical methods for solving initial value problems of differential equations often produce local solutions near the initial value point, despite the problems having larger interval solutions. Even current popular neural network algorithms or deep learning methods cannot guarantee yielding large interval solutions for these problems. In this paper, we propose a piecewise neural network approach to obtain a large interval numerical solution for initial value problems of differential equations. In this method, we first divide the solution interval, on which the initial problem is to be solved, into several smaller intervals. Neural networks with a unified structure are then employed on each sub-interval to solve the related sub-problems. By assembling these neural network solutions, a piecewise expression of the large interval solution to the problem is constructed, referred to as the piecewise neural network solution. The continuous differentiability of the solution over the entire interval, except for finite points, is proven through theoretical analysis and employing a parameter transfer technique. Additionally, a parameter transfer and multiple rounds of pre-training technique are utilized to enhance the accuracy of the approximation solution. Compared with existing neural network algorithms, this method does not increase the network size and training data scale for training the network on each sub-domain. Finally, several numerical experiments are presented to demonstrate the efficiency of the proposed algorithm.