apbs-users
|
From: Tushar Jain <jain.tush@gm...> - 2005-02-24 22:00:02
|
Hi, I am currently working in the McCammon group. I am interested in using APBS with advanced Monte-Carlo(MC) sampling strategies. I have managed to link and run APBS from within the C code that performs the MC moves. Since only a few atoms change positions from one snapshot to the next, I was wondering if there was a way to speed up APBS for the calculating the energies of the new configuration, namely perform "incremental calculations", since the electrostatic field will be largely the same at distances far away from the moved atoms. More specifically, I had the following questions: 1) Is there a way to pass APBS the solution from the previous iteration? Would such a strategy be useful in reducing time for the new solution? 2) In the case of focussing calculations, a call to initMG destroys the pbe and pmg objects of the previous calculation. Is this necessary if the goal is to reuse or repopulate these objects for incremental solutions? 3) Have any of the users implemented such incremental calculations or have insights about the best course of action to accomplish the above? Thanks, Tushar. |
|
From: Nathan Baker <baker@bi...> - 2005-02-28 15:55:14
|
Hello -- > I am currently working in the McCammon group. I am interested in using > APBS with advanced Monte-Carlo(MC) sampling strategies. I have managed > to link and run APBS from within the C code that performs the MC > moves. Great! > 1) Is there a way to pass APBS the solution from the previous > iteration? Would such a strategy be useful in reducing time for the > new solution? Yes, but it's not straightforward and it's not clear what sort of speed-up it would give you. Let me know if you really want more details. > 2) In the case of focussing calculations, a call to initMG destroys > the pbe and pmg objects of the previous calculation. Is this necessary > if the goal is to reuse or repopulate these objects for incremental > solutions? This isn't necessary; however, some arrays used within Vpmg get overwritten during solution of the PDE. You would need to institute checks for this behavior before reusing too many things. > 3) Have any of the users implemented such incremental calculations or > have insights about the best course of action to accomplish the above? I've played around with this a bit. Many PB-based dynamics methods do not update the electrostatic potential unless an atom moves more than a particular distance. I've made some initial steps towards implementing this in our parallel focusing setting; i.e., updating local/fine maps at a different frequency than global/coarse maps based on atom displacements. However, I do not currently have any estimate on errors for these types of updates. -- Nathan A. Baker, Assistant Professor Washington University in St. Louis School of Medicine Dept. of Biochemistry and Molecular Biophysics Center for Computational Biology 700 S. Euclid Ave., Campus Box 8036, St. Louis, MO 63110 Phone: (314) 362-2040, Fax: (314) 362-0234 URL: http://www.biochem.wustl.edu/~baker |