NWChem for Large systems


Clicked A Few Times
Hello,

I'm trying to run a dft calculation for a system of ~10,000 atoms, but have had problems with getting a working executable.

I have done the following to the source code:

1. $NWCHEM_TOP/src/util/nwc_const.fh at the line

Changed (~line 34):

  • Maximum number of QM atoms

     integer nw_max_atom
parameter (nw_max_atom =3000)

TO

  • Maximum number of QM atoms

     integer nw_max_atom
parameter (nw_max_atom =15000)

Changed (~line 94):

  • Maximum number of shells on a QM system
  • ...................................(must be <= nw_max_atom_shells*nw_max_atom)
     integer nw_max_shells
parameter (nw_max_shells = 10000)

TO

  • Maximum number of shells on a QM system
  • ...................................(must be <= nw_max_atom_shells*nw_max_atom)
     integer nw_max_shells
parameter (nw_max_shells = 50000)

Changed (~line 104):
  • Maximum number of basis functions on a QM system
  • ...................................(must be <= nw_max_atom_nbf*nw_max_atom)
     integer nw_max_nbf
parameter (nw_max_nbf = 10000)

TO

  • Maximum number of basis functions on a QM system
  • ...................................(must be <= nw_max_atom_nbf*nw_max_atom)
     integer nw_max_nbf
parameter (nw_max_nbf = 50000)

2. $NWCHEM_TOP/src/nwdft/include/cdft.fh

Changed (~line 108):

(MXqshells=90001,MXrad=999)

TO

(MXqshells=500000,MXrad=2000)

3. $NWCHEM_TOP/src/NWints/texas/texas_face.F

Changed (~line 521):

parameter(mxsh= 390 000)

TO

parameter(mxsh= 450 000)


Right now the code seems to run for ~5,000 atoms but the application is terminated with the exit string: Hangup after printing the screening tolerance information.

For ~10,000 atoms, I get the error: "int1e: dupl failed 0"

If anyone has any suggestions of what to do to get this to work please let me know.

Thank you

Forum Vet
Tchantaw
The most likely cause of the failure reported is due to the size of the global memory.
Please try to increase the global memory size in the in memory input line and/or use more processors.
Edo

Clicked A Few Times
Quote:Edoapra Jan 6th 3:53 pm
Tchantaw
The most likely cause of the failure reported is due to the size of the global memory.
Please try to increase the global memory size in the in memory input line and/or use more processors.
Edo


I tried to run the ~10,000 atom system with 512 and 1024 processors and get the error " dft_scf: cannot allocate rdens_atom 0" for both.

Forum Vet
Tchantaw
I have tracked down the piece of code causing the failure.
It requires an allocation of size (# atoms)^2, therefore it requires around 750MB of MA memory
(for closed shell cases, 1.5GB for open shell).
Therefore, you need to have a memory input line were the sum of stack and heap should be -- at least -- 2GB.
E.g.

memory stack 1800 mb heap 500 mb global 700 mb


Forum >> NWChem's corner >> Compiling NWChem