Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly  
We are NATO-free zone

Re: Manual parallelization of frequency calculations by computing Hessian in parts


The (numerical) calculation of Hessian in parts is possible. For that, the following
options in the group $FORCE should be used:

IASTRT=integer and IAEND=integer. These allow one to request computation of just a
                                 portion of a complete $vib info. "Vibs" will be
                                 computed starting from the atom number IASTRT and
                                 up to and including atom IAEND. Naturally, Hessian
                                 will not be computed and vibrational analysis will
                                 not be performed. The computed "$vib" information is
                                 stored in the IRCDATA and PUNCH files. Use of these
                                 options allows one to distribute large Hessian job
                                 over very large number of nodes by splitting this job
                                 into multiple independent parts.
                                 Defaults are IASTRT=1 and IAEND=overall number of
                                 atoms. These specific settings imply that a complete
                                 $vib group will be computed, followed by the
                                 computation of Hessian and vibrational analysis.

NOVIB0 (logical) When active, disables recomputation of wavefunction at
                the initial geometry ("vib0") during numerical Hessian
                restarts or when IASTRT/IAEND options are used.
                Default is .false. i.e. wavefunction at the initial geometry
                will be recomputed thus ensuring that computations of
                all "vibs" will use the same initial guess for orbitals.

On Tue Dec 13 '16 11:49am, GrEv wrote

>I wonder if there is an option to calculate particular components of the Hessian matrix, from (a) particular atom(s). Such an option would be useful to parallelize frequency calculations, by 1) computing Hessian in parts on different machines, 2) merging together the $VIB sections from those calculations, and 3) restarting the calculation to diagonalize Hessian to obtain frequencies and normal modes. This would speed up frequency calculations very much.

>Best regards,

[ Previous ] [ Next ] [ Index ]           Tue Dec 13 '16 6:40pm
[ Reply ] [ Edit ] [ Delete ]           This message read 138 times