On Thursday, January 30, 2003, at 06:50 PM, Charles Schwieters wrote: > Dear Charles, > thank you very much for your kindness in helping me with your comments. > The last time you didn't understand one of my questions. The point is > in fact > more general and has to be with the availability of general protocols > so that > starting with an extended model and providing the experimental data > one end > up with the best possible model. We are not in that stage yet, right? > Now, I am puzzled with the output energy from 'rama' that for some > calculations is negative. According with the original article E is > defined as > - - -k*log(Pi) where Pi is the probability, that I think should be > between 0 and > 1, so I don't understand how E can be negative. Do you know? > Thanks again, > Hugo
The formulation for the DELPHIC torsion and DELPHIC position terms' potentials of mean force is E = -ln Pi, where Pi is the probability of being in bin i. Pi is now defined as (nExamples(i) / volume(i)) / overallDensity. This change was necessitated by some of the averaging I do to get a defined potential energy in unpopulated regions. If there are no examples in a particular bin, I start including that bin's neighbors along each axis until I reach a certain minimum number of examples. Then I divide by the total volume of the bins I just searched to get a "local density." In order for the units to work out correctly, I need to divide this local density by the overall density (totExamples / totVolume). The upshot is that regions of space that have greater-than-average density of examples now have negative energies, and regions with less-than-average density of examples now have positive energies. Hope this clears up the confusion. --JK
