Technical University of Denmark (DTU)

Three postdoc positions in computational chemistry and machine learning for molecular science

Technical University of Denmark (DTU)

2800 Kgs. Lyngby, Denmark

The Sections for Atomic Scale Materials Modelling at DTU Energy and Cognitive Systems at DTU Compute, Technical University of Denmark (DTU), are looking for outstanding candidates for three 2-year postdoc positions within the fields of computational chemistry and machine learning for molecular science. The research positions are part of the Novo Nordisk Foundation Exploratory Interdisciplinary Synergy Programme: Self-correcting Unsupervised Reaction Energies (SURE), which brings together researchers from DTU Energy and DTU Compute.

Project descriptions
The successful candidates will use electronic structure modelling (e.g. DFT and wave function methods) and machine learning algorithms to develop a framework for uncertainty-aware prediction of chemical reaction networks. The framework development will consist of the following postdoc projects:

  1. DFT and wave function electronic structure methods will be used to calculate high-fidelity data for the thermodynamics and kinetics of selected chemical reactions. The developed methodology will be used to train the data-driven models developed in the parallel projects, as well as to validate them for the prediction of degradation reaction networks of organic electroactive molecules used in redox flow batteries. The ideal candidate will have experience in modelling of reaction mechanism of molecular systems.
  2. Molecular graph operation based methods will be established and used for reaction intermediate and product candidate generation. Machine learning predicted energies of these structures will help us create probabilistic model of the reaction networks. Furthermore, new tools will be developed to provide uncertainty guided analysis on reaction products and mechanisms. Experience in scientific programming (e.g. Python) and Bayesian statistics will be advantageous.
  3. Graph convolutional neural network models will be built for evaluating errors and uncertainties in molecular energies obtained from electronic structure simulations of varying complexity. Model training methods that can utilize multi-fidelity data will be incorporated. The developed framework will be utilized to predict energies for any given molecular structure along with uncertainties, to build probabilistic models for reaction networks. Experience in machine learning model development is expected.
The three projects will be carried out in close collaboration between the two sections and linked to other ongoing projects in the sections working on clean energy materials and machine learning for accelerated materials discovery.

Candidates should hold a PhD or equivalent degree in computer science, physics, chemistry or materials science. The candidate must have a strong background in computational chemistry, physics or materials science and/or machine learning, and are expected to have performed original scientific research within the relevant fields listed above for the specific position(s). Moreover, the successful candidate:
  • is innovative and able to work both independently and in cross-disciplinary teams
  • has good communication skills in English, both written and spoken
  • is able to work independently and take responsibility for progress and quality of projects.

Further information
If you need further information concerning these positions, please contact Prof. Tejs Vegge at or Professor Ole Winther at

Please do not send applications to these e-mail addresses, instead apply online as described below.

We must have your online application by 20 January 2020.

To view the full announcement and to apply:

Please apply via recruiter’s website.