Measure what’s easy to measure, calculate what’s easy to calculate, use AI to fill in gaps

Felix Hanke

Biovia R&D, Dassault Systems UK Limited , United Kingdom


Felix Hanke is a Senior Scientific Software Developer and Fellow of the BIOVIA Science Council at Dassault Systèmes.

He primarily focusses on developing workflows to model chemical reactions, reaction networks, and battery materials from first principles.

Felix completed his doctorate in statistical physics and did postdoctoral work in DFT development and surface science before joining Dassault Systèmes BIOVIA (then Accelrys) in 2012.


Measure what’s easy to measure, calculate what’s easy to calculate, use machine learning to fill in gaps

Successful materials modelling brings value by enabling scientists to make predictions for materials and properties that are intractable or more expensive to measure experimentally. The same goes for physics-informed machine learning models, which should ideally be built on top of both experimental and computational data. This basic philosophy creates a number of important requirements on software implementations of simulations, specifically to minimize the time user needs on any given simulation and irrespective whether they are experts or democratized non-specialist users.

Here I will outline how these concepts work in practice for simulating chemical reactions in green catalysis and in Lithium ion batteries, using recent developments in the BIOVIA software portfolio. First, I will look at the application of BIOVIA FlexTS to determine a catalytic mechanism for acetic acid hydrodeoxygenation on a novel Pt-Mo catalyst as a model system for generating renewable fuels from biomass. In a second example, the underlying automatic method to calculate reaction pathways and rates is used to calculate the molecular level degradation mechanism for an electrolyte in Lithium ion batteries. Here we also outline the ReactionFinder method used to integrate this molecular mechanism into classical molecular dynamics, which enables us to follow the evolution of the battery electrolyte degradation over long time and length scales. Finally, I will outline an approach to build a Materials Acceleration Platform, e.g. a collaborative simulation and experimental framework in which a machine learning optimizer is used to autonomously optimize the formulation for a battery electrolyte.