Article pubs.acs.org/accounts
Cite This: Acc. Chem. Res. XXXX, XXX, XXX−XXX
Single-Atom Alloys as a Reductionist Approach to the Rational Design of Heterogeneous Catalysts Georgios Giannakakis,† Maria Flytzani-Stephanopoulos,† and E. Charles H. Sykes*,‡ †
Downloaded via UNIV OF SOUTH AUSTRALIA on December 12, 2018 at 23:23:32 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.
Department of Chemical and Biological Engineering, Tufts University, 4 Colby Street, Medford, Massachusetts 02155 United States ‡ Department of Chemistry, Tufts University, 62 Talbot Avenue, Medford, Massachusetts 02155 United States CONSPECTUS: Heterogeneous catalysts are workhorses in the industrial production of most commodity and specialty chemicals, and have widespread energy and environmental applications, with the annual market value of the catalysts themselves reaching almost $20 billion in 2018. These catalysts are complex, comprising multicomponent materials and multiple structures, making their rational design challenging, if not impossible. Furthermore, typical active metals like Pt, Pd, and Rh are expensive and can be susceptible to poisoning by CO, coking, and they are not always 100% selective. Efforts to use these elements sparingly and improve their selectivity has led to recent identification of single-atom heterogeneous catalysts in which individual transition metal atoms anchored on oxide or carbon-based supports are excellent catalysts for reactions like the CO oxidation, water−gas shift, alcohol dehydrogenation, and steam reforming. In this Account, we describe a new class of single-atom heterogeneous catalysts, namely, Single-Atom Alloys (SAAs) that comprise catalytically active elements like Pt, Pd, and Ni alloyed in more inert host metals at the single-atom limit. These materials evolved by complementary surface science and scanning probe studies using single crystals, and catalytic evaluation of the corresponding alloy nanoparticles with compositions informed by the surface science findings. The well-defined nature of the active sites in SAAs makes accurate modeling with theory relatively easy, enabling the rational design of SAA catalysts via a complementary three-prong approach, encompassing surface science model catalysts, theory, and real catalyst synthesis and testing under industrially relevant conditions. SAAs constitute one of just a few examples of when heterogeneous catalyst design has been guided by an understanding of fundamental surface processes. The Account starts by describing scanning tunneling microscopy studies of highly dilute alloys formed by doping small amounts of a catalytically active element into a more inert host metal. We first discuss hydrogenation reactions in which dissociation of H2 is often rate limiting. Results indicate how the SAA geometry allows the transition state and the binding site of the reaction intermediates to be decoupled, which enables both facile dissociation of reactants and weak binding of intermediates, two key factors for efficient and selective catalysis. These results were exploited to design the first PtCu SAA hydrogenation catalysts which showed high selectivity, stability and resistance to poisoning in industrially relevant hydrogenation reactions, such as the selective conversion of butadiene to butenes. Model studies also revealed spillover of hydrogen atoms from the Pt site where dissociation of H2 occurs to Cu sites where selective hydrogenation is facilitated in a bifunctional manner. We then discuss selective dehydrogenations on SAAs demonstrating that they enable efficient C−H activation, while being resistant to coking that plagues typical Pt catalysts. SAA PtCu nanoparticle catalysts showed excellent stability in butane dehydrogenation for days-on-stream at 400 °C. Another advantage of SAA catalysts is that on many alloy combinations CO, a common catalyst poison, binds more weakly to the alloy than the pure metal. We conclude by discussing recent theory results that predict the energetics of many key reaction steps on a wide range of SAAs and the exciting possibilities this reductionist approach to heterogeneous catalysis offers for the rational design of new catalysts.
■
INTRODUCTION
opportunities exist for increased efficiency, selectivity, and activity of the chemical and environmental catalysts currently in use. Selectivity to desired products is often less than 100%, and catalyst deactivation due to coking or poisoning is a problem for many industrial catalysts. Rational design of
Chemical catalysts are the backbone of the chemical and petroleum industries, with an annual market of the catalysts themselves of more than $20 billion in 2018. Catalyst technology is responsible for the production of an estimated >$10 trillion of goods in the petroleum, power, chemicals, and food industries annually.1 Despite their widespread use, there are still many challenges that must be addressed, but great © XXXX American Chemical Society
Received: September 27, 2018
A
DOI: 10.1021/acs.accounts.8b00490 Acc. Chem. Res. XXXX, XXX, XXX−XXX
Article
Accounts of Chemical Research
Figure 1. (A) Typical volcano plot, describing the O2 reduction activity (y-axis) as a function of oxygen’s binding energy (x-axis). Reproduced with permission from ref 13. Copyright 2002 Elsevier. (B) Activation energies for N2, CO, NO, and O2 dissociation on metals surfaces plotted against the dissociative chemisorption potential energy. Reproduced with permission from ref 12. Copyright 2004 American Chemical Society (C) STM image of a single-atom alloy Pd/Cu(111) alloy with 0.01 ML Pd. Reproduced with permission from ref 15. Copyright 2013 PCCP Owner Societies.
ior by relating reaction energy (or the binding energy of the reaction intermediates) to the activation barrier (Figure 1B). Linear scaling between these two variables for both metal and alloy catalysts is very useful as a means to understanding trends in catalysis, but, implicitly limits catalytic performance, as one cannot have both low reaction barriers and weak binding of intermediates within this theoretical framework.14 The Mavrikakis group first predicted theoretically that it is possible to break the linear scaling in certain alloys termed Near Surface Alloys (NSAs) which lower the H2 dissociation barrier while maintaining weak binding of H atoms.14 This has guided experimental work, and a few examples now exist where NSAs have been used successfully under reaction conditions.16 However, a main issue is instability of these layered systems due to entropy or by the reactants causing surface segregation17 or reverse segregation.18
heterogeneous catalysts is typically difficult due to the compositional and structural complexity of the industrial catalytic systems. Furthermore, the reactor temperature, pressure, inlet flow rate, and gas composition are only few of the macroscopic variables that must be optimized. Last but not least, from an economical perspective, the scarcity and cost of the most promising catalytic elements (Pt, Pd, Rh, etc.) tend to limit their widespread use. This has been a motivator for the emergence of single metal atom catalysts, which, if applicable to a reaction and stable under working conditions, offer 100% atom efficiency by design. Besides the aforementioned issues, the particle size, shape, and coordination of the catalytic sites themselves leads to much additional complexity that has largely prevented the development of catalysts by using simple design principles. The need for a reductionist approach is apparent and was first addressed through the use of single atoms supported on metal oxides. The groups of Flytzani-Stephanopoulos,2 Gates,3 and Zhang4 have demonstrated the utility of this approach which not only reduces the number of parameters such as inhomogeneity of the particles, and local surface effects, but also elucidates how single metal atoms coordinated with −O, −N, −Cl, or other linkers act as the active sites. An alternative reductionist approach involves the stabilization of individual, isolated metal atoms in the surface of another metal host. Given the greater simplicity of metals vs oxides, such structures can be facilely produced, replicated, and employed for mechanistic investigations. The Sykes group used surface science and scanning probe techniques to develop Single-Atom Alloys (SAAs), first demonstrating facile hydrogen dissociation, spillover, and reaction on Pd-doped Cu surfaces.5−9 While dilute metal alloys have been studied by many groups,10,11 this work was the first to unambiguously demonstrate surfaces with only isolated atoms and link this structure directly to chemical reactivity. Significantly, because dissociation and reaction sites on SAAs are decoupled, SAAs escape linear scaling relationships, which limit the reactivity and selectivity of many catalysts. The >100 year old Sabatier principle12 dictates that elements like Au do not bind molecules strongly and have high reaction barriers, making reactions rate-limited by dissociative adsorption. Other metals like Fe bind molecules so strongly that while dissociation steps are facile, buildup of reaction intermediates like O or C lead to deactivation and desorption rate limits the reaction. The resulting Sabatier or “volcano plot” (Figure 1A) shows optimum catalytic activity for a few metals (typically the expensive ones like Pd, Pd, and Rh) clustered at its peak, which provide a compromise between strong and weak binding. The Brønsted−Evans−Polanyi relationship13 underpins this behav-
■
ATOMIC-SCALE STRUCTURE OF SAAs Scanning tunneling microscopy (STM), with its ∼0.1 nm x/y and ∼1 pm z resolution is an ideal way to study the atomic geometry of metal alloy surfaces.19 The Sykes lab first used this capability, coupled with techniques like temperature-programmed desorption (TPD), to ask the question: are single atoms of Pd in a more inert Cu metal catalytically active?6 These SAA surfaces are prepared by starting with very clean single crystal surfaces of metals like Cu, Au, and Ag with impurity levels less than 1 per 10,000 atoms. Purity of the host surface is a critical factor when studying ∼1% of a dopant atom. Au, Ag, and Cu metals are chosen as hosts because they weakly bind most species and hence can facilitate selective chemistry. Small amounts of more reactive species like Pt, Pd, and Ni are then added to the (111) facet of these hosts metals by electron beam evaporation in UHV.9 The (111) facet is typically studied because it is generally the most common facet on metal nanoparticles. The PdCu case is typical in that alloying occurs via incorporation of the Pd atoms in the surface layer, subsurface layer or Cu bulk depending on the alloying temparture.6 Figure 1C shows an example of STM imaging of SAA surfaces in which single Pd atoms are substituted into the surface layer of Cu(111). While the high contrast of the Pd atom vs Cu host may suggest that the Pd atom is on, not in, the surface, the apparent height of the Pd is only ∼10−20 pm above Cu, indicating that the Pd atoms are present in the surface layer.6 Alloying at higher temperatures leads the Pd atoms into the subsurface, because of the higher surface free energy of Pd than Cu.20 Therefore, under vacuum Pd prefers to be surrounded by copper atoms, so thermodynamically it is more favorable for Pd to diffuse into the subsurface layers. However, as we show later, under reaction conditions, B
DOI: 10.1021/acs.accounts.8b00490 Acc. Chem. Res. XXXX, XXX, XXX−XXX
Article
Accounts of Chemical Research
Figure 2. (A) H2 atom dissociation energetics on bare Cu(111), Pd-doped Cu(111), and bare Pd(111) surfaces. (B) Initial physisorbed state of H2 on PdCu SAA, (C) Transition state of dissociation and (D) final state. Reproduced with permission from refs 26 and 27. Copyright 2018 American Chemical Society.
microscopic reversibility dictates that they have to overcome a recombination barrier of ∼0.4 eV to desorb from the surface, which happens at ∼300 K, a value almost identical to that of Pd. Again, we see, as dictated by linear scaling, elements that bind weakly (Cu) have high dissociation barriers whereas elements with low barriers bind reaction intermediates (H atoms) more strongly. However, and most strikingly, when 1% Pd atoms are present in the Cu(111) surface, H2 is easily activated and desorbs at ∼210 K, 100 K lower than on each of the parent metals.8,24,25 The reason behind this effect is that, at the transition state (Figure 2C), the hydrogen molecule sits on top of the Pd atom and, therefore, the surface demonstrates a Pd-like chemistry during dissociation, which leads to the low dissociation barrier that is also aided by quantum tunneling effects that were previously reported.24,25 After dissociation, the H atoms prefer to sit in 3-fold hollow sites beside the Pd atom consisting of two Cu atoms and one Pd atom (Figure 2D), therefore having a more Cu-like behavior and thus weak binding. Hence, linear scaling is broken as H2 can both dissociate easily on SAA surfaces (as seen in Figure 2A) and bind weakly. STM and TPD experiments revealed that hydrogen atoms spill over from the Pd sites and populate the whole Cu(111) surface. While slightly endothermic, this spillover, favored by entropy, further weakens the binding of H on Cu, making selective chemistry a promising option for SAAs. To this end, model 1% monolayer Pd/Cu(111) SAAs were tested for hydrogenation reactions (styrene, acetylene)9 and were compared to Pd-rich model catalysts with extended Pd ensembles created by depositing a monolayer of Pd on the Cu(111) surface. These extended Pd domains dissociate not only H2 but also C−H and C−C bonds, essentially decomposing most of the reactant, as evidenced by hydrogen produced at high temperatures. On the contrary, 1% Pd SAAs ensures that Pd is dispersed as single atoms, which are not capable of breaking C−C bonds, since two adjacent Pd atoms are needed to break this bond. Instead, molecules like acetylene are hydrogenated, with an overall selectivity reaching
adsorbates like H and hydrocarbons serve to stabilize the dopant atoms at the surface. In this sense, alloy surfaces can be very dynamic at and above room temperature.17,18,21 Another important result from these STM studies is that at low coverages only individual, isolated Pd atoms exist in the surface. Only at 10% of an equivalent Pd monolayer do significant amounts of dimers appear (∼1 dimer per 10 monomers). The reason behind this very high dispersion is the negative mixing enthalpy of Pd in Cu, which means that Pd atoms would rather be surrounded by Cu than bond to other Pd atoms. In this way, thermodynamics drive the dispersion, which enables the formation of stable SAAs. The obvious advantage of the SAA approach is that the dopant atom (typically expensive elements like Pt and Pd) is used with ultimate atom efficiency.
■
REACTIVITY OF SAA MODEL CATALYSTS To probe the chemical reactivity of SAAs, TPD was employed. In terms of hydrogenation reactions, Cu is not particularly active due to its relatively high activation barrier and inability to dissociate molecular hydrogen. In order to study H atoms on Cu, an energetic molecular beam or H atom source is required.22 However, once atomic hydrogen is on Cu(111), it is stable and the surface must be heated to ∼310 K for the hydrogen to recombine and desorb.22 On the other hand, Pd is a much more catalytically active element than Cu, and can dissociate molecular hydrogen even at temperatures below 100 K. The barrier for hydrogen dissociation is practically nonexistent; in this case, hydrogen dissociates readily. Despite this difference, the Pd(111) surface must still be heated to ∼320 K, practically the same temperature as Cu(111) for the hydrogen atoms to recombine and desorb.23 Naively, one would think that Cu and Pd are very similar with respect to the energetics of their interaction with hydrogen, but this picture is incorrect; the real reason for the near identical desorption temperature is hydrogen is kinetically trapped on Cu(111) vs on Pd where it is bound much more strongly as shown in Figure 2A. Despite the weak binding of H atoms on Cu, C
DOI: 10.1021/acs.accounts.8b00490 Acc. Chem. Res. XXXX, XXX, XXX−XXX
Article
Accounts of Chemical Research 100%. This first proof of principle study highlights the utility of the surface science approach to quantify active site structure, dissociation and spillover energetics and relate these quantities to chemical reactivity. By dissociating H2 at one site with a low barrier and then spilling over to a second site where it is weakly bound linear scaling can be broken and selective hydrogenation chemistry enabled. This promising chemistry is not limited to Pd and has also been demonstrated for Pt SAAs. Platinum group metal catalysts are widely used in fuel cells and automotive catalytic converters, but also for industrial hydrocarbon processing, due to their excellent hydrogenation28 and dehydrogenation29 reactivities. Major drawbacks of Pt catalysts include their susceptibility to CO poisoning, deactivation via coke formation, and their lower selectivity in hydrogenation reactions due to strong adsorption of the reaction intermediates. The selective hydrogenation of butadiene to butene was chosen as a probe reaction to test SAA PtCu performance, for which PtCu31 and PdAu30 alloys exist and exhibit high but not 100% selectivity toward butene.30 Figure 3B shows a STM
layer, where ensemble effects become dominant, selectivity is greatly reduced. For example, when 30% ML Pt is present, 1D chains of Pt dominate the surface as seen in the STM image in Figure 3D. The corresponding TPD spectra elucidate the ability that these adjacent atoms have to break C−C bonds (Figure 3C), unlike the isolated Pt atoms. The hydrogen signal at high temperatures derives from the breaking of C−H bonds and the decomposition of the molecules through C−C bond cleavage and C deposition on the surface. This experiment highlights the point that in dilute metal alloys surface ensembles can dominate the chemistry in a very detrimental manner, even in the sub-monolayer regime. Bifunctionality offers a way to break linear scaling in heterogeneous catalysis. If one can decouple dissociation and reaction sites, linear scaling may be broken because one has two sets of activation barriers and two sets of binding energies. However, it is notoriously hard to demonstrate if a working heterogeneous catalyst is indeed bifunctional. Using welldefined model systems this problem becomes tractable. For example, in the case of hydrogenation of butadiene on Pt/ Cu(111) SAAs, CO can be used to block one of the surface sites and study the effect on reactivity. Specifically, a Pt-doped Cu surface is first exposed to H2 which dissociates and spills over on the Cu(111) surface. Sequentially, CO is deposited and preferentially binds on every Pt atom. This setup can be directly visualized in the STM images in Figure 4. There, the stationary dark spots are the CO on the Pt sites and the streaky features are mobile H atoms on the Cu surface. By scanning at elevated bias (200 mV) the mobile H can be swept away leaving only the CO which can then be removed via 5 V pulses, after which in panel C one can see a single Pt atom beneath each CO adsorption site. This demonstrates that Pt sites can be selectively blocked with CO and allows for the question of where the hydrogenation of butadiene takes place to be answered. The comparison of the TPD spectra between COfree (Figure 4E) and CO-blocked (Figure 4F) Pt sites reveals an identical product distribution of butenes. Furthermore, if the Pt sites are blocked with CO before H2 is added no reactivity is observed. These experiments definitively show that the Pt atoms dissociate H2 while Cu catalyzes the hydrogenation of butadiene to butane; the model SAA surface is indeed a bifunctional hydrogenation catalyst that escapes linear scaling (Figure 4D).
■
SYNTHESIS, CHARACTERIZATION, AND TESTING OF NANOPARTICLE SAA CATALYSTS The surface science and scanning probe studies used to realize SAAs showed many promising attributes of SAAs from high atom efficiency, escape from linear scaling relationships, tolerance to CO and carbon buildup, to a range of selective hydrogenation chemistries. The challenge then became synthesizing the same materials in nanoparticle form and testing their efficacy under industrially relevant reaction conditions. For this purpose, approximately 1 atom % of the minority metal (Pd, Pt) was deposited in inert metal (Cu, Au, Ag) nanoparticle surfaces prepared in supported or unsupported (nanoporous) form. For example, Pt0.01Cu/Al2O3 catalysts were synthesized via the galvanic replacement reaction and a series of characterization techniques were employed to verify the atomic dispersion of Pt in the Cu surface.32 Care was taken to use the purest precursor compounds (e.g., 99.999% Cu) for the host metals so that small levels of impurities did not dominate the chemistry.
Figure 3. TPD spectra showing butadiene hydrogenation over (A) 2%Pt/Cu(111) and (C) 30% Pt/Cu(111). Corresponding STM images of (B) 2% Pt/Cu(111) and (D) 30% Pt/Cu(111). Reproduced with permission from ref 32. Copyright 2015 Springer Nature.
image of a PtCu SAA surface prepared in a similar way to PdCu and corresponding TPD experiments on butane hydrogenation (Figure 3A). In a single adsorption/desorption step, 25% conversion and 100% selectivity to butenes are observed.32 The 100% selectivity is explained by butene desorbing easier than butadiene, and thus not further hydrogenating to butane.32 Consecutive cycles of desorption experiments show that the surface reactivity remains constant after 18 TPD cycles, again showing the robust nature of SAAs against poisoning. At higher concentrations of Pt in the surface D
DOI: 10.1021/acs.accounts.8b00490 Acc. Chem. Res. XXXX, XXX, XXX−XXX
Article
Accounts of Chemical Research
Figure 4. STM images of (A) H atoms and CO molecules on Pt/Cu(111) SAA. (B) Same area after H removal. (C) Same area after CO removal. Reproduced with permission from ref 33. Copyright 2016 American Chemical Society (D) Bifunctional reaction mechanism. (E) TPD spectra after butadiene hydrogenation over 3% Pt/Cu(111) CO-free surface and (F) CO capped Pt. Reproduced with permission from ref 32. Copyright 2015 Springer Nature.
Figure 5. (A) HAADF-STEM image and (B) CO-DRIFT spectra of 1% PtCu SAA NPs. (C) Temperature-programmed butadiene hydrogenation as a function of temperature over pure Cu NPs vs PtCu SAA NPs. (D) Steady state butadiene hydrogenation in the presence of excess propylene on Pt0.1Cu14/Al2O3. Reproduced with permission from ref 32. Copyright 2015 Springer Nature.
STEM imaging allows elements with a high Z difference to be distinguished, and indeed, individual Pt atoms are clearly visible in Figure 5A. Stronger proof that these nanoparticles are SAAs came from the absence of Pt−Pt coordination obtained via extended X-ray absorption fine structure measurements.32
Further corroboration was provided by DRIFTS using CO as a probe molecule, where only CO binding atop Pt was observed, and no bridged CO was present, indicating the absence of Pt− Pt ensembles at the surface32 (Figure 5B). E
DOI: 10.1021/acs.accounts.8b00490 Acc. Chem. Res. XXXX, XXX, XXX−XXX
Article
Accounts of Chemical Research
allows for the comparative study of C−H activation in methyl groups on different surfaces. Cu is known to be a good C−C coupling catalyst but product evolution (methane, ethane, ethene, propene) is limited in TPD experiments by C−H activation, which on pure Cu(111) occurs at ∼450 K. Pure Pt(111), on the other hand, can activate C−H bonds readily at ∼250 K, but strong binding leads to coking and deactivation. Pt/Cu SAA surfaces share the best properties of both elements as the same products seen from pure Cu evolved 100 K lower on the SAA without any carbon deposition. One advantage of the surface science approach is that STM can be used to image the reactants, intermediates, and occasionally the products of a chemical reaction as well as their location with respect to surface active sites. Figure 7A and B shows STM images of both Cu(111) and Pt/Cu(111) SAA surfaces after CH3I deposition and annealing to 350 K. On Cu(111) C−H activation has not yet occurred and a mix of CH3 groups and I atoms are seen on the surface. In contrast, on the 1% Pt/Cu(111) SAA, annealing to 350 K leads to the complete desorption of the hydrocarbon products leaving only the I atoms. The CH3 species are mobile on the surface and each Pt atom in the surface catalyzes multiple C−H cleavages. Based on the above, it is evident that PtCu SAA effectively lowers the barrier for C−H activation, while maintaining the high C−C coupling ability of the Cu host and its resilience to coking (Figure 7C). These promising UHV results were translated to ambient conditions and H−D scrambling in butane was used as a test of the ability of PtCu SAA nanoparticles to activate C−H bonds. Specifically, a mixture of butane and deuterium was passed through a flow reactor and incorporation of deuterium in butane was seen in the mass spectrometer as a m/z = 59 signal indicating C−H activation. On pure Cu nanoparticles, the H− D exchange light off temperature is 550 °C, while addition of just 1% Pt to the Cu dramatically lowers the light off temperature to 250 °C (Figure 7D), demonstrating that single Pt atoms can activate C−H bonds. Similar to typical Pt catalysts, Pt nanoparticles were found to deactivate after the first cycle and suffer from coke buildup (Figure 7E). The PtCu SAA NPs were effective for the industrially relevant butane conversion to butene and hydrogen and long-term activity/ stability tests showed same activity after 2 days on-stream (Figure 7F). Postreaction EXAFS and DRIFTS verified the high atomic dispersion of Pt, and no signs of restructuring were observed. Once again, due to the negative mixing enthalpy of Pt and Cu35 and the fact that Pt sits preferentially in the surface layer in the presence of hydrocarbons and hydrogen,36 SAAs are able to bridge both the pressure and structure gap between UHV studies and real world catalysis. In terms of reaction types, while the Cu-based SAA catalysts discussed in the paper as well as NiCu38,41 are used under reducing conditions (hydrogenations, dehydrogenations, C−H bond activation), Au-based SAAs (e.g., NiAu, PdAu) are suitable for use under oxidizing reaction conditions.37 Promising results from the coupling of surface science and real SAA catalysis have also been reported for the dry, nonoxidative dehydrogenation of alcohols, a class of reactions of great industrial significance, owing to the increased value of the corresponding aldehydes. The production of hydrogen instead of water (formed under oxidative conditions) eliminates the cost of separation, also yielding valuable H2. Another benefit of the SAA approach is that the stability issues of the state-of-the-art Cu-based catalysts used in ethanol
These novel SAA nanoparticles were tested under butadiene hydrogenation conditions. Figure 5C shows that addition of just 1−2% Pt led to a 50 °C lower light-off temperature for the hydrogenation reaction without compromising the selectivity, even at higher than 80% conversion, due to the ability of Pt to dissociate molecular hydrogen, the rate limiting step on Cu. Further steps to simulate industrially relevant conditions were taken, where excess propylene was also introduced in the mixture to simulate typical conditions required to convert butadiene to butenes in the propylene prepolymerization step that prevents deactivation of the catalyst. Stability tests found that almost all the butadiene is converted to butenes, while propylene is only slightly converted (
