Why is my ORCA calculation of nitrogen iodide failing?

Question

I am using ORCA version 4.2.1 to calculate on the RI-BP86/def2-SVP level of theory the molecule $\ce{NI3}$. I would assume this to be a rather trivial matter, but I am running into an error. I've only been using ORCA casually, so I'm not that experienced. I also know that version 5.0 was just released, but I did not have the time to update.
My guess is I'm doing something wrong, or missing something important. Here is the input file:

! RI BP86 def2-SVP def2/J D3BJ TIGHTSCF Opt TIGHTOPT Grid5 FinalGrid5 Freq
! LargePrint PrintBasis
%geom
Calc_Hess true # Calculate Hessian in the beginning
Recalc_Hess 5  # Recalculate the Hessian every 5 steps
end
%maxcore 1200

*xyz 0 1
N        -0.00447226283705   -0.00447228533261   -0.00447218389615
I         1.96253257409287   -0.47903013984302   -0.47903017483099
I        -0.47903014119403    1.96253259549055   -0.47903017643514
I        -0.47903017006179   -0.47903017031492    1.96253253516228
*

The molecular structure was preoptimised with GFN2-xTB, so it should be a decent guess.

At the tail of things, within the initial SCF cycle, I'll get the error:

                         ----------------------------
                         !        ITERATION    27   !
                         ----------------------------
   Total Energy        :    -947.817767792740 Eh
   Energy Change       :      -0.082996473216 Eh
   MAX-DP              :       0.392236224899
   RMS-DP              :       0.018540257659
   Orbital gradient    :       0.093887611787
   Orbital Rotation    :       0.570102706559
   Actual Damping      :       0.0000
   Int. Num. El.       :    82.00000308 (UP=   41.00000154 DN=   41.00000154)
   Exchange            :   -70.09754201
   Correlation         :    -3.88915080


          ************************************************************
          *                     SERIOUS PROBLEM IN SOSCF             *
          *                     ------------------------             *
          *            HUGE, UNRELIABLE STEP WAS ABOUT TO BE TAKEN   *
          *                         ABORTING THE RUN                 *
          ************************************************************

What is the reason for this and how can I get it to converge?

Answer 1

There was a bug with the PModel guess of iodine, which leads to a very poor initial guess for molecules containing iodine, and consequently divergent SCF calculation. This bug has been fixed in ORCA 5.0. ORCA 5.0 can be downloaded at https://orcaforum.kofo.mpg.de/app.php/dlext/?cat=13. Note that ORCA 5.0 does not support the keywords Grid5 and FinalGrid5 anymore (because we now use a new set of grids, and the new default grid is much more accurate than the old one, with about the same cost), you have to remove them before you run the input file with ORCA 5.0.

If updating to ORCA 5.0 isn't an option, you can use the PAtom guess instead. By the way, in case you encounter a similar error message in the future that is not due to a bug of the initial guess, you can probably get rid of the error by adding !NoSOSCF.

Finally, I'd say that the ORCA forum (https://orcaforum.kofo.mpg.de/index.php) is probably a better place to ask questions about ORCA.