[petsc-dev] strangness in Chebyshev estimate of eigenvalues

[Barry Smith]

> On Aug 27, 2015, at 1:58 PM, Jed Brown <jed at jedbrown.org> wrote:
> 
> Mark Adams <mfadams at lbl.gov> writes:
> 
>> Ah, you're right.
>> 
>> Would you consider this a decent hash?
>> 
>> PetscScalar v = ((PetscScalar)((79943*idx)%6871) - 3435.)/3435. +
>> ((PetscScalar)((7919*idx)%829) - 415.)/415.;
> 
> Still bullshit.  Plot this and you'll see that there are lots of
> patterns and it's just not very rough at all.  Try grid sizes of 48, 59,
> 64, 74, 75, 80, 85 to name a few.  (I looped over some grids in Octave.)
> This is a fools errand.
> 
>> I thought drand48 has been taken off the table because of Windows.  If it's
>> not I can use that.
> 
> Use drand48,

   Don't use drand48 directly; use PetscRandom. Matt promised to fix up PetscRandom to use it by default and work on all systems so all you need to do is use PetscRandom everywhere.

  Barry

> it's very simple.
> 
> http://pubs.opengroup.org/onlinepubs/007908799/xsh/drand48.html