Molcas Forum

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 86

[SOLVED] MCPDFT questions

Dear Users,

I am trying to run some MCPDFT calculations. What I want is MC-PDFT energies for each RASSCF root.

My test system is a Be atom, for which I calculate Be2+ SCF, and then do RASSCF with 2 electrons on 4 orbitals (Be 2s 2p).

The next step is CIONLY RASSCF/MC-PDFT run, followed by a separate MCPDFT run.
I assume I shoud run just one of them for the "real" calculation.
However, I have questions to both, and they produce the same questionable situations when ran separately (i.e. SEWARD>SCF>RASSCF>RASSCF/MC-PDFT and SEWARD>SCF>RASSCF>MCPDFT).
Thus, here is a whole-package test intput (it only takes seconds to run).

&SEWARD
Title
 Be atom
 
RELA 
   R02O02 

BSShow 
ECPShow 
Expert 

Basis Set
Be.ANO-rcc.Veryazov.14s9p5d2f1g.4s2p2d
   Be   0  0  0   Angstrom  
End Of Basis 

&SCF
Title 
Be2+

Occupied
 1



&RASSCF

Spin
 1
nActEl 
  2
Inactive
 1
RAS2
 4
CIRoot
 10 10 1

&RASSCF

Spin
 1
nActEl 
  2
Inactive
 1
RAS2
 4
CIRoot
 10 10 1
JOBIPH
CIRESTART
CIONLY
KSDFT
ROKS
TLSDA


&MCPDFT
KSDFT
TLSDA

The questions:
1. RASSCF/MC-PDFT produces only one energy, while I am clearly asking for ten roots. Do I need to run MS-PDFT?

2. From the value of that single energy, it is a state-average RASSCF energy, similar but not exactly equal to the first energy printed in the convergence section of the RASSCF run.
I feel like something does not add up.

3. MCPDFT prints energies root-by-root, which is what I want.
But, the DFT energies are higher than the RASSCF energies - which should not happen, as DFT adds the on-top energy.
Is it normal?

4. The bigger issue IMHO is integrated density part which for root 1 looks like this:

      Integrated total density:                                                  4.000
      Integrated alpha density before functional transformation:                 2.000
      Integrated  beta density before functional transformation:                 2.000
      Integrated alpha density  after functional transformation:                 2.509    >> BIG HM
      Integrated  beta density  after functional transformation:                 1.491    >> BIG HM

The spin is set to 1 (spin quantum number 0), meaning that alpha and beta densities should be similar. They are not, and the discrepancy increases with the root number.
Am I doint something wrong, or is the program doing something sketchy?

Thank you.
Andrew

EDIT:

I've checked another thing. I've removed the RASSCF/MC-PDFT part and did XMS-PDFT.
The input:

&SEWARD
Title
 Be atom
 
RELA 
   R02O02 

BSShow 
ECPShow 
Expert 

Basis Set
Be.ANO-rcc.Veryazov.14s9p5d2f1g.4s2p2d
   Be   0  0  0   Angstrom  
End Of Basis 

&SCF
Title 
Be2+

Occupied
 1



&RASSCF

Spin
 1
nActEl 
  2
Inactive
 1
RAS2
 4
CIRoot
 10 10 1
XMSInter



&MCPDFT
MSPDFT
KSDFT
TLSDA

The alpha-beta density differences are now much smaller, while the  XMS-PDFT energies are similar to RASSCF energies.
Still, sometimes the XMS-PDFT energy is larger than the respective RASSCF energy. I will try a larger basis.

Root    RASSCF        XMS-PDFT
 1  -14.57521642   -14.56296971
 2  -14.36824658   -14.38665458
 3  -14.36824658   -14.38665288
 4  -14.36824658   -14.38664255
 5  -14.29659983   -14.27069342
 6  -14.29659985   -14.27065410
 7  -14.27830324   -14.26422791
 8  -14.27843144   -14.26418384
 9  -14.29659972   -14.26413408
10  -14.27826343   -14.17180207

EDIT 2:
Larger basis did not change much. But, when I use GGA, the XMS-PDFT are lower than RASSCF energies. That answers question 3.

Last edited by andrewshyichuk (2021-11-30 20:21:32)

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 86

Re: [SOLVED] MCPDFT questions

I've contacted the Minesota group (prof. Laura Gagliardi) directly.
Questions 3 and 4 regard normal properties of MC-PDFT calculations. I thus mark the topic as closed.

qq270814845

Member

Registered: 2021-11-30

Posts: 1

Re: [SOLVED] MCPDFT questions

Regarding Question 3,

Note that MC-PDFT does not imply add the on-top energy to the CASSCF energy. MC-PDFT throws away the non-classical two-electron energy in CASSCF energies and then adds the on-top energy. So an MC-PDFT energy could be higher than the CASSCF energy.

We do not have much experience on this phenomenon, but it looks like using LSDA might lead to a higher MC-PDFT energy according to andrewshyichuk's EDIT 2.