Report of Visit to UK by Dr. Stefan Goedeke
(Max Planck Institute Stuttgart)
20 June - 3 July 1996
Stefan Goedeke is in Parrinello's group at the Max-Planck-Institut fuer
Festkoerperforshung in Stuttgart. He has worked intensively on
developing methods of tight-binding and ab initio pseudopotential
calculation which scale linearly with the number of atoms, making use
of both serial and parallel computers, shared and distributed memory.
From 20 June to 3 July he was sponsored by CCP5 to tour five UK
laboratories who had expressed an interest in his ideas: Oxford
University Department of Materials, The Cavendish Laboratory, The
University of Keele, The University of Edinburgh and The Queen's
University, Belfast.
The visit proved interesting to all the host institutions. In his
seminars and very many individual discussions a number of ideas were
examined. His most important contribution to date is a Green's
function based method for finding the density matrix, that makes use of
an efficient Pade approximant to the Fermi function for carrying out
the integral over energy, and which uses an efficient algorithm for
solving simultaneous linear equations. It has the merit of being
applicable even in the case where overlap of atomic orbitals is needed.
Some of his methods had been implemented in the Oxford group in the
tight binding context. The relationship to other O(N) methods,
especially ab initio methods, were of particular interest to the Keele
group, who have been developing their own ab initio methods for
molecular dynamics. Also his method of approximating the Fermi operator
bears a close similarity to the density matrix path integral approach
of Alavi in the Belfast group.
He presented quite an elegant approach to computationally efficient
pseudopotentials, which the Edinburgh group made a start on
implementing.
On the technical side, Stefan Goedeke has exceptional expertise, and
willingness to share it, in getting the best out of cache-based
computer architectures. For example, this stimulated new thinking in the Keele
group about coding for massively parallel machines such as the Cray
T3D. He has wide experience of porting codes, and among other things
could confirm and explain the recent Belfast benchmark experience that
the forthcoming R10000 processor from SGI would be slower for MD
applications than the older R8000. His visits were particularly
appreciated not only for his high quality work, but because he is able
to interact with students and staff at all levels of technical and
scientific detail.
Last modified 28 October 2002
