Multilevel tensor approximation of PDEs with random data (Q1685682)

The authors are concerned about solving elliptic partial differential equations (PDEs) with a random diffusion coefficient. The infinite-dimensional random problem is first truncated to a finite-dimensional parametric PDE by approximating the random coefficient by the \(N\)-th term Karhunen-Loeve expansion. Then the truncated problem is approximated by a standard stochastic collocation scheme. The spatial approximation is done by a hierarchy of finite element discretizations and represents the solution as a telescoping sum. To reduce the computational cost for solving the discretized PDE on every finite element level, the authors introduce the adaptive low-rank tensor techniques to evaluate each difference in the multilevel sum. Combining the low-rank tensor approximation technique with the multilevel method is proved to be an effective method for solving the elliptic PDEs with random coefficients. Numerical results are provided to illustrate the new method and justify its effectiveness.