项目介绍

2026年6月11日

A post-doctoral position is available in Physics-informed Generative Modeling and Multiscale Learning. The successful candidate will work at the interface of physics-based modeling and prob- abilistic machine learning, tackling problems that arise in molecular systems and heterogeneous materials. The research combines generative probabilistic frameworks with the mathematical struc- ture of the governing physics, addressing challenges that standard machine learning approaches cannot resolve due to high dimensionality, physical invariances, and the scarcity of data relative to the complexity of the systems involved.

In terms of Physical Modeling, this particular position requires experience in one or more of the following areas:

• Molecular dynamics and statistical mechanics of atomistic systems
• Continuum thermodynamics and numerical methods for PDEs, with applications to hetero- geneous and multiscale media

In terms of Computational and Data-driven Modeling, this particular position requires experience in one or more of the following areas:

• Variational inference and generative modeling (e.g. normalizing flows, VAEs, diffusion models)
• Numerical solution of (stochastic/deterministic) PDEs/ODEs, including multiscale and ho- mogenization methods
• Fundamentals of probability and statistics with preference given to candidates with experi- ence in Bayesian formulations
• Scientific programming (independently of language)

The Professorship of Data-driven Materials Modeling is part of the Department of Engineering Physics and Computation in the School of Engineering and Design at the Technical University of Munich. The research efforts of the group target open problems in the predictive modeling of ma- terials and molecular systems: how does microstructure dictate macroscopic material behavior, how

can we design materials with targeted properties under uncertainty, and how do atoms collectively reorganize across vastly different timescales? Addressing these questions requires methods that go well beyond standard machine learning, embedding physical laws, symmetries, and probabilistic reasoning directly into the learning framework. Interested candidates should consult our latest work here.

What to expect:

• The postdoc will have significant autonomy in shaping their research agenda within the group’s thrusts, including the opportunity to develop independent research proposals and pursue their own scientific directions.
• There will be close daily interaction with Prof. Koutsourelakis and PhD students in the group.
• Opportunities to co-supervise PhD students and to collaborate with researchers at TUM and at partner institutions internationally.
• The position offers the opportunity to present work at leading international conferences and workshops.

Qualifications:

Candidates should be proficient in scientific computing and probabilistic modeling and should have a Ph.D. by the time of appointment in any Engineering discipline, Applied Mathematics/Statistics, Computational Physics, or Machine Learning. Experience in generative modeling, or multiscale numerical methods is very desirable.

Interested candidates should apply by emailing Prof. P.S. Koutsourelakis at the following address:

contmech@mw.tum.de

with the Subject: Postdoc Position — Physics-informed Generative Modeling and Multiscale Learning and include (in PDF format):

• A CV with the names of up to three references.

• A statement of research experience, interests and goals. This should describe specific in- stances of physics-informed machine learning that you have worked on. Generic state- ments that do not address these matters will result in the automatic rejection of the application without further consideration.
• Up to 3 indicative publications/preprints.