The capture and sequestration of atmospheric CO2 has become one of the great challenges of our society as its rate of production and accumulation scales with population growth. Aqueous phase electrochemical reduction of CO2 (CO2RR) to hydrocarbon fuels, value added chemicals, and chemical precursors/feedstock shows great promise for both anthropogenic CO2 remediation and on-demand storage of energy generated from renewable sources such as solar and wind [1][2][3] .
There are, however, three critical limitations that must be addressed to further advance the potential commercial integration of CO2 electrolysis: (1) high overpotential [4][5][6][7][8] , (2) poor selectivity 4,[6][7][8][9][10][11] , and
(3) low aqueous phase solubility of CO2 12 . The high stability of the nonpolar CO2 molecule and the multiple proton/electron transfer steps required to produce the desired reduced products present a great technological challenge to the development of next generation electrocatalytic materials.
CO2RR rate and selectivity are sensitive to the general material properties of the catalyst (conductivity, reactant/intermediate/product adsorption free energy, etc.), applied overpotential, near surface pH, etc. 13 . This gives rise to a varied range of products including CO, formate/formic acid, H2, alcohols, and hydrocarbons ranging from C1 to multiple carbon species. Consequently, overall selectivity for one single product is low 11,[14][15][16] . has great utility as both an electrochemical fuel [27][28][29][30][31] and a chemical precursor [32][33][34][35] , making it a more desirable reduced product than CO. The high overpotentials, as large as 1 V in some cases 6,8,9,[17][18][19][24][25][26] , on traditional HCOO -/HCOOH evolving metals, however, is a significantly limiting barrier.
In the early surveys of the activity and selectivity of a range of polycrystalline transition metals, Pd, which like Pt is a strong CO binding metal, was found to evolve CO and H2 as the major products 36 . More recent work by Kanan 37 and Koper 18 have demonstrated that at low overpotentials, < ~200 mV, Pd nearly exclusively reduces CO2 to HCOO -/HCOOH with FEs approaching 100% at geometric current densities as high as 6.5 mA cm -2 18,37 . Increasing the overpotential beyond ~ 500 mV changes the selectivity to H2/CO with their molar ratios being a function of overpotential 38 . The 2-electron CO2RR to HCOO -(near neutral bicarbonate electrolytes), with a reversible potential of approximately 0.02 V vs. RHE 39 has been termed an electrohydrogenation process where the simultaneous transfer of a proton and electron forms an adsorbed HCOOad intermediate rather than COad 37 . The source of the high electrohydrogenation activity of Pd has been suggested to be due to its ability to form a saturated hydride phase near 0 V vs. RHE [40][41][42] , H/Pd ratios as high as 0.7 43 . The difference in observed FE and selectivity for HCOO -when comparing planar, polycrystalline electrodes to nanostructured materials is likely related to the rate of formation and degree/ease of participation of the α,β-PdH active phase, which is likely more facile on nanostructured surfaces with a higher defect density 37,41,44,45 . The hydride mediated reduction to HCOO -/HCOOH has also been observed on non-Pd hydrides 46 . The key to the selectivity of Pd is the ability of Pd to form an active hydride phase, shifting to an HCOOad intermediate rather than a COOHad intermediate which is the first electron transfer step toward COad 37,41 .
A major limitation of Pd for CO2 electrolysis is the gradual current decay initiated by the formation of a poisoning CO adlayer 37 . While the HCOO -/HCOOH FE on Pd at low overpotential is near 100%, slow but consequential rates of the 2-electron CO pathway (< 0.1% FE) 37 result in eventual deactivation. This process has received attention recently with groups focusing on the CO tolerance of Pd nanomaterials during CO2RR electrolysis. Sargent et al. 47 demonstrated that CO poisoning is suppressed on branched Pd nanoparticles that are composed of high index facets in comparison to other shape controlled particles that are bound by lower index facets. A similar shift to stable electrolysis through CO tolerance at moderate formate FEs was observed for porous Pd structures formed through electrodeposition 45 . The observed operational stability is again likely due to the high index configuration of the tortuous, porous structure. Another strategy involves a barrier Cu film on the surface of the Pd particles, demonstrating operational longevity during extended electrolysis while still exhibiting a high formate FE 48 . The presence of the underlying Pd weakens the degree of interaction between CO and Cu and limiting the thickness of the Cu film to 1 ML results in a similar product distribution to pure Pd 48 . For CO2RR, two classes of Pd alloys have been explored: (1) segregated alloys where Pd is the active surface metal 49 and (2) bimetallic surfaces where there is a cooperative effect between two dissimilar metals on the surface. The first class of Pd alloys including PdCu 50,51 , PdNi 52 , PdTe 53 have mostly been explored for improvement in the rate and selectivity of CO production. Yuhan et al. 54 have explored the second class of Pd alloys in which a mixed Pd-Sn surface is shown to promote high formate FEs at potentials as low as -0.26 V vs. RHE. More recently, Cai et al. 55 have extended the operational potential window for high FE formate production on Pd to higher overpotentials, yielding high formate production rates, by doping Pd with 7 at. % boron. They achieved a peak formate FE of 70 % at -0.5 V vs. RHE. Despite these efforts, CO poisoning for CO2RR within 0.5 V overpotential is still a major challenge for any Pd based catalyst.
Here we present the development of free-standing, nanoporous Pd-X (np-PdX) alloys (X = Ni, Co, Ag, Cu) with high HCOO -FE at low overpotentials and CO poisoning tolerance during extended CO2RR electrolysis compared to existing Pd/C electrocatalysts. Dealloying of Pd0.15X0.85 alloys results in the formation of a Pd-skinned, core-shell bicontinuous nanoporous architecture where the interior of the ligaments retain a high fraction of the X alloying component and the surface is passivated in Pd. We argue that it is the core-shell structure that results in the modification of both the electronic properties of the surface and H-solubility of the bulk solid, changing the degree of interaction with CO and mitigating poisoning. Rate of catalyst deactivation is found to be a result of the combination of alloy composition dependent CO binding strength and CO adsorption induced hydride capacity shifts. The presented freestanding np-PdX electrodes have advantages over Pd/C including morphological stability, negation of the need for ionomer binders in the catalyst layer, and CO poisoning tolerance during extended CO2RR electrolysis.
With this electrocatalyst architecture we have attained sustainable production of HCOO -, which is a highly desirable electrochemical fuel and chemical precursor, at low overpotential with industrially relevant selectivity, current density, and operational durability, both in terms of morphology retention and intermediate poisoning mitigation.
Pd/C (30 wt.% Pd): Palladium nanoparticles are prepared through organic phase solvothermal reduction 56 . Under N2 flow, 75 mg Pd(acac)2 (Palladium (II) acetylacetonate, Sigma Aldrich, assay: 99%) is dissolved in 15 mL of OAm (Oleylamine, Sigma Aldrich, assay: 70%) at 60 o C. A separate solution of 300 mg BTB (Borane tert-butylamine complex, Sigma Aldrich, assay: 97%) in 3 -4 ml OAm is quickly injected into the OAm solution at 60 o C. The temperature is increased to 90 o C at 3 o C/min and kept under stirring for 1 hour. The solution is then cooled to room temperature and mixed with 10 -20 ml chloroform for uniform dispersion of particles. The cooled solution is added drop wise to a separate solution of 60 mg carbon (Vulcan XC-72R, Fuel cell store) in 200 mL chloroform under sonication. The final mixture is sonicated for 30 minutes.
Following loading of the catalyst on the carbon support, 20 mL ethanol is added to precipitate the catalyst under centrifugation (8000 rpm, 20 minutes). Cleaning/collection cycles in hexane/ethanol mixtures are repeated at least three times. After the final cleaning, the collected particles are dried and the resulting Pd/C powder is annealed for 1 hour at 150 o C in air to remove residual surfactant and 2 hours at 250 o C in H2/Ar before use. Electrodes incorporating the Pd/C catalyst are prepared through drop casting of Pd/C ink on Ti flags (1.4 cm x 1 cm, 0.127 mm thick, Alfa Aesar). Ti flags are pre-cleaned in boiling oxalic acid (10% w/v, Alfa Aesar) for 1 hour. The catalyst ink is composed of 2 mg mL -1 Pd/C in isopropanol (2-propanol, Sigma Aldrich, assay: 99.999%). Nafion solution (Ion Power) is added to the ink in an amount that yields a dried catalyst layer with 0.01 wt.% Nafion. Formation of the catalyst layer is accomplished through drop casting of 200 µL of the catalyst ink onto a Ti flag followed by drying under vacuum.
Nanoporous palladium (np-PdX) is prepared by the electrochemical dealloying of Pd15X85 (X = Co, Ni, Cu, Ag) alloy foils. Following melting and homogenization of the alloys from high purity precursor metals (Kurt J Lesker, purity above 99.9%) in a radio frequency induction furnace (Ambrell EasyHeat 3542 LI) under Ar, the alloys are rolled into thin ribbons using a rolling mill.
Intermittent annealing under Ar is used during the rolling process to remove work hardening and promote homogenization of the precursor alloys. Electrodes are fabricated from the alloy foils by spot welding a ~0.25 cm 2 foil to a Ti wire which was subsequently sealed in epoxy, only exposing one face of the alloy foil.
As a consequence of the strong composition dependence of dealloying rates, dealloying conditions were optimized for each of the different alloys with the goal of attaining similar length scales (pore/ligament size) and residual composition. np-PdCo and np-PdNi are prepared by electrochemical dealloying (three electrode cell with Toray carbon paper counter electrode and Ag/AgCl reference electrode) in Ar purged 10 mM HCl + 0.5 M H2SO4 with potential cycling between -0.13 V and 1.02 V vs. RHE 50 mV s -1 (18 cycles). np-PdCu and np-PdAg are prepared by dealloying in 0.1 M H2SO4 with potential cycling between 0.47 V and 1.27 V vs. RHE (12 cycles) and in 0.1 M HNO3 between 0.87 V and 1.52 V (12 cycles), respectively.
The structure and surface morphology of Pd/C and np-PdX are analyzed using Transmission Electron Microscopy (TEM) (JEOL JEM2100), Scanning TEM (STEM) electron energy loss spectroscopy (EELS) (JOEL 2100F with a Quantum Gatan Imaging Filter), and Scanning Electron Microscopy (SEM) (Zeiss Supra 50VP). EELS elemental mapping of np-PdX shows the elemental distribution in the ligamentous structure. XPS measurements were performed with a PHI Veraprobe 5000 instrument equipped with Monochromated AlK alpha source. Analysis is made with X-ray settings of 100 micrometer, 25 W e-beam and photoelectrons are collected using hemispherical analyzer. Pass energy is 117 V for surveys. Dual beam neutralization is performed to compensate for depletion of photoelectrons from the surfaces.
All the electrochemical experiments are conducted in a three-electrode electrochemical cell using an Autolab PGSTAT128N potentiostat. The CO2RR is performed in 1 M KHCO3 (Sigma Aldrich, 99.7%) made with Millipore water (Milli-Q Synthesis A10). Prior to all experiments, the electrolyte is first purged with Ar followed by CO2 until saturation, ~30 min. The 1 M KHCO3 has a pH of 8.4 when Ar purged and a pH of 7.8 when saturated with CO2. All the experiments involving CO2RR, polarization curves and extended electrolysis, are performed in a custom-made two compartment gas tight glass cell separated by a Nafion membrane (Ion Power). Electrolysis experiments were run in a semi-batch mode where CO2 was continually bubbled into the catholyte to ensure continuous saturation, drive electrolyte mixing, and prevent any pH gradients from forming. After extended electrolysis, the catholyte is collected for product analysis. A Ag/AgCl electrode (BASi) is used as the reference electrode with Pt mesh as the counter electrode. All the glassware is cleaned in a 1:1 v:v solution of concentrated sulfuric and nitric acid followed by boiling in Millipore water before each use.
As Pd and its alloys can absorb hydrogen, CO stripping is used to calculate the electrochemically active surface area of the np-PdX electrodes. A monolayer of CO is deposited onto the np-PdX electrodes by holding the potential at 0.05 V vs. RHE in 0.1 M HClO4 while bubbling CO into the solution. Following passivation of the surface with CO, the electrolyte is purged with Ar for 30 minutes. The potential is then sweep anodically at a rate of 50 mV/s to oxidize the monolayer of CO. Integration of the CV curves yields the charge associated with CO oxidation which can be converted to ECSA using the conversion factor 420 µC/cm 2 . Roughness factor (RF) is calculated by dividing the ECSA by the geometric area of the electrode.
Quantification of HCOO -in the 1 M KHCO3 electrolyte after constant potential electrolysis is accomplished using a 300 MHz Varian Unity Inova NMR. A minimum of 1 hour of electrolysis is used to build up HCOO -for detection. For the samples that quickly deactivate, electrodes are periodically reactivated 37 to continue electrolysis. A sample of the electrolyte post-CO2RR is mixed (5:1) with a solution of DMSO (Dimethyl Sulfoxide, Sigma Aldrich, 99.9%) and D2O (7.5 µl DMSO in 15 ml D2O (Deuterium Oxide, Sigma Aldrich, 99.9%)) which serves as an internal standard. The 1 H spectrum of HCOO -is measured after a pre-saturation protocol.
The faradaic efficiencies (FE) of HCOO -evolution are calculated using the following equation:
where F is Faraday's constant (96485 C mol -1 ), V is the volume of catholyte to be tested for product quantification, C is the HCOO -concentration obtained from NMR analysis of catholyte and Q (integration of current over time for electrolysis period) is the total charge passed through the cathode during electrolysis. Partial formate current density is the product of formate FE and total current density (𝑗𝑗 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 = 𝐹𝐹𝐹𝐹 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 * 𝑗𝑗).
Kohn-Sham Density Functional Theory (DFT) studies of each surface were modelled as a 2 × 2 fcc (111) surface slab with five layers where atoms in the bottom two layers were fixed to coordinates from the bulk lattice of the respective alloy, and the remaining atoms were relaxed.
Pd3X (X=Ag, Co, Ni, Cu) skin alloys were represented by substituting 50% of the Pd atoms in the second (subsurface) and fourth layers with X atoms (See Supporting Information for atomic coordinate files of each structure). Electronic energies were calculated using the PBE 57 exchange correlation functional using the VASP code with projector augmented wave pseudopotentials 58,59 , and a plane wave basis set was used to represent valence electrons with a 350 eV energy cutoff.
An 8 × 8 × 1 Monkhorst-Pack sampling of the k-point grid was employed for each surface. The kpoint sampling and energy cutoffs were found to result in converged energies. The minimum energy configuration of each reference surface was determined iteratively with a conjugate gradient algorithm until the difference in steps was less than 0.1 meV. Band occupancies were broadened using Fermi smearing with a width of 0.4 eV. Spin polarized calculations are reported for the Pd3Ni and Pd3Co alloy systems as these were significantly impacted by spin-polarization while the other systems were not.
An expansive leap in both the distribution of products and product selectivity for the CO2RR occurred with the transition from bulk polycrystalline metals 9 to nanostructured materials 15,37,41,44,[60][61][62] . This is especially true for Pd. The early work by Hori et al. 1,9 showed CO and H2 as the major products for CO2RR on polycrystalline Pd. However, with the shift to carbon supported nanoparticles (Pd/C) 37,41,44,60,61 and nanostructured thin films 45 , Pd has been shown to reduce CO2 to formate ion (HCOO -) with near perfect selectivity, high faradaic efficiency, at low overpotentials in bicarbonate electrolytes 37,41,60,61 . We observe similar results for the Pd/C catalyst shown in the transmission electron micrograph (TEM) and scanning electron micrograph (SEM)
in While the initial reduction current increases with increasing overpotential, the time to deactivation or near complete loss of reduction current due to CO poisoning scales inversely with that trend. This is in line with a progressive switch from the HCOO -/HCOOH pathway to the CO dominant pathway with increasing overpotential 36,38,42 . As has been observed previously 37 , while the calculated FE for HCOO -in these low overpotential regimes is near 100 %, there is a gradual decay in reduction current during constant potential electrolysis, the rate of which increases with increasing reaction overpotential. This decay in reduction current on Pd has been attributed to gradual poisoning of the surface by CO, as Pd has a negative adsorption free energy for CO 37,63 .
While CO is not detected as a reduction product in either the liquid electrolyte or gaseous head space at low overpotential, the slow evolution of CO eventually leads to the passivation of the surface. An additional limitation of Pd/C, as has been shown previously 37 , is the time dependent sintering of the particles on the carbon support during constant potential electrolysis due to the weak physisorption of the particles onto the carbon. This agglomeration leads to the decrease in electrochemically active surface area (ECSA) over time with significant losses observed for short periods of constant potential electrolysis 37 .
To address the limitations of nanostructured Pd catalytic materials, we take advantage of a process known as dealloying to evolve a high surface-to-volume ratio, nanostructured, core-shell Pd alloy electrocatalyst. Through manipulation of dealloying conditions and protocol, precise control of pore size and distribution as well as residual composition and core-shell structure is achieved in a one-step synthesis prodcure [64][65][66][67][68] . In addition to the relative simplicity of scaling of the dealloying process, nanoporous metal electrodes are free standing, negating the necessity of carbon supports and eliminating the material agglomeration mechanism as a source of ECSA loss during extended electrolysis. and preservation of a composition near that of the precursor alloy in the interior of the ligaments is an intrinsic consequence of dealloying [64][65][66][67][68] . Dealloying in acidic electrolytes at potentials well above the equilibrium potential of the less noble component drives anodic corrosion and ensures that the surface contains only the more noble component, Pd in this case. The roughness factor (RF), which is proportional to the ECSA, increases with dealloying cycle number (Figure 4). We note that beyond a certain extent of dealloying (number of cycles), the ECSA as measured by CO stripping begins to decrease, Figure 4. We contend that this is due to the competition between movement of the etch front into the bulk of the alloy foil and post-dealloying coarsening behind the etch front which can lead to loss in surface-to-volume ratio [69][70][71] . X-Ray photoelectron spectroscopy (XPS) is used to characterize the surface electronic states of the np-PdX electrodes.
Figure S3 shows the peaks corresponding to Pd 3d3/2 (high energy band) and 3d5/2 (low energy band) spin-orbit states of zero valent Pd for all four alloy electrodes. As shown earlier 72 73 . This change in electronic structure of the surface Pd skin directly impacts the adsorption of CO. For each electrode, an exponential increase in reductive current density is observed in the absence of CO2, initiating at approximately 0 V vs. RHE as would be expected for the evolution of H2 on Pd. In the presence of CO2, we observe deactivation behavior for Pd/C, np-PdAg, and np-PdCu, Figures 5 andS1. Initially, the reductive current increases as the potential is swept cathodically, but eventually begins a quick decay at potentials beyond -0.5 V vs. RHE. The strong binding of CO on Pd causes fast deactivation as the selectivity switches from HCOO -/HCOOH to CO/H2 at higher overpotentials [36][37][38]42 . Strikingly, no deactivation is observed in the polarization curves of np-PdNi and np-PdCo, Figures 5(b) and S1.
There are two potential reasons for this: (1) the binding of CO to a Pd-skinned surface with Ni and Co in the near surface region is weaker than that on Pd, np-PdCu, and np-PdAg; (2) H-binding on np-PdNi and np-PdCo is weaker than that on the other alloys, therefore HER is enhanced. The potential takeaway from this result is that if HCOO -selectivity at low overpotential is not altered by the presence of the transition metal alloying component in the bulk and near-surface regions of the Pd-skinned nanocatalyst, the rate of CO poisoning and electrolysis current decay could be significantly reduced. and ECSA normalized, Figure 7, formate partial current densities presented here are higher than those reported by others 47 . However, few other reports normalize total and partial CO2RR currents by the real ECSA, relying on geometric area of the electrode which ignores factors such as nanomaterial morphology and loading. The CO2RR currents measured at -0.2 V RHE decrease to a stable minimum towards the maximum attainable roughness factor. Depth of porosity, which is directly proportional to ECSA, does not result in a linear increase in CO2RR current owing to reactant mass transfer limitations within the tortuous porosity. This phenomena has previously been observed for the oxygen reduction reaction (ORR) on np-PtNi 74 .
In Simple d-band models, however, do not always directly correlate to trends in binding energy. This is especially true when lateral adsorbate interactions are significant and in the cases where adsorbates induce changes in the surface or near-surface structure or composition 79 . To assess the impact of alloying and hydrogenation on CO binding, we carried out DFT calculations to determine the low energy adsorption configurations for CO* and H* on Pd and the Pd-skinned alloys. We determined the most stable configurations of different adsorbates on different surfaces by defining the binding energy (BE) for each system as:
where Eslab, Emol, and Eads are the DFT energies of a surface slab either clean or containing adsorbed species, a single adsorbate molecule in the gas phase, and the surface slab with all adsorbed species, respectively. Our computational models show that CO* binds most strongly at an hcp site on pure Pd and at an fcc site for all other surfaces. H* is bound most strongly at an fcc site for all surfaces. Reliably quantifying BEs for these systems under realistic electrochemical conditions is very challenging, but qualitative trends in relative BEs for CO* and H* on a variety of hydrogenated configurations should provide physical insights into the experimental observations.
It is hypothesized that the most active electrocatalysts would exhibit weaker H binding to better facilitate hydrogenations of CO2 into formate as well as weaker CO binding that would result in less CO poisoning. Thus, we modeled CO* on clean surfaces, CO* on hydrogenated surfaces, H* on clean surfaces, and H* on hydrogenated surfaces as well as surfaces with a pre-adsorbed CO to find those with the lowest H and CO binding energies. We investigated the possibility of forming various hydride configurations on these surfaces, including sub-surface hydrides, but all sub-surface hydride cases are found to be higher in energy than the surface adsorbed hydrogen configurations for the Pd alloys. The calculated results are summarized in Figure 10. DFT calculations show that H and CO bind most strongly to clean surfaces (as expected) with magnitudes ranging from 0.46 -0.62 eV for H and 1.71 -2.03 eV for CO on all surface models.
Furthermore, all binding energies decrease in the presence of co-adsorbed species. Interestingly, the Pd-skin/Pd3Co system is unique among this group as it yields the weakest binding energies with any low-energy surface configuration. Indeed, the observed trend in calculated H and CO binding energies across these different surfaces straightforwardly follows our experimentally observed trend in formate production and catalyst deactivation.
Furthermore, based on differences in CO binding strengths on the more hydrogenated systems, it appears that higher degrees of hydrogenation effectively destabilize CO binding more so on the Pd-skin/Pd3Co system relative to the other systems. Past work has shown that selectivity of CO2 reduction is a function of potential and gradually switches from HCOO -/HCOOH to CO/H2 with increasing overpotentials [36][37][38]41,42,55 . The strong potential dependence of the selectivity has been associated with the breakdown of a PdHx surface hydride phase due to COad formation. This hydride breakdown leads to a proportional shift in intermediate selectivity from HCOOad to COOHad with increasing overpotential 41,42,55 . Thus, for the surfaces on which strong CO adsorption is enough to induce the decomposition of the surface hydride, not only would the CO adsorption strength be increased, but the selectivity would also shift toward the COad pathway, even at low overpotentials, hastening COad passivation. The relatively strong binding of CO to Pd and its alloys ensures fast passivation with CO production, meaning that any shift in selectivity at low overpotential will not be detected as CO in the headspace as the current will quickly decay to zero after only a small amount of CO is formed. The low CO and H BEs on Pd-skin/Pd3Co calculated using DFT thus help explain the exceptional operational stability of np-PdCo as observed in Figures 8 and9.
Finally, we wish to comment on the morphological stability of these nanoporous metal catalysts. Comparative CVs before and after CO2RR load cycling and post-mortem SEMs of the np-PdX electrodes after extended electrolysis at a range of potentials indicate a negligible loss in active area and morphology, Figures S6 and S7 respectively. In addition to the morphological stability, these np-PdX electrodes are not susceptible to delamination, loss of contact between carbon supported catalyst and current collector, as is common with ionic polymer mixed, agglomerated carbon-supported composite electrodes 80 . While increasing ionomer contents improves the durability of these agglomerated electrodes, it is offset by activity losses associated with active site blocking and electrode impedance 81 . These results highlight the utility of nanoporous metal electrodes where high aerial partial current densities and selectivity for HCOO - are combined with an electrochemical and morphological stability that cannot be matched by nanoparticulate based CO2RR electrocatalysts.
In