Numerical simulation has become one of the most important tools in science and engineering. The available computer power has been steadily increasing, making it possible to do ever more complex simulations. However, in order to make efficient use of our computers, the state-of-the-art numerical algorithms must also be advanced to a similar degree. In this project, we propose a new hybrid technique designed to address partial differential equations whose solutions are largely smooth, but exhibit singularities or low regularity locally. Non-smoothness can for example occur near corners or cracks, or in media with randomly varying properties. Adaptive finite element methods (FEM) perform well for problems with low regularity, but may be computationally expensive for large-scale problems. Radial basis function (RBF) approximation methods are extremely efficient for smooth problems, but performance is degraded by low regularity. In the hybrid method, FEM and RBFs are employed where they work best, resulting in effective treatment of mixed regularity problems. The geometrical flexibility is inherent since FEM uses adaptive unstructured meshes and RBF is a meshfree method. The implementation of the algorithm will be based on domain decomposition and iterative solution techniques and parallelized to enable large-scale computations for realistic problems. The proposed project is a large undertaking and requires many different skills within the broad area of scientific computing. Together, the researcher Alfa Heryudono and the host research group at Uppsala University possess the required knowledge to succeed, and the fellowship phase is intended to be the beginning of a long-term collaboration between the researcher and the host research group.