Most simulators in industry revolve around efficiently solving a set of PDEs that describe the physical phenomena under consideration. For reservoir simulation this would be the mass conservation equations that govern the flow of hydrocarbons in subsurface reservoirs. Many of these codes have been around for a long time and have slowly evolved in feature richness. This makes these codes of great value from a user point of view but numerically very challenging. The linearization of the PDEs often lead to very ill conditioned systems and these systems are gradually getting worse as grid sizes and problem complexity grow. In this talk I will highlight the challenges faced when developing a linear solver for an industrial application as it has to be scalable, robust, virtually parameter free and ready to take advantage of next generation hardware. I will give some examples from industry and approaches explored.  

Tom Jönsthövel is HPC Architect for Schlumberger and works on improving the driving algorithms of commercial reservoir simulators by combining developments in numerical methods, hardware and software. He holds a PhD in Applied Mathematics from Delft University of Technology where he did his research on developing scalable iterative solvers for structural mechanics supervised by Kees Vuik.