Author Topic: Abnormal termination of egrad  (Read 6820 times)


  • Newbie
  • *
  • Posts: 1
  • Karma: +0/-0
Abnormal termination of egrad
« on: July 26, 2016, 01:32:59 PM »
Hi all,

I have encountered some problems using egrad.

Short background: I attempt to propagate a nuclear wave packet on the potential energy surface (PES) of the first excited electronic state, S1, of CH2O using a semiclassical approximation. Therefore, I have to run a bunch of classical trajectories on the excited state PES, and calculate the Hessian (by calling NumForce) along these trajectories. The Hessian is used in the equation-of-motion for the so-called monodromy matrix. Hence, instead of using frog to run the trajectories, I have written a few python scripts, which make the necessary calls to the TURBOMOLE executable modules. For the initial tests of the trajectory calculations, I have turned off the Hessian calculation.

The TURBOMOLE (V6.0, 2009) modules are called with the following setup of the control file:
  • C1 symmetry, and Cartesian coordinates
  • def2-TZVP basis set
  • SCF convergence criterion is 10-9
  • TD-DFT in conjunction with the RI approximation
  • BP86 (default) functional and m5 grid size
  • rpas is turned on with the irrep option "a 3"
  • The keyword "$exopt 1" is added
The trajectories are represented in a 12-dimensional phase space corresponding to the normal modes of S1. At each time step, the normal mode coordinates are transformed to Cartesian coordinates. Then, the ridft, and subsequently, the egrad modules are called. The Cartesian coordinate representation of the gradient is transformed to the normal mode representation, and Hamilton's equations-of-motion are integrated one time step. This procedure works fine as long as the vibrational energy of the trajectory is rather small, in this case less than ~7000 cm-1 (the energy of the vibrational ground state of S1 is Eg ~ 5000 cm-1). If the energy is larger (not more than 2*Eg), some trajectories are terminated before reaching the final time (500 fs in these calculations). At the configuration of termination, the ridft ends normally, and I get one of two error messages from egrad:
  • abnormal termination. These are the last few lines of the output written by egrad:
    232                         -------------------
    233                          excitation vector
    234                         -------------------
    237  dimension of super-tensorspace: 1
    239           IRREP   tensor space dimension   number of roots
    241            a               528                   3
    243  Aufbau principle violation encountered. Check occupancies!

    I don't know how to solve this problem during the trajectory calculation.
  • missing or faulty SCF orbitals. These are the last few lines of the output written by egrad:
    210   MOs are in ASCII format !
    213  reading orbital data $scfmo  from file mos .
    215  orbital characterization : scfdump=100
    218  self consistent orbitals required

    Here I have used "$scfiterlimit 100". The same problem occurs with "$scfiterlimit 200", where the output reads "scfdump=200". Should I relax the SCF convergence criterion? I have chosen 10-9 to get a well conserved total energy along the trajectories.
In order to understand these problems, I have run similar trajectory calculations in the ground electronic state, S0, of CH2O. Here, the gradient (rdgrad), and Hessian (aoforce), calculations along the trajectories can be carried out even for "large" energies (~7*Eg).



  • Global Moderator
  • Hero Member
  • *****
  • Posts: 568
  • Karma: +0/-0
Re: Abnormal termination of egrad
« Reply #1 on: August 09, 2016, 08:51:54 PM »
Hi Jakob,

point 1: Aufbau principle violation encountered. Check occupancies!

the ground state calculation probably produced a hole in the occupation. Call 'eiger' on the command line to check that. If you have a small or very small HOMO-LUMO gap and/or if the first excited state energy gets close to the ground state energy you will have to check if you ground state is stable or not.

point 2: the ground state energy calculation with ridft did not converge.

orbital characterization : scfdump=100

the fist line of the mos file (or the alpa file) contains the status of the orbitals. scfconv=<convergence_criteria> means a converged wave function. scfdump=<iteration-number> means a non-converged energy calculation after <iteration-number> number of SCF iterations.

Good luck,