Surface Segregation
Surface segregation is the tendency of one component of an alloy to be enriched (or depleted) at the surface relative to the bulk. For a dilute impurity it is quantified by the surface segregation energy, the energy needed to move one impurity atom from the bulk to a particular surface site; with the usual sign convention a negative value means the impurity prefers the surface. Segregation decides whether a deliberately added second metal stays where it was placed on a bimetallic surface or disappears into the bulk, so it matters for designing alloy catalysts, decorated surfaces and nanoparticles. Commonly cited driving forces are the loss of coordination at the surface, the difference in bond strength between host and impurity, atomic size mismatch, and d-band effects (see d-band model). Calculations based only on energies tend to overstate segregation, because the configurational entropy of an impurity in the bulk always favors dissolution.
Ni diffusion below Pt(111) on annealing (2003)
LEIS, Auger and grazing emission XPS showed that Ni deposited on Pt(111) leaves the outermost layer when heated: Ni signals fall from about 700 K and nearly vanish by 900 K, and a 650 K anneal steadily depletes surface Ni over 20 h. DFT put Ni in the second layer 0.65 eV per Ni atom below a Ni overlayer in energy, consistent with Pt segregating to the surface of Ni-Pt alloys and with the Pt-terminated, Ni-subsurface structure being the chemically active one (Kitchin et al. 2003). Summary: kitchin2003.
Oxygen-induced Pd segregation at Ag3Pd(111) (2008)
Kitchin, Reuter and Scheffler computed the segregation profile of Ag3Pd(111) in oxygen by screening 473 DFT surface structures with varied top-two-layer composition and 0 to 1 ML O (Kitchin et al. 2008). Clean surfaces under vacuum-like conditions are Ag-terminated, but rising oxygen chemical potential stabilizes O-covered structures with Pd-rich and then fully Pd first layers, reversing the profile. Small antisite nonstoichiometries in the bulk alloy change the transition oxygen pressures by orders of magnitude and alter the sequence of stable structures, and the dilute-impurity segregation model (-0.31 eV for Ag in Pd, +0.24 eV for Pd in Ag) captures the trend but errs by several tens of meV/Ų. Because Pd is scarce in Ag3Pd, its segregation sets in only close to the onset of bulk PdO formation. Summary: kitchin2008.
Coordination-dependent segregation and step decoration (2009)
Han, Kitchin and Sholl computed segregation energies with DFT (PW91) for five impurities in each of Cu, Rh, Pd and Ir hosts, over sites with coordination numbers from 6 (kinks on the chiral fcc(643) surface) to 12 (subsurface). Segregation generally became more favorable as coordination decreased. They fit a correlation that is linear in the coordination deficit (12 minus CN) multiplied by a combination of a host/impurity bond strength difference (from liquid surface tensions) and the impurity-induced shift of the surface d-band center. It reproduced Ruban et al.’s flat-surface database with a standard deviation of 0.26 eV/atom for fcc hosts plus Fe, Co and Ru, and predicted the sign of kink-site segregation for 347 pairs across the 3d, 4d and 5d metals, 206 of which favor decorating kinks. Because the model is linear in CN, the bulk-to-terrace energy also equals the terrace-to-kink energy, which lets flat-surface data be reused for stepped surfaces (Han et al. 2009). Summary: han2009; see also Chiral metal surfaces.
Hydrogen-induced segregation in CuPd catalysts (2015)
Clean CuPd surfaces in vacuum enrich in Cu, but adsorbates can reverse this. In H2-D2 exchange over a CuxPd1-x composition spread film, Gumuslu and co-workers estimated hydrogen adsorption energies from the difference between desorption and adsorption barriers (about 0.55 eV at x = 0.3 and 0.35 eV at x = 0.8), much larger than the clean-surface Cu segregation energy (0 to -0.05 eV), so adsorbed hydrogen should drive Pd to the surface. They propose this explains why the H2 adsorption barrier stops changing with composition for Pd-rich alloys (x < 0.6): the top layer saturates in Pd while UPS, which samples 2 to 5 layers, still reflects the bulk composition (Gumuslu et al. 2015). Summary: gumuslu2015.
Modeling adsorbate-induced segregation with the Gibbs isotherm (2015)
Boes and co-workers estimated the surface composition of CuxPd1-x(111) under 1 atm H2 at 800 K by solving the Langmuir-McLean form of the Gibbs isotherm with a total segregation free energy equal to the experimental vacuum value plus the hydrogen coverage times the difference in H adsorption energy between Cu(111) and Pd(111). Although clean surfaces in vacuum are Cu-enriched at every composition, the model predicts substantial Pd enrichment under hydrogen. Including this segregation was needed to reproduce measured composition-dependent H2 adsorption energies for x > 0.5; the authors describe the adsorbate term as the simplest possible form, which neglects strain differences between sites (Boes et al. 2015). Summary: boes2015a.
Segregation mapped across composition space on a single library (2016)
This Perspective highlights how composition spread alloy films let segregation be measured as a function of bulk composition rather than at one composition at a time (Kitchin 2016). On a CuxPd1-x film, XPS and low energy ion scattering over xCu = 0.05 to 0.95 and 400 to 1000 K showed Cu segregation at every composition, strongest at 700 K, and Langmuir-McLean fits yielded composition-dependent segregation enthalpies and entropies from one sample. It also stresses adsorbate-induced segregation: the lower-than-interpolated H2 adsorption barriers on Pd-poor CuxPd1-x (xPd below 0.40) were explained by hydrogen pulling Pd to the surface. Summary: kitchin2016.
Segregation across composition space in AuPd(111) (2017)
Boes and Kitchin went beyond the dilute limit by running canonical Monte Carlo on 10 × 10 × 15 AuPd(111) slabs with energies from a neural network potential trained on 3914 DFT calculations, covering 10 to 90% Au in the bulk and 700 to 1000 K. Au always enriched the top layer, the second layer was Pd-enriched and the third matched the bulk, and the 800 K top-layer compositions agreed with LEIS measurements. Fitting the Langmuir–McLean isotherm to the simulations gave a composition-dependent segregation enthalpy, about -10 kJ/mol on the Pd-rich side and approaching the dilute-limit DFT values at high Au content, fit by ΔH = -60.5x³ + 27.5x - 15.5 kJ/mol; the authors note this trend could not be obtained from dilute-limit energies alone. Surface sites showed short-range order favoring Au–Pd neighbors, and DFT tests indicated that surface relaxation has little effect on the predicted composition (Boes 2017). Summary: boes2017; see also Neural network potentials.
Acrolein-induced Pd segregation and aggregation at AgPd(111) (2021)
Liu, Yang and Kitchin combined SingleNN surrogate potentials for bulk and slab energies, a fingerprint nearest-neighbor model for acrolein adsorption energies, and semi-grand canonical Monte Carlo to predict the AgPd(111) surface composition at 373 K (Liu et al. 2021). In vacuum the first layer stays essentially pure Ag until bulk Pd exceeds 20%, with Pd enriched in the second layer. With acrolein at about 1/9 monolayer, Pd is pulled to the surface: the first layer holds about 11% Pd (one isolated Pd per acrolein) for bulk Pd of 5 to 6.5%, and near 6.5% bulk Pd Pd dimers forming bridge sites begin to appear. DFT showed that only a single surface Pd under acrolein has a net favorable segregation energy when adsorption relative to pure Ag is included. Summary: liu2021.
Ternary Cu-Pd-Au segregation and the limits of fcc(111) models (2022)
Yang, Guo, Gellman and Kitchin extended the AuPd approach to a ternary alloy, running Monte Carlo at 600 K on 10 × 10 × 15 Cu-Pd-Au(111) slabs at 24 bulk compositions with a neural network trained on 5278 DFT slabs. Au was enriched at the surface everywhere and Cu depleted everywhere, while Pd was depleted except near the PdCu edge. Compared with LEIS on composition spread films, Au-Pd exchange energies matched experiment, Au segregation over Cu was overestimated, and the Pd-Cu trend had the wrong sign. Surface relaxation and vibrational free energy were shown not to change the sign of fcc(111) segregation energies; the CuAu error was traced to PBE underbinding of Cu-Au alloys and underestimated Au surface energy. For CuPd, relaxation reverses the segregation sign on fcc(110), and Monte Carlo on relaxed CuPd(110) reproduced the measured Cu segregation, suggesting that polycrystalline films are poorly represented by ideal fcc(111) slabs (Yang et al. 2022). Summary: yang2022; see also Exchange-correlation functionals.
Ternary segregation benchmarked against composition spread films (2024)
Broderick and coworkers ran lattice Monte Carlo with a fine-tuned graph neural network potential for AgAuCu, AuCuPd and CuPdPt on (111), (110) and (100) slabs and compared surface excess maps with composition spread alloy film measurements for AgAuCu and AuCuPd (Broderick et al. 2024). The (110) orientation gave the best agreement for both ternaries, which the authors relate to atomic size mismatch being relieved on the more open facet; Ag segregation in AgAuCu was captured qualitatively but underpredicted, and Au segregation in AuCuPd was overpredicted. Pt was predicted to deplete strongly from CuPdPt surfaces. Summary: broderick2024.