Molcas Forum

sdong

Member

Registered: 2018-05-08

Posts: 2

Bug in &CASPT2 property calculation when "DENSity" is requested?

I am trying to understand how the dipole and quadrupole moments are calculated in the &CASPT2 module, but I found weird results when I include the "DENSity" keyword.

I am using trans-butadiene as a test case, and I am using Molcas 8.2.

I did a Hartree-Fock calculation in Gaussian and a CASSCF(2,1) (2 electrons in 1 orbital) calculation in Molcas (equivalent to HF), and I got the same total energy (2E-8 eV difference) and the same quadrupole moments. This says Gaussian and Molcas are using the same definition for quadrupole moments.

Then I did an MP2 calculation in Gaussian and a CASPT2 calculation using the CASSCF(2,1) reference wavefunction (should be equivalent to MP2) in Molcas. The Gaussian MP2 energy and Molcas CASPT2 energy are basically the same (3E-6 eV difference), but the quadrupole moments are different.

If I include the DENSity keyword, which is supposed to generate the full density matrix instead of the approximate density matrix, the quadrupole moment differs even more from the Gaussian quadrupole moment (in particular, the XY component has changed by more than 6 Debye*Ang). Also, if I include the DENSity keyword, the total charge of butadiene becomes -1.3330, which sounds very unreasonable.

I suspect there might be a bug related to requesting the DENSity keyword in the &CASPT2 module. If not, how should I explain this unreasonable total charge and quadrupole moment? Any comment/suggestion is appreciated.

Thank you!

-----
Summary of the results:
RHF absolute E (hartree)    :
Gaussian: -154.934328546   
Molcas: -154.9343285467

RHF Quadrupole moment (Debye*Ang):
        Gaussian  Molcas           
XX    -23.0352    -23.0352       
XY    -0.2103    -0.2103       
YY    -23.0574    -23.0573       
ZZ    -29.0810    -29.0810       
               

MP2 absolute E (hartree):
Gaussian: -155.46879602867
Molcas: -155.46879591

MP2 Quadrupole moment (Debye*Ang):
       Gaussian   Molcas (no DENS)  Molcas (with DENS)
XX    -23.4873           -23.2206             -36.2781
XY      -0.1126             -0.2045            -6.4043
YY    -23.4402           -23.1694          -29.0558
ZZ    -28.6492           -29.0859          -30.0200
charge 0.0000              0.0000            -1.3330
-----

The Molcas input is as follows:

&GATEWAY
coord=butadiene.xyz
Basis Set
H.cc-pvdz, C.cc-pvdz
group=nosym

&SEWARD
NODEleted
grid input
grid=ultrafine
end of grid input
End of input

 &RASSCF &END
CIROOT= 1 1; 1
nActEl
 2 0 0
INACtive
14
FROZen
0
RAS2
1
DELEted
0
Symmetry
1
Spin
1
CHARge
0

&CASPT2
MaxIter= 1000
DENSity
PROP

The coordinate (in angstrom) in xyz format is:

10

C 1.7414427 0.6171859 0.000000
C -1.7414427 -0.6171859 0.000000
C 0.3998134 0.6099967 0.000000
C -0.3998134 -0.6099967 0.000000
H 0.1443398 -1.5501826 0.000000
H -0.1443398 1.5501826 0.000000
H 2.302001 1.5422423 0.000000
H -2.302001 -1.5422423 0.000000
H 2.3041193 -0.3093168 0.000000
H -2.3041193 0.3093168 0.000000

Last edited by sdong (2018-05-09 00:34:10)

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Bug in &CASPT2 property calculation when "DENSity" is requested?

I don't know if it's written somewhere or if it's true, but I seem to recall reading or hearing somewhere that one shouldn't use CASPT2 in Molcas with a single-configuration space, i.e., with a HF reference as is your case.

If you make it a real CASSCF/PT2 calculation (INAC=13; RAS2=4; NACT=4 0 0), the charge comes out as 0.0 and the quadrupole is much more similar to the MP2 result (XX=-23.451, XY=-0.18408, YY=-23.219, ZZ=-28.598).

sdong

Member

Registered: 2018-05-08

Posts: 2

Re: Bug in &CASPT2 property calculation when "DENSity" is requested?

I don't know if it's written somewhere or if it's true, but I seem to recall reading or hearing somewhere that one shouldn't use CASPT2 in Molcas with a single-configuration space, i.e., with a HF reference as is your case.

Thanks! That is good to know!

The goal of my test is to know whether the &CASPT2 module uses the first-order wave function when it computes properties. Since this test is not "allowed" in Molcas, could someone confirm that the property calculation in the &CASPT2 module indeed uses the first-order wave function (when DENSity is used)?

When I was trying to confirm the quadrupole moment values using the FFPT method, I got weird results again. (This is not related to the DENSity keyword, so I should probably post a new topic. But it relates to my objective to see how CASPT2 properties should be calculated, and uses the same test case.)

What I did was, in the code I posted previously, I included

&FFPT
QUAD
ZZ 0.00001

right after the &SEWARD module. I got E(CASSCF; ZZ=0.00001)=-154.93437344 hartree. I used the formula -3*[E(ZZ=0.00001) - E(ZZ=0.0)]/0.00001 to calculate the CASSCF (in fact, RHF) quadrupole moment in the ZZ direction, and got -3*[-154.93437344-(-154.9343285467)]/0.00001 = 13.47. I am not quite sure about the unit. If it is in atomic unit or Debye*Angstrom, the value will be far from the RHF results from Molcas or Gaussian I posted previously. I have also tried a state-specific CASSCF(4,4) calculation, and got 11.76. What unit conversion am I supposed to be doing here? Or have I done something more seriously wrong?

Thanks a lot for your help!

Last edited by sdong (2018-05-10 09:21:55)