Hi Rosana –
This is a warning message intended to encourage you to explore the sensitivity of your results to the calculation parameters. In your case, I would recommend varying the “pdime” (increase it) and “ofrac” (increase it) parameters to see how much they influence your results.
However, I would also encourage you to clarify the problem you’re solving with APBS. Absolute electrostatic binding energies aren’t usually very meaningful – quite often you’re more interested in relative values of binding energies as differences calculated when testing multiple mutations, solution conditions, titration states, etc. It is likely that these relative energies are more stable with respect to grid setup (within reason).
Pacific Northwest National Laboratory
From: Rosana Collepardo [mailto:firstname.lastname@example.org]
Sent: Thursday, May 09, 2013 2:56 AM
Subject: [apbs-users] Convergence problems in very large calculation
I'm trying to calculate the electrostatic energy between two nucleosomes in a dinucleosome complex using apbs, and I'm having problems converging the calculation against grid size.
The calculation requires a lot of memory, but using fine grid spacings of 0.5 and 0.35 A requires less than the ~28GB apbs limit.
For those grid spacings the resulting internucleosome electrostatic energy (dinucleosome - nucleosome1 - nucleosome2) is not well converged and gives very large values. For instance, it changes from -341 kJ/mol for 0.5A to -563 kJ/mol for 0.35 A. (No evident errors or warning messages are reported by apbs)
The problem is that if I further decrease the grid spacing, I need to split the calculation as explained in the apbs website. I used the asynchronous parallel procedure and prize.py to calculate the dimensions of the sub grids, and then summed the energies for each calculation. A new problem is that when I split the calculation I start getting warning messages for the subgrid calculations (Unusually large potential values detected on the focusing boundary! Convergence not guaranteed for NPBE/NRPBE calculations!).
Furthermore, the resulting energy (direct sum of the energies given by each of the subgrid calculations) seems to diverge now, it is -1740 kJ/mol for a grid spacing of 0.2A and to -3697 kJ/mol for a grid spacing of 0.15A. I also tried changing ofrac, but without any noticeable improvement.
I wanted to know if anyone has experience with very large systems and if you could give me a few hints as to what I could do in order to converge this calculation.
With thanks and best wishes,