Automated Adjoint-Looping Optimization with Pseudospectral Simulations

Physics-constrained optimization can be used to study fluids systems in the context of science and engineering. In both cases, we aim to study flows which exhibit specific behaviors. We combine adjoint-looping with the Dedalus pseudospectral PDE solver to perform nonlinear optimization of initial conditions and spatially-dependent parameters. We describe how Dedalus' features (such as functional differentiation, implicit/explicit timestepping, and domain decomposition) are well-suited for adjoint-looping. Using this framework in future, we aim to study a diverse set of problems related to time-inversion, turbulence, solar physics, and shape optimization.