z-matrix for linear molecule with 3 atoms?

[Jack Daniels]

Dear all,

even as a total n00b with nwchem I feel a little embarrassed for posting this question but I simply can't get the following input (which is adopted from the examples in the manual) to work in nwchem 7.2.0:

scratch_dir .
title "FHF"

echo

charge -1

geometry units Angstrom
zmatrix
F
H  1  FH
F  2  HF    1 FHF
variables:
FH 2.0
Constants:
HF 1.0
FHF 180.0
end
end

basis noprint
 * library 3-21G
end

scf
 noprint "final vectors analysis"
 singlet
end
task scf optimize

The idea obviously is to calculate a linear F....H-F adduct - which works fine when using a xyz-type input but in order to do constraint distance scans later on I want to use the zmatrix. But this crashes right at the start with:

 !! There are insufficient internal variables: expected     2 got     3
 !! Either AUTOZ failed or your geometry has changed so much that the
 !! coordinates should be regenerated.

Putting in noautoz does not help, nor does it help to switch off symmetry. As the input above crashes before the first SCF there should also be no change in geometry. However, a single point calculation (task scf) is working fine and if I change the distances somewhat I can get the optimization to run for 1 or two cycles before it crashes with the same error. The exact same z-matrix works just fine in GAUSSIAN and ADF btw, so I am running a little out of ideas here.

Thanks in advance for any help on this!

Jack

[Edoardo Aprà]

Instead of using the zmatrix input, using the adjust keyword from an xyz input and adding a dummy x atom should suit your needs.

geometry
 f 0. 0. 0.
 h 0. 0. 2.
 f 0. 0. 3.
 x 3. 0. 2.
end
geometry adjust
 zcoord
  bond  2 3 1. hf constant
  angle 1 2 3 180. fhf constant
  end
end

https://nwchemgit.github.io/FAQ.html#input-problem-autoz-fails-to-generate-valid-internal-coordinates

https://nwchemgit.github.io/Keywords-on-the-GEOMETRY-directive.html

[Jack Daniels]

Thank you very much! This worked well for the scan. Unfortunately, I ran into the next issue/set of questions:

1. On the long run I intend doing vscf calculations with NWChem. I started by an attempt to reproduce Chabans results for H2O (MP2/TZP). I obtain (without any error messages) these values:

  RESULTS OF VIBRATIONAL SCF CALCULATION: Frequencies, cm-1
  Mode      Harmonic     Diagonal      VSCF       PT2-VSCF
   1        3924.71      4021.95      4063.26      4030.09
   2        3806.67      4033.25      4010.87      3956.20
   3        1638.94      1776.22      1686.21      1680.68

which not only disagrees with the reference but appears to be nonsense alltogether, as the harmonic freq is actually lower than the VSCF one. Input was like this:

scratch_dir .
title "H2O TZP Optimized geom"

noprint

charge 0

geometry units au
   O       0.0000000000     0.0000000000    -0.2244495306
   H      -1.4306412400     0.0000000000     0.8977705872
   H       1.4306412400     0.0000000000     0.8977705872
end

basis noprint
 * library TZP
end

vscf
 ngrid 16
 coupling pair
end

mp2
 print none
 tight
end

task direct_MP2 vscf

any thoughts on this?

2. In parallel I am trying to learn how to run VSCF on my linear molecule and this ends up with another symmtry issue, I guess: Per default NWChem lists (harmonic) frequencies for all 3N degrees of freedom but  the VSCF calculation on H2O couples and lists only the 3 virbational states - as I expected it would. 

However, this is not the case for my tri-atomic linear molecule: here, VSCF uses all 3x3=9 "modes" and (as a result?) is failing the SCF convergence for many of the coouplings involving the rotational degrees. So my question is two-fold I guess: Is this really intended behaviour to include all the 3N degrees of freedom and if so, may I trust VSCF results if *any* of the VSCF-steps does not converge?

Thanks again for any help,

Jack