Molcas Forum

lijingbai2009

Member

Registered: 2019-08-28

Posts: 29

[SOLVED] Orbital rotation during numerical Hessian calculation

Hi,

I am calculating the frequency for my SA-CASSCF optimized structure. However, I found sometimes the orbital rotated out of the active space (like a pi-orbital exchanged with an outside sigma-orbital) because of the atom displacement made by the numerical Hessian procedure. I think it could produce a wrong frequency results, right? But if I use "supsym" to constrain the active space, does it make sense?

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: [SOLVED] Orbital rotation during numerical Hessian calculation

Yes, that kind of change during the numerical differentiation makes the surface potentially discontinuous and the result could be garbage. If you constrain the active space, you'd be getting the derivatives of a constrained wavefunction. Whether or not that makes sense, depends on the particular case, I guess. An alternative could be reducing the "delta" value, such that the displacements are smaller and the active space hopefully more stable1

lijingbai2009

Member

Registered: 2019-08-28

Posts: 29

Re: [SOLVED] Orbital rotation during numerical Hessian calculation

Hi Ignacio,

Thanks for your quick response! The manual said the delta in ALASKA is used for numerical gradient only, while the delta in SLAPAF doesn't say anything about it. Where should I put the delta in my input so I can still use the analytical gradient but just change the step size for the numerical hessian calculation? I assume the MCKINLEY is still the last one in the input, right?

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: [SOLVED] Orbital rotation during numerical Hessian calculation

When MCKINLEY decides that it has to do numerical differentiation, it just calls SLAPAF to do it. If you want to change the default delta, you'd have to call SLAPAF yourself (and remove MCKINLEY):

&SLAPAF
  Numerical
  Iterations = 0
  Delta = 0.005

Note that you'd have to put this in a DoWhile loop, as with geometry optimization. Gradients will be analytical if available.

lijingbai2009

Member

Registered: 2019-08-28

Posts: 29

Re: [SOLVED] Orbital rotation during numerical Hessian calculation

I have tried many smaller "delta" values but unfortunately, none of them worked. It turned out the problem is from my active space, which is not stable and changed during the numerical differentiation. After I changed the active space, now it is stable and the frequency job worked fine.