Polaron pattern recognition
in correlated oxide surfaces
Subproject P07
The formation of polarons by charge trapping is pervasive in transition metal oxides. Polarons have been widely studied in binary compounds but comparatively much less so in perovskites.
In P07, we aim to combine advanced first-principles approaches with computer-vision and machine-learning techniques to accelerate and automatize the study of polarons and novel polaron effects in transition metal oxides.
We address and provide solutions for static and dynamical polaron properties by implementing ML and computer vision approaches to accelerate the exploration of the multi-polaron configurational space and to extend small polaron dynamics to the nanoscale. We integrate charge state encoding into the atomic features and forecast the dynamic evolution by predicting the occupation matrix. These methodological developments will be integrated in our software Leopolad (Learning of polaron dynamics) based on an equivariant graph neural network framework.
Addressing ML-augmented charge (polaron) dynamics is essential for advancing our understanding of complex materials and phenomena, and for fully leveraging ML-assisted MD. By expanding DFT capabilities with ML algorithms, we aim to extend the simulation of polaron dynamics to the nanoscale. This advancement will enable us to uncover novel effects, such as the dynamical interaction of surface polaron with adsorbates.
The advancement of ML-assisted polaron-MD will significantly contribute to the key methodological developments of the SFB, particularly in predicting multivalence states in oxides.
Cesare Franchini
PI
Expertise
Theoretical and computational modeling of quantum materials, in particular transition metal oxides in bulk phases and surfaces, to predict and interpret novel physical effects and states of matter arising from fundamental quantum interactions: electron-electron correlation, electron-phonon coupling, spin-spin exchange, spin-orbit coupling, to name the most relevant ones. The theoretical research is conducted in strong synergy and cooperation with experimental groups.
Methods:
- Density functional theory, hybrid functionals, GW, BSE
- First principles molecular dynamics
- Effective Hamiltonian
- Diagrammatic quantum Monte Carlo
- Dynamical mean-field theory
- Machine learning and computer vision
Applications:
- Polarons: formation, dynamics, polaron-mediated effects, many-body properties
- Computational surface science: energetics, reconstructions, surface polarons, polarity effects, adsorption and chemical reactions
- Quantum magnetism: all-rank multipolar spin-spin interactions beyond Heisenberg exchange
- Electronic and magnetic phase transitions
Our goals in TACO:
- Accelerated study of polaron properties by integrating molecular dynamics and machine learning methods (kernel-ridge regression, standard and convolutional neural-networks
- Implementation of automated identification of local structures in atomically resolved images using computer vision methods
- Complementing the experimental measurements with extensive first principles modeling of perovskite surfaces.
Team
Cesare Franchini
PI
Johanna Paulina Carbone
Co-PI
Daniela Mangano
PhD Student
Alex Kortstee
PhD Student
Associates
Matthias Meier
Postdoc
Luca Leoni
PhD Student
Darin Joseph
PhD Student
Former Members
Michele Reticcioli
Co-PI
Viktor Birschitzky
PhD Student
Marco Corrias
PhD Student
Florian Ellinger
PhD Student
Publications
2022
Modeling polarons in density functional theory: lessons learned from TiO2
Journal ArticleOpen AccessIn: Journal of Physics: Condensed Matter, vol. 34, no. 20, pp. 204006, 2022.
2021
Rapid oxygen exchange between hematite and water vapor
Journal ArticleOpen AccessIn: Nature Communications, vol. 12, iss. 1, no. 6488, 2021.