Re: amber-developers: fixup to max14

From: Robert Duke <>
Date: Wed, 12 Nov 2008 19:21:40 -0500

Okay, bear in mind, with reference to using pmemd, that the data structure
sizes have not been tweaked for this problem. It will be interesting to see
if it works, and amber10 + latest pmemd patches is required. I was
originally going to play with this, but so much strange stuff was going on
that I thought it best to stay out of the fray until you guys actually
figured out the wierd stuff vs. stuff that is actually potentially wrong
with the sander extra points implementation (which in turn could very well
exist in pmemd; I am sure Darden's assumptions about datasize limits were
picked up in pmemd, as the rationale behind a number of things was not
crystal clear (a situation where I will typically start off mimicking sander
and then respond to any problems that arise)). So if it just works, great!
If it doesn't, then we really need to understand the datasize limitations
(using -CB in an ifort compile for runtime array bounds checks will help in
Regards - Bob

----- Original Message -----
From: "Lachele Foley" <>
To: <>
Sent: Wednesday, November 12, 2008 7:00 PM
Subject: Re: amber-developers: fixup to max14

>> 1. Why are you doing lots of minimization in the first place?
> Because I was taught that it is always safest to minimize a
> structure before attempting a simulation. I think the idea
> is to prevent unexpected behaviors related to starting with
> a high-energy or highly-ordered system, etc.
>> (with explicit solvent) is to only do minimization if the starting
>> structure
>> is so bad that MD will fail. 14,000 cycles seems way beyond what is
>> necessary or productive.
> Dad used to call me a perfectionist... It's because the curve gets
> really flat there. I can stamp on the OCD a little if necessary.
>> 2. Related: why do ntb=0 if you have explicit solvent?
> Because using ntb=1 gave me a box that was 1/3 empty, right
> down the middle, and that caused the first n tries at a CP
> run to fail because the CV heating didn't help at all.
> I am considering trying to heat directly from the leap-generated,
> perfect-array structure, but am concerned about getting results
> like those mentioned in my answer to [1]. And, in a purely
> aesthetic sense, I find it odd to "heat" from a structure that
> really can't be stable at anything other than a very high energy,
> and even then only fleetingly. But, I can get over that easily
> enough if necessary.
>> 3. If I read correctly, you are starting with a very low density
>> of water (0.6 g/cm3)?
> Not on purpose! That's what leap made when I solvated the box. I
> was just wondering if that was a bug or a feature. It isn't that the
> solvent is hard to pack around, because I tried solvating a water
> molecule, which should pack perfectly, and got 0.6 g/cc. (and after
> I wrote this, I saw what you said below... but this is the answer...)
>> ... What you want to do is to get going with MD, get ...
> Yeah... I guess something like that will behave. I just didn't
> expect it to. The manual recommends "heating" (this really high-energy
> structure) at CV, but I wonder if I couldn't just set the desired
> temperature and let CP adjust the density.
> I did get a simulation to run long enough to have the CP adjust
> the density to something reasonable. I'll check out the structure,
> but when I looked halfway through, it still looked good. So, I
> can just use that restart for future starting structures.
>> 4. You seem not to be using pmemd. Of course, this is good in that it
>> may
>> expose sander bugs, but Bob's code is a lot less likely to have weird
>> corners
>> that mysteriously hang, and it will almost always be faster to boot.
>> After
>> we
>> twisted his arm to get extra points into pmemd, we should take advantage
>> of
>> it.
> Oh... sorry. Yes. I'll do that. Tomorrow. I hope. Thanks, Bob! I was
> only slow because the prepared configs are all for admittedly faster but
> proprietary compilers, and it might take a while for me to tweak it for
> the
> compiler I have (gfortran is likely).
>> (a) use the option closeness parameter (say set to 0.7) to get a little
>> more
>> reasonable initial density;
> Ah... thanks.
>> (b) we eventually need a TIP5PBOX setup (rather than just solvating with
>> TP5).
> Is that difficult to do? I think I thought that's what was happening, so
> I
> didn't even look for an option to pack things tighter.
>> I hope this all works--I want to find out what sucrose does with TIP5P
>> and
>> the corresponding glycam potentials!
> I did get a simulation to run long enough to have the CP adjust
> the density to something reasonable. It took about 13,000 steps
> to do it. I've only got about .97 ns of (what is probably) decent
> data now. Want it? How much? I can't do much with it for a few
> days. I'll be running a longer simulation in the meantime.
> :-) Lachele
> --
> B. Lachele Foley, PhD '92,'02
> Assistant Research Scientist
> Complex Carbohydrate Research Center, UGA
> 706-542-0263
Received on Fri Dec 05 2008 - 14:36:17 PST
Custom Search