[gmx-users] normal mode analysis (size issue)

[Erik Lindahl]

Hi,

On Jun 19, 2007, at 5:26 PM, Liwei Li wrote:

> I want to carry out normal mode analysis on a large protein (about  
> 10,000 atoms) without using coarse grain approximations. Amber is  
> out of the question since it can only handle about 5,000 atoms for  
> NMA. I am wondering if GROMACS can deal with this assuming running  
> on a 64-bit machine with 8G memory. I truely appreciate it if you  
> can help.
>

If you use release 3.3 or later you won't even need lots of memory  
since we're using sparse matrix representation and diagonalization as  
soon as you have cutoffs, which should always be the case if you're  
doing NMA in vacuo.

We regularly do such NMA on 25,000-atom systems.

The best interaction form is usually shifted interactions with a  
cutoff of 1.2 to 1.4 nm. First minimize with steepest descent, then  
_double_ precision l-bfgs. This is virtually the only case when you  
absolutely need double precision. The rlist variable should be  
~0.2-0.3nm larger to conserve energy with l-bfgs, since that helps  
convergence.

Cheers,

Erik





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20070619/14c525b2/attachment.html>