Seminar by Hartmut Maennel: Covariants for 3D Point Configurations

© 2025 EPFL

© 2025 EPFL

Join us for a seminar by Hartmut Maennel on Monday, February 10th, at 13:30 PM in MED 2 1124 (Coviz2). Hartmut is a researcher at Google DeepMind working on Covariants for 3D Point Configurations with applications to learning molecular quantum properties.

When modeling physical properties of molecules with machine learning, incorporating SO(3)-covariance is highly desirable. In this talk, Hartmut will discuss how traditional low body order features fail to achieve completeness, and formulate and prove general completeness properties for higher order methods and show that 6k–5 of these features are enough for up to k atoms. In addition, he will discuss how the Clebsch–Gordan operations commonly used in these methods can be replaced by matrix multiplications without sacrificing completeness, lowering the scaling from O (l6) to O (l3) in the degree of the features. While the focus will be on quantum chemistry applications, the methods presented can be applied to a wide range of problems involving three-dimensional point configurations.

[1] Hartmut Maennel, Oliver T. Unke, and Klaus-Robert Müller. "Complete and Efficient Covariants for Three-Dimensional Point Configurations with Application to Learning Molecular Quantum Properties." The Journal of Physical Chemistry Letters 15, no. 51 (2024): 12513-12519.