University of Cambridge > > Theory of Condensed Matter > Observe, learn, design: machine learning in physics and materials science

Observe, learn, design: machine learning in physics and materials science

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Bartomeu Monserrat.

This talk has been canceled/deleted

The standard paradigm for discovery in physics and materials science is based on an interplay between laboratory experiments and the formulation of theories, which, depending on their type and complexity, can be solved by pen and paper or through computational studies. Recently much attention has been dedicated to a third way, namely to the use of machine-learning and artificial-intelligence methods to learn from existing data our physical and materials world. In this talk I will provide a tutorial overview of such new paradigm and present examples of different complexities.

First, I will show how one can use machine learning to predict a complex thermodynamical quantity, such as the Curie temperature of ferromagnets [1]. In particular I will discuss how to develop meaningful feature attributes for magnetism and how these can be informed by experimental results only. The model presented allows one to forecast the magnetism critical temperature of a material from the sole knowledge of simple chemical information. Some discussion on how experimental data can be retrieved, curated and utilized will be provided.

Then, I will discuss how the atomic structure can be represented in a way, which is amenable to machine learning. This is a non-trivial problem, affecting the outcome of any model aiming at establishing structure- to-property relations. I will show examples in which a many-body, but short-range, representation of the atomic distribution enables us to construct ultra-accurate force fields, agnostic of the nature of the chemical bond [2]. These will be used to describe entire phase diagrams and materials under extreme conditions [3]. Furthermore, the same representation will be used to predict electronic quantities and to perform global materials optimization over a vast phase space [4].

Finally, I will show how the competition between different artificial-intelligence algorithms can be exploited to generate new structures according to known laws of chemistry. This is the case of the so called generative models, largely used in image processing, but only scarcely exploited in materials science.

[1] James Nelson and Stefano Sanvito, Predicting the Curie temperature of ferromagnets using machine learning, Phys. Rev. Materials 3, 104405 (2019).

[2] Alessandro Lunghi and Stefano Sanvito, A unified picture of the covalent bond within quantum-accurate force fields: from simple organic molecules to metallic complexes reactivity, Science Advances 5, eaaw2210 (2019).

[3] YanhuiZhang, Alessandro Lunghi and Stefano Sanvito, Pushing the limits of atomistic simulations towards ultra-high temperature: a machine-learning force field for ZrB2, Acta Materialia 186, 467 (2020).

[4] Alessandro Lunghi and Stefano Sanvito, Surfing multiple conformation-property landscapes via machine learning: Designing magnetic anisotropy, arXiv:1911.02263 (2019).

This talk is part of the Theory of Condensed Matter series.

Tell a friend about this talk:

This talk is included in these lists:

This talk is not included in any other list

Note that ex-directory lists are not shown.


© 2006-2020, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity