Molcas Forum

hbschlegel

Member

Registered: 2025-08-16

Posts: 6

Bug report MC-PDFT frequencies

The following produces MC-PDFT frequencies.

&GATEWAY
  Coord = 3

    O      0.00  0.00  0.00
    H      0.56  0.75  0.00
    H      0.56 -0.75  0.00
  Basis = cc-pVDZ
  Group = NoSym
  RICD

>>> DO WHILE

  &SEWARD

  &RASSCF
    NActEl = 8 2 2
    Charge = 0
    RAS1   = 0
    RAS2   = 6
    RAS3   = 0
    StAverage = 4
    RlxRoot = 1

  &MCPDFT
    Functional = t:PBE
    Grad

  &SLAPAF

>>> END DO

&MCKINLEY

The following produces NAN's in the frequency calculations.

&GATEWAY
  Coord = 3

    O      0.00  0.00  0.00
    H      0.56  0.75  0.00
    H      0.56 -0.75  0.00
  Basis = cc-pVDZ
  Group = NoSym
  RICD

>>> DO WHILE

  &SEWARD

  &RASSCF
    NActEl = 8 2 2
    Charge = 0
    RAS1   = 4
    RAS2   = 0
    RAS3   = 2
    StAverage = 4
    RlxRoot = 1

  &MCPDFT
    Functional = t:PBE
    Grad

  &SLAPAF

>>> END DO

&MCKINLEY

The only change is in RAS1, RAS2, RAS3.
Is there something wrong with my input of is this due to a software bug?
I have a variety of other attempts to calculate MC-PDFT frequencies - some produced NAN's and some worked properly.
(see previous post chain for further info)

Berny Schlegel

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,141

Re: Bug report MC-PDFT frequencies

The problem is not in the frequency calculation, but in the gradient directly. Analytical RASSCF-MCPDFT gradients are probably not implemented, but not detected either. I'll disable them (which will fall back to numerical gradients).

In the meantime, you can force numerical gradients by adding "Numerical" in the GATEWAY input (you can then remove "Grad" from MCPDFT).

hbschlegel

Member

Registered: 2025-08-16

Posts: 6

Re: Bug report MC-PDFT frequencies

For RAS1=0, RAS2=6, RAS3=0, numerical and analytical optimizations and frequency calculations give the same results to within the precision expected.

For RAS1=4, RAS2=0, RAS3=2, numerical optimizations and frequency calculations succeed, but analytical fails with NANs.

For RAS1=2, RAS2=2, RAS3=2, numerical optimizations and frequency calculations succeed, but analytical fails with NANs.

Since analytical gradients work for RAS1=0, RAS2=6, RAS3=0, MC-PDFT gradients appear to be implemented correctly for CAS.
The failure for RAS1=4, RAS2=0, RAS3=2 and RAS1=2, RAS2=2, RAS3=2 points to a problem with analytical gradients for RAS1, RAS3 non-zero.

A warning should be provided that MC-PDFT gradients work only for CAS but not for general RASSCF.

Hopefully MC-PDFT gradients can be implemented properly for general RASSCF in the near future.