Metal–organic framework electrocatalysis: More than a sum of parts?

Material discovery for catalysis is essential to driving forward the decarbonization of our society using the renewable energy based on wind and solar. Here, electrolysis can enable the key to store electricity in the form of chemical bonds.¹ However, electrolysis is limited by the catalyst at the electrodes, i.e., the electrocatalyst. Water electrolysis for H₂ production is an important example of an electrocatalytic process, but more exotic reactions are needed to produce our most valuable fuels and chemicals.² To effectively tailor electrocatalysts, advanced material discovery is required. Unfortunately, for electrocatalysis, there is an extremely challenging set of material constraints in terms of activity,³ selectivity,⁴ and stability.⁵ This leaves typical electrocatalysts to be metals,^3,4 metal oxides,⁵ and, recently, metal–nitrogen–carbon (MNC) structures.^6,7 The goal is to discover catalysts that fulfill these requirements and, at the same time, have radically different properties to lead to a breakthrough in the energy transition.

Metal–organic frameworks (MOFs) could lead this breakthrough by opening a new research direction in catalysis, owing to their pronounced structural differences to currently used benchmark systems. MOFs are porous solids, formed by coordination bonds between organic ligands and metal ions or clusters.⁸ MOFs have been investigated for various applications because of their high porosity, internal surface area, and chemical tunability.⁹ Their nanoporous structure can also give rise to an unusual electric double layer.¹⁰ Unfortunately, MOFs with high levels of electrical conductivity are rare¹¹ but increasing. A mixed valence MOF with a conductivity of 1 S/cm has been demonstrated.¹² However, this is still around 100 times less than that of metal–nitrogen–carbon based materials. The electrical conductivity of MOFs, along with the chemical tunability of their active sites, has allowed the exploitation of such materials in electrocatalysis.^13,14

Experimentally, the field of MOF electrocatalysis is growing rapidly.¹⁵ Hod et al. demonstrated the activity of Fe-porphyrin-based MOFs for electroreduction of CO₂ (CO₂RR),¹⁶ where Zn-based ZIF-8 and CALF20 acted as CO₂ to CO electrocatalysts.¹⁷ Wang et al. showed close to 100% Faradaic efficiency on a Co-PMOF structure for the reaction.¹⁸ For oxygen reduction reaction (ORR), the ZIF-8 MOF structure has been investigated by replacing Zn with Fe to achieve high and dense Fe–N₄ sites;^19,20 in addition, special MOF linker spacing in PCN-226 has been suggested to improve the reaction kinetics,²¹ and for oxygen oxidation reaction (OER), a bifunctional MOF catalyst has been reported.²² However, one should be careful as it is not always clear if the MOF is an active catalyst or a precatalyst.²³ In the case of porphyrin-based MOFs, they could break down to active derivatives,^6,7 and in the case of bifunctional NiFe-based MOFs, they could break down to an optimal NiFe oxide catalyst.^24–26 Furthermore, it is noteworthy that metallic Zn works for CO₂ to CO electrocatalysis, challenging the observation of active ZIF-8 and CALF20 catalysts,¹⁷ not to mention Cu-based MOFs for CO₂RR,²⁷ which could break down to metallic Cu particles that are known CO₂RR catalysts giving rise to a large variation of products.^2,28–32

In this work, we highlight the use of MOFs as electrocatalysts by studying in depth MOF-525 as a template example, due to the fact that it has been studied experimentally as a working electrocatalyst for CO₂ reduction¹⁶ and the fact that different metalations change the charge transport rate in MOF-525.³³ In addition, MOF-525 contains common features of MOFs, such as a metal node and porphyrin linker. We use density functional theory (DFT) (see the “Computational details” section), to study MOF-525 variants as depicted in Fig. 1(a). Figure 1(b) shows how the density of states for the structure changes as a function of different iron levels in the four nitrogen pockets of porphyrin. This indicates that the conductivity of this structure is a function of metal incorporation. While we acknowledge that MOF-525 could decompose to multiple different derivatives, we made the choice to construct and study two derived structures from MOF-525, which we call the TCCP molecule model [Fig. 1(c)] and the TCCP slab model [Fig. 1(d)]. This enables us to compare a MOF electrocatalytic performance with its constituent parts, and hereby estimate if the MOF and/or the degradation models are more or less active.

Fundamentally, metal and metal oxide materials are understood as catalysts.³⁴ The binding energies of the important reaction intermediates scale linearly³⁵ with each other on these surfaces, which allows for a quantification of Sabatier activity volcanoes.³⁶ However, these catalytic fundamental properties are not necessarily present for materials as complex as MOFs, with an organic framework and possible chemical changes during absorption.

MOF-525 contains 276 atoms, and we explore each atom as the possible binding site for *H, *COOH, *OH, *OOH, and *O intermediates to test where the catalytic intermediates bind, how strong they bind, and the effect on the electronic properties. However, we note that in certain cases, *COOH may react to form *H and CO₂ or *O reacts with the carbon framework to form *OH. These structures were discarded using a maximum distortion of 0.8 Ang.

Figure 2 depicts the matched binding energies within a 2 Ang minimum distance for MOF-525 for (a) *COOH vs *H and (b) O* vs *OH, while Fe-MOF-525 is shown in (c) *COOH vs *H and (d) *O vs *OH. The data points are colored by the DFT bandgap, and the diagonal in each plot is represented by a black dashed line. For MOF-525, it is observed that *H and *COOH adsorb strongest on the unsaturated nitrogen sites on the TCCP linker, whereas *O and *OH has binds strongest on the carbon framework, as depicted by the figure insets below the figure. For Fe-MOF-525, the Fe site is the strongest for all four intermediates, as also shown by the insets below. Interestingly, this analysis shows a spread on the catalytic reaction intermediates that typically scale (*COOH vs *H³⁷ and *O vs *OH³⁸). It seems that the catalytic intermediate variation on the pristine MOF-525 is larger than that on Fe-MOF-525. We believe that, in both cases, this is likely due to the formation of organic bonds that are avoided on traditional metals.

We plot these reactions as a function of the electrochemical potential in Fig. 3 for (a) MOF-525 and (b) Fe-MOF-525. This allows one to interpret when, for example, *H or *OH adsorbs as a function of potential. The maximum atomic distortion is also shown, which acts as a descriptor for which reaction may lead to the decomposition of the MOF. In addition, we add red and blue regions for visualizing where the structure undergoes reduction and oxidation. In Fig. 3(a), a few *H are adsorbed close to 0 V vs Reversible Hydrogen Electrode (RHE), which is the adsorption on the unsaturated N. This adsorption leads to a minor distortion with an ∼1.25 eV bandgap. Under oxidizing conditions, the initial oxidation of the structure takes place by *OH adsorbing around 1.5 V vs RHE or removing *H around 1.6 V vs RHE. We note that this intermediate oxidation is above the experimental carbon corrosion potential at 0.9–1.0 V vs RHE.^39,40 Fe-MOF-525 in Fig. 3(b) shows a different picture. The first *H is adsorbed on iron at negative potentials with negligible distortions and a small bandgap. The oxidation region changes with *OH adsorption from water oxidation first taking place at Fe at ∼0.5 V vs RHE. The fact that oxidizing the carbon framework still takes place at around 0.9–1.0 V vs RHE^39,40 shows that the MOF framework is probably not stable at mild oxidation potentials but could be used as a reduction reaction electrocatalyst with Fe doping.

Observing that Fe acts as an active site in MOF-525, we extend this to a series of metals (Fig. 4). Figure 4(a) shows how the density of states varies significantly around the Fermi-level (dashed line), illustrating that MOF-525 bandgap depends on metal doping and closes for certain metals. Figures 4(b) and 4(c) show that the scaling relations are intact with respect to metals. However, the slope and intersections as compared to previous work deviate, possibly due to the electronic structure. We also show the mean absolute error (MAE), which displays the simulated point errors from the line. While the MAE is low for *COOH vs *H and *OOH vs *OH (∼0.1 eV), it is significantly higher for *O, possibly due to the need of pulling two electrons from the metal site as compared to one. Figure 4(b) has a slope of ∼1 and intersection of −0.4, while previous results on porphyrin systems show a slope of 1 and intersection of 0.³⁷ The deviation in the intersection originates in the different choice of functional and CO vs using CO₂ as reference molecules, which is a known challenge in simulating CO₂RR intermediates.^41,42 The scalings shown in Figure 4(c) also deviate slightly from *OOH, *OH, and *O scalings as discussed elsewhere.⁴³ 

As MOFs could be a precatalyst²³ or simply degrade to the porphyrin motif or a derivative metal cluster, we compared DFT binding energetics of MOF-525, with a TCCP linker molecule model (blue) and a TCCP slab model (red), in Fig. 5 (see Fig. 1 for motifs). It shows that the intermediates binding at the MOF or at the derived structures are close to similar. Only a minor spread is shown on *H and *COOH intermediates in Figs. 5(a) and 5(b). For the oxygen-bound intermediates *OH, *OOH, and *O, the binding energetic spread is larger [Figs. 5(c)–5(e)], in particular for *OOH with both slope and intersection differences. This means that for oxygen reactions (i.e., ORR), there would be a catalytic activity difference depending on MOF-525 being intact during the reaction or degraded. In contrast, for Hydrogen Evolution Reaction (HER) and CO₂RR, the absolute value scatter for each metal doping on the three structures may influence the catalytic activity which we show below.

To better display the catalytic activity, the binding energies are transferred to the HER volcano [Fig. 6(a)], the ORR volcano [Fig. 6(b)], and the CO₂RR Faradaic efficiency scheme as a function of hydrogen similar to Ref. 2 [Fig. 6(c)]. When depicting the binding energies in the volcano scheme, it becomes clear that although the energetics between MOF-525 and the derived parts are linear, as shown in Fig. 5, their absolute value variations matter for activity and selectivity. A specific example to mention is the Co-based catalyst. For HER [Fig. 6(a)], Co-MOF-525 binds close to the top of the volcano on the strong binding side, while the degraded models bind weaker on the weak binding side of the volcano. For CO₂RR [Fig. 6(c)], which is highly sensitive to the *H binding, Co-MOF-525 binds *H strongly, giving rise to only H₂ evolution, while the degraded models are almost perfectly situated close to the calculated *H binding on metallic copper, which is known to give a series of interesting CO₂RR products.^28–32 This corresponds well to known Co compounds performing catalytically with a large variation for CO₂RR.^6,44–46 For ORR Fe-based MOFs, TCCP_mol and TCCP_slab structures are on the strong binding side of the volcano, which typically gives rise to four-electron ORR.⁴⁷ Fe-MOF-525 binds slightly stronger than Fe-TCCP_mol and Fe-TCCP_slab, given the latter predicted a better activity. The Co based structures should also depend strongly on the actual structure, being that Co-MOF-525 is close to the top at the strong binding side and Co-TCCP_mol and Co-TCCP_slab are on the weak side, giving rise to two-electron ORR toward hydrogen peroxide. Noteworthy is that these activity schemes are based on the computational hydrogen electrode,⁴⁸ which shows activity on the reversible hydrogen electrode scale. However, it is experimentally known that the activity for molecular and M-Nx catalysts can vary as a function of pH, and it has been debated how the activity relates to the redox potential of the catalyst.⁴⁹ Computational activity prediction as a function of pH is an ongoing challenge, with the development of grand canonical approaches⁵⁰ and constant potential methods.⁵¹ These methods are currently being employed at the electrolyte–metal interface and are beyond the scope of investigating MOFs and their derivatives. Overall, it displays that the actual structure under catalysis conditions is important for activity and selectivity.

In summary, we have theoretically explored the electrocatalytic activity of pristine and metal-loaded MOF-525. The carbon framework allows for hydrogen (*H), hydroxyl (*OH), and oxygen (*O) adsorption from aqueous reactions in an electrochemical setup. The bandgap closes for the Fe-MOF-525 structure, and the metal becomes the active site. The incorporation of different metals on both MOF-525 and its constituents with a molecular model and a slab model of the TCCP linker shows a linear binding energy behavior. However, the absolute values of MOF-525, the TCCP molecular model, and the TCCP slab structure show changes for HER and, indeed, selectivity for CO₂RR when using predictive schemes. We suggest further experimental studies of MOFs with exposed Co and Fe sites for reduction reactions as we predict these to be conductive, electrocatalytically stable, and active catalysts.

A.B. acknowledges the financial support from the Carlsberg Foundation (Grant No. CF21-0144) and the Pioneer Center for Accelerating P2X Materials Discovery (CAPeX, DNRF Grant No. P3). This project was also supported with funding from Samsung Electronics Ltd. (SAIT). Via our membership of the UK’s HEC Materials Chemistry Consortium, which is funded by EPSRC (Grant No. EP/X035859/1), this work used the ARCHER2 UK National Supercomputing Service (http://www.archer2.ac.uk).

The authors have no conflicts to disclose.

Alexander Bagger: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Aron Walsh: Conceptualization (equal); Project administration (equal); Supervision (equal); Writing – review & editing (equal).

The data that support the findings of this study are available within the article.