Chemical Bonding at Surfaces and Interfaces

By Anders Nilsson

Molecular surface science has made enormous progress in the past 30 years. The development can be characterized by a revolution in fundamental knowledge obtained from simple model systems and by an explosion in the number of experimental techniques. The last 10 years has seen an equally rapid development of quantum mechanical modeling of surface processes using Density Functional Theory (DFT).

Chemical Bonding at Surfaces and Interfaces focuses on phenomena and concepts rather than on experimental or theoretical techniques. The aim is to provide the common basis for describing the interaction of atoms and molecules with surfaces and this to be used very broadly in science and technology.

The book begins with an overview of structural information on surface adsorbates and discusses the structure of a number of important chemisorption systems. Chapter 2 describes in detail the chemical bond between atoms or molecules and a metal surface in the observed surface structures. A detailed description of experimental information on the dynamics of bond-formation and bond-breaking at surfaces make up Chapter 3. Followed by an in-depth analysis of aspects of heterogeneous catalysis based on the d-band model. In Chapter 5 adsorption and chemistry on the enormously important Si and Ge semiconductor surfaces are covered. In the remaining two Chapters the book moves on from solid-gas interfaces and looks at solid-liquid interface processes. In the final chapter an overview is given of the environmentally important chemical processes occurring on mineral and oxide surfaces in contact with water and electrolytes.

• Gives examples of how modern theoretical DFT techniques can be used to design heterogeneous catalysts
• This book suits the rapid introduction of methods and concepts from surface science into a broad range of scientific disciplines where the interaction between a solid and the surrounding gas or liquid phase is an essential component
• Shows how insight into chemical bonding at surfaces can be applied to a range of scientific problems in heterogeneous catalysis, electrochemistry, environmental science and semiconductor processing
• Provides both the fundamental perspective and an overview of chemical bonding in terms of structure, electronic structure and dynamics of bond rearrangements at surfaces

• Language: English
• Publisher: Elsevier Science
• Release date: Aug 11, 2011
• ISBN: 9780080551913

Book preview

Chemical Bonding at Surfaces and Interfaces

Anders Nilsson

Stanford Synchrotron Radiation Laboratory, Menlo Park, California, USA, FYSIKUM, Stockholm University, Stockholm, Sweden

Lars G.M. Pettersson

FYSIKUM, Stockholm University, Stockholm, Sweden

Jens K. Nørskov

Technical University of Denmark, Lyngby, Denmark

Elsevier

Table of Contents

Cover image

Title page

Preface

Chapter 1: Surface Structure

Publisher Summary

1 Why surface structure?

2 Methods of surface adsorbate structure determination

3 Adsorbate-induced surface reconstruction

4 Molecular adsorbates — local sites, orientations and intramolecular bondlengths

5 Chemisorption bondlengths

6 Conclusions

Chapter 2: Adsorbate Electronic Structure and Bonding on Metal Surfaces

Publisher Summary

1 Introduction

2 Probing the electronic structure

3 Adsorbate electronic structure and chemical bonding

4 Adsorbate systems

5 Radical atomic adsorption

6 Diatomic molecules

7 Unsaturated hydrocarbons

8 Saturated hydrocarbons

9 Lone pair interactions

10 Summary

Acknowledgement

Chapter 3: The Dynamics of Making and Breaking Bonds at Surfaces

Publisher Summary

1 Introduction

2 Theoretical background

3 Experimental background

4 Processes

5 Summary and outlook

Chapter 4: Heterogeneous Catalysis

Publisher Summary

1 Introduction

2 Factors determining the reactivity of a transition metal surface

3 Trends in adsorption energies on transition metal surfaces

4 The d-band model

5 Trends in chemisorption energies

6 Trends in activation energies for surface reactions

7 Brønsted-Evans-Polanyi relationships in heterogeneous catalysis

8 Activation barriers and rates

9 Variations in catalytic rates – volcano relations

10 The optimization and design of catalyst through modeling

11 Conclusions and outlook

Chapter 5: Semiconductor Surface Chemistry

Publisher Summary

1 Inroduction

2 Structure of semiconductor surfaces

3 Surface oxidation

4 Passivation of semiconductor surfaces

5 Reactions at passivated semiconductor surfaces

6 Adsorption of organic molecules under vacuum conditions

Acknowledgments

Chapter 6: Surface Electrochemistry

Publisher Summary

1 Introduction

2 Special features of electrochemical reactions

3 Electrochemistry at the molecular scale

4 Electrocatalytic reaction processes

5 Summary and outlook

Chapter 7: Geochemistry of Mineral Surfaces and Factors Affecting Their Chemical Reactivity

Publisher Summary

1 Introduction

2 Environmental interfaces

3 Factors affecting the chemical reactivity of mineral surfaces

4 Conclusions

Acknowledgments

Index

Preface

Molecular surface science has made enormous progress in the past 30 years. The development can be characterized by a revolution in fundamental knowledge obtained from simple model systems and by an explosion in the number of experimental techniques. Furthermore, the last 10 years has seen an equally rapid development of quantum mechanical modeling of surface processes using Density Functional Theory (DFT). The methods of surface science have been essential for the birth of nano-science and nano-technology, and more generally we are witnessing a rapid shift of the methods and concepts of surface science into a broad range of scientific disciplines where the interaction between a solid and the surrounding gas or liquid phase is an essential component. The purpose of the present book is to provide a broad overview of chemical bonding at surfaces, and to show how it can be applied in a range of scientific problems in heterogeneous catalysis, electrochemistry, environmental science and semiconductor processing.

We focus in the following on phenomena and concepts rather than on experimental or theoretical techniques, and the aim is to provide the common basis for describing the interaction of atoms and molecules with surfaces to be used very broadly in science and technology. The organization of the book reflects the general approach. We start with an overview of structural information on surface adsorbates and discuss the structure of a number of important chemisorption systems which will be further discussed in the subsequent chapters. In Chapter 2, we describe in detail the chemical bond between atoms or molecules and a metal surface in the observed surface structures. These two initial chapters set the stage for discussing chemical reactions at surfaces in the remaining parts of the book. We begin in Chapter 3 with a detailed description of experimental information on the dynamics of bond-formation and bond-breaking at surfaces. This is followed by an in-depth analysis of aspects of heterogeneous catalysis based on the d-band model, and examples are given of how modern theoretical DFT techniques can be used to actually design efficient heterogeneous catalysts. In Chapter 5, we turn our attention to adsorption and chemistry on the enormously important Si and Ge semiconductor surfaces. In the remaining two Chapters, we leave the solid-gas interface and turn our attention to solid-liquid interface processes by first studying the surface chemistry occurring on the electrodes in electrochemistry and in particular modern fuel cells for clean energy production. In the final Chapter, we give an overview of the environmentally important chemical processes occurring on mineral and oxide surfaces in contact with water and electrolytes.

It is the hope of the whole team of authors that the present effort will assist in providing a coherent and easily grasped picture of the fascinating chemistry occurring at the various surfaces that provide templates for wanted and unwanted catalysis in industry and in our environment.

Anders Nilsson, Lars G.M. Pettersson and Jens K. Nørskov

Chapter 1

Surface Structure

D.P. Woodruff,     Physics Department, University of Warwick, Coventry CV4 7AL, UK

Publisher Summary

Quantifying and understanding the structure of surfaces, and particularly of adsorbates on surfaces, is a key step to understanding many aspects of the behaviour of surfaces including the electronic structure and the associated chemical properties. Methods to determine the structure of surfaces by ab initio methods, in which the structural model and the positions of the atoms are varied to find the lowest energy configuration form the basis of the calculation of the electronic and chemical properties. This chapter illustrates some of the structural phenomena associated with adsorbate bonding at surfaces and shows how (experimental) quantitative surface structure determination can provide insight into the nature of adsorbate bonding at surfaces. It begins with a brief outline of the methods used for adsorbate structure determination. These methods are important to understand the strengths and limitations of the various methods in order to evaluate the data that arise from them. Following this, the study presents a few examples of the way that adsorbates may modify the structure of the outermost atomic layers of the surface onto which they are adsorbed, and the significance of such adsorbate-induced reconstruction. Furthermore, it describes investigations of molecular adsorbates of varying size. Finally, it explores the issues raised by careful quantitative measurements of chemisorption bondlengths, and discusses the insight they give into bonding mechanisms.

1 Why surface structure?

Quantifying and understanding the structure of surfaces, and particularly of adsorbates on surfaces, is a key step to understanding many aspects of the behaviour of surfaces including the electronic structure and the associated chemical properties. For example, any calculation of the electronic structure starts from the structure. Of course, it is now common to try to determine the structure of surfaces by ab initio methods, in which the structural model and the positions of the atoms are varied to find the lowest energy configuration which then forms the basis of the calculation of the electronic and chemical properties. Such methods have become increasingly powerful and effective in recent years, yet experimental tests of these optimised structures are crucial to ensure the integrity of such calculations, and there are certainly clear examples in the literature of the failure of these calculations to reproduce well-established experimental structural trends (e.g., CO on Pt(111) — see Section 4). A particular example of the significance of surface structure in surface chemistry is in the field of heterogeneous catalysis, in which one frequently reads references to ‘the active site’. Underlying such statements is the belief that key steps in surface chemical reactions occur at specific geometrical sites on a surface, and that understanding the nature of these sites could greatly improve our understanding of how to make more efficient catalysts. In those cases in which a catalytic system is found to be ‘structure sensitive’ it seems likely that these active surface sites may be quite specific and thus their availability is dependent on the mode of catalyst preparation.

In this chapter, the objective is to illustrate some of the structural phenomena associated with adsorbate bonding at surfaces and to show how (experimental) quantitative surface structure determination can provide insight into the nature of adsorbate bonding at surfaces. To achieve this, a brief outline of the methods used for adsorbate structure determination is first given in Section 2. Details of these methods are not the focus of this chapter, yet it is important to understand the strengths and limitations of the various methods in order to evaluate the data that arise from them. In Section 3, are presented a few examples of the way that adsorbates may modify the structure of the outermost atomic layers of the surface onto which they are adsorbed, and the significance of such adsorbate-induced reconstruction. Section 4 includes illustrations of investigations of molecular adsorbates of varying size, while in Section 5 issues raised by careful quantitative measurements of chemisorption bondlengths, and the insight they give into bonding mechanisms, are discussed.

2 Methods of surface adsorbate structure determination

2.1 General comments

In this section, some key aspects of the various methods of surface adsorbate structure determination are described. Far more detailed descriptions of the individual methods may be found elsewhere (some relevant references are given), and the objective here is rather to highlight the particular strengths, limitations and special aspects of the techniques which need to be considered when evaluating and comparing the results of applications of these methods. One particular feature which is common to the great majority of these techniques is that the structure is extracted from the experiment through some kind of trial-and-error modelling. In this approach one ‘guesses’ a possible structure and then compares the results of the experiment with the results which would be expected from the guessed structure, through a computation based on the known physical phenomena that underlie the experiment. In many cases it is possible to refine the structural model in an automated and objective fashion by varying the structural parameter values in the model calculation and searching for the best agreement with experiment, typically identified as the minimum value of some kind of reliability- or R-factor. R-factors are commonly based on a sum of the squares of the differences of the experimentally measured and theoretically computed quantities. This type of optimisation, however, is only conducted within a specific structural model. For example, one may adjust the interlayer spacings within the substrate, within a molecular adsorbate, and between the substrate and adsorbate, and may also adjust lateral positions of atoms, but typically within some applied symmetry constraints. It is then necessary to compare the results of such structural optimisations for different structural models. These models may only differ in the lateral registry of the adsorbate of the adsorbate — e.g., adsorption in atop, bridge or hollow sites — but may also include specific models of adsorbate-induced substrate reconstruction, such as changes in the atomic density of the outermost layer or layers of the substrate.

An important general limitation of this approach is that the ultimate structure determination is limited by the imagination of the researcher. If the correct structural model is not tested, the final solution will be the best structure tried, but not the correct one. Indeed, this best structure may differ fundamentally from the true structure. Notice, too, that this limitation also applies to ab initio total energy calculations to determine surface structures theoretically. Here, too, one must start from specific trial models of a structure which can then be optimised.

A second general issue in surface structure determination using the trial-and-error modelling approach is uniqueness. In any optimisation of a structural model one can find an optimal set of structural parameters which defines a minimum in the R-factor. This minimum value may represent a ‘good fit’ but is still not necessarily the correct structure. One can then compare the R-factor values associated with these local minima for different structural models, perhaps resulting in several ‘good fits’. Ideally, one of the structural models gives a significantly lower R-factor. In some cases, however, the goodness-of-fit is similar for more than one best-fit structure. The risk of this problem arising can generally be greatly reduced by ensuring that the size of the data set being used for theory-experiment comparison is large. Large data sets not only reduce the likelihood of this type of ambiguity, but also reduce the size of the variance of the R-factor and thus render significant smaller differences in minimum R-factor values. For this reason the size of the data set is an important issue in determining the reliability of any experimental structure determination, as well as its precision. Of course, there are also situations in ab initio total energy calculations in which two structures have essentially the same lowest energy. In this case one must conclude either that the two structures really are energetically almost equivalent, in which case one expects coexistence of the two structures, or that the computation contains systematic errors in the accurate description of the underlying physics.

2.2 Electron scattering

In many ways the ‘benchmark’ method of quantitative surface structure determination is low energy electron diffraction (LEED) [1—3] This was the first method to be developed in the early 1970s and still accounts for the largest number of catalogued surface structure determinations [4]. A key feature of the technique is that, like conventional X-ray crystallography of bulk solids, it exploits the long-range periodic order of the sample to concentrate the elastically scattered low energy electrons into distinct diffracted beams. This can be both a strength and a limitation. In particular, the scattered electron intensity in the diffracted beams is dominated by contributions for those parts of a surface that have good long-range order, so the technique selectively provides information on these regions. If other regions lack this long-range order, the method is ‘blind’ to them, but also the information on the ordered parts is not distorted by the presence of the disordered regions. Because the elastic scattering cross-sections of atoms at the low energies (∼30—300 eV) characteristic of LEED are very large, multiple scattering plays an important role and the structure can only be extracted through trial-and-error modelling. One further important feature of LEED is that it probes several atomic layers of the near-surface region, so getting a proper fit of experiment and theory requires not only a good description of the adsorbate geometry, but also of the substrate geometry including detailed layer relaxations and rumpling. Indeed, if these substrate relaxations are not well-described in the model, this may introduce systematic errors into the adsorbate geometry. In this sense, LEED gives the complete structure, but it is also important to describe all aspects to be confident of any of the conclusions.

Two rather different techniques that exploit the same underlying phenomenon of coherent interference of elastically scattered low energy electrons are photoelectron diffraction [5] and surface extended X-ray absorption fine structure (SEXAFS) [6,7]. Figure 1.1. shows schematically a comparison of the electron interference paths in LEED and in these two techniques. In both photoelectron diffraction and SEXAFS the source of electrons is not an electron beam from outside the surface, as in LEED, but photoelectrons emitted from a core level of an atom within the adsorbate. In photoelectron diffraction one detects the photoelectrons directly, outside the surface, as a function of direction or photoelectron energy (or both). The detected angle-resolved photoemission signal comprises a coherent sum of the directly emitted component of the outgoing photoelectron wavefield and other components of the same wavefield elastically scattered by atoms (especially in the substrate) close to the emitter. As one changes the collection angle, or the photoelectron energy (and thus the photoelectron wavelength), particular scattering paths switch in and out of phase with the directly emitted component, leading to intensity modulations. These modulations can be interpreted in terms of the structural environment of the emitter, through the use of multiple-scattering calculations for trial structures, in a fashion very similar to that used in the interpretation of LEED data. Indeed, insofar as the method of diffracted beam intensity collection in LEED involves the measurement of the intensity of these beams as a function of electron energy, the scanned-energy mode photoelectron diffraction (PhD) technique is closely similar to LEED, the photoemission intensity being measured as a function of photon, and thus photoelectron, energy. A key difference between LEED and PhD, however, is that PhD is element specific and local. The fact that the photoelectrons are emitted from an adsorbate core level with a characteristic binding energy means that the source of the photoelectron wavefield is known to be a specific elemental atom on the surface, and the structural information is centred on this atom. Furthermore, because the electron source is an outgoing spherical wave the structural information is local to the emitter atom, and does not depend on (or exploit) any long-range periodic order. This means that the technique determines the local structure independent of whether or not long-range order exists in the adsorbate layer; on the other hand, it also is unable to distinguish between those areas of the surface which have this long-range order and those which do not.

Figure 1.1 Schematic diagram showing the electron elastic scattering pathways contributing to the techniques of low energy electron diffraction (LEED), backscattering photoelectron diffraction (including the scanned-energy mode — PhD) and surface extended X-ray absorption fine structure (SEXAFS). Black disks represent substrate atoms, grey-shaded disks represent adsorbate atoms.

SEXAFS shares with photoelectron diffraction the elemental specificity and local character of the structural information content. The key difference between these two techniques is that while in photoelectron diffraction one measures the angle-derivative of the photoelectron emission cross-section, in SEXAFS one measures the total photoionisation cross-section (indirectly through the decay of the core holes created by the ionisation, leading to the emission of X-radiation or Auger electrons). SEXAFS exploits the fact that when the photoelectron wavefield emerges from the absorber atom a small fraction of this wavefield is elastically backscattered to the emitter, where it interferes with the outgoing component to modify the wavefield amplitude at the emitter; this wavefield amplitude enters the matrix element for the photoionisation cross-section as the final state. The total photoionisation cross-section is thus modulated with photon energy as the photoelectron energy changes, causing a change in the photoelectron wavelength, such that the back-scattering leads to alternately constructive and destructive interference. These modulations are the ‘extended fine structure’ of the technique’s name. They provide information primarily on the distance of the emitter from the near-neighbour scattering atoms, although some limited directional information is contained in the way the amplitude of the modulations varies with the direction of the electric vector of the incident X-rays. Because both the source of electrons and the detector (in both cases the photoelectron emitting atom) are local in SEXAFS, the structural data is even more localised than in photoelectron diffraction, and for the same reason the amplitude of the modulations, and thus the ease of achieving good signal-to-noise ratios in the measured modulations, are about an order of magnitude lower. Typically SEXAFS provides accurate nearest-neighbour distances and limited information on the direction of these neighbours and the distances to other near neighbours. In photoelectron diffraction one is significantly more sensitive to non-nearest neighbours, and also obtains more specific directional information because the direction of photoelectron detection influences the scattering path-length differences explicitly. Full structural optimisation in SEXAFS also involves modelling through trial structures, although it is commonly possible to extract good nearest-neighbour distance information directly through Fourier transform methods providing these include corrections for the influence of phase shifts in the electron scattering events.

One final key feature of photoelectron diffraction which is not shared by LEED or SEXAFS is the ability to exploit so-called chemical shifts in photoelectron binding energies of atoms of the same element in different structural and electronic environments to obtain chemical state specificity in the local structural information.

All of these electron scattering techniques are typically capable of determining interatomic distances to a precision of ∼0.02—0.05 Å, with specific cases in which somewhat worse, and occasionally even better, values are cited. For LEED and photoelectron diffraction one commonly finds the best precision for distances corresponding to atomic separations that are near-normal to the surface, with lower precision in locations parallel to the surface, a consequence of the fact that the scattered electrons are generally not detected at very grazing angles relative to the surface.

Because LEED typically involves incident beam currents of ∼1 μA into an area of less than 1 mm², the problem of radiation damage can be severe for fragile adsorbed molecules and some surfaces. This problem can be substantially reduced by using channel-plate amplified systems in the measurement of the diffracted beam intensities to permit the use of incident currents of ∼1 nA. By contrast, incident X-ray techniques have commonly been regarded as less of a problem for radiation damage. However, particularly when using modern third-generation synchrotron radiation sources that are capable of delivering high photon fluxes (∼10¹¹ photons/s) into highly focussed spots (∼50 × 50 μm), there can also be significant damage problems in photoelectron diffraction and SEXAFS unless special precautions, such as defocusing of the incident radiation, are taken.

2.3 X-ray scattering

In contrast to low-energy electrons, X-rays are very weakly scattered by atoms, a property which leads to the success of X-ray diffraction as a means of determining the structure of bulk solids through scattering from atoms over a large depth into the crystals. While this property means that the X-ray scattering signal from surfaces is weak, surface X-ray diffraction (SXRD) [8,9] experiments can be performed experimentally by measuring the surface scattering at locations in momentum-transfer space far removed from those corresponding to diffraction from the underlying bulk. Like LEED, SXRD relies on good long-range periodic order, and indeed the quality of the order required for SXRD is typically higher than LEED in order to ensure that the weak surface diffraction beams are narrow and thus more easily detected above the diffuse scattering background. The benefit of performing these more demanding experiments is that because the scattering is weak, multiple scattering plays no significant role, and direct inversion Fourier transform methods are far more useful. Nevertheless, the final structural refinement generally still involves trial-and-error modelling. The simpler theoretical description also means that it is viable to tackle more complex surfaces involving much larger surface periodicity than in LEED. The intrinsically weak scattering, however, means that it is particularly demanding in SXRD to obtain precise structural information on the very-weakly-scattering low atomic number adsorbates (such as C, N and O) which comprise some of the most chemically interesting adsorbate molecules. We should also note that in many SXRD studies, measurements of the scattered intensities are made mainly at grazing angles (where the signals are largest) which allows one only to determine the relative lateral positions of surface atoms and not the spacing perpendicular to the surface. It is possible to extract such information from SXRD experiments, however, if measurements are made for a wider range of take-off angles (corresponding to so-called ‘rod scans’ in reciprocal space). While SXRD is capable of structural precision of ∼0.01 Å, the actual precision depends strongly on the atoms being investigated and whether the position parallel or perpendicular to the surface is being determined. For low atomic number elements the location perpendicular to the surface may suffer from random errors of 0.1 Å or even significantly more.

A quite different surface structural technique which nevertheless exploits X-ray diffraction is X-ray standing wavefield (XSW) absorption [10—12]. In this technique one uses X-ray diffraction from the substrate to set up an X-ray standing wave with the same periodicity as the substrate scatterer-planes within, and outside, the crystal, due to the interference of the incident and diffracted X-rays. This standing wave can be scanned in a systematic way relative to the substrate scatterer-planes by scanning through the Bragg diffraction condition in either incidence angle or X-ray wavelength. If one measures the X-ray absorption at an adsorbate atom due to this standing wave, in such a scan, one can locate the absorber atom relative to the underlying substrate. Because the X-ray absorption is typically measured by core level photoemission, or by the X-ray fluorescence or Auger electron emission resulting from the refilling of the core hole, the energy of these emissions provides elemental specificity in the structure determination. Indeed, by exploiting chemical shifts in the core level photoemission this technique can provide chemical-state specific structural data. This added specificity is exploited at the lower photon energies typically associated with normal incidence to the Bragg scatterer-planes (NIXSW), and this variant of the technique is applicable to a wider range of materials due to its relative insensitivity to the mosaicity of the substrate crystal. Because the X-ray diffraction exploited in this technique relies only on the long-range periodicity of the substrate, there is no dependence on long-range order in the adsorbate. An important feature of XSW is that the adsorbate atom is determined relative to the extended bulk structure, because the standing wave is established in scattering from very many sub-surface layers. As such, the method provides no direct information regarding the position of the adsorbate atom relative to the nearest substrate atoms, and is completely blind to surface reconstruction (although such reconstruction may be inferred from a combination of the adsorbate location and plausible values of the chemisorption bond lengths). A further significant feature of the method is that the extraction of the basic structural parameters, the so-called coherent positions and coherent fractions, is model-independent. Moreover, in the simplest cases of single high-symmetry adsorption site occupation, the interpretation of these parameters in terms of the actual adsorbate location is trivial and unique. In more complex systems, however, simple modelling is still required to relate the measured structural parameters to a real structure. While precisions as high as 0.01 Å are sometimes claimed for this method, more typical values for adsorbates on surfaces are ∼0.03—0.05 Å.

The radiation damage problems with these incident X-ray methods are similar to those described in the previous section for photoelectron diffraction and SEXAFS, namely that there are potential problems, but they can mostly be overcome by appropriate precautions.

2.4 Ion scattering

Ion scattering methods, covering a wide range of energies from ∼1 keV to ∼1 MeV, and mainly using low atomic number ions such as H+, He+ and Li+, but also often including Ne+ at low energies, have been used in a range of surface structural studies (e.g. Refs. [13,14]). The basic physical principle exploited is of elastic scattering shadow cones, such that atoms behind a scattering atom on the incident ion trajectory may be hidden from the incident beam within a certain range of relative lateral displacements but will scatter incident ions if this lateral displacement is exceeded. The visibility of scattering from these subsurface atoms as a function of incident direction thus provides information of the relative locations of the shadower (surface) atoms and shadowed (subsurface) atoms. Similar effects occur for the outgoing scattered ions, with surface atoms ‘blocking’ the scattered ions from subsurface atoms and preventing them from reaching the detector in certain directions. These methods formally exploit the well-defined crystallography of the surface but not explicitly the long-range order of an adsorbate. They have been used mainly to investigate a range of atomic adsorbate structures and have contributed little quantitative structural information on the local adsorption geometry of molecular species, although at the higher energies they can be particularly effective in investigating adsorbate-induced reconstructions of the outermost substrate layers. The precision of these methods is generally highest for higher energy ions (∼100 keV — referred to as medium energy ion scattering or MEIS) for which the shadow cones are narrowest, when values of ∼0.02—0.03 Å may be achieved. While each ion which scatters from a surface atom causes significant local damage due to the recoil of the scattering atoms, the information on this scattering atom relates to its position before the scattering event. For sufficiently low incident flux density, therefore, these methods can provide information on surfaces essentially devoid of damage induced by the incident beams.

2.5 Spectroscopic methods and scanning probe microscopy

While the methods summarised above are primarily directed to obtaining quantitative structural information on adsorbates on surfaces, a range of other methods may provide valuable qualitative information, yet much of this information must be treated with caution.

Perhaps the most obvious methods are the scanning probe microscopies, of which scanning tunnelling microscopy (STM) is the one most commonly able to offer atomic-scale resolution. Superficially, at least, STM provides a real-space mapping of surface atoms with sub-atomic resolution, so one might wonder why one needs the far more complex and indirect surface structural methods outlined above. The answer, of course, is that STM is a probe of the spatial variations of the surface electronic structure, not of the relative locations of the atomic centres on the surface. The electronic tunnelling probability depends on the overlap of the tails of the electron wavefunctions of the occupied and unoccupied valence states just outside the tip and the surface being scanned, and to a first approximation the surface corrugation obtained in STM is a contour of constant partial electronic density of states outside the surface. For an elemental surface this usually (but not invariably — e.g., [15]) leads to the peaks of the surface protrusions being located above the atom centres, but the amplitude of the surface corrugation has no simple relationship to the relative heights of atoms above the surface, except when comparing the height of symmetrically equivalent atoms (such as those defining the height of a surface step). Moreover, on compound surfaces or elemental surfaces in the presence of adsorbates, even the simple correlation between the lateral position of atoms and atomic-scale protrusions in STM ceases to be reliable. For example, adsorbate C and O atoms on metal surfaces are commonly imaged as dips rather than protrusions (e.g., [16]). Similarly, on the TiO2(110) surface, it is generally believed that the protrusions in the STM images correspond to the surface Ti atoms despite the fact that these atoms lie physically more than 1 Å lower in the surface than the O atoms [17]. Because of these electronic effects it is also not reliable to correlate apparent lateral shifts in atomic protrusions in STM images with real lateral shifts of the underlying atoms.

Despite these very real limitations, which certainly preclude the use of STM as a source of quantitative surface structural information, the technique can play a very valuable role in elucidating surface structural phenomena. For example, in low coverages of adsorbates on a surface (a situation in which other methods may lack sufficient sensitivity) it is often possible to determine the lateral registry of the adsorbate; at least in cases of high-symmetry adsorption sites, one may distinguish atop, hollow and bridge sites. STM images can also be helpful in the case of complex structures, as a source of possible structural models which may be tested in the trial-and-error modelling in quantitative structural methods, although there are clear pitfalls in interpreting such images too literally. However, the most valuable role of STM is in identifying inhomogeneity at surfaces such as step-site adsorption, island growth, coexistent surface structures, and in gaining information on the time-evolution of surface structural changes by such processes as nucleation and growth. For example, in early studies of the structure of the Cu(110)(2 × 1)—O surface phase in which the outermost Cu layer has only half the atom density of the underlying bulk layers there were often debates about ‘where do all the Cu atoms go’ in creating the missing-row structure. Sequential STM images during the evolution of the surface show [18,19] that the phase actually forms by the addition, rather than removal, of rows of surface Cu atoms, but the answer to the converse question which is then raised, namely ‘where do all the Cu atoms come from’, is surface steps.

Quite different information on surface structure arises from some spectroscopies. Most obvious are the vibrational spectroscopies, infra-red reflection absorption and electron energy loss. In these methods the comparison of the behaviour of molecular adsorbates on surfaces with the previously characterised behaviour in coordination compounds has led to the spectral fingerprint being used to infer local geometry. Much the best-known example of this is CO adsorption, the C—O stretching frequency being used to identify single, double and higher coordination adsorption sites (atop, bridge, hollow) by comparison with the considerable body of evidence on metal carbonyls (e.g., [20]). Even for these extremely well-characterised systems, however, this indirect approach to adsorption site determination has been found to be subject to misinterpretation, most conspicuously in the case of the c(4 × 2) phases formed by CO on Ni(111) and Pd(111). In both cases the vibrational spectroscopy was interpreted in terms of bridge site adsorption (Figure 1.2), in part because the vibrational frequency was deemed consistent with bridging sites, in part because the bridge site model leads to an appealing model with a periodic CO overlayer as seen in Figure 1.2. This view defined conventional wisdom for many years until quantitative structural studies by SEXAFS [21], PhD [22] and quantitative LEED [23] on Ni(111), and subsequently by PhD on Pd(111) [24], showed the true adsorption sites to be the (two inequivalent) three-fold coordinated hollows (Figure 1.2). Core level photoemission (X-ray photoelectron spectroscopy — XPS, albeit commonly performed with soft X-ray synchrotron radiation), may also show ‘chemical shifts’ in the photoelectron binding energy of adsorbates which depend on the coordination to the substrate. In most cases this is used only as a spectral fingerprint of the existence of multiple sites, but there has been some success in using these shifts to identify local coordination (e.g., Ref [25]). It is, however, in combination with a true quantitative technique that monitors photoemission, such as photoelectron diffraction and XSW, that these shifts have their greatest value in true surface structure determination.

Figure 1.2 Plan view of the Ni(111)c(4 × 2)—CO surface phase showing the bridge site model favoured for many years on the basis of the interpretation of vibrational spectroscopy, and the mixed-hollow site model subsequently established through SEXAFS, PhD, and LEED measurements. To allow visibility of all surface atoms the C atoms are shown as the larger dark-shaded spheres and the O atoms as smaller white spheres. The full lines show the primitive unit mesh while the dashed lines show the centred (4 × 2) unit mesh.

3 Adsorbate-induced surface reconstruction

In early structural studies of adsorbates on surfaces there was an implicit assumption that the surface provided a rigid chequer board of identical sites into which atoms or molecules were adsorbed, the only structural parameters of interest being the lateral registry and the adsorbate—substrate chemisorption bondlength. It was, of course, understood that the surface could modify the adsorbate species, most obviously through partial dissociation, but also in more subtle ways, because this is the whole basis of heterogeneous catalysis. We now know, of course, that the adsorbate also induces changes in the substrate surface. In some cases this effect is quite subtle. The simplest example is just a modification of the relaxation of the surface layer(s). The outermost atomic layer(s) of a solid generally have layer spacings which differ from that of the underlying bulk as a consequence of the termination of the solid; typically the outermost layer spacing is contracted, the second layer spacing expanded and so on, although the amplitude of this relaxation damps rapidly with depth. For a close-packed low-index surface even the outermost layer spacing change may be only ∼1%, although for a more open-packed low-index surface such as fcc(110) the outermost layer spacing change may be ∼10% or more. Not surprisingly, these relaxations will change when material, including an adsorbate, is added to the surface. Typically the size of the clean surface relaxation is reduced, but in some cases larger changes may occur. This effect may also be local to the adsorbed atom. Consider, for example, the case of atomic O on Ni(100) [26,27]. At a coverage of 0.5 ML an ordered c(2 × 2) phase is formed in which the O atoms occupy alternate four-fold coordinated hollow sites in the surface (Figure 1.3). This means that in the second substrate layer half of the Ni atoms have an oxygen atom directly above them while the other half have no such O near-neighbour. This leads to a ‘rumpling’ of the second Ni layer, with the Ni atoms below the O adsorbates being 0.035 Å lower than those that are not covered in this way. This effect is marginal but detectable with LEED. Interestingly, there is also a (2 × 2)-O surface phase of Ni(100) in which the O coverage is only 0.25 ML, so only every fourth second layer Ni atom is covered by an O atom and of the remaining ¾ of the second layer atoms there is ¼ of them that are more distant from the O adsorbates than is possible in the c(2 × 2) phase. In this case the overall rumpling amplitude of the second Ni layer is larger with a value of 0.100 Å. This leads to an interesting speculation: would the effect be even larger for isolated O atoms at low coverage? The answer is not known; the great majority of such subtle substrate distortion measurements have been made by LEED on surfaces with good long-range order in the adsorbate and with relatively small adsorbate—adsorbate distances.

Figure 1.3 Plan view of the Ni(100)c(2 × 2)—O and Ni(100)(2 × 2)—C p4g surface structures. In each case the full lines show the primitive unit mesh while in the O-induced structure the dashed lines show the centred (2 × 2) mesh. In the case of the C-induced structure the outermost Ni atoms are shown as smaller more-darkly shaded spheres than those of the underlying substrate to see more clearly the relationship of this reconstructed layer to the substrate. Notice that this reconstruction also leads to some reduced Ni—Ni nearest-neighbour distances in the surface, so using the bulk atomic radii for these atoms would lead to some overlapping spheres.

A somewhat related question has been addressed, however, for the case of C on Ni(100). In this case at high coverage the influence of the adsorbate on the structure of the surface atomic layer of Ni is far more significant than these subtle changes in interlayer spacings. Figure 1.3 shows a schematic model of the structure formed by 0.5 ML of C on Ni(100). While the C atoms occupy alternate four-fold coordinated hollow sites just like O in Ni(100)c(2 × 2)—O, the Ni atoms of the outermost substrate layer are displaced tangentially within the surface plane relative to the C atoms such that the groups of four Ni nearest-neighbours of the C atom expand outwards and rotate alternately clock-wise and counter-clockwise to produce a larger (2 × 2) periodicity, but with characteristic glide symmetry lines which can be identified in the LEED pattern as being associated with a p4g space group [28]. The geometry of this structure, commonly referred to as the ‘clock’ reconstruction because of the clockwise and counter-clockwise rotations, is now rather well established by LEED [28,29], PhD [30,31], and SEXAFS [32]. The C atoms are almost coplanar with the reconstructed outermost Ni layer, and this layer expands significantly outwards away from the underlying bulk. However, the detailed mechanisms underlying this reconstruction have proved more controversial [33]. Early qualitative discussions centred on the idea that the C atom (unlike O in the unreconstructed Ni(100)c(2 × 2)-O phase) is almost small enough to penetrate the hollow site, and can do so when the hollow site is enlarged by the reconstruction. The idea that the C atom also forms a strong bond with the second layer Ni atom directly below as a result of this penetration has also been discussed, and at least one total energy calculation identifies this second layer bonding as very significant [34]. Experimental and theoretical studies of the vibrational properties of this surface led to the conclusion that the reconstruction can be regarded as a freezing of a specific nickel surface phonon mode [35—37], potentially clarifying the mechanism of the reconstruction, but this still does not provide a clear picture of the underlying driving force. It has also been shown that the reconstruction leads to a partial relief of adsorbate-induced compressive surface stress associated with the C adsorption [38]; the adsorbate-induced change in surface stress as a function of C coverage shows that the onset of the reconstruction around 0.3 ML halts the rise in compressive stress, presumably helping to lower the total energy.

One possible route to gaining more insight into the mechanism of this reconstruction is to compare the structure in this 0.5 ML ordered reconstructed surface with the local structure around C atoms adsorbed at lower coverage when the reconstruction does not occur. Notice that one feature of the higher coverage ‘clock’ reconstruction is that the local distortions of the Ni surface periodicity involve concerted interlocking movements that are only possible in a periodic structure; the same distortions applied locally around an isolated C atom would lead to much shorter Ni—Ni distances on the periphery of the distorted region. In this regard an STM study by Klink et al. [39] appeared to provide the key information. As expected, STM images at a C coverage of 0.5 ML, were found to show direct evidence for the lateral distortions of the (2 × 2) p4g-phase described above. At low coverage, however, images indicated that the C atoms adsorb in hollow sites without inducing a local clock reconstruction. On the basis of quantitative analysis of the STM images and associated line-scans, these authors deduced that there is a local lateral (radial) outward relaxation of the nearest-neighbour Ni atoms around these isolated adsorbed C atoms of 0.15 Å (albeit with an estimated precision of only ±0.15 Å). This local strain was interpreted as a signature of Ni—C near-neighbour repulsion and was taken as evidence that this repulsion is the key to the reconstruction and the compressive stress increase in the absence of reconstruction. A quantitative structure determination of the local environment of the C atoms at low coverage by PhD, however, showed that no such radial relaxation occurs in the Ni top-layer neighbours of the C atoms [31], implying that the apparent Ni atom displacements seen in the STM images are a result of the local electronic effects around the C atoms. Interestingly, the PhD structural study showed that the C—Ni nearest neighbour distances to both outermost and second layer Ni atoms were unchanged in the reconstruction; in the absence of the reconstruction the C atoms sit slightly higher above the (smaller) four-fold coordinated hollow site, while the Ni atom directly below the C atoms moves down to maintain a constant N—C bondlength when the reconstruction occurs. Indeed, the structural signature of the reconstruction in terms of bondlengths was found to be a reduction of the Ni—Ni distance within the outermost layer (which must be induced by the bonding to the C atoms), rather than any change in the Ni—C bondlengths.

Another example of adsorbate-induced surface reconstruction in which there have been studies of the structure of the precursor to the reconstruction at low coverages is the case of atomic oxygen on Cu(100). The ability of chemisorbed oxygen to form a missing row (or equivalently in terms of the equilibrium structure, an added row) reconstruction on Cu(110) has already been mentioned in 2)R45°—O structure is formed by the creation of missing rows, the rejected Cu atoms forming islands one atomic layer higher with the same reconstruction [45].

Figure 1.4 2)R45°—O surface reconstruction. The outermost layer Cu atoms are shown more lightly shaded than those of the underlying substrate to show more clearly the missing-row structure of this outermost layer. The full lines show the surface unit mesh.

2)R45°—O phase might be difficult to see if the long-range ordering is poor [46,47]. More recently, however, STM studies [48,49] appear to have resolved this issue, in that it appears that for low oxygen coverages the adsorbate forms extremely small ordered islands with a very high density of antiphase domain boundaries, such that there is little true long-range order but within the islands the periodicity is c(2 × 2). Such a surface phase is particularly easy to reconcile with earlier reports of a so-called ‘four-spot’ LEED pattern [50—52], with the general appearance of a c(2 × 2) phase but with splitting of the ½-order diffracted beams which could be attributed to antiphase domain boundaries.

2)R45°—O phase the O atoms lie only about 0.1 Å above the outermost layer Cu atoms, in the low coverage phase these atoms lie about 0.7 Å above (unreconstructed) four-fold coordinated hollow sites [43,44]. There have been a number of theoretical studies of the Cu(100)/O system aimed at elucidating the mechanism for the reconstruction; total energy calculations do show the reconstructed phase is energetically preferred, although the energy difference is surprisingly small (values of 0.3 eV [53] and 0.17 eV [54] have been found). One proposed mechanism relates to the large effective charge, anticipated to be on the adsorbed O atoms in the c(2 × 2) phase, which leads to a large surface dipole moment when the O atom is significantly above the surface. This may drive a phonon instability [55], but more recent work has cast some doubt on the true value of this charge and, indeed, even the proper definition of this effective charge [54]. Moreover, very recent experimental and theoretical work supports the view that relief of compressive surface stress plays a role in this reconstruction [56].

A quite different class of adsorbate-induced surface reconstruction is formed by those systems involving pseudo-(100) reconstruction of the outermost atomic layer; this behaviour has been found to occur on fcc(111) and (110) surfaces in several metal/adsorbate combinations. The essential driving force for such reconstructions appears to be that adsorption on a (100) surface (typically in a c(2 × 2) arrangement) is so energetically favourable that, even on a surface with a different lateral periodicity (and point-group symmetry), reconstruction of the outermost layer or layers to form this (100)-like geometry is favoured. This must occur despite the introduction of strain energy at the interface between the substrate and the reconstructed layer(s). Most of the examples of this phenomenon occur on fcc(111) surfaces, but atomic N adsorption on both Cu(110) and Ni(110) appears to involve essentially the same type of reconstruction [57]. In most cases these reconstructions are induced by atomic adsorbates, but Figure 1.5 shows a schematic diagram of the pseudo-(100) phase formed by methanethiolate, CH3S—, on Cu(111). For simplicity the methanethiolate species, formed by reaction of the surface with methanethiol (CH3SH) or dimethyl disulphide ((CH3S)2), are simply represented in the figure by the S atoms alone. The fact that this adsorbate induces a major density-lowering reconstruction of the Cu(111) surface was established more than 15 years ago by SEXAFS and NIXSW [58], but it is rather more recently that STM and qualitative LEED have shown that this reconstruction creates a near-square arrangement of surface Cu atoms [59], while MEIS has provided direct evidence of the existence of the reconstructed outermost single Cu layer [60].

Figure 1.5 registry of the overlayer and substrate. The methylthiolate species are represented by the S head-group atoms alone, shown as the darkest spheres. The Cu atoms of the reconstructed pseudo-(100) layer are shown more darkly shaded than those of the underlying substrate. For clarity the reconstructed overlayer has been omitted from the lower right-hand side of the diagram, exposing the outermost unreconstructed Cu(111) layer.

mesh has been proposed, but even if this structure is truly commensurate the surface unit mesh is 15 times the area of the substrate unit mesh, and has five different thiolate/substrate registry sites within the structure (3)rect. commensurate phase has been proposed, the Cu—Cu distances are 4% larger than that in a Cu(100) surface in one direction and almost identical to the ideal (100) surface in the other [61,62]. Within this commensurate unit mesh are 144 different reconstructed Cu atoms and 72 N atoms, most of which could be in different registry sites. With this degree of complexity it is easy to see why there are no fully quantitative surface structure determinations of any of these systems, and indeed no ab initio total energy calculations. The rationale for the existence of these surface phases given above, however reasonable it may seem, must therefore formally remain as speculation.

More generally, the existence of these surface phases may be related to the superficially quite different adsorbate-induced surface restructuring phenomenon of faceting [57]. Here, too, the driving force is the energetic favourability of chemisorption onto one or more specific orientations of a substrate, but the restructuring involves a modification of the surface morphology into a ‘hill-and-valley’ corrugated structure in which at least one of the sides of the ‘hills’ comprises planar facets of the preferred orientation. This basic phenomenon of surface faceting has a long history and was traditionally observed on a microscopic scale with conventional optical microscopy. The surfaces present on the restructured faceted surfaces are those to be seen on the equilibrium shape of a small particle of the substrate material. This shape is related to the polar diagram of the surface free energy as a function of orientation (the so-called γ-plot) by the Wulff theorem, and facet planes occur at orientations corresponding to cusps in the γ-plot. An interesting unanswered question is why pseudo-(100) reconstructions of fcc(111)surfaces occur in preference to (100) faceting if the (100)/adsorbate surface structure is so energetically favoured. Of course, the creation of a hill-and-valley surface corrugation leads to a significant increase in surface area, so the reduction in specific surface free energy must more than compensate for this area increase in order for faceting to occur. Evidently, in these systems, the strain energy cost at the substrate/pseudo-(100) layer interface is less than the energy increase in faceting due to the increase in area.

2)R45°—O surface reconstruction. Curiously, however, the Cu(410)-O surface also has a similar coverage of additional oxygen atoms coadsorbed in four-fold coordinated hollow sites in the middle of the (100) terraces (see Figure 1.6) — the very sites which appear to be unstable at high coverage on the extended Cu(100) surface. Notice, though, that if surface stress relief plays a role in destabilising the Cu(100) surface with these adsorption sites occupied, as has been suggested [56], the narrow terrace width of the (410) surface means that local lateral strain offers an alternative means of stress relief on this vicinal surface. Density functional theory (DFT) ab initio calculations and SXRD experiments [63] support the structure of Figure 1.6, but there has been no systematic theoretical study to understand why the (410) surface, with this particular terrace width, is so strongly favoured over other vicinal orientations for quite a wide range of average initial surface orientations.

Figure 1.6 Plan and side views of the structure of the Cu(410)—O surface phase. The full and dashed lines show respectively the primitive and centred rectangular surface unit meshes (that are unchanged from those of the clean surface by the O adsorption).

4 Molecular adsorbates — local sites, orientations and intramolecular bondlengths

4.1 General issues and the case of CO on metals

Molecular adsorbates provide an increasingly complex set of challenges to surface structure determination, the complexity growing steeply with the number of atoms within the molecule. A special problem is also presented by hydrogen atoms which are present in the majority of chemically interesting molecules, but to which all truly quantitative experimental structural methods are, to a greater or lesser extent, ‘blind’. In particular, H atoms are extremely weak scatterers of electrons and have no electronic core level. While there are a few examples of structure determination of atomic H adsorption phases, and a few molecular adsorbate structural studies in which marginal effects due to the presence of the H atoms have been established, there are no structurally significant determinations of the H atom positions within adsorbed molecules.

As in many areas of surface science, the most-studied molecular adsorbate structurally is CO; from a structural point of view this is a simple molecule (just two distinct atoms), while the chemistry of CO on transition metal surfaces, in particular, is chemically important in the context of heterogeneous catalysis. CO adsorption on such surfaces also shows some fascinating complexity. In particular, while it almost invariably bonds end-on to the surface through the C atom, it may adopt one-fold (atop), two-fold (bridge) or three- or four-fold (hollow) coordination sites on surfaces, with the relative energetics of these different sites being dependent on the substrate material, the substrate orientation, and the CO coverage. For example, the initial (lower coverage) adsorption site of CO on Ni(100) is atop, while on Ni(110) it is bridge and on Ni(111) it is hollow. Moreover, while CO forms a 0.5 ML coverage c(4 × 2) ordered phase on several fcc(111) metal surfaces, the local adsorption sites in this phase on Ni(111) and Pd(111) are the two inequivalent three-fold coordinated hollow sites (see the discussion in Section 2.5 and Figure 1.2 [21—24]), whereas on Pt(111) it is the atop and bridge sites which are co-occupied [64—66]. A further manifestation of the subtle energy changes involved in these different bonding sites is seen in the coverage dependence. For example, on Pt(111) and Rh(111), CO adsorbs at low coverage in atop sites only, but as the coverage increases both atop and bridge sites become occupied; indeed, on Ni(100) it appears that initial atop adsorption at low coverage changes to atop and bridge site occupation at intermediate coverage and then to pure bridge site occupation at the highest coverage [67]. Electronic structure measurements for this system have been interpreted as indicating that the energetic differences between these sites are very small, a consequence of increased π-bonding and increased σ-repulsion with increasing substrate-atom coordination number [68] (see also Chapter 2). It is perhaps not surprising, in view of this subtlety, that DFT ab initio calculations have been unable to reproduce all of these effects correctly, with hollow rather than atop sites being favoured at low coverage on both Pt(111) and Rh(111); this theoretical problem has attracted considerable debate and some proposed solutions (e.g., [69—72]).

One structural parameter of potential interest in the adsorption of simple diatomic molecules, such as CO, NO and N2, is the intramolecular bondlength. Much of the motivation for studying such adsorbates is related to the way adsorption modifies the chemistry of these species as the basis for heterogeneous catalysis. Many such reactions involve scission of the intramolecular bond, and if the adsorption is of the molecular species, one might expect the relatively greater ease of bond scission to be reflected in the molecular precursor by a weakening, and hence a lengthening, of the intramolecular bond. Indeed, quite generally, the formation of the chemisorption bond might be expected to weaken the intramolecular bond. Unfortunately, it seems that the precision achievable in current surface structural