Convergence problems; high difference between atomic guess and first step in scf

Just Got Here

2:57:32 PM PST - Wed, Dec 10th 2014

Hello,

I have some convergence problem with my system. First i have tried to use atomic energies as a initial guess:

 Sum of atomic energies:      -12168.89600910

However, in the first step of the self consistent field iteration the calculated energy is much lower (around 3000 Hartree) than the sum of atomic energies.

   convergence    iter        energy       DeltaE   RMS-Dens  Diis-err    time
 ---------------- ----- ----------------- --------- --------- ---------  ------
     COSMO solvation phase
 d=95,ls=0.0,diis     1  -9070.3026315120 -3.81D+04  2.12D+00  3.75D+03  2432.4
                                                     1.36D+00  3.75D+03

 ...

 Grid integrated density:     881.000491141720
 Requested integration accuracy:   0.10E-07
 d=95,ls=0.5,diis    20  -8896.2436450218  2.10D+01  1.03D+00  7.86D+03 39285.7
                                                     1.08D+00  7.85D+03

Furthermore, the energy does not converge. Even with a very high damping factor the energy value become lower.

charge -1

geometry units angstrom noautosym noautoz
  ...
end

basis spherical
   C library Def2-SVP
   H library Def2-SVP
   N library Def2-SVP
   O library Def2-SVP
   S library Def2-SVP
  Fe library Def2-SVP
end

cosmo
  dielec 4.0
  do_gasphase false
end

dft
 xc b3lyp
 odft
 mult 2
 noio
 direct
 grid xfine
 iterations 400
 mulliken
 vectors input atomic output NoCDFT.mos
 convergence damp 95
 convergence ncydp 400
end

Any idea how it can be fixed?

Forum Regular

5:32:36 PM PST - Sat, Feb 14th 2015
You have set do_gasphase false Hence you are doing a COSMO calculation where you are embedding you system in a field of point charges to represent the solvent. But now you are calculating the embedding charges from a unconverged charge density. At that point anything can happen. I would recommend running the gas phase calculation first. That might help a lot. Suppressing the gas phase calculation is a good idea for example in geometry optimizations where you have a sensible wavefunction from the previous point. For a single point calculation it is not recommended.

Forum >> NWChem's corner >> Running NWChem