Skip to main content

Towards a chemically accurate description of reactions on metal surfaces

Final Report Summary - REACTIONBARRIOMETRY (Towards a chemically accurate description of reactions on metal surfaces)

This project addressed four major challenges facing theorists who aim to make accurate predictions for reactions of molecules on metal surfaces. The central goal was to enable the computation of chemically accurate barrier heights for reactions with metal surfaces of catalytic interest, using so-called first principles methods. In the first challenge addressed, we have established that chemically accurate reaction barrier heights can be computed with a semi-empirical approach to density functional theory, i.e. the specific reaction parameter density functional theory (SRP-DFT) for a databse now containing six systems of molecules dissociating on metals. This involved three systems in which a polyatomic molecule interacts with a metal (methane + Ni(111), Pt(111), and Pt(211), and three H2 metal systems (H2 + Cu(111), Cu(100), and Pt(111)).

We have also tested the accuracy of a first principles method, i.e. the Quantum Monte-Carlo method, by performing a calculation of the barrier height for the benchmark H2 + Cu(111) system. With present day methodology and computer hardware, we were able to get to within 1.6 kcal/mol of the best semi-empirical value of the barrier height for this system.

We were able to achieve an accurate description of the effect of motions of the surface atoms and of surface temperature on reaction of both diatomic and polyatomic molecules through implementing so-called Ab Initio Molecular Dynamics (AIMD) calculations on several systems. In AIMD, forces acting between the atoms are calculated directly ("on-the-fly") while the simulation proceeds, from first principles. While this eliminates the need for fitting a potential energy surface, the AIMD method is expensive. We have therefore also used results of DFT and AIMD calculations to obtain a neural network fit for a diatomic molecule interacting with a mobile surface (N2 + Ru(0001)), as well as a polyatomic molecule interacting with a mobile surface (methane + Cu(111)), in collaboration with Jörg Behler. These so-called high-dimensional neural network potentials allow orders of magnitude more efficient dynamics calculations when compared with AIMD, thereby also enabling accurate calculations for systems which exhibit small reaction probabilities.

We have also performed research on ways to achieve an accurate description of the effect of electron-hole pair excitations in the metal on reaction in systems like N2 interacting with a ruthenium surface, and HCl interacting with Au(111). A collaboration with the San Sebastian team resulted in the first publication of results obtained with the Ab Initio Molecular Dynamics with Electronic Friction (AIMDEF) method. In this method, the effect of surface phonon motion and electron-hole pair excitation on the scattering of a molecule from a metal suraface can be treated simultaneously. The AIMDEF method was subsequently used to study the effect of electron-hole pair excitation on vibrational excitation of H2 scattering from Cu(111), and on reaction of HCl on Au(111).

Finally, in collaboration with Jörg Meyer we have tested a new method for describing electron-hole pair excitation, i.e. orbital dependent friction (ODF) on N2 + Ru(0001), in calculations in which also the surface phonons were taken into account using a neural network potential. This method significantly improved the description of reaction of N2 on and scattering from Ru(0001) when compared with a more simple approach to electron-hole pair excitation, i.e. the local density friction approximation (LDFA). Both the ODF and LDFA approach have advantages and disadvantages and need to be put to further tests. The significance of our results is that N2 + Ru(0001) represents the first reaction for which ODF and LDFA lead to significantly different results.