# [gmx-users] Poor exchange probability for REMD

Ben Reynwar ben at reynwar.net
Mon Nov 21 20:17:56 CET 2011

```The exchange probability will just effect the computational
efficiency.  It's entirely possible that it won't even effect that,
since the dynamics of the replica flow up and down the temperature
range may be limited by the dynamics of your protein rather than the
exchange frequency and probabilities.  Most likely, you have bigger
things to worry about than exchange probabilities.

Things I would start by doing are:

Do a decently long non-REMD simulation of the protein at the high end
of the temperature range.  Inspect the trajectory visually to make
sure that it is is sampling the configuration space you are interested
in.  If it's unfolding too much, it's going to be tough to find the
global minimum again, but if it's not unfolding enough then you might
not sample what you want to.  There is no guarantee that there even
exists an upper temperature that will work.

Look at the rates of replica movement up and down the temperature
ladder.  How long does it takes replicas to move all the way from the
bottom to the top.  You want your total simulation time to be
considerably longer than that.  Will that be feasible?

Cheers,
Ben

On Sat, Nov 19, 2011 at 7:03 PM, ÏéÇ« ¿× <iconsdkxq at yahoo.com.cn> wrote:
> Dear Mark,
>> Pre-equilibration at the right temperature is always a good
>> idea, not just for REMD.
>
>  Right temperature? I guess you mean pre-equilibrate at every replica temperatures, do you think so?
>
>> However for NVT REMD you need to
>> equilibrate the other temperatures at the same box size that
>> is sound for the temperature at which you want to make
>> observations.
>
>  If i am interested in the properties at 3OOK, i may equilibrate the system at 300K with NPT ensemble, then equilibrate system at every replica temperatures with the same box size obtained from the NPT simulation at 300K.Is it right?
>
>> Even given that, there is no strong reason to suppose that
>> a temperature distribution following a simple mathematical
>> formula should lead to equal exchange probabilities on a
>> "real" system with free energy bottlenecks. Knowing that one
>> might need to be adding more replicas at relevant
>> temperatures is something that can only be determined
>> empirically - and probably from more than 2ns.
>
>   Yes, that's right! While the exchange probabilities were unequal for neighboring temperatures and maybe very high (0.3~0.4), i think i should remove some replicas at which the exchange probabilities were higher and add some replicas to increase the upper limit of the temperature range for REMD.
>   Meanwhile, anther question, what's the proper range for the exchange probability?  Does the high exchange probability impair the properties we want to observe or just a waste of computational resources?
>   Best regards!
>   Xiangqian Kong
>
>
> --- On Sat, 11/19/11, Mark Abraham <Mark.Abraham at anu.edu.au> wrote:
>
>> From: Mark Abraham <Mark.Abraham at anu.edu.au>
>> Subject: Re: [gmx-users] Poor exchange probability for REMD
>> To: "Discussion list for GROMACS users" <gmx-users at gromacs.org>
>> Date: Saturday, November 19, 2011, 12:14 AM
>> On 18/11/2011 12:43 PM, ÏéÇ« ¿×
>> wrote:
>> > Dear GMX users，
>> >     Recently i am performing the
>> REMD simulation with Gromacs program and the temperature
>> distribution for each replica was predicted with the server
>> "http://folding.bmc.uu.se/remd/". However, after a 2-ns
>> short test simulation with 64 replicas  , i checked the
>> exchange probability for the neighboring replicas and find
>> the exchange probability was about 0.3 to 0.4 (as the file i
>> attached )but the desired probability was 0.2. Meanwhile, i
>> found the exchange probabilities fluctuated markedly for
>> each pair of  replicas while ideally we may hope they
>> were consistent with each other.  I don't know whether
>> this is acceptable or must be fixed up, or  a longer
>> simulation time and pre-equilibrium at different replica
>> temperature for each replica was needed.
>>
>> Pre-equilibration at the right temperature is always a good
>> idea, not just for REMD. However for NVT REMD you need to
>> equilibrate the other temperatures at the same box size that
>> is sound for the temperature at which you want to make
>> observations.
>>
>> Even given that, there is no strong reason to suppose that
>> a temperature distribution following a simple mathematical
>> formula should lead to equal exchange probabilities on a
>> "real" system with free energy bottlenecks. Knowing that one
>> might need to be adding more replicas at relevant
>> temperatures is something that can only be determined
>> empirically - and probably from more than 2ns.
>>
>> Mark
>>
>> >      The system i simulated
>> includes 60074 atoms which consists of 155 residues,19173
>> waters and 14 chloridions. I first equilibrium the system
>> for 2ns with NPT ensemble at 300K, then start the REMD
>> simulation for 64 different replicas (temperature ranges
>> from 300 to 386K) with NVT ensemble and the exchange attempt
>> time was 2-ps(1000 integral steps).
>> >      Now i was totally puzzled and
>> don't know how to figure out these problems,i am eager for
>> the help from you and any suggestions will be greatly
>> appreciated!
>> >      Best regards!
>> >      Xiangqian Kong
>>
>> -- gmx-users mailing list    gmx-users at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-users
>> Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search
>> before posting!
>> Please don't post (un)subscribe requests to the list. Use
>> the www interface or send it to gmx-users-request at gromacs.org.
>>
> --
> gmx-users mailing list    gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.