Klasifikace fází bez dozoru pro kvantové systémy

• Název práce v češtině: Klasifikace fází bez dozoru pro kvantové systémy
• Název v anglickém jazyce: Unsupervised phase classification for quantum matter
• Akademický rok vypsání: 2024/2025
• Typ práce: disertační práce
• Ústav: Katedra fyziky kondenzovaných látek (32-KFKL)
• Vedoucí / školitel: RNDr. Martin Žonda, Ph.D.
• Konzultanti: RNDr. Pavel Baláž, Ph.D.

Zásady pro vypracování

We offer an opportunity to learn, implement and develop machine learning techniques for phase classification in the context of condensed matter physics. More specifically, for systems of correlated electrons and quantum lattice spin models and materials. We will propose to the interested candidate several open physical questions from which she or he can choose one that will anchor the main physical theme. Some of these open questions, e.g., problems related to frustrated magnetic systems or neuromorphic computing, offer an opportunity to collaborate with local experimental groups.

The candidate should have a master's or engineering degree in physics, physical chemistry, electrical engineering, material science or any related topic. We expect a solid knowledge of quantum mechanics, statistical physics, and basics of material science. Further, the candidate should like to code and be an experienced user of at least one of the following programming languages: Python, Julia, Matlab, Fortran 90 (or newer), C++. Experience with machine learning is an advantage, but it is not a requirement.

Seznam odborné literatury

[1] Julian Arnold, Frank Schäfer, Martin Žonda, and Axel U. J. Lode, Phys. Rev. Research 3, 033052
(2022)
[2] A high­bias, low­variance introduction to Machine Learning for physicists, Pankaj Mehta et al.,
Physics Reports 810 (2019) 1­124
[3] Physicist's Journeys Through the AI World, Imad Alhousseini et al., arXiv:1905.01023

Předběžná náplň práce

The classification of phases of a quantum matter, i.e., materials which need to be described by quantum mechanics, is a crucial task for their understanding and a key step towards their usability in future applications. Here the point is, that although typical condensed matter many­body systems
have an enormous number of degrees of freedom, their stable phases can often be characterized by a small set of physical quantities, e. g., order parameters or correlation function. Yet the material or quantum device can have radically different properties in each of its phases. However, the identification and classification of phases is a complex problem with a large state space that often involves tedious procedures. Naturally, it gets even worse when it is not clear what type of phases, how many of them and where in the parameter space one should look for them, which is a
typical case in basic research.

Recent years showed that Machine­learning methods can be a remedy for these problems. They can deal with large data sets and can efficiently extract information from them. More importantly, various unsupervised techniques can be applied on row data without prior knowledge about the character of the phase diagram or its components. Besides simplifying the entire process of the phase classifications this opens a possibility to identify new phases previously missed by the researchers. In this context is especially important the recent progress in making machine learning classification interpretable.

We offer an opportunity to learn, implement and develop these techniques in the context of condensed matter physics. More specifically, for systems of correlated electrons and quantum lattice spin models and materials. We will propose to the interested candidate several open physical questions from which she or he can choose one that will anchor the main physical theme. Some of these open questions, e.g., problems related to frustrated magnetic systems or neuromorphic computing, offer an opportunity to collaborate with local experimental groups.