MS&E Colloquium: Prof. Phillip M. Duxbury, Michigan State University
Date: February 27, 2009 from 2:00 pm to 3:00 pm EST
Location: Columbia University
Morningside Campus
214 S. W. Mudd Building
Contact: For further information regarding this event, please contact Chad Gurley by sending email to cg2029@columbia.edu .
Info: Click Here to Visit Website.
Bookmark and Share  

Computational geometry, combinatorial optimization and nanostructure determination*

Phil Duxbury, Michigan State University

Crystallography is the gold standard for structure determination and provides precise information about inter-atomic distances, spin structures, charge structures and thermal factors.  Achieving precise structural characterization of non-crystalline materials is a grand challenge in experimental and computational science and is central to progress in a wide range of organic and inorganic materials, including nano-particles, proteins in solution and nano-motifs in the bulk of complex materials.  Nanostructure determination in non-crystalline materials relies on a variety of computational methods ranging from refinement based on carefully chosen starting states to reverse Monte Carl methods, with the latter being typically under-constrained while the former is usually over-constrained. Over-constrained models only provide a good fit if a high quality starting state is used, while under-constrained models often yield a high quality fit to the data even when the structure is not accurate.  We are developing new approaches to ab-initio nanostructure determination (illustrated in the figure) that take an intermediate approach where experimental data is used to place strict constraints on starting structures and additional information is added to further refine the system.  The starting structures are constrained by interatomic distance lists extracted from experimental data using novel peak extraction methods. Chemical placement is then optimized with respect to the data using coloring algorithms.

Theoretical issues key to making this approach robust and broadly applicable will be discussed, including: (i) Is there enough information in the pair distribution function to constrain the structure? How many distances are required to constrain the atomic positions? (ii) Is this distance geometry inverse problem well posed - when is a unique reconstruction possible?  (iii) How sensitive is the result to noise in the data?  (iv) Are we reinventing the wheel - how is this different than distance geometry methods used in NMR structure solution of proteins?  The status of these issues and key outstanding aspects will be elucidated.

* In collaboration with: Simon Billinge, Pavol Juhas, Chris Farrow, Luke Granlund, Suraubh Gujarathi, Corey Musolff.  Supported by the DOE.

Hosted by Prof. Simon Billinge