Re: [AMBER] Amber and Short Bonds

From: Ross Walker <ross.rosswalker.co.uk>
Date: Mon, 26 Mar 2012 09:44:30 -0700

Hi Pietro,

You'll need to provide quite a lot more information here for us to be able
to help you. Firstly where are the short bonds coming from? Are these
specialized (artificial bonds) in which you have special bond parameters? -
Or are they just standard bonds (like CT-CT say) that start out short? In
the second case here it is no wonder the system blows up and gives you NANs,
your forces will be crazy. A short time step can help but if your parameters
are not appropriate for such short bonds you are toast.

I also don't understand what you mean by changing the 'forces' doesn't help?
- Are you going into the code and editing the force array? Or do you mean
changing the 'force constant' for the bond doesn't help? In the latter case
it will make little difference to stability unless you modify the
equilibrium distance as well.

Also you should note that there are a LOT of assumptions in the code,
especially pmemd and pmemd.cuda that assume you will have reasonable length
bonds. I would suggest running with Sander first to see how it does before
using pmemd or pmemd.cuda - the more tuned code tends to assume that you
have a stable system to begin with.

All the best
Ross

> -----Original Message-----
> From: Aronica, Pietro [mailto:pietro.aronica07.imperial.ac.uk]
> Sent: Monday, March 26, 2012 8:05 AM
> To: amber.ambermd.org
> Subject: [AMBER] Amber and Short Bonds
>
> Hello,
> I'm new here, so this is more theoretical than an actual request for
> help.
> I've been dealing with really short bonds, we're talking about less
> than 0.5 , and Amber, somewhat understandably, is not too pleased with
> it. My simulations keep failing, and I've changed many parameters
> around to find out if there's a solution to this. So far, reducing the
> time step is the only thing that will allow the simulation to reach the
> conclusion: masses, charges and forces don't seem to have an impact. IS
> there something I'm overlooking? Or is the timestep the only solution?
> I'm mostly using pmemd on a GPU. When my simulations fail, they start
> displaying NaN as energy values (and restart coordinates). Sometimes
> they'll go on for a while before failing: visualising in vmd these
> simulations shows that the short bonds move around very rapidly when
> compared to the other ones until the whole structure abruptly
> disappears when the energy goes NaN.
> Thanks in advance
> Pietro Aronica
> Imperial College London
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Mar 26 2012 - 10:00:03 PDT
Custom Search