[petsc-users] Removing rigid body motion using Null Space

[Barry Smith]

> On Jul 28, 2015, at 4:33 PM, Jed Brown <jed at jedbrown.org> wrote:
> 
> Hom Nath Gharti <hng.email at gmail.com> writes:
> 
>> Thanks for your response. My example is also a elasticity problem
>> where displacment is a vector-valued.
> 
> You called it "electrostatic".  

   To be fair he said

> a simple
> elastostatic problem. I have an unconstrained sphere and I need to
> compute displacement field due to a point load on the surface.

which made it clear to me that it involved mechanics as well.



> Anyway, use the actual null space for
> your operator and run the tests I mentioned.
> 
>> 
>> On Tue, Jul 28, 2015 at 3:38 PM, Jed Brown <jed at jedbrown.org> wrote:
>>> Hom Nath Gharti <hng.email at gmail.com> writes:
>>> 
>>>> Dear all,
>>>> 
>>>> After searching old mailing lists, it appears that the rigid body
>>>> motion
>>> 
>>> "rigid body motion" refers to rotations and translations, which appear
>>> in vector-valued problems like elasticity or fluids.
>>> 
>>>> can be removed using Null Space features in PETSc.  Unfortunately, I
>>>> could not successfully implement this to a simple elastostatic
>>>> problem.
>>> 
>>> What is vector-valued in your electrostatic problem?
>>> 
>>>> I have an unconstrained sphere and I need to compute displacement
>>>> field due to a point load on the surface.
>>>> 
>>>> I tried:
>>>> call MatNullSpaceCreate(PETSC_COMM_WORLD,PETSC_TRUE,0,0,nullspace,ierr)
>>>> call MatSetNullSpace(Kmat,nullspace,ierr)
>>>> call MatNullSpaceRemove(nullspace,bvec,PETSC_NULL_OBJECT,ierr)
>>>> call MatNullSpaceDestroy(nullspace,ierr)
>>>> 
>>>> But the solution did not converge! Any help/hint would be greatly appreciated.
>>> 
>>> Check that this really is the null space of the operator.  You can use
>>> -ksp_test_null_space, for example.
>>> 
>>> Check that the preconditioner is stable.  Use a simple preconditioner on
>>> a tiny problem to start with.