Module for Ab Initio Structure Evolution features
Neural network description of interatomic interactions
Evolutionary optimization of bulk, film, and nanoparticle structures
Analysis of structural properties: space group solver, structure comparion, etc.

General information
MAISE is an open-source C code for parallel execution on Linux (version 2.5, May 25, 2020)
MAISE-NET is an open-source Python script for Linux (version 1.0, May 25, 2020)

Development team
Alexey Kolmogorov (developer, 2009-present)
Samad Hajinazar (co-developer, 2014-present)
Ernesto Sandoval (co-developer, 2016-present)
Aidan Thorn (co-developer, 2019-present)
Saba Kharabadze (co-developer, 2020-present)

News and Announcements
May 2020 A review of MAISE and MAISE-NET describes key features and predictions.
May 2020 MAISE-NET Python-based automated generator of neural networks is released.
Dec 2019 NN models are shown to outperform classical potentials for guiding ab initio searches.
Apr 2019 Multitribe evolutionary algorithm is developed to accelerate prediction of nanoparticles.
Nov 2018 Neural networks are used to accelerate prediction of stable Mg-Ca alloys.
Aug 2018 MAISE and developed NN models are made publically available on Github.
Jan 2017 Stratified construction of NN models for multielement systems is developed.

Neural Network Overview
We have automated the construction of NN models for use in unconstrained structure searches.
Our developed evolutionary sampling scheme includes configurations typically encountered in global searches.
Our stratified training scheme fits NN models from the bottom up: from elements to binaries, ternaries, etc.
Neural networks are trained on energy/forces and use Behler-Parrinello symmetry functions.

We have made all published NN models publically available on Github:
Mg, Ca, Cu, Pd, Ag, Mg-Ca, Cu-Pd, Cu-Ag, Pd-Ag, Cu-Pd-Ag

We welcome requests to develop NNs for prediction of stable alloys.
Please contact Alexey Kolmogorov for more information.

Evolutionary Search Overview
Evolutionary methods are particularly suited for global optimization of complex systems.
The EA is based on passing on beneficial traits to future generations via the survival of the fittest.
Atomic positions and unit cell parameters are mutated or crossed over to produce offspring.
MAISE EA can be run with DFT (VASP), neural networks (MAISE), or classical potentials (MAISE).

The following features are currently available for optimization of crystals, films, or clusters:
different evolution and selection options: crossover + mutation, pure mutation, or pure random search
detection and elimination of duplicate structures to avoid stagnation
start from random or predefined structures
simultaneous optimization of clusters with different sizes driven by multitribe co-evolution

Select Confirmed Predictions
2016-2017 We predicted NaSn2 to be an overlooked stable phase synthesizable at ambient pressure. [31] The phase has flat honeycomb tin layers, not known to be stable in any other tin-based material, and topologically non-trivial electronic features. The proposed NaSn2 material was discovered in a following independent work.
2006-2015 Predicted [8,9] and synthesized [30] LiB has the desired structural and eletronic features to be a long-sought-after analog to the MgB2 superconductor. The material's high-pressure synthesis and characterization was complicated by an unusually complex behavior of the starting LiBx compound.
2010-2013 Predicted [16,17] and synthesized [26] FeB4 is one of the first superconductors designed fully 'in silico': a new compound with a new crystal structure and unexpected BCS superconductivity for an Fe-based material. For more details see a APS viewpoint article and a Press release.
2012 An unexpectedly complex high-pressure tI56 crystal structure synthesized and solved for the CaB6 compound [23,24]. The 28-atom ground state structure with unfamiliar 24-boron building blocks was found with our evolutionary search without any structural parameter input, e.g. truly 'from scratch'.