Firefly and PC GAMESS-related discussion club


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



Re: Numerical gradients and CASSCF/XMQDPT2

Alex Granovsky
gran@classic.chem.msu.su


Dear Thom,

sorry for some delay with my reply.

>As I'm currently trying to find out what is going wrong, I would like to ask some more information about the initial guess that is used throughout the gradient calculation. Does each displacement start with the converged orbitals of the starting structure, or does each displacement use the converged orbitals of the previous displacement? If the latter is the case, how does this work together with the XP mode? And, at the end of a gradient calculation during a geometry optimization, which orbitals will be used as starting orbitals for the next geometry?

>Also, is there a way to control this behavior?

As to numerical gradients, XP mode and starting orbitals,
the situation is as follows. At the very initial geometry
each process group generates the same initial guess MOs as
directed by the input. These orbitals are used for the
very first MCSCF step. Then, each process group proceeds with
displacements required to complete gradient using MOs of its
previous local displacement as starting guess for the next one.
When the gradient is completed, there is always a synchronization
point where the global master sends to all process groups the saved
converged MOs for the very first geometry computed by the global
master. This geometry is always the unperturbed geometry for which
the numerical gradient is computed. Basically, the whole process
is then repeated as explained above, the only difference is that
these particular orbitals obtained from the master process are
used as the initial guess for the first MCSCF procedure of the
next set of MCSCF/XMCQDPT2 calculations required to complete
gradient. At present, there is no way to alter this behavior.

To summarize, at the beginning of the each next numerical
gradient evaluation, initial orbitals used by the different
groups are exactly the same. They become different in the course
of computations but will be synchronized again upon completion of
gradient calculations.  

Hope this helps.

Kind regards,
Alex





On Fri Oct 5 '12 6:19pm, Thomas wrote
-------------------------------------
>Dear Alex,

>I'm currently trying to do a XMQDPT2 numerical gradient calculation. Hereby, I have the problem that somewhere during the calculation of the gradient something goes wrong with the active space: the orbitals and state ordering after calculation of the gradient is completely different from those of the converged orbitals of the starting structure.

>As I'm currently trying to find out what is going wrong, I would like to ask some more information about the initial guess that is used throughout the gradient calculation. Does each displacement start with the converged orbitals of the starting structure, or does each displacement use the converged orbitals of the previous displacement? If the latter is the case, how does this work together with the XP mode? And, at the end of a gradient calculation during a geometry optimization, which orbitals will be used as starting orbitals for the next geometry?

>Also, is there a way to control this behavior?

>Thanks in advance.
>
>
>Kind regards,
>Thom


[ Previous ] [ Next ] [ Index ]           Tue Oct 9 '12 0:25am
[ Reply ] [ Edit ] [ Delete ]           This message read 813 times