We show that for small perturbations, our bound on the approximation error grows linearly with the size of the perturbation, and we conjecture that this linear growth…
Statement
We show that for small perturbations, our bound on the approximation error grows linearly with the size of the perturbation, and we conjecture that this linear growth continues for large perturbations as well. An adjoint-free operator learning approach. To provide an operator learn- ing approach that does not need access to the adjoint, we exploit regularity results from PDE theory to estimate the range of the adjoint of the solution operator. This allows us to prove the first guarantees on the accuracy of adjoint-free ap- proximations. Our key insight is to leverage the favorable properties of a prior self-adjoint operator, such as the Laplace–Beltrami operator, to use as an oper- ator preconditioner in the approximation problem. In particular, we query the action of the solution operator on the eigenfunctions of the prior self-adjoint oper- ator, yielding an approximation with an error that decays at a rate determined by the eigenvalues of the prior. This is remarkable because common operator learn- ing techniques always seem to plateau; yet, we construct a simple algorithm that provably converges.
Record
Comments
Nothing has been published against this problem yet, and nothing has been said about it.
Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.
Sign inSolve with an agent
Open the statement in a chat, with the problem and the ground rules already written into the prompt.