On Thu, Mar 6, 2014 at 5:38 PM, Mohammad Mirzadeh <mirza...@gmail.com>wrote:
> Yes. To be precise this is the set of functions I call: > > ierr = MatNullSpaceCreate(mpicomm, PETSC_FALSE, 1, &null_space, > &A_null_space); CHKERRXX(ierr); > > ierr = MatSetNullSpace(A, A_null_space); CHKERRXX(ierr); > > Verify using http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/Mat/MatNullSpaceTest.html > ierr = KSPSetNullSpace(ksp, A_null_space); CHKERRXX(ierr); > > ierr = MatNullSpaceRemove(A_null_space, rhs_, NULL); CHKERRXX(ierr); > > ierr = KSPSolve(ksp, rhs_, solution); CHKERRXX(ierr); > > Send the output of -ksp_monitor -ksp_view Matt > > On Thu, Mar 6, 2014 at 3:33 PM, Matthew Knepley <knep...@gmail.com> wrote: > >> On Thu, Mar 6, 2014 at 5:24 PM, Mohammad Mirzadeh <mirza...@gmail.com>wrote: >> >>> Hi guys, >>> >>> I have a discretization of Poisson equation with Neumann bc for embedded >>> boundary grids in such a way that that nullspace is not the usual constant >>> vector. Instead the nullspace is constant in the domain of interest and >>> zero elsewhere. >>> >>> I compute this nullspace myself and have checked it against MATLAB by >>> dumping the matrix and computing the nullspace explicitly using null >>> function -- they match and there is only this single vector. Then I take >>> this calculated vector and subtract it off the matrix and rhs. >>> >> >> "subtract it off the matrix" does not make sense to me. Are you calling >> KSPSetNullSpace()? >> >> Matt >> >> >>> However, I am having convergence issues. For instance this is the output >>> of ksp_monitor_true_residual for one particular run: >>> >>> 0 KSP preconditioned resid norm 3.033840960250e+02 true resid norm >>> 2.332886580745e-01 ||r(i)||/||b|| 1.000000000000e+00 >>> 1 KSP preconditioned resid norm 1.018974811826e+01 true resid norm >>> 1.941629896918e-02 ||r(i)||/||b|| 8.322864527335e-02 >>> 2 KSP preconditioned resid norm 5.450493684941e-02 true resid norm >>> 1.029339589324e-02 ||r(i)||/||b|| 4.412300185615e-02 >>> 3 KSP preconditioned resid norm 3.944064039516e-02 true resid norm >>> 1.030277925024e-02 ||r(i)||/||b|| 4.416322394443e-02 >>> 4 KSP preconditioned resid norm 6.286181172600e-05 true resid norm >>> 1.030243055045e-02 ||r(i)||/||b|| 4.416172923059e-02 >>> 5 KSP preconditioned resid norm 4.349133658643e-06 true resid norm >>> 1.030239080406e-02 ||r(i)||/||b|| 4.416155885630e-02 >>> 6 KSP preconditioned resid norm 9.279429568232e-08 true resid norm >>> 1.030239169298e-02 ||r(i)||/||b|| 4.416156266666e-02 >>> 7 KSP preconditioned resid norm 3.032522248740e-09 true resid norm >>> 1.030239175066e-02 ||r(i)||/||b|| 4.416156291393e-02 >>> 8 KSP preconditioned resid norm 6.533747246875e-09 true resid norm >>> 1.030239175718e-02 ||r(i)||/||b|| 4.416156294184e-02 >>> 9 KSP preconditioned resid norm 6.083185162500e-12 true resid norm >>> 1.030239175259e-02 ||r(i)||/||b|| 4.416156292220e-02 >>> 10 KSP preconditioned resid norm 5.510319622225e-12 true resid norm >>> 1.030239175259e-02 ||r(i)||/||b|| 4.416156292221e-02 >>> 11 KSP preconditioned resid norm 5.456758524534e-12 true resid norm >>> 1.030239175259e-02 ||r(i)||/||b|| 4.416156292221e-02 >>> 12 KSP preconditioned resid norm 5.456756081783e-12 true resid norm >>> 1.030239175259e-02 ||r(i)||/||b|| 4.416156292221e-02 >>> 13 KSP preconditioned resid norm 5.456755930952e-12 true resid norm >>> 1.030239175259e-02 ||r(i)||/||b|| 4.416156292221e-02 >>> 14 KSP preconditioned resid norm 5.456755930949e-12 true resid norm >>> 1.030239175259e-02 ||r(i)||/||b|| 4.416156292221e-02 >>> 15 KSP preconditioned resid norm 5.456755930949e-12 true resid norm >>> 1.030239175259e-02 ||r(i)||/||b|| 4.416156292221e-02 >>> >>> >>> As you can see, the true residual is quite large and moreover it does not >>> reduce beyond a certain point. This is using hypre as preconditioner, but >>> the situation is equally bad with several other preconditioner (ilu, sor, >>> jacobi, or even none). As for the solution itself, the error has poor to >>> none convergence under grid refinement. All this suggests that the linear >>> system is not converging in my case. >>> >>> >>> Do you have any idea/suggestions why this is happening and how I can avoid >>> it? >>> >>> >>> Thanks >>> >>> >> >> >> -- >> What most experimenters take for granted before they begin their >> experiments is infinitely more interesting than any results to which their >> experiments lead. >> -- Norbert Wiener >> > > -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener