[petsc-users] questions about pc_comosite

[Barry Smith]

> On Sep 29, 2017, at 8:26 AM, Xiangdong <epscodes at gmail.com> wrote:
> 
> Hi Barry,
> 
> Can you point me the documentation for understanding these iteration patterns? Thanks a lot.

  They are explained implicitly in the documentation for the pieces but I admit it may not be clear.

   Case 1:

     Plain GMRES, prints the preconditioned residual at each iteration by default. The operator is just application by A

  Case 2:

     The operator is multiply by A, the preconditioner is apply the identity, compute the residual, update the solution by the residual (because the second inner preconditioner is identity). The preconditioner is not the same as in case 1 because of the computation of the residual inside the preconditioner and the update of the solution.

   Case 3: 

    Same preconditioner as 2 but since the preconditioner is not the identity the unpreconditioned norm is different than the preconditioned norm hence it prints something different when you request the unpreconditioned norm.

  Barry


     
> 
> Best,
> Xiangdong
> 
> On Fri, Sep 29, 2017 at 11:24 AM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
>   They are all different iteration patterns. They definitely should present different residual results
> 
> 
> > On Sep 29, 2017, at 8:19 AM, Xiangdong <epscodes at gmail.com> wrote:
> >
> > Hello everyone,
> >
> > I have a questions about residuals reported in pc_composite. For the examples in petsc-3.7.6/src/ksp/ksp/examples/tutorials/ex1.c, I found that when I use these three options:
> >
> > 1) -pc_type none
> > 2)  -pc_type composite -pc_composite_pcs none,none -pc_composite_type multiplicative
> > 3) -pc_type composite -pc_composite_pcs none,none -pc_composite_type multiplicative -ksp_norm_type unpreconditioned
> >
> > as shown below, it reported different KSP residuals. Given that I use the identity preconditioner (with multiplicative), why does the residual vary for different options?
> >
> > Thanks.
> >
> > Best,
> > Xiangdong
> >
> > mpirun -np 1 ./extest -ksp_monitor -pc_type none
> >
> > 0 KSP Residual norm 1.414213562373e+00
> >   1 KSP Residual norm 6.324555320337e-01
> >   2 KSP Residual norm 3.779644730092e-01
> >   3 KSP Residual norm 2.581988897472e-01
> >   4 KSP Residual norm 1.906925178491e-01
> >   5 KSP Residual norm 1.616509124176e-15
> >
> > mpirun -np 1 ./extest -ksp_monitor -pc_type composite -pc_composite_pcs none,none -pc_composite_type multiplicative
> >
> >  0 KSP Residual norm 1.414213562373e+00
> >   1 KSP Residual norm 1.176696810829e+00
> >   2 KSP Residual norm 1.096908636191e+00
> >   3 KSP Residual norm 4.389821446437e-01
> >   4 KSP Residual norm 2.088364906576e-01
> >   5 KSP Residual norm 2.725851091482e-13
> >
> > mpirun -np 1 ./extest -ksp_monitor -pc_type composite -pc_composite_pcs none,none -pc_composite_type multiplicative -ksp_norm_type unpreconditioned
> > 0 KSP Residual norm 1.414213562373e+00
> >   1 KSP Residual norm 1.290994448736e+00
> >   2 KSP Residual norm 1.249516035344e+00
> >   3 KSP Residual norm 4.836894336642e-01
> >   4 KSP Residual norm 1.467306491390e-01
> >   5 KSP Residual norm 9.881776494390e-14
> 
>