[gmx-users] Normal Mode Analysis in Double-Precision

[Ran Friedman]

Inon Sharony wrote:
> Dear Ran,
>
> I'm still unsure of why I sometimes can get double-precision output
> and sometimes not. I can't find what I'm doing differently in each
> case, however I now know what to look for in order to be sure that I'm
> getting what I need. Obviously, the need for double-precision stems
> only from the need to minimize the structure well before calculating
> the Hessian. When I get a maximal force of more than 1E-6 I will get
> some negative eigenvalues (which the program sets as zero) and some
> other eigenvalues which are nearly zero. In all, there are six such
> eigenmodes which I disregard as rigid-body degrees of freedom (3
> translational and 3 rotational). The next modes have larger positive
> eigenvalues.
If the info in the .gro file is in single precision, the values in the
.tpr will be with 3 significant digits and then zeros. Sorry if I wasn't
clear about this.
> As for the fix in eigio.c: I made lines 110-112 into COMMENTS (they
> were not comments in the original file in the source). I recompiled
> the entire program (didn't know at the time that I can "make" only one
> function). Now the g_anaeig_d works fine (as best as I can tell).
Oops. I had the lines already commented, probably because I corrected
this on my machine some months ago.
Ran.