Alex Granovsky
gran@classic.chem.msu.su
the problems with determinant CI code you have encountered are
caused by two reasons. First, the spin-tracking code used by
Davidson diagonalization routine sometimes can fail. Second,
more general problem is the loss of orthogonality and spin
purity as the consequence of multiple repeated re-orthogonalizations
and reconstructions of the expansion basis. Both these problems will
be addressed in the final Firefly v. 8.0.0. Meantime, I'd suggest you
the following modification of your input:
$DET NCORE=63 NACT=10 NELS=12 NSTATE=20 ITERMX=400 WSTATE(1)=1,1,1 ispin=0 nstgss=60 $END
The ispin=0 option projects out
all but singlet states in the initial guess. This does not
automatically mean that you get only 20 singlet states as the
result of diagonalization. Most likely, you'll get some quintet
states (S=2) as well. However, this should filter out all odd S
values (triplets etc...). Similarly, ispin=1 would filter out
all even S values (singlets etc...).
The nstgss=60 option requests more
states in the initial guess procedure. This is required as some of
them are triplets and states of higher multiplicity that should be
filtered off. The Firefly version you have will silently abort if
there are not enough singlet states in the initial guess. Current
beta prints the diagnostic message, and the final Firefly 8.0 will
automate the filtration process to be fully transparent.
Hope this helps,
Alex Granovsky
On Mon Dec 20 '10 4:06pm, Thomas wrote
--------------------------------------
>Dear Alex, dear Firefly-ers,
>First of all, many thanks for the post on
(X)MCQDPT earlier this month. It has been of great help to understand this technique and apply it to some smaller molecules. However, for the system I’m currently investigating I would like to ask for some additional information.
>For the system under investigation I’m interested in finding the energies for the 3 lowest states. The system has 4 pi bonds and 2 lone pairs which are in conjugation with the pi system, so I’ve opted for a (12,10) active space which includes all p orbitals, with state averaging over the 3 lowest states. I then did a XMCQDPT2 calculation with the following parameters:
>
$XMCQDPT NSTATE=15 EDSHFT=0.02 IROT=1 THRGEN=1D-10 WSTATE(1)=1,1,1,-0 AVECOE(1)=1,1,1,-0 $END
>However, the results are a bit surprising when I look at the eigenvectors of the effective Hamiltonian (see attachment for full input and output files):
>
1 2 3 *********************************** 1 0.979176 0.088076 -0.026794 2 0.035958 0.059822 -0.016755 3 -0.035267 0.704873 0.694772 4 0.010869 0.094671 0.009798 5 0.129268 0.135014 -0.201877 6 0.033920 0.143020 -0.159014 7 0.014135 0.021763 -0.020772 8 0.001207 0.004650 -0.018419 9 -0.048716 0.049059 -0.021054 10 -0.085162 0.248572 -0.207512 11 -0.056139 0.282944 -0.224556 12 0.054204 -0.289349 0.328600 13 -0.066811 0.447878 -0.487557 14 0.016651 -0.100756 0.093465 15 0.006428 -0.070395 -0.032100
>Apparently, the CASSCF states 10 till 13 contribute significantly to XMCQDPT states 2 and 3. My question is: is it common for high-lying states to be so important? Or does it rather show a problem with the underlying CASSCF states?
>
>
>Then there is a second question I would like to ask. If I were to proceed by including CASSCF states 10-13 (as well as states 5 and 6 which seem to be of importance as well) in both the SA-CASSCF as well as the XMCQDPT part of the calculation, I would write the following input:
>
$DET NSTATE=40 ITERMX=400 WSTATE(1)=1,1,1,0,1,1,0,0,0,1,1,1,1 $END $XMCQDPT NSTATE=20 EDSHFT=0.02 IROT=1 THRGEN=1D-10 WSTATE(1)=1,1,1,0,1,1,0,0,0,1,1,1,1,-0 AVECOE(1)=1,1,1,0,1,1,0,0,0,1,1,1,1,-0 $END
>However, this gives me problems during the CI optimization: after many iterations, one or more states become unconverged and will not converge again:
>
5 -1372.2196226568 0.00000114 5 -1372.2077781435 0.00017627 5 -1372.2072391615 0.00525372 5 -1372.2039774568 0.00568757 Warning - some CI states may have just been missed! -1 CONVERGED CI STATE(S) HAVE BECOME UNCONVERGED, NOW WORKING ON 3 STATES. After a few more iterations: 73 -3820.2279388183 30217.75359972 73 -3677.7641825114 28409.89326395 73 -2930.8951122879 15383.10968857 73 -2905.6192085034 14967.16114983 73 -2821.9694978003 14105.86660842 73 -1372.4833177907 0.00000041 73 -1372.4258925356 0.00000015 73 -1372.3874360397 0.00000021 73 -1372.3748222301 0.00000020 73 -1372.3321312588 0.00000017 73 -1372.2919746805 0.00000020 73 -1372.2521978693 0.00000018