[gmx-users] Calculate observed "Acceptance Ratio" from REMD simulation

[David van der Spoel]

On 2010-08-21 15.05, henri mone wrote:
> Calculate observed "Acceptance Ratio" from REMD simulation
>
> Dear Gromacs users,
>
> I used the Gromacs how-to [1] to setup a REMD simulation. The targeted
> acceptance ratio was P_acp=15%, I used the temperature generator
> web-service [2] to get the temperature spacing. From the test
> simulation I want to calculate the observed acceptance ratio P_obs.
> The last md.log file gives a "Replica exchange statistic". I got
> following questions regarding on how to calculate P_obs.
>
> 1. To get P_obs from simulation I have to calculate the ratio between
> the successful exchange and total exchange attempts.
> Is this correct? I checked several papers and books (Frenkel/Smit,
> etc.), I only found the formula how to calculate the temperature
> spacing given P_acp. But I couldn't find a formula or definition how
> to calculate P_obs from a simulation. Is the above definition of P_obs
> correct?
>
> 2. The md.log file "Replica exchange statistic" gives in my case
> following summary [3]. For temperature T0 to T1 there wehre 664
> successful exchanges. The number of exchange attempts were 1213. The
> ratio 664/1213 gives 55%. Would the P_obs=55% , is this correct?
>
> 3. A observed ratio of P_obs=50% given a targeted acceptance ratio of
> P_acp=15% is quite high.
> What can be the reason for this large discrepancy?

Good question. We used two force fields (OPLS and GROMOS) for the 
development. We did however use Berendsen thermostats, which is not a 
good idea, and this might induce a deviation, but I would expect it to 
be small. The most obvious thing could be that the system used is not 
the one for which you input parameters on the webserver. The webserver 
wants the number of water molecules, if you supplied the number of 
atoms, that would give you a factor of three in energy.


> 4. Could the Gromacs users, admin etc. please add some more
> information to the REMD Howto [1]. In particular regarding how to
> calculate P_obs?
> It would be a really useful for the Gromacs users.
> Can I as a Gromacs user add some more information to the howto or do I
> need a login?
>
> I would appreciate any help.
>
> Thanks,
> Henri
>
>
>
> 1: Replica-Exchange Molecular Dynamics howto
>     http://www.gromacs.org/Documentation/How-tos/REMD
>
> 2: Temperature generator for REMD-simulations
>     http://folding.bmc.uu.se/remd/
>
> 3: "md.log" file summary
>     Replica exchange statistics
>     Repl  2425 attempts, 1213 odd, 1212 even
>     Repl  average probabilities:
>     Repl     0    1    2    3    4    5    6    7    8    9   10 ...
>     Repl      .55  .55  .55  .56  .56  .54  .56  .56  .56  .56  ...
>     Repl  number of exchanges:
>     Repl     0    1    2    3    4    5    6    7    8    9   10 ...
>     Repl      664  684  667  692  659  627  699  688  699  683  ...
>     Repl  average number of exchanges:
>     Repl     0    1    2    3    4    5    6    7    8    9   10 ...
>     Repl      .55  .56  .55  .57  .54  .52  .58  .57  .58  .56  ...


-- 
David van der Spoel, Ph.D., Professor of Biology
Dept. of Cell & Molec. Biol., Uppsala University.
Box 596, 75124 Uppsala, Sweden. Phone:	+46184714205.
spoel at xray.bmc.uu.se    http://folding.bmc.uu.se