1: \begin{abstract}

2: In many practical applications like parameter estimation problems in PDEs or the autoconvolution problem, the considered operator is indeed monotone on a closed convex subset $ \mathcal{M} \subsetneq \mathcal{H} $ only. Since the Lavrentiev regularized equation may not have a solution $ v_\alpha^\delta \in \mathcal{M} $ under such general assumptions, we replace this equation by the corresponding regularized variational inequality and consider $ u_\alpha^\delta \in \mathcal{M} $ satisfying

3: %

4: $$ 

5: \langle F u_\alpha^\delta + \alpha u_\alpha^\delta - f^\delta , w - u_\alpha^\delta \rangle \ge 0 \quad \textup{for each } w \in \mathcal{M}.

6: $$

7: If a solution $ u_* \in \mathcal{M} $ of the unperturbed equation $ Fu = f $ admits a source representation either of the form $ u_* = F'(u_*)z $ or $ u_* = F'(u_*)^* z $, with some $ z \in \mathcal{H} $, where $ F'(u_*)^* $ denotes the adjoint operator of the Fr\'{e}chet derivative $ F'(u_*) $ of $ F $ at $ u_* $, then convergence rates for the error $ \Vert u_\alpha^\delta  - u_* \Vert $ are available for suitable a priori parameter choices $ \alpha = \alpha(\delta) $. In this talk we present some new results in this direction.

8: \end{abstract}

9: