On Wed, 29 Sep 1999, Patrick Carl wrote:

>
Hi, It's nice to see someone use the COSMO program that I wrote.
What NDIV actually does is this:
The atoms are tesselated into 60 points each. The resulting triangles are then
subdivided NDIV-1 times to give 60*2^(NDIV-1) triangles. The GEPOL algorithm is
applied to these points and some of them are modified or removed depending on
the type of surface you requested. These fine points are still mapped to the
original 60 points. The surface areas are calculated using these fine points to get more accuracy in the diagonal elements of the A-matrix (surface-surface).
Then these fine points are combined back into the original 60 points (or less,
as some of them may have been completely removed) and the position of the
center of the new triangle may change because it is a weighted average of
SURVIVING fine triangles. So the number of points you end up with should
always be approximately 60*#atoms (factor of 2 either way, probably). You can
specify a finer division (FD) with a newer version of the code, but this is
not recommended as it takes an exhorbitant amount of time to actually do the
solvation calculation. See the following table for HCl running on one of my
previous group's DEC alphas with ND=5

        atomic units
FD E_T E_el CPU time (s)
1 -0.23115 -0.00812 1.2
2 -0.23124 -0.00829 4
3 -0.23126 -0.00833 50
4 -0.23127 -0.00835 1000

The bottleneck is clearly the solvation part of the calculation, which scales
at about N^2.5, where N is the number of surface points. The accuracy gained
in this case is 2-3 % at most.

-Cory Pye

>
> I am curious about what the keyword Ndiv in the Solvation block actually
> does. The max value is 5 however it generates about 500 point charges
> when set to this value. In the original referance it mentions the

Are you referring to the GEPOL reference or the COSMO in ADF reference,
the manual may have said in press, but it's now published
C. C. Pye and T. Ziegler, Theor Chem Acc 101 396-408 (1999)

There will be a new version coming out which will have the COSMO-RS model, as
soon as I find the time to debug it. The code is written.

> division of the unit sphere into triangles for determining the SAS,
> however the same is said in the ADF manual. I thought that the number of
> triangles desired would be as high as possible to most appropriately
> characterize the SAS and that Ndiv was this number of triangles. Am I
> wrong? Does it scale by some factor? What is this factor?
> Any info on this would be appreciated.
> Thank you all.
>
> Patrick Carl
> University of Iowa
> Chemistry Dept.
>
>
>

   *************
 ***************** ! Dr. Cory C. Pye
*** ** ** ** ! Postdoctoral Fellow
** * **** ! Theoretical and Computational Chemistry
** * * ! cory_at_ucalgary.ca
** * * ! http://www.cobalt.chem.ucalgary.ca/cory
*** * * ** !
 ***************** ! Les Hartree-Focks
   ************* ! (Apologies to Montreal Canadien Fans)
Received on 1999-10-04 07:15:42