Overview

Using Jax, demo of the adjoint optimization. The toy case is solving a Laplace equation with a simple central difference.

• forward.py demos the forward solving procedure and it is used to generate the truth solution.
• adjoint.py demos the adjoint optimization procedure.
• plot_T_K.ipynb visualizes the solution T and the design variables K.
• K_adj-N10-AbsErr15.npy is the K distribution generated by the adjoint optimization at the no. 100 iteration with the absolute error 15. For K_adj-N10-AbsErr300.npy, the absolution error is 300 at the no. 10 iteration.

The auto differentiation is employed by using Jax.

Reference: Inverting PDEs with adjoints by JoelCFD