TURBOMOLE Users Forum

TURBOMOLE Modules => Escf and Egrad => Topic started by: Lohengrin on February 25, 2007, 08:19:22 PM

Title: many problems encoutered while I used TDDFT EGRAD for optimization...
Post by: Lohengrin on February 25, 2007, 08:19:22 PM
dear all,
i often meet two sorts of troubles, especially in triplet pi-pi* excited state geometry optimizations:

1. the optimization died without any message error. it usually stopped after several cycles at an unconverged gradient -- around 0.1~0.3. and it would always stop at the same position if i redid it from a different initial geometry. eg. the first triplet pi-pi* state of acetaldehyde (C1), formaldehyde (Cs), and acetone (Cs).

2. usually occurs in triplet states as well with the following message:
MODTRACE: no modules on stack
Reference state is non-real unstable
egrad ended abnormally
error in gradient step (1)

i've heard that increasing the number of roots calculated and config of FERMI etc might help. but these didn't work at all!
anyone knows the recipe? Many Many Thanks!!
Title: Re: many problems encoutered while I used TDDFT EGRAD for optimization...
Post by: dimi on February 26, 2007, 05:42:27 PM
Hello,

it is rather unusual that egrad just stops without further notice during optimization.
The most common problems that occur in excited state optimizations are instabilities of the ground state:

The first two indicate that another electronic state of lower energy exists at the given geometry, a singlet or a triplet state,
respectively. Non-real instabilities (see also the "singlet excited state geometry optimization" post) usually result from conical intersection with the ground state. In all cases it means that the ground state description,e.g., restricted Kohn-Sham, is no longer adequate. In general, this plainly means that TDDFT cannot be applied for the given system at the given geometry.

If an instability occurs in an excited state optimization, you should check several things before giving up:

There are certainly more possibilities but these might help already.

Greetings,
Dmitrij