Page 1 of 1
bader population
Posted: Thu Jan 25, 2007 4:33 pm
by gopan
hello, may be it is a stupid question, but I cannot resist myself. :(
How many times we need to refine the EDGE, and what is the criterion ?
I asked it because I run a computation and the automatic refining went up to 100.
or it is just I am doing some thing wrong ?
-gopan
Posted: Thu Jan 25, 2007 11:40 pm
by graeme
Ah, this is a very fair question, and one that we are also trying to understand better. Theoretically, the method should only require one edge refinement. That is to say, the first analysis should only yield errors one grid point away from the true Bader dividing surface. Then, a single refinement step should fix any errors. This is the case for ideal charge densities (calculated from analytic functions). Practically, however, in the fastest (linear scaling) version of the algorithm, this does not hold. By default, the code will continue to refine the edges until no points are reassigned. I'm very surprised that you found 100 iterations - something is really strange here.
Can you test something for us on this system? Try running with the flag:
-m max
This changes the scaling of the algorithm so that it is slower for very large systems, but it should also eliminate any mis-assigning that is going on. If this make a significant difference, that would be good to know. Also, in the end (after as many refining steps as are needed) the two calculations should give the same charges.