Substantial H2020 funding has just been secured for a European multi-center road map consortium on battery research (BIG-MAP).  https://www.big-map.eu/

Within this framework, the research groups of Kersti Hermansson (Uppsala University), Gabor Csanyi (University of Cambridge), and Anatole von Lilienfeld (University of Vienna) will collaborate on the development of first principles methods tailored towards their application to battery relevant materials (electrodes, electrolytes, and interfaces).   

To perform this work we are looking for three talented postdoctoral candidates with suitable background and ambition who would be interested to work with us on either one of the three following aspects of this work for three years, respectively, starting Sep 1 2020. If interested, please contact the corresponding PI with a cover letter, CV, and 3 references during the summer of 2020.

1. Kersti Hermansson, Department of Chemistry-Ångström Laboratory, Uppsala University, Sweden

   (e-mail: [email protected])

In the Computational Materials Chemistry group of the Structural Chemistry Program at Uppsala University, we will explore and assess the applicability of various quantum mechanical (QM) approximations to battery interface systems. We will, among other things, create databases of consistent DFT data, suitable for modelling the potential energy surface of the constituent molecules and solids. We will set up, design, and perform QM calculations of varying cost efficiency and accuracy for simulations of equilibrium and redox properties of a broad selection of compounds from materials classes relevant for the BIG-MAP's targets. The resulting data will serve as input to the project tasks described below for the Cambridge (Csanyi) and Vienna (von Lilienfeld) activities in order to work toward a closed-loop scheme: QMFFQML. 

More specifically, the undertaken task involves extensive method assessment to identify the most appropriate (sufficiently accurate and most cost-effective) level of QM treatment, including semi-empirical schemes such as tight-binding calculations as validation tests for generated FFs. For example, the role of van der Waals interactions present in molecular clusters of the molecular electrolyte candidates. The workflows will typically consist of generating systematic trends among various levels of theory by considering relevant equilibrium and non-equilibrium structures. Resulting data will continuously be fed into the other tasks (see below), and feedback from both of these tasks will be used in order to define additional systems for subsequent QM treatment. 

The applicant is expected to have a solid background in quantum chemistry modelling of materials and molecules as well as a keen interest in acquiring awareness of the capabilities of ML in the context of the tasks led by the Cambridge and Vienna groups. 

Collaboration with groups at CNR (Italy), EPFL (Lausanne) and CTH (Göteborg) will will also be part of this sub-project.  

2. Gábor Csányi, University of Cambridge

The task of the Cambridge group is to create molecular force fields for electrolyte compounds based on data supplied from Uppsala, and to compute some properties using molecular dynamics that are feed forward to the Vienna group and others within the larger BIGMAP project. The approach for force field fitting is rooted in machine learning ideas, initially using the Gaussian Approximation Potential (GAP) framework, but over the course of the project, new formulations based on polynomial regression will also be tried. Apart from the molecular force fields themselves, the electrolyte-electrode interfaces will also be considered, and the force fields extended to include the solid-molecule interactions, something that has been traditionally very difficult with conventional force fields and potentials, and thus the capabilities of the general machine learning force field framework will be exploited to their full extent. 

Special attention will be given to the intermolecular part of the force field, using a hierarchy of approaches:  multiscale descriptors, electrostatic baseline models, explicit machine learning of electrostatic and vdW parameters will all play a part. 

The ideal applicant will significant experience in molecular modelling. Experience in machine learning, particularly in the materials/chemistry space, would be useful, but not essential. Similarly, experience in fitting conventional molecular force fields would be very beneficial, but not essential. 

3. Anatole von Lilienfeld, University of Vienna

In the newly established lab of Anatole von Lilienfeld at the University of Vienna, we will develop new quantum machine learning models which can serve the ranking and selection of new promising battery candidate systems. This will be done by studying and developing improved and purpose-tailored descriptors and representations, which capture the physics of those degrees of freedom which govern battery relevant properties. We will use these developments to train, test, and apply computationally efficient and transferable ML models which are applicable across many and various materials, the representative training instances being obtained from simulations with the new machine learning potentials (Csanyi group), as well as from the QM calculations (Hermansson group). In particular, we plan to consider, the chemical space of small organic electrolyte candidates (functionalizing derivatives of ethylene and propylene carbonates, the effect of doping on Li at Ni enriched cathode materials (such as Ni-oxides), as well as chemical modifications to Silicon-graphite based anode materials. After training, we will use the the quantum machine learning models to explore said materials compound spaces with well-established optimization algorithms (Monte Carlo, simplex, genetic algorithm, etc.) in order to converge towards ranked sets of promising new materials candidates with expected optimal property combinations. Our results will subsequently loop back to inform work in the Hermansson and Csanyi groups about new promising input materials worthy of further study. Collaborations with the Danish Technical University will also be part of this sub-project.