G-shell not closed

[jiangx3]

Dear All,

When I do the convergence study of graphene, I encounter the problem as followings:


Subroutine Unknown:0:ERROR
G-shell not closed
Initial G vector 908/ 909 -2 1 6 Rotated G vector 3 1 6
Through sym 3 and itim 1


I have no idea how to modify the input file based on this error. Could it be the compilation problem? I mean did I miss something when I compiled the new version abinit?

Thanks and Best regards,
XUeping Jiang

[jmullen]

Hi,

Did you ever resolve this problem? I have followed the tutorial closely, replacing graphene for Si. At every step, I have calculated converged results, but when I perform the final GW calculation as in the tutorial, the section with optdriver=3 generates this same results and I do not know where to go from here. Below is a little snippet of the configuration file from the run

--------------------

Code: Select all

# Graphene hexagonal unit cell
ndtset    3

ngkpt 12 12 1
shiftk 0.0 0.0 0.0

# Dataset 1: SCF
prtden1 1
istwfk *1

# Dataset 2: KSS file
iscf2 -2
getden2 -1
nband2 8
nbandkss2 300


# Dataset 3: SCreening
optdriver3 3
getkss3 -1
ecutwfn3 35
ecuteps3 35
ppmfrq 16.7 eV


# Dataset 4: GW
optdriver4 4
getkss4 -2
getscr4 -1
nband4 300
ecutwfn4 35
ecutsigx4 35
nkptgw4 1
kptgw4
0.3333 0.3333 0.0000
bdgw4 4 5

-- The rest is the geometry for graphene which gives a sane band structure (so I feel it is correct)


The error I receive is

Code: Select all

m_gsphere.F90:297:ERROR
  G-shell not closed
  Initial G vector  4106/ 4954     0   -1   34 Rotated G vector     0    1  -34
  Through sym   1 and itim   2

Regards
Jeff
NCSU Physics

[gmatteo]

There's a problem somewhere in the ordering of the G-vectors used for the polarizability.
This kind of error usually shows up when the energy cutoff for chi0|W is close to the
the max kinetic energy of the plane-waves stored in the KSS files.

For the time being one has to proceed by trials and errors trying different values of ecutesp
until the code doesn't stop anymore.

BTW: Are you sure you need ecutesp 35 to converge your results? I suggest you try with a much
smaller value for ecutesp and monitor the convergence of your results wrt ecuteps as discussed in
the tutorial

Matteo

[mverstra]

Also your k-point is not super close to K (need more digits) so it could be that the equivalence between G vectors for the shifted spheres around K is a bit messed up (ie one side of the sphere is not quite inside the total set of G vectors, which comes from your ground state, with properly shifted and rounded k-points. Shouldn't be, though, as it is less than 1/3, not more...

Matthieu

[jmullen]

Thanks for your comments. I have implemented both of these suggestions, however now I am receiving a SIGSEGV error part way through part 3 of the calculation. I will post again when I have more information on this problem.

Jeff Mullen
NCSU Physics

[sponce]

Dear all,

I'm also facing the same problem:

Code: Select all

 optimal value for ng0sh =  1 1 1

 m_gsphere.F90:332:ERROR
  G-shell not closed
  Initial G vector 21798/47817     2    2  -17 Rotated G vector    -2   -2   17
  Through sym   1 and itim   2

 m_gsphere.F90:332:ERROR
  G-shell not closed
  Initial G vector 21798/47817     2    2  -17 Rotated G vector    -2   -2   17
  Through sym   1 and itim   2
-P-0000
-P-0000  leave_new : decision taken to exit ...

This occurs during the sigma calculation. For the screening the calculation went smootly with:

Code: Select all

nband3     450
ecutwfn3   35   
ecuteps3   7
kptopt3  1
ngkpt3 4 4 4
nshiftk3 1
shiftk3 0.0 0.0 0.0

For the sigma calculation I have:

Code: Select all

ecuteps4   5
ecutwfn4   25       
ecutsigx4  60
kptopt4     1
ngkpt4    4 4 4
nshiftk4    1
shiftk4  0.0 0.0 0.0

As you can see I have reduce my values (following Matteo advices at the start of this topic) from ecuteps 7 to 6 and then 5. I have also try to decrease ecutwfn from 35 to 30 and then 25 (and ecutsigx accordingly).

All the calculations crash with the same error message.

Any thoughts?

Thank you

Samuel

[gabriel.antonius]

Dear Samuel,

Try the following:

Code: Select all

ecutwfn4 25
ecutsigx4 20

If it goes through, it means there is a problem with the reordering of the G-vectors.

See in 70_gw/setup_sigma.F90 :

Code: Select all

  ! gvec_p is contained in gsphere_sigx_p but the ordering might differ as we have used 
  ! a different cutoff value in merge_and_sort_kg. Thus swap the elements in gsphere_sigx_p 
  ! in order to have gvec_p(:,1:ng_kss) == gsphere_sigx_p(:,1:ng_kss)

  do ig1=1,ng_kss

   if (ANY(gvec_p(:,ig1)/=gsphere_sigx_p(:,ig1))) then
    gswp = gsphere_sigx_p(:,ig1)

    ig2=ig1; found=.FALSE.
    do while (ig2<ng_sigx.and..not.found)
     ig2=ig2+1
     found = ALL(gvec_p(:,ig1)==gsphere_sigx_p(:,ig2))
    end do

    if (.not.found) MSG_ERROR("gvec_p(:,ig1) not in gsphere_sigx_p")
    gsphere_sigx_p(:,ig1) = gvec_p(:,ig1)
    gsphere_sigx_p(:,ig2) = gswp
   end if

  end do

I had to suppress this loop, and it solved the problem: I could use ecutsigx > ecutwfn.
Let me know if it works for you!

Gabriel

[sponce]

Thank you Gabriel for your reply and help (and sorry for the super late reply).

It indeed work by removing this loop. But is the result of the calculation relevant if the ordering is messed up?

With the new version (6.15.2 or later) of Abinit the routine has been rewritten and I'm now facing with the exact same input file a norm determination problem : http://forum.abinit.org/viewtopic.php?f=11&t=1835

Kind regards,

Samuel

[maxim]

Dear abinit users,

Is it true that the GW calculations require the powerful computer? My computer freezes.

Bests, Maxim.