What is USPEX? USPEX is a method developed jointly by Artem R. Oganov, Andriy O. Lyakhov, Colin W. Glass and Qiang Zhu, and implemented in the same-name code written by Andriy O. Lyakhov, Colin W. Glass and Qiang Zhu. This method/code enables crystal structure prediction at arbitrary P-T conditions, given just the chemical composition of the material. Many previous attempts to solve crystal structure problem were plagued by low success rate and extreme computational costs that prevented full ab initio studies. USPEX avoids both of these problems. In fact, "uspekh" means "success" in Russian - which highlights a nearly 100% success rate that we find for our method.
How USPEX works: USPEX is based on a carefully tuned structure prediction-specific evolutionary algorithm. USPEX searches for the structure corresponding to the global minimum of the ab initio free energy. The quality of trial structures is judged by the ab initio free energy calculated by an external ab initio code (currently for this our code can use VASP and SIESTA, but we may include other codes as well). The use of simpler methods, e.g. based on interatomic potentials, is also possible - in this case the structure prediction is extremely fast.
Figure. Test of USPEX: 40-atom cell of MgSiO3 post-perovskite. Left - structure search using local optimisation of random structures, Right - evolutionary search with USPEX. While random search did not produce the correct structure even after 120000 steps, USPEX found the stable structure in fewer than 1000 steps. Note that in random search the distribution of the energies remains constant throughout the run (no "learning" involved). In USPEX simulations, random starting structures have the same energy distribution as in the random search - but very quickly it shifts to lower energies (result of "learning"), while some higher-energy structures continue to be produced (this "diversity" enhances the ability of the algorithm to explore new areas of the energy landscape).
Features of the code.
(1) Structure prediction using no experimental information, with just the chemical composition.
(2) incorporation of partial structural information is possible - (a) constraining search to fixed experimental cell parameters, or fixed cell shape, or fixed cell volume, (b) starting structure search from known or hypothetical structures.
(3) efficient contraint techniques, which eliminate unphysical and redundant regions of the search space. Cell reduction technique (Oganov & Glass, 2008).
(4) handling of molecules (rather than atoms), fully or partly rigid (or fully flexible) is possible.
(5) niching using fingerprint functions (Oganov & Valle, 2009).
(6) restart facilities, enabling calculations to be continued from any point along the evolutionary trajectory (if needed, with changed parameters).
(7) powerful visualisation and analysis techniques implemented in the STM4 code (by M.Valle), fully interfaced with USPEX.
(8) USPEX is interfaced with VASP, SIESTA and GULP codes. Interfacing with other codes is easy. There are development interfaces with ABINIT, LAMMPS, and MD++.
(9) submission of jobs from local workstation to remote clusters and supercomputers is possible.
(10) job submission via grid is possible (grid part written by S. Tikhonov and S. Sobolev).
(11) many new features are now in progress. to be described later...
Current limitations of USPEX. Because of the high success rate of the method, we have not seen many limitations in practice. It is efficient for systems with up to 100-200 atoms/cell. However, it becomes expensive to do USPEX simulations at a fully ab initio level for systems containing more than ~30 atoms in the unit cell (more than sufficient for solving most crystallographic and geophysical problems). Difficulties for large systems are due to the increasing cost of ab initio calculations for increasing system sizes, and also due to the rapidly increasing number of free energy minima. Our algorithm seems to be very effective in counteracting this effect and will make structure prediction for systems containing hundreds of atoms affordable in near future.
How to collaborate on USPEX. We welcome collaborations with experimentalists finding new interesting phases and wishing to find their structure, however since we have constantly a large number of such suggestions for collaboration we will have to select only those suggestions which are closest to our own interests (high-pressure crystallography). We are also happy to establish collaborations with industrial partners. And - last, but not least - we welcome all interested theoretical and computational scientists to join the development of USPEX. The best way to enquire about a possible collaboration is to e-mail Prof. A.R. Oganov.
Conditions for becoming a user of USPEX. The USPEX code is public domain, but as for any public code, there are certain conditions that users must sign to -
(i) the code is given to an individual researcher (not a group or institution), users are not allowed to distribute the code,
(ii) we encourage new users to seek advice of the USPEX team in setting the first calculations, to avoid misuse of the code (the code is user-friendly, but the method has a rather different philosophy from traditional simulation methods),
(iii) citations to the original USPEX publications must be present in all papers that used USPEX.
(iv) all new features that the users would like to implement will have to be sent to Prof. A.R. Oganov in order to be included in the common version of USPEX, maximally benefiting the user community. Users of USPEX are welcome to participate in the development of USPEX and will then be named as its coauthors, but will refrain from developing any competing codes.
4. Glass C.W., Oganov A.R., Hansen N. (2005). Predicting crystal structures of new high-pressure phases. (Invited lecture, 20th IUCr congress, 23-31 August 2005, Florence, Italy). Acta Cryst. A61, C71, abstract MS54.27.5. (pdf-file).
5. Martonak R., Oganov A.R., Glass C.W. (2007). Crystal structure prediction and simulations of structural transformations: metadynamics and evolutionary algorithms. Phase Transitions80, 277-298 (pdf-file).
6. Oganov A.R., Ma Y., Glass C.W., Valle M. (2007). Evolutionary crystal structure prediction: overview of the USPEX method and some of its applications. Psi-k Newsletter, number 84, Highlight of the Month, 142-171 (pdf-file).
7. Oganov A.R., Glass C.W. (2008). Evolutionary crystal structure prediction as a tool in materials design. J. Phys.: Cond. Mattter 20, art. 064210 (invited paper) (pdf-file).
8. Oganov A.R., Ma Y., Lyakhov A.O., Valle M., Gatti C. (2010). Evolutionary crystal structure prediction as a method for the discovery of minerals and materials.