[molpro-user] inconsistent core/valence energy partitioning in MRCI-F12 calculation

[Irikura, Karl K. Dr.]

Hi,

When calculating a vertical excitation energy using MRCI-F12, we found one molecule (CrCl) for which we get alarmingly different results when the preceding CASSCF is run with the doubly-occupied orbitals "frozen" or merely "closed."  Here is the input file:

***, CrCl, illustrating mysterious inconsistency in core/valence MRCI partitioning (6A1 state and 8B1 state)
MEMORY,100,M
angstrom

geometry={
Cr;
Cl,1,r;
}
r=2.20219779

basis={
default,AWCVTZ
set,jk
default,def2-QZVPP/JKFIT
set,df
default,aug-cc-pwCVTZ/MP2FIT
s,Cr,1125.39,341.170,73.9951,31.2459,18.9340,9.25672,6.42575,3.38099,1.74898,0.984534,0.758085,0.337239,0.128563,0.075795,0.042102
p,Cr,421.364,189.344,39.4212,18.5562,9.17740,7.23806,3.89384,1.91685,1.28017,0.662766,0.331280,0.151677,0.073572,0.057624
d,Cr,161.521,50.5342,22.2636,9.60004,6.76492,4.13610,2.11763,1.20327,0.710380,0.315499,0.126834,0.064490
f,Cr,80.4841,30.8651,10.8536,5.83615,3.98155,2.21072,1.13224,0.589006,0.283022,0.162113,0.132695
g,Cr,31.5558,13.1525,6.19049,4.29099,2.33217,1.19468,0.492783,0.203392,0.640652
h,Cr,14.3240,6.43119,3.90826,2.30624,0.920789,0.385634
i,Cr,8.00125,4.77640,2.17128,0.883034
}

! sextet state (6S, 6A1)
{rhf
        wf,sym=1,spin=5;
}
! CASSCF(21/14)
{multi
        occ,13,5,5,1;
        closed,6,2,2,0;
!        frozen,6,2,2,0;
}

! MRCI
{mrci-f12,df_basis=df,df_basis_exch=jk,ri_basis=df
        core,6,2,2,0
}
gs=energd

! octet 8B1
{rhf
        wf,sym=2,spin=7
}
{multi
        occ,13,5,5,1;
        closed,6,2,2,0;
!        frozen,6,2,2,0;
}

! MRCI
{mrci-f12,df_basis=df,df_basis_exch=jk,ri_basis=df
        core,6,2,2,0
}
xs=energd

excitation_energy=(xs-gs)*TOEV

The corresponding output file yields EXCITATION_ENERGY               =       3.30659542  EV.
But if the two "frozen" lines are uncommented, the final result is EXCITATION_ENERGY               =       4.18284549  EV.

The biggest difference I see in the output files is the section of the MRCI describing the core/valence energies.  In the first job, this section for the two MRCI calculations (lower and upper states)  is:

Nuclear energy:                      98.04037687
Core energy:                      -1476.38524047
Zeroth-order valence energy:        -25.75585178
Zeroth-order total energy:        -1404.10071539
First-order energy:                 -98.88176684

Nuclear energy:                      98.04037687
Core energy:                      -1476.38451622
Zeroth-order valence energy:        -23.65850589
Zeroth-order total energy:        -1402.00264524
First-order energy:                -100.79974424

But in the second case (with the doubly-occupied orbitals optimized during the CASSCF) I get:

Nuclear energy:                      98.04037687
Core energy:                      -1409.15674314
Zeroth-order valence energy:        -68.87794571
Zeroth-order total energy:        -1379.99431197
First-order energy:                -122.98962991

Nuclear energy:                      98.04037687
Core energy:                      -1476.23082148
Zeroth-order valence energy:        -24.00693777
Zeroth-order total energy:        -1402.19738238
First-order energy:                -100.60555506

In the second job, about 67 hartree has been moved from the core, but only for the lower state.  I spent a couple of hours thrashing keywords but have had no luck understanding or solving this problem.  I did try following the CASSCF by another CASSCF with the optimized orbitals "frozen", but that made no difference.  I tried setting the PRCORE array but couldn't obtain any effect.

Grateful for any help or insight,
Karl


-------------------------------------------------------------------
Dr. Karl K. Irikura
National Institute of Standards and Technology
100 Bureau Drive, mail stop 8320
Gaithersburg, MD 20899-8320
voice:  301-975-2510   fax:  301-975-3670
email:  karl.irikura at nist.gov<mailto:karl.irikura at nist.gov>
http://www.nist.gov/mml/csd/informatics_research/index.cfm
-------------------------------------------------------------------

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.molpro.net/pipermail/molpro-user/attachments/20131218/98c040a7/attachment.html>