Adjoint Method

A procedure for computing gradients or sensitivities of scalar-valued outputs (or functionals) with respect to inputs, parameters, or initial/boundary data by solving an associated adjoint equation that runs backward in time or uses the transpose of linearized operators.