[gmx-users] CG-MD simulation of protein, always crash with protein

[Martti Louhivuori]

On 19 Apr 2010, at 06:49, Trang wrote:
> The protein topo is too large, so I put it here. http://pastie.org/926617

The protein topology is fine, but you could try also with elastic  
bonds instead of dihedrals in the extended regions (--elastic option).  
Elastic bonds are more robust than dihedrals, especially if running  
with a larger timestep (30fs).

> Here is the system topology.
>
> ----------------------------
> #include "../martini_v2.1.itp"
> #include "../martini_v2.1_aminoacids.itp"

The last line is unnecessary. Use it only when simulating single  
aminoacids instead of a polypeptide.

> Minimized structure showed no close contact (with distance cutoff  
> 1.6, even to 1.9). Production run stopped at

That's only 0.16-0.19nm, so overlap is still possible.

> step 175552 (5266.56 ps) with Segmentation fault, preceeded with a  
> lot of Lincs warnings (I set ring_bonds = "constraints").  This is  
> the mdp file, this file goes along with the system in vaccuum that  
> crashed. I'm not sure if

What's different this time around? Last time you said you couldn't run  
it even in vacuum.

Sounds like you are just hitting statistical limits with your  
combination of (extended) dihedrals + 30fs timestep. Use a smaller  
timestep (20fs for example) or switch to local elastic bonds for the  
beta-strands... and solvate the system (without overlap) so the  
simulation makes some sense!

-martti-
--
Post-doctoral research fellow
Moleculaire Dynamica
University of Groningen
Nijenborgh 4, 9747AG Groningen, the Netherlands
tel. +(31) 50 363 4339 | fax. +(31) 50 363 4398