ORCA 6.1.1: CHELPG charges and fragment-orbital initial guesses

In allocated CPU resources:
  bash run.sh
  bash analyze.sh work
  /public/toolkit/miniconda3/envs/ase-3.29.0/bin/python plot_results.py
Use a new extraction directory for a complete repeat. The workflow uses one
CPU process and one thread per native program; no GPU. Main ORCA is launched
by its full path, once per calculation, never wrapped in mpirun or srun.
ORCA 6.1.1 is the vendor Linux precompiled package, with OpenMPI 4.1.8 built
using GCC 13.3; it was not locally recompiled for this example.

Four starting molecules are calculated once: HF/def2-SVP water, PBE/def2-SVP
CF3Cl and two PBE/aug-cc-pVDZ water fragments at their dimer coordinates.
The two fragment GBWs are merged, and one MOREAD dimer SCF is actually run.
Four further HFwater property calculations use MOREAD NoIter and change only
CHELPG grid spacing, cutoff, radius convention or dipole constraint. They
preserve the original density and perform no iterative orbital optimization.
There are five iterative single points, four NoIter property drivers and
22 native program calls in a complete fresh workflow.

CHELPG output case.pc_chelpg begins with an atom count and a comment, then
Q charge x y z lines, with positions in angstrom. It is a point-charge
listing, not an ordinary element/XYZ molecular structure file. Printed
charges and positions have six decimals. Preserve atom ordering and units.
Mulliken, Loewdin and CHELPG charges describe different atomic partitions.
The nine CF3Cl positions are independent holdout points, not the fitting
grid. The point-charge model does not exactly reproduce quantum potentials,
particularly the anisotropy responsible for a halogen sigma-hole.

The merged starting orbitals concatenate the two actual fragment sets.
Their complete overlap matrix need not be the identity. Main ORCA explicitly
reads merged.gbw, reorthogonalizes it and converges the SCF. The example
checks atom/basis ordering, every signed/permuted fragment MO coefficient,
frozen fragment density, electron count, independent Gaussian AO overlaps,
and the converged target energy. Actual GBW/JSON MO dimensions are 41+41=82;
use actual coefficient/S-matrix dimensions when constructing matrices.
The source stdout also contains other basis-function statistics; these are
not used to infer the matrix dimensions in this package.

Native text data and original scientific reference output are included.
No software binaries, GBW or binary density containers are distributed.
Check reference data without another SCF with:
  bash analyze.sh
Plotting defaults to supplied reference data; edit the data path for new data.

Official references:
https://www.faccts.de/docs/orca/6.1/manual/contents/spectroscopyproperties/population.html
https://orca-manual.mpi-muelheim.mpg.de/contents/utilitiesvisualization/utilities.html
https://www.faccts.de/docs/orca/6.1/manual/contents/essentialelements/initialguess.html
