PAW iprcel 149

[mverstra]

Hello,

I know these features are in development, so this is just for your information:

an attempt to run a PAW job with SO _and_ LDA+U _and_ iprcel 149 gave a segfault at the following line in src/77_suscep/suscep_stat.F90:

533 sussum(:)=reshape(susmat(:,:,:,:,:),(/2*npwdiel*nspden*npwdiel*nspden/))

This means in particular that there is no test to forbid iprcel in certain cases where it does not work. Or should it work?

thanks!

Matthieu


as usual, ifort 10 / openmpi on xeon cluster with CVOS

>>>>>>>>>
iprcel 149 # calculate RPA diel mat every 9 steps

nstep 100

ngkpt 8 8 1
istwfk *1

# LDA+U
usepawu 1 lpawu -1 3 upawu 0.0 5 eV jpawu 0.0 0.7 eV

# add SO coupling
pawspnorb 1 nspden 4 kptopt 3

nband 500 # in spinor case

spinat
0 0 4 0 0 0 0 0 0
0 0 4 0 0 0 0 0 0
...
...

#===========================================================

acell 2*6.253 33. Angstr
angdeg 90 90 90

.
.
.
atomic details censored
.
.
ecut 20
pawecutdg 40

toldfe 1.e-10
tsmear 0.005
occopt 7
nshiftk 1
shiftk
0. 0. 0

<<<<<<<<<

[torrent]

Hi MJV,

Should work... in principle.
This piece of code is neither related to PAW nor LDA+U or SO...

But it is particularly innefficient:

Code: Select all

   sussum(:)=reshape(susmat(:,:,:,:,:),(/2*npwdiel*nspden*npwdiel*nspden/))
   call xsum_mpi(sussum,spaceComm,ierr)
   susmat(:,:,:,:,:)=reshape(sussum(:),(/2,npwdiel,nspden,npwdiel,nspden/))

should be simply replaced by:

Code: Select all

   call xsum_mpi(susmat,spaceComm,ierr)

Marc