The ABINIT Users Mailing List ( CLOSED )

[Emmanuel Arras <emmanuel.arras@cea.fr>]

I may be wrong, but I'm guessing that the "long range" (10 bohr) 
attraction you are waiting for here (i.e. what happens in "real life") 
is due to the Van der Waals interaction, which is known not to be 
reproduced by local approximations of the exchange correlation energy 
(LDA/GGA)... So if you want to study the covalent or ionic or dipolar 
bonding, your molecule must be much closer to the surface, I'd say 3 
bohr tops, so to "see" the surface.

Then I can't help but wonder what the size of your cell is ! If you have 
a slab of ZnO, with I guess backward hydrogen passivation, fixed 
positions of deeper layer, and a resonable thickness, that is minimum 5 
bohrs already, plus 10 bohrs of void, plus say 2 bohrs of molecule, + at 
least 15 bohrs of void again (to prevent the molecule to go backward), 
that makes 22 bohrs already. This times 5x5 bohrs to "isolate" your 
molecule... This is huge !

So I's day, what you want to do is try different pinning points on the 
surface (starting much closer to the surface), with different 
orientations, look at their energy, and maybe also look at the 
liberation barrier for each, and moving barriers (with NEB like 
algorithms). As for beeing trapped in local minima, I'm afraid the 
"good" method is still to be found. You can try minima hopping, or 
ART... but nothing is perfect... far from it.

But then again, I may not understand what you want to do.

Hope that helps anyway,

Emmanuel ARRAS



David M. Wood a écrit :
> Greetings all!
>
> I'm examining the 'bonding' of a (neutral) molecule to a ZnO Zn-terminated 
> surface (using a 5 bilayer slab geometry) and, in order not to bias the 
> orientation of the molecule, started force relaxation [ionmov=3] (with the 
> molecule 'above' the surface) with the bottom atom of the molecule 10 bohr 
> above the 'top' surface atom.  To my horror, after 75 Broyden steps all I've 
> seen is small (about 0.1 bohr) oscillations about the z value where I started 
> the molecule. The molecule has *twisted* slightly in place around its 
> original axis.  With ionmov=7 (quenched Verlet molecular dynamics) the same 
> thing occurs.  I'm going to die of old age before these calculations 
> converge.  In some sense this (not my death) is reassuring  since very 
> different approaches to structural relaxation exhibit the same (annoying) 
> physics.
>
> I assume that the problem lies in intra-molecular forces far exceeding the 
> molecule-surface atom force at this distance, so that the neutral molecule 
> experiences only a weak torque due to an (inhomogeneous) electric field.  
> Even a rigid rotation of the molecule might then induce a long cascade of 
> slight readjustment of Cartesian coordinates that change the total energy 
> very little and very very slowly.  I found that optimizing isolated molecules 
> (in a large-lattice-constant cubic cell) was extremely easy by comparison.
>
> So, a several obvious questions:
>
> (0) Which mode of IONMOV is most resistant to being trapped in local minima 
> of the energy as a function, say, of the molecular orientation?
>
> (1) LARGE MAIN QUESTION: Is there a simple way within ABINIT to freeze the RELATIVE 
> coordinates of the atoms in a molecule and allow the "rigid" molecule to 
> rotate and move as a block in response to external forces (in my case, due to the atoms 
> in the slab).
>
> (2) SMALL AUXILIARY QUESTION: Is there a general protocol for such 
> relaxations: e.g., first do a not-so-well converged relaxation starting from 
> no more than 'xxx' bohr from the surface, then do a refinement once a 
> plausible near-equilibrium geometry has been found?  (Yes, xxx would 
> obviously depend on nasty details: I should estimate surface charges and 
> hence the non-uniform electric field, the molecule's dipole moment, and hence 
> the non-uniform field force and insist this exceed intra-atomic forces.)
>
> I'm using ABINIT 5.8.4 but believe my problems have little to do with ABINIT 
> per se.  Key parts of my input file:
>
> occopt 7 
> tsmear 0.01 
> ionmov 3
> toldff 1.d-6 #this is for ELECTRON SCF: 10x bigger than tolmxf:
> tolmxf 1.d-5 #somewhat finer than before (sep 15)
>
> Many thanks for any suggestions!
>
> DMW
>
>
> David M. Wood, Dept. of Physics, Colorado School of Mines, Golden, CO 80401
> Phone: (303) 273-3853; Fax: (303) 273-3919
>
>
>   

--
Emmanuel ARRAS
L_Sim (Laboratoire de Simulation Atomistique)
SP2M / INAC
CEA Grenoble
tel : 00 33 (0)4 387 86862