Date: Fri, 18 Jan 2002 09:50:29 -0800
From: David Case
Subject: Re: nmode memory requirements

On Fri, Jan 18, 2002, Michael Jakusch wrote:
>
> Total memory required : 235879241 real words
> Total memory avail : 3500000 real words
>
> So I tried to increase the value of MAXMEMX (in sizes.h), because this
> was set to 3500000 before. This resulted in the compiler to complain
> about:
>
> _nmode_.f:54:
> real*8 x (MAXMEMX)
> ^
> Array `x' at (^) is too large to handle
>
> This seems to be a problem of the compiler (g77), in the first
> place. However, even if a different compiler could do the job, this
> would still be much more than the physical ram of my machine (Athlon
> with 256M).
>
> Is normal mode analysis really so demanding that it can hardly be done
> on usual proteins???
>

The second derivative matrix is a 3N x 3N matrix, although you only need
to store half of it because it is symmetric.

So, for a 5000-atom system, you would need (15000)(15000)/2 =
112,000,000 words, or 900 MBytes just to store the matrix. Depending on
how many eigenvectors you want, you would also need a large amount of
memory to store them.

There are ways to carry out large matrix analysis problems on machines
with limited memory, but these are not implemented in Amber.

So, the unfortuante answer to your question is "yes": normal mode
calculations on "usual proteins" (looks like yours has about 300 amino
acids) is beyond the scope of PC's with the amount of memory you describe.

...regards...dac