Molcas Forum

angeloraymondrossi

Member

Registered: 2024-04-04

Posts: 19

Problem with converging Conical Intersection

Here is my input file:

& GATEWAY
coord=ethylene.xyz
basis= 6-31G*
group=C1

/*
basis = STO-3G
basis = 3-21G
basis= 6-31G*
basis= def2-TZVP
basis= aug-cc-pVTZ
*/

Constraints
   a = Ediff
  Value
   a = 0.000
End of Constraints

>>> EXPORT MOLCAS_MAXITER=300

>>> Do while

&SEWARD


>>> If ( Iter = 1 ) <<<


&SCF

>>> EndIf <<<

&RASSCF
LumOrb
Title= ethylene
Spin= 1; nActEl= 6 0 0; Inactive= 5; Ras2= 4
CiRoot
4 4; 1 2 3 4; 1 1 1 1
RlxRoot = 1

&ALASKA
PNEW

>>COPY $WorkDir/$Project.RunFile $WorkDir/RUNFILE2

&RASSCF
Title= ethylene
LumOrb
Spin= 1; Nactel= 6 0 0; Inactive= 5; Ras2= 4
CiRoot
4 4; 1 2 3 4 ; 1 1 1 1

&ALASKA
PNEW

&SLAPAF

>>> EndDo

This input closely resembles that given in DOI: 10.1021/acs.jctc.6b00384; J. Chem. Theor. Comput. 2016, 12, 3636-3653.
except that the basis is smaller so that I could run it on my laptop.

This is what is happening with different modifications:

Final state energy(ies):
      ------------------------  
::    RASSCF root number  1 Total energy:    -77.64735195
::    RASSCF root number  2 Total energy:    -77.63812256
::    RASSCF root number  3 Total energy:    -77.62236512
::    RASSCF root number  4 Total energy:    -77.61031619

       +----------------------------------+----------------------------------+
       +    Cartesian Displacements       +    Gradient in internals         +
       +  Value      Threshold Converged? +  Value      Threshold Converged? +
 +-----+----------------------------------+----------------------------------+
 + RMS + 1.2152E-01  1.2000E-03     No    + 1.7558E-02  3.0000E-04     No    +
 +-----+----------------------------------+----------------------------------+
 + Max + 1.8190E-01  1.8000E-03     No    + 3.5232E-02  4.5000E-04     No    +
 +-----+----------------------------------+----------------------------------+

 No convergence after max iterations

.#######################.
.# Convergence problem #.
.#######################.

Timing: Wall=6361.68 User=8751.86 System=2470.12

Can you please help with this.  What am I doing wrong and how can I get this to converge?

Thanks so much for helping me with this.

Kind regards,

Angelo

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,129

Re: Problem with converging Conical Intersection

You're using the "old" version, with two runfiles and manually running ALASKA. That will use an approximate and less reliable method for optimization (with no NAC vector). Moreover, if you are starting with the "(a)" geometry from the article, note that it is actually not a minimum (see section 4.1), so you'll likely find convergence issues. Also, you're specifying RlxRoot=1 for the first ALASKA and nothing for the second, which means RlxRoot=4, so you're requesting an intersection between the 1st and 4th state... and therefore also 2nd and 3rd. I doubt that's what you want, and I doubt that exists. Try with:

&GATEWAY
  Coord = 6

    C  0.000000  0.000000  0.725000
    C  0.000000  0.000000 -0.725000
    H  0.943102  0.100000  1.269500
    H -0.939150 -0.099581  1.276363
    H  0.262249 -0.911409 -1.269500
    H  0.565287  0.946139 -0.569500
  Basis = 6-31G*
  Group = NoSym
  Constraints
     a = Ediff 1 2
    Value
     a = 0.000
  End of Constraints

> do while

&SEWARD

&RASSCF
  Spin   = 1
  Charge = 0
  nActEl = 6
  RAS2   = 4
  StAver = 4

&SLAPAF

> end do

angeloraymondrossi

Member

Registered: 2024-04-04

Posts: 19

Re: Problem with converging Conical Intersection

Dear Ignacio,

Thank you for you help with this.  And I apologize for waiting so long before responding.

You are correct I should have specified RlxRoot = 1 for the first state and RlxRoot = 2 for the second state.  And I knew this too already.  For example, when using RASSCF in connection with &surfacehop, it is important to specify mdrlxroot = ???, otherwise RASSCF will choose the highest state available. 

Also, when I specified RlxRoot=2, then the "old" version does work OK.

Now, when I used the new method (a = Ediff 1 2 ... etc), that worked very well too.

But, there is one further point of discussion.  The initial geometry in your input example was very close to the converged Conical Intersection.  I took the initial geometry from the supplement of DOI: 10.1021/acs.jctc.6b00384; J. Chem. Theor. Comput. 2016, 12, 3636-3653, and that initial geometry converged very well too,

After several try with a planar C2H4 geometry,  there were big problems with CI convergence.  I finally tried a twisted ethylene structure with on CH2 group pyramidal.  It turns out that after several trials, that was similar to the ones in the paper.  DUH.  It took me a while to figure that out.

Don't you think it is antithetical to one of the underlying ideas of Computational Chemistry that a first principal calculation should be able to predict a CI structure from a normal geometry or C2H4.  The antithetical part is that I need to guess a possible structure close to one that I should be able to calculate.  If you look at the paper though, there are only a few molecules like C2H4 which really need the initial structure tickled so that it converges.

Thanks again so much for your help.

Kind regards,
Angelo