Accelerating material discovery by smart high-throughput computations

July 3, 2019 to July 5, 2019
Location : University of Liverpool


  • Dmytro Antypov (University of Liverpool, United Kingdom)
  • Linjiang Chen (Department of Chemistry, University of Liverpool, United Kingdom)
  • Kim Jelfs (Imperial College London, United Kingdom)
  • Edward Pyzer-Knapp (IBM Research UK, United Kingdom)


Leverhulme Research Centre for Functional Materials Design



The complexity and the multi-scale nature of the relationships between material's structure and its properties makes the design of a material with given properties one of grand scientific challenges of our times. This workshop brings together experts across different material classes united by the common goal of accelerating materials discovery to discuss methodologies that enable fast and accurate computations of material structures and properties and facilitate high throughput in silico screening.

Predictive calculations of material structure and properties have successfully been applied to a range of materials including cathode materials for batteries [1], organic photovoltaics [2], redox-active frameworks [3], metal oxides [4], and porous solids [5]. Different classes of materials pose different challenges associated with predicting their structures and properties but there are some commonalities. When designing new materials, do we start with existing materials and try to optimise their performance by small chemical changes or do we let the new structures emerge from the interactions of the substituents? How do we construct the search space to conduct the high throughput calculations? What computational chemistry methods do we use to get the best balance between speed and accuracy? How do we use hypothetical structures alongside reported materials to make the most of the existing knowledge? How do we utilise the information stored in databases to inform the choices for future research? How do we extract and process information from the high throughput calculations?

There is a constant improvement in the predictive power and efficiency of accurate but computationally demanding methods such as Density Functional Theory (DFT). Therefore, a number of approaches that are computationally less demanding but still make useful predictions are widely used in different forms including classical forcefields and machine learning potentials that work as a substitute to the quantum mechanics formalism [6]. Fast and accurate energy calculations are at the heart of all high-throughput calculations, whether they are related to crystal structure prediction or to the evaluation of the material properties. Here machine learning approaches offer an alternative top-down approach in which the material’s structure [7] and its functional properties [8] may be deducted from the existing knowledge (databases) or/and computed characteristics related to materials performance.

The progress in computational techniques and increasing available computer power facilitate the use of high-throughput methods for materials design. This in turn contributes to better understanding of structure-property relationship and allows one to optimise function [6]. By their nature, high-throughput methods generate a lot of data that can be used for data mining, statistical analysis, and machine learning algorithms to improve our understanding and identify new physical trends. The ability to calculate and verify experimentally the properties promotes an increasing synergistic collaboration between experiment and theory in materials design leading to unique mutual feedback circles that accelerate the discovery of new materials.



[1] Ceder, G. et al., Nature 392, 694 (1998)
[2] Hachmann, J. et al., J. Phys. Chem. Lett. 2, 2241 (2011)
[3] Sharma, V. et al., Nat. Commun. 5, 4845 (2014)
[4] Dyer, M.S. et al., Science 340, 847 (2013)
[5] Pulido, A. et al., Nature 543, 657 (2017)
[6] Smith, J.S. et al., Chem. Sci. 8, 3192 (2017)
[7] Ryan, K. et al., J. Am. Chem. Soc., DOI: 10.1021/jacs.8b03913 (2018)
[8] De Luna, P., Nature 552, 23 (2017)