TURBOMOLE Users Forum
TURBOMOLE Modules => Escf and Egrad => Topic started by: administrador.ccc on February 09, 2007, 07:11:13 AM

I am performing a B3LYP first singlet excited state geometry
optimization, symmetry a, tzvp basis set. The work made some
optimization steps but it stops with error:
reference state is nonreal unstable
egrad ended abnormally
Taking the last coord file (or even the previous coord file) and
starting a job from scratch, I got the same error.

Hello,
the message "reference state is nonreal unstable" means that the ground state solution is not stable with respect to KS orbitals becoming complex. Usually, this is an indication for a conical intersection between the ground state and some excited state at the given geometry. Your can also see it from the excitation energy becoming really small (<0.1 eV or so). In that case there is basically no remedy to this problem since it just represents the shape of the potential energy surface.
Dmitrij

Does this mean the same thing?
Reference state is unstable
egrad ended abnormally

I guess so...
Does this mean the same thing?
Reference state is unstable
egrad ended abnormally

Hello,
I have a similar problem with the triplet T1 excited state energy optimization for uranyltriauqasulphate complex within the DFT/B3LYP/defSVP (Turbomole V6.5).
I started the calculation from the optimal geometry for the ground S0 state. But soon after the optimalization (26 steps in steepest descent algorithm with dqmax=0.1) I have get the GEO_OPT_FAILED result since the last point has a
" Reference state is unstable
egrad ended abnormally
"
error message on its end.
My supervisor (who doesn't supervise particularly this project, I am playing with uranyl more like on my own) told me, that it is obvious that the DFT converged to excited state, so I should swap MOs (or KS orbitals to be accurate).
But I am not sure if just reordering the MOs from the failed calculation (or from the last point of the optimization trajectory where it still worked) would work ... and which one of the virtuals I should take instead of the currect HOMO (or HOMO1, HOMO2, ?). Since the resulting occupation is 2222...2 00..0. Is it possible to do the swap before the DFTselfconsistent equations are being solved?
I have found elsewhere (https://groups.google.com/forum/#!topic/newtonx/G7DkofJR5YM) answer (for completely different molecule and in combination with the dynamics, but the problem might be the same) that
 Unrestricted KS
(and/or)  Fractional occupation (in particular, temperature smearing)
would work.
Since I am interested in temperature smearing of KS occupation numers also with the perspective to better model 298K temperature effects, I will try the last one.
I will try both the postDFT reordering