<div>Hi Michael,</div><div>I agree with you respect the recoding. But performance difference is of 2x. </div><div>I started inserting the features in 4.5.5 thursday and within today or tomorrow</div><div>I will end. I plan one day of debug. So totally 4-5gg...I will get back these </div>
<div>days as performance improvement :)</div><div><br></div><div>Of course, implementing seriously this features will take more time both for coding</div><div>and testing. But right now I am not planning to do it seriously.</div>
<div><br></div><div>Maybe later when 4.6 will be out and if it will give me problem.</div><div><br><br><div class="gmail_quote">2012/7/16 Berk Hess <span dir="ltr"><<a href="mailto:hess@kth.se" target="_blank">hess@kth.se</a>></span><br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div text="#000000" bgcolor="#FFFFFF"><div class="im">
<div>On 07/16/2012 06:08 PM, francesco oteri
wrote:<br>
</div>
<blockquote type="cite">
<br>
<br>
<div class="gmail_quote">2012/7/16 Berk Hess <span dir="ltr"><<a href="mailto:hess@kth.se" target="_blank">hess@kth.se</a>></span><br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div text="#000000" bgcolor="#FFFFFF">
<div>Using lambda scaling for non-bonded interactions is
going to be very slow.<br>
</div>
</div>
</blockquote>
<div><br>
</div>
<div>If it is done with the actual implementation, you are
right. But if mdrun recognize rest2=yes thzn it rescales the</div>
<div>parameters at teh beginning. Then a normal MD run goes.</div>
<div> </div>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div text="#000000" bgcolor="#FFFFFF">
<div> <br>
But this would scale the protein-protein non-bonded
interactions with lambda^2.<br>
</div>
</div>
</blockquote>
<div><br>
</div>
<div><br>
</div>
<div>Why? computing charge-charge interaction is:</div>
<div><br>
</div>
<div>S*qi*S*qj = (S^2 )*qi*qj</div>
<div>but S=sqrt(Bi/B) </div>
<div>so what I obtain is:</div>
<div>(Bi/B)* qi*qj</div>
<div><br>
</div>
<div>That is exactly what I want: rescaling by (Bi/B)
protein-protein interactions</div>
</div>
</blockquote></div>
Ah, sorry, that works indeed. <br>
<br>
Cheers,<br>
<br>
Berk<div><div class="h5"><br>
<blockquote type="cite">
<div class="gmail_quote">
<div> </div>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div text="#000000" bgcolor="#FFFFFF">
<div> Is that what you want?<br>
<br>
Cheers,<br>
<br>
Berk
<div>
<div><br>
<br>
On 07/16/2012 05:06 PM, francesco oteri wrote:<br>
</div>
</div>
</div>
<div>
<div>
<blockquote type="cite"> In the paper they just say that
non bonded rescaling is obtained rescaling
<div>LG epsilon by Bi/B and the protein charges by
sqrt(Bi/B).</div>
<div><br>
</div>
<div>I did simulation using the lambda approach and I
qualitatively reproduced </div>
<div>the results.</div>
<div><br>
</div>
<div>Just to give my contribution, I think the most
efficient way to implement REST2</div>
<div>is using the lambda approach as follows:</div>
<div><br>
</div>
<div>1) in .mdp introducing a new keyword like rest2 =
yes/no</div>
<div>2) using the init_lambda value to rescale the
parameters at the bootstrap of md </div>
<div> (es. in do_md or init_replica_exchange). </div>
<div>3) running the code as normal MD and swapping the
states </div>
<div><br>
</div>
<div>Since point 1 implies modification of grompp and
point 2 requires the knowledge of </div>
<div>the data structure, I am just implementing point
3 while point 2 is demanded to external </div>
<div>scripts.</div>
<div> <br>
</div>
<div>Francesco</div>
<div>
<div><br>
<div class="gmail_quote">2012/7/16 Berk Hess <span dir="ltr"><<a href="mailto:hess@kth.se" target="_blank">hess@kth.se</a>></span><br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div>On 07/16/2012 04:42 PM, Shirts, Michael
(mrs5pt) wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> One
question, since better REST2 is one item
on the todo list for 5.0, and I<br>
want to be thinking about the
possibilities.<br>
<br>
For solute-solvent energy rescaling, how
does one calculate the changes in<br>
solute-solvent energy without doing
multiple PME calls to decompose the<br>
electrostatic energy?<br>
<br>
I'm guessing that by since your are load
multiple topologies of the system,<br>
for each of those topologies, you can
explicitly write new nonbonded<br>
pairwise parameter terms between the water
and protein. This is a pain to<br>
do (need new ij terms for every parameter
pair) but straightforward in the<br>
end to automate. Is this how your are
planning to do it?<br>
<br>
Without PME, then it's straightforward to
decompose into solute-solvent<br>
energy / solute-solute / solvent-solvent
electrostatics, though for adding<br>
long term to Gromacs, we'd want to support
PME.<br>
</blockquote>
</div>
No, even without PME this is not
straightforward.<br>
Using a plain cut-off's for electrostatics is
horrible, so I'd say you want to use
reaction-field.<br>
But with reaction-filed the correction terms
are non pair-wise, as with PME.<br>
<br>
Cheers,<br>
<br>
Berk<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div>
<div> <br>
Best,<br>
~~~~~~~~~~~~<br>
Michael Shirts<br>
Assistant Professor<br>
Department of Chemical Engineering<br>
University of Virginia<br>
<a href="mailto:michael.shirts@virginia.edu" target="_blank">michael.shirts@virginia.edu</a><br>
(434)-243-1821<br>
<br>
<br>
</div>
</div>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div>
<div> From: francesco oteri <<a href="mailto:francesco.oteri@gmail.com" target="_blank">francesco.oteri@gmail.com</a>><br>
Reply-To: Discussion list for GROMACS
development <<a href="mailto:gmx-developers@gromacs.org" target="_blank">gmx-developers@gromacs.org</a>><br>
Date: Mon, 16 Jul 2012 16:13:16 +0200<br>
To: Discussion list for GROMACS
development <<a href="mailto:gmx-developers@gromacs.org" target="_blank">gmx-developers@gromacs.org</a>><br>
Subject: Re: [gmx-developers]
calculating poteintial energy into
do_md()<br>
<br>
Hi<br>
<br>
2012/7/16 Mark Abraham <<a href="mailto:Mark.Abraham@anu.edu.au" target="_blank">Mark.Abraham@anu.edu.au</a>><br>
<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> On
16/07/2012 10:59 PM, francesco oteri
wrote:<br>
<br>
Hi,<br>
what I am trying to do is
implementing the REST2 tecnique [1]<br>
The goal is enanching the
conforlational sampling splitting
the system in<br>
two parts: solvent and solute.<br>
The two groups have a different
APPARENT temperature because the
bonded<br>
interaction within the solute atoms
are rescaled by a factor Bi/B (<br>
Bi=1/Ti*Kb, Ti = apparent
temperature for the i-th replica,
B=Boltzmann<br>
constant) and the protein-water
interactions are rescaled by
sqrt(Bi/B).<br>
So the system temperature, regulated
by the thermostat, is equal along
the<br>
different replicas but the dynamic
of the different replicas<br>
changes becaus of the different
Boltzmann factor.<br>
To implement the tecnique, the naive
approach (adopted in [2]) is:<br>
1) Generating N-different topology
(N= number of replicas).<br>
2) Removing same check at the
beginning of the Replica Exchange
code, in<br>
order to run N replicas with equal
run parameters (temperature,
pression,<br>
ecc. ecc.)<br>
3) Changing the acceptance ratio
formula.<br>
<br>
This approach has been developped
for gmx4.0.3 but I need to use<br>
gmx4.5.5 because it has a native
implementation of the CHARMM
force-field.<br>
Since the code between the two
versions is different I am trying to
figure<br>
out how to reinvent the wheel.<br>
<br>
As shown in [3] and adopted in [2]
the same result can be obtained
using<br>
the lamda dynamics in gromacs.In
this case, only one topology has to
be<br>
generated. The state A correspond to
the replica at the lowest apparent<br>
temperature, while the state B
represents the highest temperature
replica.<br>
Intermediate replicas are generated
changing the init_lambda value.<br>
<br>
What has been found, this approach
results in slowest run.<br>
<br>
<br>
So, right now I got stuck at the
point 3. In fact to find:<br>
<br>
delta = B( Va(Xb) + Vb(Xa) -
Va(Xa) - Vb(Xb))<br>
<br>
I need to calculate Va(Xb) and
Vb(Xa)<br>
<br>
<br>
Va(Xb) is determined by the
potential of replica a and the
coordinates of<br>
replica b, so it is most effectively
evaluated on the processor that has<br>
the coordinates of replica b. If the
topology is the same at the replicas
a<br>
and b, then can Va(Xb) be computed
by replica b by rescaling suitably
the<br>
quantities that contributed to
Vb(Xb), based on the knowledge that
replica<br>
a is the exchange partner? Then the
quantities that are exchanged before<br>
the test are Va(Xb) and Vb(Xa).
Should be as fast as regular REMD.<br>
<br>
</blockquote>
<br>
This is my idea, indeed<br>
<br>
</div>
</div>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div>
<div> <br>
</div>
</div>
</blockquote>
</blockquote>
</blockquote>
</blockquote>
</div>
</div>
</div>
</blockquote>
</div>
</div>
</div>
</blockquote>
</div>
</blockquote>
<br>
</div></div></div>
<br>--<br>
gmx-developers mailing list<br>
<a href="mailto:gmx-developers@gromacs.org">gmx-developers@gromacs.org</a><br>
<a href="http://lists.gromacs.org/mailman/listinfo/gmx-developers" target="_blank">http://lists.gromacs.org/mailman/listinfo/gmx-developers</a><br>
Please don't post (un)subscribe requests to the list. Use the<br>
www interface or send it to <a href="mailto:gmx-developers-request@gromacs.org">gmx-developers-request@gromacs.org</a>.<br></blockquote></div><br><br clear="all"><div><br></div>-- <br>Cordiali saluti, Dr.Oteri Francesco<br>
</div>