Hello users/devs. Sorry in advance for a long post with lots of thoughts though switching a seemingly minor setting led me down the rabbit hole.
I have seen running simulations without wrapping (iwrap=0) being argued in the list from time to time so I decided to follow this advice. Now I am considering the analyses of such trajectories. From previous experience, wrapping sometimes cause issues and analyses needed to be run with noimage. This led me to think that the opposite must also be true. So for some analyses, like HBOND and RDF, analysing the “unwrapped” (iwrap=0) output trajectories from sander/pmemd as compared to “wrapped” (iwrap=1) trajectories could produce different results.
Doing some quick tests seems to show minute differences in the RDF analyses, indicating that RDF analyses might not suffer from major artefacts from unwrapped/wrapped simulations (somewhat surprising as there is a no image option for “radial”). However, the same does not seem to be true for HBOND analysis where I observe much larger deviation. Here there is an option to use “image”, which almost reproduces the output of performing autoimage before analysing.
What would be the recommendations?
Looking at radial/rdf
Analysing unwrapped, autoimaged or autoimaged with the “noimage" option all produces slightly different results. There are only small differences though they are present.
Looking at hbond
Analysing unwrapped, unwrapped with the “image" command or autoimaged all produces slightly different results. It here seems obvious that unwrapped is a really bad choice, so I’ll ignore this.
1) When would it be important to use the “noimage” option for radial analyses? I see the options though not the motivation and I have not been able to generate a difference larger than 0.001 with my trial sets
2) Is there any preference for using autoimage before performing hbond analysis or using the image option during the analysis.
Considering both efficiency and accuracy. Performing longer, large system simulations often produces large amounts of data. From an efficiency point of view, if there is a need to “autoimage” trajectories before analysing, the time consumption for this action and the extra disk space required may indicate using iwrap=1 during simulation being more efficient for these reasons. I have no idea regarding accuracy/performance hit of using the "noimage” option with radial so I donated know if this would outweigh the benefits.
How would this compare to using iwrap=0 and image option in hbond with regard to speed/accuracy?
Is there potentially any nice overview indicating which analyses could be affected by wrapping and/or recommendations for settings not listed with the particular analyses in the manual?
Best regards
// Gustaf
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Nov 10 2020 - 03:30:03 PST
