Also always use -snes_linesearch_monitor in these cases. Also when something goes wrong in a nonlinear solver with a fancy preconditioner on the linear solver it is good to make a run with a direct solver to make sure the issue is not with the preconditioner; so -pc_type lu
> On May 15, 2015, at 2:04 PM, Dmitry Karpeyev <[email protected]> wrote: > > What does -ksp_monitor_true_residual show? > > On Fri, May 15, 2015 at 1:45 PM Fande Kong <[email protected]> wrote: > Hi all, > > I was wondering why we will get a zero residual when solving the linear step > of a nonlinear problem? I am solving a fluid-structure interaction problems > using inexact Newton method, within which fGMRES is used together with a > multilevel Schwarz preconditioner. The algorithm works well when the mesh is > small, but it catched a zero residual issue when we turn to use a little bit > large mesh. The output is listed as follows: > > 0 SNES Function norm 1.291775443677e-02 > Jacobian: rows 5575441 Column 5575441 > call DMGetInterpolation_SpmcsDMMesh > Jacobian: rows 1298448 Column 1298448 > 0 KSP Residual norm 1.291775443677e-02 > 1 KSP Residual norm 0.000000000000e+00 > Nonlinear solve did not converge due to DIVERGED_LINE_SEARCH iterations 0 > > > Any suggestions? Thanks. > > Fande, > > >
