Page 1 of 1
Huge kpoints for GW
Posted: Tue Jul 07, 2020 1:53 am
by sdwang
Dear all,
I am running the chi for a large number of k points (the symmetry is removed due to the external field), bur it is impossible for my current resources. Is there any other choice for splitting the chi calculation for huge k points? (I know the ypp can merge the QP energy for independent calculations).
Best
Shudong
Re: Huge kpoints for GW
Posted: Tue Jul 07, 2020 7:42 am
by Daniele Varsano
Dear Shudong,
the response function is calculated for each q point and ndb.pp_fragments_#q databases are written on disks. If the calculations is not finished you can run the calculation with the same input and the calculation will restart from the last calculated q point.
Best,
Daniele
Re: Huge kpoints for GW
Posted: Sat Jul 18, 2020 8:44 am
by sdwang
Dear Daniele,
I know the response process could restart from the abrupt point. My problem is the huge q points need much more memory even other parameters same as the less q points setting. Is any other choice like wannier interpolation to solve the huge q points calculation for response function?
Thanks!
Best
Shudong
Re: Huge kpoints for GW
Posted: Mon Jul 20, 2020 8:20 am
by Daniele Varsano
Dear Shudong,
as you need the screening matrix for each q point to perform a GW (or BSE) calculation there is not interpolation technique implemented.
As maybe you know you can distribute CPU over "c" and "v" bands in X_ROLES to distribute memory.
Best,
Daniele