[gmx-users] Re: Normal Mode limits for memory

[Nguyen Hoang Phuong]

> David van der Spoel wrote:
> > On Fri, 6 Aug 2004 statk002 at fiu.edu wrote:
> >
> >
> >>Hello Dr.Spoel
> >>I am trying to run a NM simulkation but keep running into memory allocation problems.
> >>
> >>The code that I input is
> >>g_nmeig_d -f nm.mtx -s 1CF3_nm.tpr -o eigenvalue.xvg -v eigenvector.trr
> >>
> >>The error I get is :
> >>Reading frame       0 time    0.000   Dimensionality of matrix: 17454
> >>Fatal error: calloc for hess (nelem=304642116, elsize=8, file gmx_nmeig.c, line 121): Cannot allocate memory
> >
> > You need 2.5 Gb memory for this operation. You could use a subset of your
> > system (only Ca for instance).
> >
>
> AFAIK, you cannot perform NM on a subset only. What people from the NM community often do is
> to switch to a more coarse-grained desciption of the system. However, such simplified forcefields
> are to my knowledge not implemented in gromacs.

or you may try with the technique called: Partial Hessian vibrational
analysis. With this method only a subblock of Hessian is diagonalized.
See, e.g. Theoretical Chemistry Accounts 107, 211, 2002

Phuong