Firefly and PC GAMESS-related discussion club

 
Learn how to ask questions correctly  
 
 

 

Alex Granovsky
gran@classic.chem.msu.su

let me further clarify what Sanya and me are trying to explain.

Of the charts of your first post, look at the E0 chart.
This quantity is computed based on the effective Fockian
obtained using the state-averaged first order density matrix.
The weights of states in averaging are defined by the
wstate and avecoe arrays of $xmcqdpt group.

The idea is very simple. E0 is computed before and without any PT
and in fact depends only on the CASSCF orbitals and roots. If you
look at the graph, you will see that multiple small jumps and
the large jump are already here! Why do you then expect
PT will result in a smooth curve? That's simply impossible.

If you look at NOs at your points # 135 and # 136, you'll see
that CASSCF orbitals are dramatically different and that there
is a huge jump in CASSCF energy between these two points.
There was clearly a bifurcation of your active space between these
two points so that the active space at point # 136 cannot be
described as a small modification of that at point # 135.

To avoid these bifurcations, you may try to increase active space
and/or to use different state-averaging in SA-MCSCF procedure.
However, there are no any universal solutions here. You need to
try and check results, etc... etc...


On Fri Aug 17 '12 1:57am, sanya wrote
-------------------------------------
>Well, this CASSCF looks really terrible :( The suspicious thing is that all states of interest consist of many configurations with small weights (instead of 1-2 dominant configurations with some small additions), and most of these configurations are double, triple, or even higher excitations relative to the closed-shell configuration. This means that you orbitals are not optimal at all. Probably, your (substantially nonlinear!) CASSCF minimization ended up in a wrong minimum. This situation frequently occurs in CASSCF. Did you notice the outlying points in your XMCQDPT curves? they may be of the same origin. The states you are trying to improve by XMCQDPT may be simply the wrong states.

>Please, keep in mind that CASSCF is NOT a black-box method. You should know in advance what are you going to see. In your example, HN3 particle (zero charge, singlet state) dissociates into N2 (zero charge, singlet state?) and HN (zero charge, singlet state, too?). Is this mechanism really possible? Does hydrogen azide decompose through a monomolecular mechanism? Are the first-stage products really N2 and HN? If so, are you confident in the charge and/or multiplicity of the resulting particles? Probably, you should try ALDET instead of GUGA to include states of different multiplicity.

>Anyway, you should start at the very beginning, i.e., with a hydrogen azide molecule, whose ground state is well described by a single close-shell reference (or by some relatively simple and understandable combination of configurations), and whose lowest excited states mostly consist of single excitations from the ground state. Even if you want to average over 3 lowest states (S0, S1, and S2), set NSTATE=5 or even more and add NTRACK=3 to $MCSCF group. This will help you detect root flipping if it occurs.

>On Thu Aug 16 '12 6:05pm, Serzhan wrote
>---------------------------------------
>>On Thu Aug 16 '12 2:13pm, sanya wrote
>>-------------------------------------
>>>The problems in QDPT frequently arise from errors or problems in CASSCF. From your outputs, I see that root flipping occurs: in 136CASSCF-10-10PT34 State 1 was closed-shell (all occupancies are close to 2 and 0), while in 135CASSCF-10-10PT34 I see fractional occupancies of orbital 5 (1.43) and 6 (0.57). Can you show us your CASSCF inputs and outputs?
>>>In addition, you'd better set $XMCQDPT group as follows:

>>> $XMCQDPT INORB=1
>>>! (instead of kstate(1))
>>>  nstate=34
>>>! (additional parameters: you need 3 states, so 3 1's)
>>> wstate(1)=1,1,1,-0 avecoe(1)=1,1,1,-0
>>>  edshft=0.02
>>> $end

>>>I think this improvement alone will not help you, unless you find out what is wrong with your CASSCF.

>>Thank you for your reply.

>>I attached the CASSCF inputs and outputs with this post. It seems the MOs in active space remain intact, while the CI roots differ at points of interest. So that can be the cause of the jump. However, the root flipping, you've mentioned, is due to the physics of the problem. So It will be great if the energy profile could be kept smooth, while root flipping is present.

>>Thanks for your help,
>>Serzhan

[ This message was edited on Sat Aug 18 '12 at 1:11am by the author ]