USPEX release notes

Continuation of USPEX pythonization effort. Experimental active learning of interatomic potential mode. Interface for MLIP code.

Continuation of USPEX pythonization effort. New approach to calculating distance between atomic structures utilizing the same atomic fingerprints. Implemented global parent pool option. Introduced auxiliary *.uspex files, which allows to use seeds for molecular crystals. New optimization types.

Previous release (2021.0) had the core of the code and prediction of bulk crystal structures. In this version (2022.0.1) we bring major functionalities - structure prediction of thin films, surfaces and nanoparticles.

This version is a beginning of reimplementation of previously developed techniques (with improvements, bug fixes, and more transparent code structure to facilitate further developments). In this version we have reimplemented structure/compound prediction for bulk crystals.

NOTE: All regimes of 10.5 version will be reimplemented in subsequent releases.

Dear USPEXers, Let me share some good news, related to (A) release of a new version, USPEX 10.5, (B) New important online utilities, (C) New version of GDIS, allowing one to visualize and analyze USPEX output, (D) New relevant publications.

1. New release 10.5 is ready. It includes minor bug fixes and improvements, and a new important feature. Thanks to Z. Allahyari and E. Mazhnik.

   Bug fixes:
   

   • General bug fixes.
   • Fixing bugs in multiobjective optimization when formula is used in optType.
   • Fixing a bug in DFTB+ interface.

   New features and improvements
   

   • Improving interface with gulp, to be compatible with gulp 5.2.
   • Multiobjective optimization feature with more than one formula expressions in optType block.
   • Adding a USPEX -u feature, to update USPEX instantly to its latest version, without downloading the whole package. Each new update will change the version name. So, the main version is 10.5 and the name after the first update will be 10.5.1. and after the second update 10.5.2. and so on.
   • New optimization types (optType=1201-1207) – calculation the elastic moduli using graph convolutional neural network (and many quantities derived from them, including hardnesses and fracture toughnesses computed using Mazhnik-Oganov model). To learn how to use this feature, see new Example35, and see online utilities.
   • Another useful example (Example36) – variable-composition compound/structure prediction with QuantumEspresso (example provided by S. di Cataldo).

2. New online utilities (USPEX Tools and Utilities — USPEX (uspex-team.org))

   • Hardness_ML – you provide the crystal structure and this program, using graph convolutional neural network, predicts bulk and shear modulus – from which calculates Vickers hardness and fracture toughness using Mazhnik-Oganov model. Thanks to E.Mazhnik.
   • Hardness3.0 – you either provide elastic constants tensor (from which you get aggregate values - bulk and shear moduli, Young’s modulus and Poisson’s ratio), or provide aggregate values. The program gives you Vickers hardness and fracture toughness calculated using Mazhnik-Oganov model. Thanks to E.Mazhnik.
   • Pressure-composition phase diagram – you provide extended_convex_hull file from USPEX output at a given pressure, and the program gives an approximate phase diagram in a range of pressures. Thanks to E.Mazhnik.

3. New visualization tool (under Linux) – thanks to H. Valencia and G. Frapper, there is support for analyzing USPEX results in GDIS (GitHub - arohl/gdis: A visualization program for the display, manipulation, and analysis of isolated molecules and periodic structures).

   Check out the recent publication: New developments in the GDIS simulation package: Integration of VASP and USPEX - Okadome Valencia - - Journal of Computational Chemistry - https://doi.org/10.1002/jcc.26697.

4. Some relevant publications
   • New model of hardness and fracture toughness:
     Mazhnik E., Oganov A.R. (2019). Model of hardness and fracture toughness of solids. J. Appl. Phys. 126, 125109.
   • Graph convolutional neural network for calculating the elastic moduli of materials:
     Mazhnik E., Oganov A.R. (2020). Application of machine learning methods for predicting new superhard materials. J. Appl. Phys. 128, 075102.
   • Understanding the physical meaning and redefining Mendeleev numbers of the elements:
     Allahyari Z., Oganov A.R. (2020). Nonempirical definition of Mendeleev numbers: organizing the chemical space. J. Phys. Chem. C124, 23867-23878.
   • New scale of electronegativities of the elements:
     Tantardini C., Oganov A.R. (2021). Thermochemical electronegativities of the elements. Nature Communications 12, 2087.
   • Demonstration of power of USPEX in comparison with other codes – for the important Ti-O system:
     Kun L., Wang J., Oganov A.R. (2021). High-pressure phase diagram of the Ti-O system. J. Phys. Chem. Lett. 12, 5486-5493.
   • Prediction of extremely complex quaternary (!) phase diagram C-H-N-O using USPEX:
     Naumova A.S., Lepeshkin S.V., Bushlanov P.V., Oganov A.R. (2021). Unusual chemistry of the C-H-N-O system under pressure and implications for giant planets. J. Phys. Chem. A125, 3936-3942.
   • Prediction and experimental synthesis of a very complex high-pressure phase Eu₈H₄₆:
     Semenok D.V., Zhou D., Kvashnin A.G., Huang X., Galasso M., Kruglov I.A., Ivanova A.G., Gavriliuk A.G., Chen W., Tkachenko N.V., Boldyrev A.I., Troyan I., Oganov A.R., Cui T. (2021). Novel Strongly Correlated Europium Superhydrides. J. Phys. Chem. Lett. 12, 32-40.
   • New extremely high-temperature superconductors predicted and synthesized:
     Semenok D.V., Troyan I.A., Ivanova A.G., Kvashnin A.G., Kruglov I.A., Hanfland M., Sadakov A.V., Sobolevskiy O.A., Pervakov K.S., Lyubutin I.S., Glazyrin K.V., Giordano N., Karimov D.N., Vasiliev A.L., Akashi R., Pudalov V.M., Oganov A.R. (2021). Superconductivity at 253 K in lanthanum–yttrium ternary hydrides. Materials Today, in press.
   • New program (AICON, version 2) for efficient calculations of thermal conductivities and thermoelectric properties:
     Fan T., Oganov A.R. (2021). AICON2: A program for calculating transport properties quickly and accurately. Computer Phys. Comm. 266, 108027.

Dear USPEXers, We are happy to announce the release of version 10.4. The changes are relatively significant and include bug fixes, optimization of new properties, interfaces to additional ab initio codes, and improvements of efficiency of VCNEB. This version runs smoother and we recommend users to switch to it from 10.3. Below is a list of changes:

1. Bug fixes:
   1. Minor bug for multi-objective optimization when the population size is very small.
   2. Bug in permutation.
   3. Reading error when IonDistances matrix in INPUT.txt is not symmetric.
   4. Inconsistent axes in volume_vs_enthalpy figure.
   5. Bugs in VCNEB.
   6. Bug in variable-composition calculation of quaternary systems (now such systems can be run).
   7. Bug in PSO.
   8. Reading seeds in 200 mode.
   9. Bugs in DFTB+ energy reading.
   10. Bugs in outputs and figures.
   11. Bugs for submitting jobs.
2. New features:
   1. Half-metalicity fitness (for 300/301 regimes).
   2. Interface to Abinit.
   3. Interface to CRYSTAL code (It is not tested).
   4. New model for calculation of hardness and fracture toughness (Mazhnik E., Oganov A.R. (2019). Model of hardness and fracture toughness of solids. J. Appl. Phys. 126, 125109) for 300/301 regimes.
3. Improvements:
   1. USPEX works in python3 now (all python2 scripts are rewritten in python3).
   2. Remote and local submission jobs are improved.
   3. VCNEB significantly improved.
   4. DMACRYS interface is improved (better than before, but still not the best).
   5. Quantum Espresso interface for variable compositions and surfaces.
   6. VASP interface - Discarding converged structures in VASP, when they have crazy CONTCARs.
   7. Updated references (for Pareto optimization).

This is a minor release, which is an improvement of v.10.2. A few important bugs were fixed.

Dear USPEXers, We are releasing version 10.2 of USPEX. This is a very major upgrade, with many new features, improved robustness, and all known bugs fixed. We have removed the condition that users cannot participate in the development of competing codes, and warmly invite everyone to use USPEX under conditions specified on our website (uspex-team.org).

1. No more dependence on Matlab: USPEX now is distributed as a compiled code, the use of which does not require any commercial products such as Matlab.
2. List of new features:
   1. Implementation of random topological structure generator (Bushlanov, Blatov, Oganov, 2019), which greatly speeds up the calculations.
   2. Magnetic optimization. Available for clusters and crystals. This allows one to predict magnetic states of materials.
   3. Now fitnesses can be minimized or maximized (min_ or max_), and mathematical expressions can be mini/maximized.
   4. Pareto optimization, enabling simultaneous optimization of two or more properties (Allahyari & Oganov, 2018). optType is now a keyblock, not a keyword.
   5. Implementation of optimization of thermoelectric figure of merit (ZT), which is computed using BoltzTraP (Nunez-Valdez et al, 2018).
   6. Implementation of optimization of birefringence and fracture toughness (Niu, Niu, Oganov, 2019).
   7. Interface with Gaussian, MOPAC, DFTB, ORCA, FHI-aims, ABINIT.
   8. Use of plane groups for generating structures of 2D-crystals.
   9. Variable-composition prediction of 2D-crystals (following Revard et al., 2016).
   10. For surfaces, non-integer reconstructions (such as sqrt(2)xsqrt(2)) are enabled.
   11. For surfaces, the code now outputs the phase diagram.
   12. Symmetry determination is now done using SPGLIB. Symmetrization can also be optionally performed during calculation of physical properties, making it more robust.
   13. Added utility “pmpaths” based on the method of Stevanovic et al. (2018), allowing prediction of the likeliest phase transition mechanisms. This will only give a rough approximation, which you need to refine using VCNEB method implemented in USPEX (pmpaths gives files that can be directly read by VCNEB@USPEX).
   14. VCNEB module of USPEX has been improved and now converges a few times faster.
   Bug fix: structure generation for quaternary systems in varcomp didn’t work before, now is fixed. Various other bugs fixed. Version 10.2 has no known bugs.
3. Changes in input:
   1. Keyword gap -> bandgap.
   2. Change of specification of fitness (optType is now a keyblock, not a keyword).
   3. For calculations of physical properties symmetrization of the structure can now be performed – for this, use parentheses “()” to specify stages of calculation where symmetrized structure will be used.
4. New mode of distribution: Now USPEX is distributed as a compiled code. This makes the code easier to install, and removes the need to have Matlab or any other commercial software.
5. Documentation:
   1. New website
   2. USPEX Manual has been updated to describe all changes and new features. Both English and Chinese versions are available.
   3. USPEX distribution now includes a number of MOL-files as examples.
   4. Improved many examples (especially Example13).
   5. Rerun all examples with the new code. Added many new examples. Now we have 34 examples of different types of calculations, and new examples include:
      1. two examples with Quantum Espresso.
      2. surface structure prediction for MgO.
      3. examples of calculations with FHI-aims, DFTB, MOPAC, Gaussian.
      4. Pareto optimization of thermoelectric figure of merit (ZT) and enthalpy.
      5. single-block (=fixed stoichiometry, but variable total number of formula units) optimization of the band gap for Si.
      6. magnetic structure prediction with USPEX.
      7. example of a TPS calculation.
      8. variable-composition prediction of stable 2D-compounds of Sn and S.
      9. variable-composition prediction of surface oxidation of Pd.
   Happy USPEXing!
6. References:
   1. Allahyari Z., Oganov A.R. (2019). Multi-objective optimization as a tool for materials design. In:.Handbook of Materials Modeling (ed. W. Andreoni, S. Yip). Volume 2 Applications: Current and Emerging Materials. Springer Verlag.
   2. Bushlanov P.V., Blatov V.A., Oganov A.R. (2019). Topology-based crystal structure generator. Comp. Phys. Comm. 236, 1-7.
   3. Núñez-Valdez M., Allahyari Z., Fan T., Oganov A.R. (2018). Efficient technique for computational design of thermoelectric materials. Comp. Phys. Comm. 222, 152-157.
   4. Niu H.Y., Niu S.W., Oganov A.R. (2019). Simple and accurate model of fracture toughness of solids. J. Appl. Phys., in press.

This is a minor release. With the help of users (especially Ah Hui), and with our own testing, we found and fixed 19 bugs and improved Examples and Tests. This version is, to the best of our knowledge, bug-free. It is fully compatible with Octave 3.4 and partially compatible with Octave 3.6/3.8/4.0. Contributors to version 9.4.4: M.S. Rakitin, Q. Zhu, M. Davari, S. Wang, F. Qi, G.R. Qian, A.R. Oganov.

This version is intended to be the current “golden standard” in computational materials discovery. Please switch to version 9.4.4, all previous versions of USPEX are obsolete.

Also, stay tuned – we are working hard on the new major release (version 10.1), which will take some time, but will feature major speedups and new functionalities.

It includes fixing a number of bugs (which should slightly speed up performance), extending functionalities, and improving documentation. This version should be nearly bug-free and is a milestone towards a very major upgrade, which will be made available in version 10. Contributors to v. 9.4.3: Q. Zhu, G.R. Qian, M. Rakitin, F. Qi, Z. Allahyari, P. Bushlanov, Q.F. Zeng, D. Dong, J. Zhang, A.R. Oganov.

1. User-friendliness:
   1. The Manual has been updated.
   2. A Chinese version of the Manual has been created (link)
   3. Several examples have been added: illustrating surface calculations, evolutionary metadynamics, generalized evolutionary metadynamics, use of USPEX in conjunction with MOPAC.
2. Interfaces with external codes:
   1. Improved interface with MOPAC for nanoparticles and crystals.
3. Internal machinery of the code, defaults, file names:
   1. Improved handling of space group symmetry and unit cell relaxation.
   2. Changed defaults for space group determination to more reliable ones: default general tolerance is now 0.08, and defaults for HIGH/MEDIUM/LOW = 0.04/0.08/0.15. We remind the users that obtained structures must be checked for the presence/absence of imaginary phonon frequencies.
   3. File non_optimized_structures renamed to gatheredPOSCARS_unrelaxed
   4. Changed default values for ionDistances, making them more robust.
   5. Softmutation rewritten and improved.
4. Debugs:
   1. A number of bugs have been fixed – for handling symmetry, restart from previous generation, distance calculations for molecular crystals, reading of band gaps, calculation of structural quasientropies and degrees of order. Some bugs affected performance of the code, so the new version should be slightly faster. No known bugs remain in the code (which is now fully compatible also with Octave3.4).
5. New methodological publication:
   1. Generalized evolutionary metadynamics (GEM) method and its applications to the discovery of novel allotropes of boron and silicon has been presented in the paper: Zhu, Q., Oganov, A. R., Lyakhov, A. O., and Yu, X. (2015). Generalized evolutionary metadynamics for sampling the energy landscapes and its applications. Phys. Rev. B, 92, 024106 (pdf).

This is a minor release with mostly bug fixes, full Octave compatibility, compatibility with 32-bit architectures, updated Manual, and several new features. Contributors to this version: Q. Zhu, Z. Allahyari and G.R. Qian.

1. New features:
   1. Re-written convex hull code, making variable-composition calculations more robust.
   2. Improved variable-composition searches for ternary, quaternary, and more complex systems.
   3. Interface with MOPAC (abinitioCode=13).
   4. To make use of the code even more convenient, we implemented intelligent defaults for several parameters, and thanks to these defaults you don’t need to specify these parameters explicitly anymore: goodBonds, valences, IonDistances.
   5. Added Example 17 for optimization of elastic properties of carbon, and Example 18 for a ternary variable-composition search for the Zn-O-H system.

New major release of USPEX, v.9.4.1. The list of changes is very large, and makes all previous versions obsolete. More than 800 modifications were introduced into the code. Contributors to this version: Q. Zhu, G.R. Qian, M. Rakitin, H.Y. Niu, F. Qi, S. Athar, M. Davari, A. Masunov, B.X. Li, A.R. Oganov.

1. New features:
   1. Optimization of the elastic properties – now users can optimize the bulk, shear, Young’s and Poisson’s moduli, Pugh’s ratio, and hardness from the Chen model, as well as the Debye temperature and acoustic wave velocities.
   2. Prediction of polymeric crystal structures is now enabled (calculationType=110).
   3. Now USPEX calculations for solids (calculationType= 300, 310, 301 and the as yet unreleased 311, which will be distributed in the next release) and surfaces (as yet unreleased calculationType=201, which will be distributed in the next release) give in the output not only the most stable structures at given conditions, but also phase diagrams. For solids they give you an idea (crude one – needs to be checked!) of what the new phases may be at higher and lower pressures, and a rough idea of transition pressures. For surfaces you get exact phase diagrams in terms of chemical potentials.
   4. Automated analysis for statistics. This is mainly needed for developers, but curious users may try this to see by how many times USPEX is faster and how much higher its success rate is, compared to another code. This feature is available only when stop_fitness is specified in INPUT.txt.
   5. Anti-composition feature enabled.
   6. Automatic evolution of variation operators (parameter control, enabled by option “1 : AutoFrac”), which speeds up the calculation by ~2 times.
   7. More powerful seeds technique: we now can organize seeds for specific generation, and more flexible of the sequence of different types of atoms in VASP5 format POSCAR file.
2. User-friendliness:
   1. Users no longer need to specify the unit cell or atomic volumes in the keyblock Latticevalues – we have implemented a special algorithm that accurately estimates it at the pressure of interest, without the need for the user to specify it. You can also use online program. The users are still able to input the volumes manually.
   2. To create INPUT.txt files, you can also use online program - we recommend this for first-time users, to learn the meaning of all keywords (volume estimation algorithm is implemented here as well).
   3. Now pressure value (in GPa) is set by the tag ExternalPressure in the INPUT.txt file. Please NO LONGER specify it in relaxation files in the Specific folder.
   4. Single-block calculations now enabled in calculationType= 300/310.
   5. Now the code comes with a number of tests (to very quickly check that the code performs all key operations) and examples (realistic accurate calculations, which users can take as templates for new calculations). In particular, there are examples of calculations for 2D-crystals, surfaces, polymers, and for the vc-NEB method (to predict phase transition pathways).
   6. New data set (USPEX.mat) which can be used for post data analysis.
   7. Re-organized output files, which are more logical and easier to analyze.
   8. More detailed Warning messages during the calculation.
   9. High-resolution pdf-format convex hull figures are now output.
   10. For all USPEX calculation interfaced with VASP, the INCAR files can be edited and made effect during the calculations.
   11. Compositions statistics enabled in calculationType=301, fitness profile in 300 and 310.
   12. Now can move the calculation folder freely, without errors (when all jobs are done).
   13. Safer Matlab data files (.mat files) saving method for unstable file systems, which can easily recover the interrupted data file from backup (.mat.backup).
   14. The Manual has been updated.
   15. USPEX installation procedure has been implemented, as well as Python-based USPEX runner. Use 'USPEX -h' to see all available options and 'USPEX -r' to run USPEX calculation.
3. Interfaces with external codes:
   1. Structures are now represented in VASP5 POSCAR format (which differs from the previous format by the presence of an extra line with element names). STM4 code for visualization is now able to read this format (please update it at http://mariovalle.name/STM4/kits/index.html).
   2. Interfaced with FHI-aims for calculation type 000 (nanoparticles).
   3. Interfaced with Tinker for calculation type 310 (molecular crystals).
   4. Improved interfaces for ATK, CASTEP, CP2K, DMACRYS, GULP, LAMMPS, Quantum Espresso, SIESTA, VASP.
4. Internal machinery of the code:
   1. Automatic testing system has been created, to help fast and deep debugging. This will help to keep 9.4.1 and all future versions reasonably bug-free.
   2. A number of bugs was fixed.
   3. Source code folders hierarchy has been updated.
   4. Global variables have been updated and unified.
   5. Python-based INPUT.txt parser has been added.

This is a significantly improved version. We have made numerous updates, introduced major new functionalities and features, debugs. Improved user-friendliness (better output, easier input for molecular crystals, easier setup for remote calculations). Please download it from our new website.

1. New functionalities:
   1. Fully debugged codes for 2D crystals and GEM. GEM is a totally new method that is under final testing and will become available soon.
   2. Variable-cell Nudged Elastic Band method (Qian et al., Comp. Phys. Comm., 2013) for predicting phase transition pathways is under final testing and will become available very soon.
   3. Updates for variable-composition algorithm. More robust algorithm. More versatile too, now we have two types of varcomp calculations: single-block and multiple-blocks.

      For single block:

      • % numSpecices
      • 1 2
      • % EndNumSpecices
      • % atomType
      • Si O
      • % EndAtomType

      It means we sample structures of compound SiO2 (with the ratio of 1:2) with a variable number of formula units.

   4. Fully functional surface calculations (calculation types 200, 201; see Zhu et al.,Phys. Rev. B, 2013
   5. New calculation type (110), to study the packing of polymers. Will become available for users shortly.
   6. New optimization types – now it is possible to optimize band gaps, dielectric constants, and newly invented figure of merit of dielectric materials (for all these types of optimizations, see Zeng et al., Acta Cryst, 2014
   7. Interfaced USPEX with LAMMPS and ATK. Added new tests to illustrate the use of these codes. Improved setup of many old test cases.
2. Technical changes:
   1. For vasp calculations, one can prepare POTCAR in the same way for fixed- and variable-composition calculations. For instance, you can just put POTCAR_Mg and POTCAR_O in the specific folder if you do Mg-O variable-composition or MgO fixed-composition calculations. USPEX will by default search if there exist POTCAR_1,2,.... If there is no POTCAR_1 file, USPEX will prepare it from POTCAR_Mg and POTCAR_O.
   2. Seeds. We no longer specify them as POSCARS_1, _2, .etc. Instead, we just put POSCARS during the calculation. USPEX will read this file and use it for the next genereation. All the seeds files will be recorded into the file named 'Seeds/Seeds_history'.
   3. Simplified MOL_* files for (310) and (110) types of calculations.
   4. 301 calculations now prepare “lattice values” according to blocks – these will be volumes not of atoms, but of the compositional blocks. For the case of MgO-SiO2 system, you should put volumes of MgO and SiO2 formula units in the INPUT.txt
   5. For 301, we also support single block calculation. For the case of elemental boron, one just need to put one block in the INPUT.txt
      • % numSpecices
      • 1
      • % EndNumSpecices

      Summary:

      Calculation Types:

      • 000: clusters, nanoparticles
      • 110: prediction of packing of 1D polymers
      • 200: surface reconstructions (fixed composition)
      • 201: surface reconstructions (variable composition, up to binary systems)
      • 300: fixed composition for 3D crystal structures
      • 310: fixed composition for 3D crystal structures (from molecular building blocks)
      • 301: variable composition for 3D crystal structures
      • 311: variable composition for 3D molecular compounds (co-crystals)
      • -200: 2D crystal structures (user defined thickness)

Release of the USPEX code version 9.3.4 is now available. All previous versions are obsolete. The input and output formats have been made simpler and more informative. Improvements in variable-composition calculations, seed technique, molecular structure prediction, massively parallel calculations. Added interface to CASTEP and LAMMPS. Several bugs fixed. Updated documentation and added new test cases.

New version (v.9.2.7) is available for download. There are significant improvements, but unless you need these improvements, you can continue using version 9.1.7.

1. New major features:

   A major new feature is released: an 'ageing' (antiseeds) technique, aimed at improving population diversity and the ability of the algorithm to find the global minimum. For more details, see Comp. Phys. Comm. 2013

2. Improved features:
   1. Optimization of order (minimization or maximization of the order or quasientropy) for alloys, without structure relaxation (for which, we introduced a new value abinitioCode=0). This allows, among many other things, an easy creation of quasirandom structures (based on the more general definition than the so-called “special quasi-random structures”). Since structure relaxation is not needed here, these calculations can be extremely fast, even on large supercells.
   2. An improvement was made in the handling of symmetry. This improvement is particularly important for fixed-cell calculations.
   3. For fixed-cell calculations, you can specify now the cell parameters, not only in the form of a 3 x 3 matrix, but also as a role of six values (three length in Angstroms and three angles in degrees), e.g.: 10.010 8.756 4.789 90.0 90.0 90.0. The former use all 3 x 3 matrices is also allowed.
   4. For the maximum number of permutations swaps (parameter howManySwaps), would have introduced an intelligent default, so that in virtually all cases, you can avoid manually setting this parameter and just rely on the default.
   5. Added interface to CASTEP (contribution of Zamaan Raza) and new test cases:
      • Generation of quasirandom structures
      • Antiseed technique (two tests, illustrating different uses of this technique)
      • Example of USPEX calculation with Quantum Espresso

Examples of USPEX calculation with CASTEP and ATK are in preparation.

New version (v.9.2.4) is available for download. Some of the new features include:

1. Improved symmetrical initialization for nanoparticles (clusters).
2. Order enhanced heredity for nanoparticles.
3. New parameter to tune the tolerance for the space group determination.
4. Better defaults, various minor bugs fixed.
5. New property (quasientropy) can be optimized.

We proudly announce the new version, v.9.1.7 - a major new release, which makes all previous versions obsolete and sets a new standard in the field. You will find major new functionalities and improvements, updated documentation and set of tests.

1. New major features:
   1. Evolutionary metadynamics (ref. [1]), a very powerful method for predicting stable structures, which also gives a large number of metastable structures that can be reached from a particular user-chosen state. For instance, you can find all carbon structures that can be made by compressing graphite (ref. [2]), carbon nanotubes, etc. This method also gives transition pathways between the structures, so in many ways it is complementary to USPEX. Test case is added as an illustration.
   2. Extension of our evolutionary algorithm to molecular crystals (ref. [3]), which enables affordable and reliable predictions. Test case is added as an illustration.
   3. PSO (Particle Swarm Optimization) method - the original version for crystal structure prediction, developed by A.Boldyrev in 2007 and its 2010 Wang-Lv-Zhu-Ma reimplementation, was improved by us by removing some unphysicalities. This corrected PSO algorithm was implemented with minor programming work on the basis of USPEX. However, the performance is generally inferior compared to USPEX - and the same is true of the older PSO versions (e.g., PSO relies on symmetry being used, its success rates and efficiency are lower, etc.). Use for testing purposes only.
2. Improved features:
   1. Debugged the softmutation operator.
   2. Improved variable-composition functionalities. Created utility extendedConvexHull.m for easy analysis of the results.
   3. Debugged the cluster (nanoparticles) prediction code.
   4. Improved symmetry features of the code.
   5. Greatly improved and updated documentation (README and Manual).
   6. Enlarged and refreshed set of tests.
   7. Graphical output, which enables rapid pre-analysis.
   8. Additional property can be optimized now - magnetic moment (thankts to R. Agarwal).