Molcas Forum

angeloraymondrossi

Member

Registered: 2024-04-04

Posts: 17

Unstable SURFACEHOP

Hello Everyone,

I am hoping that perhaps other Molcas users can provide hints on how to best use SURFACEHOP so that it DOES NOT stop in the middle of a calculation because RASSCF does not converge.  The molecule is small ... 9 atoms.  I have been working with SURFACEHOP for about 6 months, but this lack of convergence constantly occurs.

I have tweaked all the input parameters associated with DYNAMIX, i.e. DT and other variables.

Have other users experienced this problem? 

SURFACEHOP seems extremely sensitive to the molecule type. 

The irony is that there are so many other papers using SURFACEHOP that are bring published.  I want this too mommy!

But, thanks for your patience, and I hope that other users can help me.

Regards,

Angelo

P.S.  Using only DYNAMIX wiith the HOP variable turned on works, but the gold standard is the SURFACEHOP/DYNAMIX combination.

angeloraymondrossi

Member

Registered: 2024-04-04

Posts: 17

Re: Unstable SURFACEHOP

Here is an example with methanol.  It doesn't get any more succinct than this:

++ ---------   Input file   ---------

&GATEWAY

coord= meoh-opt-freq.xyz
basis= def2-TZVP
group=nosym

>>> EXPORT MOLCAS_MAXITER = 500

>>> DO WHILE

  &SEWARD

  >>> IF (ITER == 1)

    &SCF
     TITLE= MeOH
     OCCUPIED= 9
     ITERATIONS= 100

    &RASSCF
     Symmetry= 1
     Spin= 1
     nActEl= 8
     Inactive= 5
     RAS2=8
     CIroot= 3 3 1
     MDRLXR= 2

    >>> COPY $Project.JobIph $Project.JobOld

  >>> END IF

  &RASSCF
   JOBIPH
   CIRESTART
   Symmetry= 1
   Spin= 1
   nActEl= 8
   Inactive= 4
   RAS2=8
   LEVShft= 1.5
   CIroot= 3 3 1
   MDRLXR= 2

  &CASPT2
   XMULTISTATE= 3 1 2 3
   FROZEN= 4 0 0
   MaxIter= 50

  >>> COPY $Project.JobIph $Project.JobOld

  &SURFACEHOP
   TULLY
   DECOHERENCE = 0.1
   PSUB

  &ALASKA
  &DYNAMIX
   VELVer
   DT= 20.67
   VELO= 3
   THER= 0

>>> END DO

Here is a typical error:

                                         RASSCF iterations: Energy and convergence statistics
                                         ----------------------------------------------------

      Iter CI   SX   CI       RASSCF       Energy    max ROT     max BLB   max BLB  Level Ln srch  Step   QN   Walltime
          iter iter root      energy       change     param      element    value   shift minimum  type update hh:mm:ss
 reading initial CI vectors from JOBOLD
        No convergence in the CI section: MAXJT will be increased to                    24
 Nr of preliminary CI iterations:  11
        No convergence in the CI section: MAXJT will be increased to                    24
        1  11  100    0  -103.56020149    0.00E+00   1.18E-01*   3   7 1  4.71E-01*  6.37   0.00    SX    NO    0:00:01
--- Stop Module: rasscf at Sun Aug  4 15:01:58 2024 /rc=_RC_MEMORY_ERROR_ ---
*** files: meoh-surfacehop.RasOrb meoh-surfacehop.rasscf.h5 xmldump
    saved to directory /home/rossi/Downloads/workshop/OpenMolcas-projects/meoh-surfacehop
--- Module rasscf spent 4 seconds ---

>>> END DO

.#################################################.
.# Calculation terminated due to memory problems #.
.#################################################.

Can you help please.

Regards,

Angelo

angeloraymondrossi

Member

Registered: 2024-04-04

Posts: 17

Re: Unstable SURFACEHOP

One more comment.  The calculation eventually results in the dication of small molecules, e.g.  CH3-OH 2+ and other dications.  At first I felt this with a problem with the wavefunction density overlaps which drive the TULLY switches algorithm.  But the neutral molecules are also stopped by similar convergence issues.

I must be doing something incorrectly.

angeloraymondrossi

Member

Registered: 2024-04-04

Posts: 17

Re: Unstable SURFACEHOP

Hello Everyone,

There is recent 2023 article entitled Nonadiabatic Coupling in Trajectory Surface Hopping: How Approximations Impact Excited-State Reaction Dynamics by
Isabella C. D. Merritt, Denis Jacquemin, and Morgane Vacher* in  J. Chem. Theory Comput. 2023, 19, 1827−1842 that uses OpenMolcas to test various Surface Hopping methods and subsequently discusses their efficacy.

It's a well-balanced article.

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Unstable SURFACEHOP

If there are convergence issues in RASSCF, probably the active space is not adequate. It might be that the changes in geometry are too large (decrease the timestep in this case) or, more likely, that you need to reconsider the active space or state averaging scheme for the system changes.

angeloraymondrossi

Member

Registered: 2024-04-04

Posts: 17

Re: Unstable SURFACEHOP

Thanks so much for the advice. 

I tried some of the ideas.  But, you are correct, the active space might not be adequate, and I haven't modified the state-averaging scheme.  And you are correct:  this is project involving laser studies of small molecules which fragment after the laser pulse.  Most, but not all, surface hopping simulations involve vibrational/electronic spectra where the molecular structure does not radically distort as in my calculations where the molecule fragments into smaller moieties. 

Would you be able to provide a little more information on how one modifies the state-averaging?  For example, this is what I use CIroot= 3 3; 1 2 3; 1 1 1 where all states are averaged equally.  Full disclosure:  I have been hesitant to expand the active space because of increased simulation time.

Again, thanks for the advice.

Angelo

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Unstable SURFACEHOP

I mean simply changing the number of averaged states (I wouldn't recommend playing around with the weights). So, when a molecule breaks or distorts, you may need to include more or fewer states to get a... and then the challenge is finding a number that's good for the whole trajectory.