Molcas Forum

2877321934

Member

Registered: 2025-10-02

Posts: 4

How to converge a geometry optimization very close to an S0-S1 conical

Hello everyone,

I am studying a photoisomerization process using the SA-CASSCF multi-state method in OpenMolcas. The process involves scanning a dihedral angle, so I have set up a series of constrained optimizations from 0 to 180 degrees.

I've encountered a convergence issue when the dihedral angle reaches 90 degrees. The geometry optimization fails to converge and the structure oscillates, which I believe is due to its close proximity to the S0-S1 conical intersection (I located the optimized CI at a dihedral angle of 93.4 degrees).

My current goal is simply to get this constrained geometry optimization to converge successfully. I have already tried using a very small step size, and I've also tried to manually tweak the initial geometry many times, but neither approach has worked.

I would like to ask the community for advice:

Are there any specific keywords or settings in OpenMolcas that can facilitate convergence in such a difficult case, close to a conical intersection? Or is my only option to keep changing the initial guess and hope for a bit of luck?

My supervisor suggested that I calculate an accurate numerical Hessian once, and then read this Hessian in for the subsequent (constrained) geometry optimization. He mentioned this is often helpful for convergence and is analogous to using opt=calcfc or opt=rcfc in Gaussian. However, I find the description of how to handle Hessians in the OpenMolcas manual difficult to understand.

Could anyone provide some guidance on this Hessian approach or suggest an alternative strategy? Any help or suggestions would be greatly appreciated.

Thank you.

Here is the input file for the failing optimization:

&GATEWAY
 coord
scan_100.xyz
 
 basis
 6-31g*

 constraints
  d = Dihedral C6 C10 C11 C28
  Values
  d = 100 degree
 end of constraints	

 group
 nosym

 ricd

END OF INPUT

>>>>>>>>>>>>> DO WHILE <<<<<<<<<<<<<
&SEWARD
END OF INPUT
>>>>>>>>>>>>> IF ( ITER = 1 ) <<<<<<
&SCF
END OF INPUT
>>>>>>>>>>>>> ENDIF <<<<<<<<<<<<<<<<
&RASSCF
 title
 cas108

 nactel
 10 0 0

 inactive
 79

 ras2
 8

 symmetry
 1

 spin
 1

 ciroot
 4 4 
 1 2 3 4
 1 1 1 1

 rlxroot
 2
 
 lumorb

END OF INPUT

&MCLR
 iterations
 800
END OF INPUT

&ALASKA
END OF INPUT

&SLAPAF; Iterations=30; MaxStep=1.0

 cartesian
END OF INPUT
>>>>>>>>>>>>> ENDDO <<<<<<<<<<<<<<<<

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,165

Re: How to converge a geometry optimization very close to an S0-S1 conical

If there is CI nearby, there may not be a minimum on S1, i.e. the minimum may be the CI, which is not a stationary point (the adiabatic gradient is undefined).

Try locating a CI with the dihedral constraint. Then analyze the CI characterization (see https://doi.org/10.1021/acs.jctc.6b00384), and if it's peaked you're probably in such a case, and you can take this CI as the optimized constrained minimum. If it's sloped, try distorting the molecule slightly along the corresponding direction (without changing the torsion) and use that as a starting point for the optimization.

2877321934

Member

Registered: 2025-10-02

Posts: 4

Re: How to converge a geometry optimization very close to an S0-S1 conical

Thank you for your detailed and fast reply.

I apologize if my description in the original post was not clear enough. I have already located the conical intersection (CI) for the isomerization process; the dihedral angle of the optimized CI structure is 93.4 degrees.

The reason I need this specific constrained optimization is that for my paper, I need to show the complete isomerization path (a potential energy scan from 0 to 180 degrees, with a 10-degree step size). To do this, I am performing a series of constrained optimizations.

My problem is that when I try to run the optimization with the dihedral angle constrained to 90 degrees, it fails to converge, which I presume is due to its proximity to the CI.

So far, I have only tried reducing MaxStep and changing LevShift, but neither has worked.

I would like to ask if there are any other methods worth trying. (As I mentioned, I am thinking about using an accurate analytical Hessian, but I'm not sure how to do that).

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,165

Re: How to converge a geometry optimization very close to an S0-S1 conical

You've found the unconstrained CI, I suggested optimizing a constrained CI. Most likely you don't find the minimum because there's no such (stationary) minimum.

2877321934

Member

Registered: 2025-10-02

Posts: 4

Re: How to converge a geometry optimization very close to an S0-S1 conical

Hello Ignacio,

Thank you for your detailed reply, and my sincere apologies for the long delay in my response.

I was suddenly pulled into a very important collaborative project, which forced me to put this work on hold for a while. I apologize that I forgot to reply to your detailed guidance earlier, as I was overwhelmed with that project.

I have now tried your strategy and successfully located the constrained conical intersections (CIs) with the dihedral angle constrained to 90 and 100 degrees.

You mentioned needing to determine the type of CI. Could you please tell me how I can determine from the relevant output files whether these CIs are "peaked" or "sloped"?

Again, my apologies for the oversight and for being too busy to reply sooner. Thank you very much for your help!

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,165

Re: How to converge a geometry optimization very close to an S0-S1 conical

When you optimize a CI, the type should be in the log file (under "CI characterization"). Note that if it's a constrained optimization, the type may not be as meaningful (see e.g. https://doi.org/10.1039/D1CP05028A)

2877321934

Member

Registered: 2025-10-02

Posts: 4

Re: How to converge a geometry optimization very close to an S0-S1 conical

Understood, thank you for the explanation and for linking the paper. I appreciate it.