Heterogeneous catalysis is heavily used in industry and is pivotal in our efforts to make chemical processes greener and more efficient. Direct experimental identification of the catalytically active sites and reaction mechanisms in heterogenous catalysts is challenging, hindering the ability to design catalysts for particular application [1]. Our recent work [2-3] on microporous zeolitic catalysts using biased ab initio molecular dynamics highlighted the fluctional nature of the catalyst surface, challenging the conventional view of these catalysts as being static under solvation. However, achieving such computational results for realistic catalytic systems is either extremely costly using ab initio treatments only or is plainly unattainable due to unsuitability/unreliability of available empirical force fields. These are the reasons for the continuing Maddox “scandal”, i.e., that it still remains impossible to predict the structure of even the simplest crystalline solids from first principles based only on a knowledge of their chemical composition.

Therefore, the methodological objectives of the project are:

• Develop and/or fine-tune general machine learning potentials (MLPs), that will accelerate exploration of catalyst configurational space by few orders of magnitude while retaining ab initio accuracy. This will include the adoption and development of various active learning strategies, allowing for continuous improvement of MLPs, as well as deployment of the delta-learning approach for improvement of the MLP accuracy for the systems and/or reactions of interest. Importantly, we have already made a significant progress along this direction [4-5], providing important use cases on how this aim can be realized.
• Develop and fine-tune data-driven bias-free dimensionality reduction schemes in which the good collective variables accelerating rare events, i.e., chemical reactions, will be selected automatically. Starting with a few known configurations (final catalyst, initial reaction mixture or some stable long-living intermediate), this scheme will allow us to effectively explore optimal low-dimensional representation of catalyst configuration space focusing on the most likely (trans)formation paths. It will enable routine open-ended searches of complex (trans)formation in catalytic systems resolving metastable configurations and overcoming the barriers between them. Again, we have already made a significant progress along this direction [6-7], providing important use cases on how this aim can be realized.

These new tools will enable predicting structures of catalysts from a knowledge of their chemical composition and synthesis conditions. The successful candidate will be able to build on accumulated knowledge base in our group both with respect to development of machine learning potentials and generation of data-driven collective variables.

The main application side of the project is to showcase this general-purpose approach on an industrially extremely important class of catalysts with a particularly challenging configurational space, the zeolites. The grand application challenge will be the realistic atomic-level modelling of zeolite synthesis, a long-standing unresolved problem with many competing hypotheses. By charting the configurational space of transformation process from reaction mixture to a zeolite framework the successful applicant will not only settle the long-standing debate but will open new routes to novel or improved zeolitic materials by modifying existing synthetic procedures.

Publications of research group related to this topic:

[1] Grajciar, L., Heard, C.J., Bondarenko, A.A., Polynski, M.V., Meeprasert, J., Pidko, E.A., Nachtigall, P. (2018): Towards Operando Computational Modeling in Heterogeneous Catalysis. In: Chemical Society Reviews (IF=40.1), 22, 8307-8348

[2] Heard, C.J., Grajciar, L., Rice, C.M., Pugh, S.M., Nachtigall, P., Ashbrook, S., Morris, R.E. (2019): Fast Room Temperature Lability of Aluminosilicate Zeolites. In: Nature Communications (IF=17.0) doi: 10.1038/s41467-019-12752-y.

[3] Jin, M., Ravi, M., Lei, C., Heard, C.J., Brivio, F., Tosner, Z., Grajciar, L.,* van Bokhoven, J. A., Nachtigall, P., (2023): Dynamical Equilibrium between Brønsted and Lewis Sites in Zeolites: Framework-Associated Octahedral Aluminum. In: Angewandte Chemie International Edition (IF=16.8) doi:.1002/ange.202306183.

[4] Erlebach, A., Nachtigall, P., Grajciar, L., (2022): Accurate large-scale simulations of siliceous zeolites by neural network potentials. In: npj Computational Materials (IF = 12.3), doi: 10.1038/s41524-022-00865-w.

[5] Erlebach, A., Šípka, M., Saha, I., Nachtigall, P., Heard, C.J., Grajciar, L., (2024): A reactive neural network framework for water-loaded acidic zeolites. In: Nature Communications (IF = 17.0), doi: 10.1038/s41467-024-48609-2.

[6] Šípka, M., Erlebach, A., Grajciar, L., (2023): Constructing Collective Variables Using Invariant Learned Representations. In: Journal of Chemical Theory and Computation (IF = 5.8), doi: 10.1021/acs.jctc.2c00729.

[7] Šípka, M., Dietschreit, J.C.B, Grajciar, L., Gomez-Bombarelli, R. (2023): Differentiable Simulations for Enhanced Sampling of Rare Events. In: Proceedings of the 40th International Conference on Machine Learning, 202:31990-32007.

Existing collaborations related to the topic:

1. Rafael Gomez-Bombarelli (MIT, US) - supported by the MIT Global Seed Fund
2. Ben Slater (UCL, UK) - supported by the Johnson Matthey company
3. Sharon Ashbrook and Paul Wright (St. Andrews University, UK) - supported by the Johnson Matthey company
4. Alexander O'Malley (University of Bath, UK)
5. Tereza Blasco (ITQ, Valencia, Spain)
6. Joachim Sauer (Humboldt University, Berlin, Germany)

Qualifications:
Required - PhD. in Chemistry, Physics, Material Science or a related field; good knowledge of English; experience in programming (ideally Python or similar)

Advantageous, but nor required - background in Machine Learning, Statistics, Statistical Mechanics and Quantum Chemistry/Physics; experience with molecular simulations, high-performance computing (including GPU accelerated one) and Linux.

Salary: co-founding 1000 EUR/month is ensured

Co-founding resources: Department of physical and macromolecular chemistry budget

Department: Department of physical and macromolecular chemistry

Supervisor: Assoc. Prof. Lukáš Grajciar

E-mail: lukas.grajciar@natur.cuni.cz

Phone: +(420) 221 95 1298

Position available from: January 1, 2025

Deadline date for applications: 24th July, 2024

Applicants must submit required documents to:  lukas.grajciar@natur.cuni.cz (project supervisor)

and in a copy to pavla.pouskova@natur.cuni.cz (International Department)