Physics Colloquium: Miguel Marques (University of Halle, Germany) - "The second computer revolution in materials science: from density-functional theory to machine learning"
|When:||Th 24-01-2019 16:00 - 17:00|
|Where:||5111.0080 (Nijenborgh 4)|
The development of density-functional theory in the 1960s and the dissemination of computers led to a revolution in materials science. A third kind of physics, computational physics, emerged to complement its theoretical and experimental sisters. By solving complex theoretical models in a computer we had access to quantitative results for specific systems. These numerical experiments could explain experiments or be used to predict new materials and their properties. Nowadays, with the availability of ever faster supercomputers and novel computer methodologies, we are living what I would call the second computer revolution in materials science. High throughput techniques, together with ever faster supercomputers, allow for the automatic screening of thousands or even millions of hypothetical materials to find solutions to present technological challenges. Moreover, machine learning methods are used to accelerate materials discovery by replacing density-functional theory by extremely efficient statistical models.
In this talk I summarize our recent attempts to discover and characterize new materials using these novel approaches. I will start by motivating why the search for new materials is one of the most pressing problems nowadays. This is true not only to enable new technological advances, but also to circumvent availability, environmental, economical, and political problems related to many chemical elements. The strengths and weaknesses of high-throughput approaches are then discussed in the context of perovskites, and in particular for applications in photovoltaics. I will then show how relatively simple approaches, either based on data-mining or machine learning, can be used to speed up this process. These techniques turn out to be rather intuitive, as they "learn" chemical rules by gaining experience from a relatively small set of data. Finally, I speculate which role machine learning will have in the future of materials science.