Problems with frequency calculations

Clicked A Few Times

3:18:51 AM PST - Mon, Feb 29th 2016

Dear Forum

I have a problem running frequency calculations on "larger" systems (56 atoms). I am running into the ARMCI_DASSERT fail with errorcode 555.

0:0:uhf_fock_setup: could not allocate g_dens2:: 555
ARMCI DASSERT fail. src/common/armci.c:ARMCI_Error():208 cond:0

This my input, where I restart from a constrained geometry optimization (consgeo).

restart consgeo
memory heap 1000 stack 1000 global 2000 mb
freq
  animate
end
dft
 convergence energy 1d-6
 grid fine
end
geometry adjust
 zcoord
 end
end
task dft freq

I am running calculations with the nwchem_openmpi binary on a single processor. I have tried to run the same calculation with a nwchem binary I compiled myself on 8 processors with success. However, the calculation took 4 days and creates incredible large .sxmatices and .fxmatrices files (908 mb), which I believe is unnecessary for a system of this size. What is going wrong?

Forum Regular

10:12:47 AM PST - Mon, Feb 29th 2016
That error is indicating that you do not have enough global memory to allocate the array g_dens2. It appears that array is dimensioned (12*natoms)x(nbf)x(nbf), where natoms is the number of atoms in your system and nbf is the number of basis functions. You would need to increase your global memory sufficiently in order to accommodate the necessary arrays. Alternatively, you could force a numerical calculation of the Hessian, which would require less memory but would definitely take more time.

Forum >> NWChem's corner >> Running NWChem