SOC and Density Overlap

Total energy, geometry optimization, DFT+U, spin....

Moderator: bguster

Locked
eric.tenuta
Posts: 1
Joined: Mon Jun 22, 2015 5:33 pm

SOC and Density Overlap

Post by eric.tenuta » Mon Jun 22, 2015 9:27 pm

Hi there,

I am an undergraduate student working as a university researcher for the summer learning DFT and how to use Abinit for the first time. I have been trying to incorporate spin orbit coupling into my calculations, but so far with no luck.

Originally I was trying to incorporate SOC in the MAPbI3 Perovskite material, but It would run the first iteration then constantly come across the following error before it starts the second:

Code: Select all

---SELF-CONSISTENT-FIELD CONVERGENCE--------------------------------------------
 FFT (fine) grid used in SCF cycle:

 getcut: wavevector=  0.0000  0.0000  0.0000  ngfft=  80  80  80
         ecut(hartree)=     40.000   => boxcut(ratio)=   2.35650

 getcut : COMMENT -
  Note that boxcut > 2.2 ; recall that boxcut=Gcut(box)/Gcut(sphere) = 2
  is sufficient for exact treatment of convolution.
  Such a large boxcut is a waste : you could raise ecut
  e.g. ecut=   55.530725 Hartrees makes boxcut=2


 chkpawovlp : WARNING -
  PAW SPHERES ARE OVERLAPPING !
   Distance between atoms   1 and   3 is  :   2.08214
   PAW radius of the sphere around atom   1 is:   1.25215
   PAW radius of the sphere around atom   3 is:   0.90083
   This leads to a (voluminal) overlap ratio of  0.27 %

  COMPENSATION DENSITIES ARE OVERLAPPING !!!!
   Distance between atoms   1 and   3 is  :   2.08214
   Compensation radius around atom   1 is:   1.25215
   Compensation radius around atom   3 is:   0.90083
   This leads to a (voluminal) overlap ratio of  0.27 %
  THIS IS DANGEROUS !, as PAW formalism assume non-overlapping compensation densities.

  => Note that other spheres are overlapping !

       Overlap ratio seems to be acceptable (less than value
       of "pawovlp" input parameter): execution will continue.
       But be aware that results might be approximate,
       and even inaccurate (depending on your physical system) !


Even though the extended output says that the execution will continue, it does not,
Here is the input file that I had used.

Code: Select all

#CH3NH3PbI3 Perovskite : Lattice Parameters

autoparal 4

#paral_kgb 1s
#npkpt 1 # a divisor of the 216 k mesh grid points
#npspinor 1
#npband 1
#npfft 1

ndtset 2
jdtset 1 2

# DATASET 1
optcell1 0 #modification of nuclear positions, no cell or dimension optimisation
ionmov1 2  #BFGS optimization, no cell optimization since optcell=0
ntime1 99
tolmxf1 5.0d-4

# DATASET 2     
optcell2 1 #optimization of volume only, not rprim, "homogeneous dilatation of the three components of acell"
ionmov2 3  #BFGS optimization , which takes into account total energy
tolmxf2 5.0d-4
ntime2 99
getcell2 -1
getxred2 -1
dilatmx 1.15
ecutsm 0.5

kptopt 3
nshiftk 1
shiftk 0.0 0.0 0.0
ngkpt 6 6 6 #there are 216  kpts in the system

occopt 1

acell 3*1.1924170988E+01 #6.31A
angdeg 3*90.0
nsym 1 #no symmetry operations

ntypat 5
znucl 1 6 7 53 82
natom 12
typat 2 3 6*1 5 3*4
xred   
    0.9006475441E+00    0.4427964832E-04   -0.3703831657E-02 #C
    0.1368335177E+00   -0.1229545917E-03    0.2530191746E-01 #N
    0.8633161010E+00   -0.1118886484E-03    0.8256677754E+00 #H
    0.8357402862E+00    0.1440659302E+00    0.7208309989E-01
    0.8355108281E+00    0.8562696598E+00    0.7237486568E-01
    0.2092038251E+00    0.1368548391E+00    0.9576700283E+00
    0.2089778447E+00    0.8625955058E+00    0.9580401764E+00
    0.1826924427E+00    0.3599470168E-04    0.1882153186E+00
    0.4763063557E+00    0.4999998829E+00    0.4752389702E+00 #Pb
    0.4279455229E+00    0.4999647195E+00    0.9724659265E+00 #I
    0.4367785399E+00    0.8202748496E-05    0.5078656004E+00
    0.9640271920E+00    0.5000058288E+00    0.4482331528E+00

ecut 15.0
pawecutdg 40.0
pawspnorb 1
pawovlp 20
ixc 11

nstep 99
toldff 1.0d-6 #this is a global parameter and is a magnitude lower than default tolmxf (5.0d-5)
diemac 12.0
optforces 1
prtcif 1
prtden 1
enunit 1


I did try to set "pawovlp -1" but that resulted in the same error.

The pseudopotential files that I am using are the PBE files for Abinit from http://www.physics.rutgers.edu/gbrv/

I am curious if anyone else has had this problem, maybe it is a Abinit 7.10.4 specific issue?
For those of you who did come across this error and managed to fix it, if you could shed light on how you did so that would be excellent.

Thank you!

User avatar
torrent
Posts: 127
Joined: Fri Aug 14, 2009 7:40 pm

Re: SOC and Density Overlap

Post by torrent » Tue Jun 23, 2015 7:50 pm

Hi Eric,

I think the last printed message has nothing to do with the crash; it is just the last printed message...
If you run in parallel it can occur that the relevant error message is not printed.
Is it possible for you to run this job in sequential? just to see if an different error message appear ?

Just a comment: autoparal=4 is only useful on very large supercomputers; you should use autoparal=1.
Then, try to make your calculation lighter (decrease ecut, etc...)
just to see if the crash still append.
Marc Torrent
CEA - Bruyères-le-Chatel
France

Locked