(594c) Dynamical Evolution of Atomically Dispersed Catalysts: Ab Initio Molecular Dynamics Analysis of Thermal and Adsorbate-Induced Metal Atom Migration | AIChE

(594c) Dynamical Evolution of Atomically Dispersed Catalysts: Ab Initio Molecular Dynamics Analysis of Thermal and Adsorbate-Induced Metal Atom Migration


Mallikarjun Sharada, S., University of Southern California
Bac, S., University of Southern California

Catalysis is a common topic among many areas of industry from petroleum refining to health and food production. Catalysts are used in more than 80% of common methods of production that utilize chemical reactions with the industry itself worth up 33.5 billion annually.1 Most catalysts consist of rare and expensive metals, such as platinum and palladium, spurring efforts into enhancing atom efficiencies to ease their demand. Naturally, the emerging field of single atom catalysis (SACs), where metal atoms are atomically dispersed on a support interface, is of significant interest.

It has been revealed that these metal atoms often occupy the cation sublattice of the metal oxide supports, but it is not possible to observe the coordination environment, or reactants during a chemical reaction.1 These metal atoms can often become mobile at low temperatures or when a reaction intermediate becomes adsorbed,2,3 often forming small clusters of up to four atoms that may be mobile at fairly low temperatures.4 The growth of these single metal atoms into clusters, known as sintering, often leads to deactivation of these catalysts5,6 and nucleation of the crystal.7 Sintering can occur via two mechanisms: by cluster diffusion, and by Ostwald ripening.1 It has also been observed that support atoms can also diffuse from the bulk to the surface depending on the reactions conditions, as well as diffusion from the surface into the bulk, resulting in a dynamic support interface.1 For example, oxygen atoms are taken from the surface of the metal oxide interface and metal atoms, such as Fe, can diffuse into the bulk due to under coordination during the oxidation of CO.1This results in a surface that changes during the course of the reaction but can be reversed with the addition of oxygen gas to the system.

Parkinson et al. through scanning tunnel microscopy (STM) studies of adsorbate induced migration on Pd/Fe3O4 and Pt/Fe3O4,2,3 have shown that the adsorption of CO, H2, and O2 can lead to high mobility of the dispersed single atoms which results in the nucleation of small clusters of Pt or Pd. They have determined when the adsorbates are removed from the system, the clusters will redisperse into single atoms on the surface at elevated temperatures.

While microscopy methods demonstrate the importance of probing catalyst stability under reaction conditions, identifying the driving forces for metal atom mobility remains challenging. Our goal is to utilize ab initio molecular dynamics (AIMD) simulations to examine the response of the metal atom to adsorbed intermediates on picosecond timescales. Here, we present our AIMD analysis of metal atom mobility under low-temperature water-gas shift conditions for atomically dispersed Pt on rutile TiO2.


All spin polarized calculation are performed using Vienna Ab Initio Simulation Package, VASP 5.4, using the SCAN meta-generalized gradient approximation (meta-GGA).8 The Atomic Simulation Environment (ASE) is utilized for model construction and geometry relaxation (threshold=0.05 eV/A).9 All atoms are described using the default projector augmented wave (PAW) potentials available in VASP. The non-spherical contributions related to the gradient of the density in the PAW spheres are included in the calculations with the d-orbitals included in the kinetic energy density mixing. Electronic relaxations are carried out using the preconditioned conjugate gradient algorithm recommended for meta-GGA functionals. The Brillouin Zone is integrated using the Monkhorst-Pack set of k-points (1x1x1). Gaussian smearing (σ = 0.1 eV) is used along with the cutoff energy of 400eV for DFT and AIMD calculations. We employ Bader and DOS analysis to calculate charges on surface atoms and characterize the binding of the adatoms to the surface.

The AIMD calculations will be carried out in 3 steps:

  1. Temperature ramp: The DFT-optimized geometries are heated at a rate of 500K/ps from 10K to 500K. The velocities are scaled at every 2.0fs timestep and 40 steps are stored in the Broyden mixer.
  2. Equilibration: The point of equilibration is identified using the method put forth by Chodera et al.10 We use the Nosé-Hoover11 thermostat and calculate the stress tensor at a timestep of 1.0fs. Ten steps are stored in the Broyden mixer.
  • Dynamic Evolution: Five picosecond trajectories are initiated from the equilibrated structures with a 1fs timestep. Settings for mixing and electronic calculations are identical to the equilibration step.

The quantities measured during simulation are coordination number of Pt to the TiO2 surface,12 migration from one DFT site to another,13 velocities of Pt atom, and geometric analysis such as bond lengths and angles of adsorbates. We carry out hydrogen bonding analysis for the simulations that have potential hydrogen bonds forming between adsorbate and support atoms.


Eight common WGS absorbates were simulated adsorbed to Pt/TiO2: CO, H, OH, O, CO2, COOH, CHO, and H2O. The adsorbate OH was also simulated bound to a Ti atom nearby the single Pt atom because the location of the stabilizing hydroxyl is not reported in prior studies of hydroxyl-stabilized metal atoms.

While we aim to present the results of AIMD simulations for all adsorbates (manuscript in preparation), here in Figure 1, we illustrate the dynamic changes in coordination of Pt with support Ti and O atoms in the presence of (a) no adsorbate, (f) CO, and (d) OH. We find that the time averaged coordination number of Pt drops from 1.13 to 0.67 and 0.99 to 0.85 for Ti and O respectively for CO adsorption. The drop in coordination number is likely to induce migration of the Pt atom. The weaker the interaction between Pt and the surface, the more likely Pt is to migrate, which agrees with previous studies.2,3 The time averaged coordination number of Pt to Ti and O also drops when comparing the bare Pt case to OH. The time averaged coordination number of the OH(Ti5c) simulation, Figure 1(e), of Pt to O remains high at 1.12 indicating high interaction with the support surface. No migration is observed during the 5 ps trajectory for this case.


Using AIMD, we observe that the adsorption of CO leads to Pt migration while the adsorption of OH seems to inhibit migration of the Pt atom agreeing with previous experimental papers. The simulation of OH adsorbed to a nearby Pt atom with CO attached is the scope of future work to determine any competing factors between CO and OH.


  1. Parkinson, G. S. Atomic Scale Insights into Single-Atom Catalysis. Vak. Forsch. und Prax. 30, 45–49 (2018).
  2. Bliem, R. et al. Dual role of CO in the stability of subnano Pt clusters at the Fe3 O4(001) surface. Proc. Natl. Acad. Sci. U. S. A. 113, 8921–8926 (2016).
  3. Parkinson, G. S. et al. Carbon monoxide-induced adatom sintering in a Pd–Fe3O4 model catalyst. Nat. Mater. 12, 724 (2013).
  4. Xu, L., Henkelman, G., Campbell, C. T. & Jónsson, H. Small Pd clusters, up to the tetramer at least, are highly mobile on the MgO(100) surface. Phys. Rev. Lett. 95, 1–4 (2005).
  5. Campbell, C. T., Parker, S. C. & Starr, D. E. The effect of size-dependent nanoparticle energetics on catalyst sintering. Science (80-. ). 298, 811–814 (2002).
  6. Bartholomew, C. H. Mechanisms of catalyst deactivation. Appl. Catal. A Gen. 212, 17–60 (2001).
  7. Bliem, R. et al. Cluster nucleation and growth from a highly supersaturated adatom phase: Silver on magnetite. ACS Nano 8, 7531–7537 (2014).
  8. Sun, J., Ruzsinszky, A. & Perdew, J. Strongly Constrained and Appropriately Normed Semilocal Density Functional. Phys. Rev. Lett. 115, 1–6 (2015).
  9. Hjorth Larsen, A. et al. The atomic simulation environment - A Python library for working with atoms. J. Phys. Condens. Matter 29, (2017).
  10. Chodera, J. D. A Simple Method for Automated Equilibration Detection in Molecular Simulations. J. Chem. Theory Comput. 12, 1799–1805 (2016).
  11. William G. Hoover. Canonical dynamics: Equilibrium phase-space distributions William. Phys. Rev. A 31, 1695–1697 (1985).
  12. Sprik, M. Coordination numbers as reaction coordinates in constrained molecular dynamics. Faraday Discuss. 110, 437–445 (1998).
  13. Humphrey, N., Bac, S. & Mallikarjun Sharada, S. Ab Initio Molecular Dynamics Reveals New Metal-Binding Sites in Atomically Dispersed Pt 1 /TiO 2 Catalysts. J. Phys. Chem. C (2020). doi:10.1021/acs.jpcc.0c06771