Gautam R. Desiraju
University of Hyderabad, India
A chemist applauds an algorithm able to predict crystal structures from chemical composition alone.
I work in crystal engineering, a field that involves designing and constructing crystals with desired physical, chemical or pharmaceutical properties from small organic molecules. It is an experimental science based on pattern recognition and retrosynthetic strategies, in which the structure is considered as the sum of smaller, simpler parts.
Improvements to computational crystal-structure prediction could make design protocols more reliable. But this is such a difficult problem that only a handful of groups in the field work on it. In this context, I found a recent paper presenting a seemingly reliable method to be thought-provoking (A. R. Oganov and C. W. Glass J. Chem. Phys. 124, 244704; 2006).
Typically, crystal-structure prediction involves computer generation of putative crystal structures using a force field, which represents the interactions between atoms in neighbouring molecules. The correct structure is presumed to be that which minimizes the crystal's energy.
The procedure is problematic because the force fields may not be well tailored to the molecules being studied, and because the experimental structure may not be the lowest-energy arrangement. It is also impossible to explore all conceivable structures, which are mind-boggling in number.
Oganov and Glass use an evolutionary algorithm to localize the search to the most promising structures. Their approach is attractive in that it requires no system-specific knowledge — the input is just the molecule's chemical composition, not even its structure — and their ability to predict the unusual tetragonal structure of urea is impressive.
Is this the long-awaited breakthrough in crystal engineering? Perhaps not, but surely it's an important step forward.
Comments
The paper of Oganov and Glass offers indeed a signifucant advance in our ways to predict and explore crystal structures in chemical compounds. However, it is not the first work to employ evolutionary algorithms to predict crystal structures in an unbiased way. The same procedure has been applied in the context of colloidal/soft matter systems in a paper which appeared in JCP in 2005:
Dieter Gottwald, Gerhard Kahl, and Christos N. Likos: Predicting equilibrium structures in freezing processes, J. Chem. Phys. _122_, 204503 (2005).
There it has been shown, i.a., that even purely spherosymmetric interactions in one-component systems can stabilize exotic crystals, such as hexagonal, bco, diamond, tetragonal or even the A15-structure. The two works are complementary and the paper of Oganov and Glass brings the methodology clearly forward and stresses the value of evolutionary algorithms as a new, unbiased and versatile computational tool, as pointed out earlier in the work of Gottwald et al.
Posted by: Christos N. Likos | June 18, 2007 07:37 AM
I would like to thank Prof. C.N. Likos for drawing my attention to his work on the prediction of periodic structures of colloidal systems. It is very impressive to see in how many fields of science and technology evolutionary algorithms have been proven to be of use.
Both our methods employ evolution as "philosophy" of global optimisation - however, there are hardly any further similarities. E.g. the representation of structures is different (bit-strings in the work of Gottwald et al. (2005), but "physical" fractional coordinates and cell vectors in our method). We apply local optimisation to every produced structure, whereas Gottwald et al. do not. Finally, the variation operators (e.g. the ways of producing new structures) are also different (see the papers for details).
(1) Glass C.W., Oganov A.R., Hansen N., Comp. Phys. Comm. 175, 713 (2006).
(2) Oganov A.R., Glass C.W. J. Chem. Phys. 124, art. 244704 (2006).
(3) Oganov A.R., Glass C.W., Ono S. Earth Planet. Sci. Lett. 241, 95 (2006).
An algorithm very similar to the one proposed by Gottwald et al. (2005) has been developed for crystal structure prediction in
[T.S. Bush, C.R.A. Catlow, and P.D. Battle, J. Mater. Chem. 5, 1269 (1995)]. This algorithm works reasonably well for small systems (up to ~8 atoms in a fixed cell), and the work of Gottwald et al. (2005) attests to its power also for colloidal systems. For more complex systems we found it necessary to develop a very different algorithm, which has been successfully tested on systems with up to 128 atoms/cell, and with minor modifications could be adapted to deal with even more complex systems.
We believe that the power of evolutionary optimisation is still not fully exploited, and we will see many new methods based on evolutionary philosophy in many fields of science.
Posted by: Artem R. Oganov | June 25, 2007 09:00 AM
Respected sir, I am pursuing M.sc Physics in University of Hyderabad, my doubt is can we get information about Lattice parameters and atoms positions, why because sir, in X-Ray Diffraction we can get atoms positions by using I/I(o) data, through Computer programmes .
Posted by: M.Mallikarjuna Rao | September 4, 2007 04:17 PM