Electrochemical CO2 Reduction: A Classification Problem · 2017-10-26 · Electrochemical CO 2 Reduction: A Classification Problem Alexander Bagger, [a]Wen Ju, [b]Ana Sofia Varela,[c]

Electrochemical CO2 Reduction: A Classification ProblemAlexander Bagger,[a] Wen Ju,[b] Ana Sofia Varela,[c] Peter Strasser,[b] and Jan Rossmeisl*[a]

1. Introduction

In modern society, electrochemical reduction processes tomake sustainable fuels or chemicals, by the use of renewableenergy, will enable the ease of carbon cycling and hereby de-couple the economical growth from CO2 release.[1] To achievethis, the control and understanding of the electrochemical re-duction of carbon–oxygen compounds to selective products isneeded. CO2 reduction on different metal surfaces can be di-vided into three groups of materials, which reduce CO2 to dif-ferent products: 1) the reduction of CO2 to CO, 2) the reduc-tion of CO2 to formic acid (HCOOH) and 3) the reduction of COor carbon-oxygen compounds to hydrocarbons or alcohols.

1) The reduction of CO2 to CO has been extensively studiedon metal catalysts,[2] where Au[3, 4] or Ag[5–7] metal catalystsseem to be the best choice. Recent metal–nitrogen–carbon(M-NC) based catalysts are an alternative;[8, 9] due to struc-tural motifs, they may increase the selectivity of the CO2 re-

duction reaction (CO2RR) over the competitive hydrogenevolution reaction (HER).[10]

2) The reduction of CO2 to formic acid has also been shownto take place on metal catalysts.[2] In particularly Sn[11–13]

and Pb[14] have been studied. The distinction between COor formic acid products has also attracted theoretical con-siderations.[15, 16] The intermediates COOH* and HCOO*(other notations for HCOO* may be *OCHO* or HCO*O*)have been proposed for CO and formic acid products, re-spectively.[17] However, it is not clear from these intermediateswhy CO2 on specific materials reduce to CO or formic acid.

3) The further reduction of CO seems to be mainly constrain-ed, by the ability of the catalyst to bind CO, while nothaving under potential deposited hydrogen (Hupd) andamong the metals only Cu is able to fulfill these uniqueproperties.[10] However, the product distribution from COreduction depends on the Cu catalyst pre-treatment. Whilethe Cu foil catalyst[2] produces mainly methane (CH4) andethylene (C2H4), the main product of a catalyst composedof reduced copper oxides (Cu-OD)[18, 19] is ethanol (C2H5OH)and only a small fraction of methane and ethylene. In addi-tion, different plasma treatments of Cu show tuneability ofthe ethylene to methane ratio.[20] Besides the Cu catalysts,Fe-NC and Mn-NC based catalysts show traces of meth-ane,[8] while all catalysts suffer from HER.

Understanding the difference between single carbon prod-ucts and multiple carbon products has lead to both experi-mental facet studies[21, 22] and theoretical considerations.[23–25]

Combining these efforts, different reaction schemes have beenproposed.[15, 26–28] Understanding the reduction to multiplecarbon products is important in CO2 electroreduction, but thisis not a topic considered here.

In this work, we propose four non-coupled binding energies ofintermediates as descriptors, or “genes”, for predicting theproduct distribution in CO2 electroreduction. Simple reactionscan be understood by the Sabatier principle (catalytic activityvs. one descriptor), while more complex reactions tend to givemultiple very different products and consequently the productselectivity is a more complex property to understand. We ap-proach this, as a logistical classification problem, by groupingmetals according to their major experimental product fromCO2 electroreduction: H2, CO, formic acid and beyond CO* (hy-drocarbons or alcohols). We compare the groups in terms ofmultiple binding energies of intermediates calculated by densi-

ty functional theory. Here, we find three descriptors to explainthe grouping: the adsorption energies of H*, COOH*, and CO*.To further classify products beyond CO*, we carry out formal-dehyde experiments on Cu, Ag, and Au and combine these re-sults with the literature to group and differentiate alcohol orhydrocarbon products. We find that the oxygen binding (ad-sorption energy of CH3O*) is an additional descriptor to explainthe alcohol formation in reduction processes. Finally, the ad-sorption energy of the four intermediates, H*, COOH*, CO*,and CH3O*, can be used to differentiate, group, and explainproducts in electrochemical reduction processes involving CO2,CO, and carbon–oxygen compounds.

[a] A. Bagger, Prof. Dr. J. RossmeislDepartment of Chemistry, University of CopenhagenUniversitetsparken 5, Copenhagen (Denmark)E-mail : [email protected]

[b] W. Ju, Prof. Dr. P. StrasserDepartment of Chemistry, Chemical Engineering DivisionTechnical University Berlin10623 Berlin (Germany)

[c] Dr. A. S. VarelaInstitute of ChemistryNational Autonomous University of MexicoMexico City 04510 (Mexico)

The ORCID identification number(s) for the author(s) of this article canbe found under:https://doi.org/10.1002/cphc.201700736.

An invited contribution to a Special Issue on CO2 Utilisation

ChemPhysChem 2017, 18, 1 – 9 � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim1 &

These are not the final page numbers! ��These are not the final page numbers! ��

ArticlesDOI: 10.1002/cphc.201700736

In this paper, we classify metals for the electrochemical re-duction of CO2. This is to understand what surface propertiesdetermine the main product of CO2 reduction: H2, CO, formicacid and beyond CO*. Where beyond CO* denotes CO reduc-tion into hydrocarbons or alcohols, and the asterisk signifiesthe adsorption of the CO intermediate on the surface. Further-more, we carry out formaldehyde reduction experiments to un-derstand how the beyond CO* products are divided into partlyreduced products (alcohols) or fully reduced products (hydro-carbons). We separate Hori’s[2] experimental data for variousmetals into four groups defined by their major product, assign-ing colours and displaying the faradaic efficiency in Figure 1.These groups are then compared with multiple binding ener-gies of intermediates.

Our hypothesis is that the binding energies of relevant inter-mediates will determine trends in catalytic properties of eachmetal catalyst, and thus give the “genes” for the reduction endproducts. This is under the assumption that the metals are af-fected similarly by an interaction with the electrochemical en-vironment. Further, without knowing the exact reaction path-way the information needed is assumed to be contained in thebinding energies. While, we acknowledge that the CO2RR maytake place on defects, the metal to metal information for aclassification is assumed to be within the binding energies cal-culated for the 111 facet. This will allow us to analyze andmatch the binding energies of intermediates with the productsfrom CO2 and formaldehyde reduction reaction experiments.As a result, we arrive at four weakly coupled intermediateswhich can be used to explain products from the reduction ofCO2, CO and carbon-oxygen compounds. Calculating the fourbinding energies of these intermediates, for new materials,may be used to predict the product distribution of a hypothet-ical new catalyst.

2. CO2 Reduction Descriptors: H2, HCOOH orCO

Many electrochemical reactions have a single product or a fewsimilar products : hydrogen evolution reaction (HER), oxygenevolution reaction (OER) and oxygen reduction reaction (ORR).This allows the use of one key descriptor that defines the activ-ity of the reaction. From the descriptor (simulation/theoretical)and activity (experimental), the Sabatier volcano can be de-fined for the HER[29] and the OER/ORR[30, 31] electrochemical re-actions.

The CO2 reduction reaction is a more complex problem,since it has multiple very different products. Mapping a prod-uct distribution onto one key descriptor, is simply not reasona-ble. Instead of comparing descriptors and activity, we considerthis as a logistic classification problem.

Objectively analyzing the product distribution (faradaic effi-ciency) from the experimental CO2 electroreduction data onmetals from Hori[2] allows us to classify groups in terms of theirmajor product. The metals divide into four groups with themajor products : H2, HCOOH, CO and beyond CO, see Figure 1.We acknowledge that Pd in Hori’s data almost has similar H2

and CO faradaic efficiency, although there is �30 % currentnot accounted for, which could originate from hydrogen ab-sorption. Further, hydrogenated Pd has been shown to pro-duce high amount of formic. However, as our model is themetallic 111 facet of Pd, Pd has been added to the H2 group.Further, Ga does not follow the periodic table product trend,and has by others been proven to produce formic acid inoxide form[32] and are thus omitted in this work. Moreover, wenote that Cd and Zn are metals that have the lowest faradaicefficiency for their main product and Cu is the only metal togo beyond CO. The colors given for the four groups inFigure 1 are applied throughout the work to display the clas-sification.

First, we will distinguish between the groups that produceH2, CO and HCOOH, by calculating relevant intermediates.Known and proposed intermediates for the reaction leading tothese three products are the H* binding energy,[29] the COOH*binding energy[10, 17, 26] and possibly the HCOO*[16, 17] bindingenergy. While HCOO* can be bonded through one or bothoxygen(s), we find that generally the double oxygen bondedintermediate is stabilized by more than 0.5 eV compared tothe single oxygen bonded intermediate. Furthermore theCOOH* generally binds ontop, but can for very reactive metalssit in a bidentate binding configuration, which binds thoughboth the carbon and the oxygen (only Ti metal of our calcula-tions).

To classify the three products by the three intermediates, wecreate a three-dimensional binding-energy plot in Figure 2 aalong with the projections into the DEHCOO* vs. DECOOH* plane(Figure 2 b), the DECOOH* vs. DEH* plane (Figure 2 c) and theDEHCOO* vs. DEH* plane (Figure 2 d). Where Hg for a few bind-ing intermediates have more than two layers fixed, due to anlarge reconstruction of the surface. The structure of each inter-mediate is displayed to the right in Figure 2 a on Cu. In the Fig-ures 2 b–d, the solid black line represent DGH* = 0 (1

2H2(g)$H*),

Figure 1. Major product classification of metal catalysts for electroreductionof CO2 from experiments by Hori,[2] shown in a cropped periodic table withcolors and major product faradaic efficiency. Four groups are identified: H2

(red), formic acid (yellow), CO (blue) and beyond CO* (cyan), which will beapplied throughout the work. As Ga does not follow the periodic table prod-uct trend, the metal has been omitted, and Ti—due to very strong bind-ing—is indicated by arrows in the following classification plots.

ChemPhysChem 2017, 18, 1 – 9 www.chemphyschem.org � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim2&

�� These are not the final page numbers!�� These are not the final page numbers!

Articles

which shows the metals that have under potential depositedhydrogen. The diagonal dashed black lines represent the rela-tive stability of the adsorbates without functional dependenterrors and thermodynamic corrections.[16, 33, 34]

Investigating the three-dimensional binding-energy plot inFigure 2 a, we can see a distinction between the three productgroups in terms of the three binding energies. However, it isnot clear that each of the three descriptors are a classificationdescriptor for the specific product. By projecting the ellipsoidsinto the planes of intermediate binding energies in Figure 2 b–d, we directly observe a problem in using the DEHCOO* andDECOOH* descriptors for formic acid and CO, respectively. In Fig-ure 2 b these two energies are plotted against each other, andthey can not be used to classify (separate) the formic acid orCO product, since Ag, Cu and Zn are grouped together withthe formic acid producing metals. Furthermore, by investigat-

ing DEHCOO* against DEH* in Figure 2 d, we observe that nearlyall metals are below the diagonal. This shows that the bindingenergy of HCOO* is generally stronger than H* (except Au ofthe relevant metals). Discarding how the first proton reducesCO2 into HCOO*, the majority of metal surfaces should eitherbe covered by HCOO* or start producing formic acid prior toHER. However, this has not been experimentally ob-served.[2, 11–14] Therefore, we conclude, that HCOO* is probablynot an important reaction intermediate to distinguish betweenformic acid and CO products.

On the contrary, the DECOOH* binding energy descriptor inFigure 2 c matches both onset potentials for CO2 reduced toCO and formic acid for all metals in those groups, while the H*binding energy descriptor seems to be able to separate thethree product groups. 1) the H2 producing metals which haveunder potential deposited hydrogen (Hupd), 2) the CO produc-

Figure 2. Combined plot of the three proposed descriptors in CO2 electroreduction towards H2, CO and HCOOH (formic acid). In (a) the three-dimensionalspace of descriptors are plotted with the coupled binding-energy ellipsoids that represent the first s confidence interval of the calculations with respect toAu, while the Au error ellipsoid is given with respect to the slab and the gas phase (see computational details). Metals have been given color related to theirclassification group in Figure 1 and each background plane has been giving a color to match the 2d projections of energies and ellipses in (b), (c) and (d) forthe DEHCOO* vs. DECOOH* , DECOOH* vs. DEH* and DEHCOO* vs. DEH* planes, respectively. In (b), (c) and (d) the solid black lines shown the thermodynamic of hy-drogen underpotential deposited (DGH* = 0, for 1

2H2(g)$H*) and the diagonal dashed black lines indicate the relative stability of the two adsorbates withoutfunctional dependent errors and thermodynamics.[16, 33, 34]

ChemPhysChem 2017, 18, 1 – 9 www.chemphyschem.org � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim3 &

These are not the final page numbers! ��These are not the final page numbers! ��

Articles

ing metals which can adsorp H* at the potentials where CO2 isreduced and 3) the formic acid producing metals which havenone or negligible amounts of hydrogen adsorbed at the rele-vant potentials.

This leads us to investigate the difference between the DEH*and DECOOH* intermediate as a function of DEH* (see Fig-ure 3 a). From the left to the right, we can define the three re-gions with respect to the DEH* binding energy representingthe three products. As Zn and Cd are close to the CO or formicacid distinction region, marked by a dashed black line, bothmetals have minor faradaic efficiency for the opposite productHCOOH or CO, respectively. Furthermore, in Figure 3 b, we plotthe CO2 reduction reaction faradaic efficiency as a function ofthe DEH* binding energy. The two figures indicate that reach-ing 100 % faradaic efficiency for CO on metals might be theo-retically impossible. This is supported by the DEH*�DECOOH*values (see Figure 3 a), which are below zero for all CO produc-ing metals, showing that H* can adsorp at potential where CO2

is reduced. On the other hand, formic acid producing metalshave positive DEH*�DECOOH* values, which show that formicacid producing metals have negligible amounts of hydrogen.The positive DEH*�DECOOH* values may be the reason foralmost 100 % faradaic efficiency for the CO2 reduction reactionfor these metals.

3. Beyond CO*: Alcohols or Hydrocarbons

We will now use a similar approach to understand productsbeyond CO* (hydrocarbons or alcohols). Previously, we foundthat two descriptors, DECO* and DEH* , could explain the fur-ther reduction to products beyond CO*.[10] We show this rela-tion with the group colors in Figure 4. Cu is the only metal cat-alyst to give products beyond CO, which is because Cu bindsCO* while not having Hupd. We now want to determine the“genes” of the fully reduced products (hydrocarbons) or thepartially reduced products (alcohols). This will allows us to un-

derstand why the Cu catalysts mainly produces hydrocarbonsin CO2 reduction reaction, while more valuable products couldbe alcohols.

To differentiate between conversion towards alcohols andfully reduced products, we consider a model system of proton-ated intermediates from carbon-oxygen aldehyde compoundssuch as CH2O. Formaldehyde as a carbon-oxygen compound isof interest in understanding methane vs. methanol, and it hasbeen suggested as an intermediate towards methane.[26] Fur-

Figure 3. In (a), the experimental product classification by the DEH*�DECOOH* and the DEH* descriptors are shown. Metals producing H2 can be classified ashaving Hupd. Metals producing CO can be classified as having a coverage of hydrogen, although not Hupd. Metals producing formic acid can be characterizedas having negligible hydrogen coverage, which can be seen by the values of DEH*�DECOOH* being around or above 0 eV. Noting that the CO or formic acidseparation is slightly blurred due to Zn, which produces a small amount of formic acid, and Cd, which produces a small amount of CO (dashed black line plot-ted with equidistance to Zn and Cd). In (b), the CO2 reduction reaction faradaic efficiency as a function of DEH* indicates that the hydrogen binding energy ispotentially the theoretical limiting factor for improving the faradaic efficiency of the CO2 reduction reaction.

Figure 4. The binding energies of the intermediates DECO* and DEH* can beused to separate the Cu metal catalyst into its own group, and hence, ex-plain the beyond CO* group. Where the beyond CO* group bind CO* whilenot having Hupd. The colors match the groups in Figure 1. The ellipsoids rep-resent the coupled first s confidence interval of binding energies with re-spect to Au, while the Au error ellipsoid is given with respect to the slaband the gas phase (see computational details). Furthermore, the black linesshow the thermodynamics of adsorbed or none-adsorbed hydrogen(DGH* = 0, for 1

2 H2$H*) or CO (DGCO* = 0, for CO$CO*)).

ChemPhysChem 2017, 18, 1 – 9 www.chemphyschem.org � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim4&

�� These are not the final page numbers!�� These are not the final page numbers!

Articles

thermore, due to scaling relations,[30, 35] reactions that producehigher order alcohols (ethanol, propanol and ect.) or fully re-duced products (ethane, ethylene, propane and ect.) can prob-ably be described by the descriptor found here. The thermody-namics of formaldehyde reduction reaction to either methane(hydrocarbon) or methanol (alcohol) are given as [Eqs. (1) and(2)]:

CH2Oþ 2ðHþ þ e�Þ ! CH3OH URHE ¼ 0:28 V pr: ðHþ þ e�Þð1Þ

CH2Oþ 4ðHþ þ e�Þ ! CH4 þ H2O URHE ¼ 0:58 V pr: ðHþ þ e�Þð2Þ

Where the equations are calculated at standard thermody-namic conditions. It shows that formaldehyde reduction toeither alcohol or hydrocarbon are thermodynamically favoredreactions at URHE = 0 V and the thermodynamics toward hydro-carbons is more favourable than alcohols.

Considering the first protonation step of formaldehyde, twodifferent activated intermediates can be formed: the CH2OH*or the CH3O*, which are bonding through the carbon or theoxygen atom, respectively (see structures to the right of Fig-ure 5 a). As a hypothesis, we compare the oxygen bonded in-termediate and the carbon bonded intermediate difference asa descriptor for alcohol or hydrocarbon products. Thus, we cal-culate the intermediate binding energies for formaldehyde re-duction on relevant metals in Figure 5 a. The diagonal indicates

Figure 5. In (a), the DFT energy relation between the DECH2 OH* and DECH3 O* descriptors as a measure for methanol or methane product from carbon–oxygen-compound reduction is shown. In (b) and (c), the experiments carried out on formaldehyde reduction at different potentials and pH 7 are shown for methanoland methane, respectively. Methanol formation on Cu, Ag and Au metal catalysts exhibits a potential dependency, while the constant methane signal isaround two orders of magnitude lower.

ChemPhysChem 2017, 18, 1 – 9 www.chemphyschem.org � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim5 &

These are not the final page numbers! ��These are not the final page numbers! ��

Articles

the stability for the two adsorbates without any thermodynam-ics or water corrections. The metals above or around the diag-onal will bond through oxygen and give alcohols as an endproduct, while the metals below tend to bind through carbonand give the fully reduced hydrocarbons.

This hypothesis requires validation. In literature [36–38] , thefollowing metals have been shown to reduce formaldehyde tomethanol: Hg, Pb, In and Sn. These reference experiments sup-port the hypothesis, as Hg, Pd, In and Sn are above the diago-nal in Figure 5 a. However, as Cu, Ag and Au are metals of highinterest in CO2 electroreduction, we perform formaldehyde ex-periments on these metals.

The electrochemical reduction experiments were carried outin neutral and buffer potassium phosphate solution with 1 mM

CH2O as the reactant, see Figure 5 b, c. Gaseous and liquidphase products (H2, CH4 and CH3OH) were analyzed in gas-online-GC and liquid-GC, respectively. During the bulk electrol-ysis, faradaic efficiency towards HER is above 98 % (not shown)for these metals. We note that in CO2 reduction this is muchlower and the result of the low faradaic efficiency for the form-aldehyde reduction could be a difficult steric situation. Despitethe limited faradaic efficiency for the CH2O reduction reaction,the carbon based products, methane and methanol, wereclearly detected on these metallic foils. From control experi-ments for the three metals in our setup at URHE =�1.0 V for aCO2 saturated solution, we do not find any CH3OH production.By comparing Figure 5 b,c, the methane formation is found tobe constantly two orders of magnitude lower than the metha-nol formation and also seems to be potential independent.Whereas the methanol formation is potential dependent andfollows the order of the CH3O* intermediate binding energyfor Cu, Ag and Au. However, the absolute onset potential forthe reaction is shifted negatively as compared to the bindingenergy, which could be a result of the hydrogen coverage,[39]

functional dependent errors[33] or other. We also see that Au,which is on the limit between the CH3O* or CH2OH* intermedi-ate energy, produces slightly more methane than Cu or Ag.Carrying out experiments on Ir, Pd or Pt metals, which shouldhave higher methane turnover, is virtually impossible due toexcessive HER. The excessive HER can already be read from theanalysis combined in Figure 1 and Figure 2. Furthermore, it ex-plains why in the literature formaldehyde experiments areavailable only for the Hg, Pb, In and Sn metals. Since thesemetals binds through oxygen via the CH3O* intermediate theydo not produce any H2, due to the very weak hydrogen bind-ing energy.

The formaldehyde reduction experiments and descriptorsshow that the proposed formaldehyde intermediate[26] can notbe an intermediate on the path to methane in CO2 or CO re-duction, which agrees with previous kinetic studies.[27] Extend-ing this result by the scaling relation to the reduction of otheraldehydes (or carbon-oxygen compounds) concludes that alco-hol products observed in CO2 reduction is a result of an inter-mediate which has been bonded through oxygen at somepoint. This is also consistent with formaldehyde, glyoxal andglycol-aldehyde reduction experiments on copper[40] and theobservation of acetaldehyd as an intermediate for the high

ethanol faradaic efficiency of the Cu-OD catalyst.[39, 41] Further-more, it shows that weak binding metals allow the CH2O* in-termediate to detach and approach the surface again with theoxygen first.[42] Strong binding metals keep the carbon at thesurface, which allows all oxygens to get reduced. This is thegeneral case for Cu catalysts[2, 20] and weak binding metals asAu can lose their carbon binding and produce CH3OH.[43]

4. Conclusions and Outlook

We have calculated key binding energies for non-coupled in-termediates to identify the “genes” of CO2 electroreductionproducts. We have grouped the results from CO2 and formal-dehyde electrochemical reduction experiments using key de-scriptors.

We have found that the binding energies of DECOOH* andDEH* can explain the H2, CO or formic acid products in CO2 re-duction. Considering only the hydrogen binding energy of themetal as the important descriptor, allows for the classificationinto three groups. Metals with Hupd forming mainly H2, metalswith H* at the CO2 reduction potential forming mainly CO, andmetals with very little or no H* at the CO2 reduction potentialsforming formic acid.

For products beyond CO*, we consider the DECO* and DEH*binding energy, which separates the Cu metal from all othermetals calculated here by not having Hupd while binding CO*.Beyond CO*, the Cu metal catalyst can fully reduce CO’s to hy-drocarbon products or partly reduced alcohol products. Wehave investigated the carbon binding (DECH2OH* ) and theoxygen binding (DECH3 O* ) intermediates and related these toformaldehyde reduction experiments to distinguish betweenhydrocarbon or alcohol formation. From the experiments wehave found a strong correlation between the alcohol formationand the preferred oxygen binding for the catalysts tested hereand found in literature.

As a final conclusion we have found four non-coupled inter-mediates: DEH* , DECOOH* , DECO* and DECH3 O* , that can be usedto explain product groups and selectivity distributions in elec-trochemical CO2 reduction and possibly also other reductionprocesses.

Computational Details

The metal structures are composed of a 111 surface slab FCC 3 �3 � 4 unit cell, with fixed bottom layers. The electronic calculationsare carried out at the Generalized Gradient Approximation DensityFunctional Theory (GGA-DFT) level with the projector augmentedwave method together with the BEEF-vdW functional[44, 45] as imple-mented in the GPAW software.[46, 47] We apply a (4 � 4 � 1) k-pointsampling, a grid-spacing of 0.18, a vacuum of minimum 10 �, thedipole correction and all the structures are relaxed to a forcebelow 0.05 eV ��1.

To estimate errors and measure scalings, we calculated the ensem-bles from the Bayesian Error Estimation Functional (BEEF-vdW).[44, 45]

However, to plot and obtain the covariance ellipsoids (three-di-mensional plots) and ellipses (two-dimensional plots), the bindingenergy ensembles were referenced to Au by [Eq. (3)]:[10]

ChemPhysChem 2017, 18, 1 – 9 www.chemphyschem.org � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim6&

�� These are not the final page numbers!�� These are not the final page numbers!

Articles

DEA* ¼ EMetal;A* � EMetal � ðEAu;A* � EAuÞ þMean DEGasphaseAu;A*

� �ð3Þ

where A* is the adsorbate, EMetal;A* is the ensemble consisting of2000 binding energy values on the Metal, EMetal is the metal ensem-ble energies, EAu;A* � EAu

� �is the chosen Au reference ensemble

and Mean DEGasphaseAu;A*

� �shifts the binding energy ensemble with the

constant binding energy of Au. The gas phase binding energy en-semble of Au is calculated as [Eq. (4)]:

DEGasphaseAu;A* ¼ EAu;A* � EAu � EGasphase

A ð4Þ

where EGasphaseA is the vacuum calculation of the adsorbate(s). In this

work the CO2, CO, H2 and CH2O gas phase references has beenused. The ensembles, DEA* and DEGasphase

Au;A* , can be used to give ascatter plot with 2000 binding energy values. However, we con-struct the covariance ellipsoids and ellipses as follows:

Mathematically, two ensemble energies, ie. DEMetal;A* and DEMetal;B* ,can be turned into error ellipses with a first s confidence intervals(notation has been shortened from now on). First the covariancematrix is defined by Equation (5):

covðDEX;DEYÞ ¼s2

X sYsX

sXsY s2Y

!ð5Þ

where DEX are the ensemble binding energies and sX are the var-iances calculated from the ensemble. The covariance matrix can beused to illustrate two ensembles simultaneously from the eigenval-ues and eigenvectors found by singular value decomposition ofthe covariance matrix [Eq. (6)]:

covðDEX;DEYÞ ¼ USU* ; S ¼a 0

0 b

!; U ¼

x1 y1

x2 y2

!ð6Þ

where S contains the eigenvalues in the diagonal and U the eigen-vectors. The eigenvectors represent the orthogonal orientationwhich maximizes the description of correlation and the eigenval-ues the size (or weight) in each direction. However, the eigenval-ues must be normalized to fit the ellipse equation (x2

a2sþ y2

b2s¼ 1) and

represent the first s confidence intervals by [Eq. (7)]:

as ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffia*c2ðfractile; df Þ

p; a 2 a;b ð7Þ

where as is the normalized eigenvalue to the first s confidence in-terval and c2 is the chi-squared probability density function with dfdegrees of freedom.

For the plots where free-energy lines are drawn together with DFTenergies, we used the zero point, entropy and heat capacity valuesfrom Chan et al.[34] Furthermore, we corrected the H2 energy by0.09 eV, due to functional errors[16, 33] and additionally, we applied awater correction for the binding energy of CO* of 0.1 eV.[26] Thesecorrections were only used for the free energy lines.

All structures with total energies, ensembles and plottingmethod are available on the webpage: http://nano.ku.dk/eng-lish/research/theoretical-electrocatalysis/katladb/.

Electrochemical Measurements and ProductsAnalysis

The electrochemical measurements were controlled with an EC-LabSP-300 Potentiostat. The resistance between reference electrodeand working electrode was measured by using potential electro-chemical impedance spectroscopy (PEIS). 50 % of the resistancewas corrected by the software and the rest was manually correct-ed. CH2O electrolysis was carried out in a custom-made two-com-partment cell, in which the working electrode was separated fromthe counter electrode by Nafion membrane (NR 212) to hinder there-oxidation of the products on the counter electrode. The glass-ware was cleaned in aqua regia and afterwards in concentratedHNO3 for 1 h, respectively, rinsed and sonicated with 80 8C DI-water several times, and dried at 60 8C in an oven. The Nafionmembrane was rinsed in 5 % H2O2 solution and activated in 10 %H2SO4 at 90 8C for 2 hours. After rinsing with DI-water into neutral,the membrane was kept in the fresh electrolyte. Each compart-ment of the cell was filled with 40 mL N2 purged electrolyte. A Ptmesh 100 (Sigma–Aldrich 99.9 %) was used as counter electrode(CE) and a leak-free Ag/AgCl electrode (Hugo Sachs Elektronik Har-vard apparatus GmbH) was used as the reference electrode. Theelectrolyte was prepared with 0.05 m KH2PO4 + 0.05 m K2HPO4 (Hon-eywell, pH of 7). The metal foil (Cu, Ag, Au, purity above 99.998 %,Alfa Aesar) was mechanically polished with 100 nm diameter dia-mond suspension, then sonicated in DI-water/aceton/DI-water,each for 2 min. At 15 min, 60 min, 120 min, 180 min, 240 min ofbulk electrolysis at constant working potential, a sample of the gaswas analyzed by gas chromatography (Shimadzu GC 2014, Haye.-Sep Q (Col-No. CS 1015-03) + Haye.Sep R (Col-No. CS 1015-07),TCD and FID detector) to quantify the instant Production Rate andFaradaic Selectivity of the gaseous products. Additionally, 0.5 mL ofthe electrolyte during the reaction was sampled by liquid sampler(Custom made @Duratec) and then analyzed by high performanceliquid chromatograph (HPLC Agilent 1200, Zimmer Chromatogra-phy Column, RID detector) and liquid injection gas chromatogra-phy (Shimadzu GC 2010 plus, OptimaWax capillary, FID Detector)for liquid/solvated products.

Acknowledgements

We acknowledge discussions with Daniel Heestermans Svendsento treat our ensembles statistically. This work was supported byClimate-KIC under the EnCO2re project, the Carlsberg Foundation(grant CF15-0165) and the Innovation Fund Denmark (grand so-lution ProActivE 5124-00003A). P.S. acknowledges financial sup-port by the German Federal Ministry for Education and Research(BMBF) under project “CO2EKAT” 03SF0523.

Conflict of interest

The authors declare no conflict of interest.

Keywords: classification · CO2 reduction · electrochemistry ·formaldehyde reduction · scaling relation

[1] B. Obama, Science 2017, DOI : 10.1126/science.aam6284.[2] Y. Hori, A. Murata, R. Takahashi, J. Chem. Soc. Faraday Trans. 1989, 85,

2309 – 2326.

ChemPhysChem 2017, 18, 1 – 9 www.chemphyschem.org � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim7 &

These are not the final page numbers! ��These are not the final page numbers! ��

Articles

[3] Y. Chen, C. W. Li, M. W. Kanan, J. Am. Chem. Soc. 2012, 134, 19969 –19972.

[4] H. Mistry, R. Reske, Z. Zeng, Z.-J. Zhao, J. Greeley, P. Strasser, B. R.Cuenya, J. Am. Chem. Soc. 2014, 136, 16473 – 16476.

[5] Q. Lu, J. Rosen, F. Jiao, ChemCatChem 2015, 7, 38 – 47.[6] M. Ma, B. J. Trzesniewski, J. Xie, W. A. Smith, Angew. Chem. Int. Ed. 2016,

55, 9748 – 9752; Angew. Chem. 2016, 128, 9900 – 9904.[7] H. Mistry, Y.-W. Choi, A. Bagger, F. Scholten, C. Bonifacio, I. Sinev, N. J.

Divins, I. Zegkinoglou, H. S. Jeon, K. Kisslinger, E. A. Stach, J. C. Yang, J.Rossmeisl, B. R. Cuenya, Angew. Chem. Int. Ed. 2017, DOI: 10.1002/ange.201704613

[8] A. S. Varela, R. S. Nastaran, J. Steinberg, W. Ju, H.-S. Oh, P. Strasser,Angew. Chem. Int. Ed. 2015, 54, 10758 – 10762; Angew. Chem. 2015, 127,10908 – 10912.

[9] V. Tripkovic, M. Vanin, M. Karamad, M. E. Bjçrketun, K. W. Jacobsen, K. S.Thygesen, J. Rossmeisl, J. Phys. Chem. C 2013, 117, 9187 – 9195.

[10] A. Bagger, W. Ju, A. S. Varela, P. Strasser, J. Rossmeisl, Catal. Today 2017,288, 74 – 78.

[11] Y. Chen, M. W. Kanan, J. Am. Chem. Soc. 2012, 134, 1986 – 1989.[12] M. F. Baruch, J. E. Pander, J. L. White, A. B. Bocarsly, ACS Catal. 2015, 5,

3148 – 3156.[13] W. Luc, C. Collins, S. Wang, H. Xin, K. He, Y. Kang, F. Jiao, J. Am. Chem.

Soc. 2017, 139, 1885 – 1893.[14] C. H. Lee, M. W. Kanan, ACS Catal. 2015, 5, 465 – 469.[15] R. Kortlever, J. Shen, K. J. P. Schouten, F. Calle-Vallejo, M. T. M. Koper,

Catal. J. Phys. Chem. Lett. 2015, 6, 4073 – 4082.[16] J. S. Yoo, R. Christensen, T. Vegge, J. K. Nørskov, F. Studt, ChemSusChem

2016, 9, 358 – 363.[17] J. T. Feaster, C. Shi, E. R. Cave, T. Hatsukade, D. N. Abram, K. P. Kuhl, C.

Hahn, J. K. Nrskov, T. F. Jaramillo, ACS Catal. 2017, 7, 4822 – 4827.[18] C. W. Li, J. Ciston, M. W. Kanan, Nature 2014, 508, 504 – 507.[19] A. Verdaguer-Casadevall, C. W. Li, T. P. Johansson, S. B. Scott, J. T.

McKeown, M. Kumar, I. E. L. Stephens, M. W. Kanan, I. Chorkendorff, J.Am. Chem. Soc. 2015, 137, 9808 – 9811.

[20] H. Mistry, A. S. Varela, C. S. Bonifacio, I. Zegkinoglou, I. Sinev, Y.-W. Choi,K. Kisslinger, E. A. Stach, J. C. Yang, P. Strasser, B. R. Cuenya, Nat.Commun. 2016, 7, 12123.

[21] K. J. P. Schouten, Z. Qin, E. P. Gallent, M. T. M. Koper, J. Am. Chem. Soc.2012, 134, 9864 – 9867.

[22] K. J. P. Schouten, E. P. Gallent, M. T. M. Koper, ACS Catal. 2013, 3, 1292 –1295.

[23] F. Calle-Vallejo, M. T. M. Koper, Angew. Chem. Int. Ed. 2013, 52, 7282 –7285; Angew. Chem. 2013, 125, 7423 – 7426.

[24] J. H. Montoya, C. Shi, K. Chan, J. K. Nørskov, J. Phys. Chem. Lett. 2015, 6,2032 – 2037.

[25] R. B. Sandberg, J. H. Montoya, K. Chan, J. K. Nørskov, Surf. Sci. 2016, 654,56 – 62.

[26] A. A. Peterson, F. Abild-Pedersen, F. Studt, J. Rossmeisl, J. K. Nørskov,Energy Environ. Sci. 2010, 3, 1311 – 1315.

[27] X. Nie, M. R. Esopi, M. J. Janik, A. Asthagiri, Angew. Chem. Int. Ed. 2013,52, 2459 – 2462; Angew. Chem. 2013, 125, 2519 – 2522.

[28] R. Kas, R. Kortlever, H. Ylmaz, M. T. M. Koper, G. Mul, ChemElectroChem2015, 2, 354 – 358.

[29] J. K. Nørskov, T. Bligaard, A. Logadottir, J. R. Kitchin, J. G. Chen, S. Pande-lov, U. Stimming, J. Electrochem. Soc. 2005, 152, J23 – J26.

[30] I. C. Man, H.-Y. Su, F. Calle-Vallejo, H. A. Hansen, J. I. Martnez, N. G.Inoglu, J. Kitchin, T. F. Jaramillo, J. K. Nørskov, J. Rossmeisl, ChemCat-Chem 2011, 3, 1159 – 1165.

[31] M. Busch, N. B. Halck, U. I. Kramm, S. Siahrostami, P. Krtil, J. Rossmeisl,Nano Energy 2016, 29, 126 – 135.

[32] T. Sekimoto, M. Deguchi, S. Yotsuhashi, Y. Yamada, T. Masui, A. Kurama-ta, S. Yamakoshi, Electrochem. Commun. 2014, 43, 95 – 97.

[33] R. Christensen, H. A. Hansen, T. Vegge, Catal. Sci. Technol. 2015, 5,4946 – 4949.

[34] K. Chan, C. Tsai, H. A. Hansen, J. K. Nørskov, ChemCatChem 2014, 6,1899 – 1905.

[35] E. M. Fern�ndez, P. G. Moses, A. Toftelund, H. A. Hansen, J. I. Mart�nez, F.Abild-Pedersen, J. Kleis, B. Hinnemann, J. Rossmeisl, T. Bligaard, J. K.Nørskov, Angew. Chem. Int. Ed. 2008, 47, 4683 – 4686; Angew. Chem.2008, 120, 4761 – 4764.

[36] S. Clarke, J. A. Harrison, J. Electroanal. Chem. Interfacial Electrochem.1972, 36, 109 – 115.

[37] P. G. Russell, N. Kovac, S. Srinivasan, M. Steinberg, J. Electrochem. Soc.1977, 124, 1329 – 1338.

[38] S. Kapusta, N. Hackerman, J. Electroanal. Chem. Interfacial Electrochem.1982, 138, 295 – 313.

[39] I. Ledezma-Yanez, E. P. Gallent, M. T. M. Koper, F. Calle-Vallejo, Catal.Today 2016, 262, 90 – 94.

[40] K. J. P. Schouten, Y. Kwon, C. J. M. van der Ham, Z. Qin, M. T. M. Koper,Chem. Sci. 2011, 2, 1902 – 1909.

[41] E. Bertheussen, A. Verdaguer-Casadevall, D. Ravasio, J. H. Montoya, D. B.Trimarco, C. Roy, S. Meier, J. Wendland, J. K. Nørskov, I. E. L. Stephens, I.Chorkendorff, Angew. Chem. 2016, 128, 1472 – 1476.

[42] K. P. Kuhl, T. Hatsukade, E. R. Cave, D. N. Abram, J. Kibsgaard, T. F. Jara-millo, J. Am. Chem. Soc. 2014, 136, 14107 – 14113.

[43] E. R. Cave, J. H. Montoya, K. P. Kuhl, D. N. Abram, T. Hatsukade, C. Shi, C.Hahn, J. K. Nørskov, T. F. Jaramillo, Phys. Chem. Chem. Phys. 2017, 19,15856 – 15863.

[44] J. Wellendorff, K. T. Lundgaard, A. Møgelhøj, V. Petzold, D. D. Landis,J. K. Nørskov, T. Bligaard, K. W. Jacobsen, Phys. Rev. B 2012, 85, 235149.

[45] A. J. Medford, J. Wellendorff, A. Vojvodic, F. Studt, F. Abild-Pedersen,K. W. Jacobsen, T. Bligaard, J. K. Nørskov, Science 2014, 345, 197 – 200.

[46] J. J. Mortensen, L. B. Hansen, K. W. Jacobsen, Phys. Rev. B 2005, 71,035109.

[47] J. Enkovaara, C. Rostgaard, J. J. Mortensen, J. Chen, M. Dujak, L. Ferrighi,J. Gavnholt, C. Glinsvad, V. Haikola, H. A. Hansen, H. H. Kristoffersen, M.Kuisma, A. H. Larsen, L. Lehtovaara, M. Ljungberg, O. Lopez-Acevedo,P. G. Moses, J. Ojanen, T. Olsen, V. Petzold, N. A. Romero, J. Stausholm-Møller, M. Strange, G. A. Tritsaris, M. Vanin, M. Walter, B. Hammer, H.H�kkinen, G. K. H. Madsen, R. M. Nieminen, J. K. Nørskov, M. Puska, T. T.Rantala, J. Schiøtz, K. S. Thygesen, K. W. Jacobsen, J. Phys. Condens.Matter 2010, 22, 253202.

Manuscript received: July 1, 2017

Revised manuscript received: August 16, 2017Accepted manuscript online: September 5, 2017Version of record online: && &&, 0000

ChemPhysChem 2017, 18, 1 – 9 www.chemphyschem.org � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim8&

�� These are not the final page numbers!�� These are not the final page numbers!

Articles

ARTICLES

A. Bagger, W. Ju, A. S. Varela, P. Strasser,J. Rossmeisl*

&& –&&

Electrochemical CO2 Reduction: AClassification Problem

CO2 reduction: A product classificationof metal catalysts for the electroreduc-tion of CO2 is shown in a periodic tablewith colors and compared with thethree-dimensional space for the pro-

posed descriptors towards H2, CO, andHCOOH (formic acid). The color repre-sentation in the three-dimensional plotallows one to group and classify eachmajor product with descriptors.

ChemPhysChem 2017, 18, 1 – 9 www.chemphyschem.org � 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim9 &

These are not the final page numbers! ��These are not the final page numbers! ��