Molcas Forum

jotoldo

Member

Registered: 2018-04-12

Posts: 4

Convergence problems in S1 and CI geometry optimizations

Dear all,

I am trying to optimize the geometries for a set of molecules (same atoms with different substituent positions) in the S1 and conical intersection using CASPT2. It seems the hypersurface is very flat and S1 is close to the CI.  I have tried different active spaces and state average calculations, but for most of them I have a non-convergent profile for the geometry, i.e. the optimization energy keeps oscillating with very small changes in the energy. I have tried to decrease the Cutoff in Alaska (Cutoff=1.0d-06) and set "maxstep=0.2" in Slapaf, but I still have convergence problems. Is there any suggestion about how to overcome this problem?
My active space contains 6 electrons and 7 orbitals and the molecule has a positive charge. These are my input files:

For CI optimization:

&GATEWAY
 coord=$CurrDir/OPT_S1.Opt.xyz
 basis=ANO-RCC-VDZP
 group=NoSymm
 ricd

Constraints
  a  =  Ediff 1 2
  Value
  a  =  0.000
End  of  Constraints

>>>EXPORT MOLCAS_MAXITER=300

>>COPY $CurrDir/OPT_S1.RasOrb $Project.RasOrb

>>>  Do  while

&SEWARD

&RASSCF
  FileOrb = $Project.RasOrb
  Spin=  1
  Charge = 1
  CiRoot=  2 2 1

&CASPT2
 Multistate = 2 1 2
 Imagin = 0.1

&SLAPAF
 Maxstep = 0.2

>>>  EndDo

And for S1:

&GATEWAY
  Coord = OPT_S1.Opt.xyz
  Basis = ANO-RCC-VDZP
  Group = Nosymm
  RICD

>>> COPY $CurrDir/OPT_S1.RasOrb  $Project.RasOrb

>>> Export MOLCAS_MAXITER=100

>>> Do While
&SEWARD
DoAnalytical

&RASSCF
  FileOrb=$Project.RasOrb
  CIRoot = 3 3 1
  RlxRoot = 2
  Charge = 1
  Levshift=1.0
  Iterations= 50 50

&CASPT2
  MultiState = 3  1 2 3
  ImaginaryShift = 0.1

&ALASKA
 Cutoff=1.0d-06

&SLAPAF
 Iterations= 50
 Maxstep = 0.2

>>> End Do

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Convergence problems in S1 and CI geometry optimizations

ALASKA cutoff and will do nothing for a CASPT2 optimization (numerical gradients). I don't have a solution, but just a confirmation that optimizations with numerical gradients close to state crossings are expected to be difficult, you may have to settle with nonconverged geometries.

jotoldo

Member

Registered: 2018-04-12

Posts: 4

Re: Convergence problems in S1 and CI geometry optimizations

Dear Ignacio,

Thank you very much for your reply. Do you recommend adjust the Threshold in &SLAPAF setting, for example, the values of 0.01 and 3.0D-3?
And what do you think it is better: trying to converge using another basis set, such as 6-31G(d), or keep the ANO-RCC-VDZP and settle with nonconverged geometries?

Best,

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Convergence problems in S1 and CI geometry optimizations

I'd rather change the thresholds. With high enough values you can turn nonconverged geometries into converged But check carefully the behaviour of the geometry and energy (and possible state crossings) of your optimization to get a picture of what's happening and decide accordingly.

jotoldo

Member

Registered: 2018-04-12

Posts: 4

Re: Convergence problems in S1 and CI geometry optimizations

Dear Ignacio,

Thank you again!  I have changed the thresholds in SLAPAF module, let's see
I also read that there is a possibility to change for Baker convergence criterion, but I think it is not so used, at least, I don't remember seeing this on papers. Do you think changing to Baker criteria (and maybe change the thresholds) can help in this case?  Or does not make sense to change it?

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Convergence problems in S1 and CI geometry optimizations

Convergence criteria are just different combinations of how small the gradient, displacement and energy difference must be. Changing them is no different than running many iterations and picking up the geometry that looks (to you) close enough to the minimum. Or in other words, is a way for you to specify what you consider "close enough". Depending on the system and application it may have different meanings.

jotoldo

Member

Registered: 2018-04-12

Posts: 4

Re: Convergence problems in S1 and CI geometry optimizations

Thank you very much for your reply, Ignacio! It helps me a lot!