Re: [Libmesh-users] Different answers on the same number of processors

[Steffen P. <ste...@tu...>]

I do not have installed the 2.3.0 version of PETSc,
but I just ran your test problem using PETSc 2.3.1
without any suspicious behavior on multiple processors.=20

Steffen


> -----Urspr=FCngliche Nachricht-----
> Von: lib...@li...=20
> [mailto:lib...@li...] Im=20
> Auftrag von Tim Kr=F6ger
> Gesendet: Mittwoch, 1. November 2006 09:29
> An: lib...@li...
> Betreff: [Libmesh-users] Different answers on the same number=20
> of processors
>=20
> Dear all,
>=20
> The PETSc FAQ contains an item "Why do I get different answers on a=20
> different numbers of processors?".  But I face the situation that I=20
> even get different answers on the same number of processors. =20
> If I run=20
> the attached test program with
>=20
> mpirun -np 4 ./test
>=20
> on a one processor system (just for test purposes) with PETSc=20
> 2.3.0, I=20
> get one of the following four answers:
>=20
> -------------->  Steps: 11, Residuum: 1.931766949217518759e-13
> -------------->  Steps: 11, Residuum: 2.0180254300703374762e-13
> -------------->  Steps: 11, Residuum: 1.9510877808875828115e-13
> -------------->  Steps: 11, Residuum: 2.0159508477041602036e-13
>=20
> Of course, the difference is small, but for my real application the=20
> difference is essentially larger.  In particular, my real application=20
> sometimes crashes.  This might be due to a programming error,=20
> but this=20
> error is difficult to find if the behaviour is not reproducible.
>=20
> Can someone help me?  Thank you very much in advance.
>=20
> Best Regards,
>=20
> Tim
>=20