We are trying to simulate hydroxyl groups on Fe surfaces using a cluster model.

The cluster consists of 10 Fe atoms in a 7,3 configuration. The central Fe atom
has an OH group added. We found that to get a converged SCF for the cluster two
approaches were possible:

1) alter the multiplicity of the system until you achieve a value that
converges. This gave us large spin values on the cluster (20-30).

2) Use a smearing parameter. We started with a large value (0.5) and reduced it
until the SCF began to fail again. The optimum value we have fixed on is 0.2.

We have found that method 1 is difficult to use since you have to alter the
multiplicity from the clean cluster to the cluster with OH added. It also
seemed unphysical to us to have such a large residual spin. So we are
continuing with method 2.

Our problem is that as we change the angle of the Fe-O-H bond the geometry
optimiser will run for a few steps and then proclaim that the structure is
optimised for almost any angle we start with. If we plot the energy obtained as
a function of bond angle we obtain a reasonable looking graph with a maximum at
180 degrees and a minimum somewhere near 100 degrees. However if we use our
graph to estimate the potential energy gradient and then adjust the minimiser
tolerance to 100 times lower than our estimate the structure DOES NOT optimise
to the minimum on the graph. The Fe-O and OH bond lengths are altered by the
minimiser as we change the bond angle but the angle itself never gets moved by
more than a few degrees.

During all this work the Fe atoms are fixed and only the OH positions are
altered. The co-ordinates are input in cartessian form.

Has anyone similar experiences?

Dave Willock

Dr David J. Willock
Department of Chemistry
University of Wales, Cardiff
P.O.Box 912
Cardiff
CF1 3TB

Tel. 01222 874 000
Received on 1998-06-26 19:47:01