esp restrain


Click here for full thread
Forum Vet
3:02:39 PM PDT - Fri, Oct 26th 2012
I didn't forget about you, was in China for a week and since I have been back I looked at this.


It took me a long time to finally get a handle on the weight issue. If you look at the grid you define and the number of grid points you have a large number of grid points. However, in the paper you refer to, they only define a couple of radial shells, which end up at best to be about 2500 grid points for the potential to be fit on (the paper uses radial points at least 1.4 and at most 2.0 times the vdW radius). Why is this important, you should check the formulas in the paper. There is a dependency on the number of grid points. As a result, the more grid points you have the smaller the effect of the harmonic constant becomes.

I included an input deck with a different setup for defining the grid, which ends up closer to the on in the paper. We can never match the numbers in the paper, simply because we have a square cubic grid, while the paper uses spherical grids. However, with the grid points I generate in the input deck I do see the restraining effect as expected for the default constants.

Bert

and
Quote:P99 Oct 18th 12:24 pm
Unfortunately, I can not reproduce articles models 2 and 3 in NWChem too.
I have the impression that there is a problem with the units of “weight”.
Both the article and the documentation of NWChem says that it is the atomic units.
However, if you increase the hyperbolic constant in 50 times, and harmonic constant in 30 times, the results agree with the article.
Similarly, if the increase is 30 times the constant of my implementation (model 1), the results agree with the article.
So, I think that my (model 1) code is working properly, but it is necessary to deal with the input constants in all models.

The input file for methanol can be found here. Here are also modified sources. No binaries.
I allow the user defined target charges not only for harmonic functions, but for the hyperbolic too.



Quote:Bert Oct 17th 7:44 pm
Before you start coding it is key to understand the code, how it solves the equations you are interested in. You should not make guesses. What I recommend you do is to take the RESP paper and match up the coded equations to those in the paper. This will give you an understanding of what the code does and how it relates to the paper. Using the "q0" parameter should give you equation 1, at least when I match up the code. The b variable seems to match up with your b-variable in your equation 3.

I am not going to open files with binaries because of security issues. Have you done some tests that allow you to compare to the published data? This will allow you to test for correctness.

Bert