Heterogeneous catalysis plays a key role in strategic fields such as pollution control, materials and chemicals synthesis, and fuel production. The combination of several metals has long been recognized as an important route for improving the activity, selectivity and/or stability of supported nanocatalysts through various geometric and electronic effects1,2,3,4. In recent years, the renewed interest for nanoalloys, i.e. well-defined multimetallic nanoparticles, has been driven by great advances in atomic-scale characterization techniques (e.g. aberration-corrected transmission electron microscopy, AC-TEM) and computer simulation methods (based e.g. on density functional theory, DFT)5,6,7,8,9,10. Together with the development of nanocatalyst preparation methods11,12,13,14,15,16, this progress allows researchers to better model and understand surface reaction processes, thereby enabling a “rational design” of more efficient catalysts.

Although several groups have succeeded in synthesizing solid-solution nanoalloys of bulk-immiscible elements using colloidal methods11,13,14, these structures are generally metastable and cannot resist the thermal treatment needed for removing the stabilizing agents. As the thermodynamic properties of bulk alloys can influence the structure of their nanosized counterparts, some of us have recently compared the well-known bulk-miscible Au-Pd system7,12,16 to the rarely studied bulk-immiscible Au-Rh system11,13,17 in terms of mixing behaviour and reactivity at the nanoscale9,15,18. As seen for 3 nm-sized nanoparticles (NPs) anchored on well-defined rutile titania nanorods, Au and Pd atoms form a solid-solution alloy, whilst Au and Rh atoms segregate into single-phase domains within the NPs15. Nevertheless, in both cases the interaction of the active metal (Pd, Rh) with gold and titania inhibits the surface oxidation or sulfidation of the former15.

AuRh/TiO2 catalysts have proved to be efficient in several reactions including the hydrogenation of tetralin in the presence of H2S, during which the bimetallic catalyst showed improved sulfidation resistance15, and the hydrodeoxygenation of guaiacol, for which the AuRh catalyst performed more selectively than Au and Rh ones19, both carried out in large excess of hydrogen around 300 °C. In this work, with the aim of gaining insights into the structure of the catalytically active phase, the catalyst submitted to thermal treatments in hydrogen has been investigated in details by aberration-corrected scanning transmission electron microscopy (AC-STEM). Moreover, DFT calculations have enabled us to rationalize the striking bimetallic catalyst structures observed experimentally.

Results and Discussion

Structural analysis of AuRh nanoparticles

Unsupported nanoparticles

Au-Rh nanoparticles were synthesized by conventional colloidal chemical co-reduction in water, using chloride salts as metal precursors, polyvinyl alcohol (PVA) as surfactant, and NaBH4 as reducing agent, as previously reported15 (see Methods section). Figure 1 shows representative AC-STEM images of as-prepared AuRh NPs embedded in PVA (dried AuRh@PVA colloids, sample 1, Au63Rh37 average composition as determined by elemental analysis), which are round-shaped and ca. 3 nm in size. A variety of structures were identified, including alloyed fcc single-crystal, alloyed multiply twinned, and probable Rh@Au core-shell and Au-Rh Janus configurations. The structural inhomogeneity of as-prepared NPs is inherent to the NP synthesis method12. It likely arises from the fact that the atom migration is limited by the low (generally room) temperature of the reactant solution.

Figure 1
figure 1

AC-STEM characterization of unsupported nanoparticles.

HAADF images of AuRh@PVA colloids (sample 1) deposited on lacey C grid, showing single-crystal (a), multi-twinned (b), Rh@Au core-shell (c), and Janus (d) structures.

Supported nanoparticles heated to 350 °C

For the preparation of AuRh/TiO2 supported catalysts, a powder of single-phase rutile titania nanorods synthesized using a hydrothermal method (see Methods section) was added to the acidified colloidal suspension, which led to the immobilization of PVA-embedded NPs on the titania support. The average size of the bimetallic particles was 3.3 ± 1.0 nm (sample 2). Segregation was observed by STEM in the form of Au or Rh single-phase domains randomly distributed within the nanoparticles, as previously reported15.

As previously shown by infrared spectroscopy and TEM, the heating of AuRh/TiO2 to 350 °C under H2 flow leads to the complete removal of the PVA surfactant from the NP surface without significant change of the mean particle size15. However, a drastic change in the morphology and chemical structure of the NPs occurs, as shown by the AC-STEM high-angle annular dark field (HAADF) images of Fig. 2a–d and Supplementary Fig. S1. The NPs are faceted and frequently adopt a Janus-type structure with Rh mostly located at the interface between TiO2 and Au, as suggested by the Z-contrast of the images in Fig. 2b–d (Au atoms appear brighter than Rh atoms)20.

Figure 2
figure 2

AC-STEM characterization of supported Janus nanoparticles.

(a–d) HAADF images of AuRh/TiO2 (sample 2, Au63Rh37 average composition) pretreated in H2 at 350 °C. Metal particle size: 3.7 ± 1.0 nm. (e,f) Ball model resulting from the reconstruction of the image in (d), viewed along the [011] direction (e) and a tilted direction away from the zone axis (f).

A statistical analysis21,22,23 was performed on the high-resolution image of the particle shown in Fig. 2d. After determining the number of atoms for each atomic column, a simple back-projection was carried out (see details of the procedure in Supplementary Information, including Supplementary Figs S2–S4). The bright part of the particle in Fig. 2d is well modelled by an fcc truncated octahedron (the most stable structure for Au particles of a few nanometres)24,25 of ca. 1700 Au atoms with a (111) facet parallel to the titania surface (Fig. 2e,f). With respect to regular truncated octahedron, it is further truncated along a (111) plane by a thick layer of Rh (containing ca. 700 atoms from the atom-counting analysis). Thus, the hypothesis of a faceted Janus nanoparticle with Rh located at the interface between Au and TiO2 gives satisfactory results, with a relatively smooth transition between the surfaces of the Rh and Au parts, i.e. roughly equal numbers of Rh and Au atoms along the e-beam direction.

Supported nanoparticles heated to 700 °C

To assess the thermal stability of the catalyst structure above 350 °C, AuRh/TiO2 was submitted to constant heating up to 700 °C in H2 flow. After this treatment, in addition to pure Au NPs and smaller Rh NPs, the sample presents ca. 10 nm-sized bimetallic particles having a segregated structure with Rh or both metals connected to TiO2, and the Au side covering partially the Rh side. This is shown by the representative AC-STEM images and EDX maps of Fig. 3 and Supplementary Figs S5–S7. This roundish structure is referred to as “ball-cup”26 and noted RhballAucup later on. Unlike standard Janus particles, which have a planar (Au-Rh) interface, ball-cup ones have a curved interface with one metal (Rh) partly embedded in the other (Au). The large size of the NPs suggests that these particles may result from the thermally-induced coalescence of smaller ones.

Figure 3
figure 3

AC-STEM characterization of supported ball-cup nanoparticles.

(a) HAADF image of AuRh/TiO2 (sample 3, Au55Rh45 average composition) heated in H2 flow to 700 °C. (b) HAADF image of the region corresponding to the white square in (a). (c) EDX map of the region corresponding to the yellow square in (a). (d) HAADF image of an individual particle.

DFT modelling of free and supported clusters

In order to rationalize the STEM results showing metal segregation in unsupported and supported Au-Rh nanoparticles, a theoretical study was undertaken using DFT computer simulations (DFT-PAW-PBE method, VASP code27,28,29,30, see Methods section for details).

Unsupported clusters

79-atom clusters of truncated octahedral (TO) shape were considered as models for investigating the equilibrium mixing behaviour of Au-Rh nanoparticles. Several nanoalloy models were constructed, covering different compositions and morphologies, such as mixed (ordered alloy), core-shell, Janus and ball-cup particles. These general structural types were then extended to larger TO clusters (up to 260 atoms). The stabilities of various chemical configurations following local geometry optimization are compared in Fig. 4a (the mixing energy31 refers to the energy variation due to alloying with respect to the pure clusters, see also Table 1 for selected compositions). As a result, RhcoreAushell and AucoreRhshell are the most and least stable configurations, respectively. These findings are consistent with the higher cohesive and surface energies of Rh as compared to Au in the bulk state9,15. Figure 4a shows that the mixing energies of the intermediate structures between the core-shell and the corresponding pure metal particles lie on a straight line for both RhcoreAushell and AucoreRhshell. The second most stable structure is RhballAucup, the extreme end of this type being the Rh core covered with a half Rh and half Au shell. The Janus-type structures lie slightly higher in energy than the RhballAucup structures. The inverse AuballRhcup type, in which the core is Au-rich, is found to be less stable than the Janus type. When the size of the particle increases (up to 260 atoms), the same stability order is preserved32.

Table 1 Results of DFT calculations.
Figure 4
figure 4

Results of DFT calculations.

(a) Mixing energy versus atomic composition for 79-atom Au-Rh nanoalloy TO clusters and monometallic counterparts. The most stable cluster (Rh19@Au60) is enlarged in insert. (b) Structure and corresponding adsorption energy for clusters of selected compositions adsorbed on TiO2(110). Blue, yellow, cyan, and red spheres represent Rh, Au, Ti, and O atoms, respectively. Only one layer of the TiO2 slab is shown for simplicity. (c) Schematic view of the energetics of free and supported Au50Rh29 clusters.

Supported clusters

Then, Au, Rh and Au-Rh clusters were placed and locally relaxed on the rutile TiO2(110) surface, which is a model for the experimentally used TiO2 nanorods (that exhibit 80–90% (110) facets)15. The clusters were placed between bridging O rows of TiO2(110), which is the adsorption position maximizing metal-support interaction. Figure 4b reports the adsorption energies (energy gain due to metal-support interaction) of pure metal clusters and core-shell, Janus and ball-cup nanoalloys. The adsorption strength is higher for the Rh cluster than for the Au cluster one because the Rh-O interaction is stronger than the Au-O one6. Consistently, for bimetallic clusters, the adsorption energy is mainly determined by the type of metal atom in contact with the surface. By comparing the mixing energy of the free cluster (Δ) with its supported counterpart (Δ′) one can evaluate the effect of the support on the clusters, i.e. determine whether a particular mixing type is stabilized or destabilized. For Janus and RhballAucup structures, the mixing energy becomes negative when the clusters are adsorbed on the surface through Rh facets, while it becomes more positive for Au facets (see Fig. 4c and Table 1). Since there is no possibility for Rh-titania contact, RhcoreAushell is destabilized on the surface. To determine whether the destabilization of RhcoreAushell and the stabilization of Janus and RhballAucup through Rh-surface interactions are sufficient to cause a crossover in stability, we constructed Janus, RhballAucup and RhcoreAushell clusters with the same composition (Au50Rh29, i.e. close to experimental sample composition) and compared their total energies on the surface (Table 1). For this particular composition, RhballAucup becomes the lowest total energy structure, surpassing RhcoreAushell (see also Fig. 4c). For the Janus structure, although the stabilization is not enough for crossover, the energy gap to RhcoreAushell decreases from 7 eV to less than 1 eV upon adsorption.

In summary, while the Rh@Au core-shell structure is by far the most stable in the unsupported state according to DFT calculations, in the supported state this configuration competes with anisotropic segregated structures. Remarkably, the Janus (350 °C annealing in H2) and RhballAucup (700 °C) configurations experimentally observed for the bimetallic NPs are also predicted for small clusters at equilibrium. This suggests that these structures are close to equilibrium and the energetics is likely controlled by the positive Au-Rh mixing enthalpy and the difference in surface/interface energies rather than by the large lattice mismatch between Au and Rh15. Indeed, small clusters can easily accommodate the (small) strain whereas larger nanoparticles relax it through interfacial defects or non-epitaxial relationships between Au and Rh regions, as seen in Fig. 2d for the Janus structure and Fig. 3b for the ball-cup one. Consistent with our direct observation of Au/Rh/TiO2 stacking after low-temperature annealing, Han et al. suggested from DFT calculations on Au-Ir/TiO2 slabs the presence of Ir near the TiO2 surface, which would strengthen the metal adhesion and lead to the experimentally observed higher stability against sintering of (bulk-immiscible) Au-Ir NPs as compared to Au NPs33. The high-resolution AC-STEM images of Fig. 3(b,d) show that the RhballAucup structure, which had never been reported for any bimetallic system to our knowledge, is quite complex and should be further investigated in the future. The transition from the low-temperature structure to this ball-cup one might result from a thermally-activated partial dewetting of the NPs from the support together with their tendency to enrich their surface with gold.


Through the combination of model catalyst synthesis, AC-STEM-EDX characterization and DFT simulations, original nanostructures could be evidenced and investigated for the Au-Rh system supported on rutile titania. After mild annealing in hydrogen (350 °C) corresponding to typical catalyst activation and reaction conditions, a Janus structure with Rh in contact with the support is formed. This stacked chemical configuration (Au/Rh/TiO2) is driven by the Au-Rh demixing tendency, the lower surface energy of Au, and the preferential affinity of Rh with the substrate. After severe annealing (700 °C), a RhballAucup configuration, intermediate between Rh@Au core-shell and Janus, is preferred. These two types of structures are expected for other oxide-supported bimetallic systems and may lead to cooperative catalytic effects, such as those observed for AuRh/TiO2 in hydroprocessing reactions.


Experimental methods

TiO2 rutile nanorods were prepared using a simplified procedure based on a hydrothermal method reported by Li and Afanasiev34. 10 g of commercial Degussa P25 TiO2 (50 m2/g) and 100 mL of 15 wt% H2SO4 solution were mixed in a Teflon reactor and placed in a sealed autoclave kept at 200 °C for 15 days. The obtained solid was washed several times with 0.1 M NH4NO3 to remove adsorbed sulphate, then washed with distilled water, dried at 100 °C overnight, and calcined at 350 °C in air for 2 h.

Au-Rh NPs were prepared by a colloidal chemical (co)reduction route adapted from Toshima, Prati, Hutchings, and co-workers12,35,36,37,38. The metal precursors were HAuCl4.3H2O (Strem Chemicals, 99.9%, 49 wt% Au) and RhCl3.nH2O (Sigma-Aldrich, 99.9%, 38–40 wt% Rh). In a first step, a 200 mL aqueous solution containing the two metallic precursors was prepared by adding the amounts of precursors necessary for reaching a total metal loading of 3 wt%, with 50:50 at% Au:Rh composition. Next, a 1 wt% aqueous solution of a stabilizing agent, polyvinyl alcohol (PVA, Mw = 10,000) was added to the preceding solution while keeping always a mass ratio mPVA/mAu+Rh of 1.2. A solution of 0.1 M NaBH4, freshly prepared and kept at 0 °C before use, was then dropped under stirring to the metallic precursors solution with a molar ratio nNaBH4/nAu+Rh of 5. Stirring was then maintained for 30 min to allow the complete decomposition of the remaining NaBH4 excess. The solution was then acidified to pH 3.5 by addition of HCl 0.01 M in order to favour the sol immobilization onto the TiO2 support. The amount of support necessary for reaching the final metal loading was then added and stirring was kept for 3 h. Finally, the material was filterted, washed with hot distilled water (70 °C) several times, and dried at 100 °C overnight. An ex situ treatment consisting in heating the samples to 350 °C or 700 °C (10 °C/min, 3 h plateau) in hydrogen flow (5 mL/min, 1 atm) was applied using a dedicated bench.

The Au and Rh loadings of the Au-Rh/TiO2 catalysts were determined by inductively coupled plasma optical emission spectroscopy (ICP-OES) using an Activa spectrometer from Horiba Jobin Yvon. In order to dissolve them completely, the samples were treated with a mixture of H2SO4, aqua regia and HF at 250–300 °C. The measured metal loadings were 1.5–2.0 wt% for Au (target 2.0 wt%), and 0.6–0.7 wt% for Rh (target 1.0 wt%). As a result, the overall Au:Rh composition was comprised between 63:37 at% (samples 1 and 2; sample 2 is the supported counterpart of sample 1’s unsupported colloids) and 55:45 at% (sample 3). The unsupported and supported Au-Rh NPs are denoted AuRh@PVA and AuRh/TiO2, respectively, in the main text.

Scanning transmission electron microscopy (STEM) observations were conducted using a JEM-2100F (FEG, 200 kV) equipped with a probe aberration corrector, a high-angle annular dark field (HAADF) detector, and an energy dispersive X-ray (EDX) spectrometer. For STEM investigations, the dry samples were crushed in air using glass slides and were casted on copper TEM grids covered with holey carbon (supported nanoparticles) or lacey carbon (unsupported nanoparticles).

Theoretical methods

The calculations were performed using density functional theory (DFT) as implemented in the VASP code27. The generalized gradient approximation (GGA) was employed within the Perdew-Burke-Ernzerhof (PBE) parameterization for the exchange-correlation energy functional28. All the calculations were spin-polarized, with valence electrons treated explicitly, while the ionic cores were represented by the projected augmented wave (PAW) method29,30. To avoid spurious periodic interactions, unsupported clusters were placed into a sufficiently large supercell that ensures ~10 Å separation by vacuum. For the same reason, the supported cluster studies were carried out on a 3 × 3 TiO2(110) surface using a slab of 9 atomic layers. Although the lateral cluster separations are ~7 Å and ~9 Å on the 3 × 3 TiO2(110) surface, the total energy change was found to be less than 0.02 eV for both Au79 and Rh79 clusters. Therefore, there was no need to go to the larger surface cell size of 4 × 4 TiO2(110), which is computationally much more expensive. The Γ point was used to sample the Brillouin zone. All atoms except those of the bottom three atomic layers of the TiO2(110) slab were relaxed until the forces on the atoms became lower than 0.01 eV/Å, and the electronic ground states were determined by requiring a total energy convergence of 10−6 eV. For the stability comparisons, a mixing energy term was calculated31:

in which the total energy (Etot) of the nanoalloy AmBn is compared to those of the pure metal (A or B) clusters of the same size (m + n). Hence, a negative value of Δ means an energy decrease upon mixing, and therefore a more stable cluster. To determine the support effect on nanoalloy energetics, Δ′ was defined in the same manner as Δ by replacing the Etot values with the supported-cluster counterparts (Etot′). The adsorption energies were calculated using the total energy differences of the separated and combined cluster and support systems:

Additional Information

How to cite this article: Piccolo, L. et al. Understanding and controlling the structure and segregation behaviour of AuRh nanocatalysts. Sci. Rep. 6, 35226; doi: 10.1038/srep35226 (2016).