Abstract:
We consider the problem of constructing a quasiclassical variational functional for a one-dimensional homogeneous wave equation in a pentagon-shaped domain. Using the variational method for hyperbolic equations proposed by V. M. Filippov, we obtain a variational functional involving path and iterated integrals in the characteristic variables. This form of the variational functional is adapted for training neural networks that approximate solutions of boundary-value problems in mathematical physics, and increases the efficiency and speed of learning.
Keywords:variational principle, wave equation, neural network, loss functional