Really deep SCF energies DFT (with dscf)

[marilu98]

Dear all,
I'm computing ground state energies at B3LYP/6-31+G*  (defining manually the basis file) and for some of my systems the SCF (within dscf module) do not converge and gives really deep energies (of the order of -10^6 hartree).
I'm quite sure about the basis specified since it works for other "similar" systems.

I tried to modify $scforbitalshift, $scfdamp, gridsize, type of guess, but without success. I'd like to try to modify the threshold for basis linear dependencies and the threshold for neglecting one electron integral. Do you know if there are some keywords to do the latter within dscf module ?

Did any of you encounter this type of problems ?

I thank you all in advance.

Best regards
Marilu

[uwe]

Hello,

my guess would be that the functional is causing those numerical errors, this can happen if the basis set generates some really unusual density and density derivatives at certain points.

Does the same input run with a predefined basis set (not a manually modified one)?

Best Regards

[marilu98]

Thank you for your reply.

I don't think that is a problem of the functional (I tried also other xcf like tpssh and pbe0).
With aug-cc-pVDZ I have no problems, as well as with bs (external or built in) without diffuse functions (6-31G*, 3-21G*, cc-pVDZ,...).

The problem is that B3LYP/6-31+G* level is not working for some of my systems (namely 12 out of 48), so it is not a systematic error. My guess it's that for these systems linear dependencies are more crucial....thus, I'd like to remove some of them (if it is possible).

Thank you once again

Best regards

[uwe]

Hi,

is the input structure prone to linear dependencies of the basis set if diffuse functions are used? Like a fullerene where all diffuse functions 'flood' the inner part of the fullerene?
As far as I can see aug-cc-pVDZ is not less diffuse than 6-31+G* (I just downloaded both basis sets for carbon from https://www.basissetexchange.org/ and compared them).
Is it possible to post the structure of one of those cases (coord file and also the basis set file you have used)?
Best Regards

[marilu98]

Hi,
thank you very much once again. In fact, I have problems with systems that contain triple CC bonds.
Here an example (it is from the paper J. Chem. Theory Comput. 2022, 18, 2, 1046–1060):
$coord
    9.27006233815159     -4.56563200912835      0.00013379544421       h
    4.64141236904137     -4.32483962923246      0.00013182805034       h
    5.05918122897159      3.75748513925418     -0.00010634112513       h
    9.66505382813530      3.54143798936859     -0.00010887127944       h
   -3.67501791919848     -3.89309077933371     -0.00010509711842       h
   -3.25648850932071      4.19556853915043      0.00011925041120       h
   -8.32979991840199     -3.64789074931242     -0.00010594314881       h
   -7.93772671848856      4.41415218923358      0.00011623176269       h
  -15.81144469698963     -0.50727138483899     -1.67389502958465       h
  -15.75527819702021     -3.44057121940902     -0.00010840943037       h
  -15.81145089699207     -0.50737682191914      1.67386171976020       h
   16.41873589675920      2.04080322952146     -0.00006053284106       h
   13.80513209730624      3.27359241934086     -1.68893740959647       h
   13.80512039725591      3.27368692926867      1.68873124973594       h
    8.35084343836196     -2.73888994941181      0.00007836070629       c
    5.74629668879938     -2.60666967941831      0.00007860126843       c
    4.49886966908027     -0.26110449798891      0.00001062063877       c
    5.97959980891925      1.93352382967140     -0.00005539221908       c
    8.59585792826932      1.80436951956161     -0.00005627037481       c
    9.81767446813259     -0.53413285394070      0.00001026385847       c
    1.82241367963363     -0.12226715902900      0.00000922356424       c
   -0.45911807593796     -0.00608386365412      0.00000770327957       c
   -3.13570901941989      0.13323481802946      0.00000705000125       c
   -4.60739961919282     -2.07607416960626     -0.00005649241763       c
   -4.37115621920461      2.48437982950287      0.00006990720545       c
   -7.21746430856807     -1.93788686959910     -0.00005681008059       c
   -6.97894704856623      2.61135661953761      0.00006833816584       c
   -8.43051093826820      0.40687496393239      0.00000522736041       c
  -11.22667899787731      0.66458740692910      0.00000708666194       c
  -15.12622359708390     -1.48762030975725     -0.00004640146910       c
   14.37355899722264      2.24752188963408     -0.00007215975899       c
  -12.34226999752111      2.65789218946130      0.00005136653552       o
  -12.41664749751367     -1.59239539976045     -0.00004424018933       o
   13.13702549744587     -0.95154658274258      0.00001396280839       s
$end

I agree with you: 6-31+G* has less diffuse functions with respect to aug-cc-pVDZ... this is why I can't spot properly the problem.
The only difference is that aug-cc-pVDZ is implemented and I used 6-31+G* as implemented in QChem.

Best regards

[marilu98]

I'm now trying the same calculation with HF (to have a better guess to feed a DFT calculation) and I still have extremely oscillating scf energies. Hereafter some cycles:

 ITERATION  ENERGY          1e-ENERGY        2e-ENERGY     NORM[dD(SAO)]  TOL
   2  -975974.79994451    -4311.6171875    -973025.53918    0.467D+11 0.102D-11
          Norm of current diis error:  4.4348
          max. resid. norm for Fia-block=  1.448D+05 for orbital     25a
          max. resid. fock norm         =  1.454D+05 for orbital     25a
          Delta Eig. = **************** eV

                                              current damping :  0.850
 ITERATION  ENERGY          1e-ENERGY        2e-ENERGY     NORM[dD(SAO)]  TOL
   3  -1019817.0084406    -4249.1926270    -1016930.1722    0.584D+09 0.102D-11
          Norm of current diis error:  5.1341
          max. resid. norm for Fia-block=  4.385D+01 for orbital      1a
          max. resid. fock norm         =  3.580D+02 for orbital      1a
          Delta Eig. = **************** eV

                                              current damping :  0.900
 ITERATION  ENERGY          1e-ENERGY        2e-ENERGY     NORM[dD(SAO)]  TOL
   4  -1019817.9718195    -4261.4228516    -1016918.9054    0.251D+05 0.102D-11
          Norm of current diis error:  2.9852
          max. resid. norm for Fia-block=  7.470D+01 for orbital      1a
          max. resid. fock norm         =  6.127D+02 for orbital      1a
          Delta Eig. = **************** eV

                                              current damping :  0.950
 ITERATION  ENERGY          1e-ENERGY        2e-ENERGY     NORM[dD(SAO)]  TOL
   5   92468.401837716    -4271.7119141     95377.757324    0.251D+05 0.809D-11
          Norm of current diis error:  1.5651
          max. resid. norm for Fia-block=  3.087D+03 for orbital      1a
          max. resid. fock norm         =  1.669D+04 for orbital      1a
 mo-orthogonalization: Cholesky decomposition
          Delta Eig. = **************** eV



Note the positive energy of iteration 5.
Hereafter my control file generated using define:

$title
$symmetry c1
$user-defined bonds    file=coord
$coord    file=coord
$optimize
 internal   off
 redundant  off
 cartesian  on
 global     off
 basis      off
$atoms
h  1-14                                                                        \
   basis =h 6-31+G*
c  15-31                                                                       \
   basis =c 6-31+G*
o  32-33                                                                       \
   basis =o 6-31+G*
s  34                                                                          \
   basis =s 6-31+G*
$basis    file=basis
$scfmo    scfdump=30   format(4d20.14)
$scfiterlimit       100
$scfconv        7
$thize     0.10000000E-04
$thime        5
$scfdamp   start=1.000  step=0.050  min=0.100
$scfdump
$scfintunit
 unit=30       size=0        file=twoint
$scfdiis
$maxcor    500 MiB  per_core
$scforbitalshift  closedshell=.4
$drvopt
   cartesian  on
   basis      off
   global     off
   hessian    on
   dipole     on
   nuclear polarizability
$interconversion  off
   qconv=1.d-7
   maxiter=25
$coordinateupdate
   dqmax=0.3
   interpolate  on
   statistics    5
$forceupdate
   ahlrichs numgeo=0  mingeo=3 maxgeo=4 modus=<g|dq> dynamic fail=0.3
   threig=0.005  reseig=0.005  thrbig=3.0  scale=1.00  damping=0.0
$forceinit on
   diag=default
$end


The basis file contains the external basis set.
I'm using Turbomole7.6
I have two questions:
1) is there any way to do basis set projection (i.e. use a smaller basis set first and the use the basis set needed).
2) I cannot impose hcore guess. By following the manual (i.e. using $scfmo none) I still have a EHT starting vectors in the output. Do you have any   experience with this? (dscf module)

Thank you to everybody in advance.
Best regards,
Marilu

 

[uwe]

Hi,

what did help for me was to
- use RI-DFT
- set $scftol 1d-15  as I described here: https://forum.turbomole.org/index.php/topic,401.0.html

With Turbomole 7.7 (should also work with 7.6 but probably not older versions), I used this simple control file as input:

$coord file=coord
$scftol 1d-15
$basis file=basis
$symmetry c1
$atoms
    basis = 6-31+G*
    jbas  = universal
$dft
  functional b3-lyp
$rij
$marij
$end

coord file contained your coordinates and basis file was what I have downloaded from https://www.basissetexchange.org/ .

Attached the input and the output of ridft. It converged in 15 SCF iterations.

I do not think that DFT is causing problems here as Hartree-Fock shows the same behavior.

The geometry has almost Cs  symmetry, and if you enforce Cs (call define and in the geometry menu 'desy 0.1', this will find Cs and also moves the coordinates to the exact positions of the symmetric structure), the calculation also runs fine without the need to set any keywords like $scftol.

Using other diffuse basis sets like def2-SVPD or aug-cc-pVDZ might be the better option.

Best Regards