<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<br>
Hi again,<br>
<br>
It seems that your test was too short for you to see the kind of
crashes we have been observing.<br>
Right after the first tests we also decided to move to the
"v-rescale" thermostat. All my latest tests have been done with this
and sometimes with both (to make sure it was not affecting the
results).<br>
<br>
A new test using a nstlist=4 and a tau_t=0.05 resulted in a crash
after 2.07 ns. The system distorted on the main chain's NH of the
last residue of HEWL.<br>
<br>
If you care to take a look at my system, here goes the files needed
to reproduce my crashes:<br>
<br>
<a class="moz-txt-link-freetext" href="http://dl.dropbox.com/u/3245083/Lyso_RF-test_GMX-4.5.x.rar">http://dl.dropbox.com/u/3245083/Lyso_RF-test_GMX-4.5.x.rar</a><br>
<br>
Once again, thank you for all the effort.<br>
Cheers,<br>
Miguel<br>
<br>
<br>
On 13-12-2011 17:02, Berk Hess wrote:<br>
<span style="white-space: pre;">> Hi,<br>
> <br>
> I did one more check. I can run a 128 residue protein for 100
ps with<br>
> both nstlist=4 and nstlist=5 without warnings. But I do see
that the<br>
> protein is much more above the target temperature than the
water.<br>
> <br>
> One difference I did notice is that you are using a Berendsen<br>
> thermostat. I would always use the Bussi thermostat
(v-rescale)<br>
> instead. With that thermostat you can make tau_t arbitrarily
small. <br>
> So I would try tau_t=0.05<br>
> <br>
> Cheers,<br>
> <br>
> Berk<br>
> <br>
> On 12/12/2011 09:27 PM, Miguel Machuqueiro wrote:<br>
>> <br>
>> Hi Berk, and all the devel guys.<br>
>> <br>
>> Once again, thank you for not giving up on us.<br>
>> <br>
>> It took me a while to do the tests to check your
suggestions and my<br>
>> results are not very famous.<br>
>> <br>
>> A decrease in the nstlist from 5 to 4 was not able to
provide<br>
>> stable MD simulations in my HEWL system. All attempts
(several by<br>
>> now) ended up in lincs crashed with several
stepxxxxxx.pdb files<br>
>> being generated. I also tried simulations using nstlist
values of<br>
>> "2" and "1" to be sure: - The system with a value of "2"
held on<br>
>> for a while, but eventually crashed after ~15 ns. I
reproduced this<br>
>> crash in 3 replicates. - The system with a nstlist of "1"
went<br>
>> through the 50 ns test without any problems.
Unfortunately, this<br>
>> simulation was very slow; 4 times slower than PME
(remember that<br>
>> nstlist=5 gives RF simulations much faster than PME).<br>
>> <br>
>> The idea that RF in previous versions of Gromacs was
stable<br>
>> probably due to error cancellation raises in me several
concerns. <br>
>> Unfortunately, with the correction in the current
implementations,<br>
>> I am not able to use the RF option for the treatment of
the long<br>
>> range electrostatics.<br>
>> <br>
>> If you are interested, I could send you the files used to
run my<br>
>> tests. Even though we modified our code in order to use
GMX version<br>
>> 4.0.7 (in use at the moment), we are still very
interested in using<br>
>> the latest and faster versions of Gromacs.<br>
>> <br>
>> Thank you and all the guys doing the hard work! Miguel<br>
>> <br>
>> <br>
>> <br>
>> On 27-11-2011 14:59, Berk Hess wrote:<br>
>>> Hi,<br>
>>> <br>
>>> I found a solution for your problem!<br>
>>> <br>
>>> The basic issue is that a time step for the long
range forces of<br>
>>> 5*2 fs = 10 fs is too long in a system where charges
move fast<br>
>>> (probably mainly the water hydrogens). With the old,<br>
>>> non-reversible, scheme there must be some
cancellation of <br>
>>> errors, which makes the integration stable (although
still very<br>
>>> inaccurate). With the new, reversible, integration
scheme the<br>
>>> hydrogens in the protein heat up by a factor 1.5 with
a<br>
>>> T-coupling strength of tau_t=0.1 (worse than the old
scheme)<br>
>>> and they become unstable. Changing the long time step
to 8 fs<br>
>>> (nstlist=4 with dt=2fs) reduces the energy drift by a
factor of<br>
>>> 4 and makes protein simulations stable. With this 8
fs the energy<br>
>>> drift, for my protein test system, with the new
scheme is a<br>
>>> factor 3 lower than with the old scheme.<br>
>>> <br>
>>> The proper solution is using PME. But you said you
couldn't use<br>
>>> that for whatever reason. I hope that that is not
because you<br>
>>> need pair forces. Note that the reaction field force
is NOT a<br>
>>> pair force, it is a collective effect of all charges
in the<br>
>>> cut-off sphere, very similar to PME.<br>
>>> <br>
>>> Cheers,<br>
>>> <br>
>>> Berk<br>
>>> <br>
>>> On 2011-11-03 18:52, Miguel Machuqueiro wrote:<br>
> </span><br>
<br>
<br>
-- <br>
============================================<br>
Miguel Machuqueiro<br>
Department of Chemistry and Biochemistry<br>
Faculty of Sciences, University of Lisbon<br>
Campo Grande, Edifício C8 (sala 8.5.47)<br>
1749-016 Lisboa, Portugal<br>
Tel. : +351 217500112 (int.ext.28547)<br>
Mobile: +351 967562285<br>
E-mail: machuque at fc.ul.pt<br>
www1: <a class="moz-txt-link-freetext" href="http://webpages.fc.ul.pt/~mamachuqueiro">http://webpages.fc.ul.pt/~mamachuqueiro</a><br>
www2: <a class="moz-txt-link-freetext" href="http://intheochem.fc.ul.pt">http://intheochem.fc.ul.pt</a><br>
______________________________________________<br>
<br>
<br>
</body>
</html>