"Machine learning in chemistry and beyond" (ChE-651) seminar by Miguel Caro "Machine learning local atomic properties: accurate prediction of XPS spectra of carbon materials"


Event details

Date 10.05.2022 15:1516:15  
Speaker Miguel Caro is originally from a small town (Cartaya) in southwestern Spain. He graduated with a Physics degree from University of La Laguna, Tenerife, Spain. He then moved to Cork, Ireland, where he pursued a PhD in computational condensed-matter physics under Prof. Eoin O’Reilly at the Tyndall National Institute. His thesis work, for which he was awarded his PhD in 2013, focused on theory of III-N alloys, a material system widely used for optoelectronic applications. After the PhD, he moved to Aalto University, Finland as a postdoc in 2013. In 2017 he obtained the Academy of Finland Postdoctoral Researcher grant and since 2020 he is Academy of Finland Research Fellow. Dr. Caro’s current research interests concern the atomistic simulation of real materials, especially carbon-based materials, using a battery of simulation tools and methodologies, from density functional theory to machine learning.
Location Online
Category Conferences - Seminars
Event Language English

n recent years great advances have been made in the development of machine learning based force fields. These force fields feed on reference DFT total energies and forces and, once trained, can make predictions from the knowledge of the atomic positions alone. Less attention has been paid to predicting other atomistic properties of materials, such as spectroscopic features. Fortunately, the same methodological and computational tools that have been developed for force fields are very well suited to learn a variety of "local" atomic properties. In this presentation I will talk about these local property models and discuss how we have employed them to learn adsorption energies [1], Hirshfeld volumes used to parametrize van der Waals corrections [2], and core-electron binding energies [3]. I will particularly focus on the latter, and show how these local models, trained from a combination of multilevel reference data, can be used to efficiently and accurately predict XPS spectra of complex materials.

[1] M.A. Caro, A. Aarva, V.L. Deringer, G. Csányi, and T. Laurila. Chem. Mater. 30, 7446 (2018).
[2] H. Muhli, X. Chen, A.P. Bartók, P. Hernández-León, G. Csányi, T. Ala-Nissila, and M.A. Caro. Phys. Rev. B 104, 054106 (2021).
[3] D. Golze, M. Hirvensalo, P. Hernández-León, A. Aarva, J. Etula, T. Susi, P. Rinke, T. Laurila, and M.A. Caro. arXiv preprint arXiv:2112.06551.

Practical information

  • Informed public
  • Free


  • Kevin Maik Jablonka, Solène Oberli, Puck van Gerwen


  • Kevin Maik Jablonka, Solène Oberli, Puck van Gerwen