self-consistent GW + PAW crash

roginovicci · Post by **roginovicci** » Thu Mar 17, 2016 5:14 pm

Abinit crashes when performing self-consistent GW calculations within PAW formalism. If gwcalctyp > 10 it fails if <10 goes clear. With NCPP all goes fine
Input file:

Code: Select all

ecut 30

#  PAW
pawecutdg 60
usexcnhat=-1
pawovlp 20


# DATASET 1 : Sigma calculation
#
optdriver   4            # Sigma run.
irdkss      1  
irdscr      1
gwcalctyp  12   
gwpara      2            # Parallelization over bands.
symsigma    1            # To enable the symmetrization of the self-energy matrix elements.
ecutwfn    28            # Cutoff for the wavefunctions.
ecuteps     8            # Cutoff in the correlation part.
ecutsigx   24            # Cutoff in the exchange part.
nband       254           # Number of bands for the correlation part.
icutcoul    3            # old deprecated value of icutcoul, only used for legacy

gwmem 0
# List of k-points for GW corrections.
nkptgw     13
kptgw  
 0.00000000E+00  0.00000000E+00  0.00000000E+00

 < ----------------------- skip -------------------------------->

# Definition of the k-point grid
   kptopt    1
   nshiftk   1
   shiftk    0. 0. 0.
   ngkpt     4 4 4

istwfk  *1         # This is mandatory in all the GW steps.

# Use only symmorphic operations
symmorphi 1 #

Crash log:

Code: Select all

 *************** QP Energies *******************
 calc_vhxc_braket : calculating v_xc[n_val] (excluding non-linear core corrections)
 For spin  1 Min density rhor =     0.207196E-02
 For spin  1 Min density nhat =    -0.118870E-04
 For spin  1 Min density trho-nhat =     0.207196E-02
 using usexcnhat =  0
 E_xc[n_val]  = -24.2329 [Ha]. <V_xc[n_val]> =  -0.3816 [Ha].
Assembling PAW strengths for the bare Hamiltonian
 
--- !WARNING
message: |
    symdij does not symmetrize dijhartree term!
src_file: m_pawdij.F90
src_line: 4700
...
 
 Will calculate 1377 <b,k,s|O|b',k,s> matrix elements in calc_vhxc_me.
*** glibc detected *** /share/apps/abinit/abinit-7.10.5//bin/abinit: double free or corruption (!prev): 0x0000000011e2d970 ***
======= Backtrace: =========
/lib64/libc.so.6[0x34ab075f4e]
/lib64/libc.so.6[0x34ab078cf0]
/share/apps/opt/intel/composer_xe_2013.1.117/compiler/lib/intel64/libifcore.so.5(for_deallocate+0xb9)[0x2aad585fda39]
/share/apps/abinit/abinit-7.10.5//bin/abinit[0x989416]
/share/apps/abinit/abinit-7.10.5//bin/abinit[0x4c1de1]
/share/apps/abinit/abinit-7.10.5//bin/abinit[0x414626]
/share/apps/abinit/abinit-7.10.5//bin/abinit[0x40a1a1]
/share/apps/abinit/abinit-7.10.5//bin/abinit[0x4081ac]
/lib64/libc.so.6(__libc_start_main+0xfd)[0x34ab01ed5d]
/share/apps/abinit/abinit-7.10.5//bin/abinit[0x4080a9]
======= Memory map: ========
00400000-02bee000 r-xp 00000000 00:15 1319876                            /share/apps/abinit/abinit-7.10.5/bin/abinit
02ded000-02f9d000 rw-p 027ed000 00:15 1319876                            /share/apps/abinit/abinit-7.10.5/bin/abinit
02f9d000-0723d000 rw-p 00000000 00:00 0 
09061000-13c37000 rw-p 00000000 00:00 0                                  [heap]
34aac00000-34aac20000 r-xp 00000000 08:01 665357                         /lib64/ld-2.12.so
34aae1f000-34aae20000 r--p 0001f000 08:01 665357                         /lib64/ld-2.12.so
.....

Is it possible to perform self-consistent GW calculations with PAW?

p.s. .Version 7.10.5 of ABINIT
.(MPI version, prepared for a x86_64_linux_intel13.0 computer)

roginovicci · Post by **roginovicci** » Mon Mar 21, 2016 10:42 am

I'm sorry I've posted wrong input in previous post.

Actually problem arises when full set of k-points required.
In my case Abinit crashes when array kptgw contains point
2.50000000E-01 2.50000000E-01 2.50000000E-01

In the screening step calculation that point do exist in q-mesh:

Code: Select all

 ==== Q-mesh for the screening function ====
 Number of points in the irreducible wedge :    13
 Reduced coordinates and weights : 
 
     1)     0.00000000E+00  0.00000000E+00  0.00000000E+00       0.01563
     2)     2.50000000E-01  0.00000000E+00  0.00000000E+00       0.09375
     3)     5.00000000E-01  0.00000000E+00  0.00000000E+00       0.04688
     4)     2.50000000E-01  2.50000000E-01  0.00000000E+00       0.09375
     5)     5.00000000E-01  2.50000000E-01  0.00000000E+00       0.18750
     6)    -2.50000000E-01  2.50000000E-01  0.00000000E+00       0.09375
     7)     5.00000000E-01  5.00000000E-01  0.00000000E+00       0.04688
     8)     2.50000000E-01  2.50000000E-01  2.50000000E-01       0.03125    <<<------- Here it is.
     9)     5.00000000E-01  2.50000000E-01  2.50000000E-01       0.09375
    10)    -2.50000000E-01  2.50000000E-01  2.50000000E-01       0.09375
    11)     5.00000000E-01  5.00000000E-01  2.50000000E-01       0.09375
    12)    -2.50000000E-01  5.00000000E-01  2.50000000E-01       0.09375
    13)     5.00000000E-01  5.00000000E-01  5.00000000E-01       0.01563

Mutta · Post by **Mutta** » Fri Mar 25, 2016 3:58 am

Dera Sir
I have also with previous post. "Abinit crashes when performing self-consistent GW calculations within PAW formalism. If gwcalctyp > 10 it fails if <10 goes clear. With NCPP all goes fine".

My input file:

TiO2_rutile_band_g0w0_lda_paw.in: (6.46 KiB) Downloaded 476 times

And the crashes shown also :

--- !WARNING
message: |
symdij does not symmetrize dijhartree term!
src_file: m_pawdij.F90
src_line: 4700
...

Version 7.10.4 of ABINIT
.(MPI version, prepared for a x86_64_linux_gnu4.4 computer)

How can I fix it

Thank you

Abdulmutta Thatribud
Department of Science (Physics), Prince of Songkla University

gmatteo · Post by **gmatteo** » Tue Apr 05, 2016 12:23 pm

I will have a look at it.
Note, however, that for the time being it would suggest to use norm-conserving pseudos for self-consistent GW calculations.
Recent papers showed that GW with PAW requires specialised pseudopotentials in order to get accurate results.

roginovicci · Post by **roginovicci** » Thu Apr 07, 2016 5:34 pm

I was using datasets from GPAW project. The single shot G0W0 approximation shows similar results as we we obtained with VASP package, thus it seems these datasets is suitable, while there is no (as far as I know) special NC pseudopotential database except several potentials for selected atoms from ONCVPSP project. But the ONCVPSP potentials constructed with multiple projectors per channel thus inclvkb variable is only possible to set to 0 which leads to accuracy loss

.

Paul Gutiérrez · Post by **Paul Gutiérrez** » Sat Oct 14, 2017 5:40 am

gmatteo wrote:I will have a look at it.
Note, however, that for the time being it would suggest to use norm-conserving pseudos for self-consistent GW calculations.
Recent papers showed that GW with PAW requires specialised pseudopotentials in order to get accurate results.

Hi, gmatteo,

Could you share the papers about GW with PAW, please?

Thanks al lot,

ABINIT Discussion Forums

self-consistent GW + PAW crash

self-consistent GW + PAW crash

Re: self-consistent GW + PAW crash

Re: self-consistent GW + PAW crash

Re: self-consistent GW + PAW crash

Re: self-consistent GW + PAW crash

Re: self-consistent GW + PAW crash