[petsc-users] inconsistent SNES/KSP behavior

[Matthew Knepley]

On Thu, Sep 24, 2015 at 9:50 AM, Gideon Simpson <gideon.simpson at gmail.com>
wrote:

> So I may be imagining this, but I like to think that the following
> happened.  I submitted the same job script to my cluster, twice, and I got
> different results.    This is a SNES problem where I am using the MUMPS
> linear solver.  Here is what I mean:
>

It looks like MUMPS uses randomization, since the initial true residuals
are the same, but the preconditioned is different.

  Matt


> *Trial 1:*
>
> Generating initial guess with a0 = 1.95, b0 = 2.09, sigma = 2, xmax = 100,
> nx = 2560000
>
> 2560001
> 10240006
> Writing to disk as guess.bin
>
>  Solving sigma = 2
>   0 SNES Function norm 1.691829175711e-01
>     0 KSP preconditioned resid norm 6.098266966934e-04 true resid norm
> 1.691829175711e-01 ||r(i)||/||b|| 1.000000000000e+00
>     1 KSP preconditioned resid norm 1.698748932236e-12 true resid norm
> 5.163872517578e-10 ||r(i)||/||b|| 3.052242266367e-09
>   1 SNES Function norm 1.395434035402e-04
>     0 KSP preconditioned resid norm 7.922333994913e-07 true resid norm
> 1.395434035402e-04 ||r(i)||/||b|| 1.000000000000e+00
>     1 KSP preconditioned resid norm 1.839375186797e-14 true resid norm
> 3.034980160221e-13 ||r(i)||/||b|| 2.174936316030e-09
>   2 SNES Function norm 1.394208676802e-04
>  sigma = 2, a = 1.94758, b = 2.09002, reason =4
>
> *Trial 2:*
>
> Generating initial guess with a0 = 1.95, b0 = 2.09, sigma = 2, xmax = 100,
> nx = 2560000
>
> 2560001
> 10240006
> Writing to disk as guess.bin
>
>  Solving sigma = 2
>   0 SNES Function norm 1.691829175711e-01
>     0 KSP preconditioned resid norm 6.098266972456e-04 true resid norm
> 1.691829175711e-01 ||r(i)||/||b|| 1.000000000000e+00
>     1 KSP preconditioned resid norm 1.497022256767e-11 true resid norm
> 1.953323112042e-09 ||r(i)||/||b|| 1.154562848357e-08
>   1 SNES Function norm 1.396785292629e-04
>     0 KSP preconditioned resid norm 1.430883443301e-05 true resid norm
> 1.396785292629e-04 ||r(i)||/||b|| 1.000000000000e+00
>     1 KSP preconditioned resid norm 5.226190977119e-14 true resid norm
> 3.918710626146e-12 ||r(i)||/||b|| 2.805521111100e-08
>   2 SNES Function norm 1.396634713981e-04
>     0 KSP preconditioned resid norm 1.693020976611e-05 true resid norm
> 1.396634713981e-04 ||r(i)||/||b|| 1.000000000000e+00
>     1 KSP preconditioned resid norm 6.991045787498e-14 true resid norm
> 4.339991163168e-12 ||r(i)||/||b|| 3.107463332911e-08
>   3 SNES Function norm 1.395099855381e-04
>     0 KSP preconditioned resid norm 8.023057268371e-06 true resid norm
> 1.395099855381e-04 ||r(i)||/||b|| 1.000000000000e+00
>     1 KSP preconditioned resid norm 3.476126449145e-14 true resid norm
> 2.800497344039e-12 ||r(i)||/||b|| 2.007381287610e-08
>   4 SNES Function norm 1.394785126062e-04
>  sigma = 2, a = 1.94758, b = 2.09002, reason =4
>
>


-- 
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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150924/75de7ec6/attachment.html>