On optimal regularization methods for the backward heat equation (Q1915228)

In a Hilbert space setting the abstract heat equation \[ u_t + A u = 0 \quad (0 \leq t < T) \] backward in time is considered as an operator equation \[ K(t) q(t) = z, \] where the temperature \(q(t) = u(x,t)\) for \(0 \leq t < T\) has to be determined from data \(z(x) =\) \(u(x,T)\). The operator \(A\) is assumed to be linear, densely defined, selfadjoint and positive with a discrete spectrum and eigenvalues \(\lambda_i\) tending to infinity as \(i \to \infty\). The authors especially take into account the case \(A = -\Delta\) with homogeneous boundary conditions in the Hilbert space \(L^2\). For noisy data and under the a priori condition \(|q(0)|\leq E\) the error of regularized solutions in a rather general sense is evaluated. All regularization methods considered depend on a regularization parameter \(\alpha\) which can be chosen such that optimal error bounds are obtained. In detail, theorems are proven on optimality for the method of Tikhonov regularization and iterated versions of this method, for the method of asymptotical regularization and some quasireversibility methods.