That is, for all δp ∈ P the functional sensitivities can be computed via a single linear solution for ψ rather than repeated application of the linearized forward operator, F'(p)δp.