Purpose of m_undrrelxr in AMReX_MLABecLaplacian.H

[ankithadas]

Hey AMReX team

I quite didn't understand the purpose of the variable m_undrrelxr in AMReX_MLCellLinOp.H. I can see it is used in abec_gsrb in MLABecLaplacian but I couldn't figure out where the values inside m_undrrelxr are assigned. I believe m_undrrelxr is initialized to quite_NaNs. My question is what values does this variable store and why is it used in the Gauss Seidel RB solver ? Any help would be much appreciated

Kind regards,
Ankith

[WeiqunZhang]

m_undrrelxr is initialized here

amrex/Src/LinearSolvers/MLMG/AMReX_MLCellLinOp.H

Line 1578 in f3598d2

MLCellLinOpT<MF>::prepareForSolve ()

, where it calls functions like

amrex/Src/LinearSolvers/MLMG/AMReX_MLLinOp_K.H

Line 377 in f3598d2

void mllinop_comp_interp_coef0_x (int side, int i, int j, int k, int blen,

.

It is used for convenience. For cells near the physical domain boundary or coarse/fine boundary, the stencil there is different. A normal 1d Laplacian stencil might be something like (1, -2, 1). But if the left point is outside a homogeneous Neumann boundary, the stencil becomes something like (N/A, -1, 1). In GSRB, -1 instead of -2 should be the diagonal component of the matrix. m_underrelxr is there to help us to go from -2 to -1 (or something else depending the boundary condition).