Electrochemical synthesis of propylene from carbon dioxide on copper nanocrystals

The conversion of carbon dioxide to value-added products using renewable electricity would potentially help to address current climate concerns. The electrochemical reduction of carbon dioxide to propylene, a critical feedstock, requires multiple C–C coupling steps with the transfer of 18 electrons per propylene molecule, and hence is kinetically sluggish. Here we present the electrosynthesis of propylene from carbon dioxide on copper nanocrystals with a peak geometric current density of −5.5 mA cm−2. The metallic copper nanocrystals formed from CuCl precursor present preponderant Cu(100) and Cu(111) facets, likely to favour the adsorption of key *C1 and *C2 intermediates. Strikingly, the production rate of propylene drops substantially when carbon monoxide is used as the reactant. From the electrochemical reduction of isotope-labelled carbon dioxide mixed with carbon monoxide, we infer that the key step for propylene formation is probably the coupling between adsorbed/molecular carbon dioxide or carboxyl with the *C2 intermediates that are involved in the ethylene pathway.

The conversion of carbon dioxide to value-added products using renewable electricity would potentially help to address current climate concerns. The electrochemical reduction of carbon dioxide to propylene, a critical feedstock, requires multiple C-C coupling steps with the transfer of 18 electrons per propylene molecule, and hence is kinetically sluggish. Here we present the electrosynthesis of propylene from carbon dioxide on copper nanocrystals with a peak geometric current density of −5.5 mA cm −2 . The metallic copper nanocrystals formed from CuCl precursor present preponderant Cu(100) and Cu(111) facets, likely to favour the adsorption of key *C 1 and *C 2 intermediates. Strikingly, the production rate of propylene drops substantially when carbon monoxide is used as the reactant. From the electrochemical reduction of isotope-labelled carbon dioxide mixed with carbon monoxide, we infer that the key step for propylene formation is probably the coupling between adsorbed/molecular carbon dioxide or carboxyl with the *C 2 intermediates that are involved in the ethylene pathway.
The electrochemical conversion of carbon dioxide (CO 2 ) to value-added products driven by renewable electricity serves as a promising strategy for alleviating the negative impact of excessive anthropogenic carbon emissions 1,2 . With copper-based catalysts, electrochemical CO 2 reduction has shown an appreciable activity for the production of multiple C 1 (refs. 3-6) and C 2 chemicals [7][8][9] . Although C 3+ terminal oxygenates such as n-propanol and n-butanol could be produced from CO 2 reduction 10-12 , C 3+ hydrocarbons, such as propylene (CH 2 =CH-CH 3 ), have rarely been observed as products. Propylene, a critical chemical feedstock, has reached an annual global capacity of 130 Mt in 2019, requiring an input of energy equivalent to the one from about 190 million barrels of crude oil and entailing around 80 Mt of CO 2 emission (https://cen.acs.org/ energy/Periodic-Graphics-Environmental-impact-industrial/97/i24). The electrosynthesis of propylene from CO 2 , yielding a negative carbon footprint, is an attractive strategy for producing this indispensable feedstock for the polymer industry and is yet to be achieved.
The electroreduction of CO 2 to propylene involves the transfer of 18 electrons per propylene molecule and requires multiple C-C coupling steps 13 , posing kinetic and thermodynamic barriers for driving this reaction: where E° is the thermodynamic equilibrium potential and RHE represents the reversible hydrogen electrode. All potentials cited in this work are scaled against RHE unless otherwise stated. Lee et al. observed propylene formation from electrochemical CO 2 reduction over chloride-induced biphasic Cu 2 O-Cu catalysts with a low production rate of 72 μA cm −2 and a Faradaic efficiency (FE) of 0.9% at −1.8 V versus RHE 14 . Recently, Pablo-García et al. proposed that the production of propylene could be traced to the allyl alkoxy (CH 2 =CHCH 2 O) intermediate, easy desorption of which in an alkaline microenvironment results in the unfavourable production of propylene 15 . This conclusion helps to explain why propylene is rarely being produced/detected in CO 2 reduction, which contrasts with the production of ethylene [16][17][18] . In-depth understanding of the reaction pathway Article https://doi.org/10.1038/s41557-023-01163-8 a potential range from −0.475 to −0.675 V, with the highest total FE of 6.2% at −0.50 V (Fig. 2a) and the maximum cathodic current density of 21.4 mA cm −2 at −0.675 V (Fig. 2b). As a comparison, sputtered polycrystalline copper films that consist of 50-80 nm particles deliver a poorer performance towards propylene formation with an optimized FE of 1.1% peaking at −0.625 V and a partial current density of −2.89 mA cm −2 at −0.675 V (Fig. 2c, Supplementary Fig. 4 and Tables 3 and 4).
The surface active sites of CuNC and polycrystalline copper catalysts were further assessed by lead underpotential deposition (UPD) and hydroxide (OH − ) adsorption to reveal the difference in their catalytic activity. Two cathodic peaks at approximately −0.02 and −0.06 V versus RHE, which could be respectively assigned to the UPD of lead on Cu(111) and Cu(100) facets, are observed ( Supplementary Fig. 5) [25][26][27] . The area ratios of Cu(111) and Cu(100) were calculated to be 1.3 for CuNCs and 2.7 for polycrystalline copper, respectively (Supplementary Table 5). On the other hand, OH − adsorption peaks on Cu(100) and Cu(111) are present at ~0.36 and ~0.46 V versus RHE 28,29 , respectively ( Supplementary Fig. 6). The area ratios of Cu(111):Cu(100) determined from OH − adsorption for both CuNC and polycrystalline copper catalysts are consistent with the ones calculated from the lead UPD (Supplementary Table 6). Moreover, OH − adsorption on both catalysts after 10 min of electrolysis at −0.60 V shows that the distribution of facet orientations remains almost unchanged (Supplementary Fig. 7 and Supplementary Table 6). Note that noticeable background currents were observed in the above cyclic voltammograms due to the unavoidable oxygen and carbon component within the porous GDL substrate. Switching the substrate to a non-porous one, such as glass, renders a flat baseline and shows consistency with the above analysis (Supplementary Fig. 8 and Supplementary Table 5). Nevertheless, we selected to perform cyclic voltammetry with the GDL as the substrate to assess the real active sites in the flow cell.
We then normalized the partial current of propylene against the electrochemical surface area of two catalysts (Supplementary Table 5). Interestingly, CuNCs rival polycrystalline copper in terms of the intrinsic activity towards propylene formation, as evidenced by j specific and turnover frequency (TOF) (see calculation in Methods and Supplementary Figs. 9 and 10). The kinetics based on the Tafel analysis of propylene also reveal that CuNCs outperform polycrystalline copper with a smaller Tafel slope of 75 mV dec −1 (Fig. 2d). The enhanced intrinsic activity of CuNCs may be due to their in situ reconstruction, resulting in an appropriate distribution of Cu(100) and Cu(111) facets on a rough surface, as indicated by the lead UPD and OH − adsorption measurements (Supplementary Figs. 8 and 9). Previous studies have proposed that a mixture of different copper facets is active for catalysing CO 2 conversion to multi-carbon products 25,28 . The propagation of carbon chains is thermodynamically and kinetically favoured on Cu(100) 30,31 , while coexistence of Cu(111) could provide a conjoined interface for stabilizing key intermediates for multi-carbon products 22 .
Besides C 3 products, C 2 hydrocarbons and oxygenates are also produced with remarkable selectivity and reaction rate. The FE of C 2 products increases from 32.1% to 66.0% once the applied potential shifts from −0.475 to −0.550 V and remains at around 60% from −0.550 V to −0.675 V (Extended Data Fig. 4). Meanwhile, the FE of methane is suppressed to <0.4% regardless of the applied potential (Supplementary Fig. 11). At −0.55 V, the FE ratio between C 2+ /CH 4 reaches a value of up to 1,200, showing the excellent selectivity of CuNCs in catalysing C-C coupling. The partial current density for C 2 products peaks at −335.5 mA cm −2 ( Supplementary Fig. 12).
The addition of Cs + , although it does not induce any notable morphological difference in the CuNCs ( Supplementary Fig. 13), improves both the activity of CO 2 reduction and hydrogen evolution, regardless of the anion component of the additives (Supplementary Fig. 14). First, larger metal cations such as Cs + with a softer hydration shell have a higher concentration near the surface of the catalyst and deliver a favourable coordination with negatively charged intermediates, that towards propylene formation is warranted for designing catalysts for this reaction.
Here we synthesize copper nanocrystals (CuNCs), the surfaces of which predominantly consist of Cu(100) and Cu(111) facets, enabling the electrosynthesis of propylene from CO 2 reduction with appreciable selectivity and production rate. By carrying out well-designed control experiments, including the reductions of CO, CO 2 /CO, CO 2 /He and 13 CO 2 /CO mixtures, we propose that propylene generation shares a highly protonated *C 2 intermediate with ethylene generation, and *CO is unlikely to be the *C 1 intermediate that couples with *C 2 species for propylene formation. This contrasts with the n-propanol pathway where *CO is proposed to be the key *C 1 precursor participating in the *C 1 -*C 2 coupling.

Characterization of electrocatalyst
A CuCl layer, resulting from an electrochemical roughening of a copper film on a gas diffusion layer (GDL, Supplementary Fig. 1) 19 , was electrochemically pre-reduced to form copper NCs. X-ray diffraction and high-resolution X-ray photoelectron spectroscopy (XPS) confirm the presence of CuCl after electrochemical roughening (Fig. 1a,b and Extended Data Fig. 1) 20 . The surface Cu 2+ species shown by XPS may result from the oxidation of Cu 0 /Cu + once the sample is exposed to air and moisture (Fig. 1b). CuCl exhibits aggregated cuboids ∼500 nm in size (Fig. 1c). After pre-reduction in the CsI-containing KOH electrolyte, the film shows metallic copper features with dominant Cu(100) and Cu(111) facets (Fig. 1a) and the disappearance of chloride from the XPS spectrum demonstrates the effective removal of residual chloride via pre-reduction ( Fig. 1b and Extended Data Fig. 1). The surface after pre-reduction shows densely arranged copper nanoparticles and nanocubes 30-80 nm in size (Fig. 1d). We note here that CsI has little effect on forming nanocubes during pre-reduction ( Supplementary Fig. 2). However, keeping electrolytes for pre-reduction and subsequent CO 2 reduction identical is essential for avoiding any possible reconstruction of copper, indirectly caused by air exposure during electrolyte replacement ( Supplementary Fig. 2).

Electrosynthesis of propylene from CO 2 on CuNC catalyst
In an electrochemical flow cell (Extended Data Fig. 2), CuNCs catalyse the conversion of CO 2 with high reaction rates at moderate potentials in aqueous 1 M KOH with additional 0.2 M CsI (refs. 23,24) (Supplementary Fig. 3). For example, the total geometric current density reaches about −0.6 A cm −2 with a partial current density towards CO 2 reduction of about −0.4 A cm −2 at −0.675 V.
The products detected on our CuNC catalyst include C 1 -C 3 hydrocarbons/oxygenates and hydrogen (Extended Data Fig. 3 and Supplementary Tables 1 and 2). Strikingly, the formation of propylene emerges at a potential of only −0.475 V (Fig. 2a), corresponding to an overpotential of ∼600 mV. If a more negative bias is applied, the FE of propylene increases and reaches a maximum value of 1.4% at −0.60 V, which is 1.6-fold higher than the one reported by Lee et al. on Cu/Cu 2 O biphasic catalysts in a H-type cell 14 . A cathodic current density of 5.46 mA cm −2 for propylene production is achieved at −0.65 V (Fig. 2b), delivering an improvement factor of 65 as compared with the previously reported value 14 . The formation of C 3 chemicals, including propylene, n-propanol and allyl alcohol, is observed on CuNCs over Article https://doi.org/10.1038/s41557-023-01163-8 is, CO 2 − , thus promoting the CO 2 reduction rate 32 . Second, hydration of Cs + ions could induce a lower local pH, which improves the activity of CO 2 reduction by dissolving more CO 2 molecules, and also promotes the kinetics of hydrogen evolution 33,34 . Moreover, change in the local electric field introduced by hydrated Cs + may also help improve both the CO 2 reduction rate and hydrogen evolution rate 24,35 . The complexity of the Cs + effects leads to the observed trend: with the addition of 0.2 M CsOH into 1 M KOH electrolyte, selectivity of C 2+ products is enhanced at potentials greater than −0.6 V versus RHE and hydrogen selectivity is improved at all the tested potentials ( Supplementary Fig. 14). It is also noted that the catalyst requires 100 mV smaller overpotential to achieve the optimum formation of C 2+ products in the presence of Cs + , consistent with our recent finding that current density is a critical factor determining C-C coupling activity 36 . On the other hand, the I − anions also could improve the geometric current density at potentials less than −0.60 V if we compared the activity measured in CsOH + KOH electrolyte and CsI + KOH ( Supplementary Fig. 15) 37 . Ogura et al. proposed that the specifically adsorbed halides facilitate the electron flow from the electrode surface to the vacant orbital of CO 2 (ref. 38). Akhade et al. reported that a small quantity of KI improves current density at more negative potentials on a copper electrode by enhancing the reaction energetics of *CO coupling due to the presence of I − ions in the electrochemical double layer 39 . This is consistent with the increased CO 2 reduction current observed on our CuNCs at potentials less than −0.6 V ( Supplementary  Fig. 15). We also observed that the addition of I − suppresses methane formation, different from the results shown in the study by Strasser and co-workers in which an enhanced methane formation was reported 40 . This difference is probably due to the higher local pH in our alkaline reaction system that facilitates the generation of hydrocarbon product.
In summary, the distribution of cationic and anionic species in the local microenvironment could induce complex impacts, such as stabilization of intermediate, specific adsorption on electrode and repelsion of reacting species, which could impair or favour CO 2 reduction. In our system, the presence of Cs + and I − in the electrolyte was found to increase the CO 2 electroreduction rate on CuNC catalysts without sacrificing their intrinsic selectivity towards C 2+ products at potentials greater than −0.60 V versus RHE.
The stability of our CuNCs was evaluated by 16 h electrolysis at −273.7 mA cm −2 , corresponding to a potential of −0.60 V versus RHE ( Fig. 2e and Supplementary Fig. 16). With the periodic removal of precipitated salt ( Supplementary Fig. 17) 41,42 , CuNCs show durable performance towards propylene formation with the FE and partial current density remaining at 95% of the initial values after 16 h (Fig. 2e). However, we observed a reconstruction of the catalyst from nanocubes to nanospheres that consist of agglomerates with particle sizes of 10-30 nm (Extended Data Fig. 5). This could be due to the high surface free energy of nanocubes which induces aggregation after long-term electrolysis 43 , although the possibility of this morphology change originating from an unavoidable oxidation during cell disassemby could not be ruled out. We further performed OH − adsorption on the CuNC catalyst immediately after 16 h CO 2 reduction ( Supplementary Fig. 18). The OH − adsorption peaks on both Cu(100) and Cu(111) facets are still present at ~0.37 and ~0.48 V versus RHE, respectively. Further analysis of their charges evidenced a ratio of 1.05 between the active area of Cu(111) and Cu(100) (Supplementary Table 7), implying that the feature of nearly equal distribution of Cu(100) and Cu(111) facets remains almost unchanged despite the huge morphology reconstruction.

Identification of the intermediates for propylene production
The substantial production rate of propylene achieved on our CuNCs makes this material an excellent model catalyst for further mechanistic analysis 15 . With the general belief that the formation of C 3 compounds involves a key step of coupling between *C 1 and *C 2 species 44 , we first elucidate the structure of *C 2 species by a, X-ray diffractograms of CuCl film, pre-reduced CuNCs and GDL substrate. θ denotes the angle of X-ray incidence. b, High-resolution XPS spectra of Cu 2p and Cl 2p of CuCl and pre-reduced CuNCs. c-e, Scanning electron micrographs of CuCl film formed by roughening of sputtered copper film (c), CuNCs after pre-reduction (d) and CuNCs after CO 2 reduction at −0.60 V versus RHE for 10 min (e). f, High-resolution transmission electron micrographs of CuNC catalysts. g-j, Lattice fringes of Cu(100) facet (g,h) and Cu(111) facet (i,j). d in (g-j) denotes the interplanar spacing of each facet. Scale bars: c, 500 nm; d,e, 100 nm; f, 10 nm; g-j, 5 nm.
Article https://doi.org/10.1038/s41557-023-01163-8 analysing the linear correlation between the partial current density (j) of propylene and the j of a specific C 2 product (a statistical analysis is shown in Extended Data Fig. 6). The linearity between j propylene and j C2 , as indicated by the R 2 values of the fitting curves, becomes poorer if the C 2 product contains more oxygen atoms Fig. 3a,b and Supplementary Fig. 19). This indicates that *C 2 intermediates involved in *C 1 -*C 2 coupling for propylene production are probably highly protonated ones, such as *OCH=CH 2 (ref. 45). We also introduced CO into the feed gas to form mixtures of CO/CO 2 to tune the production rate of the products of interest (Fig. 3c- e and Supplementary  Tables 8-10) 46,47 . Strikingly, the absolute production rate of propylene also follows the same trend as that of ethylene (Fig. 3c). For example, with 20% of CO 2 being substituted by CO, the production rate of ethylene is enhanced by 14% compared with the one using 100% CO 2 . Simultaneously, the production rate of propylene increases by 25%.
Surprisingly, only a trace amount of propylene is detected from the reduction of 100% CO with a cathodic partial current density of <0.10 mA cm −2 and a production rate of <90 pmol s −1 cm −2 at potentials from −0.50 to −0.75 V versus RHE (Supplementary Tables 9 and 10) although CuNCs are still capable of producing sufficient *C 2 intermediates as indicated by the FE of ethylene (Fig. 3d and Supplementary  Figs. 20 and 21). This interesting observation highlights that the active *C 1 intermediates for *C 1 -*C 2 coupling towards propylene formation might be missing in CO reduction. First, although formate is not a product of CO reduction 45 , the possibility of *OCO − as the *C 1 intermediate is disproved by co-reduction of CO and HCOO − which shows a low rate of <35 pmol s −1 cm −2 for propylene production, similar to the value observed in CO reduction (Extended Data Fig. 7). Secondly, all *C 1 intermediates involved in the pathway of CO → CH 4 conversion are unlikely to be involved in *C 1 -*C 2 coupling for propylene production because the reduction of CO leads to the appreciable formation of methane (Supplementary Table 9). Moreover, the possibility of formaldehyde being the key *C 1 intermediate, as proposed by a recent mechanistic study on the routes towards C 3 products 15 , is also excluded by carrying out CO reduction using formaldehyde-containing electrolytes (Extended Data Fig. 7). Thus, the key *C 1 fragments involved in propylene pathway are likely to be two species, that is, molecular/adsorbed CO 2 or *COOH. This is strongly corroborated by the observation that the production rate of propylene increases by a stunning factor of 14, from 86 pmol s −1 cm −2 in CO reduction to 1.24 × 10 3 pmol s −1 cm −2 in the reduction of a mixture comprising 80% CO and 20% CO 2 (Fig. 3e). Note that the reduction of an 80% He + 20% CO 2 mixture leads to a production rate of 200 pmol s −1 cm −2 , corresponding to only 1/6 of the value observed in the reduction of 80% CO + 20% CO 2 mixtures (Fig. 3e). Hence, *C 1 intermediates resulting from CO 2 or reduction of *CO 2 couple with the *C 2 intermediates stemming primarily from CO reduction, leading to the production of propylene with substantially improved rate in the reduction of 80% CO + 20% CO 2 .
In contrast, the pathway towards the formation of n-propanol is slightly different. It is proposed that *CO is the key *C 1 species for n-propanol production 10,22,48 . On the basis of our observation that the maximum production of n-propanol occurs if large amounts of CO and C 2 H 4 are simultaneously formed ( Supplementary Tables 1 and 3), we propose that the coupling of *CH 2 CH/*CH 3 Table 8) because the *CO intermediate could be either due to the direct adsorption of feed CO or from the reduction of CO 2 . Because the local pH becomes higher and the number of protons decreases with the incorporation of CO into the reactant stream due to the fact that CO does not react with OH − like CO 2 , the production rate of n-propanol declines slightly once the percentage of CO increases in the CO/CO 2 mixture ( Supplementary Fig. 22). For allyl alcohol, the change of production rate with different reactant mixtures is difficult to ascertain due to the low levels of allyl alcohol produced, approaching the detection limit of 1 H NMR ( Supplementary Fig. 22).

Quantitative gas chromatography-mass spectrometry analysis for 13 CO 2 / 12 CO reduction
We further combined isotopic labelling experiments with gas chromatography-mass spectrometry (GC-MS) to gain more insights into the *C 2 and *C 1 intermediates for C-C coupling to propylene production. Standard gas of ethylene and propylene show consistent mass-to-charge signals (m/z) compared to the standard mass spectra of two chemicals (Fig. 4a,b and Supplementary Fig. 23). Ethylene and propylene produced from 12 CO 2 reduction show the same ionized molecules and fragments as compared to their respective standard gas (Fig. 4c,d), with the two highest peaks at m/z = 28 and m/z = 41 representing 12 C 2 H 4 + and 12 C 3 H 5 + , respectively. In the reduction of 13 CO 2 , the m/z ratios of these two highest peaks increase by 2 (m/z = 30, 13 C 2 H 4 + ) and 3 (m/z = 44, 13 C 3 H 5 + ) for ethylene and propylene, respectively (Fig. 4e,f), indicating the production of 13 C 2 H 4 and 13 C 3 H 6 .
If the ratio of 13 CO 2 / 12 CO is 80%/20%, the highest peak of ethylene appears at m/z = 30, representing the dominant presence of 13 C 2 H 4 . The peaks at m/z = 29 and 28 exhibit slightly higher intensity compared to the standard 13 C 2 H 4 spectrum (Fig. 4g), indicating the additional formation of 13 CH 2 12 CH 2 (from the 13 CO 2 -12 CO pathway) and 12 C 2 H 4 (from the 12 CO-12 CO pathway), with a low percentage of 12 C 2 H 4 . The MS signals of propylene also show a wide range of m/z ratio from 40 to 45, with the main fragment peaks locating at 44 and 45 (Fig. 4h), indicating the formation of 13 C 3 H 6 .
The mass spectrum of ethylene detected from the reduction of the mixture of 13 CO 2 / 12 CO = 20%/80% shows the highest peak at m/z = 28 and two smaller peaks at m/z = 29 and 30 (Fig. 4i), indicating that the majority of the formed ethylene has two 12 C atoms. In comparison, the mass spectrum of propylene exhibits the strongest peak at m/z = 42, which is similar to the reference 12 C 3 H 6 (Fig. 4d) except that the m/z value is shifted by 1 unit (Fig. 4j). Additional weak peaks located at m/z = 44 and 45 are also observed. This result indicates that the majority of propylene has two 12 C atoms and one 13 C atom and arises from the coupling of 13 CO 2 /* 13 COOH intermediates with the * 12 C 2 species that are produced from 12 CO reduction. This quantitative analysis supports our 24   . c,e,g,i, Mass spectra of ethylene under 12 CO 2 feeding (c), 13 CO 2 R feeding (e), 80% 13 CO 2 and 20% 12 CO co-feeding (g) and 20% 13 CO 2 and 80% 12 CO co-feeding (i) conditions. d,f,h,j, Mass spectra of propylene under 12 CO 2 feeding (d), 13 CO 2 feeding (f), 80% 13 CO 2 and 20% 12 CO co-feeding (h) and 20% 13 CO 2 and 80% 12 CO co-feeding (j) conditions. Article https://doi.org/10.1038/s41557-023-01163-8 hypothesis that the key *C 1 and *C 2 intermediates for propylene generation are likely to be molecular/adsorbed CO 2 or *COOH and highly protonated *C 2 , that is, *OCH=CH 2 , respectively.

Identification of *C 3 intermediates for propylene generation
To gain further insights into the key *C 3 intermediates for propylene production, we performed the electrochemical reduction of allyl alcohol, propionaldehyde, hydroxyacetone and propylene glycol by dissolving them in KOH + CsI electrolyte. The possibility of decomposition of these compounds in alkaline solution over the test period is ruled out by the observation of their fingerprint signals in 1 H NMR spectra after electrolysis. Interestingly, only the reduction of allyl alcohol leads to the production of propylene, and propylene is absent without applied bias ( Fig. 5a and Supplementary Fig. 24), consistent with a recent study showing a noticeable yield of propylene from electroreduction of allyl alcohol on oxide-derived copper catalyst 15 . We also reveal a linear relationship between the generation rate of propylene and the feed amount of allyl alcohol (Fig. 5b), but only <3.0% of propylene produced from CO 2 reduction could originate from the direct reduction of allyl alcohol precursor (see calculation in Methods and Supplementary Tables 11 and 12), indicating that the major *C 3 intermediate for propylene production is unlikely to be allyl alcohol.

Discussion
Based on the above mechanistic analysis, we highlight the key steps and crucial intermediates for propylene production (Fig. 6). CO 2 is first reduced to *COOH, which undergoes further reduction to *CO with elimination of a water molecule 45 . These *CO intermediates could undergo C-C coupling to form *C 2 species, which are hydrogenated to form either ethylene or ethanol 45,49,50 , with the former being favoured on The vertical black arrow indicates that the concentration increases from 1 mM to 50 mM sequentially. Inset: production rate of propylene as a function of the feed rate of allyl alcohol.  Fig. 6 | Proposed mechanism for the electroreduction of CO 2 to propylene on copper catalyst. CO 2 is first reduced to *CO intermediate. The coupling between two *C 1 intermediates leads to the formation of *C 2 intermediate, which is further reduced to C 2 products such as ethylene. The coupling between possible *C 1 and *C 2 intermediates leads to the formation of C 3 products such as propylene. The direct reduction of allyl alcohol also leads to the formation of propylene.
Article https://doi.org/10.1038/s41557-023-01163-8 our CuNC catalyst. Some of the adsorbed *C 2 intermediates that feature a carbon double bond and are highly protonated, such as *OCH=CH 2 , could undergo coupling with either molecular/adsorbed CO 2 or *COOH intermediates, followed by multiple proton-coupled electron transfer steps to form allyl alcohol and propylene.
The production of propylene on our CuNCs benefits from their nanostructure. The catalyst is composed of nanocrystals with prevalent Cu(100) and Cu(111) facets, providing active centres for binding the key *C 1 and *C 2 species and improving the intrinsic activity of CuNCs towards propylene production as compared to polycrystalline copper. The CuNCs are endowed with specific sites for CO 2 reduction, allowing them to reach substantial geometric current densities for propylene formation. We further investigated a large library of copper-based monometallic, bimetallic and even trimetallic electrocatalysts for improved propylene generation (Extended Data Fig. 8). We have achieved a peak FE of 1.83% for propylene generation with CuNCs coated with a 2 nm silver layer, which may result from the increased ethylene production due to the improvement of local CO concentration by the introduction of silver active sites. Moreover, enhancing the flow rate of CO 2 to 20 cm 3 min −1 leads to an increase in FE propylene by 14% on the CuNCs ( Supplementary Fig. 25).
Transfer of 18 electrons is required for the production of one propylene molecule from three CO 2 molecules. The intermediate species involved in this process are more numerous than the ones revealed by our study. Nevertheless, the mechanism uncovered in our study offers opportunities for designing advanced catalysts for the efficient production of this crucial chemical feedstock. Although still far from large-scale practical implementation, this study opens a pathway to the electrosynthesis of propylene via electrochemical CO 2 reduction.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41557-023-01163-8.

Synthesis of CuCl-derived copper catalysts
The synthesis procedure for the CuCl layer was modified from a previous study 1 . Here, a 920 nm-thick Cu film (99.995%) was deposited via sputtering (DP650, Alliance-Concept) onto a GDL (38BC, Fuel Cell Store). The Cu/GDL substrate was then electrochemically roughened in 0.1 M KCl electrolyte by repeating five cycles of triangular potential scans. Each triangular potential scan includes three steps: (1) a chronoamperometric step at an applied potential of 0.24 V versus RHE for 10 s; (2) a linear sweep voltammetric step from 0.24 to 1.74 V versus RHE at a scan rate of 500 mV s −1 ; (3) a chronoamperometric step at an applied potential of 1.74 V versus RHE for 5 s. The as-prepared GDL-supported CuCl films were then rinsed thoroughly with deionized water and dried using compressed air. The CuCl films were pre-reduced in a custom-designed flow cell at a constant current density of −30 mA cm −2 for ~80 s. The aqueous electrolyte was prepared by dissolving 56.11 g of KOH (Reactolab) and 51.96 g of CsI (99.999%, Alfa Aesar) into 1 litre of deionized water.

Characterizations of catalysts
The surface morphologies of copper-based samples were acquired using a field emission scanning electron microscope (Zeiss Merlin). Image acquisition was carried out via an in-lens detector under an accelerating voltage of 15 kV. High-resolution transmission electron micrographs of the catalysts were obtained using a transmission electron microscope (Talos, FEI) equipped with a high-angle angular dark field detector. X-ray diffraction was measured on a Bragg-Brentano instrument (Cu Kα radiation, λ = 1.5409 Å) with a grazing incident beam. XPS was performed using a PHI VersaProbe II scanning XPS microprobe (Physical Instruments). Analysis was carried out using a monochromatic Al Kα X-ray source of 24.8 W power with a beam size of 200 μm.

Electrochemical CO 2 reduction and product analysis
The electrolysis was performed using a Gamry (Interface 1000) potentiostat and each reduction reaction lasted for ∼3,000 s. iR correction was made automatically via a current interrupt mode. A custom-built flow cell containing a cathodic chamber and an anodic chamber separated by an anion-exchange membrane (Fumasep FKS-50, Fumatech) was used ( Supplementary Fig. 3). The catholyte was identical with the one used for pre-reduction (1 M KOH containing 0.2 M CsI). The anolyte was prepared by dissolving 56.11 g of KOH into 1 litre of deionized water. The as-prepared catholyte and anolyte were respectively pumped into the cathodic and anodic liquid chambers at the same flow rate of 0.25 cm 3 min −1 . Before and during the electrochemical reaction, the cathodic and anodic gas chambers were flowed continuously with various feeds at rates of 10 and 5 cm 3 min −1 , respectively, controlled by two mass flow controllers (Alicat Scientific). The flow rate of the cathodic chamber was also confirmed at the GC outlet by a soap-bubbled volumetric meter. The gases employed in this study include 12 CO 2 (99.999%, Carbagas), 13 CO 2 (Sigma-Aldrich, 99.0 at% 13 C), 12 CO (99.997%, Carbagas), helium (99.9999%, Carbagas) and their mixtures. The ratio between 12 CO 2 / 12 CO, 13 CO 2 / 12 CO or 12 CO 2 /He was tuned by adjusting the corresponding flow rate of each gas, with the total flow rate being 10 cm 3 min −1 .
For each measurement, fresh working electrodes and electrolytes were used. The geometric surface area of the working electrode was 0.33 cm 2 . A gas diffusion layer with sputtered 200-nm-thick platinum (99.995%) was used as the counter electrode and a Ag/AgCl electrode (saturated KCl, Pine) was used as the reference electrode. The electrochemical potential of the Ag/AgCl was calibrated daily against an RHE (HydroFlex, Gaskatel) immersed in 0.1 M HClO 4 solution, prepared by diluting 0.862 ml of HClO 4 (70%, ACS reagent, Merck) into 100 ml deionized water. All of the potentials measured in this work were referenced to the RHE using the following conversion: E RHE (V) = E Ag/AgCl (V) + 0.197 + 0.0591 × pH During the electrolysis three gas aliquots were automatically injected into an online GC (Trace ULRTA, Thermo). The first aliquot was sampled at ∼600 s after the start of reaction to ensure adequate equilibrium of gas phase. The gas products were separated by a micropacked shincarbon column (Restek) and quantified by a pulse discharge detector (Vici). The oven was programmed as follows: (1) holding at 60 °C for 3.5 min; (2) increasing to 180 °C with a ramp rate of 40 °C min −1 and holding at 180 °C for 2.5 min; (3) increasing to 260 °C with a ramp rate of 40 °C min −1 and holding at 260 °C for 3.5 min. The total run time was 14.5 min. The pulse discharge detector signals were calibrated using standard gas mixtures (Carbagas) with all relevant gases, including H 2 , CO, CH 4 , C 2 H 4 , C 2 H 6 and C 3 H 6 . The liquid products that dissolved in the electrolyte were collected after electrolysis. The electrolyte was mixed with D 2 O (99.9% deuterium, Aldrin) and an internal standard consisting of 25 mM phenol (>99.0%, Sigma-Aldrich) and 5 mM dimethyl sulfoxide (DMSO, 99.7%, Acros Organics). Then the prepared solution was analysed on a 1 H NMR spectrometer (Avance III HD 600 MHz, Bruker) and water suppression technique was applied ( Supplementary Fig. 5). The areas of the product peaks on the left of water peak were normalized against the peak area of DMSO, and the areas of the product peaks on the right of water peak were normalized against the area of phenol. Standard solutions with known concentrations of reference compounds of HCOONa (for HCOO − , >99.0%, Fluka Analytical), CH 3 COONa (for CH 3 COO − , >99.0%, Sigma-Aldrich), C 2 H 5 OH (≥99.8%, Fisher Scientific), C 3 H 7 OH (≥99.8%, Fisher Scientific) and C 3 H 6 O (>99.0%, Sigma-Aldrich) in 1 M KOH were used for calibration.
The partial current density of each product is calculated by multiplying the FE by the total geometric current density. The average of multiple measurements was used in this work.

Calculation of TOF
The TOF, in units of nmol s −1 cm −2 , is calculated based on the specific surface areas of catalysts determined by lead UPD measurements: 18 × 96,485.3 C mol −1 (1) where j propylene (specific) represents the partial current density of propylene against the specific surface area. The TOF value given in this work corresponds to the average of three independent measurements.

Isotope-labelling experiments and product analysis
Isotopically labelled 13 CO 2 (Sigma-Aldrich, 99.0 at.% 13 C) was also used as the reactant gas. Pure 13 CO 2 and mixed 13 CO 2 -12 CO feeding experiments were carried out using the same protocol as mentioned above. Gas products were collected into a withdrawal valve after the system reached equilibrium and were injected by microsyringe into a GC-MS instrument (Agilent 7890B/5977A). A GC equipped with a capillary column (HP-Plot/Q, Agilent) was used for chemical separation with helium (99.9999%, Carbagas) as the carrier gas. The GC was interfaced with an MS (5977A, Agilent), which was operated with a filament current of 34.6 μA and electron energy of 70 eV in electron ionization mode. The data acquisition and processing were performed with GC/ MSC MassHunter Acquisition software. The signals were referenced to National Institutes of Standard (NIST) library.

Electrochemical OH − adsorption and lead UPD
Electrochemical OH − adsorption on copper active sites was performed by CV using N 2 -saturated 1 M KOH solution with 0.2 M CsI additive as