Hello,

I am trying to do a PAW+GW calculation. Before the heavy GW part, I need prepare the ground state electron density and wavefunctions. So, first step I did a normal SCF calculation. The second step is a nscf calculation (iscf=-2) to get wavefunctions using the converged electron density from step 1. As GW usually needs a lot of empty bands to get converged, I included 3008 nbands in the nscf calculation.(cause nband<2000 is not enough to converge the GW bands according to my convergence study). However, when I did nscf calculation with nband=3008, my calculation stop with the following error:

--- !ERROR

src_file: fxphas.F90

src_line: 188

mpi_rank: 3

message: |

The eigenvector with band 634 has zero norm.

This usually happens when the number of bands (nband) is comparable to the number of planewaves (mpw)

Action: Check the parameters of the calculation. If nband ~ mpw, then decrease nband or, alternatively, increase ecut

...

I checked the out file, npw=8165 which is a little larger than nband. As the message said, should I increase the ecut to get rid of this error? For ground state calculation, I did convergence study on the ecut which is 15 Ha in the present case. I am wondering if increasing the ecut is the common recipe to include thousands of bands in GW calculations for PAW? Is there another way to deal with this error? Cause increasing the ecut will greatly increasing the computational cost for GW.

Following is my input file:

enunit 2

ecut 15

pawecutdg 30

occopt 4

tsmear 0.01

nstep 300

irdden 1

tolwfr 1.0d-12

nline 6

kptopt 1

ngkpt 2 2 1

nshiftk 1

shiftk 0.0 0.0 0.0

iscf -2

nband 3008

nbdbuf 8

istwfk *1

acell 2*3.319 44.949 angstrom

angdeg 90 90 120

natom 20

ntypat 4

typat 1 1 1 1 2 4 2 2 3 2 2 3 2 2 4 2 1 1 1 1

znucl 21 34 42 74

xred ......... (atomic positions)

Xiaoming Wang

Postdoc

Rutgers University