Fischer-Tropsch synthesis using carbon dioxide, water and electricity


 The Fischer-Tropsch (FT) synthesis of fuels from CO and H2 lies at the heart of the successful and mature Gas-to-Liquid technology; however its reliance on fossil resources comes with the burden of an undesirable carbon footprint. In contrast, the electroreduction of CO2 (CO2RR) powered by renewable electricity has the potential to produce the same type of fuels, but in a carbon-neutral fashion. To date, only ethylene and ethanol are attainable at reasonable efficiencies and exclusively on copper. Herein, we report that the oxygenated compounds of nickel can selectively electroreduce CO2 to C1 – C6 hydrocarbons with significant yields (Faradaic efficiencies of C3+ up to 6.5%). While metallic Ni only produces hydrogen and methane under CO2RR and FT conditions respectively, we show that polarized nickel (Niδ+) sites facilitate ambient CO2RR via the FT mechanism. The catalysts yield multi-carbon molecules with an unprecedented chain growth probability values (α) up to 0.44, which matches many technical FT synthesis systems. We anticipate that the integration of the herein proposed electrochemical-FT scheme with fuel cells may provide at this seminal stage up to 7% energy efficiency for C3+ hydrocarbons, inaugurating a new era towards the defossilization of the chemical industry.

including n-hexane (C 6 ) were produced with a remarkable total FE = 14%, with outstanding selectivity to C 3+ hydrocarbons (FE C3+ = 6.5%) on some of the Ni oxygenate-derived electrocatalysts ( Fig. 1a; X-d: X = NP, NC, NB, NHC, NH, and NO; d means derived). This performance is more than twice the best values reported in the literature, but with a far outstanding range of multi-carbon products formed -25 carbonaceous molecules including branched hydrocarbons were detected (Supplementary Tables 1 and   2 Table 3) did not produce any detectable C 4+ hydrocarbons. Possibility of the Ni samples being contaminated by Cu before and after CO 2 RR could also be eliminated by detailed elemental analyses . Given the remarkably similar performances within the family of Ni materials, we chose NP-d, i.e., the most selective one toward C 2+ products (FE C2+ = 14.1%), to unveil the features behind their ability to reduce CO 2 to long-chain hydrocarbons.
The formation of C 1 -C 6 hydrocarbons, including branched ones such as isobutane and isobutene suggests polymerization of adsorbed C 1 intermediates (CO or CH x , x = 1, 2) following a FT mechanism (Supplementary Fig. 2; gas chromatography-mass spectrometry (GC-MS) of products in Supplementary Figs. 11-15; isotope-labeled 13 CO 2 experiments in Supplementary Fig. 11). Indeed, the distribution of C 1 -C 5 hydrocarbons 11 produced at the optimal potential of − 1.2 V versus reversible hydrogen electrode (RHE, all potentials reported hereafter are referenced to this scale) does follow the Anderson-Schulz-Flory (ASF) model ( Supplementary Fig. 16). We found linear correlations in the ASF plots for pristine and NiOmodi ed Ni discs, s-Ni and Ni oxygenate-derived samples ( Fig. 1b and Supplementary Fig. 17). The formation of long-chain hydrocarbons is more favored on NP-d electrocatalysts (α = 0.42) compared to s-Ni (α = 0.30), and becomes even more apparent as the number of carbon atoms grows (Fig. 1c). The α value for C 1 -C 5 hydrocarbons on NP-d electrocatalysts already matches those found for existing low range of technical FT synthesis catalysts 7 , like Fe (α = 0.39) 10 . Furthermore, we note that all the Ni electrocatalysts showed stable C 2+ hydrocarbon FEs over at least 12 h of continuous operation, which emphasizes their amenability for practical development (Fig. 1d, Supplementary Fig. 18).

Identi cation of the active sites
To identify the nature of the Ni active sites, we probe the chemical and physical properties of various materials. The highly defective, sputter-deposited s-Ni with a large fraction of geometric defects yielded only FE C2+ = 2.2%, which is ∼1/7 that of NP-d (Fig. 1a, Supplementary Fig. 7 and Supplementary Note 1).
This suggests that the contribution of under-coordinated Ni sites to CO 2 RR performance is negligible ( Supplementary Fig. 19). Positively charging (or polarization) of Cu sites in Cu matrices can enhance C-C coupling in CO 2 RR 12 . Therefore, we investigated the degree of Ni oxidation in the Ni oxygenate-derived electrodes during CO 2 RR. The oxidation state of Ni in all freshly-prepared Ni oxygenate compounds was + 2 (according to X-ray absorption near-edge spectroscopy, XANES, Fig. 2a, Supplementary Fig. 6 and Ni 2+ /Ni 0 redox transitions observed by cyclic voltammetry 13 , Fig. 2b). We then determine how the oxidation state of Ni evolved during CO 2 RR at − 1.2 V using operando XANES (Fig. 2c). Notably, the location of the absorption of the Ni K-edge indicated the existence of stable Ni δ+ sites. We found, by linear combination tting of the spectra (Supplementary Table 4), that 65% of the initial bulk Ni 2+ remained after 5 min of reaction, stabilizing at 34% after another 10 min. The concentration of phosphorus during CO 2 RR was also monitored using ICP-OES and found to be ~ 1.0 at. % (P/Ni) after prolonged electrolysis ( Supplementary Fig. 20). Thus, the Ni δ+ sites are likely to be in the form of Ni δ+ -(PO 4

) 3− (Supplementary
Figs. 20-21), or Ni δ+ -(OH) − species due to the local alkaline pH at the electrolyte/electrode interface under CO 2 RR conditions 14 . More details on the stability of the phases can be found in Supplementary Note 2 and Table 5. Overall, we found that polarized nickel sites can be stabilized by the local electrochemical environment 15 .
We performed CO stripping voltammetry to determine the properties of the Ni δ+ . Pristine Ni discs, s-Ni, and commercial NiO (NiO-c) were measured as representative samples for metallic Ni sites, under-coordinated metallic Ni sites, and Ni sites fully coordinated with oxygen, respectively. Pristine Ni showed a sharp CO stripping peak centered at 0.54 V (Fig. 2d), which is due to the expectedly strong adsorption of CO (for Ni(100), ΔE *CO = − 1.32 eV, Supplementary Table 6 Supplementary Fig. 25). The weaker CO binding energy on Ni δ+ sites can be attributed to fewer d-electrons available for backdonation in the Blyholder model 18 . Free from the CO poisoning which occurs on metallic Ni, polarized Ni electrocatalysts may thus allow subsequent reduction of adsorbed *CO species toward multi-carbon compounds.

Mechanistic analysis
To investigate the FT-mechanistic route promoted by Ni δ+ sites on NP-d, different C 1 feedstocks other than CO 2 (i.e., CO, CH 2 O, HCOOH) were electrochemically reduced. These were converted to multi-carbon products with lower FEs (Supplementary Fig. 26) and production rates (Fig. 3a). As for their co-feeding, the addition of CO or CH 2 O to CO 2 dramatically improved formation rates, with a CO 2 -CO co-feed having the best performance (Fig. 3a). In contrast, the presence of HCOOH showed a negligible impact.
Since a CO 2 feed overperformed the CO one, and they are accepted to be mechanistically linked via CO 2 → *COOH → *CO 19 , we propose that *COOH is a relevant intermediate in a polymer chain initiator role.
Veri ably, the CO 2 -CH 2 O feed did perform better than the CO-CH 2 O feed. Since CO or CH 2 O addition did not trigger changes in the α values compared to that of the pure CO 2 feed ( Supplementary Fig. 27), it can be deduced that neither CO nor CH 2 O interact directly with the existing adsorbates for C-C coupling.
Instead, we propose that the carbonaceous intermediates generated from CO or CH 2 O reduction are the ones to undergo C-C bond formation. CO electroreduction to CH and CH 2 is exothermic by 0.3 and 0.5 eV, respectively; thus both species can be formed on the surface (Supplementary Table 6). In parallel, *CH 2 is the generally proposed intermediate for electrocatalytic reduction of CH 2 O 20,21 , since CH 2 O protonation and consequent reduction to *CH 2 and H 2 O is highly exothermic at cathodic biases (ΔE < − 1 eV for U = − 1.2 V and q Ni < 0.56 |e − |, Supplementary Fig. 28). This is in line with the observed formation of C 1 to C 4 hydrocarbons from CH 2 O electroreduction ( Supplementary Fig. 2, Supplementary Table 7). Thus, we propose either *CH 2 or *CH to be one of the C-C coupling units (Fig. 3b). All in all, experimental observations support a mechanism where the rst C-C bond formation on Ni would proceed via the *CH or *CH 2 -*COOH coupling pathway (initiation), followed by subsequent *CH or *CH 2 insertions (chain growth steps). Here, we note that on Cu, the highest production rate of multi-carbon products was achieved using pure CO, which further highlights the stark mechanistic differences between Cu and Ni for CO 2 RR ( Supplementary Fig. 29).
To further understand the role of the Ni δ+ sites, DFT simulations at the PBE-D2 22 Table 9).
Following the qLSR dependencies, the proton-coupled electron transfer (PCET) assisted *CH 2 + *COOH coupling to form *OCCH 2 is exothermic at − 1.2 V (Fig. 3c) and becomes increasingly favored upon nickel charging (ΔE OCCH2 = − 0.64 eV on Ni 0 to − 3.61 eV on Ni 2+ ). The alternative, potential independent *CH 2 + *CO coupling becomes exothermic on nickel sites with Bader charges larger than + 0.6 |e − | ( Supplementary Fig. 33), although it is less favorable than the *CH 2 + *COOH path by more than 1 eV. This result, thus, con rms the role of *COOH to activate the C-C coupling, and rationalizes our experimental observations of higher multi-carbon product formation rates when a CO 2 -CH 2 O mixture is used in lieu of CO-CH 2 O (Fig. 3a-b). Furthermore, since *CH + *COOH coupling is only 0.5 eV more endothermic than *CH 2 + *COOH ( Supplementary Fig. 33), both pathways are open for CO 2 -CO co-feed upon polarization of the active sites. In contrast, for CO 2 -CH 2 O co-feed, only the *COOH + *CH 2 coupling can occur, thus limiting the FT performance ( Fig. 3a-b). In general, Ni δ+ sites are responsible for the enhanced C 2+ selectivity on NP-d electrocatalysts -they facilitate *CH/CH 2 insertions towards long-chain hydrocarbons during CO 2 RR (Fig. 3b, Supplementary Fig. 34).
It is notable that CO desorption on Ni becomes less endothermic as the polarization degree increases, whilst on Cu(100), this process is almost thermoneutral (Fig. 3d). Thus, in the case of nickel, the polarized sites anchor CO strongly enough to enable its conversion into *CH/CH 2 , while still preventing *CO surface poisoning characteristic of metallic Ni. In comparison, C 3+ formation on Cu is hindered by the easy elimination of CO and C 2 towards the solvent.
Overall, nickel polarization is key to open the rst C-C coupling route and favor the subsequent *CH/CH 2 insertion via an optimal tuning of intermediates' binding energy.
Prospects for future implementation of e-FTS Finally, we elaborate on the prospects for practical implementation of the e-FTS. The absence of promising electrocatalysts has precluded the study of this strategy, from the system engineering standpoint, to produce long-chain hydrocarbons. So far, environmental and technoeconomic assessments have considered hybrid schemes where CO 2 and H 2 O are separately transformed by electrocatalytic means into syngas feeding a FT reactor 26-28 . These studies conclude that they will not be economically appealing in the foreseeable future, compared to the fossil-fueled FT process (Fig. 4a), and may display even a larger environmental ngerprint in some cases 28 . Strikingly, the simplicity of e-FTS (Fig. 4b) calls for its potential implementation, as roughly estimated below, though solid conclusions can only be drawn after more comprehensive analyses.
The electricity-to-fuels (considered as C 3+ compounds) e ciency 29 attainable in an ideal electrolyzer with no ohmic losses and a state-of-the-art anode 30 (ca. 0.25 V overpotential) would be γ e−C3+ ∼ 3.5% considering yields in Fig. 1a (Supplementary Note 4). More importantly, the nature of the e-FTS would enable its integration including a partial recirculation of electrical energy not stored in the form of carbon products via the transformation in a fuel cell of the hydrogen byproduct (Fig. 4b). Considering the use of a state-of-the-art polymeric electrolyte membrane fuel cell with a hydrogen-to-electricity e ciency γ FC = 51% 31 , the total electricity-to-C 3+ products e ciency γ C3+ , Eq. (1), could ideally reach ca. 7%. Since ethylene is also a highly valuable product, including it would make γ C2+ reach ca. 12.5%.
At this point, we highlight that the theoretical limit of a less environmentally deleterious FT process where H 2 is produced electrolytically is 66% 32 . These encouraging estimations display the potential of developing e-FTS-based schemes to produce synthetic fuels, leaving this concept far from being a mere academic curiosity. NO was synthesized by calcinating the NH as prepared through the procedure aforementioned. The asprepared NH was calcined in a tube furnace under Ar atmosphere at 578 K for 6 h. The NO sample was collected after being cooled down to room temperature under Ar ow.

Synthesis of sputtered Ni
A gas diffusion electrode (Sigracet 38 BC, Fuel cell store) with a Vulcan XC 72 (0.5 mg cm − 2 ) coating was used as the substrate for Ni sputtering. Approximately 30 nm thick Ni lms were sputtered onto the substrate using a Ni target (99.99%, Chemicals Testing & Calibration Laboratory) at a sputtering rate ~ 3 nm min − 1 in a Discovery®-18 sputtering system at a base pressure of < 10 − 6 Torr.

Synthesis of NiO/Ni
The NiO/Ni samples were prepared in two sequential steps 33,34

Preparation of working electrodes
The catalyst ink was prepared by adding 20 µL Na on 117 solution (~ 5wt.% in a mixture of lower aliphatic alcohols and water, Sigma-Aldrich) into 1 mL of the sample dispersion in methanol. The wellmixed catalyst ink was coated onto a 2 cm × 4 cm gas diffusion electrode (Sigracet 38 BC, Fuel cell store) by airbrushing with N 2 as a carrier gas. After drying at room temperature overnight, a 2 cm × 2 cm of the as-prepared gas diffusion electrode was assembled into a ow cell electrolyzer for electrocatalytic CO 2 reduction performance test. The working area of the electrode is 1 cm 2 . As for the sputtered Ni sample, a 2 cm × 2 cm sputtered Ni gas diffusion electrode was used. The electrodes for CO-stripping experiments were prepared by drop-coating 30 µL of the catalyst ink on a graphite electrode (15 mm in diameter).
To prepare the ink for the Cu gas diffusion electrode, 6.36 mg of Cu nanoparticles (25 nm, Sigma-Aldrich) ( Supplementary Fig. 36), 10 mg of Vulcan XC 72, and 20 µL Na on 117 were dispersed in 1 mL methanol. The Cu gas diffusion electrode was prepared by coating the ink using airbrushing with N 2 as a carrier gas.
CO stripping measurement 0.1 M KHCO 3 was used as the electrolyte for CO stripping experiments. The electrodes were rstly reduced at − 1.0 V for 10 min under Ar (15 mL min − 1 ) in a H-type cell. The CO was adsorbed on the working electrode by continuously owing CO (99.97%, Linde Gas Singapore Pte. Ltd.) into a H-type cell at a ow rate of 15 mL min − 1 for 10 min, while holding the potential at − 0.8 V. The gas was subsequently switched to Ar for 10 min to purge CO traces from the working electrode compartment. Cyclic voltammetry was then carried out by sweeping the potential from 0 to 1.0 V at 50 mV s − 1 . A freshly polished Ni disc, commercial NiO (NiO-c) ( Supplementary Fig. 36), and sputtered Ni gas diffusion electrode were measured as references.
Instrumentation X-ray diffraction (XRD) patterns were collected on a D8 Advanced Powder Diffractometer (Bruker) with a K a1 germanium monochromator for Cu radiation (λ = 1.5406 Å). X-ray photoelectron spectroscopy (XPS) measurements were carried out on a K-alpha XPS spectrometer (Thermo ESCALab 250i-XL) using Al K a Xray radiation (1.486.6 eV). Carbon C1s at the position at 284.6 eV was used as a reference to correct the charging effect. X-ray absorption measurements at the Ni K-edge were performed at the XAFCA facility in Singapore Synchrotron Light Source (SSLS). An IFEFFIT package was employed to analyze the XAS data 35 . Fourier-transform infrared spectroscopy (FTIR) spectra were collected on a Bruker Alpha spectrometer.
Raman spectroscopy was performed using a Raman microscopy system (LabRAM HR Evolution, Horiba Scienti c) with a He-Ne laser (λ = 633 nm). Transmission electron microscopy and energy-dispersive X-ray spectroscopy (EDS) mapping were conducted on a JEOL JEM-2010 transmission electron microscope. Scanning electron microscopy (SEM) and corresponding EDS analysis were performed on a eldemission SEM (JEOL JSM-7600F). ICP-OES was carried out on a Perkin Elmer Avio 500 Inductively Coupled Plasma-Optical Emission Spectrometer. Gas chromatography-mass spectrometry (Shimadzu GCMS-QP2020) was used to distinguish the straight and branched hydrocarbons.
Electrocatalytic reduction of CO 2 The electroreduction of CO 2 (CO 2 RR) experiments were conducted using a three-electrode ow cell electrolyzer on a Gamry Reference 3000 potentiostat/galvanostat/ZRA. An anion exchange membrane (AMV, AGC Asahi Glass) was used to separate the cathodic and anodic compartments. The as-prepared gas diffusion electrode, Ag/AgCl (saturated KCl), and Pt foil were used as the cathode, reference electrode, and anode. 0.1 M KHCO 3 as electrolyte was circulated through the electrochemical cell using a peristaltic pump. Chelex® 100 (100 mesh, Sigma-Aldrich) was used to purify all the electrolytes of possible trace metal contaminants. The Gaseous CO 2 (99.999%, Linder Gas Singapore Pte. Ltd) passed through the gas chamber with a ow rate of 5 SCCM. The applied potentials were corrected using postmeasurement iR drop correction with the uncompensated resistance (R u ) of 86 ohm measured by Electrochemical Impedance Spectroscopy (EIS) at 1.0 MHz. Before CO 2 RR, the cathode was pre-reduced at − 1.2 V for 1 min with CO 2 ow. The gas products during the 5000 s electrocatalysis period were analyzed by an on-line gas chromatograph (Shimadzu GC2014) equipped with one thermal conductivity detector and two ame ionization detectors. The liquid products were analyzed using a headspace-gas chromatograph (Agilent, 7890B and 7697A) with a ame ionization detector and high-performance liquid chromatography (Agilent 1260 In nity). The amount of each product was calculated by subtracting the values from control experiments (Supplementary Table 10 rate for all the cases was 5 mL min − 1 , except for the case of CO 2 -CO in which the total gas ow rate was 10 mL (5 ml min − 1 for CO 2 and CO, respectively). In the cases without CO 2 or CO gas, Ar with a gas ow rate of 5 mL min − 1 was used instead. The concentration used for the liquid reactants is 0.04 M for CH 2 37 , with PBE as density functional 22 and van der Waals dispersion introduced through the DFT-D2 method 38,39 , with our reparametrized C 6 coe cients 40 . PAW was employed to represent inner electrons 41,42 and the monoelectronic states corresponding to valence electrons were expanded as plane waves with a kinetic energy cutoff of 450 eV. To address the role of nickel polarization, we modeled the catalysts as Ni(100) p(3×3) doped with near-surface oxygen atoms, which we xed in 1 to 4 hollow sites around the nickel center where intermediates are adsorbed (Supplementary Fig. 30). As metallic references, we considered Ni(100) p(3×3) and Cu(100) p(3×3), the state-of-the-art catalyst for C-C coupling so far 2 . For the most oxidized system, we took the rock-salt structure, antiferromagnetic NiO(100) p(3×3), so that nickel charging could be extended to Ni 2+ oxidation state 43 . To properly reproduce NiO electronic properties, we applied a Hubbard correction U eff = 5.3 eV = U -J, with J = 1 eV, which are values typically found in literature [43][44][45] . All simulations related to Ni, O-doped Ni, and NiO were spin-polarized. All the employed slabs contained four layers, with the two outermost relaxed and the rest xed to the bulk distances. Vacuum thickness between periodic repetitions of the slabs accounted for at least 12 Å. We sampled the Brillouin zone by a Γ-centered k-points mesh from the Monkhorst-Pack method 46 , with a reciprocal grid size smaller than 0.03 Å −1 . Each selected C 1 -C 4 intermediate and H was placed on Ni 0 , Ni δ+ , or Ni 2+ sites to assess the in uence of active site polarization on its binding energy.
Since adsorption was limited to one side of the periodic cell, we imposed a dipole correction to correct artifacts from the asymmetric slab model 47 . We reported all formation and adsorption energies using as energy references: CO 2 (g), H 2 (g), and the clean surface, either Ni(100), O-doped Ni(100), NiO(100), or Cu(100). We followed the computational hydrogen electrode formalism to de ne the relative energy between the H + and the H 2 (g) at U = 0 V and standard conditions and to correct formation energies for contribution of electric potential in case of proton-coupled electron transfer (PCET) steps 23