Wrong ordering of energy levels
Clicked A Few Times
11:54:24 AM PDT - Fri, May 11th 2012
Hi,

I noticed that sometimes converged DFT calculations give the Kohn-Sham orbitals (which are printed right after the converged energy) in a funny order. Around the Fermi energy, the orbitals are not ordered with increasing energy such that some orbitals are said to be occupied when in fact they shouldn't be. For example :

Vector  835  Occ=2.000000D+00  E=-1.725603D-01

             MO Center=  4.1D-01,  7.7D-01,  1.2D+00, r^2= 3.1D+01

  Bfn.  Coefficient  Atom+Function         Bfn.  Coefficient  Atom+Function

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

 2155      0.290222  203 Pd dyz           1697     -0.287172  184 Pd dxz

 1435     -0.285856  173 Pd dyz           2225      0.280435  206 Pd dxz

 1721     -0.259931  185 Pd dxz           2177      0.241332  204 Pd dxz

 1745     -0.234405  186 Pd dxz           1673     -0.232804  183 Pd dxz

 2035      0.220113  198 Pd dyz           2032     -0.213596  198 Pd dxy



Vector  836  Occ=2.000000D+00  E=-1.708588D-01

             MO Center=  1.0D+00, -4.2D-01,  1.1D+00, r^2= 3.1D+01

  Bfn.  Coefficient  Atom+Function         Bfn.  Coefficient  Atom+Function

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

 1502     -0.206055  176 Pd pz            1915      0.198638  193 Pd dyz

 1645      0.194995  182 Pd py            1315     -0.191369  168 Pd dyz

 1193      0.186496  163 Pd dxz           1888      0.174989  192 Pd dxy

 1478      0.166567  175 Pd pz            2297     -0.160579  209 Pd dxz

 1455     -0.158688  174 Pd dxx           1699     -0.158116  184 Pd dyz



Vector  837  Occ=0.000000D+00  E=-1.737251D-01

             MO Center=  1.2D+00, -2.3D-01,  9.2D-01, r^2= 2.7D+01

  Bfn.  Coefficient  Atom+Function         Bfn.  Coefficient  Atom+Function

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

 2032     -0.243681  198 Pd dxy           1384      0.239949  171 Pd dxy

 1336     -0.235336  169 Pd dxy           1624     -0.229898  181 Pd dxy

 1192      0.195941  163 Pd dxy           1460     -0.178903  174 Pd dzz

 2176      0.178709  204 Pd dxy           1241      0.178167  165 Pd dxz

 1408     -0.177786  172 Pd dxy           1506     -0.171399  176 Pd dyy



Vector  838  Occ=0.000000D+00  E=-1.722297D-01

             MO Center=  8.4D-01, -9.0D-01,  1.1D+00, r^2= 2.8D+01

  Bfn.  Coefficient  Atom+Function         Bfn.  Coefficient  Atom+Function

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

 1433      0.224043  173 Pd dxz           1792     -0.204859  188 Pd dxy

 1264      0.199807  166 Pd dxy           1432     -0.196589  173 Pd dxy

 1673     -0.180613  183 Pd dxz           1480      0.174413  175 Pd dxy

 1555     -0.169257  178 Pd dyz           2177      0.166209  204 Pd dxz

 1815     -0.158712  189 Pd dxx           1697     -0.156071  184 Pd dxz

Is there a reason for this? Couldn't this affect other values calculated from the converged density since it comprises orbitals that shouldn't be occupied in reality?

Thanks,

Chloe

Forum Regular
2:15:12 PM PDT - Thu, May 17th 2012
Multiple possible reasons
Hi Chloe,
I think there can be multiple possible reasons for the behavior you are seeing. Whatever the reason is at convergence the total energy is a minimum for the corresponding orbital occupations. Possible reasons include: the choice of density functional, the use of level shifters, the symmetry of the system, etc.
With regards to the choice of functional this sort of thing is less likely to happen if you pick a functional including exact exchange. Alternatively I would guess that the quasi Newton-Raphson solver is also less likely to produce these kinds of answers.
If you are willing to send me you input I can have a look to see if I can give you some more detailed answers.
Hubertus dot vanDam at PNNL dot gov.

Clicked A Few Times
8:31:54 AM PDT - Mon, May 28th 2012
Hi Huub,

This happened to me both with B3LYP and LDA. I only use DIIS for convergence and always use no symmetry (c1). As for the solver, I use all the defaults (is that Newton-Raphson then ?).

Even if the energy is ok, can I trust the properties computed from the converged density?

I can send you the input file if you want, but it is rather large. The example above is from a calculation that took over a week on a supercomputer.

Thanks for your help,

Chloe


Forum >> NWChem's corner >> General Topics