Identification of Sulfur-Tolerant Bimetallic Surfaces Using DFT Parametrized Models and Atomistic Thermodynamics
Nilay İnoğlu and John R. Kitchin, “Identification of Sulfur-Tolerant Bimetallic Surfaces Using DFT Parametrized Models and Atomistic Thermodynamics”, ACS Catalysis 1(4), 399-407 (2011). https://doi.org/10.1021/cs200039t
This summary was written by a language model from the paper and has not been checked line by line. Rely on the published paper for details.
Question
Sulfur compounds in fossil-derived feeds poison transition metal catalysts, and alloying is one way to make a metal more tolerant. The difficulty is combinatorial: even a modest set of metals and surface motifs gives more alloy surfaces than DFT can comfortably evaluate one by one, and a low S binding energy alone does not guarantee that a structure will survive under reaction conditions, because segregation and the gas environment matter too. The paper asks whether a cheap, DFT-parametrized algebraic model of the surface d-band width can predict S adsorption energies across hundreds of bimetallic surfaces, and whether an atomistic thermodynamic treatment can then pick out alloys that are both stable and more S-tolerant than the parent metal (İnoğlu 2011).
Approach
The screening space was binary combinations of ten late transition metals (Fe, Ni, Cu, Ru, Rh, Pd, Ag, Ir, Pt, Au) in four motifs on fcc(111): a heteroepitaxial overlayer, a subsurface underlayer (sandwich type), a 0.25 ML surface dopant and a 0.25 ML subsurface dopant, 370 surfaces in all. All structures were treated as fcc at the host lattice constant (an approximation for Fe and Ru). DFT calculations used DACAPO with ultrasoft pseudopotentials, PW91, a 340 eV cutoff, 4-layer 2x2 slabs with the top two layers relaxed, a 6x6x1 k-point grid and spin polarization where needed, with S at 0.25 ML in the fcc hollow. The d-band width was taken as the square root of the second moment of the projected d band about the Fermi level (İnoğlu 2011).
d-Band widths for all 370 surfaces were estimated with the group’s “solid-state table”, a tight-binding inspired formula summing neighbor contributions that scale as the product of characteristic d-orbital radii to the 3/2 power over the fifth power of interatomic distance, plus an atomic offset and a facet constant (İnoğlu 2010; İnoğlu 2010). A subset of 110 overlayer structures was computed directly with DFT to calibrate a correlation between S adsorption energy and d-band width. Stability was assessed with a surface free energy that combines the clean host surface energy, an impurity segregation energy (assumed independent of alloy concentration), the S adsorption energy and the S chemical potential, the latter set by the H2S/H2 ratio and referenced so that bulk sulfur sits at zero. The onset of poisoning is the chemical potential at which the S-covered surface free energy equals the clean one, so a more tolerant surface needs a higher S chemical potential before it poisons (İnoğlu 2011).
Key findings
On Pd(111) and Pt(111), S adsorption energies at six coverages from 0.25 to 1 ML correlated nearly linearly with the d-band center, which was itself linear in the d-band width, supporting width as a reactivity descriptor. Estimated and DFT d-band widths for the 110 overlayers agreed with a mean absolute error below 4.5%, with outliers mostly Ag or Au overlayers whose large lattice mismatch causes relaxations not captured by unrelaxed host distances. Raw correlations between S adsorption energy and width were poor for Ag- and Au-containing systems, attributed to filled d bands and Pauli repulsion. Excluding those, shifts in adsorption energy versus shifts in width relative to the pure metals collapsed onto one line with slope 1.59 (uncertainty 0.045). The model then reproduced 86% of DFT S adsorption energies within 15% and 92% within 20%, with a mean absolute error below 6% after removing systems the d-band picture cannot describe. Reconstructions (rumpling in Ir@Fe, Ru@Ni and Ag@Cu) were the main failures (İnoğlu 2011).
Propagating the width and slope uncertainties gave a mean standard deviation below 0.018 eV in predicted adsorption energies and about 3% in the poisoning onset. Across the database, alloying with a metal of smaller lattice parameter (compressive strain) or larger d-orbital radius (more d overlap) broadens the d band, lowers its center and weakens S binding. For Cu, poisoning of the pure metal begins near a S chemical potential of -0.93 eV (minus signs are lost in the extracted text but follow from the stated scale running from minus infinity to zero). Cu@Fe and Cu@Ni overlayers, Cu sandwiching 4d and 5d metals, and Ag, Ir, Pt or Au dopants weaken S binding; a Cu-Ir-Cu structure is more tolerant, while Cu on Ru (tensile strain) poisons earlier, near -1.30 eV. The predictions were consistent with experimental reports of S-tolerant Pd-Cu and Pd-Au hydrogen separation membranes, Au-modified Pt and Rh, stronger S binding in Cu-Ag than in Ag, and Pd-Rh combinations more tolerant than the parent metals (İnoğlu 2011).
Significance
The work chains a cheap electronic-structure model, a DFT-calibrated descriptor correlation and atomistic thermodynamics into a coarse screen that reduces a large alloy space to a shortlist for more accurate study. The authors note its approximations (bulk impurity segregation energies, no strain contribution to stability, unrelaxed geometries) and suggest the same route could extend to other adsorbates through scaling relations such as SH to S (İnoğlu 2011).