PAW+DFPT+GGA has unreasonable energy values

Dear everyone,

I calculate the Se2Mo elastic constants and piezoelectric properties using the PAW+DFPT+GGA(Se.GGA_PBE-JTH.xml,Mo.GGA_PBE-JTH.xml). However the results give some unreasonable energy values as below. The errors are also reported in the forums ( and not solved。

Please give me some suggestions. Thank you.

Qiang Li

iteration_state: {dtset: 3, }
solver: {iscf: 7, nstep: 40, nline: 4, wfoptalg: 10, }
tolerances: {toldfe: 1.00E-12, }

iter 2DEtotal(Ha) deltaE(Ha) residm vres2
-ETOT 1 -701760.65247417 -6.984E+05 3.645E+01 2.897E+17
ETOT 2 -49209390380.263 -4.921E+10 7.120E+05 2.045E+22
ETOT 3 -10552868127.241 3.866E+10 8.377E+05 1.082E+22
ETOT 4 -765158638.97658 9.788E+09 1.616E+05 1.174E+21
ETOT 5 -328606905.80232 4.366E+08 2.190E+05 2.164E+20
ETOT 6 -285982265.96698 4.262E+07 2.339E+04 1.552E+20
ETOT 7 -126967387.93959 1.590E+08 5.587E+04 7.697E+19
ETOT 8 -53706375.221736 7.326E+07 1.250E+04 2.930E+19
ETOT 9 -57720945.789142 -4.015E+06 2.587E+04 3.018E+19
ETOT 10 -94194367867.836 -9.414E+10 1.713E+07 3.913E+22
ETOT 11 -384991891410.32 -2.908E+11 6.568E+07 1.496E+23
ETOT 12 -36551214042.881 3.484E+11 2.122E+07 4.576E+22
ETOT 13 -205481270109.56 -1.689E+11 5.039E+06 9.851E+22
ETOT 14 -89283335120.502 1.162E+11 1.997E+07 6.265E+22
ETOT 15 -803690232494.15 -7.144E+11 9.567E+06 3.450E+23
ETOT 16 -450853856217.79 3.528E+11 3.788E+07 2.767E+23
ETOT 17 -2121346612896.9 -1.670E+12 3.431E+08 8.809E+23
ETOT 18 -7.34499723495663E+14 -7.324E+14 4.176E+10 3.065E+26
ETOT 19 -5.63337223721916E+14 1.712E+14 1.541E+11 3.093E+26
ETOT 20 -1.82127807321210E+14 3.812E+14 6.711E+10 9.866E+25
ETOT 21 -1.38585260115070E+15 -1.204E+15 1.148E+11 5.971E+26
ETOT 22 -1.85799200262427E+15 -4.721E+14 9.617E+10 8.497E+26
ETOT 23 -1.22152216435784E+16 -1.036E+16 4.086E+11 5.449E+27
ETOT 24 -4.72721586153121E+16 -3.506E+16 6.590E+11 2.157E+28
ETOT 25 -1.13397096852396E+17 -6.612E+16 3.492E+12 4.856E+28
ETOT 26 -1.56161417490567E+19 -1.550E+19 3.997E+14 6.457E+30
ETOT 27 -1.52792414370269E+18 1.409E+19 4.666E+14 1.734E+30
ETOT 28 -3.27796815393749E+18 -1.750E+18 1.481E+15 1.566E+30
ETOT 29 -5.63493846164628E+19 -5.307E+19 1.581E+16 2.231E+31
ETOT 30 -4.66932036183611E+19 9.656E+18 7.006E+15 2.440E+31
ETOT 31 -8.28079591771165E+20 -7.814E+20 2.533E+16 3.597E+32
ETOT 32 -4.75626230752273E+21 -3.928E+21 1.053E+17 2.005E+33
ETOT 33 -6.93628489422145E+21 -2.180E+21 1.750E+17 3.311E+33
ETOT 34 -5.71010084387757E+22 -5.016E+22 4.875E+18 2.420E+34
ETOT 35 -1.37611319784882E+22 4.334E+22 3.745E+18 1.049E+34
ETOT 36 -2.12580639293762E+23 -1.988E+23 1.409E+19 8.999E+34
ETOT 37 -1.86232926681585E+24 -1.650E+24 2.852E+20 7.471E+35
ETOT 38 -1.97315995886578E+24 -1.108E+23 6.801E+20 1.054E+36
ETOT 39 -9.26233829865064E+25 -9.065E+25 3.149E+21 3.864E+37
ETOT 40 -1.63474793427571E+26 -7.085E+25 2.236E+21 7.441E+37
telast_2.out (63.3 KB) (3.71 KB)

  1. Currently a ABINIT is only available PAW+DFPT+LDA
  2. If you want a DFPT+GGA you need NC pseudopotential
  3. For 2D structures there is a problem even with PAW+DFPT+LDA, NC pseudopotentials work correctly

What I would do

  1. Start with LDA+PAW
  2. Check other PAW pseudopotential
  3. I would reduce ecut to 500 eV (for PAW 800 eV somewhat redundant)
  4. Try occopt = 3 tsmear =0.01
  5. “ngkpt 12 12 1” Is that so much needed? I always use kptrlen~35
  6. Generate own PAW
  7. Use NC pseudopotential

Dear gswylq,

I see in your ouput file that the GS wavefunctions are not well converged. Indeed residm~1e-7 while it should be 1e-15 or lower (1e-20). You have to improve that before doing any RF computation.


1 Like

Hello gswylq and mmazdz,

DFPT+GGA+PAW does work, actually, since several years now, but I have recently also had problems with DFPT+GGA+2D systems (LDA is fine). This is quite probably your issue: the very low density in the vacuum is a problem for the DFPT perturbed xc potential evaluation. Move to a JTH or GBRV potential for LDA and see what happens.

You should also try Laurent’s recommendation first, the others from mmazdz may or may not help with your ETOT convergence issue:

  • The ecut and ngkpt have to be converged beforehand and not reduced arbitrarily. Experience is that DFPT often requires a bit more kpt than the ground state
  • the smearing can definitely help if your system is semiconducting and becoming more metallic with perturbation, but 0.01 is quite a large smearing, especially with Fermi Dirac occupations
  • I would not generate a PAW dataset randomly - you will lose a lot of time testing it. Going to NC potentials is always a good way to check if your problem is intrinsic in the physics or not
1 Like

As far as I know DFPT+GGA+PAW works for phonons, but not for elastic constants (this was the case at least in version 9.4.1). Produces non-physical results