Active sites and mechanisms for H2O2 decompositionover Pd catalystsAnthony Plaucka, Eric E. Stanglandb, James A. Dumesica,1, and Manos Mavrikakisa,1

aDepartment of Chemical and Biological Engineering, University of Wisconsin, Madison, WI 53706; and bCore R&D–Inorganic Materials and HeterogeneousCatalysis, The Dow Chemical Company, Midland, MI 48674

Contributed by James A. Dumesic, February 24, 2016 (sent for review August 18, 2015; reviewed by Alexis T. Bell and Jens K. Nørskov)

A combination of periodic, self-consistent density functional theory(DFT-GGA-PW91) calculations, reaction kinetics experiments on aSiO2-supported Pd catalyst, and mean-field microkinetic modelingare used to probe key aspects of H2O2 decomposition on Pd in theabsence of cofeeding H2. We conclude that both Pd(111) andOH-partially covered Pd(100) surfaces represent the nature of theactive site for H2O2 decomposition on the supported Pd catalystreasonably well. Furthermore, all reaction flux in the closed catalyticcycle is predicted to flow through an O–O bond scission step in eitherH2O2 or OOH, followed by rapid H-transfer steps to produce theH2O and O2 products. The barrier for O–O bond scission is sensitiveto Pd surface structure and is concluded to be the central param-eter governing H2O2 decomposition activity.

catalysis | density functional theory | hydrogen peroxide | palladium |microkinetic analysis

Hydrogen peroxide (H2O2) is a desirable oxidant because theonly by-product from its reduction is water (1). The largest

demand for H2O2 is in the pulp and paper industry (2). In addition,there are applications for H2O2 in catalytic oxidations (3) such as theepoxidation of propene to propylene oxide (4). However, the currentproduction process for H2O2, the anthraquinone process, is complex,energy intensive, and only economic for large-scale productions (5).The direct synthesis of hydrogen peroxide (DSHP) from H2

and O2 is an appealing alternative that has the potential to enablesmall-scale integrated production of H2O2 (5). A commercialDSHP process has not yet been implemented, although Degussa/Headwaters announced the successful integration of a pilot plantfor the DSHP and a pilot plant for the manufacture of propyleneoxide in 2005 (6). García-Serna et al. (7) have discussed theeconomic viability of a DSHP process in comparison with theexisting industrial-scale anthraquinone process.The primary catalytic challenge for the DSHP is identifying cat-

alysts that can maintain high selectivity for H2O2 at industrially rel-evant H2O2 concentrations. The complete reduction of O2 to H2O ismore thermodynamically favorable than the partial reduction of O2

to H2O2:

H2ðgÞ+O2ðgÞ→H2O2ð1Þ ΔG298K° =−120.4  kJ=mol,[Reaction 1�;

H2ðgÞ+O2ðgÞ→H2Oð1Þ+ 1/ 2O2ðgÞ ΔG298K° =−237.1  kJ=mol,[Reaction 2�:

Consequently, H2O2 decomposition,

H2O2ð1Þ→H2Oð1Þ+ 1/ 2O2ðgÞ ΔG298K° =−116.7  kJ=mol,[Reaction 3�;

and H2O2 hydrogenation by H2,

H2O2ð1Þ+H2ðgÞ→ 2H2Oð1Þ ΔG298K° =−353.8  kJ=mol,[Reaction 4�;

are also thermodynamically favorable reactions. The optimalDSHP catalyst must therefore selectively produce H2O2 at highrates and preserve H2O2 from decomposition.Pd is widely recognized as the most effective transition metal

for the DSHP; however, Pd generally exhibits poor selectivity inthe absence of promoters and is highly active for the H2O2 de-composition reactions. Experimental strategies to improve se-lectivity of Pd-based catalysts include the following (8, 9): addingstrongly coordinating anions (e.g., CN−, Cl−, Br−) and acids tothe solvent; alloying Pd with other noble metals, namely Au orPt; and controlling the nature of the catalyst support material.The performance of Pd-based catalysts has also been shown tostrongly depend on the oxidation state of Pd (i.e., catalyst pre-treatment conditions, O2:H2 feed ratio, etc.) (8).A common goal in many of these modifications is the minimi-

zation of the H2O2 decomposition reactions (reactions 3 and 4)(10). Importantly, H2O2 has been identified as the primaryproduct on Au-Pd catalysts at low H2 conversion (11, 12), whereasthe subsequent H2O2 decomposition reactions decrease overallyield as the reaction progresses. Experiments also suggest that theactive sites for H2O2 synthesis (reaction 1) and decomposition onPd-based catalysts may be different (12, 13). In particular, Hutchingsand coworkers (14) demonstrated that the H2O2 decompositionreactions on a Au-Pd/C catalyst could be suppressed by pretreatingthe carbon support with HNO3, and this resulted in a stable catalystwith >95% selectivity for H2O2 during the DSHP reaction. A de-tailed understanding of the active site(s) and elementary reactionmechanisms for these undesired reactions would benefit the identi-fication of improved catalysts. Nonetheless, there is still much workto be performed to elucidate the optimal structure and composition

Significance

The use of hydrogen peroxide (H2O2) for catalytic oxidations islimited by the energy-intensive and wasteful process by whichH2O2 is currently produced—the anthraquinone process. Thedirect synthesis of H2O2 (DSHP) is a promising alternative pro-cess, yet catalysts active for this reaction (Pd being the mostwidely studied) are generally hindered by subsequent H2O2 de-composition. Through a combined theoretical and experimentalapproach, our work (i) provides an understanding of the natureof Pd active sites responsible for H2O2 decomposition and (ii)identifies a single type of elementary step that controls the rate.These structural and mechanistic insights are important for de-signing improved DSHP catalysts and for developing transition-metal–catalyzed oxidations that efficiently use H2O2.

Author contributions: A.P., E.E.S., J.A.D., and M.M. designed research; A.P. performedresearch; A.P., J.A.D., and M.M. analyzed data; and A.P., E.E.S., J.A.D., and M.M. wrotethe paper.

Reviewers: A.T.B., University of California, Berkeley; and J.K.N., Stanford University.

The authors declare no conflict of interest.1To whom correspondence may be addressed. Email: jdumesic@wisc.edu or emavrikakis@wisc.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1602172113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1602172113 PNAS | Published online March 22, 2016 | E1973–E1982

CHEM

ISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Mar

ch 1

4, 2

020

of Pd-based catalysts for the DSHP; some recent important contri-butions on this front can be found in refs. 11 and 15–18.This paper will focus on the mechanism and the nature of the

dominant active site(s) responsible for H2O2 decomposition(reaction 3, with no H2 present) under conditions relevant to theDSHP process. We highlight key factors that will aid in theidentification of improved DSHP catalysts that exhibit minimalH2O2 decomposition activity. These findings are also relevant totransition-metal–catalyzed oxidation reactions that use H2O2—

whether produced in situ or fed as reactant—but whose effi-ciency may be limited by catalytic H2O2 decomposition (19–22).

Materials and MethodsDensity Functional Theory. Periodic Pd(111) and Pd(100) slabs were chosen asrepresentative models for the planar surfaces of the supported Pd nano-particles used in the experiments. Pd(111) and Pd(100) have the lowestsurface free energies among the clean Pd facets, and a truncated octa-hedron minimizes the total surface free energy for Pd particles >3–5 nmbased on a Wulff construction (23, 24). Therefore, in the absence of strongparticle-support interactions, both of these Pd surfaces are expected to bein high abundance.

All density functional theory (DFT) calculations were performed using theDACAPO total energy code (25, 26), using the self-consistent PW91 gener-alized gradient approximation (GGA-PW91) (27, 28) to describe the ex-change correlation energy and potential, and ultrasoft pseudopotentials(29) to describe the ionic cores. Electron density was determined by iterativediagonalization of the Kohn–Sham Hamiltonian, Fermi population of theKohn–Sham states (kBT = 0.1 eV), and Pulay mixing of the resulting electrondensity (30). The total energy was then extrapolated to kBT = 0 eV. TheKohn–Sham one-electron valence states were expanded using a plane wavebasis with kinetic energy below 25 Ry.

The (111) and (100) metal surfaces were modeled using a slab geometrywith a periodically repeated (2 × 2) unit cell and four atomic layers; thiscorresponds to 1/4 monolayer (ML) coverage of a single adsorbate placed inthe unit cell. The surface Brillouin zone was sampled using 18 special Chadi–Cohen (31) k-points for the (111) slabs, and a (6 × 6 × 1) Monkhorst–Pack (32)k-point mesh for the (100) slabs. All slab layers were fixed for calculations onthe (111) slabs [previous calculations show minimal effect of surface re-laxation on the calculated energetics for a similar system (33)], whereas thetop two metal layers were allowed to relax in the (100) slabs. A distance of14 Å of vacuum separated successive slabs in the z direction, and adsorptionwas only permitted on one of the two available surfaces with the electro-static potential adjusted accordingly (34, 35). The equilibrium PW91 bulk Pdlattice constant has been calculated previously (33) to be 3.99 Å [experi-mental value is 3.89 Å (36)]. Calculations involving O2 were performed spin-polarized.

Binding energies (BEs) are reported based on the total energy of the metalslab with the adsorbate on it (Eads) with respect to the total energies of thefree gas-phase adsorbate (Egas) and the clean slab (Eclean). All reported DFTresults have been corrected for the zero-point energy (ZPE). The minimumenergy paths for elementary steps were calculated using the climbing im-age–nudged elastic band method (37, 38) with at least seven intermediateimages, and the transition state was verified by identification of a singleimaginary frequency along the reaction coordinate. Vibrational frequencieswere calculated by diagonalization of the mass-weighted Hessian matrixand using the harmonic oscillator approximation (39).

Experiments. A 0.09 wt% Pd/spSiO2 (spSiO2 denotes the spherical silica sup-port) catalyst was prepared for reaction kinetics experiments. The spSiO2 wassynthesized by a modified Stöber process (40) described in a previous pub-lication (41), resulting in spherical silica particles (∼100–200 nm in diameter)with no internal pore structure and a Brunauer–Emmett–Teller surface areaof 21 m2/g (41). The Pd was loaded onto the spSiO2 by vacuum evaporativeimpregnation (in a rotary evaporator) using a solution of Pd(II) acetate dis-solved in dichloromethane. The dried material was reduced in a quartz cellfollowing a procedure described in ref. 42 to promote formation of large Pdparticles (average particle size of 5.6 ± 2.4 nm determined from scanningtransmission electron microscopy images of the Pd/spSiO2 catalyst) thatbetter compare with the Pd(111) and Pd(100) DFT models. This heat treat-ment procedure involved a temperature ramp to 673 K (10 K/min) and 3-hhold at 673 K in flowing H2 (30 mL/min), followed by cooling to roomtemperature under flow of Ar (30 mL/min) and passivation with 1% O2 in Ar.The Pd surface site density was determined by irreversible CO uptake ex-

periments at 300 K, using an apparatus and procedure described previously(43), and applying a surface stoichiometry of 2:3 for CO:surface Pd atom (42).

H2O2 decomposition experiments were performed in a 50-mL Parr In-strument Company Hastelloy C-276 autoclave containing an overheadmagnetic stirrer (Magnetic Drive A1120HC6CH, Parr Instrument Company,Moline, IL), a fixed thermocouple, and a pressure gauge. A Teflon liner wasused in all experiments to minimize contact of H2O2 with metallic compo-nents in the autoclave. Blank experiments were performed before each re-action to ensure negligible contributions to H2O2 decomposition from thestirrer/thermowell/liner; these wetted parts of the reactor were passivated using25 vol% HNO3 in cases where significant H2O2 decomposition was measured inthe absence of catalyst. The bare spSiO2 support (no Pd loaded) was shown to beinert toward H2O2 decomposition over all conditions studied.

In a typical reaction, the autoclave was loaded with catalyst, sealed, andpurged with Ar. The autoclave was then pressurized to 450 psi with 4% H2 inAr (Airgas), and held at 323 K for 1 h to reduce the passivated Pd nano-particles. After cooling to room temperature, the autoclave was againpurged with Ar, pressurized to 115 psi with Ar, and cooled to the desiredreaction temperature using a refrigerated bath circulator (ARCTIC A25;Thermo Scientific). The H2O2 feed solution (12.5 g of 0.08–0.60 M H2O2 inH2O) was prepared by dilution of a nonstabilized 30 wt% H2O2 solution (<10ppb Cl−; Gigabit; KMG) in ultrapure water (18 MΩ·cm), cooled to reactiontemperature, and then charged into the autoclave using a HPLC pump(Chrom Tech Series 1). The resulting pressure in the autoclave before re-action was 150 psi. Stirring (1,200 rpm) was then started. Conversion of H2O2

was determined by titration of the final solution with 0.05 M Ce(SO4)2 usingferroin as indicator.

Initial reaction rates were calculated by fitting a line through a plot of themoles H2O2 consumed versus time for conversions under 15% and normal-izing to the total number of Pd surface atoms determined by the CO uptakeexperiments; all conversion versus time data points were replicated at leasttwo times. The apparent activation energy barrier was determined over atemperature range of ∼25 K, and the apparent reaction order with respectto H2O2 was determined by varying the feed concentration of H2O2 with allother reaction parameters constant. The apparent reaction order with re-spect to the O2 product was determined by varying PO2 in the gas phaseusing a 25% O2 in Ar mixture, with all other conditions invariant. Thismixture was introduced immediately after charging the autoclave with theH2O2 feed, before stirring.

Microkinetic Model. A mean-field microkinetic model was developed to de-scribe the experimentally measured reaction rates, reaction orders, andapparent activation barrier. The model parameters were defined using aprocedure described in our previous work (44–46), using the ZPE-correctedBEs and activation energy barriers determined through DFT as initial guesses;preexponential factors and entropies were derived from the DFT-calculatedvibrational frequencies. The maximum adsorbate coverage permitted was1 ML, and adsorption/desorption steps were assumed to be quasiequilibrated. Inthe case that the microkinetic model-predicted adsorbate coverage exceededthe minimum adsorbate coverage in the context of the unit cell used in theDFT calculations (1/4 ML), the DFT calculations were repeated with the appro-priate spectator species coadsorbed in the unit cell. Note that, although theexperimental measurements were performed in a three-phase system usingconditions relevant to a DSHP process, no corrections were made to the DFTcalculations to reflect potential interaction with the liquid phase. Further detailsof the microkinetic model formulation and parameter sets are provided inSupporting Information.

ResultsThe decomposition of H2O2 has been studied both in the vaporphase (47) and aqueous phase (48) (thermal, noncatalyzed), andover a variety of materials including metal oxides (49–51) andmetal ions in solution (52, 53). Based on these studies, we havecompiled an encompassing network of 17 elementary reactionsinvolving four closed-shell species (H2O2, H2O, O2, H2) and foursurface intermediates (O, H, OH, OOH), which are shown inTable 1. Elementary reactions are classified as follows: adsorp-tion/desorption, O–O bond scission, dehydrogenation, and hy-drogen transfer. Note that the majority of calculations presentedbelow on Pd(111) are based on a previous publication (33), andthese results will not be described in detail here aside fromnoting key differences between binding properties and reactionenergetics on Pd(111) and Pd(100). Reference 54 also presents a

E1974 | www.pnas.org/cgi/doi/10.1073/pnas.1602172113 Plauck et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 1

4, 2

020

subset of DFT results on Pd(100). The “*” appended to a speciesdenotes adsorption at a single surface site, or an unoccupiedsurface site if the “*” stands alone.

Thermochemistry and Binding Configurations of Reaction Intermediateson Clean Pd Surfaces. Table 1 summarizes the most stable adsorptionsites and binding energies for all surface species on Pd(111) andPd(100). Images of the individual adsorbates in their preferredbinding geometry on Pd(100) can be viewed in Fig. 1; refer to ref. 33for the corresponding images on Pd(111). The Pd(100) facet is moreopen than the Pd(111) one and binds all intermediates more strongly.The BE of atomic hydrogen (H*) on Pd(100) is −2.74 eV, only

0.04 eV stronger than its BE on Pd(111). H* preferentially bindsto the fcc site on Pd(111) and the hollow site on Pd(100). Fur-thermore, H* is expected to be mobile on Pd(100), as the BE ofH* on a bridge site of Pd(100) is only 0.14 eV less stable thanthat on the hollow site. Similarly, on Pd(111), the BE of H* con-strained to a bridge site is 0.14 eV less stable than that on the fcc site.Atomic oxygen (O*) has the same site preferences as H* on

both Pd(111) and Pd(100). The binding strength of O* on Pd(100)is −3.90 eV, which is 0.26 eV stronger than that on Pd(111).Moreover, O* has a strong preference for the hollow site onPd(100), with the next best adsorption site (bridge) being lessstable by 0.48 eV. On Pd(111), the next best adsorption site forO* is the hcp site, which is 0.15 eV less stable than O* binding tothe fcc site.Hydroxyl (OH*) is often proposed to be the initial interme-

diate generated during H2O2 decomposition, resulting fromhomolytic O–O bond cleavage in H2O2 (55). OH* binds moststably to the bridge site on both Pd(111) and Pd(100) with theO–H bond tilted away from the plane perpendicular to the sur-face. The BE of OH* on Pd(100) is −2.43 eV, which is 0.40 eVstronger than the BE on Pd(111). OH* binding at the hollow siteof Pd(100) is only 0.05 eV weaker than on the bridge site, which issimilar to the difference in energy between OH* binding at the fccsite and bridge site on Pd(111) (0.08 eV).Hydroperoxyl (OOH*) is considered an important intermediate

in the DSHP and was identified spectroscopically during the gas-phase reaction of H2 and O2 on Au/TiO2 using inelastic neutronscattering (56). OOH* binds through its nonhydrogenated oxy-gen atom to a top site on Pd(111) with the hydroxyl group po-sitioned over an adjacent bridge site and the O–H bond pointingaway from the surface, whereas OOH* binds through its non-hydrogenated oxygen atom to a bridge site on Pd(100) with thehydroxyl group positioned over an adjacent hollow site and theO–H bond pointing away from the surface. The binding energy

of OOH* on Pd(100) is −1.28 eV—stronger than its bindingenergy on Pd(111) by 0.34 eV. However, the site preference forOOH* on Pd(100) and Pd(111) is weak: on Pd(100), OOH* canalso bind to top and hollow sites with less than 0.12 eV differencein binding energy from its most stable adsorption site; on Pd(111),OOH* can also bind through its nonhydrogenated oxygen atom tobridge sites with less than 0.03 eV difference in binding energyfrom its most stable adsorption site.Molecular oxygen (O2*) has the largest disparity in binding

strength between Pd(111) and Pd(100); the BE on Pd(100) is−1.27 eV, which is 0.77 eV stronger than that on Pd(111). O2*binds flat on Pd(100) centered over a hollow site, whereas onPd(111) O2* binds across a hcp site with one O atom at a bridgeposition and the other at a top position. Interestingly, Long et al.(57) used probe molecules and electron spin resonance spec-troscopy to show that Pd(100) can more readily activate O2through excitation of ground-state triplet O2 to reactive singletO2. This result is in agreement with our calculations; O2* retainssome of its magnetic moment on Pd(111) (33) but a negligiblemagnetic moment on Pd(100). The strong affinity of Pd(100) forO2* and O* are reflected in the tendency to reconstruct to a ki-netically stable (√5 ×√5)R27° surface oxide phase under moderatechemical potentials of O2 (58). The next best adsorption site for O2on Pd(100) is a top–top site with a binding energy of −0.85 eV.The binding energies of H2O* and H2O2* are weak (<0.4 eV)

on both Pd(111) and Pd(100). H2O* preferentially binds to topsites on both Pd(111) and Pd(100) with the O–H bonds parallelto the Pd surface. The binding energy of H2O* on Pd(100) is−0.30 eV. H2O2* also preferentially binds to top sites and adoptsthe trans configuration on both Pd facets; one oxygen atom isbound to a top site with its hydrogen atom pointing slightly awayfrom the surface plane, and the other oxygen atom is positionedover an adjacent (fcc or hollow) site with its hydrogen atompointing toward the surface.Other potential intermediates include aquoxyl (OOHH*, an

isomer of H2O2* with both hydrogen atoms on the same oxygenatom) and trihydrogen peroxide (HOOHH*). Similar to ourfindings on Pd(111) (33), neither of these species is stable onPd(100), i.e., adsorption of aquoxyl and trihydrogen peroxidestructures on Pd(100) results in spontaneous decomposition to(O* + H2O*) and (OH* + H2O*), respectively.Table 1 provides the calculated O–O bond lengths for the

adsorbed dioxygen species (O2*, OOH*, and H2O2*) and thecorresponding values calculated in the gas phase. There is sig-nificant expansion of the O–O bond in both O2* and OOH*upon adsorption, whereas the O–O bond length in H2O2* re-mains within 2% of its calculated gas-phase value. The largerO–O bond expansion on Pd(100) compared with Pd(111) suggestsa weaker O–O bond strength on the more open surface for all ofthe dioxygen species.

Activation Energy Barriers of Elementary Steps. The calculated ac-tivation energy barriers (Ea) and reaction energies (ΔE) arereported with respect to reactant and product states at infinite

Table 1. Calculated BEs of adsorbed species, their preferredadsorption sites, and O–O bond lengths (dO–O) on Pd(111) andPd(100)

Pd(111)† Pd(100)

SpeciesAdsorption

site BE, eV dO–O, Ň

Adsorptionsite BE, eV dO–O, Å

‡

H* fcc −2.70 N/A Hollow −2.74 N/AO* fcc −3.64 N/A Hollow −3.90 N/AOH* Bridge-tilted −2.03 N/A Bridge-tilted −2.43 N/AOOH* Bent-top −0.94 1.46 Bent-bridge −1.28 1.51H2O* Top −0.22 N/A Top −0.30 N/AH2O2* Top −0.32 1.48 Top −0.36 1.49O2* Top-bridge −0.50 1.35 Hollow −1.27 1.41

Reference energy corresponds to the adsorbate in the gas phase far awayfrom the metal surface. N/A, not applicable.†Data based on ref. 33.‡Calculated gas-phase dO–O for O2, OOH, and H2O2 are 1.24, 1.35, and 1.48 Å,respectively.

Fig. 1. (A–G) Side and top-down views of the preferred binding sites for alladsorbates on Pd(100). Blue spheres are hydrogen, red spheres are oxygen,and gray spheres are Pd atoms.

Plauck et al. PNAS | Published online March 22, 2016 | E1975

CHEM

ISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Mar

ch 1

4, 2

020

separation, unless stated otherwise. Table 2 summarizes the resultsfor all elementary steps on Pd(111) and Pd(100). Transition-stategeometries for the elementary steps are shown in Fig. 2.O–O bond scission.At least one type of O–O bond scission step canbe involved in the decomposition mechanism of H2O2. OH* and/or O* fragments are the direct products of O–O bond scission.Both Pd(111) and Pd(100) can readily break the O–O bond inH2O2* and OOH*, but there is a significant difference in theability of these facets to dissociate O2*.

H2O2* + * → OH* + OH*. H2O2* decomposes to two OH* onPd(100) with a barrier of 0.05 eV and a reaction energy of −2.29 eV.The corresponding barrier and reaction energy on Pd(111) are0.18 eV and −1.53 eV. This step occurs through a similar mech-anism on both Pd(111) and Pd(100) whereby H2O2* rotates fromits most stable position on a top site to the transition state at whichthe O–O bond is elongated and both OH groups are bound toadjacent Pd atoms across a bridge site. However, at the transitionstate, the O–H bonds are on the same side of H2O2 moleculeon Pd(100), whereas they are on different sides of the molecule onPd(111). Following O–O bond scission, the two OH* relax tobridge sites in their final coadsorbed state, stabilized through ahydrogen bond.

OOH* + * → O* + OH*. OOH* decomposition to O* and OH*on Pd(100) is nearly spontaneous with a barrier of 0.02 eV and areaction energy of −1.83 eV—similar to the energetics on Pd(111).The O–O bond scission occurs over a hollow site on Pd(100)and a hcp site on Pd(111).

O2* + * → O* + O*. The dissociation of O2* on Pd(100) occursover a hollow site, whereby the O–O bond stretches from 1.41 Åin the initial state to 1.90 Å in the transition state. The reactionenergy is −0.98 eV and the barrier is 0.30 eV on Pd(100), whichis 0.55 eV lower than the corresponding barrier on Pd(111).The reverse of O2* dissociation, O* recombination, represents

a potential pathway for formation of the O2 product; this stephas been proposed in a number of papers (59–61). Our calcu-

lations show that O* recombination has prohibitively high bar-riers—and is thermodynamically unfavorable—on both Pd(111)and Pd(100); the activation barrier exceeds 2 eV on Pd(111) and1 eV on Pd(100). These results are in agreement with temper-ature programmed desorption experiments for O2 desorptionfrom Pd(111) (62) and Pd(100) (63), in which the evolution of O2from these Pd single crystals after preadsorbing O* at near-ambient temperatures is only observed at temperatures exceed-ing 600 K. Furthermore, in the context of the DSHP, Lunsford(64) used a mixture of [18O2 + 16O2] with H2 over a Pd/SiO2catalyst and observed that no H2

16O18O was formed, indicatingthat O*/OH* recombination reactions were not relevant toH2O2 formation.Dehydrogenation. Because in this study we are only investigatingthe decomposition of H2O2 (reaction 3) in the absence of H2 asreactant, H* can only be derived from dehydrogenation of sur-face species through O–H bond scission. Barriers for O–H bondscission are generally lower on the more open Pd(100) facetcompared with those on Pd(111).

H2O2* + * → OOH* + H* and OOH* + * → O2* + H*. The O–Hbonds in H2O2* and OOH* are more difficult to break than theO–O bond, based on the activation barriers in Table 2. OnPd(100), the O–H bond in H2O2* that is pointing toward thesurface is cleaved over a bridge site. The activation barrier is0.44 eV, and the reaction energy is −0.29 eV.For OOH* on Pd(100), the O–H bond is also broken over a

bridge site. This breaking requires rotation of the O–H bondtoward the surface, starting from the most stable OOH* geometry.The corresponding activation barrier and reaction energy for O–Hbond cleavage in OOH* are 0.52 and −0.67 eV on Pd(100).

H2O* + * → OH* + H* and OH* + * → O* + H*.Dehydrogenationsof H2O* and OH* require a larger activation energy comparedwith H2O2* and OOH* dehydrogenations. On Pd(100), thebarrier to break the O–H bond in H2O* is 0.67 eV, and the re-action is thermoneutral. OH* dehydrogenation is more difficult

Table 2. Elementary steps considered for the decomposition of H2O2

No. Elementary step

Pd(111)† Pd(100)

Ea, eV ΔE, eV Ea, eV ΔE, eV

1 H2O2 + * ↔ H2O2* — −0.32 — −0.362 H2O* ↔ H2O + * — 0.22 — 0.303 O2* ↔ O2 + * — 0.50 — 1.274 H* + H* ↔ H2 + 2* — 1.11 — 1.195 H2O2* + * ↔ OH* + OH* 0.18 −1.53 0.05 −2.296 OOH* + * ↔ O* + OH* 0.08 −1.50 0.02 −1.837 O2* + * ↔ O* + O* 0.85 −1.23 0.30 −0.988 OH* + * ↔ O* + H* 1.02 0.07 1.03 0.179 H2O* + * ↔ OH* + H* 1.10 0.37 0.67 0.0010 OOH* + * ↔ O2* + H* 0.59 −0.20 0.52 −0.6711 H2O2* + * ↔ OOH* + H* 0.62 0.05 0.44 −0.2912 H2O* + O* ↔ OH* + OH* 0.33 0.33 0.00 −0.5113 H2O2* + O* ↔ OOH* + OH* 0.04‡ −0.44 0.14‡ −0.8714 H2O2* + OH* ↔ OOH* + H2O* 0.00 −0.16 0.00 −0.1715 OOH* + O* ↔ O2* + OH* 0.00 −0.27 0.02 −0.8116 OOH* + OH* ↔ O2* + H2O* 0.00 −0.38 0.00 −0.1317 H2O2* + O2* ↔ OOH* + OOH* 0.20 −0.02 0.00 0.00

Energetics are reported with respect to either reactants/products at infinite separation (steps 1–11) orcoadsorbed for H-transfer reactions (steps 12–17) because these reactants/products are generally stabilizedthrough hydrogen bonding. Elementary steps are classified as follows: adsorption/desorption (steps 1–4);O–O scission (steps 5–7); dehydrogenation (steps 8–11); and H transfer (steps 12–17). Ea and ΔE representthe calculated activation energy and reaction energy in the forward direction. —, no activation barriersare calculated for adsorption/desorption steps.†Data for steps 1–12 on Pd(111) are based on ref. 33.‡Activation energy corresponds to breaking Pd–O bonds to lift O* from its preferred binding site (fcc orfourfold hollow).

E1976 | www.pnas.org/cgi/doi/10.1073/pnas.1602172113 Plauck et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 1

4, 2

020

and has a barrier of 1.03 eV on Pd(100), and the reaction is slightlyendothermic. The transition state for O–H cleavage in both H2O*and OH* occurs over a hollow site on Pd(100).Hydrogen transfer. Formation and cleavage of O–H bonds in whichthe Pd surface is directly involved have significant activationbarriers (>0.4 eV). Alternatively, the Pd surface can mediate Htransfer between oxygenated intermediates—without involvingan explicit H* species. These elementary steps involve nearlyspontaneous H transfer in the exothermic direction on bothPd(111) and Pd(100) (Table 2). The activation energy barriersand reaction energies in this section are reported with respect tocoadsorbed reactant and product states, because these states aregenerally stabilized through hydrogen bonding (∼0.1–0.4 eV perhydrogen bond) with respect to the infinitely separated reactantsand products.The hydrogen atom is always transferred between O atoms

involved in hydrogen bonding in the most stable coadsorbedconfiguration. Importantly, the H-transfer steps represent po-

tential pathways for formation of both the H2O (H transfers toO*/OH*) and O2 (H transfers from H2O2*/OOH*, retaining theO–O bond) products of H2O2 decomposition.

H transfer to O.H2O2*, OOH*, and H2O* can all directly transfera H atom to O*; the activation energy barriers for these steps are0.04, 0.00, and 0.33 eV on Pd(111), with reaction energies of−0.44, −0.27, and 0.33 eV. The corresponding activation energybarriers on Pd(100) are 0.14, 0.02, and 0.00 eV with significantlymore exothermic reaction energies of −0.87, −0.81, and −0.51 eV.

H transfer to OH. H2O2* and OOH* can also directly transfer aH atom to OH*. We calculate that these steps proceed withnearly zero activation energy barrier on Pd(111) and Pd(100).The reaction energy for H transfer from H2O2* to OH* is weaklyexothermic [−0.16 eV on Pd(111) and −0.17 eV on Pd(100)].The reaction energy for H transfer from OOH* to OH* is moreexothermic on Pd(111) (−0.38 eV) than that on Pd(100) (−0.13 eV).An additional H-transfer step that was explored is H transfer

from H2O2* to O2*. This reaction is nearly thermoneutral on

Fig. 2. Side and top-down views of the transition-state geometries for O–O bond scission (A–C), dehydrogenation (D–G), and H-transfer (H–M) elementarysteps on Pd(100). Blue spheres are hydrogen, red spheres are oxygen, and gray spheres are Pd atoms. Elementary step numbers are in reference to Table 2.Bond lengths (dx-y, in Å) refer to the bond being broken in the forward reaction, as written. Note that in I, the transition state for step 13 involves breakingPd–O bonds to lift O* from its preferred binding site, followed by spontaneous H transfer from H2O2* to O*.

Fig. 3. Schematic representation of reaction pathways for H2O2 decomposition on clean surfaces. The numbers by the black arrows correspond to the el-ementary steps from Table 2. The overall reaction for each of the three complete mechanisms described in this figure is as follows: 2 H2O2 → 2 H2O + O2.

Plauck et al. PNAS | Published online March 22, 2016 | E1977

CHEM

ISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Mar

ch 1

4, 2

020

Pd(111), with a reaction energy of −0.02 eV and an activationenergy barrier of 0.20 eV. On Pd(100), the reaction is thermo-neutral with negligible barrier.

Catalytic Cycles and Potential Energy Surfaces. Based on the ele-mentary steps above, several mechanisms are available to com-plete the catalytic cycle on clean Pd facets, as summarized in Fig.3: “O*-assisted,” “OH*-assisted,” “O*+O*-recombination,” and“direct dehydrogenation.” A complete mechanism for directdehydrogenation is not shown for simplicity, as both the DFTcalculations and microkinetic modeling results suggest that directdehydrogenation steps are characterized by much higher barriersand therefore not relevant under the reaction conditions exploredin this study. The first step in all other pathways is H2O2 ad-sorption followed by homolytic O–O bond cleavage to form twoOH* species. The second H2O2* species can also adsorb anddirectly decompose to two OH*; these OH* species can thendisproportionate to form H2O* and O*—necessitating therecombination of two O* to form O2* (O*+O*-recombinationmechanism). Alternatively, two channels exist that bypass thethermodynamically unfavorable and highly activated O* recom-bination step; both involve consecutive H-transfer steps fromthe second H2O2 molecule to the O*/OH* fragments with re-tention of the original O–O bond in H2O2 (O*-assisted andOH*-assisted mechanisms).The potential energy surfaces for all pathways are displayed in

Fig. 4 for both Pd(111) and Pd(100). Based on the DFT calcu-lations alone, the O*-assisted and OH*-assisted pathways notonly provide the most energetically efficient route to form theproducts, but are also mutually competitive on both Pd(111)and Pd(100). However, the deep potential wells associatedwith the strongly bound O*/OH* fragments indicate that thereis a strong thermodynamic driving force to populate the surfaceswith O*/OH*—especially Pd(100). Therefore, the active surfaceunder reaction conditions may be partially covered by O*/OH*.Note that on O*/OH*-modified surfaces, there is an increased

probability of H transfer from H2O2* before O–O bond scission;the required O–O bond scission step may then occur in OOH*rather than in H2O2*, slightly altering the succession of ele-mentary steps proposed in Fig. 3.

Kinetics Experiments and Microkinetic Modeling. The results fromour kinetics experiments are shown in Tables 3 and 4. The ex-perimentally determined activation energy barrier of 53.3 ±3.0 kJ/mol indicates that there is a significant variation in H2O2decomposition rate with reaction temperature under conditionsrelevant to the DSHP. The nearly first-order dependence onconcentration of H2O2 is in agreement with other experimental

studies of H2O2 decomposition on Pd under similar conditions oftemperature and H2O2 concentration (13, 65). We also observedthat the addition of O2 to the gas phase did not significantlyaffect the decomposition rate of H2O2 up to O2 partial pressuresof at least 37 psi, indicating negligible product inhibition over theconditions studied. This finding is in agreement with the result ofChoudhary and Samanta (66), who observed only a minor dif-ference in the reaction rate for H2O2 decomposition overPd/Al2O3 (in the absence of H2) in a semibatch reactor whenflowing either O2 or N2 through the liquid phase.Initial estimates for microkinetic model rate parameters are de-

rived from the DFT-calculated energetics. The reactor is simulatedas a continuous stirred tank reactor. The turnover frequencies forH2O2 decomposition obtained from this model (i.e., the rate ofH2O2 converted per surface site) are used to calculate reactionorders and apparent activation barrier for comparison with theexperimental data.The rates, reaction orders, and apparent activation barriers

predicted by the microkinetic model were initially in poor agree-ment with the experimental data when using purely DFT-derivedparameters from Pd(111) or Pd(100). We subsequently usedsensitivity analysis to identify the sensitive DFT-derived BEs ofsurface species and transition-state energies. We then fit model-predicted reaction rates to experimental rates (such that the re-sidual error is less than 20%) by modifying sensitive parameters,

Fig. 4. Potential energy surfaces (thermochemistry only) for reaction pathways from Fig. 3 on clean Pd(111) and Pd(100) based on the DFT-derived ener-getics. Energies are referenced to two H2O2 molecules in the gas phase. The “j” separating two adsorbates denotes infinite separation from each other. The“(g)” denotes a gas-phase species. Insets compare O–H and O–O bond scission barriers in H2O2. “TS” denotes transition state.

Table 3. Reaction rates obtained from the kinetics experimentson Pd/spSiO2

Run† Temperature, K y(O2) x(H2O2)Experimental rate,mol·molPds

−1·s−1

1 307 0.00 0.60 71.82 307 0.00 0.30 31.53 307 0.00 0.15 17.14 307 0.00 0.08 10.55 297 0.00 0.15 7.96 285 0.00 0.15 3.47 307 0.25 0.15 16.78 307 0.13 0.15 16.29 307 0.06 0.15 16.9

†y denotes mole fraction in the gas phase (balance Ar), and x denotes molarity(moles per liter) in the liquid phase at the start of reaction. Pds denotes surfacePd atoms determined by CO uptake. Total pressure was 150 psi, and the stir ratewas 1,200 rpm for all experiments. Reaction rates reported here correspond tothe initial rates measured from conversion versus time data at<15% conversion,which was approximately linear in this regime. Each reported rate representsthe average from at least two repeated sets of conversion versus time data.

E1978 | www.pnas.org/cgi/doi/10.1073/pnas.1602172113 Plauck et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 1

4, 2

020

constraining the adjustment of parameters such that (i) the de-viation between DFT-derived BEs and activation barriers on agiven Pd facet and the corresponding values from the microkineticmodel should be within the ∼0.1- to 0.2-eV error bars generallyattributed to DFT calculations (67), and (ii) the adsorbate cov-erage used in the DFT calculations should be consistent with thecoverage predicted by the microkinetic model (that is, the solutionshould be self-consistent with respect to coverage).Next, we present the results from two model solutions that

satisfy the above criteria, but differ in both the coverage and identityof the most abundant surface intermediate, and are denoted as the“O*-coverage solution” and the “OH*-coverage solution.”O*-coverage solution. Fig. 5A shows a parity plot comparing theexperimental H2O2 decomposition rates with the microkineticmodel predictions for the O*-coverage solution [initial estimatesof parameters are derived from DFT calculations on cleanPd(111), and Supporting Information provides details of the pa-rameter adjustments used to obtain this solution]. There is goodagreement between model-predicted and experimental reactionrates, and the microkinetic model is able to accurately reproducethe experimental activation barrier and reaction orders (Table 4).The microkinetic model predictions for the surface coverage

for the most abundant intermediate, O*, range from 0.13 to0.16 ML (Fig. 5B), with the remaining sites being vacant. Mar-ginal changes to the clean surface energetics are expected fromadsorbate–adsorbate interactions at such low O* coverage (68),and furthermore this adjusted parameter set compares well withthe DFT-derived parameters on clean Pd(111); the maximumdeviation in binding energy or activation barrier is a 0.24 eVdestabilization of the binding energy of O2* on Pd(111).Therefore, a partially O*-covered Pd(111) surface is a plausiblerepresentation of the active site for H2O2 decomposition.

On the other hand, DFT-derived parameters on Pd(100) de-viate significantly (>0.35 eV for OH*, OOH*, and O2*) fromthis O*-coverage solution parameter set. Pd(100) binds inter-mediates too strongly, and the low predicted O* coverage is notexpected to destabilize intermediates on Pd(100) sufficiently forconsistency with the O*-coverage solution parameters.The O*-coverage solution predicts that the dominant reaction

pathway is the O*-assisted pathway shown in Fig. 3 (sequence ofelementary steps from Table 2: 1, 5, −12, 2, 1, 13, 16, 2, 3), withsome reaction flux through the parallel OH*-assisted pathway(sequence of elementary steps from Table 2: 1, 5, 1, 14, 2, 16, 2, 3).The rate of O* recombination to form O2* is negligible, anddehydrogenation reactions are also inactive (atomic H is onlytransferred between surface intermediates). The kinetic relevanceof each elementary step was also analyzed using Campbell’s de-gree of rate control (69, 70):

XRC,i =kir

�∂r∂ki

�Ki,eq ,kj

,

where ki and Ki,eq are the rate constant and equilibrium constantfor step i, and r is the overall reaction rate. O–O bond scission inH2O2* (step 5 of Table 2) carries the highest degree of ratecontrol over the reaction conditions examined, shown in Table5; the remaining rate control is distributed between the subse-quent H-transfer reactions.OH*-coverage solution.Using the DFT calculations on clean Pd(100)to derive initial estimates of parameters, a second solution wasidentified that also gave agreement with the experimental data set(Fig. 5C and Table 4). In this case, the model-predicted surfacecoverage is ∼0.5 ML of OH* (Fig. 5D) and is therefore not self-consistent with the clean Pd(111) and Pd(100) surface models usedin the DFT calculations. To ensure a solution self-consistent incoverage, we recalculated the binding energies of surface inter-mediates and the activation energy barriers for steps carrying sig-nificant reaction flux (as predicted by the OH*-coverage solution)in the presence of 0.5 ML of OH* spectators, i.e., two OH* wereadded to the unit cell and allowed to relax in the DFT calculations.The DFT-derived parameter set for the OH*-modified

Pd(100) surface is found to be in close agreement with the adjustedparameter set from the OH*-coverage solution (BEs shown inTable 6, with further details in Supporting Information). The DFTcalculations show that 0.5 ML of OH* destabilizes most inter-mediates and transition states investigated on Pd(100) relative tothe clean Pd(100) calculations. The binding energies of O*,OH*, O2*, and OOH* are weakened by >0.5 eV, whereas thebinding energies of H*, H2O*, and H2O2* are not significantlyaffected. In addition, the activation energy barriers for O–Obond breaking in OOH* and H2O2* increase by 0.39 and 0.56 eV,respectively. Activation energy barriers for H transfer fromH2O2* or OOH* to OH* or O* remain small (<0.2 eV). Themaximum deviation in binding energy or activation barrier be-tween the OH*-coverage solution and DFT calculations onOH*-modified Pd(100) is a 0.18-eV decrease in the activation

Fig. 5. (A and C ) Parity plots of experimental and model-predicted re-action rates for H2O2 decomposition. Refer to Table 3 for reaction condi-tions at each of the points in A and C. Pds denotes surface Pd atomsdetermined by CO uptake. Red points are varying temperature; blue pointsare varying O2 partial pressure; black points are varying feed concentra-tion of H2O2. (B and D) Microkinetic model-predicted surface coverages ofthe most abundant surface intermediates (0.15 M H2O2 in H2O feed with150 psi Ar in gas phase). Plots on the Left refer to the O*-coverage solu-tion, and plots on the Right refer to the OH*-coverage solution obtainedfrom the microkinetic model. Insets in B and D provide graphical repre-sentations of the nature of the active sites as concluded through this study[nearly clean Pd(111) and OH*-modified Pd(100)]. Blue spheres are hy-drogen, red spheres are oxygen, and gray spheres are Pd atoms.

Table 4. Experimental and microkinetic model-predictedreaction orders and apparent activation energy barriers (Eapp)

Species Experiment†O*-coverage

solutionOH*-coverage

solution

H2O2 0.92 ± 0.08 1.00 0.97O2 −0.01 ± 0.03 −0.01 0.00Eapp, kJ/mol 53.3 ± 3.0 53.1 56.6

Reported experimental error is the SE from linear regression.†Subsequent catalyst batches yielded reaction orders and an apparent bar-rier within ∼15% of the values reported here.

Plauck et al. PNAS | Published online March 22, 2016 | E1979

CHEM

ISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Mar

ch 1

4, 2

020

barrier for O–O breaking in OOH*. Therefore, OH*-modifiedPd(100) also appears to be a feasible representation of the activesite for H2O2 decomposition on Pd.The dominant reaction pathways predicted for the OH*-cover-

age solution are shown in Fig. 6. At high OH* coverage, immediateH transfer from H2O2* to OH* is predicted to be nearly quasie-quilibrated (step 14 of Table 2). The O–O bond breaks in OOH*,and this step carries the highest degree of rate control (Table 5).Hydrogen transfers from OOH* to OH* (step 16 of Table 2) andfrom OOH* to O* (step 15 of Table 2) carry the remaining re-action flux to form O2* and H2O*; hydrogen transfer from H2O*to O* (step 12 of Table 2) is also nearly quasiequilibrated.

DiscussionThe microkinetic modeling results suggest that both the close-packed Pd(111) and more open Pd(100) facets can contribute tothe total H2O2 decomposition activity. Furthermore, on both Pdfacets, all reaction flux is predicted to go through an O–O bondbreaking step (in either H2O2* or OOH*), followed by successiveH-transfer steps to O*/OH* adsorbates. The relevant surfacecoverage of O*/OH* is then a function of the ability of the Pdsurfaces to generate the O*/OH* fragments through O–O bondbreaking [which can vary strongly with surface coverage, as seenfrom calculations on the OH*-modified Pd(100)], and theavailability of H-donating species (H2O2* and OOH*) to reduceO*/OH* to H2O* through the rapid H-transfer reactions. Thismechanism is comparable to the redox mechanism discussedin refs. 65 and 71. The direct dehydrogenation and O*+O*-recombination pathways (Fig. 3) are predicted to be inactive overall experimental conditions examined.Although H-transfer steps carry some degree of rate control in

the microkinetic model solutions, the DFT calculations showthat the activation barriers for these steps are nearly insensitiveto the surface structure of the Pd substrate. Interestingly, ex-perimentally measured activation energy barriers for the gas-phase H-transfer reactions of H2O2 or OOH• to OH• or O•radicals are also readily accessible (<0.2 eV) around room tem-perature (72). The action of the metal substrate is then to generatethe O*/OH* species through O–O bond breaking, and localize theH-transfer event to the surface. H2O2*, OOH*, and H2O* aremobile on both Pd(111) and Pd(100) based on the small differ-ences in binding energies among the available binding sites andtherefore can diffuse across the Pd surface to find—and reactwith—the O*/OH* fragments. Additionally, we note that O*strongly prefers the threefold and fourfold hollow sites on Pd(111)and Pd(100), respectively; and O* must be lifted slightly from itspreferred binding site to accept a H atom. This behavior is reflectedin the low activation energy barrier generally calculated for Htransfers to O*, compared with the virtually zero barrier calculatedfor H transfers to OH*—which is more accessible at its mostfavorable binding site (bridge site) on Pd(111) and Pd(100).Breaking of the O–O bond in either H2O2* or OOH* carries the

majority of rate control, suggesting that strategies to reduce H2O2

decomposition activity must focus on tuning surface reactivity to-ward the O–O bond. Retention of the O–O bond in dioxygenspecies is generally acknowledged to be a key factor in the se-lective synthesis of H2O2 by the DSHP both in theoretical (33,73, 74) and experimental literature (8, 12, 64, 75), and ourresults here quantitatively highlight this as the central param-eter governing the subsequent H2O2 decomposition activity onthe Pd surface.In view of the aforementioned findings, reduced Pd nano-

particles would be expected to be an ineffective catalyst for theDSHP due to high activity of Pd for O–O bond breaking [0.18-eV and 0.05-eV barrier to break O–O bond in H2O2 on cleanPd(111) and Pd(100), respectively]. Extensive surface poisoningmay be necessary to inhibit H2O2 decomposition on Pd, whichour results suggest can readily occur on surface facets of sup-ported Pd nanoparticles that are generally in high abundance[the (111) and (100) facets]. Indeed, the experimentally measuredH2O2 decomposition activity of supported Pd nanoparticles can beeffectively quenched upon adding halides (along with acids,whose role may partly be to facilitate halide adsorption) to thereaction medium, often at Pd:halide atomic ratios close to orexceeding 1:1 (66, 75–78). Unfortunately, there are limited fun-damental studies that examine the halide coverage necessary toachieve this effect.Some of the most successful experimental catalysts to-date for

the DSHP are based on alloys of Pd with Au, on which the sub-sequent decomposition reactions of H2O2 are partially or com-pletely inhibited. DFT calculations indicate that dilution of Pdsurfaces with Au significantly increases barriers for O–O bondscission (79). However, experiments demonstrate that Au itself isgenerally an ineffective catalyst for the DSHP and gives slow rates(14, 80), likely due to the significant activation energy barrier re-quired to dissociate H2 on Au (81) and weak adsorption of O2(33). Promising search directions for improved DSHP catalystsmay include bimetallic systems in which an active component (e.g.,Pd) is effectively isolated in a relatively inert component (e.g., Au)resulting in reduced O–O bond breaking capacity but retention ofH2 dissociation (82) capacity; similar catalysts have proven veryeffective for the electrocatalytic synthesis of H2O2 (83, 84).Last, the presence of H2 in the reactor feed (not considered in

the present work) has been shown to enhance the overall H2O2decomposition activity over Pd-based catalysts (66, 85). Choudharyand Samanta (66) observed that, on unmodified Pd, H2 both in-creases the H2O2 decomposition rate and consumes H2O2 throughcomplete hydrogenation to H2O (reaction 4). Moreover, althoughadding chloride or bromide to an acidified reaction medium canquench H2O2 decomposition on Pd, H2O2 hydrogenation activityremains (66); this observation may indicate significant differencesin the active site(s) and rate-controlling step(s) responsible forH2O2 decomposition versus H2O2 hydrogenation—althoughO–O bond breaking is required in both reactions. Tentative

Table 5. Degree of rate control (XRC) calculated for kineticallyrelevant reaction steps for reaction condition 3 of Table 3

No. Elementary step

XRC, O*-coveragesolution

XRC, OH*-coveragesolution

5 H2O2* + * ↔ OH* + OH* 0.59 0.006 OOH* + * ↔ O* + OH* 0.02 0.5513–16 (H transfers) 0.40 0.43

XRC is given for both the O*-coverage solution and the OH*-coveragesolution at this experimental condition. Elementary step numbers (No.) arein reference to Table 2. XRC for the H transfers is the sum over all steps listed.

Table 6. DFT-calculated BEs of adsorbed species on Pd(100) inthe presence of 0.5 ML of OH*, and comparison with the BEs inthe OH*-coverage solution

Pd(100) + 0.5 ML OH* OH*-coverage solution

Species Adsorption site BE, DFT, eV BE, eV

H* Hollow −2.56 −2.56O* Hollow −3.16 −2.99OH* Top-tilted −1.68 −1.83OOH* Hollow-upright −0.61 −0.51H2O* Top −0.26 −0.32H2O2* Top −0.36 −0.46O2* Hollow −0.14 −0.14

E1980 | www.pnas.org/cgi/doi/10.1073/pnas.1602172113 Plauck et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 1

4, 2

020

explanations addressing the role of H2 have been proposed,such as maintaining the Pd surface in the reduced state (59) forfacile O–O bond breaking. Additionally, direct hydrogenationof H2O2 was shown to be a highly activated step using DFT cal-culations (86). The influence of subsurface hydrogen or evenPd-hydrides may also be relevant (87). The next stage of this workis to investigate the mechanistic role of H2 in accelerating H2O2decomposition on Pd, in addition to probing the nature of theactive site(s) in the presence of H2.

ConclusionsBoth the close-packed (111) and more open (100) facets canrepresent the active site for SiO2-supported Pd nanoparticles. TheDFT results show that O–O bond scission is facile on both Pdfacets, such that O* and OH* intermediates are readily produced.Furthermore, H2O2* and OOH* can reduce O* and OH* to H2Othrough thermodynamically driven H-transfer reactions, liberatingO2. The alternative step to produce O2 (recombination of O*) isboth thermodynamically and kinetically unfavorable. In addition,steps involving dehydrogenation through direct O–H bondcleavage over Pd are less favored than the H-transfer steps.Microkinetic models based on two parameter sets are able to

describe the experimental data for a SiO2-supported Pd catalyst:the first set corresponds to a Pd surface partially covered in <0.2ML of O*, and these adjusted parameters are consistent with theDFT-derived parameters on clean Pd(111); the second set cor-responds to a Pd surface covered in ∼0.5 ML of OH*, and these

adjusted parameters are consistent with the DFT-derived pa-rameters on a Pd(100) surface with OH* spectators. Therefore,the microkinetic model suggests that both Pd(111) and Pd(100)can contribute to H2O2 decomposition activity. Experimentalidentification of dominant surface species during H2O2 de-composition on Pd might be realized by in situ X-ray photo-electron spectroscopy measurements in a similar manner to workperformed on Pt for the oxygen reduction reaction (88).Consistent with the insights from DFT calculations, the domi-

nant reaction pathways involve O–O bond breaking in eitherH2O2* or OOH* followed by H-transfer reactions between vari-ous reaction intermediates. Breaking of the O–O bond is identi-fied as the key parameter governing H2O2 decomposition activity,because this step carries the highest degree of rate control.

ACKNOWLEDGMENTS. A.P. thanks Assistant Professor Fuat E. Celik for hisinitial guidance with the density functional theory calculations, as well asYunhai Bai and Benjamin Chen for their comments on the article. This materialis based on work supported as part of a Dow Chemical Company UniversityPartner Initiative with the University ofWisconsin–Madison, under Dow Agree-ment 235744C. Computational time was used at supercomputing resourceslocated at Environmental Molecular Sciences Laboratory (EMSL), a nationalscientific user facility at Pacific Northwest National Laboratory (PNNL); theCenter for Nanoscale Materials (CNM) at Argonne National Laboratory (ANL);and the National Energy Research Scientific Computing Center (NERSC). EMSL issponsored by the Department of Energy’s Office of Biological and EnvironmentalResearch located at PNNL. CNM and NERSC are supported by the US Departmentof Energy, Office of Science, under Contracts DE-AC02-06CH11357 and DE-AC02-05CH11231, respectively.

1. Centi G, Cavani F, Trifirò F (2001) Selective Oxidation by Heterogeneous Catalysis(Springer, New York).

2. Walsh PB (1991) Hydrogen peroxide: Innovations in chemical pulp bleaching. Tappi

Journal 74(1):81–83.3. Strukul G (1992) Catalytic Oxidations with Hydrogen Peroxide as Oxidant (Kluwer

Academic, Dordrecht, The Netherlands).4. Nijhuis TA, Makkee M, Moulijn JA, Weckhuysen BM (2006) The production of propene

oxide: Catalytic processes and recent developments. Ind Eng Chem Res 45(10):3447–3459.5. Centi G, Perathoner S, Abate S (2009) Direct synthesis of hydrogen peroxide: Recent

advances. Modern Heterogeneous Oxidation Catalysis Design, Reactions and

Characterization, ed Mizuno N (Wiley-VCH, Weinheim, Germany), pp 253–287.6. Madden S, Ross T (2005) Degussa/Headwaters Pilot Plant Validates New Tech-

nology (Business Wire, New York). Available at www.businesswire.com/news/home/

20050309005685/en/DegussaHeadwaters-Pilot-Plant-Validates-Technology. AccessedDecember 3, 2014.

7. García-Serna J, et al. (2014) Engineering in direct synthesis of hydrogen peroxide: Tar-

gets, reactors and guidelines for operational conditions. Green Chem 16(5):2320–2343.8. Samanta C (2008) Direct synthesis of hydrogen peroxide from hydrogen and oxygen:

An overview of recent developments in the process. Appl Catal A Gen 350(2):133–149.9. Campos-Martin JM, Blanco-Brieva G, Fierro JLG (2006) Hydrogen peroxide synthesis:

An outlook beyond the anthraquinone process. Angew Chem Int Ed Engl 45(42):6962–6984.10. Choudhary VR, Samanta C, Choudhary TV (2006) Factors influencing decomposition

of H2O2 over supported Pd catalyst in aqueous medium. J Mol Catal A Chem 260(1-2):

115–120.11. Abate S, et al. (2013) On the nature of selective palladium-based nanoparticles on nitrogen-

doped carbon nanotubes for the direct synthesis of H2O2. ChemCatChem 5(7):1899–1905.12. Edwards JK, Freakley SJ, Carley AF, Kiely CJ, Hutchings GJ (2014) Strategies for de-

signing supported gold-palladium bimetallic catalysts for the direct synthesis of hy-

drogen peroxide. Acc Chem Res 47(3):845–854.13. Pospelova TA, Kobozev NI (1961) Palladium-catalysed synthesis of hydrogen peroxide

from the elements. III. Active centres for the catalytic decomposition of hydrogen

peroxide on palladium. Russ J Phys Chem 35:584–587.14. Edwards JK, et al. (2009) Switching off hydrogen peroxide hydrogenation in the di-

rect synthesis process. Science 323(5917):1037–1041.15. Ouyang L, et al. (2015) The origin of active sites for direct synthesis of H2O2 on Pd/TiO2

catalysts: Interfaces of Pd and PdO domains. J Catal 321:70–80.16. Arrigo R, et al. (2014) Dynamics of palladium on nanocarbon in the direct synthesis of

H2O2. ChemSusChem 7(1):179–194.

17. Edwards JK, et al. (2012) The effect of heat treatment on the performance andstructure of carbon-supported Au-Pd catalysts for the direct synthesis of hydrogen

peroxide. J Catal 292:227–238.18. Deguchi T, Yamano H, Iwamoto M (2015) Kinetic and mechanistic studies on direct

H2O2 synthesis from H2 and O2 catalyzed by Pd in the presence of H+ and Br− in water:

A comprehensive paper. Catal Today 248:80–90.19. Monnier JR (2001) The direct epoxidation of higher olefins using molecular oxygen.

Appl Catal A Gen 221(1-2):73–91.20. Laufer W, Meiers R, Holderich W (1999) Propylene epoxidation with hydrogen peroxide

over palladium containing titanium silicalite. J Mol Catal A Chem 141(1-3):215–221.21. Meiers R, Holderich WF (1999) Epoxidation of propylene and direct synthesis of hy-

drogen peroxide by hydrogen and oxygen. Catal Lett 59(2-4):161–163.22. Jenzer G, Mallat T, Maciejewski M, Eigenmann F, Baiker A (2001) Continuous epoxi-

dation of propylene with oxygen and hydrogen on a Pd-Pt/TS-1 catalyst. Appl Catal AGen 208(1-2):125–133.

23. Popa C, et al. (2015) Structure of palladium nanoparticles under oxidative conditions.

Phys Chem Chem Phys 17(3):2268–2273.24. Lim B, et al. (2009) Shape-controlled synthesis of Pd nanocrystals in aqueous solutions.

Adv Funct Mater 19(2):189–200.25. Hammer B, Hansen LB, Norskov JK (1999) Improved adsorption energetics within

density-functional theory using revised Perdew-Burke-Ernzerhof functionals. Phys Rev

B 59(11):7413–7421.26. Greeley J, Nørskov JK, Mavrikakis M (2002) Electronic structure and catalysis on metal

surfaces. Annu Rev Phys Chem 53:319–348.27. Perdew JP, et al. (1992) Atoms, molecules, solids, and surfaces: Applications of the

generalized gradient approximation for exchange and correlation. Phys Rev B

Condens Matter 46(11):6671–6687.28. White JA, Bird DM (1994) Implementation of gradient-corrected exchange-correla-

tion potentials in Car-Parrinello total-energy calculations. Phys Rev B Condens Matter50(7):4954–4957.

29. Vanderbilt D (1990) Soft self-consistent pseudopotentials in a generalized eigenvalue

formalism. Phys Rev B Condens Matter 41(11):7892–7895.30. Kresse G, Furthmuller J (1996) Efficiency of ab-initio total energy calculations for metals

and semiconductors using a plane-wave basis set. Comput Mater Sci 6(1):15–50.31. Chadi DJ, Cohen ML (1973) Special points in the Brillouin zone. Phys Rev B 8(12):

5747–5753.32. Monkhorst HJ, Pack JD (1976) Special points for Brillouin-zone integrations. Phys Rev

B 13(12):5188–5192.

Fig. 6. Dominant reaction pathways predicted by the microkinetic model for the OH*-coverage solution. The numbers by the black arrows correspond to theelementary steps numbers given in Table 2.

Plauck et al. PNAS | Published online March 22, 2016 | E1981

CHEM

ISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Mar

ch 1

4, 2

020

33. Ford DC, Nilekar AU, Xu Y, Mavrikakis M (2010) Partial and complete reduction of O2

by hydrogen on transition metal surfaces. Surf Sci 604(19-20):1565–1575.34. Neugebauer J, Scheffler M (1992) Adsorbate-substrate and adsorbate-adsorbate in-

teractions of Na and K adlayers on Al(111). Phys Rev B Condens Matter 46(24):16067–16080.

35. Bengtsson L (1999) Dipole correction for surface supercell calculations. Phys Rev B59(19):12301–12304.

36. Mermin ND, Ashcroft NW (1976) Solid State Physics (Harcourt College Publishers, FortWorth, TX).

37. Henkelman G, Uberuaga BP, Jonsson H (2000) A climbing image nudged elastic bandmethod for finding saddle points and minimum energy paths. J Chem Phys 113(22):9901–9904.

38. Henkelman G, Jonsson H (2000) Improved tangent estimate in the nudged elasticband method for finding minimum energy paths and saddle points. J Chem Phys113(22):9978–9985.

39. Greeley J, Mavrikakis M (2003) A first-principles study of surface and subsurface H onand in Ni(111): Diffusional properties and coverage-dependent behavior. Surf Sci540(2-3):215–229.

40. Stober W, Fink A, Bohn E (1968) Controlled growth of monodisperse silica spheres inthe micron size range. J Colloid Interface Sci 26(1):62–69.

41. Alba-Rubio AC, et al. (2014) Pore structure and bifunctional catalyst activity ofoverlayers applied by atomic layer deposition on copper nanoparticles. ACS Catal 4(5):1554–1557.

42. Zou WQ, Gonzalez RD (1992) The preparation of silica supported Pd catalysts: Theeffect of pretreatment variables on particle size. Catal Lett 12(1-3):73–86.

43. Spiewak BE, Shen J, Dumesic JA (1995) Microcalorimetric studies of CO and H2 ad-sorption on nickel powders promoted with potassium and cesium. J Phys Chem99(49):17640–17644.

44. Gokhale AA, Kandoi S, Greeley JP, Mavrikakis M, Dumesic JA (2004) Molecular-leveldescriptions of surface chemistry in kinetic models using density functional theory.Chem Eng Sci 59(22-23):4679–4691.

45. Grabow LC, Mavrikakis M (2011) Mechanism of methanol synthesis on Cu throughCO2 and CO hydrogenation. ACS Catal 1(4):365–384.

46. Grabow LC, Gokhale AA, Evans ST, Dumesic JA, Mavrikakis M (2008) Mechanism ofthe water gas shift reaction on Pt: First principles, experiments, and microkineticmodeling. J Phys Chem C 112(12):4608–4617.

47. Tessier A, Forst W (1974) Mechanism of hydrogen-peroxide pyrolysis. Can J Chem52(5):794–797.

48. Takagi J, Ishigure K (1985) Thermal decomposition of hydrogen peroxide and its ef-fect on reactor water monitoring of boiling water reactors. Nucl Sci Eng 89(2):177–186.

49. Lousada CM, Johansson AJ, Brinck T, Jonsson M (2012) Mechanism of H2O2 de-composition on transition metal oxide surfaces. J Phys Chem C 116(17):9533–9543.

50. Hiroki A, Laverne JA (2005) Decomposition of hydrogen peroxide at water-ceramicoxide interfaces. J Phys Chem B 109(8):3364–3370.

51. Lin SS, Gurol MD (1998) Catalytic decomposition of hydrogen peroxide on iron oxide:Kinetics, mechanism, and implications. Environ Sci Technol 32(10):1417–1423.

52. Haber F, Weiss J (1934) The catalytic decomposition of hydrogen peroxide by ironsalts. Proc R Soc Lond A Math Phys Sci 147(861):332–351.

53. Kwan WP, Voelker BM (2002) Decomposition of hydrogen peroxide and organiccompounds in the presence of dissolved iron and ferrihydrite. Environ Sci Technol36(7):1467–1476.

54. Tian P, Ouyang L, Xu X, Xu J, Han Y-F (2013) Density functional theory study of directsynthesis of H2O2 from H2 and O2 on Pd(111), Pd(100), and Pd(110) surfaces. Chin JCatal 34(5):1002–1012.

55. Lousada CM, Johansson AJ, Brinck T, Jonsson M (2013) Reactivity of metal oxideclusters with hydrogen peroxide and water—a DFT study evaluating the performanceof different exchange-correlation functionals. Phys Chem Chem Phys 15(15):5539–5552.

56. Sivadinarayana C, Choudhary TV, Daemen LL, Eckert J, Goodman DW (2004) Thenature of the surface species formed on Au/TiO2 during the reaction of H2 and O2: Aninelastic neutron scattering study. J Am Chem Soc 126(1):38–39.

57. Long R, et al. (2013) Surface facet of palladium nanocrystals: A key parameter to theactivation of molecular oxygen for organic catalysis and cancer treatment. J Am ChemSoc 135(8):3200–3207.

58. Simmons GW, Wang YN, Marcos J, Klier K (1991) Oxygen adsorption on Pd(100)surface: Phase transformations and surface reconstruction. J Phys Chem 95(11):4522–4528.

59. Ntainjua NE, et al. (2008) The role of the support in achieving high selectivity in thedirect formation of hydrogen peroxide. Green Chem 10(11):1162–1169.

60. Choudhary VR, Samanta C, Jana P (2007) Decomposition and/or hydrogenation ofhydrogen peroxide over Pd/Al2O3 catalyst in aqueous medium: Factors affecting therate of H2O2 destruction in presence of hydrogen. Appl Catal A Gen 332(1):70–78.

61. Paunovic V, Ordomsky VV, Sushkevich VL, Schouten JC, Nijhuis TA (2015) Direct syn-thesis of hydrogen peroxide over Au-Pd catalyst: The effect of co-solvent addition.ChemCatChem 7(7):1161–1176.

62. Zheng G, Altman EI (2000) The oxidation of Pd(111). Surf Sci 462(1-3):151–168.63. Zheng G, Altman EI (2002) The oxidation mechanism of Pd(100). Surf Sci 504(1-3):

253–270.64. Lunsford JH (2003) The direct formation of H2O2 from H2 and O2 over palladium

catalysts. J Catal 216(1-2):455–460.65. Bianchi G, Mazza F, Mussini T (1962) Catalytic decomposition of acid hydrogen per-

oxide solutions on platinum, iridium, palladium and gold surfaces. Electrochim Acta7(4):457–473.

66. Choudhary VR, Samanta C (2006) Role of chloride or bromide anions and protons forpromoting the selective oxidation of H2 by O2 to H2O2 over supported Pd catalysts inan aqueous medium. J Catal 238(1):28–38.

67. Singh S, et al. (2014) Formic acid decomposition on Au catalysts: DFT, microkineticmodeling, and reaction kinetics experiments. AIChE J 60(4):1303–1319.

68. Grabow LC, Hvolbaek B, Norskov JK (2010) Understanding trends in catalytic activity:The effect of adsorbate-adsorbate interactions for CO oxidation over transitionmetals. Top Catal 53(5-6):298–310.

69. Campbell CT (2001) Finding the rate-determining step in a mechanism: ComparingDeDonder relations with the “degree of rate control.” J Catal 204(2):520–524.

70. Dumesic JA (1999) Analyses of reaction schemes using De Donder relations. J Catal185(2):496–505.

71. Voloshin Y, Manganaro J, Lawal A (2008) Kinetics and mechanism of decompositionof hydrogen peroxide over Pd/SiO2 catalyst. Ind Eng Chem Res 47(21):8119–8125.

72. Atkinson R, et al. (2004) Evaluated kinetic and photochemical data for atmosphericchemistry: Volume I—gas phase reactions of Ox, HOx, NOx and SOx species. AtmosChem Phys 4:1461–1738.

73. Grabow LC, Hvolbaek B, Falsig H, Norskov JK (2012) Search directions for direct H2O2

synthesis catalysts starting from Au12 nanoclusters. Top Catal 55(5-6):336–344.74. Rankin RB, Greeley J (2012) Trends in selective hydrogen peroxide production on

transition metal surfaces from first principles. ACS Catal 2(12):2664–2672.75. Burch R, Ellis PR (2003) An investigation of alternative catalytic approaches for the

direct synthesis of hydrogen peroxide from hydrogen and oxygen. Appl Catal B 42(2):203–211.

76. Melada S, Pinna F, Strukul G, Perathoner S, Centi G (2006) Direct synthesis of H2O2 onmonometallic and bimetallic catalytic membranes using methanol as reaction me-dium. J Catal 237(2):213–219.

77. Ntainjua N E, et al. (2009) Effect of halide and acid additives on the direct synthesis ofhydrogen peroxide using supported gold-palladium catalysts. ChemSusChem 2(6):575–580.

78. Liu Q, Lunsford JH (2006) Controlling factors in the direct formation of H2O2 from H2

and O2 over a Pd/SiO2 catalyst in ethanol. Appl Catal A Gen 314(1):94–100.79. Ham HC, Hwang GS, Han J, Nam SW, Lim TH (2009) On the role of Pd ensembles in

selective H2O2 formation on PdAu alloys. J Phys Chem C 113(30):12943–12945.80. Edwards JK, et al. (2005) Direct synthesis of hydrogen peroxide from H2 and O2 using

Au-Pd/Fe2O3 catalysts. J Mater Chem 15(43):4595–4600.81. Hammer B, Norskov JK (1995) Why gold is the noblest of all the metals. Nature

376(6537):238–240.82. Tierney HL, Baber AE, Kitchin JR, Sykes ECH (2009) Hydrogen dissociation and spillover

on individual isolated palladium atoms. Phys Rev Lett 103(24):246102.83. Jirkovský JS, et al. (2011) Single atom hot-spots at Au-Pd nanoalloys for electro-

catalytic H2O2 production. J Am Chem Soc 133(48):19432–19441.84. Verdaguer-Casadevall A, et al. (2014) Trends in the electrochemical synthesis of H2O2:

Enhancing activity and selectivity by electrocatalytic site engineering. Nano Lett 14(3):1603–1608.

85. Landon P, et al. (2003) Direct synthesis of hydrogen peroxide from H2 and O2 using Pdand Au catalysts. Phys Chem Chem Phys 5(9):1917–1923.

86. Li J, Staykov A, Ishihara T, Yoshizawa K (2011) Theoretical study of the decompositionand hydrogenation of H2O2 on Pd and Au@Pd surfaces: Understanding toward highselectivity of H2O2 synthesis. J Phys Chem C 115(15):7392–7398.

87. Todorovic R, Meyer RJ (2011) A comparative density functional theory study of thedirect synthesis of H2O2 on Pd, Pt and Au surfaces. Catal Today 160(1):242–248.

88. Casalongue HS, et al. (2013) Direct observation of the oxygenated species duringoxygen reduction on a platinum fuel cell cathode. Nat Commun 4(2817):1–6.

89. Madon RJ, et al. (2011) Microkinetic analysis and mechanism of the water gas shiftreaction over copper catalysts. J Catal 281(1):1–11.

90. Hwang H, Dasgupta PK (1985) Thermodynamics of the hydrogen peroxide-watersystem. Environ Sci Technol 19(3):255–258.

91. Michaelides A (2006) Density functional theory simulations of water-metal interfaces:Waltzing waters, a novel 2D ice phase, and more. Appl Phys A Mater Sci Process 85(4):415–425.

92. Sha Y, Yu TH, Liu Y, Merinov BV, GoddardWA (2010) Theoretical study of solvent effectson the platinum-catalyzed oxygen reduction reaction. J Phys Chem Lett 1(5):856–861.

93. Carrasco J, Santra B, Klime�s J, Michaelides A (2011) To wet or not to wet? Dispersionforces tip the balance for water ice on metals. Phys Rev Lett 106(2):026101.

94. Santra B, et al. (2013) On the accuracy of van der Waals inclusive density-functionaltheory exchange-correlation functionals for ice at ambient and high pressures.J Chem Phys 139(15):154702.

E1982 | www.pnas.org/cgi/doi/10.1073/pnas.1602172113 Plauck et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 1

4, 2

020