Electrocatalysis for the Hydrogen Economy

Abstract

This chapter deals with the concept of “hydrogen economy”, which was introduced by John O.M’ Bockris in 1972. We summarize the fundamental principles and the progress for the reactions relevant to the hydrogen economy, namely the hydrogen and oxygen evolution for water electrolyzers, and the hydrogen oxidation and oxygen reduction for fuel cells. The activity of each reaction can be correlated to a single descriptor, i.e. the adsorption energy of a key reaction intermediate, following a volcano-type relationship. Highly active materials can be prepared with the aid of modern computational and experimental tools. Nevertheless, to develop catalysts that are substantially more active and reach the performance of ideal catalysts, the focus must be placed on materials that can break the energetic scaling relations between intermediates. The systems of choice are acidic water electrolyzers or fuel cells, using noble metals for the catalytic material, despite the great progress made in the field of alkaline systems. However, to realize the concept of hydrogen economy on a large scale, the electrode material for either reaction must combine activity, stability and abundance.

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

In Jules Verne’s 1874 novel “The mysterious island”, the engineer Cyrus Smith explains to his fellow prisoners his belief that “water will one day be employed as fuel, that hydrogen and oxygen which constitute it, used singly or together, will furnish an inexhaustible source of heat and light, of an intensity of which coal is not capable” [1]. One hundred years later, in a letter to Science, John O’M. Bockris gave a scientific outlook to these words by introducing the term “hydrogen economy”: Bockris’ vision for a paradigm shift in meeting the increasing global energy needs with hydrogen acting as the primary energy carrier [2]. Bockris used the term “economy” to emphasize the energetic, economic, ecological and societal aspects of his idea. In brief, this concept was originally based on converting the plentiful and inexpensive electrical energy delivered by nuclear stations to chemical energy, by splitting water and thereby producing hydrogen in onsite water electrolyzers. After transportation to distribution stations and to final locations (houses, factories, vehicles, trains, aircrafts, etc.), the produced hydrogen would be used in an onsite fuel cell to deliver electrical energy. The advantages of Bockris’ concept were obvious: the rapidly increasing energy demand would be met at a lower cost without polluting the environment, while the dependence on fossil fuels would become minimal.

When Verne was writing his book, he was probably aware of two independent observations which set the scientific stage for the “hydrogen economy”: In 1789, the Dutch Adriaan Paets van Troostwijk and the German-Dutch Johan Rudolph Deiman observed that gas bubbles form on two gold wires immersed in water, when they are connected to an electrostatic generator, and they realized that they had “split” water into hydrogen and oxygen [3]. Fifty years later, in 1839, the British Sir William Grove described in a letter to the Philosophical Magazine how he was able to produce electricity by connecting two platinum wires, immersed in an acidic solution through two glass tubes, one filled with hydrogen and one with oxygen [4]. The observation that oxygen and hydrogen can recombine by producing electricity was actually mentioned a few months before Grove’s publication, by the German Christian Friedrich Schönbein [5]. In principle, Van Troostwijk and Deiman were the first ones to split water into its components, while Grove and Schönbein had prepared the first, very primitive fuel cell.

The original concept of Bockris in 1972 included the utilization of nuclear energy for supplying the necessary energy to produce H₂ via water splitting and to set the hydrogen economy in motion. In the years that followed, however, a modification of this idea became necessary: the Chernobyl accident in 1986 raised severe concerns for the use of nuclear power, and concomitantly, there was a boost in the interest for the exploitation of renewable energy sources. Therefore, the modern interpretation of the “hydrogen economy” concept involves the utilization of renewable instead of nuclear energy, as the first step to split water and generate hydrogen. It is evident, however, that the central idea of John Bockris to use hydrogen as the primary means of storing and transporting energy remained unaltered. In fact, the concept of “hydrogen economy” is nowadays at the focus of academic, technological and industrial interest and plays a key role in the research conducted in the fields of electrochemistry, (electro-)chemical engineering, materials science, computational chemistry and others.

The focus in this chapter is the core of the ‘hydrogen economy’ principle, namely the electrochemical reactions of hydrogen and oxygen evolution for the water electrolyzers, and the hydrogen oxidation and oxygen reduction reaction for the low-temperature hydrogen–air fuel cells. We will (i) demonstrate the fundamental principles that govern the above mentioned reactions, (ii) summarize the important progress that was made in the field over the last decades, and (iii) discuss the future prospects towards the realization of Bockris’ and Verne’s vision.

2 Water Electrolyzers and Fuel Cells—The Basic Principle

In water electrolysis cells (water electrolyzers) the oxygen evolution reaction (OER) occurs at the anode (positive electrode) and the hydrogen evolution reaction (HER) occurs at the cathode (negative electrode) (see Scheme 1, left). Inversely in a hydrogen–air fuel cell, the hydrogen oxidation reaction (HOR) takes place at the anode (negative electrode) and the oxygen reduction reaction (ORR) occurs at the cathode (positive electrode) (see Scheme 1, right). The overall reaction occurring in a water electrolyzer (forward reaction, left-to-right) or in a hydrogen–oxygen fuel cell (back reaction, right-to-left) is described by (1):

$$2{\text{H}}_{2} {\text{O}} \, \rightleftharpoons \, {\text{O}}_{2} + 2{\text{H}}_{2} .$$

(1)

Scheme 1

Principle of operation of water electrolyzers (left) and hydrogen–oxygen fuel cells (right) under acidic conditions

At standard conditions, the Gibbs free energy of the forward non-spontaneous water splitting reaction is +474.4 kJ per 1 mol of O₂ evolved. Thus, water splitting can take place only with the supply of electrical energy, i.e. by applying a potential of at least 1.229 V between the two electrodes. The Gibbs free energy of the reverse spontaneous reaction is equal but of opposite sign.

2.1 Fundamentals of the ORR and the OER

The oxygen reduction and the reverse water oxidation are multi-step processes and involve the transfer of up to four electrons [6]:

$${\text{O}}_{2} {\text{ + 4H}}^{ + }\, \text{ + }\, 4 {\text{e}}^{ - } \, \rightleftarrows \, 2 {\text{H}}_{2} {\text{O}}\quad {\text{E}}_{{{\text{O}}_{2} /{\text{H}}_{2} {\text{O}}}} = \, + 1.229\,{\text{V}}\,{\text{vs}}\,{\text{RHE}}$$

(2)

where E is the standard equilibrium potential of the reaction with respect to the reversible hydrogen reference electrode (RHE). Such multi-electron processes proceed sequentially through the formation of several intermediates. Depending on the followed mechanism, reaction intermediates can be *OOH, *O, *OH, *HOOH or dissolved H₂O₂, where the asterisk denotes adsorbed species. The distinction between the mechanisms is based on the number of proton/electron transfer steps that precede the O–O bond cleavage (for the ORR) or formation (for the OER) [7,8,9]. The only possible intermediate of the overall reaction (2) that can desorb is hydrogen peroxide [10]:

$${\text{O}}_{2} + 2 {\text{H}}^{ + }\, + 2 {\text{e}}^{ - } \, \rightleftarrows \, {\text{H}}_{2} {\text{O}}_{2} \quad {\text{E}}_{{{\text{O}}_{2} /{\text{H}}_{2} {\text{O}}_{2} }} = \, + 0.695\,{\text{V}}\,{\text{vs}}\,{\text{RHE}}$$

(3)

$${\text{H}}_{2} {\text{O}}_{2} + 2 {\text{H}}^{ + }\, + 2 {\text{e}}^{ - } \, \rightleftarrows \, 2{\text{H}}_{2} {\text{O}}\quad {\text{E}}_{{{\text{H}}_{2} {\text{O}}_{2} /{\text{H}}_{2} {\text{O}}}} = \, + 1.709\,{\text{V}}\,{\text{vs}}\,{\text{RHE}} .$$

(4)

The participation of H₂O₂ in the reaction sequence remains debateable. The standard equilibrium potentials shown in (3) and (4) are calculated for a solution with the standard activity of H₂O₂ equal to one. Thus, in thermodynamic terms it cannot be excluded that a small amount of H₂O₂ forms as an intermediate from the ORR at potentials more positive than +0.695 V_RHE or from the OER at potentials less positive than +1.709 V_RHE [10]. Indeed, the comparison between the O₂ and H₂O₂ reduction shows that the two reactions follow the same trends in different electrolytes or surfaces (low- and high-index Pt facets), indicating that the ORR is likely to proceed through an H₂O₂-mediated pathway [11,12,13]. Similar studies on the oxidation of H₂O₂ under conditions relevant to the OER are not known.

The ideal catalyst would carry out both reactions at the reversible potential, i.e. at +1.23 V_RHE. This would be achieved only if there was no uphill step during either the ORR or the OER at this potential; namely, if the difference in the free energy of the species involved in each step was zero [14]. Real catalysts, however, deviate from the ideal scenario. Density functional theory (DFT) calculations allow the calculation of the free energies of the species [14] and show that there are three endothermic steps for the ORR, i.e. OOH* formation, *O hydrogenation and *OH desorption (see the black diagram from left-to-right in Fig. 1), while the OER is associated with the strongly endothermic *O + *OH recombination (see black diagram from right-to-left). These calculations are made for Pt(111) at the reversible potential (+1.23 V_RHE). To render all steps exothermic for each individual reaction an overpotential is required, i.e. a potential as low as +0.78 V_RHE for the ORR and as high as +2.55 V_RHE for the OER (Fig. 1). The comparison between the solid and dashed diagrams shows the deviation of free energies for Pt(111) from the ideal catalyst. The description above includes the assumption that the ORR and the OER proceed via the same intermediates, i.e. *OOH, *O and *OH. It must be mentioned, however, that Fig. 1 shows a unified scheme for the ORR and the OER for an oxygen-free Pt(111) surface, and thus it must be assessed only qualitatively because the OER for instance takes place on an oxide-covered surface.

Fig. 1

Reproduced from [14] with permission from Elsevier

Free energy diagram for the ORR (left-to-right) or the OER (right-to-left) for a bare Pt(111) surface (solid diagram) or an ideal catalyst (dashed diagram), at the reversible potential (1.23 V), and at a potential for which all steps become exothermic (+0.78 V for the ORR or +2.55 V for the OER). All free energies are expressed with respect to the free energy of liquid water.

The deviation of real catalysts from the ideal performance is imposed by the scaling relations between the free energies of adsorption of the ORR or OER intermediates that hold for metal surfaces [15, 16]. The free energies of adsorption of *OOH, *O and *OH on (111) surfaces at a given potential, E, are related to each other through the following equations [14]:

$$\left( {\Delta {\text{G}}_{{\text{ * }{\text{OOH}}}}^{{\text{E}}} + 3\times {\text{E}}} \right){ = 0} . 5 3\,{ \times }\,\left( {\Delta {\text{G}}_{{\text{ * }O}}^{{\text{E}}} \text{ + }\, 2{ \times }{\text{E}}} \right) + 3. 3 3 \,{\text{eV}}$$

(5)

$$\left( {\Delta {\text{G}}_{{\text{ * }{\text{OH}}}}^{{\text{E}}} { + }\, 1\times {\text{E}}} \right) = 0. 5 0\times \left( {\Delta {\text{G}}_{{\text{ * }{\text{O}}}}^{{\text{E}}} + 2\times {\text{E}}} \right) + 0. 0 4\,{\text{eV}} .$$

(6)

The factors multiplied by the potential, E, represent the number of electrons required to form the respective species from water, i.e. 1 for *OH, 2 for *O and 3 for *OOH. Scaling relations also apply for metal oxides with the expressions being slightly different than for (111) metal surfaces [17]:

$$\left( {\Delta {\text{G}}_{{\text{ * }{\text{OOH}}}}^{{\text{E}}} + 3\times {\text{E}}} \right) = 0. 6 4 \, \left( {\Delta {\text{G}}_{{\text{ * }{\text{O}}}}^{{\text{E}}} + 2\times {\text{E}}} \right){ + }\, 2. 4 0 \,{\text{eV}}$$

(7)

$$\left( {\Delta {\text{G}}_{{\text{ * }{\text{OH}}}}^{{\text{E}}} + 1\times {\text{E}}} \right) = 0. 6 1 \, \left( {\Delta {\text{G}}_{{\text{*}{\text{O}}}}^{{\text{E}}} + 2\times {\text{E}}} \right) - \, 0. 5 8\,{\text{eV}} .$$

(8)

Therefore, independent of the nature of the metal or metal oxide, the difference $\Delta {\text{G}}_{{\text{ * }{\text{OOH}}}} - \Delta {\text{G}}_{{\text{ * }{\text{OH}}}}$ will be ca. (3.29 – 2×E) eV for (111)-metals or (2.98 – 2×E) eV for metal oxides, which is 0.83 or 0.52 eV higher than what is required for the ideal catalyst for either the ORR or the OER, respectively. These differences are fixed for all metal or metal oxide surfaces, respectively, as long as (i) the reaction intermediates involved are the same and (ii) adsorbates bind in the same way, i.e. through the same atom. The thermodynamic limitations introduced by scaling relations can in principle be avoided using catalysts that bind the intermediates on different sites, as this would allow tuning the adsorption energy of one adsorbate independently of the other [15].

The scaling relations allow the derivation of standard equilibrium potentials as a function of a single free energy of adsorption, e.g. the ΔG_O:

• For ORR on (111) surfaces, assuming an O₂ → *OOH → *Ο + *OH → 2*OH → Η₂Ο pathway:

  $${\text{E}}_{{{\text{O}}_{2} /*{\text{OOH}}}} \, = \,\frac{1}{{{\text{e}}_{{\text{o}}} }} \times \left( {\Delta {\text{G}}_{{\text{O}}} + 1.59} \right)$$

  (9)
  $${\text{E}}_{{*{\text{OOH}}/*{\text{OH}}}} \, = \,\frac{1}{{{\text{e}}_{{\text{o}}} }} \times \left( {0.03\Delta {\text{G}}_{{\text{O}}} + 3.29} \right)$$

  (10)
  $${\text{E}}_{{*{\text{OH}}/{\text{H}}_{2} {\text{O}}}} = \frac{1}{{{\text{e}}_{{\text{o}}} }} \times \left( {0.50\Delta {\text{G}}_{{\text{O}}} + 0.04} \right).$$

  (11)

• For the OER on metal oxides, assuming an ${\text{H}}_{2} {\text{O}}\, \to \,*{\text{OH}}\, \to \,*{\text{O}}\, \to \,*{\text{OOH}}\, \to \,{\text{O}}_{2}$ pathway:

  $${\text{E}}_{{*{\text{OH}}/{\text{H}}_{2} {\text{O}}}} \, = \,\frac{1}{{{\text{e}}_{{\text{o}}} }} \times \left( {0.61\Delta {\text{G}}_{{\text{O}}} \, - \,{\mkern 1mu} 0.58} \right)$$

  (12)
  $${\text{E}}_{{*{\text{O}}/*{\text{OH}}}} \, = \,\frac{1}{{{\text{e}}_{{\text{o}}} }} \times \left( {0.39\Delta {\text{G}}_{{\text{O}}} \, + \,0.58} \right)$$

  (13)
  $${\text{E}}_{{*{\text{OOH}}/*{\text{O}}}} \, = \,\frac{1}{{{\text{e}}_{{\text{o}}} }} \times \left( { - 0.36\Delta {\text{G}}_{{\text{O}}} + 2.40} \right)$$

  (14)
  $${\text{E}}_{{{\text{O}}_{2} /*{\text{OOH}}}} \, = \,\frac{1}{{{\text{e}}_{{\text{o}}} }} \times \left( { - 0.64 + 2.52} \right).$$

  (15)

Thus, the ΔG_*O can be used as a single descriptor for the construction of thermodynamic volcano-type plots. For example, Fig. 2 shows the standard equilibrium potentials for the various steps of the ORR or the OER [15]. Note that contrary to Fig. 1 which was drawn for Pt(111) for both reactions, here the volcanoes are constructed for metal (111) and for metal oxide surfaces for the ORR and the OER, respectively. The thick curves in Fig. 2 represent the onset potential for each reaction, and the volcano develops by a change in the potential-determining step¹ as the metal–oxygen bond becomes weaker from left to right. For instance, on the left-hand side of the volcano (strong *O binding) the ORR is limited by *OH desorption while on the right-hand side (weak *O binding) the potential-determining step is the formation of *OOH. The optimum catalyst, i.e. at the top of the volcano, achieves a compromise between strong and weak *O binding. For the OER, there are four potential-determining steps depending on the free energy of adsorption for *O, but the most relevant ones near the top of the volcano are the O–O bond formation for strong *O binding (left-hand side) and the *OH dehydrogenation for weak *O binding (right-hand side). For both reactions, the top of the volcano is accompanied with a deviation from the reversible potential, i.e. with an overpotential, which cannot be avoided when a metal or a metal oxide surface is used for the ORR or the OER, respectively, as long as the above scaling relations are satisfied.

Fig. 2

Reproduced from [15] with permission from Elsevier

Thermodynamic volcano plots for the ORR on metal (111) surfaces and for the OER for metal oxides, as derived from the solution of Eqs. (9)–(15). The thick curves represent the potential at which the free energy of all steps of the respective reaction is negative. The potential-determining step is denoted in the figure legends. The ORR was assumed to follow the O₂ → *OOH → *Ο + *OH → 2*OH → Η₂Ο pathway, and the OER the H₂O → *OH → *Ο → *OOH → O₂ pathway. The dashed curve represents the standard potential for the O₂/H₂O couple.

The volcano plot in Fig. 2, derived purely based on thermodynamics, is in principle different from an (experimental) volcano plot that has an activity term on the vertical axis and is thus based on kinetics. However, the E_PDS is a reliable measure of the overall reaction rate, as long as a relation between the activation energy and the reaction energy (Brønsted–Evans–Polanyi principle [19]) applies. Therefore, thermodynamic volcanoes are typically in good, qualitative agreement with kinetic volcanoes.

3 Oxygen Evolution Reaction

The oxygen evolution takes place on oxide surfaces. The oxide may undergo redox transitions depending on the potential, which makes it more complicated to rationalize structure–activity relationships. Trasatti summarized the early efforts to develop activity descriptors for the OER and showed that the experimental OER activity for some rutile- or spinel-type oxides follows a volcano-type relationship, with the descriptor being the standard enthalpy of the lower-to-higher transition [20].

Rutile-type oxides of precious metals have attracted most of the interest in acidic electrolytes, because they are supposed to be more stable toward dissolution compared to oxides of non-noble metals. The activity of precious metal oxides in acid increases in the order Au < Pt < Pd < Rh < Ir < Ru [21]. The same activity trend is also observed for nanoparticulate catalysts [22]. The activities of RuO₂ and IrO₂ are lower in alkaline compared to acid; however, RuO₂ remains the most active material with the IrO₂ following [23]. These activity trends are consistent with the *O binding energy on the metal oxides; for example RuO₂ and IrO₂ bind O with nearly the optimal adsorption energy [24]. One important feature of the OER on noble metal oxide surfaces is that the mechanism of the reaction seems to be different among the oxides. Differential electrochemical mass spectrometry (DEMS) studies using isotope-labelled oxide (¹⁸O) at the surface and non-labelled ${\rm H}^{16}_{\rm 2}$O for the electrolyte showed that the metal oxide can participate in the O₂ formation on some surfaces (e.g. gold or ruthenium oxide) [25, 26], while on others O₂ forms only from water adsorbed at the surface (e.g. platinum oxide) [27].

All noble metal oxides are unstable towards dissolution under the highly oxidizing potentials of the OER. The combination of electrochemistry with online elemental analysis of the electrolyte has shown that the dissolution of precious metals and their oxides must be distinguished between steady and transient dissolution, i.e. during potentiostatic or potentiodynamic conditions respectively [21]. Transient dissolution occurs mostly during the reduction of the metal oxide, and to a lesser extent during the reverse oxidative transition, but always during changes in the metal or metal oxide state. On the contrary, steady dissolution occurs with a constant rate at a fixed potential or current. IrO₂ and RuO₂ exhibit transient dissolution, while the steady dissolution (relevant for the OER) is very pronounced for RuO₂ only (see Fig. 3), due to the oxidation of the rutile-type RuO₂ phase to RuO₄ [28, 29]. The same trend is maintained also in alkaline solution, where the absolute dissolution rates are actually higher compared to acidic for either IrO₂ or RuO₂ [30, 31]. Therefore, despite its slightly higher overpotential for the OER, IrO₂ is the state-of-the-art catalyst for the OER because of its better stability compared to RuO₂. One other approach to combine the higher activity of RuO₂ and the better stability of IrO₂ has been to use mixed ruthenium–iridium oxides [32, 33].

Fig. 3

Reproduced from [21] with permission from Wiley

Online monitoring of the dissolution rate of polycrystalline Ru, Ir and Pt during (left): potential ramps from 0.05 to 1.5 V_RHE with 2 mV s⁻¹ and (right): galvanostatic polarization from 0.05 to 1.6 mA cm⁻². Electrolyte: 0.1 M H₂SO₄.

By decreasing the particle size of nanoparticulate RuO₂ or IrO₂ catalysts, not only the surface-to-volume ratio but also the area-normalized OER activity increases [34,35,36]. This trend has been attributed to the higher proportion of less active ordered planes at larger particles [34,35,36]. Therefore, in terms of mass-normalized activity the goal for RuO₂ or IrO₂ catalysts is to decrease the particle size. The stability, however, of the state-of-the-art IrO₂ catalysts decreases with decreasing the particle size [36].

Because of the amount of precious iridium needed for IrO₂ catalysts, approaches to reduce the iridium content by mixing it with another non-noble material have been explored. For instance, Nd₃IrO₇ or Pb₂(Pb_xIr_2-x)O_7-y were found to be about as active as IrO₂ in strongly alkaline solutions [37]. Double perovskites based on iridium and a second metal oxide, such as Ba₂NdIrO₆, can combine up to three times improved OER activity in acid and three times lower noble metal content compared to state-of-the-art IrO₂, while the stability of these materials is comparable to IrO₂ [38].

However, the main approach to replace expensive and scarce noble metals has been to develop catalysts based entirely on non-noble transition metals. In acidic solutions, IrO₂ clearly outperforms these materials which are additionally not stable (see Fig. 4). Therefore, the electrolyte for non-noble OER catalysts typically needs to be neutral or alkaline, since the dissolution of the electrode is minimal and the OER activity similar to that of IrO₂ in this medium (see Fig. 4). Neutral electrolytes, however, are expected to face issues of lower conductivity and local pH changes even for strongly buffered solutions. Different classes of such materials have been investigated; for example, rutile-type Co-, Ni- and Mn-based oxides, the activity of which increases in 0.1 M KOH in the order MnO_x < NiO_x < CoO_x and they are somewhat less active compared to IrO_x [39, 40]. This activity trend is explained in terms of the oxygen binding energy of the materials [24]. Interestingly, the activity of mixed oxides such as NiCoO_x, NiFeO_x, CoFeO_x, etc. is higher than the activity of the individual metal oxides, almost approaching that of IrO₂ in the same solution [40]. Other interesting classes of materials are spinel- or perovskite-type electrodes. The first systematic investigations on spinels such as NiCo₂O₄, NiLa₂O₄, Co₃O₄, etc. were made by Tseung and by Trasatti in alkaline solutions [41,42,43]. Early studies on ABO₃ perovskites (A: a lanthanide, B: a first-row transition metal) were carried out by Bockris, who tried to identify the parameters that control the electrocatalytic activity. Bockris proposed that the rate-determining step is the desorption of OH and hypothesized that the OER activity of perovskites follows a volcano-type relationship where the activity descriptor is the adsorption energy of OH on the transition metal used in the perovskite [44, 45]. The analysis of Bockris already allowed the early prediction of some active perovskites, such as LaNiO₃ or LaCoO₃ [45]. Using molecular orbital principles, Shao-Horn and co-workers [46] first derived a volcano-type relationship for a series of perovskite-type oxides of first-row transition metals and then predicted that Ba_0.5Sr_0.5Co_0.8Fe_0.2O_3-δ (BSCF) will be located at the top of this volcano. The very high activity of this material was verified experimentally, and it is by 10 times higher than the state-of-the-art IrO₂ in alkaline solution. The same group later showed that the BSCF particles in fact undergo quick amorphization during the oxygen evolution, which leads to an activity increase [47].

Fig. 4

Reprinted from Ref. [40] with permission from the American Chemical Society

Plots of catalytic activity, stability, and electrochemically active surface area for OER electrocatalysts in acidic (top) and alkaline (bottom) solutions. The x-axis and y-axis show the overpotential required to achieve 10 mA cm⁻² per geometric area at time t = 0 and t = 2 h. The diagonal dashed line is the expected response for a stable catalyst. The unshaded white region highlights the region where the overpotential to achieve 10 mA cm⁻² at time t = 0 and 2 h is < 0.5 V. Note the break and change in scale in both axes at overpotentials > 0.5 V.

Another approach, finally, is the so-called “in situ formed” catalysts. Starting from a neutral, phosphate-buffered Co²⁺-containing solution and an indium tin oxide (ITO) substrate, Kanan and Nocera showed that oxygen is evolved at positive bias, probably through the oxidation of Co²⁺ to Co³⁺, precipitation of Co³⁺–HPO₄ ^2– on the ITO, and consequent oxidation of Co³⁺ to Co⁴⁺. The in situ grown film on the ITO is active for the OER and the formed Co⁴⁺ is reduced to Co²⁺, so a new catalysis round can start [48].

4 Oxygen Reduction Reaction

Of all the monometallic catalysts, platinum exhibits the highest activity for the ORR, being located closest to the top of the volcano [49]. The trends in the ORR activity for model low- and high-index single-crystal surfaces depend on the electrolyte, which highlights the importance of the structure-sensitive adsorption of electrolyte ions on platinum. In particular, the ORR activity increases in the order (100) < (111) ≈ (110) in HClO₄, (111) < (100) < (110) in H₂SO₄ and (100) < (110) < (111) in KOH [50,51,52,53]. The activity of (111) increases by the introduction of (111) or (100) steps in HClO₄ and H₂SO₄, but has the opposite effect in KOH [13, 54]. On the other hand, the activity of (100) is not influenced by the introduction of (111) or (110) steps in HClO₄ and H₂SO₄, but increases in KOH [13, 54]. The sensitivity of the ORR activity on the surface atom arrangement indicates that the platinum activity can be tuned by finding the appropriate surface geometry. Indeed, high-index facets are more active than the low-index crystals in HClO₄, with Pt(331) and Pt(221) being the most active of all [51].

To enhance the utilization of platinum atoms in a real system, the ratio of surface to bulk atoms (surface-to-volume ratio) needs to increase using for instance finely dispersed platinum nanoparticles on a high-surface-area support. However, the decrease of the particle size does not merely increase the surface-to-volume ratio: the properties of surface atoms at nanoscale are different than in extended surfaces due to structural and electronic effects. The origin of such effects lies in the size- and shape-dependent distribution of surface atoms to various short- and long-range terraces and to different steps and defect sites [55,56,57]. To develop design rules for ORR catalysts from pure platinum, the impact of surface atom arrangement must be rationalized, e.g. by the introduction of activity descriptors that include structure-sensitive parameters such as the generalized coordination numbers [58, 59].

If the particles are represented as truncated octahedra, then the ratio of surface atoms on (111) and (100) terraces decreases for smaller particles [60, 61]. This is predicted to decrease the area-normalized activity toward the ORR, especially in the size range 2–10 nm (particle size effect) [60, 61]. Such a prediction has been confirmed experimentally only when uniformly defined ultrathin catalyst layers are prepared; such layers prevent any O₂ diffusion limitations within the catalyst layer [62, 63]. The mass-normalized activity is a trade-off between the better Pt utilization and the lower reaction rate by decreasing the particle size. Thus, the mass-normalized activity is maximized at ca. 3 nm [63, 64], which is the typical size for state-of-the-art carbon-supported Pt catalysts for the ORR.

The orientation of surface atoms can be also altered by the nanostructure shape. For instance, nanoparticles with dominant (100) terraces (e.g. cubic particles) are more active in H₂SO₄ but those with dominant (111) terraces (e.g. cubooctahedral or tetrahedral particles) are more active in HClO₄ and KOH. This is consistent with the fact that Pt(100) is more active than Pt(111) only in H₂SO₄ [65, 66]. Nanoparticles that exhibit a high proportion of high-index facets are in general much more active than particles of a similar size [67,68,69], which is attributed to the higher activity of high- versus low-index facets. The stability of shape-controlled nanoparticles under the ORR conditions is however an issue, especially for those enclosed by high-index facets, and eventually these particles reshape due to degradation issues that will be discussed below [70, 71].

Even though the ORR takes place at potentials below +1.0 V_RHE, the biggest challenge currently for Pt-based ORR catalysts is their exposure to potentials at which the catalyst or the catalyst support is unstable [72]. The cathode transiently experiences potentials as high as 1.4 V during start-up or shut-down of the fuel cell, as a result of the reverse-current decay mechanism [73]. Since the long-term performance is affected by processes during the entire operation, from start-up to regular load to shut-down, assessing the stability of ORR catalysts for a potential up to 1.5 V represents better the conditions in a fuel cell. Platinum dissolves mostly transiently, i.e. during the reduction of the metal oxide. Apart from metal dissolution, real ORR catalysts can suffer from other degradation mechanisms which are induced by the catalyst exposure to highly positive potentials [74, 75]: (i) particle migration and coalescence, (ii) dissolution of platinum from small particles and redeposition at larger particles (Ostwald ripening) and (iii) corrosion of the support which enhances particle mobility and detachment. The above mechanisms, which may take place in parallel [75], eventually result in a loss of platinum surface area. In addition, the dissolved platinum from the cathode may be reduced by hydrogen that permeates through the membrane, forming a large “Pt band” on the membrane [76].

A particularly interesting class of platinum materials that are aimed to mitigate stability issues are the “nanostructured thin-film” (NSTF) electrocatalysts innovated by 3M. They consist of non-conductive organic crystalline whiskers as the support, on which an ultrathin layer of platinum is deposited [77]. The polycrystalline thin-film catalyst morphology and the absence of a conventional conductive high-surface-area carbon support allow the NSTF catalysts to combine a polycrystalline-like surface-normalized ORR activity, with a low amount of noble material required and an improved stability, since they are free from issues such as support corrosion, Pt particle migration and coalescence, etc. [77, 78]. However, the catalyst morphology introduces also disadvantages, such as the higher susceptibility to flooding at low temperature of operation, which decreases limiting currents and makes water management more complicated.

Returning to the volcano plot, the problem with platinum is that it binds oxygen a bit too strongly, i.e. it is located at the left side of the ORR volcano in Fig. 2 [79, 80]. In a series of patents from the United Technologies Corporation (UTC) in 1980s, it was disclosed that binary or ternary PtCr, PtV or PtCrCo alloys are more active for the ORR than pure platinum [81,82,83,84,85]. Since then, the strategy to weaken the Pt–O bond to move closer to the top of the volcano has been to use such Pt–M catalysts, where M is a late transition metal such as Cu, Ni, Co, Fe, etc. (see Fig. 2) [79, 86,87,88,89,90,91] or alloys with more than one alloying element [92,93,94]. Alloying platinum leads to a modification of the surface electronic properties, even though the atoms of the alloying metal are not present at the surface as they are not stable under the ORR conditions when in contact with the electrolyte. The enhancing effect of the alloying metal atoms to the intrinsic ORR activity has been attributed to bonding interactions between surface Pt and sub-surface M atoms (electronic or ligand effect), and to the compressed arrangement of the surface Pt atoms due to the shorter M–M interatomic distance below the surface (geometric or strain effect) [89, 91]. Even though it is hard to decouple the two effects, the consequence of both is to alter the chemisorption properties of the surface Pt atoms, and the alloying metal atoms have a stronger impact when they are located closer to the surface (see Fig. 5) [90, 91, 95, 96]. Therefore, ideally the alloying M metal must be located in the second layer.

Fig. 5

Reproduced from Ref. [91] with permission from the Nature Publishing Group

Left Polarization curves for the ORR on polycrystalline platinum (grey curve), skeleton-Pt₃Fe (blue dashed curve) and skin-Pt₃Fe (red curve), in 0.1 M HClO₄ at 50 mV s⁻¹. Right Volcano-type plots of the experimentally measured ORR activity for various skin-Pt₃M surfaces at a temperature of 333 K.

The activity enhancement for bimetallic Pt surfaces may depend strongly on the surface atom arrangement: for Pt₃Ni-skin² surfaces in HClO₄, the ORR activity increases by ca. 9 times for the Pt₃Ni(111) versus Pt(111) and by only 2–2.5 times for Pt₃Ni(100) versus Pt(100) or for Pt₃Ni(110) versus Pt(110) [97]. This alters the order in which the ORR activity increases, from Pt₃Ni-skin(100) to Pt₃Ni-skin(110) to Pt₃Ni-skin(111) in HClO₄ (compare with Pt(100) < Pt(111) ≈ Pt(100) in the same solution) [97].

The situation is rather complex for nanoparticulate bimetallic catalysts, when the goal is to obtain the maximum possible activity enhancement. The preparation method and the post-treatment can have a significant impact on parameters such as the size or the shape of the nanostructure, porosity, alloy composition, etc., which in turn influence the ORR performance. As expected from the findings on extended bimetallic surfaces described above, the ORR activity enhancement depends strongly on the shape of the nanostructure. The ORR activity on carbon-supported octahedral Pt₃Ni particles (i.e. with a high ratio of (111) terraces) is by ca. 7 times higher than the commercial Pt/C catalyst, whereas cubic Pt₃Ni particles (i.e. with a high ratio of (100) terraces) show little enhancement [98].

To increase platinum utilization, to decrease the amount of noble metal needed, and to maximize the alloying effect, the approach that gained the largest interest is the synthesis of nanostructures with a pure Pt shell and a Pt–M core, so-called core–shell materials. The Pt shell is supposed additionally to protect the non-noble M metal atoms in the core, where M is typically a late transition metal such as Cu, Co, Ni, etc. However, as it will be described below, this is not really the case. Nanostructures with a highly ordered core are more active and stable than disordered particles [99,100,101,102,103] with the explanation being still under debate. For a detailed description of the methods for the preparation of core–shell structures the interested reader is referred to relevant reviews [104, 105]. In the following, we will briefly mention only the two main preparation methods:

In dealloying, the less noble metal atoms of a bimetallic alloy are rapidly and selectively dissolved. Classical dealloying involves the selective dissolution of the less noble metal, i.e. of the metal with the least positive standard reduction potential [106, 107]. The Strasser group showed that core–shell nanoparticles can be prepared by an “electrochemical dealloying” process, in which an M-rich platinum alloy precursor, e.g. PtCu₃, is subject to cycling within a potential region where M is unstable [89, 108, 109]. The interesting feature of dealloyed core–shell particles is that the resulting structure and its activity/stability can be controlled by parameters such as the initial alloy composition, the dealloying or post-treatment conditions, the size of the particle precursor, etc. [109].

Another method of core–shell preparation is based on the spontaneous galvanic replacement of the non-noble material at the surface of the core by platinum. This can be achieved using PdM (e.g. PdCu, PdCo) cores which result in a PdCu@Pt or PdCo@Pt core–shell structure. Adzic and co-workers pioneered a more elegant method of galvanic replacement, the so-called monolayer method; a copper monolayer, formed by underpotential deposition on the surface of an M metal core, is spontaneously oxidized by platinum ions which are reductively deposited. This results in a core–shell structure with a monolayer of platinum enclosing an M core. The metal core M can be Pd, Au, Ir, or even bimetallic such as PdNi. The method was originally developed on model single-crystalline “cores” and extended to nanostructured electrodes [90, 110,111,112].

Except for core–shell structures, other interesting approaches for the preparation of bimetallic materials include alloys based on 3M’s NSTF technology (e.g. Pt₃Ni₇-NSTF) [113] or the Pt₃Ni-skin nanoframes [114]; the latter enhance the ORR activity by 16 times compared to standard Pt/C catalysts.

The stability of bimetallic electrodes is an important issue, which limits their application in real systems. The dissolution of platinum removes Pt surface atoms from the protective shell, leading to the exposure of the otherwise protected non-noble metal atoms to the electrolyte [115]. This eventually results in the dissolution of the alloying metal atoms from the first atomic layers and the thickening of the protective Pt shell, so the enhancing effect from alloying decays with time. Even worse is the fact that the dissolved metal ions deteriorate the fuel cell performance further. For example, copper ions can be reduced and deposited at the anode of the fuel cell [116], while nickel or cobalt ions can be deposited on the membrane [113, 117].

DFT calculations predicted that alloys of platinum with rare earths (e.g. Pt₃Y and Pt₃Sc) will be not only more active, but also more stable than pure platinum or other Pt alloys [80]. The argumentation is based on the higher (more negative) enthalpy for alloy formation of such alloys compared to those with late transition metals, which is likely to render the diffusion of the alloying metal to the surface kinetically more difficult [80, 118]. The correlation between the enthalpy for alloy formation and the kinetic barrier for diffusion of the alloying element has been demonstrated in a later publication [119]. The higher activity of model Pt₅M alloys (M: a rare earth or alkaline earth element such as Sc, Y, La, etc.) for the ORR was evidenced experimentally by the Chorkendorff group and was attributed to strain [118, 120,121,122]. The same group also showed the enhanced activity with nanostructured electrodes [123, 124]. Long-term accelerated degradation tests on such catalysts indeed showed that the selective dissolution of the second metal is slower; however, there is still an activity loss with time and a concomitant thickening of the platinum shell, for either extended or nanostructured surfaces [120, 125].

Due to the high cost and the limited crustal abundance of platinum, extensive efforts to find alternative catalysts have been made. A good alternative to platinum would exhibit high activity for the ORR, low cost, sufficient abundance and stability under the operation conditions of a fuel cell.

Among other monometallic surfaces, mostly palladium has been considered as a potential candidate. Palladium is the metal closest to platinum in the volcano plot and its price is lower. For flame-annealed palladium single-crystal electrodes, the ORR activity increases in the order Pd(110) < Pd(111) < < Pd(100) [126]. The trend may be different for Pd(hkl) crystals annealed by other methods, e.g. inductive heating [127]. Overall, the problems with palladium are that (i) its activity must still increase to compete with platinum, (ii) it is significantly more susceptible to dissolution than Pt [21] and (iii) it is still a rather scarce metal, so a large-scale technology would increase the demand and the price for Pd substantially.

Other approaches include the use of noble metal-free catalysts, such as chalcogenides, oxides, carbides, nitrides, etc. [128]. The activity of such materials, however, remains low compared to platinum. The most promising way to replace platinum with catalysts free of noble metals is a class of materials noted generally as Me–N_x/C_y, where Me is a transition metal, for example, iron (Fig. 6) [129]. The first report on such materials showed that cobalt phthalocyanine can reduce O₂ in acidic solution; [130] later it was shown that also other Me–N₄ chelates can catalyse the reaction in either acid or alkaline [131, 132]. The activity of those catalysts was decreasing with time because of catalyst decomposition by the formed H₂O₂ but thermal treatment improves the stability of the catalyst likely by making it more active for H₂O₂ decomposition [133, 134]. It has been, however, questionable whether the metal centre is still a component of the active site or if the pyrolysis results in a new catalyst, consisting of nitrogen and carbon [135,136,137,138]. Yeager and co-workers showed later that it is not necessary to use Me–N₄ chelates as the precursor, but the catalyst can also be synthesized at high temperature by mixing a Me salt with a carbon–nitrogen source [139], while Dodelet and co-workers showed that the sources of carbon and nitrogen can be also different [140].

Fig. 6

Reproduced from Ref. [129] with permission from the Nature Publishing Group

ORR polarization curves for Fe–N–C catalysts and a Pt/C benchmark catalyst, recorded at room temperature in O₂-saturated pH 1 electrolyte at a rotation rate of 1,600 rpm. The current was corrected for the background current measured in N₂-saturated electrolyte. The potential was corrected for the Ohmic drop. The catalyst loadings were 818 μg_Fe-N-C cm⁻² or 16 μg_Pt cm⁻². The subscript shows the wt% Fe in the catalyst precursor, i.e. prior to any pyrolysis; Fe_0.5-900 and Fe_0.5-950 were obtained after pyrolysis of Fe_0.5 at 900 °C or 950 °C, respectively.

The above progress indicated that (i) an important parameter for this class of materials is the density of Me–N₄ sites per unit volume, independent of whether they serve as the active sites or as the precursors for the generation of active sites after pyrolysis and (ii) the design of catalyst can be tailored by the careful choice of the transition metal, the nitrogen, and the carbon source as these can be independently added, as well as by following appropriate treatment procedures.

Following this notion, the Dodelet group synthesized a very active Fe–N/C catalyst by increasing substantially the density of active sites [141]. They used initially a mixture of a highly (micro)porous carbon support, a pore filler and ferrous acetate, which was subject to ball milling to force the pore filler and the iron precursor into the pores, and then was pyrolyzed first in Ar and then in NH₃ to generate the Fe–N₄ moieties on the carbon support. The high density of such moieties makes this catalyst active for the ORR, so the activity reaches the ca. 90% of the activity of Pt/C [141]. The Zelenay group used heteroatom polymer precursors such as polypyrrole (PPy) or polyaniline (PANI) as the template for nitrogen and carbon, aiming to a uniform distribution and high density of Me–N₄ sites. They initially showed that Co–PPy/C catalyst exhibits respectable stability and activity without any pyrolysis step [142]. Later, they synthesized pyrolyzed Fe-, Co- and FeCo-PANI/C catalysts with activity that almost matches the one of Pt/C and promising stability which was proposed to be due to a graphitized carbon phase [143].

The ORR activity and selectivity on Me–N₄ chelates (where Me: Fe, Mn, Co) is related to the binding energy of O₂, which in turn follows the redox potential for the Me(III)/Me(II) transition [144]. Catalysts with weak O₂ binding (i.e. positive redox potential) reduce O₂ to H₂O₂ and the onset potential shifts more positive in the RHE scale by making the solution more alkaline. On the other hand, catalysts with strong O₂ binding (i.e. negative redox potential) reduce O₂ to H₂O₂ and the onset potential is independent of the pH in the RHE scale.

Overall, the progress made in the last ten years in ORR catalysis with non-noble metals is impressive [128, 145]; however, the issue of stability and long-term performance remains. In fact, non-noble metal catalysts also suffer from degradation when they are exposed to potentials above 0.9 V in acidic solutions. The main origin of degradation is the oxidation of carbon, which leads to the destruction the ORR-active FeN_xC_y sites, while iron leaching from iron particles occurs already at lower potentials (<0.7 V) without though affecting the ORR activity [146].

5 Fundamentals of the HER and the HOR

The hydrogen evolution and the reverse oxidation reaction involve the transfer of two electrons:

$$2{\text{H}}^{ + } \, + \,2{\text{e}}^{ - } \, \rightleftarrows \,{\text{H}}_{2} \quad {\text{E}}\, = \,0\,{\text{V}}\,{\text{vs}}\,{\text{RHE}}.$$

(16)

The generally accepted mechanism for the HER/HOR involves the following three elementary steps [147]:

$${\text{H}}^{ + } \, + \,*\, + \,{\text{e}}^{ - } \, \rightleftarrows \,{\text{H}}_{{{\text{ads}}}}\quad{\text{Volmer}}\,{\text{step}}$$

(17)

$${\text{H}}_{{{\text{ads}}}} \, + \,{\text{H}}^{ + } \, + \,{\text{e}}^{ - } \, \rightleftarrows \,{\text{H}}_{2} \quad{\text{Heyrovsky}}\,{\text{step}}$$

(18)

or

$$2{\text{H}}_{{{\text{ads}}}} \, \rightleftarrows \,{\text{H}}_{2} \, + \,2^{*} \quad{\text{Tafel}}\,{\text{step}}.$$

(19)

Thus, both reactions involve the intermediate formation of H_ads, the binding energy of which can be used as a descriptor for the construction of volcano plots (Fig. 7), following a similar analysis as for the OER/ORR. The ideal catalyst for both reactions adsorbs hydrogen with ΔG_*H = 0, meaning that the energy of *H is the same as that of H₂ and H⁺. In that case, there are no uphill or downhill steps during any of the two electron transfer steps in the corresponding energy diagram, and thus this ideal case is associated with zero overpotential. If the binding is too strong (ΔG_H < 0), then the two reactions are associated with overpotential, because the Heyrovsky/Tafel step limits the HER and the Volmer step limits the HOR. The opposite occurs for too weak *H binding (ΔG_H > 0).

Fig. 7

Reproduced from [15] with permission from Elsevier

Thermodynamic volcano plots for the HER and the HOR, with the energy of hydrogen adsorption as the descriptor, assuming a Volmer-Tafel/Heyrovsky pathway.

6 Hydrogen Evolution Reaction

Trasatti has provided the most traditional expression of a kinetic volcano for the HER [148], using the M–H bond energy as the activity descriptor. Platinum and other Pt-group metals are located near or at the top of the volcano. In the interpretation of such an expression, however, one should take into account that the energies of *H adsorption were calculated for bare metal surfaces, which is not really representative of the conditions during the HER on all metals [149].

In fact, platinum catalyses the HER in acidic solutions with a very high rate, so that the current is entirely limited by H⁺ diffusion [150, 151]. Therefore, conventional RDE measurements cannot be used to estimate the kinetics of the HER on platinum [151]. Moreover, no significant impact of the surface structure was found for the HER on low-index Pt single crystals in acid [152]. The development of platinum-free HER catalysts in acid has been explored to replace platinum in acid electrolyzers, with sulfides, carbides or phosphides being the most promising materials [153].

Contrary to acidic solutions, platinum is not as good a catalyst in alkaline (see Fig. 8) [151]. The reaction is structure sensitive on Pt, which also contrasts with acidic solutions, with the activity increasing in the order Pt(111) < Pt(100) < Pt(110) [154]. The reason for the observed structure sensitivity only in alkaline solutions and for the lower reaction rate compared to acid is still unknown [155]. The lower activity of Pt in alkaline solution implies higher cell voltage in an alkaline electrolyzer, thus other catalysts must be developed to carry out the reaction in alkaline solutions as fast as platinum does in acid. Nickel is a state-of-the-art catalyst for the HER in alkaline solution [156]; however, the catalyst deactivates and the cell voltage increases in the long term [157], which has been attributed to the formation of hydrides [158]. One approach has been to modify platinum with Ni(OH)₂ clusters, which was found to increase the HER activity in alkaline compared to pure Pt by a factor of 8 [155]. The promoting effect of this modification was attributed to the enhancement of water splitting by Ni(OH)₂, while *H recombination still occurs on platinum, in a reaction that proceeds in a bifunctional manner.

Fig. 8

Reprinted from Ref. [151] with permission from the Electrochemical Society

HOR/HER polarization curves on polycrystalline platinum in 0.1 M HClO₄ (a) and 0.1 M KOH (b) at 1600 rpm. Solid black curves represent the voltammograms before iR-correction (E_RDE), dotted-red lines after iR-correction (iR-free) and dashed black curves the Nernstian diffusion overpotential.

7 Hydrogen Oxidation Reaction

Platinum catalyses also the HOR with a very high rate in acid, so that the current is entirely limited by H₂ diffusion [150, 151]. The reaction is so fast that the current is fully preserved even when the surface of platinum is covered by a surface modifier such as calix[4]arene to a coverage of 98% [159]. Similar to the HER, the high reaction rate makes it impossible to correctly measure kinetics for the HOR with RDE measurements. Following the concept of the “floating electrode”³ which allows extremely high mass transport rates of a gaseous reactant, Kucernak and co-workers managed to record HOR curves with limited hydrogen mass transport limitations [160], and the measured current densities were in the order of 0.5 A ${\rm cm}^{-2}_{\rm Pt}$ (Fig. 9).

Fig. 9

Reprinted from Ref. [160] with permission from the Royal Society of Chemistry

HOR polarization curves using a “floating electrode” with a 2.2 μg cm⁻² Pt/C catalyst exposed to H₂, run in 4.0 M HClO₄, at 10 mV s⁻¹. The inset shows the comparison of the floating electrode measurement with the HOR on polycrystalline platinum using an RDE (6800 rpm) in 0.5 M HClO₄.

Similar to the HER, the HOR deviates from the ideal Nernstian behaviour in alkaline solutions on platinum [151]. Indeed, the reaction is structure sensitive only in alkaline media, with the activity increasing in the order Pt(111) ≈ Pt(100) << Pt(110) at low positive overpotential [154]. At high potentials, the inhibition from the adsorption of oxygenated species is stronger for Pt(100) and weaker for Pt(111) [154]. The lower rate of the HOR in alkaline solution implies that a higher amount of Pt (about 100 times) is required in an alkaline fuel cell to achieve the same performance as for an acidic fuel cell. Nickel-based catalysts are the only materials to date, to be successful in achieving HOR activities comparable to platinum in alkaline solutions [161].

8 Summary

In conclusion, the electrocatalysis community has developed a good understanding of the reactions involved in the hydrogen economy; the HER and OER for water electrolyzers, and the HOR and ORR for fuel cells. The activity of each reaction on different surfaces follows a volcano-type relationship where the adsorption energy of a key reaction intermediate can be used as the descriptor, for example, *H for HOR and HER and *O for ORR and OER. The highest activity is obtained for a catalyst that exhibits moderate *H or *O binding, accordingly. However, for reactions involving more than one intermediate (and typically involving more than two electron and proton transfers), such as ORR and OER, energetic scaling relations between intermediates (specifically *O, *OH and *OOH) limit the extent to which a reversible catalyst with a very small overpotential can be achieved. A minimal overpotential of ca. 0.25–0.3 V for both ORR and OER appears very difficult to overcome. Computational and experimental tools offer today the opportunity to prepare highly active materials that approach the performance of the optimal catalyst predicted by current models. However, future models and computational methods should focus on materials that can break the energetic scaling relations between intermediates, to develop catalysts with significantly enhanced properties compared to the current state-of-the-art electrocatalysts.

Despite the massive efforts that have been devoted to the development of materials that operate in alkaline environments and with lower/no noble metal content, the best systems remain to be acidic water electrolyzers or fuel cells, using Pt for the HER, the HOR and the ORR, and IrO₂ for the OER. However, such systems can be a solution only for the small scale: to realize Bockris’ hydrogen economy on a large scale we need to find a combination of materials for the two electrodes that are abundant and but also stable in the long term. This is today the biggest challenge, as it will decrease the cost per device, but will also massively increase the number of devices that can be used worldwide.