The ABINIT Users Mailing List ( CLOSED )

[Matteo Giantomassi <gmatteo@pcpm.ucl.ac.be>]

> Dear Matteo
>
> We've tried to comment the line number 1236 of the subroutine
> screening.F90 but cannot get it compiled. Could you please give me a
> litlle bit more details (code) of the line that needs commenting? I
> wonder if the line number of this piece of the subroutine has been
> alterred. Thank you.

 Sorry, the version of screening I have on my laptop is different from that 
contained
 in the official 5.4.3 version, thus I suggested the wrong the line number.

 Your version of screening.F90 should be commented from line 1235 up to line 
1237.
 In particular it's important to skip the call to the outeps routine.
 At the end you should have a commented block similar to:

   ! Write the independent matrix elements on external file (useful to perform 
convergence study in nomegasf)   <<---- line 1233
   ! FIXME should write also the header
!   write(message,'(a,3(f10.6),a)')&
!&   ' epsilon^-1(G,G'') at q = ( ',(q(i,iq),i=1,3),' ) [r.l.u.]'
!   call 
outeps(dtset%npweps,gvec,nomega,omega,ngpi,gptab,gptabo,ip2fp,epsm1,gmet,message,unitem1,prtvol)
!DEBUG
!write(*,*)'checking epsm'
!call 
symf12(dtset%npweps,nomega,epsm1,omega,nop,op,ngpi,gptab,gptabo,ip2fp,phgt,2) 
!only check
!ENDDEBUG

Best Regards, 
Matteo Giantomassi

> Best regards
> Thanusit Burinprakhon
>
>
> > On October 10, 2007 6:01:57 PM ICT, "Matteo Giantomassi"
> <gmatteo@pcpm.ucl.ac.be> wrote:
> > > I have tried GW calculation with the same input file (with or without
> > > mkmem=0) using paralleised Abinit  V5.4.3, and without commenting the
> > > line number 1236 of the subroutine csigme.F90.
> >
> > Well, I hope you've commented screening.F90 instead of csigme.F90.
> > Also because in csigme.F90 the line 1236 doesn't exist!
> >
> > > The jobs have been
> > > succesfully completed.  This seems OK to me. Is it still better to
> > > comment the line, though?
> >
> > If you want to run a screening calculation (optdriver=3) with
> > version 5.4.3, you have to comment that line.
> >
> > BTW: the mkmem=0 option is not yet implemented when optdriver=4.
> > You might try the parallelism over bands (i.e gwpara=2), if you really
> need
> > to decrease the memory per processor,
> >
> > Regards,
> > Matteo Giantomassi