Mechanistic routes toward C₃ products in copper-catalysed CO₂ electroreduction

Abstract

The electrocatalytic CO₂ reduction (eCO₂R) reaction powered by renewable electricity holds promise for the sustainable production of multi-carbon chemicals and fuels. On Cu-based catalysts, ethylene and ethanol (C₂) have been produced in appreciable amounts. C₃ products (mostly terminal oxygenates) have limited yields, whereas propylene is puzzlingly absent. Herein, we devise a divide-and-conquer strategy to explain the formation of the C₃-backbone and elucidate the mechanism responsible for the observed selectivity by combining network graphs, density functional theory, and experiments to prune the network and benchmark the identified path. Our approach concludes that the most frequently reported products, propionaldehyde and 1-propanol, originate from the coupling of CH₂CH with C(H)O. While propylene and 1-propanol share common intermediates, the former is barely produced due to the unfavourable formation of allyl alkoxy (CH₂CHCH₂O), whose crucial nature was confirmed experimentally. This work paves the way for tailoring selective routes towards C₃ products via eCO₂R.

---

Introduction

Developing functional catalysts for the electrochemical CO₂ reduction (eCO₂R) to complex products lies at the core of new efforts to develop sustainable technologies.¹ Among available materials, copper-based electrocatalysts occupy a pivotal role due to their ability to form the C₂₊ backbone for high-value fuels and commodity chemicals.^1,2 The type and amount of products formed are sensitive to the applied potential, electrolyte, and the preparation protocol of Cu.^3–6 The established mechanism to the C₂ fraction advocates that CO₂ first reduces to CO, which dimerises to OCCO⁻ and subsequently reduces to hydrocarbons and alcohols. Typically, the main C₂ product is ethylene (up to ca. 74% Faradaic efficiency FE),⁷ although an exceptional ethanol selectivity (ca. 91% FE)⁸ has been reported on Cu clusters. For C₄ products, the aldol condensation of acetaldehyde (C₂) gives crotonaldehyde, which reduces to 1-butanol, albeit with low yields.⁹ Among C₃ compounds, 1-propanol can be produced with appreciable yields (∼23% FE),^10,11 whereas propylene (the corresponding C₃ olefin; 0.36 eV less stable than 1-propanol, Table S1†) has only been detected as a trace product (<0.1% FE).⁹ This puzzling outcome contrasts with the vast formation of ethylene, which is less stable than ethanol by 0.47 eV, Table S1.† Furthermore, 2-propanol, which is the most thermodynamically stable C₃ alcohol (0.17 eV lower than 1-propanol, Table S1†), has never been observed in eCO₂R.¹² Reports have indicated that the formation of the C₃ backbone at high CO concentrations and relatively mild applied potentials (−0.36 to −0.56 V vs. RHE)^11,13 requires asymmetric sites on oxide-derived Cu (OD-Cu) catalysts. Nonetheless, the reasons behind the low selectivity to C₃ products in eCO₂R at a molecular level is very limited. This is due to the large number of elementary steps (>10³) that prevent the use of standard reaction sampling tools based on explicit density functional theory (DFT) and reaction profile analysis.

Herein, we analyse electrocatalytic routes towards C₃ products through a divide-and-conquer strategy based on the generation of the network graph and computational reaction profiles combined with key electrochemical experiments involving C₁, C₂, and C₃ reagents. This new methodological approach enables us to (i) identify the most likely C₁–C₂ coupling steps towards C₃ intermediates; (ii) elucidate the bifurcation points to different C₃ products; and (iii) pinpoint kinetic bottlenecks hindering propylene production.

Experimental and computational methods

The electrocatalytic reactor used for our experiments is a gas-tight cell consisting of two compartments separated by a Nafion 211 membrane with gas-flow inlet and outlet ports. The cell has an OD-Cu working electrode, a gas diffusion layer (GDL) carbon paper counter electrode, and a leak-free Ag/AgCl (3 M KCl) reference electrode. Triplicate measurements were done, with the average values and standard deviations presented in the ESI.† Extended details about reagents, catalysts preparation, electrochemical measurements, detection limits, and product analyses are shown in Experimental and computational details in the ESI.† The OD-Cu catalyst was obtained from CuO (see Fig. S1† for XRD analysis). The physicochemical and catalytic properties of this material have been discussed elsewhere.^9,14

We initially reduced CO₂ in 0.1 M KHCO₃ at −0.95 V vs. RHE to maximise the production of multi-carbon products,¹⁴ as shown in Fig. 1 (see Fig. S2† for polarisation curves). C₁ and C₂ products account for 53% of the FE, whereas 7% corresponds to C₃ products and the balance is H₂. Our observed product distribution agrees with selectivity trends presented in the literature, which are summarised in Fig. S3,† and the paths to C₂ products are shown in Fig. S4–S5.† To unravel the selectivity patterns observed from both the literature and our experiments, our workflow entails: (i) building the reaction network by encoding the corresponding structural graphs; (ii) sampling the intermediates by DFT; (iii) computing all C₁–C₂ backbone couplings by DFT; (iv) pruning the network of non-viable backbone formation routes by probing the products from CO, formaldehyde, and methanol co-reduction with C₂ reactants (Table 1), with particular attention to missing products; (v) computing all routes from the C₃ backbone to the final products using DFT and linear-scaling relationships (Table S15†) to identify the best routes towards propanol and propylene; (vi) experimental benchmarking of the main predicted routes via electrocatalytic tests with key intermediates.

Fig. 1 C₁–C₃ products formed from the electrocatalytic CO₂ reduction on oxide-derived copper in 0.1 M KHCO₃ at −0.95 V vs. RHE. Detailed results are presented in Table S2.† Hydrogen FE is 37.4%.

Table 1 Summary of C₃ products and their corresponding formation rates observed experimentally from the electrolysis of C₁ or a mixture of C₁ and C₂ compounds on OD-Cu. The C₁ and C₂ compounds used are listed in the topmost row and leftmost column respectively, while the experimental conditions are indicated in the footnotes. A colour code has been added to highlight the higher C₃ formation rates. For reference, the formation rates of C₃ products from eCO₂R are given in parenthesis, coloured scale is indicated in the lower bar. The full set of experiments and product distributions are shown in Tables S2–S14†

---

While the routes to C₁ and C₂ products can be probed manually, as shown in the literature,^15–20 the analysis of routes to C₃ products demands automation. The full network containing all C₁, C₂, and C₃ intermediates has 463 elements, represented here as nodes in a graph (see “Graph representation of the reaction network” in Experimental and computational details). The energies of intermediates (referenced to CO₂, H₂O, and H₂) were evaluated with the computational hydrogen electrode (CHE)^21,22 containing the DFT energy obtained with a PBE-D2 formulation^23–25 (corrected for metal overbinding), our in-house developed implicit solvation model,^26,27 and the polarisation term^9,28 (see eqn S1–S4† in the Experimental and computational details). The D2 contribution on H* and CO* adsorption is small: 0.04 and 0.14 eV, respectively. Intermediates are linked by 2266 steps (edges linking the nodes in the graph): 55 C₁–C₁ and 636 C₁–C₂ couplings, 683 C–H and 305 O–H hydrogenations, and 587 C–O(H) cleavages. To ensure the desired accuracy, 586 out of all C₁–C₂ couplings (Tables S16–S19,† 92% of total), 10 C–O(H) breakings (Table S20†), and 8 hydrogenations (Table S21†) were explicitly obtained via nudged elastic band (DFT-NEB)²⁹ and confirmed by vibrational analysis. Initial guesses for NEB were generated automatically (Experimental and computational details in ESI† and Notes S1–S4). Linear scaling relationships (LSR) were employed in the initial fast-sampling of C–H and O–H hydrogenations as they are reliable for these cases^20,30,31 (Table S15†). The transition states for key hydrogenation steps in the main path were further refined with DFT-NEB. Heyrovsky-type reactions for C–OH breakings and C–H formations were considered (Note S5†). Tests on density functionals, LSR, and charge analysis are described in Note S6 and Fig. S6–S9,† and demonstrate that our strategy provides an excellent cost-efficiency balance.

Benchmark electrochemical experiments (Table 1) involving the reduction of selected C₁ and C₂ compounds and their mixtures were conducted at mild overpotentials (−0.4 V vs. RHE) in alkaline pH, where the production of multi-carbon products is expected to be boosted.^32,33 In the case of aldehydes, which undergo side reactions in alkaline media,⁹ electrolysis was performed in neutral potassium phosphate buffer (PPB) at −1.0 V vs. RHE as the optimum condition for the production of propylene (Table S3†). To avoid interference of the parasitic hydrogen evolution reaction (HER) in assessing reactivity, we compared production rates, instead of Faradaic efficiencies, of the carbonaceous products formed under different conditions. Additional information on the experimental conditions can be found in Note S7.†

Results and discussion

Routes to main C₁ and C₂ products and precursors

The main outcome of previous computational studies on the manually-analysed network for eCO₂R to C₁ and C₂ products over Cu(111) or Cu(100) can be summarised as follows:^15–20 CO₂ adsorbs on the metal and reduces to CO, a precursor for methanol and methane (C₁). Alternatively, CO dimerises to OCCO⁻, which has been claimed to be the rate-determining step in the formation of the C₂ backbone.^15–19 On Cu(100), this process is endergonic by 0.84 eV at −0.4 V vs. RHE, (with an activation barrier of 1.42 eV), but can be promoted on defective surfaces.³⁴ Additionally, OCCO⁻ can be stabilised by electrolyte cations^35,36 and the electric field at the cathode.^28,36,37 OCCO⁻ is further protonated and then dehydrated to CCO upon a proton-coupled electron transfer step. The latter intermediate is subsequently hydrogenated to CH₂CHO.^1,38 If hydrogenolysis occurs on the remaining O atom, CH₂CH is formed, leading to ethylene. If hydrogenation occurs on the α-C (C_αH₂CHO), ethanol is formed (Fig. S4 and S5†).

Exploring C₁–C₂ coupling reactions

To unravel the most likely reaction routes towards C₃ backbone formation, a sequential analysis combining available reports, experiments and theory was applied. The possible C₁–C₂ couplings shown in Fig. 2a and Tables S16–S18† emerge from combining 10 C₁ (–CH_xO* or –CH_x*) and 70 C₂ precursors that can be hydrocarbons (CH_yCH_z*–) or oxygenates (CH_yOCH_zO*–, CH_yOCH_z*–, CH_yCH_zO*–). This gives rise to oxygenate-to-oxygenate, hydrocarbon-to-oxygenate, and hydrocarbon-to-hydrocarbon C–C coupling reactions. Thus, six families of intermediates, namely C₃H_x, 1-C₃OH_x, 1,3-C₃O₂H_x, 2-C₃OH_x, 1,2-C₃O₂H_x and 1,2,3-C₃O₃H_x, x = 0–8 (families 1–6 in Fig. 2a) are formed, where the prefix numbers show the position of the oxygenated functional group, irrespective of it having alcohol, alkoxy, aldehyde, or ketone character. Molecular fragments with carboxylate, carboxylic acid, ethers, or cyclic backbones were not considered, as these functionalities have not been found experimentally in the pool of C₃ products.

Fig. 2 Screening process to narrow down the C₁–C₂ coupling steps. a C_xH_yO_z backbones that can be obtained through C₁–C₂ couplings. Only the fully hydrogenated product is shown with the bond formed marked in orange. The numbers label six families of molecules and the colour stands for their abundance reported from CO₂ reduction experiments, found either from the literature or this work. High, low, or zero relative abundance is shown in orange, purple, and black, respectively. Backbones 3 and 5 can be formed from two combinations of C₁ and C₂ intermediates. b and c Reaction and activation energies for CH_xCH_y–CH_zO couplings between a C₁ oxygenate and a C₂ hydrocarbon, ΔE and E_a in eqn S3–S6.† Most likely steps indicated in bold. The full set of 10 C₁ × 70 C₂ coupling reactions is shown in Tables S16–S18.† Further details can be found in Notes S1–S2 and S8–S9.†

We then compared the C₃ products formed from the electrolysis of mixtures of CO with different C₂ molecules (glyoxal, ethylene glycol, oxalate, acetate, ethanol, acetaldehyde) at open-circuit potential and −0.4 and −1.0 V vs. RHE (Tables 1 and S6 and S8–S9†). At −0.4 V vs. RHE, all these mixtures generate 1-propanol at rates much larger than the reduction of CO alone (Table S4†). Mixtures with ethylene glycol gave the highest yield, while those with ethanol and acetate gave the lowest. Propylene was not detected (detection limits of gaseous products are equivalent to 0.5 μmol cm⁻² h⁻¹, see ESI† Experimental and computational details). Overall, if a set of products with a given C_xO_y backbone is not observed experimentally, then such couplings can be considered unlikely, and the routes pruned from the network.

To further verify the nature of the active C₁ fragment leading to 1-propanol, electrocatalytic tests of the C₂ compounds with either methanol or formaldehyde (Table 1) were conducted. 1-Propanol was not detected in some of these experiments, though we observed allyl alcohol and propylene from the electrolysis of a formaldehyde and acetaldehyde mixture. Moreover, the reduction of CH₂O itself produced only CH₃OH and CH₄ (Table 1 and S4†), but it does not produce C₂ and C₃ compounds as it is hardly broken into the more reactive CH₂ and CHO species (Table S22†). CH₃OH, on the other hand, was electrochemically inert (Tables S12 and S13†). The unique predominance of 1-propanol in experiments using CO indicates that *CO (or a derivative like *CHO) is instrumental in promoting 1-propanol formation. This is further confirmed by the absence of 1-propanol in experiments starting with CH₂O or CH₃OH (Table 1).

After considering the experimental input, we then switched to theory to explore the C₁–C₂ coupling reactions based on the reaction energies (ΔE, Table S16†), activation energies obtained by DFT-NEB (E_a, Table S17†), and complemented by the electrochemical driving force computed as the polarisation variation upon reaction (ΔΔQ_B in eqn S6, Table S18†). More favourable values are shown in brighter colours in Fig. 2b and c. The most likely candidates were then selected among all couplings, which reduced the set to CH₂CH–CHO and CH₃CH–CHO. In the following paragraphs, we describe how the different coupling families are retained or discarded during the analysis of the network based on abovementioned literature, experimental, and theoretical analyses:

1,2,3-C₃O₃H_x backbone. Early computational studies proposed that the C₃ backbone was formed via trimerisation of *CO^39,40 (family 6 in Fig. 2a, 1,2,3-C₃O₃H_x). We computed this potential reaction and found a relatively high activation barrier of E_a = 0.96 eV. Furthermore, the CO-trimer reverts 0.14 e⁻ to the surface, so the net reaction is therefore expected to be hindered under reductive potentials (Table S19†). Alternatively, a sequential process can be envisaged where CO dimerises to OCCO⁻, which further reacts with CHO (E_a = 0.73 eV, ΔΔQ_B = −0.81 e⁻), or COCHO with CO (E_a = 0.78 eV, ΔΔQ_B, = −0.39 e⁻) to form COCOCHO as the base of the 1,2,3-C₃O₃H_x backbone. Only the latter reaction would be promoted at more reducing potentials. However, should this reaction occur, glycerol would likely appear as a product of CO₂ reduction, but this has not been reported in the literature. Thus, the absence of glycerol as a product combined with the medium to high computed barriers and electrochemical penalties suggest that 1,2,3-C₃O₃H_x (6 in Fig. 2a) intermediates are unlikely to participate in the main mechanistic route.

1,2-C₃O₂H_x and 1,3-C₃O₂H_x backbones. There are two families of C₃O₂H_x intermediates, with O atoms in different positions: 1,2-C₃O₂H_x (5) and 1,3-C₃O₂H_x (3, Fig. 2a). Among the products derived from 1,2-C₃O₂H_x, only 1-hydroxyacetone has been reported in the literature, albeit in trace quantities.² Indeed, some CH_xCH_yO*–CH_zO* pairs have low coupling barriers, such as the coupling of CH₂CO, CH₃CO, and CH₃COH with CHO (up to 0.32 eV, Table S17†). Remarkably, the CHO coupling is expected to be strongly promoted under reductive potentials (ΔΔQ_B = −0.52 |e⁻|, Table S18†). However, these C₂ intermediates have higher potential energies than other structural isomers (Fig. S5†). As such, their concentration is expected to be too low at −0.4 V vs. RHE to form any significant amount of 1,2-C₃O₂H_x products, as confirmed by the range of products observed in our experiments (Tables S8–S13†). The formation of 1,3-C₃O₂H_x products is expected to proceed from CH₂OCH–CHO(H) (E_a ≤ 0.34 eV) or CH₂OHCH–CHO (E_a < 0.16 eV). These reactions would occur as chemical steps and are not favoured under eCO₂R conditions (according to computational charge considerations, Table S18†). Our simulation results may also explain why 1,3-C₃O₂H_x products (viz. 1,3-propanediol and 3-hydroxypropanal) have not been experimentally observed for Cu-based catalysts.

2-C₃OH_x backbone via H_xC₁–C₂H_yO coupling. We are now left with the paths that generate mono-oxygenated C₃ intermediates. 2-C₃OH_x products can be produced from the coupling of a CH_xCH_yO*– fragment, such as CH₃CHO*– and CH₂CHO*–, with a C₁ hydrocarbon, –CH_x* (family 4 in Fig. 2a, Tables S16–S18†). This pathway may expectedly yield 2-propanol, which is 0.17 eV more stable than 1-propanol (Table S1†), whereas intermediates leading to these two species show similar stabilities (Fig. S10†). However, 2-propanol was not experimentally detected (Table 1). Previous experiments on Cu-based catalysts have only detected small amounts of acetone,² in line with our present results (0.2% FE, Fig. 1, Table S2†). Acetone is likely produced by coupling CH₃CO with CH₂, (ΔE = −1.57 eV; E_a = 0.28 eV) and the further hydrogenation of the unsaturated aliphatic carbon atom.

1-C₃OH_x backbone. Most of the C₃ products detected in our experiments belong to the 1-C₃OH_x family (2 in Fig. 2a), namely 1-propanol (CH₃CH₂CH₂OH, 1-C₃OH₈), with propionaldehyde and allyl alcohol (CH₃CH₂CHO and CH₂CHCH₂OH, 1-C₃OH₆) produced at smaller rates (Table 1).^2,41 Considering the experimentally observed scarcity of C₃O₂H_x and C₃O₃H_x products, we infer that there is only one oxygen atom present during the coupling, either on the C₁ or the C₂ moiety. Reported experiments^42–45 show that during eCO₂R, the maximum production of 1-propanol occurs when high amounts of CO and C₂H₄ are formed simultaneously. Indeed, the lowest activation barriers are found for the highly exothermic CH₂CH–CHO and CH₃CH–CHO(H) couplings (ΔE < −1.50 eV, E_a ≤ 0.19 eV, Fig. 2b and c). As CH₂CH is a precursor of C₂H₄, and CHO(H) is directly formed from CO, we conclude that all such paths are highly likely. Couplings involving C₂ moieties less hydrogenated than CH₂CH or CH₂CHO (another C₂H₄ precursor) are therefore less likely. In the remaining region, most C₁–C₂ couplings are highly activated (Fig. 2c). Thus, CH₂CHCHO* and CH₃CHCHO* intermediates are common precursors for C₃ products. To a lesser degree, CH₂CHCO* can also be formed if the coupling starts with –CO instead of –CHO, Fig. 3. Finally, reactions involving C₂ oxygenated precursors (Tables S16–S18†) have higher barriers, such as the CH₂–CCH₂O coupling (while the process is exothermic, ΔE = −1.61 eV, it exhibits a non negligible activation barrier, E_a = 0.39 eV). Upon reaction, part of the electronic density of CH₂CHCHO* is returned to the surface (−0.30 e⁻, Table S18†). The reaction is therefore unfavoured at strongly reductive potentials, which explains the decrease in 1-propanol production as the potential becomes more negative.^1,2

Fig. 3 CH₂CHO–CO coupling on OD-Cu models.³⁴ a–c Initial, transition, and final state respectively. A surface cavity with high oxygen affinity assists the C–O bond breaking of the CH₂CHO* precursor. A neighbouring polarised site, Cu^δ+, weakly adsorbs CO.

The activation barriers of the transition states associated with the formation of key C₃ intermediates are sensitive to surface geometry and ensembles. Since defective Cu surfaces have been reported as selective to propanol formation,^39,44 we assessed the role of defects on OD-Cu models³⁴ for the concerted coupling of CH₂CH(O)*–CO* to yield the simplest C₃ precursor, CH₂CHCO, Fig. 3. The Cu and CuO_x structures were optimised for 10 ps through ab initio molecular dynamics and recurrent morphological motifs occured upon surface reconstruction.³⁴ Out of twelve surface motifs assessed (Fig. S11†),^34,46 an active site consisting of a surface cavity (Fig. 3a, centre) and a neighbouring Cu^δ+ atom (Fig. 3a, right) is the most suitable for promoting the coupling. While CH₂CHO* is trapped at the surface cavity, the high oxygen affinity of this site leads to the breaking of its C–O bond to give CH₂CH* (Fig. 3b). On the other hand, CO adsorption is almost thermoneutral on the polarised copper site. Thus, the CH₂CH* fragment can easily couple to the weakly bound CO* to form the C₃ backbone (exergonic by 0.13 eV; Fig. 3c). In absence of polarized Cu sites (Fig. S11†), this step is endergonic by at least 0.6 eV, thus confirming the instrumental role of surface polarization. This coupling mechanism may explain the high selectivity toward 1-propanol (FE = 23% at −0.44 V vs. RHE) achieved on highly defective Cu surfaces containing a large number of surface cavities.^11,39,44 This concept can also be extended to other key C₁–C₂ coupling reactions from moieties directly derived from CH₂CHO and CO, such as CH₂CH–CHO and CH₃CH–CHO. Going beyond pure copper catalysts, we propose that intermetallic alloys containing high oxygen affinity elements coupled with weak CO binding sites could be highly selective to C₃ as well. For instance, Cu–Ag alloys exhibited enhanced propanol selectivity depending on the silver atomic ratio,^13,47 suggesting a CO spillover mechanism⁴⁸ from Ag domains to facilitate the formation of CH₂CH–CO.

Routes to C₃ products

Routes to 1-propanol. Once the C₃ backbone is formed, the C₃ subnetwork (Fig. 4) starting from CH₂CHCHO* (orange) and CH₂CHCO* (black) can be employed to analyse selectivity trends. The colour code of the boxes in Fig. 4 represents the computed relative stability of the intermediates (thermodynamics), while the thickness of the lines linking the intermediates accounts for the barriers (thicker lines stand for faster steps). The hydrogenation of CH₂CHCO* gives CH₂CHCHO*, which then evolves via CH₃CHCHO* → propionaldehyde (CH₃CH₂CHO*) → propanoxy (CH₃CH₂CH₂O*) → 1-propanol. The existence of this path is confirmed experimentally, since the electrochemical reduction of propionaldehyde on OD-Cu yielded predominantly 1-propanol (Fig. 5c, Table S23†). Alternatively, the 1-propanol formation can proceed through CH₂CH₂CO* → CH₂CH₂CO*H → CH₃CH₂COH* → CH₃CH₂CHOH* → 1-propanol (bottom path in Fig. 4, and 5a).

Fig. 4 Computed subnetwork for CH₂CHCO and CH₂CHCHO conversion to propylene (C₃H₆) and 1-propanol (C₃H₇OH) at −0.4 V vs. RHE (full network in Fig. S12–S14†). The colours of the boxes scale with the relative DFT energy of their intermediates (eqn S3†). Relevant intermediates which can desorb and be used as probe molecules are drawn in 3D. We used allyl alcohol (C₃H₅OH) and propionaldehyde (C₃H₆O) as reactants in our experiments to confirm the pathways predicted by the network (Tables S23–S24†). The thickness of the arrows connecting the intermediates account for the activation energies (E_a, obtained by LSR. Those obtained explicitly by DFT are denoted by *).

Fig. 5 a and b Energy profiles for electrocatalytic reduction of key C₃ compounds on Cu(100), using H₂, CO₂, and H₂O as thermodynamic sinks, and shifting the energy reference to make a propionaldehyde and b allyl alcohol the zero. Corresponding products for experimental electroreduction of c propionaldehyde and d allyl alcohol at −0.4 V and −1.0 V vs. RHE on OD-Cu: propane (grey), propylene (orange), and 1-propanol (purple). Full product distributions are shown in Tables S23 and S24.† Other energy profiles at 0.0 V, −0.4 V, and −1.0 V vs. RHE are shown in Fig. S15-S17.† The (*) symbol refers to species adsorbed on the surface. Detailed DFT values can be found in Tables S20–S21.†

Routes to propylene. Mono-oxygenates can be converted to propylene via dehydration reactions starting from CH₂CHCO(H)*, CH₂CHCHO(H)*, CH₃CHCHO(H)*, CH₂CHCH₂O(H)*, and CH₃CHCH₂O(H)*, where (H) represents an optional hydrogen. The corresponding barriers of these ten reactions were computed (Table S20†). Most C–O(H) bonds are relatively difficult to activate (E_a > 1.0 eV), thus, we depict the ones showing relatively lower barriers (CH₂CHCH₂–OH* and CH₃CHCH₂–OH*, E_a = 0.17 and 0.94 eV, respectively) in Fig. 5a and b. A selectivity switch to propylene occurs when the aldehyde carbon on CH₂CHCHO* is hydrogenated to form CH₂CHCH₂O*, which in turn produces allyl alcohol (CH₂CHCH₂OH). To generate propylene, OH is eliminated from the allyl alcohol intermediate, which is then hydrogenated (Fig. 5b). However, this path is not fully selective, as allyl alcohol can also undergo hydrogenation to CH₃CHCH₂OH* to form 1-propanol. The C–OH bond breaking in allyl alcohol (CH₂CHCH₂OH) has a low barrier of 0.17 eV and it is strongly promoted by reducing potentials, with a net charge gain of 0.87 e⁻ (Table S20† and Fig. 5b).

Therefore, the production of propylene could be traced to the allyl alkoxy (CH₂CHCH₂O) intermediate, which is also a direct precursor of allyl alcohol. This proposition was verified experimentally by reducing allyl alcohol on OD-Cu (Fig. 5d), which gave noticeable amounts of propylene as theoretically predicted. Moreover, allyl alcohol (1.97 μmol cm⁻² h⁻¹) was detected alongside propylene (1.44 μmol cm⁻² h⁻¹) from the reduction of a mixture of acetaldehyde and formaldehyde (Table 1). In this combination, the most likely path occurs when acetaldehyde loses an acidic α-hydrogen (H_α–CH₂CHO)⁹ to form CH₂CHO, which dehydrates to form CH₂CH. The latter compound reacts with CH₂O to form CH₂CHCH₂O* (Fig. 4 and S12–S14†) which is mainly selective towards allyl alcohol and propylene, but not 1-propanol (Table 1). Interestingly, we note that CO₂ reduction produced 1-propanol (FE = 4.4%) and allyl alcohol (FE = 1.2%) (Fig. 1), while propylene was absent. This can be rationalised by a mild eCO₂R interface alkalinisation, which occurs under reaction conditions,⁴⁵ favouring the desorption of allyl alkoxy (protonated in solution into allyl alcohol) and thus preventing propylene synthesis. Overall, these observations strongly suggest the key role of allyl alcohol in the route to propylene.

From a broader perspective, the low activity of Cu catalysts for eCO₂R to C₃ compounds, particularly propylene, could be improved through engineering at different scales. Currently, the most explored approaches to promote multi-carbon products include engineering catalyst surfaces with a high density of defects to improve activity, and optimizing the electrolyte and reactor conditions to alleviate mass transport limitations and tuning the environment at the electrode-electrolyte interface. Modifications at the process level could benefit from three different approaches: (i) one single reactor recycling C₂ (or C₁) eCO₂R products to ensure a high concentration of active intermediates; (ii) independently optimised catalysts and reactors to produce C₁ (Cu or Ag-based catalysts) and C₂ (on an oxide-derived Cu catalyst) intermediates, which mix in a third reactor dedicated to the coupling to form the C₃ backbone or alloys containing close domains of both; (iii) a process able to produce the relevant intermediate allyl alcohol (for which an effective catalyst is not yet known) which is then converted to propylene in a second unit. Although there are some experimental indications in the literature of the potential of strategies (i) and (ii),¹³ the detailed understanding of the reaction network and the elucidation of key intermediates presented in this work ultimately serve to direct future works toward realizing these solutions.

Conclusions

In conclusion, we have performed an integrated mechanistic analysis of the eCO₂R to C₃ products with all the individual steps available in our open database.⁴⁹ Methodological implementations including structural graph network generation, fast energy screenings, and network pruning of irrelevant paths through experimental input allow the effective sampling of the complex C₃ network pointing out the difficulties found when only a part of the reaction network is sampled.⁵⁰ C₂ and C₃ products were found to share a common precursor, CH₂CHO*. Our findings rationalise the generally observed low selectivity of eCO₂R toward C₃ products, as well as their enhancement on nanostructured Cu catalysts: (i) C₃ backbones are formed via the sluggish coupling of CO or CHO with CH₂CH*, preferentially at defects and (ii) all C₃ precursors end up containing at least one O atom, i.e., CH_x + C₂H_y couplings are highly unlikely. The most stable mono-oxygenated intermediate CH₂CHCHO* gives access to propylene, propionaldehyde, and 1-propanol. The inaccessible allyl alkoxy intermediate is identified as the most likely kinetic trap preventing propylene production as indicated by simulations and further reinforced with the electrolysis of allyl alcohol leading to propylene. Our mechanistic understanding paves the way towards the development of advanced electrocatalysts that promote C₃ products, particularly alkenes.

Author contributions

S. P.-G., Software, methodology, data curation, investigation, visualisation, writing – original draft. F. L. P. V., L. R. L. T., Methodology, data curation, investigation, visualisation, writing – original draft. R. G.-M., A. J. M., methodology, data curation, investigation, visualisation, validation, supervision, writing – original draft. F. D., Data curation, investigation, visualisation, writing – review & editing. B. S. Y., J. P.-R., N. L., Conceptualisation, funding acquisition, supervision, project administration, writing – review & editing.

Conflicts of interest

There are no conflicts of interest to declare.

Acknowledgements

This work was financed by the Spanish Ministry of Science and Innovation (RTI2018-101394-B-I00, Severo Ochoa Centre of Excellence CEX2019-000925-S 10.13039/501100011033), ETH Research Grant (ETH-47 19-1), the National University of Singapore Flagship Green Energy Program (R143-000-A64-114, R143-000-A55-733 and R143-000-A55-646), and Ministry of Education of Singapore (R143-000-B52-114). This publication was created as part of NCCR Catalysis, a National Centre of Competence in Research funded by the Swiss National Science Foundation. The Barcelona Supercomputing Centre – MareNostrum (BSC-RES) is acknowledged for providing generous computational resources. The authors thank Mavis Pei Lin Kang from NUS for assisting with XRD analyses.

Notes and references

1. S. Nitopi, E. Bertheussen, S. B. Scott, X. Liu, A. K. Engstfeld, S. Horch, B. Seger, I. E. L. Stephens, K. Chan, C. Hahn, J. K. Nørskov, T. F. Jaramillo and I. Chorkendorff, Chem. Rev., 2019, 119, 7610–7672.
2. K. P. Kuhl, E. R. Cave, D. N. Abram and T. F. Jaramillo, Energy Environ. Sci., 2012, 5, 7050.
3. B. Schmid, C. Reller, S. Neubauer, M. Fleischer, R. Dorta and G. Schmid, Catalysts, 2017, 7, 161.
4. Y. Kwon, Y. Lum, E. L. Clark, J. W. Ager and A. T. Bell, ChemElectroChem, 2016, 3, 1012–1019.
5. X. Wang, A. Xu, F. Li, S. F. Hung, D. H. Nam, C. M. Gabardo, Z. Wang, Y. Xu, A. Ozden, A. S. Rasouli, A. H. Ip, D. Sinton and E. H. Sargent, J. Am. Chem. Soc., 2020, 142, 3525–3531.
6. Y. Y. Birdja, E. Pérez-Gallent, M. C. Figueiredo, A. J. Göttle, F. Calle-Vallejo and M. T. M. Koper, Nat. Energy, 2019, 4, 732–745.
7. J. Li, Z. Wang, C. McCallum, Y. Xu, F. Li, Y. Wang, C. M. Gabardo, C. T. Dinh, T. T. Zhuang, L. Wang, J. Y. Howe, Y. Ren, E. H. Sargent and D. Sinton, Nat. Catal., 2019, 2, 1124–1131.
8. H. Xu, D. Rebollar, H. He, L. Chong, Y. Liu, C. Liu, C. J. Sun, T. Li, J. V. Muntean, R. E. Winans, D. J. Liu and T. Xu, Nat. Energy, 2020, 5, 623–632.
9. L. R. L. Ting, R. García-Muelas, A. J. Martín, F. L. P. Veenstra, S. T.-J. Chen, Y. Peng, E. Y. X. Per, S. Pablo-García, N. López, J. Pérez-Ramírez and B. S. Yeo, Angew. Chem., Int. Ed., 2020, 59, 21072–21079.
10. D. Gao, I. Sinev, F. Scholten, R. M. Arán-Ais, N. J. Divins, K. Kvashnina, J. Timoshenko and B. Roldan Cuenya, Angew. Chem., Int. Ed., 2019, 58, 17047–17053.
11. J. Li, F. Che, Y. Pang, C. Zou, J. Y. Howe, T. Burdyny, J. P. Edwards, Y. Wang, F. Li, Z. Wang, P. De Luna, C. T. Dinh, T. T. Zhuang, M. I. Saidaminov, S. Cheng, T. Wu, Y. Z. Finfrock, L. Ma, S. H. Hsieh, Y. S. Liu, G. A. Botton, W. F. Pong, X. Du, J. Guo, T. K. Sham, E. H. Sargent and D. Sinton, Nat. Commun., 2018, 9, 4614.
12. L. Mandal, K. R. Yang, M. R. Motapothula, D. Ren, P. Lobaccaro, A. Patra, M. Sherburne, V. S. Batista, B. S. Yeo, J. W. Ager, J. Martin and T. Venkatesan, ACS Appl. Mater. Interfaces, 2018, 10, 8574–8584.
13. X. Wang, Z. Wang, T. T. Zhuang, C. T. Dinh, J. Li, D. H. Nam, F. Li, C. W. Huang, C. S. Tan, Z. Chen, M. Chi, C. M. Gabardo, A. Seifitokaldani, P. Todorović, A. Proppe, Y. Pang, A. R. Kirmani, Y. Wang, A. H. Ip, L. J. Richter, B. Scheffel, A. Xu, S. C. Lo, S. O. Kelley, D. Sinton and E. H. Sargent, Nat. Commun., 2019, 10, 1–7.
14. D. Ren, J. Fong and B. S. Yeo, Nat. Commun., 2018, 9, 925.
15. A. J. Garza, A. T. Bell and M. Head-Gordon, ACS Catal., 2018, 8, 1490–1499.
16. J. D. Goodpaster, A. T. Bell and M. Head-Gordon, J. Phys. Chem. Lett., 2016, 7, 1471–1477.
17. Y. Huang, Y. Chen, T. Cheng, L.-W. Wang and W. A. Goddard, ACS Energy Lett., 2018, 3, 2983–2988.
18. T. Cheng, H. Xiao and W. A. Goddard, Proc. Natl. Acad. Sci. U. S. A., 2017, 114, 1795–1800.
19. Y. Zheng, A. Vasileff, X. Zhou, Y. Jiao, M. Jaroniec and S.-Z. Qiao, J. Am. Chem. Soc., 2019, 141, 7646–7659.
20. Q. Li, R. García-Muelas and N. López, Nat. Commun., 2018, 9, 526.
21. J. K. Nørskov, J. Rossmeisl, A. Logadottir, L. Lindqvist, J. R. Kitchin, T. Bligaard and H. Jónsson, J. Phys. Chem. B, 2004, 108, 17886–17892.
22. A. A. Peterson, F. Abild-Pedersen, F. Studt, J. Rossmeisl and J. K. Nørskov, Energy Environ. Sci., 2010, 3, 1311–1315.
23. J. P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. Lett., 1996, 77, 3865–3868.
24. S. Grimme, J. Comput. Chem., 2006, 27, 1787–1799.
25. N. Almora-Barrios, G. Carchini, P. Błoński and N. López, J. Chem. Theory Comput., 2014, 10, 5002–5009.
26. M. Garcia-Ratés and N. López, J. Chem. Theory Comput., 2016, 12, 1331–1341.
27. M. Garcia-Ratés, R. García-Muelas and N. López, J. Phys. Chem. C, 2017, 121, 13803–13809.
28. R. B. Sandberg, J. H. Montoya, K. Chan and J. K. Nørskov, Surf. Sci., 2016, 654, 56–62.
29. G. Henkelman and H. Jónsson, J. Chem. Phys., 2000, 113, 9978–9985.
30. M. García-Mota, B. Bridier, J. Pérez-Ramírez and N. López, J. Catal., 2010, 273, 92–102.
31. S. Pablo-García, M. Álvarez-Moreno and N. López, Int. J. Quantum Chem., 2021, 121, e26382.
32. C. T. Dinh, T. Burdyny, G. Kibria, A. Seifitokaldani, C. M. Gabardo, F. Pelayo García De Arquer, A. Kiani, J. P. Edwards, P. De Luna, O. S. Bushuyev, C. Zou, R. Quintero-Bermudez, Y. Pang, D. Sinton and E. H. Sargent, Science, 2018, 360, 783–787.
33. J.-J. Lv, M. Jouny, W. Luc, W. Zhu, J.-J. Zhu and F. Jiao, Adv. Mater., 2018, 30, 1803111.
34. F. Dattila, R. García-Muelas and N. López, ACS Energy Lett., 2020, 5, 3176–3184.
35. E. Pérez-Gallent, G. Marcandalli, M. C. Figueiredo, F. Calle-Vallejo and M. T. M. Koper, J. Am. Chem. Soc., 2017, 139, 16412–16419.
36. J. Resasco, L. D. Chen, E. Clark, C. Tsai, C. Hahn, T. F. Jaramillo, K. Chan and A. T. Bell, J. Am. Chem. Soc., 2017, 139, 11277–11287.
37. K. Jiang, R. B. Sandberg, A. J. Akey, X. Liu, D. C. Bell, J. K. Nørskov, K. Chan and H. Wang, Nat. Catal., 2018, 1, 111–119.
38. R. Kortlever, J. Shen, K. J. P. Schouten, F. Calle-Vallejo and M. T. M. Koper, J. Phys. Chem. Lett., 2015, 6, 4073–4082.
39. Y. Pang, J. Li, Z. Wang, C. S. Tan, P. L. Hsieh, T. T. Zhuang, Z. Q. Liang, C. Zou, X. Wang, P. De Luna, J. P. Edwards, Y. Xu, F. Li, C. T. Dinh, M. Zhong, Y. Lou, D. Wu, L. J. Chen, E. H. Sargent and D. Sinton, Nat. Catal., 2019, 2, 251–258.
40. H. Xiao, T. Cheng and W. A. Goddard, J. Am. Chem. Soc., 2017, 139, 130–136.
41. Y. Hori, A. Murata and R. Takahashi, J. Chem. Soc., Faraday Trans., 1989, 85, 2309–2326.
42. Y. Hori, I. Takahashi, O. Koga and N. Hoshi, J. Phys. Chem. B, 2002, 106, 15–17.
43. D. Ren, N. T. Wong, A. D. Handoko, Y. Huang and B. S. Yeo, J. Phys. Chem. Lett., 2016, 7, 20–24.
44. T. Zhuang, Y. Pang, Z. Liang, Z. Wang, Y. Li, C. Tan, J. Li, C. T. Dinh, P. De Luna, P. L. Hsieh, T. Burdyny, H. H. Li, M. Liu, Y. Wang, F. Li, A. Proppe, A. Johnston, D. H. Nam, Z. Y. Wu, Y. R. Zheng, A. H. Ip, H. Tan, L. J. Chen, S. H. Yu, S. O. Kelley, D. Sinton and E. H. Sargent, Nat. Catal., 2018, 1, 946–951.
45. F. L. P. Veenstra, N. Ackerl, A. J. Martín and J. Pérez-Ramírez, Chem, 2020, 1–16.
46. F. Dattila, R. García-Muelas and N. López, Dataset associated to: Active and Selective Ensembles in Oxide- Derived Copper Catalysts for CO₂ Reduction, ioChem-BD, 2020,  DOI:10.19061/iochem-bd-1-165.
47. A. Herzog, A. Bergmann, H. S. Jeon, J. Timoshenko, S. Kühl, C. Rettenmaier, M. Luna Lopez, F. T. Haase and B. Roldan Cuenya, Angew. Chem., Int. Ed., 2021, 60, 7426–7435.
48. P. Iyengar, M. J. Kolb, J. R. Pankhurst, F. Calle-Vallejo and R. Buonsanti, ACS Catal., 2021, 11, 4456–4463.
49. R. García-Muelas, S. Pablo-García and N. López, Dataset associated to: Mechanistic Routes toward C₃ products in Copper-Catalyzed CO₂ Electroreduction, 2020,  DOI:10.19061/iochem-bd-1-175.
50. M. T. Tang, H.-J. Peng, J. H. Stenlid and F. Abild-Pedersen, J. Phys. Chem. C, 2021, 125, 26437–26447.

---

Footnotes

† Electronic supplementary information (ESI) available: Experimental and computational details, supplementary notes, figures, and tables. See DOI: 10.1039/d1cy01423d

‡ These authors contributed equally to this work.

---