My Community

Hi.

I am presently working with a rather large unit cell, such that the system capabilities of the computer i am running vasp on has become the bottleneck for the calculations.

I want to do a slightly denser k-point mesh for the DOS, but the memory requirements are prohibitive (20 irreducible k-points). Can i, in this case where i am using a constant electron density (ICHARG = 11), take the irreducible k-points and their weights as listed in the outcar for a certain mesh, split these into multiple calculations with explicit k-point lists, and add the resulting partial DOSCAR values together? Or is there a problem with this approach?`

Thank you.

Hi,

The method you propose is indeed possible but requires some work. Care must be taken that the individual calculations are compatible to each other (Fermi energy could be a concern). Keep in mind that the tetrahedron method (ISMEAR=-5) will not work in this case.

Alternatively, as of VASP 6.3.0, an optional k-point mesh can be supplied via the KPOINTS_OPT file. This can be used to perform DOS calculations on fine grids. In this case, VASP simultaneously operates only on a subset of these k-points, which substantially decreases the memory requirements. All this can be done in a single run. The results can then be read from the vasprun.xml (node <dos comment="kpoints_opt">) or, preferably, from the vaspout.h5 file if VASP was compiled with HDF5 support ("/results/electron_dos_kpoints_opt/").