BOMD of two cationic fragments (coulomb explosion)

Questions
#1

Dear all,

I am trying to do a rough simulation of the dynamics of a Coulomb explosion of a dimer made by an argon atom and a nitrogen molecule. We start from eq. geometry and put a positive charge on each unit, so overall it’s a dication, with the one hole in the argon, one in the nitrogen molecule. The charges repel and fly away from each other.

I am trying to simulate the dynamics but I keep having convergence issues, in particular when the units start to drift away.
This is my input:

$molecule
2 3
Ar  -0.00495633   0.00000000   1.59505954
N    0.56436532   0.00000000  -2.10450086
N   -0.54048683  -0.00000000  -1.94386377
$end

$velocity
  -5.73850449e-06    0.00000000e+00   -2.61252976e-05
  -7.05048896e-04    0.00000000e+00   -3.00609711e-05
   7.32695734e-04    0.00000000e+00    1.55926842e-04
$end

$rem
JOBTYPE        AIMD
EXCHANGE       wB97x-D
BASIS          def2-TZVP
TIME_STEP      5         
AIMD_STEPS     1500    
AIMD_METHOD    BOMD  
SYMMETRY         FALSE    
SYM_IGNORE       TRUE
UNRESTRICTED    true
SCF_ALGORITHM   DIIS_GDM
SCF_MAX_CYCLES  250
SCF_CONVERGENCE   6
$end

The geometry and velocities come from a Wigner distribution of the vibrational ground state.
Is there a way to perform this kind of dynamics in QChem?
How to get rid of the convergence error?

I tried also

SCF_ALGORITHM   RCA_DIIS
SCF_CONVERGENCE   6    
FOCK_EXTRAP_ORDER   6

But it crashes even faster.

The error is always:

gen_scfman_exception: SCF failed to converge

 Q-Chem fatal error occurred in module libgscf/gen_scfman/gen_scfman_main.C, line 246:

 Error in gen_scfman


 Please submit a crash report at q-chem.com/reporter

Thank you for the help

#2

For very high-energy BOMD, you may need a time step shorter than 1 fs as the energy scales get high and this can cause the Verlet algorithm to take a too-large step, putting you into a hard-to-converge region of configuration space.

That said, as fragments drift apart (or just for large atomic separations generally), one often has a difficulty SCF convergence problem because there can be multiple low-lying SCF solutions. The only “trick” is to vary the convergence algorithm (and definitely increase MAX_SCF_CYCLES). In recent versions of Q-Chem (since 6.3, I think), you can use SCF_ALGORITHM = ROBUST, which will cycle through several different options in case of convergence problems.

Finally, perhaps try turning off the Fock matrix extrapolation and using the guess from the previous time step.