Correlations between experiments and simulations for formic acid oxidation

Electrocatalytic conversion of formic acid oxidation to CO2 and the related CO2 reduction to formic acid represent a potential closed carbon-loop based on renewable energy. However, formic acid fuel cells are inhibited by the formation of site-blocking species during the formic acid oxidation reaction. Recent studies have elucidated how the binding of carbon and hydrogen on catalyst surfaces promote CO2 reduction towards CO and formic acid. This has also given fundamental insights into the reverse reaction, i.e. the oxidation of formic acid. In this work, simulations on multiple materials have been combined with formic acid oxidation experiments on electrocatalysts to shed light on the reaction and the accompanying catalytic limitations. We correlate data on different catalysts to show that (i) formate, which is the proposed formic acid oxidation intermediate, has similar binding energetics on Pt, Pd and Ag, while Ag does not work as a catalyst, and (ii) *H adsorbed on the surface results in *CO formation and poisoning through a chemical disproportionation step. Using these results, the fundamental limitations can be revealed and progress our understanding of the mechanism of the formic acid oxidation reaction.

To establish reasonable context to our study we conducted a relative extensive literature study of what we gauge to be relevant papers concerning the formic acid oxidation reaction (FAOR) for direct formic acid fuel cell (DFAFCs) applications. The papers by Kolb 1 and Adzic 2 are prime examples of how in-depth model studies on extended surfaces can help researchers as the right questions. Cyclic voltammograms CVs of Pd(hkl) and Pt(hkl) have been compiled in Figure S1. Figure S1 allowed us establish the following: -Pt(100) appears to exhibit the lowest onset potential at ~0.25 VRHE, however as the catalyst shows complete inactivity on the cathodic sweeps, i.e. severe poisoning is suspected. In terms of FAOR onset order (FAOR peak position) it appears to follow that Pt(100)>Pd (111)>Pt (111)>Pd (110)>Pt (110)≈Pd (100).
-By considering the highest current as metric for activity 3 the order is then Pd(100)>Pt(100)> Pd(110)> Pd (111)>Pt (110)>Pt (111). I.e. it appears the more open fc structures shows better intrinsic activity. Figure S1 helped us to come to grip with the fact that more than one parameter is relevant to give insight to the FAOR performance and not just focus on maximum current. E.g. having an catalyst with extremely high current but high onset potential, such as Pd (100), essentially result in a DFAFC with very low cell voltages. From Figure 1 it appears that open (100)-like fcc surfaces allow for the highest currents, while closed (111)-like surfaces exhibit least differences between anodic-and cathodic sweeps, i.e. less prone to poisoning effects. It also appears as if Pt (100) allows for the lowest FAOR onset. 2 Unintuitive, it however appears that Pd (100) or Pd(911) is intrinsically more active. 1,3,5 We would like to note similar studies have been reported for Ir(hkl).
However, FAOR currents are extremely low (potentially due to irreversible surface oxidation), regardless we are confident that pure Ir is a poor FAOR catalyst. 6,7 Expanding on Figure S1 it is important to note what is actually desirable traits for a FAOR catalyst when looking at its CV; FAOR onset is desired to occur at low overpotential while exhibiting high currents (high intrinsic activity and limited poisoning). The anodic and cathodic sweeps should be comparable indicating little to no formation of poisoning species and/or catalyst restructuring. Finally, the electrode should be stable and active at relevant operational potentials, this is easier observed through chronoamperometric (CA) measurements, see Figure   S2. The reason why current generally have been highlighted as the main performance metric is obvious given the main drawback of a DFAFC compared to a direct methanol fuel cell (DMFCs) lies in the energy density, where the overall 8ereaction in a DMFC follows: Conversely, the overall reaction in DFAFC only concerns 2e -: The DFAFC consist to two half reactions at the anode and the cathode, i.e. the formic acid oxidation reaction and the oxygen reduction reaction (ORR), respectively:  (111) or Pd (111) not exhibit high FAOR at potentials above their respective CO oxidation potentials? Such question is strongly implied by Figure S1, and highlighted by the four different potential regions I-IV. Region I denotes the potential range in which FAOR theoretically should be able to occur, see Figure S1. Region II the earliest observed FAOR onset, note the anodic and cathodic sweeps are congruous with one another, suggesting one specific reaction pathway in which poisoning species are formed. Region III designate the region in which both CO oxidation and FAOR takes place, high activity in this potential region may be achieved through partly oxidation of HCOOH to CO followed by CO oxidation toward CO2. Its should be stressed, that any FAOR activity in region III is of no consequence in a real DFAFC as ORR overpotential will make any cell potential negligible and stack cost too high. 20 Finally, Region IV in Figure S1 designates a region in which FAOR is limited either due to formed site blocking species (other than * CO) 21 or the adsorption of other spectator species, such as * OH. We note that high activities in the III-IV potential region in Figure S1 are not relevant for DFAFC applications as significant ORR overpotentials are expected. 22 Moreover, at potentials above 1.1 VRHE Pd and Pt is known to dissolve. 23,24 Having the ability to detect adsorption onset is crucial to identify which adsorbates, spectator species and reaction intermediates (both warranted and unwarranted) are present during FAOR, is crucial as such information will identify the most likely reaction pathways and possible   35,37,38,41 FAOR pathway. It is worth noting that during potential cycling a multitude of possible reaction pathways and possible adsorption events are able to cloud any FAOR trend, e.g. from partial FAOR forming unwarranted COxHy surface species (Purple), 9,10,39 site blockage due to hydroxide adsorption (blue) or formation of CO from direct/indirect disproportionation (red) 40 has been proposed. Even CO2 reduction reaction by cycling too cathodic (brown) 36 have been suggested in literature (reverse CO2 reduction reaction). Note some of the reaction pathways like hydroxyl adsorption and CO-oxidation often takes place at relatively high potentials for Pt and Pd based catalysts. Moreover, while * CO and * COxHy (and * OH/ * H) are the most mentioned poisoned species mentioned in FAOR in literature, recent focus have shifted to formate itself (yellow), which in various arrangements may itself cause self-poisoning of the catalyst. 25,35,42,43 From Figure S3 the all the relevant reaction pathways (and intermediates) have been established.
As we a priori not know the true reaction route(s) and our identification of this/these pathways(s) are limited due to the fact that CO, formate, H and OH adsorption is difficult to detect with traditional FTIR techniques under in situ FAOR conditions [30][31][32][33] . Hence, herein we have opted for the simpler approach of testing catalysts with specific properties and compare these with widely studied reference systems before finally correlating these results with a broad self-consistent theoretical framework. Consequently, it is important to figure out which types of catalyst have been considered most relevant for the field. Given that our framework should encompass the results, obtained for the most studied catalysts in literature. In literature there has generally been four approaches to optimize FAOR performance (note other exists 44,45 ): 1) Use intrinsically active materials such as Pt and Pd (and possible Rh 11 ).
2) Design single-site catalyst; surround some active with inert Au, 46-48 Ag, 49 N/C 50 or other 51-54 as this should disallow the formation of CO during formic acid oxidation. Note, Rh and Ir single-sites have also been reported. [55][56][57][58] 3) Include ad-atoms on Pt or Pd catalyst as elements such as Cd, 59 Sn, 60 Sb, [61][62][63] Bi, 42,[64][65][66][67] Pb, 39 Te, 68 and Tl 45,69 as these elements have been proposed to mitigate both CO and/or formate poisoning. 4) Alloy Pt or Pd, typically with Ru 70,71 as this has been shown to mitigate CO poisoning in DMFC systems. However, a range of other alloys systems bulk 49,72-83 and surface alloys 84 have also been proposed. We note most alloys exhibit similar FAOR performance regardless of Pt's or Pd's alloying elment. 85 Tuning of catalyst shape and -morphologies and even active area have resulted in very conservative improvements in catalysts FAOR performances. A range of fundamental work on different (non-Pd, non-Pt, non-Rh) pure metal catalysts exist Ru, 86 Os, 87 Au. 26,88 However, currents magnitudes and/or onset potential was poor in all instances.
Several FAOR studies concerns the Pd-Au and Pt-Au systems. Recent work by Zhang and coworkers on Pt-Au NP single-site catalyst suggested some FAOR improvement using a novel, albeit not straightforward synthesis. 46 However, looking into the mass activity performance on many of these systems, we noted that all Pt-Au NPs systems' catalytic performances found in literature [89][90][91][92][93] are quite similar to one another; in-fact it seems to be true for all PtxMx-1/PdxMx-1 alloys. 49,72,81-85,73-80 However, we wanted to see if previous results on Pt-Au systems arises from some unknown propensity towards forming Pt/Pd dual-or triple-sites during synthesis formation.
One simple approach to test this is to form PtxMx-1 (M=Au, Ag, Hg and x>3) intermetallic.
Fortunately work by Arnau et al. 94 have reported Hg4Pt catalyst preparation by modifying Pt/C NPs and forming Pt-Hg. Our idea was to use this well-tested catalysts and see if we observe any mass activity improvement relative to the pure seed catalyst.
Other work by Lim and co-wokers 67 (among others) report Pt/C NP modification using Bi. Such model systems seems to exhibit very high activities and are consequently very interesting to us.
Besides the work on various metal-nitrogen-carbon (MNC)and Pt/Pd alloyed catalyst it is worth mentioning that a range of fundamental works exist investigating the effects of electrolyte pH, 25    The deposition of Bi on Pt and the mercury alloying followed the procedure developed by others. 107 We also tried depositing Pb following others work 39 . Unfortunately we found that Pb was unstable even at rather low potential <0.6 VRHE, see Figure S6a. Similarly, we found that cycling Pt-Bi/C with Ohmic drop compensation stripped off Bi when going too high in potential. Hence, the Pt-Bi/C system was only considered stable in the 0.0-0.8 VRHE range, see Figure S6b.    CVs for Pt/C, Pd/C, Pt-Hg/C and Pd-Hg/C catalyst was conditioned and obtained in the potential limits from 0.00 to 1.05 VRHE, whereas for Pt-Bi/C conditioning and CVs was limited to 0.0 to 0.8 VRHE. Following the conditioning the aforementioned CVs were collected at varying can rates, see Figure S8.

Additional electrochemical data
As we needed additional insight to pure metal catalyst for our theoretical framework we tested Ni, Ag and Cu wires as ad hoc test to elucidate whether CO-and H-binding were sufficient descriptors for FAOR activity, see Figure S10.
Due to one of our initial hypothesis, we hoped to observe CO-vibration absence for the Pt-Hg/C relative to the pure Pt/C system during in situ Fourier-transform infrared spectrometry (FTIR).
Unfortunately, we were never able to observe CO-vibration changes in HClO4 while varying the potential. However, in flowing CO directly into the cell and conducting CO-stripping experiments revealed that if changes in CO-coverages were in-fact occurring these would not be observable in HCOOH containing electrolyte, see Figure S10.  The most surprising aspect of the FTIR study of Figure S11 was that contrary to our expectation the CO-stripping peak did not seem to shift significantly with the lanthanide contraction 102 and consequently the CO-binding strength 109 . In regards to the FAOR data the Pt5M data strongly suggests that binding of CO is unlikely the limiting factor for most of the observed FAOR performances presented in within the field, and consequently any theoretical model should take this into account.
From our realization, that disproportionation from * H and HCOOH (or * CHOO) towards * CO and H2O was a real issue in Pt-(and perhaps) Pd-based catalysts, we wondered if we by shifting the HCOOH adsorption beyond the pKa of 3.75 could minimize disproportionation. Hence, CVs in pH 4.2 in formic acid and acetate was attempted, see Figure S12. Figure S12 suggested that tuning pH was unlikely to influence FAOR activity in any meaningful way. This was done by going to pH's above formic acid's pKa of 3.75. Although a change in activity is visible the onset hardly varies. Any effects could simply be due to acetate adsorption.
Or in other words, changing pH appears not to have an effect on the hysteresis, suggesting the poisoning seen as hysteresis is independent on HCOOH's pKa, i.e. facilitated by something that is not HCOOH.
It was generally noted, that getting reproducible results of the FAOR performance was very difficult in lieu of the NP ink system. It seemed that the catalyst surfaces were very dependent on the cleanliness and age of the electrodes on which ink was deposited. This very real dependence on advantageous adsorption and its effect on FAOR have been highlighted by adding HCl into our HClO4/HCOOH electrolyte during CVs, see Figure S13. The devastating effect HCl has on FAOR highlights how certain adsorbates can significantly lower FAOR activities.

Density Functional Theory (DFT) Calculations Details
All calculations were performed using Density Functional Theory (DFT), using the programs ASE version 3.19.0, 110 and GPAW version 19.8.1. 111 Calculations were done at the Generalized Gradient Approximation (GGA) level of theory, using the grid mode, with the BEEF-VdW functional. 112 All calculations are done in vacuum. We utilize a k-point sampling appropriate for the specific structure and a vacuum of minimum 10 Å. All the structures are relaxed to a force below 0.05 eV /Å. As model structure, we use the (111)-fcc facet to represent the metal catalyst, which is a fair choice when analysis is carried out on eV scale (although we note in Figure S1 that (100)-facets may be more active than (111)     Rep. 2017, 7 (1), 1-11. https://doi.org/10.1038/s41598-017-17978-8.