Algorithm Discovers New Superconductor

In nature, random genetic variations produce results that are sorted by survival. Over time, the best working variations reproduce more often, creating the process we call evolution. Evolutionary algorithms follow the same concept, trying small variations to advance the solution of a problem. An evolutionary algorithm designed to predict stable crystal structures has discovered a new superconductor.

Scientists create first computer-designed superconductor – []

More than five years ago, Kolmogorov, then at Oxford University, began studying boron-based materials, which have remarkably complex structures and a wide range of applications. He developed an automated computational tool to identify previously unknown stable crystal structures without any input from experiment. His “evolutionary” algorithm emulates nature, meaning it favors more stable materials among thousands of possibilities. (Kolmogorov is a computational physicist, but he also dreams of holding a compound in his hands that he predicted in silico.)

The search revealed two promising compounds in a common iron-boron system, which came as a surprise. Moreover, graduate student Sheena Shah’s calculations indicated that one of them should be a superconductor at an unusually high temperature of 15-20 Kelvin for the considered (so-called “conventional”) type of superconductivity.

Module for Ab Initio Structure Evolution (MAISE)


  • The module enables an evolutionary algorithm (EA)-based search for ground states in multi-component alloys
  • Only the composition is required as an input; the crystal structure optimization is unrestricted
  • MAISE can be linked with any external package that calculates the total energy of a given structure, such as VASP
  • The code is fully automated: it runs in the background under LINUX submitting and processing VASP jobs via serial or parallel queue
  • The current version is 4.0
  • MAISE (Scottish Gaelic for beauty, grace, elegance) was written by Aleksey Kolmogorov at the University of Oxford in 2009

Evolutionary methods are particularly suited for finding global minima in complex systems with a large number of degrees of freedom.
The EA is based on passing on beneficial traits to future generations via the survival of the fittest.
For compound prediction, atomic positions and unit cell parameters are used as genes that are swapped and mutated to produce offspring.
Efficient performance of EA requires the tuning of many parameters, such as generation size, selection criteria, crossover and mutation.
MAISE takes advantage of the best strategies reported in the recent years [1-4] as well as the extensive in-house tests to achieve fastest convergence.
The following features are currently available:

  • different evolution and selection options: crossover + mutation, pure mutation, or pure random search
  • detection and elimination of duplicate structures to avoid stagnation
  • plane and periodic cuts for crossover
  • start from random or predefined structures
  • use of building blocks for dealing with large complex systems (to be published)

One should keep in mind that no theoretical method can guarantee that a predicted crystal structure is the true ground state
because every search method has constraints, most common ones being the composition and the unit cell size.
However, identification of phases in a given system more stable than the known ones can stimulate a targeted experimental search for new materials.

Evolutionary Computation
Genetic Algorithms

Comments are closed.