Finding optimal surface sites on heterogeneous catalysts by counting nearest neighbors

See allHide authors and affiliations

Science  09 Oct 2015:
Vol. 350, Issue 6257, pp. 185-189
DOI: 10.1126/science.aab3501

Accounting for surface coordination

The exploration of heterogeneous catalysts using first-principles calculations can be daunting because the large number of atoms and possible surface geometries. Calle-Vallejo et al. describe a simpler metric for assessing optimal reactivity: a weighted average of surface coordination that includes second-nearest neighbors (see the Perspective by Stephens et al.). The calculations identified three approaches for introducing cavity sites into the platinum(111) surface to improve its performance for the oxygen reduction reaction used in fuel cells.

Science, this issue p. 185, see also p. 164


A good heterogeneous catalyst for a given chemical reaction very often has only one specific type of surface site that is catalytically active. Widespread methodologies such as Sabatier-type activity plots determine optimal adsorption energies to maximize catalytic activity, but these are difficult to use as guidelines to devise new catalysts. We introduce “coordination-activity plots” that predict the geometric structure of optimal active sites. The method is illustrated on the oxygen reduction reaction catalyzed by platinum. Sites with the same number of first-nearest neighbors as (111) terraces but with an increased number of second-nearest neighbors are predicted to have superior catalytic activity. We used this rationale to create highly active sites on platinum (111), without alloying and using three different affordable experimental methods.

Slight changes in the surface structure of catalytic materials can have large impacts on the products and energetics of chemical reactions (1, 2). However, not all sites at catalytic surfaces have the same activity or selectivity, which gives rise to the concept of structure sensitivity (35). Generally, identification of active sites is a challenging task that requires the combination of several approaches (6, 7). A rational way of designing catalysts should first identify the optimal active sites and then engineer surfaces where the presence of such sites is maximized. Modern computational screening techniques can provide the atomic-scale insight needed to elaborate simple catalyst design rules (2, 8). Such rules are the starting point to engineer target active sites on catalytic surfaces (912). The connection between these two steps must be straightforward and clear, which is not trivial in practice. The difficulties originate from an important detail: Existing computational techniques outline optimal energetic properties, which can be met by countless materials. Therefore, it is desirable to create procedures that generate more precise design rules.

Consider the specific case of the oxygen reduction reaction (ORR), Embedded Image, a key reaction for proton exchange membrane fuel cells that is normally catalyzed by Pt, which is scarce and expensive. Although many challenges remain (13), rational catalyst design has opened new avenues for the ORR through volcano-type activity plots (8, 1416). These plots are based on the Sabatier principle, which states that good catalysts balance the strength of adsorption and desorption of key reaction intermediates. Volcano plots typically correlate surface adsorption energies with estimates of the catalytic activity of materials. By means of density functional theory (DFT) calculations, they predict that optimal ORR catalysts must bind hydroxyl species (*OH) ~0.1 eV (2.3 kcal/mol) more weakly than Pt(111) (8, 15). Several studies have used this criterion to find alloy catalysts with high ORR activities (8, 12, 15, 17). However, this condition can be met by numerous materials, including metals and alloys (15), oxides (18), functionalized graphitic materials, and porphyrins (19, 20). Therefore, screening routines are used to select suitable candidates from large databases in which the composition, structure, and adsorption energies of materials are known beforehand. Creating such databases demands considerable time and computational expenses.

We introduce here “coordination-activity plots,” which outline the geometric structure of optimal active sites. The method is illustrated on the ORR, for which we devise Pt catalysts that are 3.5 times more active than Pt(111).

Usually, trends in adsorption energies for small species on extended surfaces of a given transition metal are well described by the coordination number (cn) of the surface sites (21, 22). However, cn loses its accuracy when nanoparticles (NPs, the typical form of metals on high–surface area catalysts) are considered because of “finite-size effects” (23, 24), so more sophisticated descriptors are needed (25). For example, in Fig. 1A, all sites in blue have nine nearest neighbors, that is cn = 9, but the adsorption energies of *OH calculated with DFT [see fig. S21 and tables S2 and S3 in the supplementary materials (26)] can differ by more than 0.5 eV (11.5 kcal/mol). A simple strategy to allow for direct comparison of NPs and extended surfaces is the use of “generalized” coordination numbers (Embedded Image) (27), which introduce a weight to each first-nearest neighbor atom j, corresponding to its own coordination number [cn(j)] The formula used to estimate Embedded Imagefor a site i is (27)

Fig. 1 Adsorption-energy trends (in eV) described by generalized coordination numbers Embedded Image.

(A) Sites with cn = 9 (blue) on NPs of various sizes and on the extended (111) surface. The six surface nearest neighbors (yellow) and the three in the subsurface (white) are marked. Despite the identical coordination (cn = 9), Embedded Image can differ by ~0.5 eV. The differences are due to the second-nearest neighbors. (B) Trends in Embedded Image and Embedded Image described by Embedded Image, including extended surfaces (brown) and truncated octahedron (●), cuboctahedron (■), and tetrahedron (▲) NPs: Pt586 (yellow), Pt201 (blue), Pt147 (gray), Pt79 (magenta), Pt68 (orange), and Pt38 (green). The reactions used to calculate the adsorption energies appear as insets. Least-squares fits are provided together with mean and maximum absolute errors (MAE and MAX).

Embedded Image (1)

The sum includes all of the first-nearest neighbors, and the division by the maximum number of first-nearest neighbors in the bulk (cnmax) ensures that Embedded Imagespans the range between 0 and 12 in face-centered cubic metals, similarly to conventional cn. For instance, the blue site on Pt38 in Fig. 1A has six neighbors with cn = 6 (yellow) and three with cn = 12 (white), so Embedded Image [section S3.1 in (26) shows how to compute Embedded Imagefor all sites under study]. This simple extension explains the variations in Fig. 1A and results in the linear relation in Fig. 1B, where the trends in adsorption energies of *OH and *OOH, Embedded Image and Embedded Image, on all sites on Pt NPs of various sizes (0.7 to 2.6 nm) and shapes (truncated octahedron, cuboctahedron, tetrahedron) and extended surfaces are presented; data are reported for terraces, edges, corners, steps, kinks, and metal adatoms [see tables S2 and S3 in (26)]. Figure S21 shows that Embedded Imagedescribes adsorption-energy trends more accurately than cn. Additionally, because Embedded Image is arithmetical, its assessment does not require DFT calculations.

Following the ORR model in (16), the two potential-determining steps are [see Fig. 2 and section S1 in (26)] (i) the first proton-electron transfer, in which O2 is transformed into *OOH; or (ii) the last proton-electron transfer, in which *OH is transformed into H2O. Thus, within this model, the ORR activity depends primarily on Embedded Image and Embedded Image. An activity plot is formed when the reaction energies of steps (i) and (ii) are evaluated as a function of a given descriptor. If adsorption energies are used as descriptors, only the optimal adsorption properties will be identified (2, 15, 16). If structural parameters are used as descriptors, the outcome will be the geometry of optimal sites. The choice between energetic and geometric descriptors is determinant, as illustrated in Fig. 2A, where a coordination-activity plot is presented [see also fig. S1 in (26)]. This plot shows that optimal Pt surface sites for the ORR possess Embedded Image. In agreement with energetic volcano plots (8, 15), Embedded Image on optimal catalysts is ~0.15 eV weaker than on Pt(111). Besides, the additional structural prediction Embedded Image can be used to guide experiments.

Fig. 2 Coordination-activity plot.

(A) Potentials for the two limiting steps on extended surfaces and NPs. Points B and C (in light blue) are given for two cavities on Pt(111). The potential-determining step on the left (low coordination – strong binding) and right (high coordination – weak binding) sides of the volcano are indicated. Theoretical overpotentials (ηORR) are the vertical difference between the points and the equilibrium potential (red dashed line). Optimal catalysts have Embedded Image and *OH adsorption energies ~0.15 eV weaker than Pt(111) (area in gray). (B) Top view of a six-atom cavity on Pt(111) with Embedded Image. (C) Five-atom cavity on Pt(111) with Embedded Image.

First, note that Embedded Image for (111) terraces in extended surfaces and sufficiently large NPs (Figs. 1 and 2). If the top of the volcano is found at Embedded Image, optimal catalytic sites must have more neighbors than (111) terraces. Sites with cn = 10 [e.g., bottom of (100) step sites] or cn = 11 [e.g., troughs of (110) facets or bottom of (111) step sites] have Embedded Imagevalues between 8.75 and 9.50, which exceed the optimal value and are problematic because of steric hindrance, weak adsorption energies, and proximity to undercoordinated sites, resulting in adsorbate diffusion to neighboring sites.

If no more first-nearest neighbors can be added to Pt(111) sites, a way of producing sites with Embedded Image is by changing the coordination numbers of such first-nearest neighbors—namely, the number of second-nearest neighbors of the active site. Figure 2, B and C, shows two one-layer-deep cavities on Pt(111), corresponding to the removal of six (B) and five (C) adjacent surface atoms. In the configuration in Fig. 2B, the atom in the middle of the cavity (blue) has cn = 9, but its six in-plane nearest neighbors (yellow) have cn = 10 [one more neighbor compared to Pt(111)] and its three subsurface neighbors (white) have cn = 12, so Embedded Image. Similarly, the active site in Fig. 2C has Embedded Image. The overpotentials for these sites are lower than on Pt(111) by ~0.10 to 0.13 V. Several other configurations may exist, but the design rule is clear, general, and simple: Enhanced Pt(111) sites for the ORR must have an increased number of second-nearest neighbors, so that cn > 9 for the first-nearest neighbors of the active site. Such a conclusion could not be obtained using cn as descriptor in Fig. 2, as it does not account for second-nearest neighbors [see fig. S21 in (26)].

We used these theoretical guidelines to engineer active sites at Pt(111) with one of three approaches illustrated in Fig. 3A [see section S2 in (26)]: (i) We stripped away Cu atoms electrochemically from the top layer of an ordered Cu/Pt(111) surface alloy (SA) (9); (ii) we exchanged a deposited Cu-overlayer with Pt ions in solution via galvanic displacement (28) to form Pt “surface islands”; or (iii) we formed a subsurface Pt oxide, which we then reduced during a cathodic potential scan, causing desorption of some Pt atoms from the surface (29, 30) to create both (desirable) small and (undesirable) large cavities. These approaches create surfaces with different adsorption energies of *OH compared to Pt(111), as shown in Fig. 3, B to D [see also figs. S3 to S5, S8, S10, and S16 to 19 in (26)].

Fig. 3 Electrochemical experiments on pristine and defective Pt(111).

(A) Schematics of the approaches used to create defects at Pt(111). Cu atoms appear in red and Pt atoms in gray or black, according to the depth with respect to the surface layer. (B to D) Left: cyclic voltammograms characterizing Pt(111) upon the following electrochemical modifications: (B) dealloying of a Cu/Pt(111) SA (red); (C) five galvanic displacements (blue); and (D) electrochemical destruction of Pt(111) (green). Right: integrated anodic parts of the corresponding voltammograms, scan rate dE/dt = 50 mV s−1, Ar-saturated 0.1 M HClO4. The positive shifts indicate weakened *OH adsorption energies and enhanced ORR activity compared to Pt(111).

For Pt-based ORR catalysts, the most insightful part of the cyclic voltammograms is the *OH adsorption-desorption region (0.6 to 1.0 V in Fig. 3, B to D). For the three modified electrodes, a noticeable weakening of the *OH adsorption is observed, as the two peaks at ~0.8 V appear at more positive potentials. The interaction between *OH and the surface is quantified in the right panels of Fig. 3, B to D. Sizable positive shifts are observed for dealloyed Cu/Pt(111) SA (~46 mV), and for Pt(111) electrodes after galvanic displacement (~91 mV) and electrochemical oxidation (~78 mV). Thus, these catalysts bind *OH more weakly than pristine Pt(111) (see figs. S6 to S11). Normally, the *OH adsorption potentials predicted from volcano plots compare well to experimental onset potentials for *OH adsorption (15, 17). Here, the experimental shifts in the *OH adsorption peaks (~0.05 to 0.09 V) with respect to Pt(111) in Fig. 3 are also in agreement with those in Fig. 2A (0.10 to 0.13 V).

The activities of Pt(111), various (111) defective electrodes, and similarly treated polycrystalline Pt electrodes (Ptpc) presented in Fig. 4 show that specific defects can increase the activity ~3.5 times compared to Pt(111). This increase in catalytic activity cannot be explained by the modest increase in the number of accessible surface adsorption sites [maximum 15%; see (26)]. However, defects do not enhance the ORR activity of all Pt electrodes. Figure 4A shows that defects on Ptpc did not noticeably enhance the ORR activity [see also figs. S13 and S14 in (26)]. This is attributed to different corrosion mechanisms on Pt facets that lead to the formation of dissimilar defects (31).

Fig. 4 Comparison between catalysts in this and in previous studies.

(A) Kinetic current densities in O2-saturated 0.1 M HClO4 for defective Ptpc (dotted line); Pt(111) (black); dealloyed Cu-Pt(111) SA, (SA)dealloyed (orange); Pt(111) electrodes modified via galvanic displacement, Pt(111)1GD (1 Cu monoloayer displaced, green) and Pt(111)5GD (5 monoloayer Cu displaced, red); and electrochemical destruction (10 cycles, blue), Pt(111)ED. (B) ORR activities for defective Pt(111)ED and Pt3Ni (6); Cu-Pt(111) near-surface alloy (17), Pt3Y and Pt3Sc (8), Pt5Gd (12); Pt5Y, Pt3Zr, Pt3Hf, and Pt5La (15). The potential for comparison is 0.9 V.

Because only specific kinds of defects are beneficial, “template” methods generating uniform surfaces with abundant target defects are needed to enhance the ORR activity. This is important for the design of highly active NPs. Convex NPs (Fig. 1) have numerous undercoordinated sites that are not ORR active, and only (111) sites on sufficiently large NPs are similar to extended Pt(111) (see Fig. 2A). Thus, concave geometries are recommendable to introduce sites with Embedded Image and enhance the ORR activity (10).

Figure 4B shows the ORR current densities of the most active defective surface in this work and those of state-of-the-art Pt-based catalysts. Pt(111) with cavities possesses ORR activities that exceed those of several active alloys. Therefore, the optimal electronic and coordination configuration of Pt ORR catalysts is close but not identical to that of (111) terraces, and alloying or adding second-nearest neighbors have similar beneficial effects.

Figure S22 in (26) shows the ORR coordination-activity plot for gold, where optimal sites correspond to (100) terraces or possess Embedded Image. In (26), we outline the extension of Embedded Image to alloys and other compounds. Coordination-activity plots could be used to model other catalytic reactions such as H2O2 production, CO2/CO reduction, and nitrate reduction, as the relation between adsorption energies and surface coordination exists on metals such as Cu, Ag, and Au and for adsorbates such as *O, *O2, *H2O, *H2O2, *CO, *CH, *N, and *NO3 (21, 22, 24, 27). Thus, this work opens up the path in heterogeneous catalysis for the design of optimal surface sites utilizing coordination rationales.

Supplementary Materials

Materials and Methods

Figs. S1 to S22

Tables S1 to S3

References (3239)

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: We received funding from The Netherlands Organisation for Scientific Research (NWO), Veni project 722.014.009; the European Union’s FP7/2007-2013 program, grant 303419 (PUMA MIND); the Cluster of Excellence RESOLV (EXC 1069) funded by Deutsche Forschungsgemeinschaft, Helmholtz-Energie-Allianz (HA-E-0002), and SFB 749; and the Cluster of Excellence Nanosystems Initiative Munich. We thank Stichting Nationale Computerfaciliteiten (NCF), Institut du Développement et des Ressources en Informatique Scientifique, Centre Informatique National de l’Enseignement Supérieur (project 609, GENCI/CT8), and Pôle Scientifique de Modélisation Numérique for CPU time. We thank A. Pathan, M. Schmuck, and A. Zychma (RUB) for supporting measurements. The DFT data appear in tables S1 to S3.
View Abstract

Stay Connected to Science

Navigate This Article