The use of XMCQDPT2 in the situations you have described
results in energies which are better approximations to the
answers from the fully uncontracted theory.
As you may remember, the fully uncontracted limit of XMCQDPT2
is exactly size-extensive and separable theory. With XMCQDPT2,
this limit is gradually achieved upon extension of the model space,
the limit itself corresponds to the dimension of a model space
being equal to the overall number of SCFs.
Here, one can consider the use of XMCQDPT2 as a way how the results
from MRMP2 can be improved. In particular, this improves their
size-consistency. In addition, CASCI coefficients are allowed
One should be careful performing such calculations over a wide range
of molecular geometries. To avoid discontinuities on the PES one needs
to ensure that the CASSCF states forming the model space are always
the same i.e. that there are no discontinuities in the structure
of the model space. This may be difficult to achieve! This may require
changing the dimension of an Effective Hamiltonian. In the worst case,
one may even need to resort to MRMP2.
On Fri Mar 8 '13 1:12am, Thomas wrote
>I would like to ask a question about the use of XMCQDPT2 and MRMP2 for ground state calculations.
>Let us first assume the following case. I'm studying the reactivity of a particular molecule and am only interested in the ground state. Because not all points on the ground state PES can be properly described by a single-determinant method, I'm using CASSCF. On the geometries found with CASSCF, I would like to use a MR-PT2 method to account for dynamic electron correlation.
>Now, I've checked several points on the PES with XMCQDPT2 and found that for all of them the ground state is already well-described at the CASSCF level. When I check "EIGENVECTORS OF THE EFFECTIVE HAMILTONIAN" in the XMCQDPT2 output, I see that the lowest PT2 root is always predominantly described by the lowest CASSCF root. Contributions from higher lying CASSCF states are all below 0.1. Based on previous advice on this forum, I subsequently changed my XMCQDPT2 input to:
$XMCQDPT NSTATE=10 EDSHFT=0.02 WSTATE(1)=1,-0 AVECOE(1)=1,-0 $END
>However, because the description at the CASSCF level is already good, I assume I could also use MRMP2 instead of XMCQDPT2. My question is: what would the value of XMCQDPT2 over MRMP2 be in a case like this? Does the use of an effective Hamiltonian in XMCQDPT2 make any difference here?
>By the way, I realize that this is a very general question and that the answer could depend on the system under investigation. However, even if this is the case I would still be interested in the personal experience that some of you might have with XMCQDPT2 vs MRMP2.
>Thanks in advance for any explanation.
[ This message was edited on Sat Mar 9 '13 at 0:38am by the author ]