This is an open access article published under an ACS AuthorChoice License, which permits copying and redistribution of the article or any adaptations for non-commercial purposes.
Article Cite This: Acc. Chem. Res. XXXX, XXX, XXX−XXX
pubs.acs.org/accounts
Energies of Adsorbed Catalytic Intermediates on Transition Metal Surfaces: Calorimetric Measurements and Benchmarks for Theory Charles T. Campbell*
Downloaded via 91.243.93.135 on March 17, 2019 at 12:50:20 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.
Department of Chemistry, University of Washington, Seattle, Washington 98195-1700, United States CONSPECTUS: Better catalysts and electrocatalysts are essential for the production and use of clean fuels with less pollution and improved energy efficiency, for making chemicals with less energy and environmental impact, for pollution abatement, and for many other future technologies needed to achieve environmentally friendlier energy supply and chemicals industry. Crucial for rational design of better catalyst and electrocatalyst materials is knowledge of the energies of elementary chemical reactions on late transition metal surfaces. This knowledge would also aid in designing more efficient and stable photocatalysts and batteries for harvesting and storing solar energy. These are all crucial for sustainable living with high quality. Herein, I review measurements of surface reaction energies involving many of the most common adsorbates formed as intermediates on late transition metal surfaces in catalytic and electrocatalytic reactions of interest for energy and environmental technologies. I focus on calorimetric measurements of the heat of molecular and dissociative adsorption of gases on single crystals (i.e., single crystal adsorption calorimetry, or SCAC) that allow the heats of formation of adsorbed intermediates in well-defined structures to be directly determined. Adsorption reactions are often irreversible, and in such cases SCAC is required to get these heats, since the other methods for measuring adsorption energies (equilibrium adsorption isotherms and temperature-programmed desorption) work only for reversible adsorption. Common examples of irreversible adsorption reactions are ones that produce adsorbed molecular fragments or adsorbed molecules such as olefins and aromatic molecules that bind very strongly to non-noble metals. When the heats of formation of different adsorbed molecular fragments are compared to each other, and to their values on different metal surfaces, they reveal which properties of the metal surface and the molecular fragments determine metal− adsorbate bond strengths, and clarify differences in catalytic reactivity between different metals. When combined with earlier adsorption energy measurements, these heats also provide a database of reliable energies of adsorbed catalytic intermediates that serve as crucial benchmarks to guide the development of improved computational methods for calculating the energetics of elementary steps on late transition metal surfaces (i.e., reaction energies and activation barriers), such as density functional theory. The energy accuracy of such computational estimates is crucial for the future of catalysis research and catalyst discovery.
1. INTRODUCTION Gerhard Ertl was awarded the 2007 Nobel Prize in Chemistry partially for demonstrating the importance of surface reaction energy diagrams in understanding why some materials make better catalysts than others. That is, the energies of the adsorbed intermediates and transition states involved in the elementary reaction steps that occur between reactants and products are the most important properties that determine the rate and selectivity of one solid surface compared to another. These are the same energies that are required to build accurate microkinetic models of surface-catalyzed reactions.1 The standard-state entropies of the adsorbed intermediates and transition states (or kinetic prefactors) are also required to predict rates,2 but these tend to be rather similar for different materials within the same class,3,4 so that the energy differences usually dominate differences in relative activities of different catalysts within the same class of materials. Thus, there is a tremendous motivation to know the energies of adsorbed catalytic reaction intermediates and transition states on different surfaces. The groups of Ertl, Somorjai, King, © XXXX American Chemical Society
Madix, White, and many others did beautiful experiments to estimate these energies. Since it is nearly impossible to cleanly isolate adsorbed species or one of their elementary reaction steps on polycrystalline or powder surfaces, the most impactful measurements focused on single crystal surfaces. Most adsorbate energies on such single crystal surfaces have been determined by measuring the activation energies for the forward and reverse elementary steps and subtracting them, or by equilibrium adsorption isotherms. Since these methods require reversible elementary steps, they were able to reveal adsorbate energies, but only for species that are reversibly adsorbed, which limited results to small molecularly adsorbed species and a few atomic adsorbates (H, O, N) that can be reversibly produced from diatomic gases on some transition metal surfaces.5 Olefins and aromatic molecules only adsorb reversibly on noble metals, so their adsorption energies could not be determined on other transition metals by those Received: November 19, 2018
A
DOI: 10.1021/acs.accounts.8b00579 Acc. Chem. Res. XXXX, XXX, XXX−XXX
Article
Accounts of Chemical Research approaches. In 1991, Sir David King’s group6,7 introduced an impressive calorimeter that was capable for the first time of directly measuring the heat of adsorption on single crystal surfaces, and thus extended our knowledge of adsorbate energies on transition metals to more complex molecular fragments and ethylene.8,9 They termed this method single crystal adsorption calorimetry (SCAC). Our group introduced a new type of heat detector for SCAC that greatly extended its sensitivity and temperature range of applicability.10,11 Results from that version of SCAC will be reviewed below. All of these methods for measuring adsorbate energies are very demanding technically. They require first that many experiments be done with surface spectroscopies to learn what species are produced from what gases on what surfaces, and at what temperatures. Only then can the adsorbate energy actually be determined, and this only in special cases. The work involved is too demanding to use these methods to determine a database of energies sufficiently large to guide the design of new catalyst materials at the desired level. Nevertheless, they have already revealed much information regarding the systematics of surface reaction energies that have greatly aided our ability to fundamentally understand many catalytic reaction mechanisms and their kinetics. This has made it possible to explain some of the differences in reactivity and selectivity between different transition metal or bimetallic surfaces. Some examples of this from my own group will be presented below. Computational methods for density functional theory (DFT) calculations with periodic boundary conditions are fast and provide good energy accuracy for late transition metal surfaces. When these were developed about 20 years ago, surface scientists and catalysis researchers realized that they finally had a method with predictive ability for surface chemistry. As a consequence, there was a revolution in surface chemistry. Today there are hundreds of research groups worldwide that employ DFT with periodic boundary conditions (referred to below as “periodic DFT”) to aid in their studies of surface chemical reactions. This explosive growth is due to the new basic understanding of surface reaction mechanisms and surface reactivity that DFT provides and, more recently, by its applications to aid in discovering new and improved catalytic and electrocatalytic materials. Already, groups led by Nørskov, Mavrikakis, Greeley, Metiu, Neurock, Abild-Pedersen, Bligaard, Grabow, and others have had successes in using periodic DFT to identify catalysts and electrocatalysts with better performance and to reveal structure−performance relations that can help accelerate such discovery.12−14 Clearly, fast computational chemistry using approaches like periodic DFT will have a tremendous positive impact on heterogeneous catalysis research in the next decades. Indeed, it provides perhaps our greatest hope for developing better catalysts, electrocatalysts, and photocatalysts to meet the technological challenges required for sustainable chemicals and fuels industries. Unfortunately, the accuracy of periodic DFT in predicting adsorbate energies is still insufficient for it to realize anywhere near its full potential in aiding basic catalysis research or catalyst discovery, as shown below. Major errors for aromatic adsorbates were made obvious by our 2004−2006 measurements of the heats of benzene and naphthalene adsorption on Pt(111), which were 50−140 kJ/mol higher than DFT estimates.15,16 This is generally attributed by theoreticians to the fact that standard periodic DFT methods do not treat van
der Waals (vdW) interactions between adsorbates and metal surfaces properly,17,18 and efforts to correct this are intense. For the same reason, such methods also give very poor adsorption energies for linear alkanes19 compared to our experimental measurements by TPD on Pt(111), graphite(111) and MgO(100).20 The recognition that standard DFT methods do poorly with vdW interactions has led to an explosion in efforts to introduce vdW corrections to DFT. The annual number of papers that mention both vdW and DFT in the topic has increased almost 15-fold since 2004, to 18,000 in 2017. This highlights the importance of experimental benchmarks to the improvement of computational methods for catalysis and in fact all areas of surface science and catalysis. For smaller systems, it has been common to improve fast computational methods like DFT by comparison to benchmarks calculated with theoretical methods that are highly accurate but much more costly, like coupled cluster calculations. However, such comparison is not available for solid surfaces since there are no computational methods that give the desired benchmark-level chemical accuracy for adsorbates with such large systems. Thus, the experimentally measured adsorbate energies described herein are essential for improving the energy accuracy of fast theoretical methods for computational research in catalysis by transition metals, and in all other aspects of surface science. A substantial improvement in computational energy accuracy would be transformative for research not only in catalysis and surface science, but also in many areas of materials science and engineering. In the long term, the most important impact of these experimental adsorbate energies is likely to be helping to improve accuracy in new approaches to their theoretical calculation. In catalysis research, energy accuracy refers to the accuracy in the energy differences between different surface structures, for example, between key adsorbed intermediates on different surfaces, between different intermediates on the same surface, or between one adsorbed intermediate and its transition states for conversion to other species. These energy differences are crucial, since they determine the catalyst’s mechanism, rate, selectivity and long-term stability against poisoning. The accuracy required on relative adsorbate energies to predict rates is really quite demanding, since an error of only 20 kJ/ mol in the relative energy of an adsorbed intermediate or transition state translates into an error in reaction rate of over 400-fold at a reaction temperature of 400 K, at least for species with a large “degree of rate control”. (For the definition of degree of rate control, see ref 1.) Such a large error in rate could easily lead one to overlook promising materials in computational screening of catalyst materials, or to predict that the wrong reaction pathway dominates the mechanism on a well-known catalyst. Yet the relative errors in DFT methods often exceed 20 kJ/mol. For example, the relative error between di-σ-bonded cyclohexene and benzene adsorbed on Pt(111) is 23, 20, 52, and 48 kJ/mol for PW-91, PBE, RPBE, and BEEF-vdW.19 The conversion between these two intermediates occurs in several important reactions on Pt catalysts, including the hydrogenation of benzene and the dehydrogenation of cyclohexene and cyclohexane,21 and is even more important since these are the simplest prototypes of whole classes of dehydrogenation and hydrogenation reactions involving aromatics. B
DOI: 10.1021/acs.accounts.8b00579 Acc. Chem. Res. XXXX, XXX, XXX−XXX
Article
Accounts of Chemical Research
2. SINGLE CRYSTAL ADSORPTION CALORIMETRY (SCAC) Single crystal surfaces offer tremendous advantages for studying surface chemistry compared to powdered catalysts or other polycrystalline samples, since they afford much easier determination of the chemical formulas and geometric structures of the adsorbates formed upon gas adsorption. They also are the best type of surfaces for comparing experiments to theory, since the most accurate methods for computational surface chemistry use DFT with periodic boundary conditions, i.e., models for extended single crystal surfaces. Also, elementary reaction steps have different rate constants and activation energies on different single crystal facets of the same metal, so that kinetic studies and their interpretation in terms of microkinetic models (i.e., based on elementary steps) are also greatly simplified when performed on single crystal surfaces. Thus, for adsorption calorimetry to be most powerful, it must be done on single crystal surfaces, where the adsorbates are well-defined. Following the pioneering SCAC design from the group of Sir David King at Cambridge University,6,22−24 our group greatly improved the sensitivity and temperature range of measurements for adsorption microcalorimetry on single crystal surfaces by developing a new type of heat detector.10,11,25−33 The SCAC method gives a direct measure of the heats of chemical reactions that are induced by impinging a pulsed molecular beam of gas onto the single crystal surface in the clean conditions of ultrahigh vacuum (UHV). The gas pulses are typically 100 ms long and repeated every 2−5 s. The sticking probability for each pulse is measured by the King-and-Wells method. Our new heat detector design is based on a 9 μm-thick pyroelectric polymer ribbon that is bent into an arch and pressed against the back face of the single crystal (which is ∼1 μm thick). This is orders of magnitude more sensitive than King’s infrared optical pyrometry method, and covers a much broader temperature range to cryogenic conditions. Its precision is very high, showing a pulse-to-pulse standard deviation of just 1.3 kJ/mol with only 1.5 × 1013 molecules per cm2 (∼1% of a monolayer) per gas pulse.11 Its absolute accuracy is better than 3%.34−36 A typical heat signal response is shown in Figure 1, for cyclohexene adsorption on Pt(111) at 100 K. Here, it adsorbs molecularly, initially as di-σ bonded cyclohexene, with a heat of adsorption that drops from ∼130 to 50 kJ/mol with coverage up to 0.25 ML. (One ML or monolayer is defined throughout as a coverage of one species per surface metal atom.) At 263 K, it instead dissociates initially, forming adsorbed π-allyl-c-C6H9 (2-cyclohexenyl) plus a H adatom, with a heat that drops from ∼160 kJ/mol initially to ∼90 kJ/mol at saturation coverage (0.18 ML). The heat signals are calibrated by shining light pulses of the same spatial and temporal profile as the molecular beam pulses onto the sample. In the example of Figure 1, the heat signals had the same exact line-shape as that seen from these light pulses (which is instantaneous and thus measures the instrument response function). This means that the heat of adsorption was deposited instantaneously (or too fast to give rise to significant signal broadening relative to this instrument response function). The heat signals from adsorption events are sometimes broader than those from the laser calibration. This allows us to also extract dynamical information from the heat signal line-shape. An example is shown in Figure 2, for the
Figure 1. Raw calorimetry signal and the molecular beam intensity versus time for three pulses of cyclohexene adsorption on Pt(111) at 100 K, where it adsorbs molecularly. Each gas pulse (seen in the molecular beam intensity profile shown here) consists of 0.011 ML in the ∼4 mm beam diameter, and deposits only ∼0.25 μJ of heat into the Pt crystal. Reproduced with permission from ref 31. Copyright 2008 American Chemical Society.
Figure 2. Example of the deconvolution of a heat signal into a fast step followed by a slow step, for the case of methyl iodide dissociation on clean Pt(111) at 270 K. The red solid curve is the experimental signal for the third CH3I pulse (which increases the coverage from 0.037 to 0.055 ML). The simulated heat pulse (purple dashed curve) of total heat 212 kJ/mol is the sum of the fast molecular adsorption step (#S1, blue), followed by a slow dissociation step (#S2, green) that has a first-order time constant τ (and rate constant k = 1/τ). Reproduced with permission from ref 37. Copyright 2013 Elsevier.
two-step dissociative adsorption of CH3I on Pt(111). Here, we were able to determine not only the total heat, but the separate heats for the fast molecular adsorption step (136 kJ/mol) and the slower dissociation step (76 kJ/mol), as well as the rate constant for this second step (4.2 s−1).37 We later developed methods to directly deconvolute the instrument response function from the signal, to directly give the heating power due to adsorption/reaction versus time.38 The surface temperature can be controlled from 100 to ∼380 K for our SCAC measurements. It is usually desirable to increase the rates of the elementary steps involved so that such pulse broadening does not occur, thereby eliminating the need for the complex fitting exemplified by Figure 2, which reduces the accuracy of the net reaction energy found. Thus, we typically choose the highest temperature where the net C
DOI: 10.1021/acs.accounts.8b00579 Acc. Chem. Res. XXXX, XXX, XXX−XXX
Article
Accounts of Chemical Research reaction of interest occurs, but below the temperature where its products further decompose. Sometimes pulse broadening still occurs there.
3. CALORIMETRIC MEASUREMENTS OF ADSORBED CATALYTIC INTERMEDIATES ON Pt(111) AND Ni(111) SURFACES We applied these SCAC techniques first to measure the heats of formation of several adsorbed intermediates on Pt(111). Specifically, we were able to determine the heats of formation of adsorbed benzene, naphthalene, di-σ-cyclohexene and cyclohexenyl, −OH, −OCH3, formate, −CH, −CH2, −CH3, and −C(CH3)3 on Pt(111).15,16,21,31,34−36,39−45 These include the simplest examples of six ubiquitous classes of adsorbed intermediates proposed in countless catalytic mechanisms: alkoxy groups, carboxylates, carbynes, carbenes, alkyls, tertiary alkyls, di-σ-bonded olefins, and π-allyls. To our knowledge, these results, taken together with our similar but later results on Ni(111) described below, include the only measurements of the heats of formation of molecular f ragments on any late transition metal surfaces except for a few specific adsorbates. Those included atomic H, O, and N, whose values were determined on several transition metals by temperatureprogrammed desorption (TPD) or equilibrium adsorption isotherms (EAI), and various C2Hx species on several faces of Pt, Ni, Pd, and Rh and −CH on Ni(100) and (110). These heats for −CH and C2Hx species were determined by King’s group using SCAC.8,9 The results from our studies of the dissociative adsorption of methyl iodide and di-iodomethane allowed us to determine the energy diagram for methane activation on Pt(111), as summarized in Figure 3. Given the importance of catalytic methane conversion to the future clean utilization of US shale gas, these types of energetics are of current interest.
Figure 4. Differential heat of dissociative adsorption of CH3I to form adsorbed methyl and an adsorbed iodine atom on Ni(111) at 160 K as a function of dissociated CH3I coverage. Each data point represents a pulse of ∼0.0032 ML of CH3I gas and is the average of nine experimental runs. The equation is the best fit of that functional form to the data in the range below 1/7 ML. The effect of defect sites (probably step edges) can be seen in the ∼20 kJ/mol higher heat for the very first gas pulse. Reproduced with permission from ref 46. Copyright 2017 American Chemical Society.
value we found for CH3−Pt(111) bonds. DFT calculations from previous literature have systematically underestimated the bond energy of methyl to Ni(111), giving that methyl binds more weakly to Ni(111) than Pt(111), the opposite trend from these SCAC experiments.46 This is an important difference, since the SCAC result (that alkyls bind more strongly to Ni than Pt) explains the wellknown much higher activity of Ni catalysts than Pt for C−C bond-breaking reactions, such as hydrogenolysis.46 That is because a metal should have a lower activation barrier for cleaving a C−C bond in an adsorbed hydrocarbon if the two fragments produced bond more strongly to that metal compared to another metal, according to Brønsted−Evans− Polanyi (BEP) relations. The DFT results are inconsistent with the activity difference. Being able to accurately measure (or better, predict accurately by fast calculations) the relative energies of key adsorbed intermediates on different surfaces is clearly the key to explaining differences in activity and selectivity between different catalyst materials, and will also be essential for designing new catalytic materials based on metals, bimetallics, alloys, oxides, mixed oxides, etc. We have measured the heats of adsorption for many other reactions where dissociation occurs on both Ni(111) and Pt(111) surfaces. An example for olefin adsorption on Pt(111) is shown on Figure 5. In this case the olefin is cyclohexene, which adsorbs at lower temperature molecularly, or partially dissociating the CC double bond, to make di-σ-bonded cyclohexene. At the temperature here, it loses an H to the Pt surface and makes the π-allyl species, cyclohexenyl. The strong decrease in heat of adsorption with coverage is common for such large hydrocarbon adsorbates, and is probably related to their large dipole moments47,48 and cumulative pairwise dipole−dipole repulsions between adsorbates. We also studied the adsorption of several aromatic hydrocarbons on Pt(111) and Ni(111) surfaces by SCAC. The results for benzene are shown in Figure 6. As seen, there is only a slight difference in the heats of formation and bond energies of benzene on these two surfaces. Benzene is only ∼6 kJ/mol more stable on Ni(111) than Pt(111). This was
Figure 3. Experimental enthalpy landscape for stepwise dehydrogenation of methane on Pt(111), adapted with permission from ref 45. Copyright 2014 American Chemical Society. We also confirmed the barrier between CH4,ad → CH3,ad + Had,36 shown here as well. All energies are at low coverage (0.04 ML), which minimizes the effect of coadsorbed iodine adatoms on energies, but exceeds the coverage where defects influence the heat (
