Molcas Forum

Davinor

Member

Registered: 2015-12-01

Posts: 4

unreasonably high GRID_IT charge

Hello,

I was doing some calculations on a grid of charges created by GRID_IT, which works fine for smaller ions.
For heavier ions, however, i encounter grid points with extremely high charges that add up to >1000 total charge
(after multiplying by the grid volume and summing up of course).

Varying any thresholds in the calculations or integrals didn't help, as did switching off relativity. Close to the nucleus
this fails for a number of atoms and basis sets. Any idea on what is going wrong or how to fix it?
A sample input for La(III) is posted below. Choosing a finer grid also does not improve the result, but produces more
points with increased charge. The error is also independent of the version of Molcas i use.

 &SEWARD &END
 Basis set
 	La.ano-rcc-vtzp    
	La .0 .0 .0
 End of Basis set

 &RASSCF &END
Symmetry
 1
Spin
 1 
nActEl
 6 0 0
Inactive
 24                        
Ras2
  3                      
Iterations
 200 100 
ciroots 
 1 1 1
End of input  

  &GRID_IT    
all           
ascii
noluscus
NPOINTS
  50 50 50
gap
  7.0

in the .grid file then there is

...
0.676338322625E+00
0.256639094136E+01
0.856216054475E+01
0.706768428043E+02
0.136614980143E+06
0.706768428043E+02
0.856216054476E+01
0.256639094136E+01
...

close to the nucleus. Shifting the grid slightly also does not help.

valera

Administrator

Registered: 2015-11-03

Posts: 129

Re: unreasonably high GRID_IT charge

Grid_it was never made for accurate density calculations - the main purpose of the code is to produce cartesian grid which  can be used for producing of isosurfaces - so, smothness is more important than precision. especially close to atoms.
I also would try to do the following test - make a dense grid (e.g. atomic like) using (as I recall) GRID keyword so x y z coordinates explicitly given, and integrate the density. If you get wrong numbers of electrons - yes, the case catch my attention

Davinor

Member

Registered: 2015-12-01

Posts: 4

Re: unreasonably high GRID_IT charge

Dear Valera,

thanks for your reply. I tested your suggestion and created a dense grid with some 8 million points around La(III).
It took Molcas almost an hour to even read the giant input file, but it finally managed to give me the density.
Things improved (for some reason) compared to the grid generated by grid_it itself, but still the total number of
electrons integrates up to over 565 (which should be 54).

Also the density around the nucleus is still high

0.101575376300E+02
0.147346765161E+02
0.359156282221E+02
0.985203859564E+02
0.193206508486E+03
0.497991345946E+03
0.121652549097E+04
0.497991345946E+03
0.193206508486E+03
0.985203859564E+02
0.359156282221E+02
0.147346765161E+02
0.101575376300E+02

If there is any other way to compute the density i would be glad to use that, but grid_it was the only one i found
in the manual.

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: unreasonably high GRID_IT charge

You can generate densities with Molden (though it doesn't support h functions), Pegamoid and possible other programs too. I think part of the problem is you are sampling the density at the nucleus, which is very high, and therefore a simple-sum integration does not work very well. For what it's worth, a density computed with Pegamoid in a 14-bohr-side cube with 50×50×50 points gives an "integrated" density of 44, which improves with finer grid and larger volume.

valera

Administrator

Registered: 2015-11-03

Posts: 129

Re: unreasonably high GRID_IT charge

Davinor,
I prefer to move this discussion out.

Davinor

Member

Registered: 2015-12-01

Posts: 4

Re: unreasonably high GRID_IT charge

Pegamoid densities acutally work. It has some trouble when there are dummy atoms without basis functions,
but inserting some zeros by hand does the trick.

Thank you both very much for your suggestions and help.