<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
On 15/05/2012 6:17 PM, David de Sancho wrote:
<blockquote
cite="mid:CALUsGp3u5xtDPDPoXcuq4aQr7KqGRd929m2Eo4apQJKc4AyPsQ@mail.gmail.com"
type="cite">Dear all
<div>I have been following Justin Lemkul's tutorial for the
lysozyme simulations</div>
<div><a moz-do-not-send="true"
href="http://bevanlab.biochem.vt.edu/Pages/Personal/justin/gmx-tutorials/lysozyme/index.html">http://bevanlab.biochem.vt.edu/Pages/Personal/justin/gmx-tutorials/lysozyme/index.html</a></div>
<div>I am using Gromacs 4.5.5. </div>
<div><br>
</div>
<div>My concern is with the energy conservation in the
implementation of Bussi's velocity-rescaling thermostat.</div>
</blockquote>
<br>
Check out their paper and see what they claim for conservation
properties.<br>
<br>
<blockquote
cite="mid:CALUsGp3u5xtDPDPoXcuq4aQr7KqGRd929m2Eo4apQJKc4AyPsQ@mail.gmail.com"
type="cite">
<div> In step 6 of the tutorial an NVT equilibration is run using
tcoupl = V-rescale. The temperature equilibrates quite rapidly
as shown in Justin's webpage. Essentially, T fluctuates around
its equilibrium value after ~ 2 ps. However looking at the
"conserved energy" I find that there is a drift that does not
seem to plateau even by the end of the 100 ps run (see
attachment).</div>
</blockquote>
<br>
Different observables equilibrate over different time scales.<br>
<br>
<blockquote
cite="mid:CALUsGp3u5xtDPDPoXcuq4aQr7KqGRd929m2Eo4apQJKc4AyPsQ@mail.gmail.com"
type="cite">
<div><br>
</div>
<div><font style="background-color:rgb(255,255,255)">I have tried
to sort this out myself by using the following settings:</font></div>
<div>
<div style="font-family:arial,sans-serif;font-size:13px"><font
style="background-color:rgb(255,255,255)">(1) change
lincs_iter from 1 to 2.</font></div>
<div style="font-family:arial,sans-serif;font-size:13px"><font
style="background-color:rgb(255,255,255)">(2) change from
PME to PME-Switch, which for NVE Gromacs recommends as more
accurate (I also modified: rlist=1.0, rcoulomb=1.0,
rvdw=1.4).</font></div>
<div style="font-family:arial,sans-serif;font-size:13px"><font
style="background-color:rgb(255,255,255)">(3) change to
vdwtype = Shift, so that errors due to cutoffs were
eliminated.</font></div>
</div>
<div style="font-family:arial,sans-serif;font-size:13px">
<font style="background-color:rgb(255,255,255)">None of this
seems to help.</font></div>
<div style="font-family:arial,sans-serif;font-size:13px"><font
style="background-color:rgb(255,255,255)"><br>
</font></div>
<div style="font-family:arial,sans-serif;font-size:13px">
<font style="background-color:rgb(255,255,255)">Actually Justin
himself has helped me and found that with the following
settings the conservation is considerably better</font></div>
<div><font style="background-color: rgb(255, 255, 255);"
face="arial, sans-serif">
<div>
=== shift settings ===</div>
<div>vdwtype = shift</div>
<div>coulombtype = PME</div>
<div>rlist = 1.4</div>
<div>rcoulomb = 1.4</div>
<div>rvdw = 1.0</div>
<div>rvdw_switch = 0.8</div>
<div>Still there is a drift in the "conserved quantity" which
seems a quite severe problem.</div>
</font></div>
</blockquote>
<br>
"Severe" sounds like an over-description. The drift is 0.15% in the
conserved quantity, beginning from a non-equilibrated starting point
and only continuing for about the shortest equilibration period
anybody could imagine using these days. There's a lot of
approximations going on (PME, rigid bonds, static point charges) and
maybe drift at this level is not significant. Still, it should get
smaller if you just run for longer.<br>
<br>
Mark<br>
</body>
</html>