different colvars trajectory with two LAMMPS versions

[karthik3327]

[jhenin]

Thanks for the report, but could we have a little more information? At least the colvars input, and some info on the simulation.

[karthik3327]

For a small system, I ran ABF simulations with two different LAMMPS versions..

trajectory length is 19ns.

In 6Apr13, sampling is almost uniform . But in the other version, it very much different.. I continued the run further, but not able to get uniform sampling in the latest version..

I can the see difference from colvars.traj files too

I will mail the input files...

Thanks

Karteek Kumar

[karthik3327]

Please find the attachment

On Thu, Jul 4, 2013 at 6:06 PM, Jérôme Hénin notifications@github.comwrote:

Thanks for the report, but could we have a little more information? At least the colvars input, and some info on the simulation.

— Reply to this email directly or view it on GitHubhttps://github.com/colvars/colvars/issues/13#issuecomment-20474771 .

[jhenin]

Sorry, Github won't let you attach files here. You can send them directly to my email address ( "lastname"@ibpc.fr ).

[jhenin]

Okay, another point is that these are stochastic simulations, so they may not be reproducible due to numerical details: the July 1 simulation could be an "accident". Did you run several? If not, can you try and reproduce the behavior?

[karthik3327]

Now, I ran different windows for several times with two versions and I can see significant difference.. I can clearly say this may not be due to stochastic simulations

[jhenin]

Okay, I would need one thing: can you run very short simulations, say, 100 time steps, with the two code versions, with the following options:

colvarsTrajFrequency 1

(in the colvar: )

outputAppliedForce yes outputSystemForce yes

And then send me for each simulation:

• the log of the standard output
• the colvars.traj file

Thank you, Jerome

On 5 July 2013 06:47, karthik3327 notifications@github.com wrote:

Now, I ran different windows for several times with two versions and I can see significant difference.. I can clearly say this may not be due to stochastic simulations

— Reply to this email directly or view it on GitHubhttps://github.com/colvars/colvars/issues/13#issuecomment-20501186 .

[karthik3327]

what should be the full samples value??

whether log of standard output means log.lammps

Thanks Karteek Kumar

[jhenin]

On 5 July 2013 16:10, karthik3327 notifications@github.com wrote:

what should be the full samples value??

All other parameters should be kept the same as your current ones.

whether log of standard output means log.lammps

Yes, that's it.

Jerome

[jhenin]

Karteek: thank you for reporting this. There was indeed a bug introduced on June 10, leading to system forces being calculated incorrectly in the presence of biases. It should be fixed by commit 76442825320cf8d7d87ea5c4aa5d38c5ff113ae2. In my tests, I can now reproduce the behavior of the April version. Please give it a try!

[karthik3327]

I have recompiled LAMMPS with new version with modified files.. but still I can see the difference.

[akohlmey]

it is easy to mix up files the way the LAMMPS build system works. would you mind trying with a precompiled binary from here http://git.icms.temple.edu/rpm/ ? just download the most suitable lammps-latest rpm and then unpack it with rpm2cpio and run the lmp_g++ executable by itself. thanks.

[akohlmey]

is this resolved now or not? please let us know

[karthik3327]

It's running fine now Thanks Karteek kumar

On Thursday, July 18, 2013, Axel Kohlmeyer wrote:

is this resolved now or not? please let us know

— Reply to this email directly or view it on GitHubhttps://github.com/colvars/colvars/issues/13#issuecomment-21184200 .