Computational Modeling of New Surface Catalysis Systems by Means of Ab-initio Methods as well as Novel Machine-Learning Force-Field Approaches

Catalysis at liquid interfaces (CLINT) covers different novel approaches to make chemical catalysis more effective and sustainable. Traditionally, there are two main branches in chemical catalysis. In homogeneous catalysis, the whole process takes place in a single phase, usually a liquid. The catalyst is a small, well-defined molecule, for example a metal complex. The advantage of this approach is a distinct and easy-to-tailor reaction mechanism, the disadvantage, however, is that the catalyst needs to be separated from the product after the process, which is tedious in most cases. In heterogeneous catalysis, on the other hand, catalyst and reactants are located in two separate phases. The catalyst, usually a metal surface, constitutes a solid phase, the reactants and products are in the gas phase. Thus, the separation of product and catalyst is easy, with the catalyst usually staying fixed in the reactor. There are, however, no well defined single catalytically active places anymore. Further, the solid surfaces tend to be successively poisoned and deactivated by unwanted side products.

CLINT aims at combining the advantages of both approaches. The catalyst is placed in a separate phase, like in heterogeneous catalysis, but contains well defined single-atomic reaction sites, like in homogeneous catalysis. This is achieved by either placing single catalytically active metal atoms like platinum in a liquid metal like gallium, which itself acts like a passive solvent (SCALMS), or by coating a metal surface with an ionic liquid film (SCILL). In both cases, the degree of poisoning is reduced significantly. In SCALMS, the active atoms act as well defined single reaction sites, in SCILL, the properties of the catalyst can be fine-tuned by the choice of the ionic liquid. 

While the simulation of homogeneous and heterogeneous catalysts is standard since many years, the simulation of SCALMS or SCILL systems is much less straightforward. There exist no isolated catalyst molecules or highly symmetrical metal surface facets, but instead a complex dynamical liquid system with significant electronic interplay of the components. Periodic DFT, applied to representative surface slabs containing 100-400 atoms, is the method of choice to cover all relevant electronic properties.  Describing the liquid, on the other hand, requires molecular dynamics (MD) simulations, which are very expensive with DFT. Therefore, we increasingly utilize ML-FFs especially for SCALMS systems to describe dynamical phenomena in the ns time scale and in systems containing 1000s to 10000s of atoms.

Methods:

Our main workhorse is periodic DFT, applied with the Vienna ab-initio simulation package (VASP) [1, 2]. We are using the PBE exchange-correlation functional [3] with the D3 empirical dispersion correction scheme [4, 5]. Apart from optimizing geometries and calculating, e.g., adsorption energies, we also use DFT to calculate the element-resolved density of states (DOS) or core level energies. DOS overlaps between adsorbates and catalytically active atoms can be used to predict reactivity trends, Bader partial charges are suited to explain electronic interactions and core level energies can be compared to experimental XPS spectra to assign the peaks and help identifying species. For SCILL systems, the calculation of infrared adsorption spectra and scanning tunneling microscopy (STM) images is also important.

Besides DFT, ML-FFs are crucial for our studies, since many relevant simulations like MD samplings over several ns with DFT are either extremely expensive (costing millions of CPU-hs per simulation) or directly impossible. We are mostly using the VASP ML-FF, which can be learned on-the-fly with periodic DFT [6, 7]. One starts a MD simulation with DFT, reference data is collected, and if enough data for the current part of the system’s configuration space has been sampled, the method switches to ML-FF to speed up the sampling by two orders of magnitude, until a new region of configuration space barely represented in the collected training data has been reached and it switches back to DFT. After learning, the ML-FF can be applied to much larger systems containing 1000s of atoms, where DFT cannot be used at all. Representative screenshots of ML-FF MD simulations can be taken to calculate DOS or core level energies with DFT, combining the strengths of both methods.

Results

We studied the element-resolved diffusion coefficients of liquid Ga and GaNi with ML-FF MD simulations, to characterize the dynamic behavior of SCALMS systems more exactly. We found out that adding 2% of Ni to liquid Ga almost halves the overall diffusion of the system, whereas a further increase to 10 % Ni almost had no change, both in agreement with neutron scattering experiments [8, 9]. The effect of site-isolation of Rh atoms distributed in Ga, in comparison to Rh metal, has been studied by DFT, finding that the DOS of isolated atoms significantly narrows, in agreement with the idea of well-defined single atom sites in SCALMS [10]. In the next steps, we took a closer look at the actual catalysis performed by SCALMS. In close comparison with experimentalists, we did ML-FF simulations of GaNi surface slabs to estimate the temperature-dependent concentration of Ni atoms at the surface and the interaction of them with hydrogen atoms and molecules, revealing that hydrogen atoms (but not molecules) were able to keep the Ni atoms longer at the surface [11]. A systematic study of Pt atoms dissolved into different liquid metals (Ga, In, Sn) has been done with DFT and ML-FF simulations. We could show that the experimental catalytic activity depends on the surface concentration of active Pt atoms and the DOS overlap between them and the gaseous reactants and can thus be tailored by changing the solvent metal [12]. In a real catalyst, SCALMS based on Ga as solvent is always prone to surface oxidation. The structure and thickness-dependent stability of Ga oxide layers on GaPt SCALMS has been studied by DFT and ML-FF simulations, revealing that thin oxide layers are easier to reduce with hydrogen than thicker ones [13].

For SCILL systems, we mostly look at adsorption patterns of ionic liquid molecules on different metal surfaces to elucidate the interactions between them and the metal. In collaboration with traditional force field application by another group, we can then investigate both the larger scale adsorption structures and the detailed electronic interactions. So far, this method has been applied to ionic liquids adsorbed on Au(111) [14,15] and Pt(111) [17]. Further, we optimized intermediates of the dehydrogenation of diethylamine on Pt(111) [16].

Besides the simulation of CLINT systems, we use our computational resources to perform simulations of related surface-science systems. We already simulated the adsorption of bromide atoms on a Pd(111) surface and on a boron nitride nanomesh pattern formed on that surface [18, 19]. For the latter, we also performed ML-FF simulations, showing that we could learn a ML-FF able to simulate the whole nanomesh cell using only data of characteristic parts of it. The dehydrogenation of a liquid organic hydrogen carrier molecule on a platinum surface has been studied by DFT [20]. Finally, the adsorption and metalation of modified porphin molecules on copper [21] and the diffusion of hydrogen on nickel, including the description of nuclear quantum effects [22], have been studied with DFT and ML-FF as well.

[1] G. Kresse; J. Hafner, Ab initio molecular dynamics for liquid metals, Phys. Rev. B 1993, 47, 558.

[2] G. Kresse; J. Furthmüller, Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set, Comp. Mater. Sci. 1996, 6, 15-60.

[3] J. P. Perdew; K. Burke; M. Ernzerhof, Generalized Gradient Approximation Made Simple, Phys. Rev. Lett. 1996, 77, 3865.

[4] S. Grimme; J. Anthony; S. Ehrlich; S. Krieg, A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu, J. Chem. Phys. 2010, 132, 154104.

[5] S. Grimme; S. Ehrlich; L. Goerigk, Effect of the damping function in dispersion corrected density functional theory, J. Comput. Chem. 2011, 32, 1456-1465.

[6] R. Jinnouchi; J. Lahnsteiner; F. Karsai; G. Kresse; M. Bokdam, Phase Transitions of Hybrid Perovskites Simulated by Machine-Learning Force Fields Trained on the Fly with Bayesian Inference, Phys, Rev. Lett. 2019, 122, 225701.

[7] R. Jinnouchi; F. Karsai; C. Verdi; R. Asahi; G. Kresse, Descriptors representing two- and three-body atomic descriptors and their effects on the accuracy of machine-learned inter-atomic potentials, J. Chem. Phys. 2020,152, 234102.

[8] A. Shahzad; F. Yang; J. Steffen; C. Neiss; A. Panchenko; K. Goetz; C. Vogel; M. Weisser; J. P. Embs; A. Görling; I. Goychuk; T. Unruh, Atomic diffusion in liquid gallium and gallium-nickel alloys probed by quasielastic neutron scattering and molecular dynamic simulations. J. Phys.: Condens. Matter 2024, 36, 175403.

[9] I. Goychuk; A. Panchenko; A. Shahzad; J. Steffen; C. Neiss; K. Götz; C. Vogel; M. Weisser; B. Khairalla; J. P. Embs; J. Englhard; J. Bachmann; H. Hildebrand; N. Denisov; P. Schmucki; A. Görling; T. Unruh. Effects of isotope and chemical incoherence on self-diffusion of atoms in liquid alloys of gallium, with isotopes of nickel: experiments based on quasielastic neutron scattering versus molecular dynamics simulations. Accepted by Phys. Rev. B.

[10] T.-E. Hsieh; S. Maisel; H. Wittkämper; J. Frisch; J. Steffen; R. Wilks; C. Papp; A. Görling; M. Bär, Unraveling the Effect of Isolation on Shallow d States of Gallium-Rhodium Alloys. J. Phys. Chem. C 2023, 127, 20484-20490. 

[11] A. Sogaard; T.-E. Hsieh; J. Steffen; M. Wu; S. Carl; Y. R. Ramzi; S. Maisel; J. Will; J. Frisch; N. Taccardi; M. Haumann; E. Spiecker; A. Görling; M. Bär; P. Wasserscheid, Single Atom Sites in Ga-Ni Supported Catalytically Active Liquid Metal Solutions (SCALMS) for Selective Ethylene Oligomerization, submitted to Chem. Phys. Chem.

[12] M. Moritz; S. Maisel; N. Raman; H. Wittkämper; C. Wichmann; M. Grabau; D. Kahraman; J. Steffen; N. Taccardi; A. Görling; M. Haumann; P. Wasserscheid; H.-P. Steinrück; C. Papp, SCALMS: Liquid Metal Catalysis with Ternary Alloys, ACS Catal. 2024, 14, 6440-6450.

[13] N. Madubuko; T.-E. Hsieh; N. Vorlaufer; S. Carl; J. Steffen; A. Mölkner; N. Taccardi; J. Frisch; J. Will; M. Haumann; A. Görling; E. Spiecker; P. Felfer; M. Bär; P. Wasserscheid, Effect of H2 Pre-treatment on Ga-Pt Supported Catalytically Active Liquid Metal Solutions (SCALMS) for propane dehydrogenation – from model systems to lab scale reactor. Submitted to Angew. Chem. Int. Ed.

[14] H. Bühlmeyer; J. Hauner; R. Eschenbacher; J. Steffen; S. Trzeciak; N. Taccardi; A. Görling; D. Zahn; P. Wasserscheid; J. Libuda, Structure Formation in an Ionic Liquid Wetting Layer: A Combined STM, IRAS, DFT and MD Study of [C2C1Im][OTf] on Au(111). Chem. Eur. J. 2023, 29, e202301328.

[15] J. Hauner; H. Bühlmeyer; J. Steffen; S. Trzeciak; N. Taccardi; A. Görling; D. Zahn, P. Wasserscheid; J. Libuda, Temperature-dependent structure formation in the wetting layer of the ionic liquid [C2C1Im][OTf] on Au(111). J. Phys. Chem C 2024, 128, 3894-3906.

[16] R. Eschenbacher; J. Steffen; K. Farrugia; N. Taccardi; P. Wasserscheid; A. Görling; J. Libuda, Reactivity of a model SCILL: Influence of co-adsorbed [C2C1Im][OTf] on the dehydrogenation of dimethylamine on Pt(111). Surf. Sci. 2024, 743, 122453.

[17] H. Bühlmeyer; L. Knörr; J. Steffen; R. Eschenbacher; J. Hauner; A. Görling; J. Libuda, Adsorption and thermal evolution of the carbonyl-functionalized ionic liquid [5-oxo-C6C1Im][NTf2] on Pt(111): A combined IRAS, STM and DFT study. Chem. Eur. J. 2024, e202403900.

[18] E. M. Freiberger; J. Steffen; N. J. Waleska-Wellnhofer; A. Harrer; F. Hemauer, V. Schwaab; A. Görling; H.-P. Steinrück; C. Papp, Bromine Adsorption and Thermal Stability on Rh(111): A Combined XPS, LEED and DFT Study, Chem. Phys. Chem. 2023, 24, e202300510.

[19] E. M. Freiberger; J. Steffen; N. J. Waleska-Wellnhofer; F. Hemauer; V. Schwaab; A. Görling; H.-P. Steinrück; C. Papp, Bromination of 2D Materials, Nanotechnology 2024, 35, 145703.

[20] V. Schwaab; F. Hemauer; J. Steffen; N. K. Waleska-Wellnhofer; E. M. Freiberger; M. Steinmetz: A. Görling; P. Wasserscheid; H.-P. Steinrück; C. Papp, Model catalytic studies on the thermal dehydrogenation of the benzaldehyde/cyclehexamethanol LOHC system on Pt(111). Chem. Eur. J. 2024, 30, e202402793.

[21] M. Shaker; M. Muth; J. Steffen; A. C. dos Santos; S. Jaekel; R. Adhikari; P. Gazetas; C. Oleszak; A. de Siervo; N. Jux; A. Görling; O. Lytken; H.-P. Steinrück, Coverage- and temperature-induced self-metalation of tetraphenyltransdibenzoporphyrin on Cu(111). J. Phys: Condens. Matter 2024, 37, 085001.

[22] J. Steffen, A. Alibakshi, Hydrogen Diffusion on Ni(100): A Combined Machine-Learning, Ring Polymer Molecular Dynamics, and Kinetic Monte Carlo Study. J. Chem. Phys. 2024, 161, 184116.