1: \begin{abstract}

2:     Numerically solving high-dimensional random parametric PDEs poses a challenging computational problem.

3:     It is well-known that numerical methods can greatly benefit from adaptive refinement algorithms, in particular when functional approximations in polynomials are computed as in stochastic Galerkin finite element methods.

4:     This work investigates a residual based adaptive algorithm, akin to classical adaptive FEM, used to approximate the solution of the stationary diffusion equation with lognormal coefficients, i.e. with a non-affine parameter dependence of the data.

5:     It is known that the refinement procedure is reliable but the theoretical convergence of the scheme for this class of unbounded coefficients remains a challenging open question.

6:     This paper advances the theoretical state-of-the-art by providing a quasi-error reduction result for the adaptive solution of the lognormal stationary diffusion problem.

7:     The presented analysis generalizes previous results in that guaranteed convergence for uniformly bounded coefficients follows directly as a corollary.

8:     Moreover, it highlights the fundamental challenges with unbounded coefficients that cannot be overcome with common techniques.

9:     A computational benchmark example illustrates the main theoretical statement.

10: \end{abstract}

11: