Firefly and PC GAMESS-related discussion club

 
Learn how to ask questions correctly  
 
 

 

Alex Granovsky
gran@classic.chem.msu.su

sorry for somewhat delayed reply. Please see my comments below.

>I want to do a 2D relaxed surface scan with MCSCF, and then do a energy calculation on this surface with XMCQDPT2, I have the following questions:

>1) Can I first perform an optimization with MCSCF and then do a XMCQDPT2 energy calcultion with one input file?
>If not, the work will become complicated as I need to extract all the optimized geometries and MCSCF MOs from output and punch files, respectively. And I also need to prepare a number of XMCQDPT2 input files.

No, unfortunately it is impossible to perform a MCSCF geometry
optimization followed by the XMCQDPT2 calculation at the final
geometry using a single input file at present. We can consider
implementation of this feature as a minor improvement to be added
to the next (hopefully final) Firefly v. 8.0.0 RC public build as
this should be rather straightforward. Meanwhile one needs to
extract the data on the final geometry and orbitals
from the output files.

>2) If the answer for the first is no, I plan to write a perl script to do this work rather than use the RSURF in firefly,

To write a perl script (or to use some other tool) to handle
output and punch files generated by the runtyp=RSURFACE would
be somewhat more efficient.

> then I have another question about the natural orbitals:
>When the MCSCF was done, there are two types of orbitals in the punch file, one is MCSCF NATURAL ORBITALS, and the other is MCSCF OPTIMIZED ORBITALS. From this discussion club I know the following XMCQDPT2 calculation need the MCSCF OPTIMIZED ORBITALS, my question is:
>   which type of orbitals should be used in the next MCSCF calculation, where the geometry is close to the previous one.

You can use any of them as they correspond to the same wavefunction.
In practice it is somewhat better to use natural orbitals as
subsequent MCSCF job will typically converge faster using these
orbitals.

>I ask this question because I saw the following paragraph in the Molpro manual:
>"For a CASPT2 calculation, the zeroth-order Hamiltonian can be brought to a block-diagonal
>form when (pseudo)canonical orbitals are used.  This leads to fastest convergence.  It is there-fore recommended that in the preceding MULTI calculation the orbitals are saved using the CANONICAL directive (note that the default is NATORB)."

>It seems that the CASSCF should use the natural orbitals and the CASPT2 should use the (pseudo)canonical orbitals.

To be precise, XMCQDPT2 can be run using any converged MCSCF orbitals
and there is even no need to provide virtual orbitals as a part of input.
The reason of this is that by default XMCQDPT2 program performs
orbital canonicalization as the very first steps of its operations.

This requires CASCI integral transformation step, CASCI step,
computation of (state-averaged) one-particle density matrices,
and finally computation of effective Fock oparator.

One can avoid these steps by specifying

 $xmcqdpt iforb=0 $end 

in the input file. In this case, one must provide a full set of
well-converged canonicalized MCSCF orbitals. With Firefly
v. 8.0.0 this can be done by adding:

 $mcscf iforb=.t. sd=.t. $end 

to the input file controlling mcscf stage of XMCQDPT2 computations.
In addition, if these computations are performed as a separate step,
one needs to add

 $contrl wide=.t. $end

to the input above to increase the precision used to punch orbitals.
In the latter case, one needs to use the following cards in the input:

 $guess guess=moread norb=the_overall_number_of_mos $end
 $xmcqdpt inorb=2 $end 

You can find some extra information on this and related topics in
the following thread:

http://classic.chem.msu.su/cgi-bin/ceilidh.exe/gran/gamess/forum/?C352a1b60fby4-8294-827-00.htm

Hope this helps.

Kind regards,
Alex Granovsky

[ This message was edited on Sun Sep 30 '12 at 8:06pm by the author ]