dolfin-adjoint 2018.1: automated adjoints for FEniCS and Firedrake

Abstract

Adjoint models play an important role in scientific computing. They enable for instance sensitivity and stability analysis, goal-oriented mesh adaptivity and optimisation. However, the derivation and implementation of adjoint models is challenging, especially for models governed by non-linear or time-dependent partial differential equations (PDEs). In The implementation, named dolfin-adjoint, automatically and robustly derives adjoint models for models written in the finite element software FEniCS However, the assumption that the model consists of a sequence of variational problems can be limiting. For instance when considering Dirichlet boundary conditions that are not explicitly stated in the variational formulation, when considering complex functionals that cannot be represented as an integral, or when coupling FEniCS to other non-PDE models.