[petsc-dev] strangness in Chebyshev estimate of eigenvalues

[Jed Brown]

Mark Adams <mfadams at lbl.gov> writes:

> Is there a good way to get a deterministic rand wrt number of processors?
> Seeding rand for each entry, with its global ID would work I assume.

No, no, no.  That's basically the same nonsense as your pseudo "hash"
function.  PRNGs are designed so the stream of random numbers produced
From any given seed passes some statistical randomness tests, but
generally not that the first number from each stream seeded by a
sequence of seed values passes those tests.  Some PRNGs may behave
alright in this case, but the simple ones do not.

> Cheby is otherwise deterministic, with a deterministic PC, and we have
> equation numbers to make this work. Eigen estimates are a real pain and the
> source of a lot of time consuming AMG errors.  It would be nice to have it
> not depend in any way on processor count.

As Barry says, the matrix (numbering) typically depends on process count.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 818 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20150828/7ddf46cf/attachment.sig>