AW: Tutorial example of replica exchange

From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Thu Mar 10 2016 - 16:53:22 CST

Using explicit solvent is possible, but cause of the required potential energy histogram overlap of neighbored replicas, you need lots of replicas as the energy differences between temperatures are too high to reach a reasonable exchange probability otherwise. There are improved algorithms which address this issue e.g. TIGER2 – Temperature Intervals with Global Exchange of Replicas.

 

Using npt is technically possible, but you need to implement a way to compute a suitable target pressure at each temperature. On the other hand, there’s a problem with proving the balanced condition with npt, since it changes the dimensions in conformational space and so can’t be proven to produce a boltzman weigthed sampling, which as I understood generally means, than you can’t know if the folding pathway your REMD takes, can happen in reality the same way.

 

Norman Geist

 

Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im Auftrag von Dr. Eddie
Gesendet: Donnerstag, 10. März 2016 19:26
An: namd-l_at_ks.uiuc.edu
Betreff: Re: namd-l: Tutorial example of replica exchange

 

Thanks Josh. However, now this has raised another question: can you perform a REMD with explicit solvent?

 

If you don't have PBC and you heat to 600k won't the water just evaporate away. Even if the sample was not heated that much couldn't the protein reach the surface of the water sphere?

Sorry for all the questions, but that you for your help!

Eddie

 

On Thu, Mar 10, 2016 at 10:04 AM, Josh Vermaas <vermaas2_at_illinois.edu <mailto:vermaas2_at_illinois.edu> > wrote:

Hi Eddie,

I don't think there is. The way I like to think about replica exchange methods is that you are running N parallel simulations that happen to be in communication. Just because two replicas exchange temperature/biases, it doesn't prevent each simulation from being its own continuous simulation (albeit one where the conditions keep changing). In this way the periodic boxes don't swap between simulation, and whatever barostat you picked will still try its darndest to keep the replica at the set pressure.

-Josh

 

On 03/10/2016 09:46 AM, Dr. Eddie wrote:

Thank you Norman, that helps!

 

For anyone else who had the same/similar problem I solved it by compiling Namd (2.11) from source and got it working with mpi (not CUDA) with the steps above. All multicore binaries (w/wo cuda) failed with those errors (see above).

 

Now that it runs, I cannot find any information on periodic BC or using a Piston with REMD. Is there some limitation?

Thanks,

Eddie

 

On Thu, Mar 10, 2016 at 2:03 AM, Norman Geist <norman.geist_at_uni-greifswald.de <mailto:norman.geist_at_uni-greifswald.de> > wrote:

Hey,

 

1) The jobX.conf will just put together the *_base.namd the *_remd.namd and and will source the replica.namd script which actually does the REMD. Also it is used to source additional restart.files from job1 and higher. You will get output files named like the job. So it’s just some kind of output separation if you want to restart the simulation multiple times.

2) Shouldn’t you specify how many processes to launch or threads to use? Otherwise try to use a non GPU version an see if it helps. There’s seems to be a problem with CUDA REMD, especially when using GBIS.

 

Let us know

 

Norman Geist

 

Von: owner-namd-l_at_ks.uiuc.edu <mailto:owner-namd-l_at_ks.uiuc.edu> [mailto:owner-namd-l_at_ks.uiuc.edu <mailto:owner-namd-l_at_ks.uiuc.edu> ] Im Auftrag von Dr. Eddie
Gesendet: Dienstag, 8. März 2016 23:56
An: namd-l_at_ks.uiuc.edu <mailto:namd-l_at_ks.uiuc.edu>
Betreff: namd-l: Tutorial example of replica exchange

 

Hi all,

I am trying to learn how to REMD with namd 2.11 by following the (spartan) tutorial <https://urldefense.proofpoint.com/v2/url?u=http-3A__www.ks.uiuc.edu_Research_namd_2.11_ug_node69.html&d=BQMFaQ&c=8hUWFZcy2Z-Za5rBPlktOQ&r=zFfoK61upjM5BwyoRAsX8dLq7rwWm8aw7r7dqtjgcCE&m=0tR0d-bxwpiIUv6mbDGaj1Wr4Hu_Mwp3MMx8x1Imbjs&s=K4ShHOBc3RHhIQu07hNgw172edRLwiuO7S4_Ur-6y0U&e=> . I have two problems:

1) I don't understand what jobX.conf are doing exactly and why two are commands are needed (first job0.conf then job1.conf)

2) I can't get it to run:

~/bin/NAMD_2.11_Linux-x86_64-multicore-CUDA/lib/replica/example$ mpirun ./../../namd2 +idlepoll +replicas 8 job0.conf +stdout output/%d/job0.%d.log

Charm++: standalone mode (not using charmrun)

Charm++> Running in Multicore mode: 1 threads

--------------------------------------------------------------------------

mpirun noticed that process rank 0 with PID 23811 on node heartofgold exited on signal 11 (Segmentation fault).

--------------------------------------------------------------------------

 

Nothing comes out so I can't see any errors:

 

~/bin/NAMD_2.11_Linux-x86_64-multicore-CUDA/lib/replica/example$ ls output/*

output/0:

total 0

 

output/1:

total 0

 

output/2:

total 0

 

output/3:

total 0

 

output/4:

total 0

 

output/5:

total 0

 

output/6:

total 0

 

output/7:

total 0

 

 

I've tried using +idlepoll since this is a gpu build, but this changed nothing. Any help or idea would be most welcome!

Thanks,

Eddie

 

-- 
Eddie
 
 
-- 
Eddie

This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:21:51 CST