Design Principles of Perovskites for Thermochemical Oxygen Separation.

Separation and concentration of O2 from gas mixtures is central to several sustainable energy technologies, such as solar-driven synthesis of liquid hydrocarbon fuels from CO2 , H2 O, and concentrated sunlight. We introduce a rationale for designing metal oxide redox materials for oxygen separation through "thermochemical pumping" of O2 against a pO2 gradient with low-grade process heat. Electronic structure calculations show that the activity of O vacancies in metal oxides pinpoints the ideal oxygen exchange capacity of perovskites. Thermogravimetric analysis and high-temperature X-ray diffraction for SrCoO3-δ , BaCoO3-δ and BaMnO3-δ perovskites and Ag2 O and Cu2 O references confirm the predicted performance of SrCoO3-δ , which surpasses the performance of state-of-the-art Cu2 O at these conditions with an oxygen exchange capacity of 44 mmol O 2 mol SrCoO 3-δ(-1) exchanged at 12.1 μmol O 2 min(-1)  g(-1) at 600-900 K. The redox trends are understood due to lattice expansion and electronic charge transfer.

Introduction

Renewable chemical fuels can be synthesized through solar‐driven electro‐, photo‐, and thermochemical splitting of CO2 and H2O.1 The latter approach utilizes the entire spectrum of concentrated solar radiation as high‐temperature process heat for the production of CO and H2 (syngas) via metal oxide redox cycles.1b, 2 A critical drawback of this approach is the inert gas consumed to lower the partial pressure of oxygen (pO2) for shifting the thermodynamic equilibrium of the reduction step to lower temperatures.3 This, in turn, requires separation of O2 from the product gases for recycling the inert carrier gas and closing the material cycle.3a, 4 The separation of O2 has been a requirement in a variety of commercial applications such as oxy‐combustion, autothermal gasification of carbonaceous feedstock, and O2 removal to avoid catalyst passivation by O2 in fuel cells and when deoxygenating biofuels to make these more akin to petroleum‐derived fuels.5 Industrially, O2 can be separated from air by pressure swing adsorption (PSA) using zeolites and carbon molecular sieves, by ceramic mixed ionic‐electronic conducting (MIEC) membranes,6 and by cryogenic distillation. PSA and MIEC membranes cannot produce high‐purity inert gas,6b and separating O2 from gas mixtures at low pO2 using membranes relies on a stripping gas with even lower pO2. These separation technologies further require an input of electrical work ranging from 100 to 350 kWh per metric ton O2,6, 7 which penalizes the solar‐to‐fuel energy conversion efficiencies. Since solar thermochemical cycles inherently suffer from heat losses, it would be beneficial to utilize an oxygen separation technology driven by waste heat.

Thermochemical solid‐state O2 separation (TSSOS) using metal oxide redox materials such as Cu2O/CuO,3a, 8 Mn3O4/Mn2O3,8 and CoO/Co3O4 8, 9 utilizes low‐grade process heat and does not require electricity. TSSOS has the potential to separate and concentrate O2 at low pO2 via temperature‐swing.3a The current state‐of‐the‐art TSSOS redox material, Cu2O, has a maximum oxygen exchange capacity (Δδ, i.e., the difference in the oxygen non‐stoichiometry between reducing and oxidizing conditions) of about 200 mmol O2 per mol Cu2O, exchanged at approximately 10 μmol min−1 g −1 when cycled between 1120–1450 K.3a We show below that Cu2O cannot be employed with low‐grade process heat at 600–900 K. With the aim of augmenting the O2 exchange capacities and rates of TSSOS redox materials for a more energy‐efficient O2 separation process that utilizes low‐grade solar thermal energy at lower temperatures, such as waste heat from solar fuel production processes, we evaluate perovskites that offer high O2 conductivities and a stable crystal structure over a large range of oxygen non‐stoichiometry.6a,c, 10 The O2 exchange capacity characterizes the trade‐off between high energy conversion efficiencies at low temperature during the endothermic reduction and high rates and extend of the oxygen separation process at high oxide reduction temperatures. For a perovskite with ABO3− stoichiometry—where A and B are metal cations in twelve‐ and six‐coordinated interstices—the TSSOS redox cycle can be represented by Equations (1) and (2):(2)

Conceptually, O2 is stripped from a gas mixture at low pO2 through oxidation of the perovskite at low temperatures. This yields as an output of the oxidation step an inert gas with a lowered pO2 while concentrated O2 is evolved from the solid at an elevated temperature and increased pO2 through partial reduction of the metal oxide. In principle, these reactions are analogous to the electrochemical oxygen reduction and evolution reactions (ORR/OER), where the bonding of O/OH or OH/OOH reaction intermediates to the catalyst surface controls the catalytic activity of the electrode surface.10, 11 The ideal catalytic activity of a surface is determined by an intermediately strong bonding of the key reaction intermediates, which facilitates coverage of the surface with reactants and desorption of products from the surface, as described by the Sabatier principle.12 Analogously, we hypothesize that metal oxide redox materials for removal of O2 from gas mixtures with a lower pO2 than the pO2 during the extraction of O2 from the solid can be characterized with an intermediately strong binding of the lattice oxygen.

Results and Discussion

To test this hypothesis, we screened the redox energetics of binary metal oxides across the periodic table using experiment‐based thermochemical data.13 Figure 1 A shows the free energy (ΔG) of the oxidation and reduction reactions for 32 solid metal oxide and six metal/metal oxide pairs14 versus the thermochemical oxide stability.

Figure 1

A) Free energy of the oxide oxidation at 600 K and oxide reduction at 900 K (ΔG rxn) versus the enthalpy of the oxide reduction at 298 K (ΔH red). All data are at 1 bar.14 Dashed lines are linear regressions. B) The limiting free energy of the redox cycle (ΔG rxn,lim) versus ΔH red. The colored compositions were examined experimentally, with blue and red marked materials limited by the oxide reduction and oxidation, respectively, and purple marking materials that facilitate a redox trade‐off. C) DFT‐models of the oxidized and reduced SrCoO3 and SrCoO2.5 surface, representatively for strontium cobaltite.

The analysis utilizes the enthalpy of the oxide reduction at room temperature as a descriptor10–12 of the correlated reaction energetics, which is equivalent to the amount of energy required to break metal–oxygen bonds, as shown in Table S2 given in the Supporting Information. Generally, either one of the two reactions is slightly more endergonic, thereby limiting the O2 exchange capacity. Figure 1 B shows the limiting free energy of a redox cycle near the intersection of both correlations. As indicated by the volcano‐shaped curve, the ideal redox material binds oxygen strongly enough to oxidize the oxide at relatively low temperatures—stronger than the Ag/Ag2O reference—but weakly enough to reduce the oxidized redox material at moderately higher temperatures—weaker than the Cu2O/CuO reference. The ideal metal oxide compositions are where these effects balance, located near the top of the volcano curve. Ideally, this region corresponds to negative free energies for both reactions. For the temperatures chosen in our analysis, this can be achieved with rare materials such as Rh2O/RhO or toxic materials such as PbO/Pb3O4.15 Generally, the volcano‐like shape of this correlation is due to the fact that the amount of energy absorbed for breaking metal–oxygen bonds during the reduction step correlates with the amount of heat liberated when forming these bonds during the oxidation step. Thus, as shown in Figure S1, the location of the volcano‐top can be determined from only computing the reduction enthalpy as the entropy of O2 gas participating in either reaction is the same for all specific redox couples and as entropic contributions of the solids introduce significant deviation from these correlations only at significantly higher temperatures when approaching melting and boiling points.

To tailor inexpensive and non‐toxic metal oxides, we calculated the free energy of oxygen vacancy formation (ΔG v[O]) using density functional theory (DFT) for twelve perovskites that have attracted attention for solid‐oxide fuel cells,6a, 10 air separation,6c and solar‐thermal applications.16 Stoichiometric ABO3(010) and oxygen‐deficient ABO2.5(010) facets (A=Sr, Ba, or La; and B=Mn, Co, Ni, or Cu) were modeled (Figure 1 C), using the grid‐based projector‐augmented wave (GPAW) and atomic simulation environment (ASE) electronic‐structure code.17 Figure 1 B shows the thermochemical stability and the reaction energetics as calculated from the scaling of ΔG v[O] and the redox energetics of the bulk oxides (see the Supporting Information).18 The analysis predicts an ideal O2 exchange capacity for SrCoO3, relative to too strong and too weak oxygen binding for BaMnO3 and BaCoO3, respectively.

We validated this descriptor‐based design for metal oxide redox materials by means of dynamic O2 exchange experiments using SrCoO3−, BaMnO3−, and BaCoO3−, synthesized via the Pecchini method3b and commercial Ag2O and Cu2O as reference materials. The composition and surface morphology of all solids were characterized using high‐temperature X‐ray diffraction (HT‐XRD) and scanning electron microscopy (SEM). The O2 exchange capacity and exchange rates were determined by thermogravimetric analysis (TGA).

Figure 2 A and B display dynamic TGA runs for SrCoO2.95, BaCoO2.58, BaMnO2.94, Ag2O, and Cu2O (initial stoichiometries) that were cyclically reduced at 900 K and 0.2 bar O2 (simulating air) and oxidized at 600 K and 0.035 bar O2 (simulating the composition of the gas phase from reducing ceria for solar‐driven splitting of CO2 and H2O19). As expected, Cu2O and Ag2O oxidize and reduce strongly to CuO and Ag, respectively, essentially without exchanging O2 reversibly. Compared to these reference materials, we find augmented O2 exchange capacities for the perovskites. SrCoO2.95 reaches a maximum O2 exchange capacity of 44±0.012 mmol O2 per mol SrCoO2.95 and a maximum O2 exchange rate of 12.1±0.003 μmol min−1 g −1 whereas BaCoO2.58 and BaMnO2.94 perform at much lower capacities of 3.4±0.015 and 0.5±0.015 mmol O2 per mol of perovskite and lower exchange rates of 0.8±0.003 and 0.04±0.005 μmol min−1 gperovskite −1, respectively. As predicted by DFT, the performance of BaCoO3− and BaMnO3− appears limited by reoxidation and reduction, respectively. This theory‐based screening of twelve perovskites identifies a well‐known composition, SrCoO3, that shows a significantly augmented performance for this novel application compared to the O2 exchange capacities and rates of the reference materials at the same conditions. Additionally, with an O2 exchange rate of 12.1 μmol min−1 gperovskite −1, SrCoO3− outperforms the state‐of‐the‐art Cu2O/CuO cycle, which cannot be used with low‐grade process heat at 600–900 K and which has an O2 exchange rate of only 10 μmol min−1 g −1 (for both, oxide reduction and oxidation) at significantly higher—and thereby economically less attractive—temperatures of 1120–1450 K and comparable pO2.3a

Figure 2

Dynamic O2 exchange: TGA runs of A) Ag2O and Cu2O and B) SrCoO3−, BaCoO3−, and BaMnO3−. C) Lattice constants of SrCoO3−, BaCoO3−δ, and BaMnO3− derived from HT‐XRD analyses of oxide reduction at 0.2 bar pO2 (empty circles) and oxide oxidation at 0.035 bar pO2 (filled circles). Error bars are standard deviations within a 68 % confidence interval.

To support that the TGA data is indicative of reversible O2 exchange, Figure 2 C shows the perovskite lattice constants computed using data from HT‐XRD analysis. Although all perovskites exhibit thermal expansion upon heating, only SrCoO3− shows a major difference in the lattice expansion at varied pO2, which can be attributed to the formation and filling of O vacancies.20 This, along with the electronic structure trends and the TGA analysis, suggests that SrCoO3− is particularly suitable as a redox material for TSSOS. To understand how the oxide composition controls the O2 exchange, Figure 3 A plots ΔG v[O] versus the DFT‐calculated bond length between the transition metal and the nearest O atom (d B—O) at ABO2.5(010). Generally, we observe stable O vacancies (negative ΔG v[O] values) correlating with large d B—O values. This is in agreement with the lattice expansion shown in Figure 2 C, as the lattice expansion due to a higher chemical potential for oxygen in the gas phase decreases the bonding of oxygen in the solid, which, in turn, increases the length of the metal–oxygen bond. Although the slope of this correlation is essentially due to the metal at the B‐site interstices, the absolute value of d B—O is governed by the metal at the A site, as demonstrated by the consistently higher d B—O values of the Ba versus the Sr compounds. However, the scatter of the correlation shown in Figure 3 A suggests that the trends in the free energy of the O vacancy formation and in the metal–oxygen bond length cannot alone be rationalized with geometric arguments.

Figure 3

Correlation of Δ G v[O] with A) d B—O and B) q O (empty, light‐gray, and dark‐gray symbols mark SrBO2.5, LaBO2.5, and BaBO2.5, whereas circles, squares, diamonds, and triangles mark ACuO2.5, ANiO2.5, AMnO2.5, and ACoO2.5). C) Charge density differences (CDD) of the marked surface after O vacancy formation relative to the stoichiometric surface and the reference gas‐phase O2 (given at the height of the transition metal cation).

Figure 3 B plots ΔG v[O] versus the DFT‐calculated partial charge21 of oxygen (q O). We find that perovskites that form O vacancies easily contain weakly charged oxygen. BaCoO3, for instance, accumulates less charge at the O anion than SrCoO3 and BaMnO3, which correlates with the facile reduction of BaCoO3. Generally, ΔG v[O] scales with q O, with the slope corresponding approximately to the ionization energy of oxygen, −13.62 eV per electron22 and the intercept reflecting entropic contributions to ΔG v[O] and the reference chemical potential of oxygen in the gas phase (see the Supporting Information). The quality of this linear correlation suggests that the trend in the free energy of the O vacancy formation is controlled by the enthalpy of breaking metal–oxygen bonds, that is, by the quantity of electric charge transferred from the O atom to the lattice when forming the O vacancy. This is illustrated with Figure 3 C, which shows the difference in the charge density distribution due to O vacancy formation at SrCoO3(010) and BaMnO3(010). Although the partial charge of the O anion yielding the vacancy is approximately equal at both surfaces (Figure 3 B), the charge transfer from the bonding O 2p states to the 3d states of the transition metal is significantly higher at BaMnO3(010) relative to SrCoO3 (010). We note that this analysis describes the width of the 2p–3d gap for 3d transition metals at the B site of ABO3 perovskites. Incorporating other metal cations into the B interstices, such as lanthanides with f states, may alter these trends due to a different entropy‐controlled shape of the states that are accepting electrons when forming O vacancies.

In summary, analogous to the enthalpy difference of bulk oxide reduction, we suggest that the charge transfer from O 2p to B‐site‐metal 3d states explains the higher O2 exchange capacity of SrCoO3− versus BaMnO3− at the atomic scale.

In accord with charge transfer controlling the formation of O vacancies at the oxide surface, our Sabatier analysis employs the enthalpy of the bulk oxide reduction as a descriptor of the redox energetics. To demonstrate how this information can be used practically to predict the O2 exchange capacity of a metal oxide, we have determined the O2 exchange capacity for five metal oxides as the difference of the oxygen non‐stoichiometry at equilibrium between O2 evolution at 900 K and 0.2 bar pO2 and O2 fixation at 600 K and 0.035 bar pO2 (Figure 4). The plot shows that the O2 exchange capacity resembles the volcano‐shaped trend of the limiting redox energetics. Relative to this trend across three orders of magnitude are minor deviations, such as for BaCoO3−, which are presumably due to differences in surface morphology, crystal structure, and the computational versus experimental non‐stoichiometry (see the Supporting Information). This demonstrates how computing the enthalpy of the oxide reduction from first principles allows predicting the O2 exchange capacity of metal oxides.

Figure 4

Predicting the O2 exchange capacity and the limiting free energy of solid‐state O2 separation from the enthalpy of the oxide reduction at 298 K and 1 bar.

Although the present article focuses on the thermodynamics of O vacancy formation, the kinetics of conducting these vacancies from and to the surface are of equal importance when designing metal oxides for solid‐state O2 separation. The defect chemistry of several perovskite families, including La2NiO4+ with a relatively high mobility of O interstitials, has been investigated previously.23 In these perovskites excess oxygen is incorporated as interstitial O2− or O− anions with anion Frenkel pairs being the predominant intrinsic lattice defects.23 Oxygen transport may be anisotropic23 or isotropic, such as in Sr0.75Y0.25CoO2.625.24 A low Frenkel energy may yield high O vacancy concentrations, in PrBaCo2O5.5 for instance,25 with an ordered sublattice of the A cations ensuring high oxygen ion mobility.25 Similarly, DFT was used previously to predict and understand oxygen conduction trends in metal oxides.18 These and other studies23, 26 outline the prospects of an advanced understanding of oxygen conduction for designing advanced redox materials.

Conclusions

Based on the Sabatier principle applied to the bonding of lattice oxygen atoms, we implemented a descriptor‐based design principle for predicting the O2 exchange capacity of metal oxides and perovskites in particular. The computations were validated through dynamic O2 exchange experiments using Ag2O, Cu2O, and three perovskites and rationalized based on the composition‐dependent bond geometry and charge transfer during the formation of O vacancies at the metal oxide surface. SrCoO3− was identified as an ideal material for solar‐driven thermochemical separation of O2. In a broader context, the presented principles may also aid the design of oxygen conductors for related applications, such as solid‐oxide fuel cells and air separation using dense ceramic membranes.

Experimental Section

Thermochemical equilibrium calculations—To guide the design of redox materials, the thermochemical equilibrium of binary bulk metal oxides, O2, and their reduction products was determined at the specified pO2 and temperatures from tabulated free‐energy data, with an absolute accuracy of 1–10 kJ mol−1.13 Per convention, negative free energy differences mark exergonic reactions. At the computed conditions, the correlations of ΔG and ΔH red for the metal oxide oxidation and reduction have average relative errors of 12.4 % and 17.6 %, respectively. These values increase with increasing temperature due to entropic contributions and indicate that the provided volcano plot could be employed for identifying materials active for thermochemical solid‐state O2 separation.

Electronic structure calculations—Twelve perovskite surfaces were modeled using DFT, performed with the GPAW code.17b,c Exchange‐correlation interactions were treated by the revised Perdew–Burke–Ernzerhof (RPBE) functional.27 Atomic configurations were handled in ASE.17a A Fermi–Dirac smearing of 0.1 eV was used to achieve convergence, and the structure optimization results were extrapolated to 0 K. The linesearch Broyden–Fletcher–Goldfarb–Shanno (BFGS) algorithm was employed to optimize the atomic geometries until the maximum force was less than 0.05 eV Å−1. The utility of DFT+U methods for surface calculations is not determined in general and was found unnecessary for many metal oxides.18, 28 Here, for all DFT calculations the generalized gradient approximation (GGA) was used without a Hubbard U term as we found previously that the Hubbard U correction did not improve the description of surface reactivity with the employed models.18 The cubic bulk structures of ABO3 compositions consisted of one metal atom (Sr, Ba, or La) at the twelve‐coordinated A‐site interstices, one metal atom (Mn, Co, Ni or Cu) at the six‐coordinated B‐site interstices, and three oxygen atoms that were allowed to optimize their positions (relax). The bulk structures had periodic boundary conditions in all directions and were modeled using a ‐point sampling of 4×4×4. Compositions containing Mn, Co, or Ni were modeled using spin‐polarized calculations and, to avoid reminiscent stress in the calculations, the lattice constants were chosen as the DFT‐calculated bulk lattice constants. Table S1 provides a summary of the lattice constants and magnetic moments along with a discussion of the accuracy of the employed DFT methods that predict the lattice constants within ±0.36–3.89 % of experimental values. The AO‐terminated ABO3(010) facet was chosen for modeling the perovskite surfaces as this facet was identified as being the thermodynamically most stable surface of cubic perovskites for various compositions.18 The surface models consisted of one upper ABO3(010) layer that was allowed to relax and one lower ABO3(010) layer constrained to the bulk geometry. The surfaces were periodically repeated in the directions parallel to the surface and were modeled with 10 Å of vacuum perpendicular to the surface. The Brillouin zone of surface models (slabs) was sampled using 4×4×1 ‐points. The partial charge density was determined for all atoms contained in the surface models through Bader decomposition.21 To model the perovskite surfaces at different O vacancy concentrations, one third of the stoichiometric lattice oxygen in the upper surface layer was removed while the oxygen concentration in the lower surface layer was maintained. This yielded “reduced” and “oxidized” surface models with A2B2O5(010) and A2B2O6(010) stoichiometry, marked with the conventional ABO3(010) and ABO2.5(010) notation, respectively. The free energy of forming O vacancies (ΔG v[O]) at the surface was computed as follows [Eq. (3)]:18, (3)

where G v, G s, and GO r are the free energies of the perovskite surface with the O vacancies, the stoichiometric surface and the reference energy of the liberated lattice oxygen [taken as the energy difference of stable H2O and H2 in the gas phase, see Eq. (3) in the Supporting Information], such that negative free energies indicate exergonic reactions. The formation of O vacancies as computed corresponded to formation of one monolayer O vacancies equivalent to an oxygen non‐stoichiometry of δ=0.5 in ABO2.5 (Figure 1 C). Details on converting the DFT‐computed electronic energy to Gibbs free energy at 298.15 K and 1.013 bar, the reference energies, and scaling relations18 to estimate the bulk formation energies are provided in the Supporting Information. The error of DFT‐computed adsorption energies (employed as a descriptor of surface reactivity, comparable to the energy of forming surface O vacancies in this work) was estimated previously to be 0.08 eV.29

Perovskite synthesis—Three perovskites, namely SrCoO3−, BaCoO3−, and BaMnO3−, were synthesized using a modified Pecchini method, employing stoichiometric amounts of Mn(NO3)2⋅4 H2O (Alfa Aesar, 98 %), Sr(NO3)2 (Alfa Aesar, 98 %), Co(NO3)2 (Alfa Aesar, 97.7 %), Ba(NO3)2 (Alfa Aesar, 99 %), C2H6O (Alcosuisse, 96.1 %), and C6H8O7 (Fluka, ≥99.5 %). The solid products were ground using mortar and pestle, uniaxially pressed into pellets (10 metric tons, 6 and 25 mm in diameter), and sintered in air at 1473 K for 5 h (SrCoO3− and BaCoO3−) or in pure O2 at 1273 K for 48 h (BaMnO3−). To achieve the desired oxidation state, the sintered SrCoO3− was ground, fully immersed in NaClO (Migros, <5 % in H2O), washed with deionized water, and subsequently dried for at least 2 h at 473 K. Ag2O (Merck, ≥99 %) and Cu2O (Johnson Matthey Alfa, 99.5 %) were used as reference materials.

Solid‐state analysis—XRD and HT‐XRD were performed in the Bragg Brentano geometry using CuKα radiation (20–80° 2θ, 0.06° min−1 scan rate, 45 kV/20 mA output, PANalytical/X‘Pert MPD/DY636, Philips). The original oxygen content of the perovskites was SrCoO2.95, BaCoO2.58, and BaMnO2.94 before the redox cycling, as determined using TGA (STA 409/C/3, Netzsch) of the complete reduction of the metal oxides in 5 % H2 in Ar at 823 K and 1 bar. To estimate changes in the lattice constants, the perovskites were reduced by heating from 600 to 900 K in 100 K steps at 0.2 bar pO2 and thereafter oxidized by cooling from 900 to 600 K in 100 K intervals at 0.035 bar pO2. The morphology of all materials was analyzed using SEM (15 kV accelerating voltage, TM‐1000, HITACHI) and is shown in detail in the Supporting Information.

Thermochemical cycling experiments—Starting materials (0.1 g) were placed in an Al2O3 crucible supported with an Al2O3 rod on the microbalance of the TGA (±0.1 μg). The materials were thereafter exposed to a gas flow (constant flow rate of 200 mL min−1 at 273 K and 1 bar) with specified pO2, which was adjusted by mixing O2 (99.5 %, Messer) and N2 (99.999 %, Carbagas) using three electronic mass flow controllers (MFC400, Netzsch; accuracy ±1 %, precision ±1 mL min−1). The mass change of the samples was recorded during two consecutive redox cycles with oxide reduction at 900 K and 20 vol % O2 and oxide oxidation at 600 K and 3.5 vol % O2, respectively (±1 K). Heating and cooling was performed at +10 K min−1 and −10 K min−1, respectively. To correct for buoyancy, blank runs were performed using the same measurement conditions employed for the experimental runs.

The oxygen exchange capacity (dimensionless) was defined as the difference in the oxygen non‐stoichiometry of the metal oxides after the oxide reduction (δ red) and after the oxide oxidation (δ oxi):(4)

where m red and m oxi are the metal oxide mass (in g, determined using TGA), after the oxide reduction, and after the oxide oxidation, and MO is the molar mass of oxygen (in g mol−1).The uncertainties of the oxygen exchange capacities and rates were estimated using error propagation from the accuracies of the experimental analysis.

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re‐organized for online delivery, but are not copy‐edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

miscellaneous_information

Click here for additional data file.