Perovskites in catalysis and electrocatalysis

See allHide authors and affiliations

Science  10 Nov 2017:
Vol. 358, Issue 6364, pp. 751-756
DOI: 10.1126/science.aam7092


Catalysts for chemical and electrochemical reactions underpin many aspects of modern technology and industry, from energy storage and conversion to toxic emissions abatement to chemical and materials synthesis. This role necessitates the design of highly active, stable, yet earth-abundant heterogeneous catalysts. In this Review, we present the perovskite oxide family as a basis for developing such catalysts for (electro)chemical conversions spanning carbon, nitrogen, and oxygen chemistries. A framework for rationalizing activity trends and guiding perovskite oxide catalyst design is described, followed by illustrations of how a robust understanding of perovskite electronic structure provides fundamental insights into activity, stability, and mechanism in oxygen electrocatalysis. We conclude by outlining how these insights open experimental and computational opportunities to expand the compositional and chemical reaction space for next-generation perovskite catalysts.

Catalysis is essential for addressing some of the most pressing societal and environmental challenges, ranging from the reduction of fossil fuel emissions to the production of sustainable fuels and chemicals (1). The use of fossil fuels in power plants and transportation to meet increasing global energy demands leads to increased emissions of molecules that are harmful to human health (e.g., CO and NOx) and greenhouse gases (such as CO2) into the atmosphere. Catalysis is central to efficiently converting harmful emissions to benign molecules such as H2O (2, 3) or N2 (4). Reducing CO2 emissions requires replacing fossil fuels with solar energy, in combination with using energy storage devices (5), capturing and sequestering CO2 (6, 7), or both. Electrochemically storing renewable or electron energy in materials and molecules can solve the problem of energy supply intermittency between day and night in the current solar infrastructure. Although lithium ion batteries are being adopted for energy storage in transportation and stationary applications today (8, 9), electrochemical production of materials and molecules that can store greater energy per unit weight or volume is much needed for storage at scale. Electrochemical reduction of H2O, metal oxides, N2, and CO2 (10) to make energy carriers or sustainable fuels (H2, metals, NH3, and CxHyOz species including CO, alcohols, and hydrocarbons) would not only provide large-scale and seasonal energy storage, but also enable distributed energy generation for transport and buildings and distributed synthesis of chemicals and materials (1).

Unfortunately, the efficiency of such electrochemical devices is low because of the slow kinetics of key reactions (11). For example, although electrochemically produced CO and HCOOH (formic acid) can be competitive with chemically synthesized products (12), liquid fuels such as methanol are still challenging to produce at market costs (13). Improving the efficiency of these devices to produce clean air and sustainable fuels would require (electro)catalysts that consist of earth-abundant elements and have activity and stability far greater than that of current catalysts (14).

Catalysis is an interdisciplinary area, where physics and chemistry converge to allow mechanistic understanding of reaction kinetics (15). Catalysts stabilize reaction intermediates along a particular reaction pathway and, in the simplest description, speed up the reaction by reducing the energy barrier of the short-lived transition state, the maximum energy state between two reaction intermediates. However, because the energy of the transition state is determined by the reaction intermediate energies, known as linear free energy relations (16, 17), these intermediate energies primarily determine the reaction kinetics of the pathway. Recent developments in catalyst design principles called activity descriptors (1822)—parameters that can govern activity over many orders of magnitude—and the combination of materials science, computational science, and inorganic chemistry allow rapid catalyst discoveries (18).

Although noble metal–based catalysts are used widely for catalysis of air pollutants (e.g., Pt, Pd, and Rh) and electrocatalysis of oxygen reduction (e.g., Pt) in fuel cells (23), inorganic, noble metal–free materials that have high activity, selectivity, and stability are essential to meet global environmental and energy needs (24). The search for earth-abundant and cheap catalysts has brought perovskites to the forefront for catalyzing relevant reactions such as the oxidization of toxic hydrocarbons and CO (25), reduction of NOx (24) for the treatment of automotive gas exhaust and environmental clean air applications (26), and oxygen electrocatalysis to electrochemically enable efficient generation and use of sustainable fuels (19, 20, 27). In this Review, we discuss the versatility of perovskites for catalyzing reactions relevant to the abatement of air pollutant emissions and the storage and conversion of sustainable fuels.

Perovskite structure and properties

Perovskites have a general formula of ABX3, where smaller transition metal ions on the B site reside in corner-sharing octahedra of X anions, and larger A-site cations have 12-fold coordination with X (Fig. 1). Calcium titanate (CaTiO3) was the first mineral discovered in this structural family, which was named “perovskite” after the Russian mineralogist Lev Perovski (28). Owing to the flexible electronic structure of the perovskite oxide family, represented by the distribution of electronic states [or density of states (DOS)] (Fig. 1), they exhibit diverse physical and chemical properties in bulk, including ferroelectricity in PbTiO3 (29), colossal magnetoresistance in LaxCa1−xMnO3 (30), ion conductivity for protons in (Ba,Sr)CeO3 (31), lithium mobility in (Li,La)TiO3 (32), anion-intercalation pseudocapacitors in LaMnO3–δ (33), and oxygen ion diffusion in LaGaO3 for solid-state devices such as solid oxide fuel cells (SOFCs) (34).

Fig. 1 Electronic and chemical structure of perovskites.

(Center right) ABO3 perovskite crystal structure. (Center left) The electronic DOS contributions from the oxygen (O 2p) and metal (B 3d) states, which make up perovskite electronic structure. (Outer) Applications for catalytic processes (24), bulk ion diffusion for gas sensors and fuel cells (89), solid-state ferroelectric devices (29), and superconducting properties (36, 37).

Moreover, enhanced and exotic properties of perovskites have been discovered by introducing strain and heterointerfaces (35). For example, high mobility and superconductivity in a two-dimensional (2D) electron gas was unexpectedly discovered at the atomically sharp interface between thin films of LaAlO3 grown on SrTiO3, both of which are insulating materials in bulk (36, 37). Furthermore, perovskite surfaces can catalyze several reactions (Fig. 2), including oxidation of small molecules such as CO (38, 39), hydrocarbons (40) and NOx (26), (photo)electrochemical splitting of H2O (20, 41), and reduction of CO2 (42), N2 (43), and O2 (19). The flexibility of the electronic and crystal structure and chemical versatility of ABO3 perovskites can be used to establish design principles for highly active, selective, and stable catalysts. Through careful materials design in perovskites, electronic structure can be tailored to the thermodynamic energies of a variety of reactions (Fig. 2) to minimize electrochemical and chemical reaction barriers.

Fig. 2 Chemical and electrochemical reactions involving perovskite oxides.

(Top) Electronic structure and thermodynamic reaction energies at pH 13 aligned on an absolute energy scale. (Bottom) Numerous (electro)chemical reactions relevant for perovskites (19, 20, 26, 3843).

Molecular orbital descriptors in catalysis

Mechanistic studies of the oxidation of CO at metal surfaces (44, 45) have contributed to the fundamental understanding of catalysis. Not only is this reaction of technological importance, but it has also served as a model for understanding the thermal and electrochemical oxidation of hydrocarbons (46) and alcohols (47). Insights from studies of Pt, Pd, or Rh supported on oxides (44, 48)—model systems for catalyzing CO oxidation—suggest that CO oxidation on metals proceeds by the Langmuir-Hinshelwood mechanism, in which surface-adsorbed CO and dissociatively adsorbed O2 react on the surface to form CO2, which then desorbs (48). The rates depend on the surface-binding energies of CO and O2, where the most active catalyst binds the reaction intermediates neither too strongly nor weakly (48). This concept on a general level is known as Sabatier’s principle of catalysis (1720). As a result, the surface-adsorbate binding is critical for discussing the catalytic activity of perovskites.

The nature of the metal-oxygen bond in the perovskite structure provides a basis for tuning perovskite electronic structure to control surface-binding energetics. A coordination chemistry approach is useful for describing these bonds. In the bulk perovskite structure, the B-site metal binds with six O atoms in an octahedral BO6 coordination. The transition metal d and oxygen 2p atomic orbitals hybridize, or mix, to form σ orbitals (Embedded Image and Embedded Image metal orbitals have high spatial overlap with oxygen) and π orbitals (dxy, dyz, and dxz metal orbitals have low spatial overlap with oxygen). The σ and π antibonding (σ* and π*) orbitals are known as eg and Embedded Image orbitals in perovskites, respectively, with the σ* orbital being more destabilized, or higher, in energy. The degree of stabilization depends on the symmetry of the atomic orbital overlap, called the crystal field stabilization. Similar concepts can be applied to the perovskite surface. At the surface layer, exposed B sites have the coordination environment BO5, with the apical (vertical) oxygen removed (Fig. 3A). This geometry further breaks symmetry, splitting the eg and Embedded Image states into distinct energy levels.

Fig. 3 Eg orbital filling of perovskites and catalytic reactions.

(A) Electronic configuration and relevant metal orbitals of first-row transition metals for a BO5 configuration. The (B) CO (g) σ orbital (left), (C) NO (g) π* orbital, and (D) O2 (g) π* orbital donate electrons to the empty eg orbital. Simultaneously, the π* orbital of CO (g) receives electrons from the occupied perovskite Embedded Image orbital in (B). (E) Correlation of perovskite eg occupation with catalytic activity for CO, propene (C3H6), and isobutylene (C4H8) oxidation (56, 57, 59) (F). NO oxidation for LaxMnO3+δ (orange) and La1–xSrxCoO3 (blue-green) (61, 62). (G) Electrochemical aqueous oxygen reduction reaction (ORR) (19). Activity metrics for CO oxidation, propene and isobutylene oxidation, NO oxidation, and ORR are the inverse temperature (1000 K–1) required to achieve a 1 μmol m−2 s−1 conversion rate, conversion rate (in micromoles per meter squared per second) at 573 K, conversion rate (in micromoles per meter squared per second) at 500 K, and overpotential to achieve a current density of 25 μA cm−2, respectively. Chemistries shown are B = Cr (yellow), Mn (orange), Co (blue-green), Fe (green), Ni (white), and mixed (purple). [Adapted with permission from (19, 57, 59, 61, 62)]

These concepts describe the relevant frontier orbitals of perovskite chemistries that dictate surface reactivity and catalytic activity (49). Specifically, eg-parentage orbitals of the active transition-metal B-site ion in the BO5 coordination capture surface-adsorbate bonding well because of their favorable vertical orientation toward surface-bound intermediates. As shown in Fig. 3A, the filling of these eg-like states depends on the number of d electrons and the spin state of transition metal ions. For first-row transition metals, the corresponding eg occupancies for Cr3+, Mn3+, Fe3+, Co3+, and Ni3+ are 0, 1, 2, 1, and 1, respectively (Fig. 3A) (50). When CO, NO, or O2 adsorbs on the surface B sites end-on (Fig. 3, B to D) (51), the relevant adsorbate molecules interact with vertically oriented eg-like orbitals, which allows for stronger overlap than with Embedded Image states. These interactions can determine the energy gained by adsorption and desorption of adsorbates on B ions.

Specifically for the CO molecule, a lone-pair σ orbital can donate electrons to the perovskite eg orbital while the CO π* orbital receives back electrons from the Embedded Image orbitals (Fig. 3B) (52). NO and O2 molecules prefer a tilted configuration, allowing their π* orbital to overlap with the B-site eg orbital (Fig. 3, C and D). As a result, the eg occupancy approximates the strength of adsorbate binding to the surface, with decreased eg filling corresponding with increased adsorbate binding. These relations connect perovskites’ bulk chemistry to their surface reactivity and catalytic activity in the search for catalysts with optimal binding strengths.

Molecular orbital descriptors and catalytic activity

These coordination chemistry concepts provide useful rationales for determining activity trends. The surface interactions of CO, NO, and O2 can be extended to three highly relevant model reactions encompassing clean air and electrochemical energy storage applications—namely, CO and hydrocarbon oxidation, NO oxidation, and oxygen electrocatalysis.

The basis for understanding CO oxidation kinetics on oxides originates from the Langmuir-Hinshelwood scheme for metal surfaces (48); on oxide surfaces, adsorbed CO on a metal site combines with a neighboring oxide O to form CO2 and a surface O vacancy, resulting in a vacancy-mediated mechanism referred to as the Mars–van Krevelen mechanism (53, 54). This framework on perovskites is supported by density functional theory (DFT) calculations for Co-substituted SrTiO3 showing that low oxygen vacancy formation can trigger the Mars–van Krevelen–type mechanism for perovskites (55). Energetically, the reaction kinetics are controlled by the binding strength of oxygen and CO on the surface, which depends on the eg filling of the transition metal ion (49). Voorhoeve et al. showed that the CO oxidation activity trend for perovskites exhibits a volcano shape as a function of B-site eg filling [LaBO3, La0.85Sr0.15CoO3, and La0.7Pb0.3MnO3); B = a first-row transition metal] (56, 57). Reaction conditions with a 2:1 CO:O2 ratio at atmospheric pressure and relatively low temperatures (100° to 300°C) were measured (Fig. 3E). Decreasing the eg filling from 2 to 1 increased the activity, and LaCoO3 (eg ≈ 1) and La0.7Pb0.3MnO3 (eg ≈ 0.7) showed the highest activity among the perovskites studied (56, 57).

Such observations lend support to subsequent catalyst research centered on Mn- and Co-based perovskites. Metal substitution on the A and B site can alter transition metal oxidation state on the B site, modifying the eg filling and CO oxidation activity. For example, Chan et al. have shown that the CO oxidation activity of LaxSr1–xMnO3 and LaxSr1–xCoO3 increases as nominal eg filling increases from 0.2 to 0.8 from La0.2Sr0.8MnO3 to La0.8Sr0.2MnO3 (39). Complementary mechanistic studies can help elucidate possible rate-limiting steps that underlie eg as an activity descriptor, whether they are dissociative oxygen adsorption (58) or surface reaction of adsorbed CO and oxygen (25, 38).

Additionally, because CO oxidation serves as a template reaction for oxidation of complex and toxic hydrocarbons (45), the energetics of surface-carbon and surface-oxygen bonding can similarly dictate the activity of these reactions. A similar volcano-type dependence of propylene (C3H6) and isobutylene (C4H8) oxidation (59) as a function of eg filling was observed, where the maximum activity was approximately centered on eg ≈ 1 (Fig. 3E). Despite the additional complexity of multiple oxidation steps for full conversion of the reactants to CO2, the fundamental requirement that the ideal catalyst binds carbon and oxygen species neither too strongly nor too weakly can still apply. This finding encourages continued use of molecular orbital concepts in the highly relevant realm of carbon oxidation chemistry. Perovskite catalysts continue to gain interest in this area owing to their low material costs and earth abundancy, in addition to flexible chemical tailoring for targeting reactions in complex environments such as diesel exhaust—in spite of oxidation activities that are still an order of magnitude lower than those of noble metal catalysts (25).

NO oxidation on perovskites has emerged as an area for perovskite catalyst development, because NOx emissions in automotive and industrial exhaust necessitate catalysts that are cheaper and more earth-abundant than commercial platinum catalysts (60). NO oxidation on perovskites commonly proceeds similarly to CO oxidation on oxides through the vacancy-mediated Mars–van Krevelen mechanism (53, 54); oxygen and NO first adsorb on the oxide surface, where NO reacts with O to create an O vacancy. The resultant NO2 can then desorb from the surface.

In the context of technological applications, such as lean NOx traps, surface storage and conversion of adsorbed and reacted nitrogen species such as nitrogen dioxide (NO2) and nitrates (NO3) on the surface are critical, because subsequent reduction of these nitrogen species to N2 is a key target (60). Kim et al. found that the NOx conversion rate with perovskites such as La0.9Sr0.1CoO3 and La0.9Sr0.1MnO3 was higher than with commercial state-of-the-art Pt catalysts, demonstrating that perovskite chemistries could offer practical alternatives in NOx abatement technologies (26).

In designing perovskite catalysts in the framework of the described mechanism, the kinetics of NO oxidation can be tuned by changing the adsorption strengths of NO, O, and NO2 on the surface. In Fig. 3F, we show that previously reported activity of NO oxidation for LaxMnO3+δ (61) and La1–xSrxCoO3 (62) increased with decreasing eg filling in the range from 1 to ~0.5 by increasing the Mn4+/Mn3+ and Co4+/Co3+ ratios, respectively. According to current mechanistic understanding, this result can be attributed to higher enhancement of the reactant NO gas adsorption, removing NO gas from the feed gas and forming surface NOx (NO, NO2, and NO3). Continued mechanistic studies will be useful in identifying the energetics of NO2-NO3 conversion, NO2 desorption, and other possible rate-limiting surface conversion processes (63). Nevertheless, such trends highlight the immense opportunities in tuning oxide electronic structure to enhance the activity of NO oxidation and, possibly, related nitrogen-based chemical reactions.

Further, the eg occupancy’s control of surface-oxygen interactions can also capture the energetics of electrochemical reactions. Low-temperature electrocatalysis encompasses both the oxygen evolution reaction (OER) and the oxygen reduction reaction (ORR) in aqueous environments. In the conventional ORR (O2 + 2H2O + 4e → 4OH) mechanism reported for oxides and metals in alkaline solution, O2 displaces OH, followed by formation of OOH, O, and then regeneration of OH (19, 64). This four-electron process limits the efficiency of energy storage and conversion devices such as electrolyzers (1, 2, 5), fuel cells (23), and metal-air batteries (8). However, both ORR and OER reaction kinetics are primarily determined by the surface-oxygen binding; in ORR specifically, too weak of an oxygen binding energy results in the first OH-to-OO displacement step becoming rate-limiting, whereas too strong of an oxygen binding energy results in O-to-OH formation being the rate-limiting step (19). Through the use of thin film electrode preparation methods that better capture intrinsic electrocatalytic activity by minimizing effects such as electrical or mass transport resistances (65), activity trends resulting from these surface chemistry energetics have been discerned. For instance, ORR activity on perovskites in basic solution exhibits a volcano trend as a function of eg-like filling of transition metal ions (BO5); low eg filling in LaCrO3 (eg0) can result in B–O2 bonding being too strong, whereas high eg filling in LaFeO3 (eg2) can lead to too weak of an O2 interaction (Fig. 3G) (19). These effects on ORR activity are well captured over four orders of magnitude, with a peak in perovskite ORR activity at an eg filling of ~1 for LaMnO3, LaCoO3, and LaNiO3, whose activities compare well with those of the state-of-the-art Pt-based catalysts (19, 23).

In a similar vein, for OER (the reverse reaction), the activity of perovskites exhibits a volcano shape as a function of eg filling of transition metal ions, which was used to discover the highly active Ba0.5Sr0.5Co0.8Fe0.2O3–δ with an eg filling of ~1.2 (20). This approach subsequently played a critical role in the discovery of SrNb0.1Co0.7Fe0.2O3 (66) and CaCu3Fe4O12 (67) by tuning eg filling to ~1.2 in the bulk chemistry, while also explaining why LaCoO3 nanoparticles exhibit a higher intrinsic OER activity than the bulk form: The nanoparticles have greater electron spin than in the bulk and an eg occupancy of ~1.2 (68). These examples show that controlling the eg filling of transition-metal ions can guide catalyst design by exploring the wide range of perovskite chemistries.

The compositional flexibility of the perovskite structure also opens interesting material design parameters, such as strain, for (electro)catalysis. Fontcuberta et al. used polarizable x-ray absorption measurements (x-ray linear dichroism) on La0.7Sr0.3MnO3 to show that bulk strain breaks the ideal BO6 octahedral symmetry, resulting in asymmetry in d-orbital electron occupation (69). Tensile strain favors the filling of the in-plane Embedded Image orbitals, whereas compressive strain favors the filling of the out-of-plane Embedded Image orbitals (69, 70). In line with this reasoning, careful investigation of strained LaNiO3 thin films for OER (eg = 1) showed that increased compressive strain enhances OER activity by increasing occupancy of the more symmetrically relevant Embedded Image orbital from a nominal eg filling of 1 toward the ideal filling of ~1.2 (71). These molecular orbital concepts applied to catalysis illustrate the usefulness of electronic descriptors across technologically relevant reactions encompassing carbon, nitrogen, and oxygen chemistries.

Perovskite band structure in (electro)catalysis

Despite the success of molecular orbital (MO) concepts and the establishment of eg filling of transition metal ions as a useful activity descriptor (Fig. 3), in some cases, it is difficult to determine the eg filling of active metal sites responsible for (electro)catalysis where the surface spin state (and thus eg occupancy) is not well known. For instance, extensive debate over the cobalt spin state is ongoing (72). In addition to these practical concerns, the MO treatment does not effectively capture the metal-oxygen bond covalency or the sharing of electrons between the metal and oxygen atoms. This aspect is critical because highly covalent late-transition-metal perovskites can have both metal and oxygen as active sites (73, 74), which is not captured by MO theory because it considers the transition metal as the sole active site. Moreover, highly active OER catalysts such as Ba0.5Sr0.5Co0.8Fe0.2O3–δ and La0.2Sr0.8CoO3–δ become amorphous in bulk and on the surface under OER conditions, resulting in the formation of edge-sharing octahedra from corner-sharing octahedra (75, 76). These structures are similar to electrodeposited metal oxides (77), in which increased electrochemically active surface area and metal oxy(hydroxide) surface structures are responsible for high activity (78).

Bulk electronic structure of perovskites in a crystalline band theory framework has emerged as a useful perspective for explaining (electro)catalytic activity trends and mechanistic details of (in)stability, because the bulk electronic structure dictates surface adsorption energetics (79) and surface and bulk stability (75, 76). This relationship between bulk and surface properties allows the use of electronic structure as a guiding principle for catalyst design. The metal-oxygen bonds in perovskites have mixed ionic-covalent character (80) because of the energetic similarity (covalency) and spatial overlap (hybridization) of metal 3d orbitals and O 2p states, which has been shown to influence catalytic activities (19, 20, 73, 81). Metal 3d bands consist primarily of the σ* and π* (eg and Embedded Image) antibonding states, whereas O 2p states consist mostly of σ, π, and pure nonbonding oxygen states from the coordination chemistry treatment. The DOS of perovskites have been studied extensively by x-ray absorption and emission spectroscopy (82, 83) and DFT studies (8486). The metal electronegativity on the B site, through the choice of transition metal and the design of its oxidation state, can be used to tune the covalency and hybridization of metal 3d and O 2p states (81). In practice, replacing the B site with a more electronegative atom (e.g., replacing manganese with cobalt) or oxidizing the B site from Bn to Bn+1 lowers the metal 3d states into the O 2p states (Fig. 4A, bottom), which increases the metal-oxygen covalency and hybridization—as experimentally demonstrated by soft x-ray absorption of the O K edge, complemented with computed metal 3d and O 2p DOS (81, 87). As the Fermi level moves down toward the O 2p states, the energy penalty to create O vacancy is reduced (85), allowing for electronic and/or oxygen ion conduction in perovskites, a property that has found applications in gas sensors (88) and SOFCs (89).

Fig. 4 Perovskite band structure and oxygen electrocatalytic activity and mechanism.

(A) Correlations between charge-transfer energy (Δ) and OER activity (red) (22), surface exchange activity (blue) (85), and the OER active site identity (73) in the 3d transition metals. aq, aqueous; s, solid; g, gas. (B) Relationship of charge-transfer energy to relevant energetics and rate-determining steps: oxygen vacancy formation (blue) (119121), oxygen binding energy (red) (84), and electron transfer energy (gray) (22). [Adapted with permission from (22)]

In the context of high-temperature oxygen electrocatalysis relevant for SOFCs, experimental and DFT studies showed that increasing covalency, quantified by the O 2p-band center (energy difference between the weighted center of the O 2p band and the Fermi level) or charge-transfer energy (energy difference between unoccupied metal 3d- and occupied O 2p-band centers), correlates with increased measured oxygen surface exchange kinetics (O2 + 4e → 2O2–) at temperatures of ~1000 K (Fig. 4A) (85). This parameter correlates with ORR activity in SOFC cathode materials in which oxygen transport dominates the reaction kinetics; thus, the relation between charge-transfer energy and ORR activity is attributed to a reduced energy penalty for oxygen vacancy formation (Fig. 4B) and a reduced barrier for the rate-limiting steps (85).

For low-temperature OER, increasing covalency correlates with experimental OER activities (Fig. 4A) and stability under OER in alkaline solution (21, 22). For example, increasing covalency from LaCoO3 to Pr0.5Ba0.5CoO3–δ increases the OER activity while the perovskite surface remains stable (21). As covalency is increased even further (e.g., Ba0.5Sr0.5Co0.8Fe0.2O3–δ), perovskites become unstable [converting to amorphous (oxy)hydroxides] under OER (75, 76). As a result, an optimum covalency, as measured by the O 2p-band center, gives rise to both high activity and stability, which is the case for Pr0.5Ba0.5CoO3–δ. These effects originate from the positive correlation between covalency and oxygen vacancy formation: Too high of a driving force for O vacancy formation leads to structural loss of the perovskite phase, resulting in surface amorphization (21, 75, 76). The correlation between the computed O 2p-band center and experimental OER activity of perovskites can be attributed to the relationship between higher covalency and reduced binding of oxygenated species on the surface metal site, and thus increased activity (Fig. 4B) (84).

Metal-oxygen covalency has also been useful for understanding recent mechanistic insights into OER in aqueous alkaline solutions. Attempts at predicting OER activity from the computed binding energy of oxygenated species (OH, O, and OOH) (41) on the metal site in the conventional mechanism (four electron-proton coupled transfer reactions) (64, 90) have shown limited agreement with activity values found experimentally (84). In addition, some highly covalent perovskites exhibit pH-dependent OER activity on the reversible hydrogen electrode (RHE) scale, suggesting that nonconcerted proton-electron transfers participate in catalyzing the OER (73). Increasing covalency in LaxSr1–xCoO3 by introducing the more electronegative Co4+ with increased Sr doping can trigger the involvement of lattice oxygen in OER, where the oxygen is also an active site (73, 74) (Fig. 4A, top).

These observations suggest that a more complex interplay exists between oxide electronic structure and catalytic activity, and the conventional OER mechanism might not be the single governing mechanism across oxide chemistries typically studied (22, 67, 73, 74). Recent work shows that as charge-transfer energy is decreased, the rate-determining step transitions from electron transfer–limited, to proton-electron–coupled, then to proton transfer–limited (Fig. 4B) (22), a trend that holds across a wide breadth of chemistries ranging from insulators to metals. In this treatment, the origin of the electron transfer and surface deprotonation energetic barriers were quantified by aligning the perovskite DOS with the OER thermodynamic energy, showing high electron-transfer energies at higher charge-transfer energies and weaker oxygen-adsorbate binding energies at lower charge-transfer energies (Fig. 4B) (22). A Marcus model was applied to verify the potential energy surfaces for sequential and concerted proton-electron transfer pathways (22). These studies highlight the importance of understanding the physical origin of experimental OER activity trends with electronic descriptors and the need to promote surface deprotonation from oxides to discover new catalysts with enhanced activity. Moreover, continued study of these mechanistic findings can motivate new catalyst design strategies, including support-catalyst interactions (91), surface engineering such as phosphate functionalization promoting proton-transfer limiting steps (92), and multiple active site design (67).

Catalyst discovery: Machine learning and artificial intelligence

High-throughput DFT calculations (9395), machine learning (96, 97), and artificial intelligence (98101) provide opportunities to use the activity descriptors and physical insights discussed above for accelerated catalyst discovery. Key developments for machine learning include identifying key descriptors to rationalize and predict catalytic activity (96) and assisting and expediting the use of computational tools such as DFT to calculate fundamental material properties (Fig. 5) (102). Hong et al. used statistical learning to analyze experimental OER data across a wide range of chemistries, showing that the most important descriptors for OER activity include the metal electron occupancy and metal-oxygen covalency (96).

Fig. 5 Computational and experimental tools for catalysis.

Multiple tools are available to understand and control the (electro)chemical interface for catalyst design and accelerated materials discovery. CTR, crystal truncation rod diffraction; XANES, x-ray absorption near-edge structure; TEM, transmission electron microscopy; FTIR, Fourier transform infrared spectroscopy; XPS, x-ray photoelectron spectroscopy; ML, machine learning; MD, molecular dynamics.

Machine learning can also expedite the process of high-throughput DFT screening. Ab initio computation of a complex (electro)chemical reaction with numerous reaction intermediates, competing processes, and different active sites can result in high computational costs, rendering high-throughput screening incredibly difficult. The development of interatomic potentials (e.g., neural network potentials) (98, 99) built by interpolating ab initio potential energy surfaces on a set of structures can allow for the screening of a large number of candidate catalysts and reaction paths without sacrificing accuracy and predicting power. This approach enabled the atomic structure prediction of Au-Cu nanoalloy in an aqueous environment (100) and transition metal–doped ceria (CeO2) nanoparticles (101), a key step for active site identification. Machine learning can also accelerate the search for transition states, a bottleneck in DFT calculations (103). It can provide more “intelligent” initial guesses for DFT investigations, as demonstrated in the study of the reaction network for syngas conversion on Rh(111), where it allowed the prediction of the most likely reaction mechanism (97). These new tools can have tremendous impact in predicting the performances of and guiding the search for new catalysts.

Defining active sites in situ

The integration of computational tools (DFT and machine learning) and experimental approaches (electrochemistry and spectroscopy) provides additional opportunities for targeted catalyst design (Fig. 5). Through developments in synchrotron-based surface scattering, spectroscopy, and surface-enhanced vibration spectroscopy, the ability to observe chemical and structural changes at the solid/liquid and solid/gas interfaces under catalytic conditions and time scales is increasingly accessible. Surface diffraction techniques such as crystal truncation rod and coherent Bragg rod analysis can experimentally identify oxygenated adsorbates (H2O, OH, and O) under catalytic conditions, not only on single crystal metal surfaces (104, 105) but on oxide surfaces as well (106, 107). This approach also provides atomistic detail about perovskite surface stability, both in its pristine state to quantify strontium surface segregation in La1–xSrxCoO3 (108) and under operating OER conditions to map ruthenium dissolution in SrRuO3 (109).

Numerous synchrotron-based soft x-ray absorption and x-ray photoelectron spectroscopies are useful for monitoring chemical dynamics at the electrified interfaces between metals and liquids (110), oxides and liquids (111) and oxides and oxides (112), along with laboratory-source vibrational spectroscopies such as sum-frequency generation (113), surface-enhanced Raman spectroscopy (114), and surface-enhanced infrared spectroscopy (115) for detection of molecules in low concentrations and reaction intermediates under operating conditions. Moreover, observation of catalytic transient states is increasingly possible. For example, femtosecond time-resolved infrared spectroscopy can probe photogenerated oxyl radicals (Ti-O) on the SrTiO3 surface by monitoring subsurface vibration changes (116), and ultrafast infrared spectroscopy can detect photoinduced molecular changes such as proton-transfer dynamics of OH ions (117). In addition, using state-of-the-art ultrafast pump-probe spectroscopy, Öström et al. observed the O-CO transition state between the CO and O surface intermediates in the CO oxidation reaction on the Ru(0001) surface (118). These techniques provide exciting opportunities to understand the details of the solid/gas, solid/liquid, and electrified solid/liquid interfaces under operating catalytic conditions, elucidating the nature of adsorbed species, active sites, and reaction mechanisms.

Future directions

The bridging of perovskite electronic structure with catalysis has allowed greater fundamental understanding of catalytic processes and improved catalyst design for technologically relevant reactions. We highlight how understanding perovskite electronic structure can rationalize surface energetics, catalytic activity trends, and even mechanistic insights for reactions involving carbon, nitrogen, and oxygen chemistries. Expanding the scope of reactions for perovskite catalysts, such as the kinetically challenging electrochemical CO2 reduction or ammonia synthesis reactions, remains an important goal for addressing environmental challenges. Efforts toward this goal can be enabled by current and future advancements in computational and experimental studies of perovskite material properties. Catalyst design will also benefit from taking advantage of perovskite chemical flexibility to expand its compositional space to target key catalytically relevant material properties. Finally, surface engineering and the role of strain and heterointerfaces present exciting opportunities in the search for highly active, next-generation catalysts.

References and Notes

  1. Acknowledgments: This work was supported in part by the Skoltech-MIT Center for Electrochemical Energy and Eni.

Stay Connected to Science

Navigate This Article