[Wannier] error while running

Vivek Ranjan vranjan at ncsu.edu
Fri Feb 24 14:41:06 CET 2012


On 21 Feb 2012, at 21:51, Vivek Ranjan wrote:

>> While running the wannierize part of the calculation, I got the following error:
>>
>> Error in allocating m_matrix in overlap_read
>>
>> Any hint for the cause of it....

>I've never seen or had that error reported before. You haven't really given us much to work with. The more info you provide  the greater the chance someone can help (eg running W90 as library or standalone. Does it fail with all compilers, or just one (and which one). What sort of system is it etc etc)

Let me make some wild guesses.

The precise line of code is
  allocate ( m_matrix( num_wann,num_wann,nntot,num_kpts),stat=ierr)
  if (ierr/=0) call io_error('Error in allocating m_matrix in overlap_read')

So one obvious way to trigger that error message is if m_matrix is too
large to fit in the available RAM. That's actually pretty unlikely
unless you have an immense system - and you can tell by checking the
number of wannier functions (num_wann), number of kpoints (num_kpts)
and noting that nntot cannot be greater than 12. Note also that
wannier90 should print out a memory estimate in the wout file.

If this is not the issue - then possibly something has corrupted
memory - and a bad version of lapack/blas is usually the cause.

Do other calculations work? What is special about the one that fails?



Jonathan


Hi Jonathan,

I am trying to calculate a 116 atoms system with C, N, H, S atoms in
it. I tried (2,2,1), (6,6,2), and (12,12,4) kpoints grid. Basically,
running a scf, nscf, wannier90.x to generate nnkp file, then pw2wan
and finally wannier again to generate wannier center. I am using
espresso 4.1 for scf, nscf, and pw2wan part of the calculation. The
calculations are done on an IBM power6 machine. I have performed scf,
nscf, and pw2wan using parallel implementations. I run wannier90.x in
serial mode. The memory estimate for (6,6,2) grid was only less than
504 MB. The (2,2,1) grid did not fail. But the larger ones (12,12,1)
and (6,6,2) failed. The error goes like this:

 *============================================================================*
 |                              MEMORY ESTIMATE                               |
 |         Maximum RAM allocated during each phase of the calculation         |
 *============================================================================*
 |                            Wannierise:          504.04 Mb                  |
 *----------------------------------------------------------------------------*

 Starting a new Wannier90 calculation ...

 Time to get kmesh              0.290 (sec)
 Exiting.......
 Error in allocating m_matrix in overlap_read

I am clueless what could be the reason, other than the fact that as
that it crashes for larger kpoints...

Thanks,

Vivek



More information about the Wannier mailing list