size of sf_chi0q0

GW, Bethe-Salpeter …

Moderators: maryam.azizi, bruneval

Locked
xw213
Posts: 7
Joined: Sat Sep 03, 2016 5:53 am

size of sf_chi0q0

Post by xw213 » Tue Sep 13, 2016 5:18 am

In Tutorial on Many-Body calculations in parallel, for calculating the screening by using Hilbert transform method, it is said that the size of the matrix of sf_chi0q0 will decrease when increasing the number of processors by setting gwpara=2. Is this true? I mean I did change the number of processors, but the matrix size is always the same, which is 3.8GB. Then the code stop and print out a "out-of-memory" error. Following is my input, any idea about that?

ndtset 2
# scr
optdriver1 3
gwcalctyp1 2
ecuteps1 10
nfreqim1 5
nfreqre1 20
freqremax1 10
inclvkb1 0
spmeth1 1
nomegasf1 100
# sigma
optdriver2 4
gwcalctyp2 2
symsigma2 1
getscr2 -1
ecutsigx2 30
nkptgw2 1
kptgw2 0.0 0.0 0.0
bdgw2 92 93
# control parameters
enunit 2
ecut 25
occopt 4
tsmear 0.01
nstep 200
nline 6
istwfk *1
nband 1000
getwfk 1
ecutwfn 20
fftgw 11
gwmem 0
gwpara 2
# k point
kptopt 1
ngkpt 2 2 1
nshiftk 1
shiftk 0.0 0.0 0.0
# geo structure

Locked