UEDGE Points of Investigation: Needs of a scalable preconditioner How is convergence affected by … Access to the exact Jacobian via coloring? Adding neutral gas terms? Decreasing the time step? The method of domain partitioning? The answers have improved our results from the solid lines to the dashed lines
UEDGE Investigation Results: Designing a scalable preconditioner Exploiting the exact Jacobian via coloring ASM is now a viable (scalable) solver What effect do neutral gases play? Neutral gases often demand a direct solver Choosing a reasonable time step? ASM converges scalably for Δt≈10-4 1D vs. 2D domain decomposition Without neutrals, 1D is preferred Our PCFieldSplit solves the neutral terms with LU and other terms with ASM. Scalability LU FS 4.82 7.98