Page 1 of 1
The dfpt scale
Posted: Fri Sep 09, 2011 8:51 am
by xijinyang
Hi, I just use abinit's dfpt to calculate phonon's structure.But I don't know the scale of dfpt. I want to know the relationship between the atom numbers N of cell and the calculation time. However I don't know how large size of cell I can calculate. Can you help me ? Thanks!
Re: The dfpt scale
Posted: Fri Sep 09, 2011 12:25 pm
by jzwanzig
The answer is highly dependent on the symmetry of your system. The phonon code is very smart in its use of symmetry, and will only compute the irreducible perturbations and construct the rest of the dynamical database by symmetry operations. Your best bet is to run some model systems of increasing size to get a feel for how long a given set of perturbations will take on your available computer hardware.
Re: The dfpt scale
Posted: Sun Sep 18, 2011 7:17 pm
by vistawanted
The calculation time highly depends on your atoms number (more, much slower), atoms species (more, much slower) and symmetry (poorer, much slower). Of course, k-mesh, q-mesh, cutoff and so on, also should be considered.
As a reference, this is my test result:
CPU: Intel i7 870
RAM: 12GB
OS: CentOS 5.6,
Compiler: Intel Fortran 2011,
q-mesh: 4x4x4,
k-mesh: 13x13x13,
symmetry: Fm-3m,
Atom Species: 2,
Atom Number: 2,
cutoff: 20Ha,
MPI: MPICH2,
time: 2 hours, 6 threads parallel