[gmx-users] publishing with development version of Gromacs

[chris.neale at utoronto.ca]

> I am running protein simulations in an explicit membrane  
> environment, and have found that the CVS development version of  
> Gromacs works much faster than the released version. Is it okay to  
> publish simulations performed on the development version? What kind  
> of validation would I need if I were to do this? For instance, if I  
> compared the first 10 ns of a simulation between the development and  
> released versions and found that they were substantially the same,  
> could I complete the simulations (40 more ns) with the CVS version?

I personally never run with the cvs version because I would never want  
to try to publish that version. However, if you are determined to do  
this then really the only important thing is that your experiments are  
reproducable. This means that you are going to need to ensure that a  
future reader may obtain the _exact_ version of gromacs that you used.  
I am not sure if this is currently possible via the gromacs site...  
probably not. Therefore you will need to provide a download of the  
version that you used. I am not even sure that you have licence to do  
that. Developers?

Some would say that you only need to personally stand behind the fact  
that future releases of gromacs are going to reproduce any results  
that you find. First of all, I am not sure how you would know that  
this is true if you are not a developer and second of all I don't  
personally believe that that condition is stringent enough.

Another point: how enormous is your system? 40ns should be possible  
with gromacs 3.3.1 during a couple of months of runtime for regular  
sized membrane proteins while using a single processor and less time  
while running in parrallel. Why bother with the hassle?

This is, of course, just my own thoughts on the topic.

If your run-time benifit is related to scaling then you could also  
consider my g_desort program that is available via user contributions.  
And then there is the single vs double precision topic.

Chris.