A GPU-based spectral-element solver for gaseous low-Mach-number combustion

A GPU-accelerated high-order spectral element code is developed to solve the low-Mach-number chemically reactive Navier-Stokes equations. The formulation, based on the open-source code nekRS, includes detailed chemistry and transport, and a Strang splitting method to decouple the flow and chemistry sub steps, enabling quicker computations while preserving temporal accuracy. A variety of Runge-Kutta explicit integration techniques are employed for the chemistry integration, and computational speedup numbers are presented. The numerical approach is first validated through simulations of canonical 0D and 1D hydrogen-air flames. To assess code performance for more practical problems encountered in gas turbine combustion, unsteady 3D simulations are conducted for non-premixed hydrogen-air swirling jet flames at moderate Reynolds numbers. Strong and weak scaling results are presented, along with an investigation into the effects of the global equivalence ratio on flame stability.