“…A mean approximation, combined possibly with a Chebyshev semi-iteration, is shown to be sufficient to efficiently approximate this inverse for quite a wide range of parameters, leading to practical P LRM , P LRC , P OPM , P OPC preconditioners, where the subscript M stands for "mean" and C for Chebyshev. We remark that the development of robust preconditioners for small values of the regularization parameter is not obvious and poses some interesting mathematical and computational challenges which, surprisingly, are similar to those encountered in deterministic OCP when the control acts locally, either on a portion of the domain [14], or on a portion of the boundary [20]. We further stress that our analysis does not assume that the random bilinear form is uniformly bounded and coercive with respect to the randomness, which is a frequent simplifying hypothesis in the literature, see, e.g., [44,55,3].…”