Acceleration of Source Iteration using the Dynamic Mode Decomposition

We present a novel acceleration technique for improving the convergence of source iteration for discrete ordinates transport calculations. Our approach uses the idea of the dynamic mode decomposition (DMD) to estimate the slowly decaying modes from source iteration and remove them from the solution. The memory cost of our acceleration technique is that the scalar flux for a number of iterations must be stored; the computational cost is a singular value decomposition of a matrix comprised of those stored scalar fluxes.On 1-D slab geometry problems we observe an order of magnitude reduction in the number of transport sweeps required compared to source iteration and that the number of sweeps required is independent of the scattering ratio in the problem. These observations hold for an extremely heterogeneous problem and a 2-D problem. In 2-D we do observe that the effectiveness of the approach slowly degrades as the mesh is refined, but is still about one order of magnitude faster than source iteration.