[petsc-users] Parameter to pick KSP for estimating eigenvalues for Chebyshev

[Garth N. Wells]

On 17 June 2015 at 20:33, Barry Smith <bsmith at mcs.anl.gov> wrote:
>
>    Hmm. How can I reproduce this? I tested my fix with both -pc_type mg and -pc_type gamg and it changed in both cases -mg_levels_esteig_ksp_type cg
>

I've tried again, and it works. Thanks. I must have screwed up and not
linked to the patched version of PETSc.

Will the patch find its way into the dev version?

Garth

>   Barry
>
>> On Jun 17, 2015, at 12:59 PM, Garth N. Wells <gnw20 at cam.ac.uk> wrote:
>>
>> On 17 June 2015 at 15:43, Barry Smith <bsmith at mcs.anl.gov> wrote:
>>>
>>>   I introduced a bug in my cleanup. The kspest eigenestimator was created AFTER the KSPSetFromOptions_Chebyshev() hence the options for controlling the KSP estimator were never processed.
>>>
>>>   I've attached a patch, please let me know how it goes.
>>>
>>
>> I've applied the patch, but I still can't change the Krylov method for
>> estimating the eigenvalues.
>>
>> Garth
>>
>>>  Barry
>>>
>>>
>>>> On Jun 17, 2015, at 6:24 AM, Garth N. Wells <gnw20 at cam.ac.uk> wrote:
>>>>
>>>> It used to be possible to pick the Krylov solver used to estimate the
>>>> eigenvalues for Chebyshev, via the parameter system
>>>> ("mg_levels_est_ksp_type" and "gamg_est_ksp_type"). I know there was
>>>> some clean-up, with the parameters becoming
>>>> "mg_levels_ksp_chebyshev_esteig_foo", but an option to pick the Krylov
>>>> solver appears to be missing. Or have I overlooked it?
>>>>
>>>> Garth
>>>
>>>
>