This documentation aims to give the short overview of USPEX code (v. 9.4.4) and real examples, which were already tested, describing possible problems and the ways to solve them.
Here you can find guidance for the use of the Quantum Espresso and CRYSTAL codes together with USPEX. The interface for the CRYSTAL was recently written at Aalto University and not yet included in the official USPEX release. The interface is available under the MIT license: https://github.com/MikhailKuklin/Interface_USPEX9.4.4_CRYSTAL. Should you have any furhter questions, please, contact Mikhail Kuklin or Antti Karttunen.
Contents of the guide
- USPEX input file INPUT.txt
- USPEX submission on Wihuri and Puhuri clusters
- Managing USPEX jobs
- Description of USPEX output files
- Examples of QE/USPEX input files
- Examples of CRYSTAL/USPEX input files
- Examples of CRYSTAL/USPEX input files for magnetic systems
Short Introduction to USPEX
Crystal structure prediction represents one of the major problems in physical sciences. We know a lot of different crystals such as wurtzite, sphalerite, cinnabar, rocksalt, etc., but what we aim to find the crystal structure of the absolutely new material and have no idea how it can look like? One way is to use our own chemical intuition and build the structures from already known crystals, but then there is a huge risk that you don't screen enough candidates. Another option is to use automatic algorithms for the search of the structure. Depending on the approach, you should set up the search parameters (experimental data, search criteria, and other technical details) and, as an outcome, you will, or you are supposed to get the desirable structure.
USPEX (Universal Structure Predictor: Evolutionary Xtrallography) is an evolutionary algorithm developed for the search of the 0D, 1D, 2D, or 3D structures. USPEX itself is a set of techniques for reliable structure prediction interfaced with ab initio codes for calculation of the total energy or other properties of candidates. Total energy, as a criterion within the simulation, is called fitness parameter within the simulation. Evolutionary algorithms have some advantages over other currently known techniques:
- no need for experimental data
- self-improving and "memory" feature - method choose the best structure from the screened generation and use them for construction of a new set
- start from "the scratch" - all initial structures are generated randomly
Evolutionary algorithms: working principle
In general, the working principle of the evolutionary algorithms such as USPEX can be described by the flowchart shown in Figure 1.
At the beginning (Start), type of the run and system have to be specified.
After that, the algorithm randomly generates a set of the structures (Initialization). Such set is called generation or population.
The next step is evaluation of the property of interest based on which the algorithm chose the best structures from the generation. Such parameter is called fitness. In most of the cases, such parameter is electronic energy of the structure.
Then, when the first generation is considered, and the best structures (in case of having electronic energy as the fitness - the lowest energy structures) are chosen, next generation has to be built. In USPEX it is implemented in such way that structural perturbations are induced on the best structures found in the previous generation. Such perturbations are called variation operators (heredity, lattice/atoms mutations, rotation, etc.). More details about variation operators in the USPEX can be found in description of the input files.
Number of considered generations depends on whether the best structure remains the same withinl generations. For example, we specified to stop the run when the same structure is found to be the most stable one in ten generations. In this case, halting criteria is achieved (Criteria achieved?) after ten generations, and the run is finished.
Figure 1. Flowchart of the working principle of evolutionary algorithms. Figure: Mikhail Kuklin
Key USPEX literature
- Glass, C.W., Oganov, A.R., Hansen, N. Comp. Phys. Comm. 2006, 175, 713-720 (DOI). General description of USPEX.
- Oganov A.R., Ma Y., Glass C.W., Valle M. Psi-k Newsletter 2007, 84,142-171 (URL). Overvew of the USPEX including various examples, description of the cell reduction and fingerprint function techniques.
- Oganov, A.R., Glass, C.W. J. Phys.: Condens. Matter 2008, 20, 064210 (DOI). Short overview of the crystal prediction methods and advantages of the USPEX comparing to them.
- Zhu Q., Oganov A.R., Glass C.W., Stokes H.T. Acta Cryst. B 2012, 68, 215-226 (DOI). Description of the USPEX techniques for molecular crystal.
- Lyakhov, A.O., Oganov, A.R., Valle, M. Computer Physics Communications 2010, 181, 1623-1632 (DOI). Technical paper describing fingerprint function, coordinate mutation, local order techniques. For advanced users.
- Lyakhov, A.O., Oganov, A.R., Stokes, H.T., Zhu, Q. Computer Physics Communications 2013, 184, 1172-1182 (DOI). Technical paper describing how random structures are built, "smart" variation operators, and preferable local environments. For advanced users.