Predicting optical and vibrational spectra for molecules from ML-enhanced DFT

PhD n°9


EU mobility rules apply. In principle, applicants can have any nationality and any current residence (although immigration rules apply, favoring
EU applicants). Candidates who have already been awarded a PhD degree are not eligible. In addition, candidates who have already spent more than 12 months in the Netherland within the last 3 years are not eligible (unless as part of a procedure for obtaining refugee status under the Geneva Convention).


The Researcher at SCM will develop and apply new methods to apply machine-learning (ML) techniques to predict spectroscopic properties of molecules. Initially we will focus on vibrational (such as IR) and optical (UV-Vis) properties, with possible later extensions to magnetic and chiral
properties. These properties are currently calculated with DFT in ADF and, for some, with the faster but more approximate DFTB method.
The new methods will either focus on direct prediction of spectra, bypassing DFT altogether, or focus on hybrid DFT (or DFTB) / ML methods to improve current spectroscopic predictions in either speed or quality. This will be done by delta-ML methods to train a neural network on the difference of Fock matrix in the ground state for DFT (in some suitably localized basis set) and DFTB as well as on matrices related to perturbed 1st-order Fock matrices relevant for response properties. The ML methods will also be developed by using the expertise in Optimal Transport (OT) theory of the Theoretical Chemistry VUA, for example by using its distance metric to define differences between spectra.

  • Develop methods, to be implemented in SCM’s software suite, for improved spectra prediction of molecules, using ML and OT, starting from geometry and/or fast DFTB-based spectra
  • Develop ML-based methods inside SCM’s DFT(B) codes to improve the matrices needed to calculate the spectra

Expected Results:

  • ML-based method to predict optical and vibrational spectra for molecules directly from geometry
  • Improved DFTB molecular spectra by training difference between DFT and DFTB matrices

Planned secondment(s): 6 months

  • Intersectoral (academic):
    • AALTO, P. Rinke, M10-M12: training on ML techniques.
    • JENA, S. Botti (also visiting MLU), M20-M22: joint application of vibrational spectra methods on interfaces.

Enrolment in Doctoral degree(s):

Vrije Universiteit Amsterdam


Stan van Gisbergen

Download the full description of this position (pdf)

PhD n°: PhD n°9
Country: Netherlands
This job is no longer accepting applications.