This page lists the current PhD or PostDoc openings in our group. If you are interested in working with us, but no openings matching your profile, you are still welcome to submit a general inquiry.
In any case, please consider the application guidelines before you contact us.
In inverse materials design one wishes to discover novel materials in a targeting fashion – namely by systematically traversing a design space for those material structures, which best match the desired material properties. Commonly such methods employ a statistical surrogate (e.g. within a Bayesian Optimisation framework) such that the search only requires few queries to an underlying physical model – typically a first-principle model based on density-functional theory (DFT). Due to the non-linear nature of DFT such simulations are not only costly, but advanced techniques based on employing gradient information are generally not employed. However, the advent of algorithmic differentiation (AD) techniques in DFT codes, such as our in-house density-functional toolkit (DFTK), makes it now feasible to employ gradient-based approaches for materials design.
Within this project we will explore the opportunities of gradient-based Bayesian optimisation to accelerate inverse materials design. In particular we will rely on the AD capabilities of DFTK to integrate with recent advances with first-order Bayesian optimisation procedures (e.g. [1]). Your work will be integrated with similar efforts in our group, e.g. to develop multi-task Gaussian Process surrogates[2] for materials modelling or to exploit analytical error estimates within statistical surrogates. Within ongoing collaborations with other materials simulation groups at EPFL your advances can be directly developed and tested within the scope of practical materials modelling problems.
You are motivated to tackle a challenging interdisciplinary research topic and push the state of Bayesian Optimisation methods for inverse materials design.
You obtained your PhD in statistics, computational mathematics or a related subject.
You have an excellent academic track record and demonstrated prior research experience in Bayesian optimisation, experimental design or inverse problems with an application in physics or engineering simulations.
You enjoy collaborating with researchers from a diverse background and look forward to acquiring the broad skillset required for cross-disciplinary research in materials modelling.
You are experienced in working with larger scientific codes in a collaborative software development environment. You have a solid experience with the Julia programming language or you are fluent in a related language (Python, Matlab) and are curious to code in Julia.
You are fluent in written and oral English.
You enjoy occasionally supervising undergraduate students on topics related to your research.
Bonus skills for this application are considerable experience in quantum-chemical or materials simulations or high-performance computing.
The activities of the MatMat group revolve around understanding modern materials simulations from a mathematical point of view – and to come up with ways to make such simulations faster and quantify their errors. You will become part of a young and energetic team, fully integrated with both the mathematics and the materials institutes as well as multiple cross-disciplinary initiatives, such as the NCCR MARVEL. Within the proposed topic you will be able to bring in your prior expertise, but also be able to get to know the exciting theory and practice of material modelling. EPFL's main campus is beautifully located at the lake Geneva shore hosting a stimulating community of interdisciplinary-minded researchers. Funds to disseminate your work at suitable conferences as well as potential visits to our international network of collaboration partners are provided.
The position will be a fixed-term position (CDD) for initially 18 months. Further extensions depend on progress and the funding situation. For more information on working at EPFL see also the website on current employment conditions.
Screening of candidates is done on a rolling basis until a suitable candidate has been found. The expected starting date is the first half of 2025.
| [1] | J. Wu, M. Poloczek, A. Wilson, P. Frazier. Bayesian Optimization with Gradients. NeurIPS (2017). |
| [2] | K. Fisher, M. F. Herbst, Y. Marzouk. Multitask methods for predicting molecular properties from heterogeneous data. Journal of Chemical Physics (2024). arXiv:2401.17898 |
