Surface of ice as viewed from combined spectroscopic and computer

May 17, 1995 - question of the liquid interfacial layer1 is muted. Nevertheless, the surface of ice, whether crystalline or amorphous, is a complex di...
7 downloads 0 Views 2MB Size
J. Phys. Chem. 1995,99, 16534-16548

16534

FEATURE ARTICLE Surface of Ice As Viewed from Combined Spectroscopic and Computer Modeling Studies J. Paul Devlin” Department of Chemistry, Oklahoma State University, Stillwater, Oklahoma 74078

V. Buch Department of Physical Chemistry and the Fritz Haber Research Center, The Hebrew University, Jerusalem Received: May 17, 1995; In Final Form: August 31, 1995@

The structure of the ice surface and its interaction with adsorbates are investigated by several experimental tools, combined with computer modeling. Spectroscopic features characteristic of icy surfaces were identified and assigned. Adsorbate spectroscopy is used to probe both the adsorbate layer and the ice surface structure. These results are potentially informative of basic questions, such as cooperative aspects of H-bonding and the mechanism of ice vaporization, and of diverse practical questions, such as the role of icy particles in atmospheric chemistry and physics as well as the chemistry of interstellar space. Methods are described for the preparation and spectroscopic study of microporous amorphous ice and cubic ice nanocrystals with surface to volume ratios that make it possible to obtain low-noise infrared and Raman spectra of the vibrational modes localized near the surfaces and of the fundamental modes of small molecule adsorbates. The assignment of the bands of several of the surface-localized modes is reported, on the basis of primarily the calculated vibrational excitations for simulated structures of both amorphous and crystalline ice. The usefulness of these spectra is enhanced by conversion to difference spectra that compare high surface area and low surface area samples. Bands have been assigned to each of the three important types of surface water molecules, as revealed by the simulated structures and spectra: molecules with non-H-bonded or dangling-H(D) atoms, molecules with a dangling-0 coordination, and 4-coordinated surface water molecules. The experimental difference spectra have also been used to display the influence of small adsorbate molecules on surfacelocalized vibrations of each type of water molecule. This influence is apparent through the shifting and enhancement of bands of surface-localized modes, the response of the modes of the adsorbate molecules, and the determination of site-selective heats of adsorption of small molecules using the assigned ice modes. Computer modeling in conjunction with ab initio calculations was used to analyze and interpret adsorbate spectra and to elucidate the influence of factors such as the extent of surface disorder on gas-surface interactions. The results suggest significant modification of the ice surface structure with respect to the cubic crystalline interior, toward loss of lateral order.

I. Introduction The fundamental properties of the surface of ice are the subject of the present article. When these properties are considered at temperatures below 250 K, the longstanding question of the liquid interfacial layer’ is muted. Nevertheless, the surface of ice, whether crystalline or amorphous, is a complex disordered system with a greater molecular mobility than that of bulk ice, a system that we can only hope to understand fully through the convergence of information from many sources. Fortunately, partly because of the great sensitivity of the frequencies of H20 stretching modes to hydrogen bond strengths, vibrational spectra represent a remarkably versatile probe of the molecular environment at ice surfaces. A fuller appreciation of the information in the spectroscopic data has emerged during the past few years because of advances in the computational simulation of the structure2 and spectra3 of complex extended (H-bonded) systems. Here, using a combination of spectroscopic and simulation results, we focus attention on insights gained into the interrelated Abstract published in Advance ACS Abstracts, October 15, 1995.

0022-365419.512099-16534$09.00/0

structural, dynamic, and adsorbent properties of bare ice and ice covered with small-molecule adsorbates. In particular, we examine the behavior of the intramolecular modes of H2O (D20) molecules at ice surfaces that range from that of crystalline ice, for which bulk structural order influences even the ice surface bilayer,“,5to the more fully disordered surface of microporous amorphous ice formed from the vapor under energy-limited conditions (12 K).6 Through the use of techniques that produce samples with large surface-to-volume ratios, high-quality FT-IR spectra of the surface molecules for a considerable range of ice structures, both bare5%’-’ and complexed with a variety of interesting smallmolecule adsorbate^,^.'*-^^ have been determined in recent years. A computational scheme developed by us to study the anharmonic OH stretch excitations in ice was used to assign the vibrational modes of surface water molecules5 (referred to as surface-localized modes to differentiate from sample shapedependent “surface modes”). At the same time this combination of theory and vibrational spectroscopy gives insights to the ice surface structures and the surface reconstruction. Further, analysis of changes in the ice surface modes effected by 0 1995 American Chemical Society

Feature Article

J. Phys. Chem., Vol. 99, No. 45, I995 16535

Figure 1. Simulated surface of ice, top bilayer: (a) perfect crystalline (1 1 1) surface; (b) relaxed surface, obtained by simulated heating of the crystalline surface to 200 K and recooling to 83 K, (c) more disordered surface model obtained by heating to 250 K and recooling; (d) surface c covered by a monolayer of CF4 adsorbate simulated at 83 K and 0.06 Torr.

(b) adsorbates, as well as surface perturbations of the vibrational spectra of the small adsorbate molecules, adds significantly to the general understanding of the nature of the ice surfaces. The different ice surfaces (Figures la-c and 2a) each contain three subsets of water molecules that are recognizable from the simulation^^.^ and invoked in the interpretation of the experimental spectra: two- or three-coordinated molecules with a dangling hydrogen, two- or three-coordinated molecules with a dangling oxygen coordination, and four-coordinatedmolecules with distorted tetrahedra. (These three types of surface molecules will be identified throughout this article as d-H, d-0, and s-4. The high-energy two-coordinated molecules, which almost invariably have both a d-H and a d-0, are only common for the microporous amorphous ice surfaces at low temperatures). Recognition of the surface-localized modes of these subsets of surface water molecules enables us to characterize site-specific and site-selective interactions of adsorbate molecules with the ice surface and to determine site-specific optical adsorption isotherms.l9 Why Study Ice Surfaces? To our knowledge, ice is the only simple molecular solid for which surface-localized modes have been characterized for nanocrystalline and larger particles, though vibrational modes observed in smaller clusters (including (H20)n)20undoubtedly bear some relationship to surface molecule vibrations; and there is an emerging body of spectroscopic data for the surfaces of zeolites and related materials.2* Ice is also uniquely accessible to computational modeling, due to the availability of extensive information on the intermolecular interactions in water-containing systems. Thus, among molecular substances, ice is unquestionably unique in the richness of surface information accessible through spectroscopic and simulation studies and could be regarded as a model system for the study of the surface properties of molecular solids. The ultimate objective of such studies may include insights to (a) surface

-

Figure 2. (a) Model amorphous cluster ( H 2 0 ) 4 ~ 0 generated by simulation of vapor condensation at T 20 K.& (b) Same cluster, covered by a simulated overlayer of H2 adsorbate at 12 K and 0.5 pm.

structures of ices formed under different preparative conditions, (b) binding sites and binding mechanisms for binding of adsorbates of different molecular properties (e.g. hydrophobic, hydrogen bonding, polar, acidic) to the ice surface, (c) reactivity of adsorbates, e.g. with respect to ionization, and its dependence on factors such as surface quality and structure, temperature, etc., (d) the formation of strained bonding configurations during surface reconstruction (in particular of H-bonded solids), (e) the mobilities of surface and adsorbate molecules as a function of temperature, (f) mechanisms of surface reconstruction and vaporization, (g) H-bonding cooperative effects as revealed in the strength of surface complexes with adsorbate molecules, and (h) the influence of surfaces/adsorbateson interior molecular mobilities/relaxation. In addition to these fundamental interests, the interaction of molecules with the surfaces of crystalline ice, amorphous ice, and icy acid hydrates has emerged as a prime concern of diverse areas of scientific study. Ice and acid hydrate particles are generally recognized as key players in the chemistry of the stratosphere with the surfaces, sites of which serve to concentrate and activate otherwise stable molecules, of particular interest. As a result, several recent studies have focused on the behavior of molecules like CFxC1,,'4 HN03F2 HC1,14*23 and N20524 exposed to icy surfaces. Similarly, on the basis of the available evidence, one can legitimately refer to the surface of amorphous ice as an important chemical factory of interstellar space; particularly in dense interstellar clouds, where molecular formation and reaction are thought to often depend on the concentration of atoms and molecules on/within "dirty" amorphous ice that has accreted on refractory dust particles.25 Supporting evidence is accumulating, such as the recognition of the 3 pm

Devlin and Buch

16536 J. Phys. Chem., Vol. 99,No. 45, 1995 ice band26 and of the vibrational fundamental of H2 adsorbed on amorphous ice,27both having been observed in absorption toward stellar objects. Closely related to the interstellar ice chemistry is the chemistry of comets, since comets are thought to be aggregates of relatively unprocessed ice-covered dust particles of the presolar cloud from which the solar system was formed.** The ice surface also plays a role in the icing of materials on earth, and its interplay with crystalline point defects may be involved in the generation and interparticle transfer of charge that ultimately leads to electrical displays in storms.29

11. Experimental Methods of Study of Ice Surfaces Articles have reported s u ~ a c espectra of thin films of amorphous/crystalline ice on metal surfaces using infrared reflection However, the infrared methods that we have used, to obtain the surface spectral data for both amorphous and crystalline ices described in this article, depend on the measurement of FT-IR adsorbance spectra by direct transmittance. The procedures used for the sampling of amorphous ice and ice nanocrystals differ markedly and will be described separately. Measurement of Surface Spectra of Amorphous Ice. Mayer and Pletzer showed in an early study that amorphous ice prepared by slow deposition of water vapor in vacuum onto substrates held at temperatures below -90 K is “microporous”, as deduced from BET gas adsorption measurement^.^' More recently it was recognized that the large surface-to-volume ratio associated with the microporosity makes it easy to observe spectra of the surface-localized modes of pure amorphous ice’ (although some sign that this was possible was noted as early as 1976).3’ Our series of infrared and Raman studies can be divided into measurements of the surface of pure amorphous ice and of the surface covered with small molecular adsorbates. The FT-IR spectra of the bare surface of the nanometer-sized pores and channels have been studied using standard low-temperature vapor-deposition method^.^ As prepared at 12 K using deposition rates of -10 mirdpm, films of amorphous ice a few microns thick have approximately 5% of the water molecules located at the surface of the nanopores. The Raman spectra of such deposits have been measured using an interference-enhanced internal-reflection technique in which the laser beam impinges from the substrate side of the interface. The constructive interference gives rise to strong Raman signals even for a poor Raman scatterer, such as amorphous ice, or dilute scatterers, such as adsorbed H2.33 Both infrared and Raman measurements of 12 K amorphous ice deposits have been used to monitor the uptake of molecular adsorbates. Only the gases Hz and Ne have been observed to permeate the microporous ice at 12 K. At temperatures below 15 K, gases such as N2, CO, C h . CF4, HC1, and SO2 form separate solid films that overlay the amorphous ice. However, molecules from such films penetrate to and adsorb on the pore walls when the ice is warmed above a characteristic temperature that ranges from 18 K for N2 to 55 K for CF4.I5 Other small molecules, such as COz, CzH2, C2H4, and benzene, do not diffuse into the pores regardless of the sample temperature but can be studied on an amorphous surface by using a less attractive approach: simultaneous codeposition with the water vapor.I6 The equilibrium uptake of gases other than H:! has been studied at higher temperatures with most results obtained in the 80-90 K range using pressures varying from hundredths of a torr to 1 atm. Such studies have required the use of a double cell (resembling the cluster cell of Figure 3) with the amorphous ice deposited on an end window of an inner metal cell.

\

NANOCRYSTALS

Figure 3. Cartoon of the cluster cell.

Measurement of the Surface Spectra of Crystalline Cubic Ice. Our studies with ice nanocrystals have depended on their formation in the gas phase following the rapid expansion of a -1% mixture of H20 (D20) in N2(g) or He(g) into a cluster cell precooled to -75 K. The cylindrical cluster cell (Figure 3), which is 5 cm in diameter and 15 cm in length, is a miniaturized version of the static cluster cell first described by Ewing et al.34 The original spectroscopic studies were of aerosols of suspended nanocrystals, but the limited lifetimes of the aerosols (5-15 min) restricted the number of FT-IR scans, made equilibrium gas-adsorption studies impossible, prohibited studies of a single sample at other than a single temperature and ruled out the adjustment of ice particle size, degree of crystallinity, or surface regularity through sample annealing. More recent studies have capitalized on a much more versatile sampling approach that is based on the collection of the nanocrystals as a suspended network of particles on the end windows of the cluster cells. The resulting samples, which have many desirable characteristics that the gas-phase samples lack,5~’~ produce spectra that are not distinguishable from that of the gas-phase samples. On the basis of the band intensity of the d-H surface-localized vibration,’ the average effective particle size remains near 20 nm and volatile adsorbates have full access to the particle surfaces. Such samples are stable over periods of weeks at temperatures below 110 K and have been used repetitively to study different nonreactive volatile adsorbates without significant degradation. Further, upon warming above -120 K the smaller nanocrystals volatilize and their water molecules are transferred to the surfaces of the larger particles within the network of nanocrystals. This permits the measurement of very useful difference spectra for samples containing effectively the same amount of ice at the same temperature but with different ratios of surface-to-bulk water. Such difference spectra reflect the conversion of surface molecules to bulk molecules and thus permit the identification of surface-localized vibrational modes.5 111. Computer Simulation Methodologies The computational methods used to characterize ice surfaces and the interactions with small molecule adsorbates have been as varied as the experimental spectroscopic methods/objectives. Here we present brief descriptions of methods developed and used for quite different purposes ranging from characterization of the structural and dynamic aspects of ultracold microporous ice, modeling of surface-site distributions and vibrational/ rotational spectra of adsorbed H2, and prediction of the vibrational spectra of variably relaxed surfaces of crystalline

Feature Article and amorphous ice and of CF4 adsorbed on such varied surfaces. In addition, ab initio calculations were used to map and analyze potential energy surfaces and vibrational frequencies of water mixed dimers with small gaseous molecules.35 The calculations were carried out using second- and fourth-order perturbation theory to account for electronic correlation, and the analysis was carried out using computational tools developed in ref 36. The dimer potentials were used to construct model potentials for the adsorbate-ice system^.^^.^^ Moreover vibrational analysis of the dimers was shown to provide interesting insights into vibrational spectroscopy of the adsorbate-ice surface systems. For Vapor-Deposited Amorphous Ice. The experimental formation of amorphous ice through the slow condensation of water vapor in vacuum at low temperature was simulated by solving the classical equations of motion for a sequence of water molecules impinging at random on a water cluster.6 A pairwise TIPS2 intermolecular potential was employed along with ridgid water molecules. The temperature of the cluster was rescaled to 10 K after each molecular condensation to represent the experimental removal of condensation energy by a cold substrate. An analysis of the simulated cluster growth revealed much about the formation dynamics of the amorphous ice. The dominant factor determining the highly strained amorphous ice structure is the inhibition of relaxation in the low-temperature condensation. Roughly speaking, a molecule freezes instantaneously upon sticking to the surface, with only very limited local relaxation toward improvement of the hydrogen bond network of the ice during the growth process. For the purposes of this article the critical factor is that the directional nature of the hydrogen bonds resulted in the formation of an irregularly shaped cluster, (H2O)450, with a highly convoluted surface structure (Figure 2a), sections of which are suggestive of incipient nanopores, as have been deduced from experimental gas-adsorption measurement^.^' The convoluted surface can be characterized as rich with two- and three-coordinated water molecules (with d-H(D) and/or d - 0 sites) though a large number of s-4 molecules with distorted tetrahedra are also present. The simulated annealing of amorphous ice, using trajectories of the (H20), cluster with thermal energies corresponding to much higher temperatures, was marked by a strong reduction in the relative number of two-coordinated water molecules and an increase in the s-4 molecules, effects that accompany a general reduction of the surface irregularity as surface molecules scramble to achieve higher coordination number^.^ For HZAdsorbed on Amorphous Ice. Several aspects of the behavior of H2 molecules adsorbed on amorphous ice have been s i m ~ l a t e d ' ~using . ~ ~the , ~ amorphous ~ ice cluster (H20)m to represent the surface texture (see Figure 2). Our most advanced mode133.37employed the path integral Monte Carlo (PIMC3*) technique, which was adjusted to include nuclear symmetry effects for quantum rotor H2 molecules, and treatment of the open adsorbate-surface system. This methodology was used to reproduce and analyze the experimentally observed preferential adsorption of ortho-H2 with respect to para-H2 on the ice surface3' and to interpret the rotational Raman line shapes of the hydrogen adsorbate in terms of random fluctuations in gas-surface interactions. Moreover PIMC calculations of hydrogen uptake on amorphous ice under astrophysically relevant conditions, combined with laboratory measurements, were used39 to interpret the origin of the recently observed interstellar condensed H2 infrared feature.27 The gas-surface interaction used in these calculations was constructed by assuming pair-additivity,and deriving the H2..-H20 pair interaction from ab initio calculations.35b

J. Phys. Chem., Vol. 99, No. 45, I995 16537

For Modeling the Surface-LocalizedModes of Cubic Ice. Our recent spectroscopic and simulation studies have focused predominantly on the vibrational modes of the surface molecules of crystalline cubic ice, the type of crystalline ice that normally forms at temperatures below 200 K. The simulations of the surface spectra have been carried out using an adaptation of a methodology originally developed to interpret the vibrational spectra of bulk cubic and amorphous ice and the effect of isotopic dilution on those spectra.3c Success in satisfactorily reproducing the great range of spectra available from the dilution of H20 ices with intact D20 molecules encouraged the extension of the approach to an analysis of the surface-localized modes. A primary challenge in simulating the surface-localized modes of cubic ice is to scale the degree of surface relaxationl reconstruction to legitimately represent the surface of the nanocrystalline ice particles that are formed experimentally by very rapid cooling. As discussed further below, there is evidence that surfaces of the particles deviate significantly from the perfect crystalline structure. Several factors are expected to drive the surface toward a more disordered structure than that of the nanocrystal interior. Formation of a perfect crystalline surface would be associated with numerous water molecules with unsaturated hydrogen bond coordination. Some surface water molecules may increase their coordination by forming additional (albeit strained) bonds, at the expense of lateral order. Furthermore, the surface molecules have more freedom of motion than the bulk, and some of the surface disorder may be induced by entropy. Finally, rapid cooling of nanocrystals by collisions with carrier gas molecules may also contribute to surface disorder. Direct simulation of all the complex processes taking place during cooling and freezing of a nanoparticle is not feasible. To study surface properties, we then adopt a semiempirical approach: we generate ad hoc ice surface models of different structures, use them to evaluate observable properties, and employ comparison to experiment to assess the validity of the models. Specifically, the starting point is a patch of surface of about 30 x 30 A, containing 12 layers (or 6 bilayers) of ice. The oxygen lattice is cubic crystalline (( 111) or (001) faces were used); the H atoms are distributed randomly within the wellknown ice rules. The initial crystalline ice structure is then heated to 200-300 K using molecular dynamics with a TIPS2 potential, while keeping the bottom two layers rigid. Finally, the structure is recooled to the experimental temperature. By varying the maximal temperature and the duration of the heating and cooling runs one can obtain models of different extents of surface crystallinity (see Figure la-c; a related scheme was used in the past by Goes4 to study surface melting). The infrared spectra of the reconstructed surfaces as well as of the internal layers of the thin cubic ice slab have been simulated using a quantum mechanical scheme with a Morse basis localized on the 0-H(D) bonds.5 We constructed the potential as a sum of intramolecular terms, adopted largely from the gas phase, and the intermolecular TIPS2 potential. This representation of the vibrational dependence of the potential was adopted from ref 40; in this model, the well-known OHfrequency downshift upon formation of the hydrogen bond is due to OH bond lengthening by electrostatic interactions with neighboring water molecules. The variation of energy with bond stretch within this potential was fitted to a Morse function for each bond. Vibrational excited states were represented by an expansion in a basis of the Morse eigenstates for the different bonds with only single-bond excitations included in the expansion. Oscillator strengths of the 0-H(D) bonds were assumed

16538 J. Phys. Chem., Vol. 99, No. 45, 1995

Devlin and Buch

to vary linearly with the shift of the bond frequency from the gas phase value.4’ The vibrational Hamiltonian in the Morse basis included both intramolecular and intermolecular coupling between the 0-H(D) oscillators and intramolecular coupling between OH(D) bonds and the bending overtone, with the coupling parameters determined by fitting the overall shape of H20 and D20 bulk ice spectra. This fitting showed that both the long- and shortrange intermolecular coupling can be modeled as a dipoledipole interaction. For Modeling the Spectrum of Adsorbed CF4. The spectrum of adsorbed CF4 represents a special case. The exceptionally large oscillator strength of the triply degenerate antisymmetric C-F stretching mode, with equal dipole oscillations in all directions, causes intermolecular vibrational-dipole coupling that dominates the observed band complex of this mode 260C 2780 2760 rarenumbers 2740 2720 2700 2680 for a CF4 monolayer on the “spherical” ice nanocrystals. Since the coupling, as revealed in the relative positions and shapes of Figure 4. Infrared spectra in the d-D stretching mode region of thin intense T (transverse) and L (longitudinal) components of the films: (a) amorphous pure DzO ice at 15 K; (b) same as part a with band complex, is a function of the surface density and 50% unexchanged H20;(c and d) sample of part b annealed 10 min at 60 and 120 K; (e) 15 K deposit of isotopically scrambled 4% D20/ distribution of the CFq oscillators, it is exceptionally sensitive 32% HDO mixture after 60 K annealing (from ref 7 ) . to several interesting parameters of the system, including CF4ice and CF4-CF4 interaction potentials which, along with abundance: 30 each of two-coordinated and three-coordinated surface irregularity, determine the uptake and site distributions surface water molecules with unsaturated donor sites. of a simulated adsorption. Modeling of this system was carried This computed distribution of surface defects impacts the out using grand canonical Monte Carlo simulations of the uptake interpretation of the infrared spectra obtained for 12 K amorof adsorbate molecules on model patches of the ice surface phous ice deposits. These spectra, as exemplified in Figure 4, (constructed as discussed in the previous section). The antihave two bands in the 0-H(D) stretch region (3720/2749 and symmetric CF stretch spectra were obtained by diagonalizing a 3696/2728 cm-I) near the values typical of unbonded 0-H vibrational Hamiltonian for adsorbate configurations sampled g r ~ ~ p ~Recognition, . ~ ~ from , ~ the ~ computational ~ . ~ ~ ~results, ~ in the simulation. The long-range coupling of the vibrational of the comparable probability of two specific types of d-H at dipoles was modeled using the magnitude of the dipole derivative obtained from the TO-LO splitting in pure C F ~ ( S ) . ~ ~the amorphous ice surface prompted the assignment of the bands to the d-H(D) of two-coordinated (3720/2749 cm-’) and threeA semiquantitative fitting of the computed to the observed band coordinated (3696/2728 cm-’) surface water molecules. This complex, for submonolayer to multilayer coverage and for a assignment has been confirmed (a) by simulation of the stretch variety of surface conditions, has been used to infer the values frequencies expected for the differently coordinated surface sites of the gas-surface interaction parameters and the degree of and (b) by observation of a parallel response to thermal interesting and useful is the surface i r r e g ~ l a r i t y . ~Most ~ annealing of the experimental intensities of the two bands and discovery that, through the fitting of the computed to the the simulated populations of the two- and three-coordinated H2O experimental spectra, information about the extent of surface molecule^.^^'^ Simulated annealing of the clusters to 140 K is disorder at the molecular level can be inferred. accompanied by the smoothing of the cluster surface and an order-of-magnitude reduction of the number of two-coordinated IV. Bare Amorphous Ice Surface surface water molecules, while the number of three-coordinated Structure and Structural Relaxations. The simulation of surface H20 molecules is relatively unaffected. Similarly, amorphous ice formation at 10 K resulted in an (H20)450 cluster annealing of the experimental thin films of amorphous ice to with a highly irregular surface structure and a large fraction of 60 K (or 90 K) caused the 3720/2749 cm-’ band to disappear, incompletely coordinated (surface) water molecules (Figure 2).’* while the 3696/2728 cm-’ band retained much of its intensity The water molecules sticking to the surface during the slow (Figure 4), affirming the assignment to the surface twocondensation process (in which amorphous ice is formed both coordinated and three-coordinated molecules, respectively. experimentally and in the simulation) have insufficient energy Insight to Amorphous Ice Properties. Amorphous ice to permit surface reconstruction to a lower potential energy formed at 12 K apparently grows with a highly irregular surface structure. However, despite the high potential energy of the that results in a microporous structure with 5-10% of the 10 K cluster, just two of the 121 incompletely coordinated molecules on the ice surface. This may seem contradictory to molecules form a single H-bond to the surface, testimony to recent electron diffraction data that identifies such ice as of high the improbability of singly coordinated water molecules of ice density.46 However, the results are compatible if it is recognized existing under any conditions. that the diffraction measurements are not representative of the Of the other 119 incompletely coordinated molecules, 30 were micropores but rather the incompletely bonded condensed structure of higher density than that of crystalline ice. Paradoxitwo-coordinated and 89 were three-coordinated. The twocoordinated H20 each formed one donor and one acceptor cally, at higher temperatures (30-60 K) as infrared bands of H-bond so that each retained one unsaturated donor site and the surface groups weaken, signaling the loss of surface area one unsaturated acceptor site (the d-H(D) and the d - 0 of this and two-coordinated d-H(D), the condensed structure simultaarticle). The eighty-nine three-coordinated molecules were neously becomes less dense as the water molecules establish divided between 30 with an unsaturated donor site (d-H) and fuller coordination, forming an open random tetrahedral network.46 59 with an unsaturated acceptor site (d-0). Thus, the simulation indicated the presence of two types of d-H bonds of similar There is little doubt that above 80 K the d-H and the d - 0 of

e

Feature Article amorphous ice belong almost exclusively to three-coordinated surface water molecules, with very few two-coordinated and essentially no one-coordinated surface molecules that survive the thermal energies. Further, the three-coordinatedsurface H20 molecules vanish as T approaches 120 K; reflecting the final collapse of the ice nanopores. These results have a clear application to the understanding of three important properties of ice: the mechanism of vaporization, the temperature dependence of the dielectric loss, and the irreversible exothermic relaxation during annealing of amorphous ice near 125 K. On the basis of the response of the vaporization rate of ice to the presence of dopants, Davy and Somorjai proposed a vaporization mechanism which presumed that ice surfaces are rich with two-coordinated H20 molecule^,^' molecules that convert to escape-prone one-coordinated molecules through orientational-defect activity. Though there may be salient aspects to their defect mechanism, the presumed abundance of two-coordinated surface H20 is discounted by our results. Similarly, some ambiguity in the interpretation of dielectric relaxation data for microporous amorphous ice is eliminated by the combined spectroscopic and simulation results. It has been shown that amorphous ice formed near 77 K exhibits a significant dielectric loss upon warming through the 80- 120 K range, a loss that vanishes irreversibly upon sintering the ice near 100 K in a vacuum.48 This dielectric loss was related to the onset of rotation of surface water molecules that have one or two dangling bonds, a rotation that necessarily vanishes as the nanopores collapse in the 100 K environment. Since H20 molecules with two dangling bonds (Le., two-coordinated surface H20) are rare above -80 K, this relaxation would seem to relate more specifically to the onset of rotation by threecoordinated surface H20 molecules. It is well-known that failure to anneal amorphous ice above 120 K, prior to calorimetric scans designed to observe the endothermic glass transition, will result in obscuration of the transition by an exothermic r e l a ~ a t i o n .The ~ ~ vanishing of the dielectric loss mentioned above and the loss of the infrared band intensity of three-coordinated groups as a result of annealing in the 100-120 K range (Figure 4) identifies the collapse of the nanopores as a general effect of this annealing. Preferential Deuterium Bonding of Surface Water Molecules. It is known that, given the choice, HDO molecules preferentially bond to a proton acceptor through the 0 - D bond. This is because localization of the D atom in the bond is associated with lower zero-point energy than localization of the H atom. This preference has been studied in some detail for the dimer of HDO, and results for the variation of the preference with temperature have shown that the zero-point vibrational energies stabilize the D-bond vs the H-bond by -60 ~ m - I . 4 ~ Similarly one might expect that the attachment of an HDO molecule to an ice surface, to form a three-coordinated molecule with a d-H or a d-D group, would favor the former case because of preferential D-bonding. In fact, the relative population of d-H and d-D of surface HDO molecules of microporous ice formed at 12 K is sensitive to the rate of growth of the film. For slow deposition rates (1 p d 2 0 min), the attachment of the HDO apparently follows random statistics, a reflection of the molecular orientations at impact. A similarly statistical distribution is obtained in a classical simulation of low-temperature HDO condensation. As a result, an isotopically scrambled mixture containing 66% deuterium gives the statistically expected value of 2.0 for the ratio of the intensity of the three-coordinated dangling-D of D20 to that of HDO. However, if the 12 K sample is warmed above 60 K or if the deposit is made at a faster rate, that ratio changes

J. Phys. Chem., Vol. 99, No. 45,1995 16539

I

Figure 5. Comparison of the infrared band of the 0-D stretch mode at 80 K: (a) annealed amorphous ice (125 K); (b) crystalline thin film vapor deposit, made at 155 K and scanned at 80 K; (c) unannealed cubic ice nanocrystals formed and scanned at 74 K.

to '4.0, indicative .of a significant preference for deuterium bondingG8The switch from a statistical distribution to preferential D-bonding near 60 K is consistent with the presumed onset of rotational motion of the surface-bonded molecules.48 These results are also consistent with a picture of kinetically controlled condensation at very low temperatures; that is, HDO orientation after sticking is determined by the impact geometry rather than by the relative stability of O**.H versus 0.e.D hydrogen bonds. Heating to 60 K is sufficient to activate local relaxation toward more stable 0.a.D bonding, resulting in the observed counterintuitive temperature effect.

V. Bare Cubic Ice Surface Structure and Structural Relaxation. If an ice crystal is cleaved perpendicularly to the c-axis, the instantaneous surface structure that results from the breaking of H-bonds, prior to surface reconstruction, is readily imagined (Figure la). The surface water molecules form an array of perfect hexagonal rings. The array is nonplanar, since to preserve the tetrahedral bonding geometry, every second 0 atom of water is displaced by -1 8, below its nearest neighbors (forming a bilayer). From the Bemal-Fowler ice rules it is necessary that the cleavage produce an equal number of three-coordinate d-H and d - 0 surface defects in the top hexagonally-ordered layer. Further, a number of s-4 molecules, equal to the total of the dangling groups, will also be exposed in the bottom (indented) surface layer. This distribution of surface groups is one of high energy, so in the presence of sufficient thermal energy, the surface restructures as it relaxes. In our experimental studies, the relaxation of the surface of the ice nanocrystals occurs as they form during the rapid cooling of a dilute mixture of water vapor in a carrier gas. A condensation and crystallization sequence apparently releases sufficient energy that the nanoparticles do crystallize during the rapid fall from room temperature to -80 K, with amorphous nanoparticles obtained only when the rate of cluster-cell pressurization and the percent water content are reduced substantially. The usual crystalline nature of the particles is apparent from the relatively narrow band width of the 0-D stretch of the few percent HDO often included in the samples (Le,, less than 20 cm-l in width compared to the 60-70 cm-' of the amorphous phase, Figure 5).50 It can be imagined that the structure of a typical nanocrystal, including that of the surface, becomes frozen as the particle cools rapidly below -200 K.5' The detailed description of the resulting surface is a special interest of this study with the most direct information obtained by simulating the surface of a slab of cubic ice as a function of the increasing thermal energy of

Devlin and Buch

16540 J. Phys. Chem., Vol. 99, No. 45, 1995

a, 0.14

$ 0.12

e

0.1

3 0.06 a 0.06 0.04 0.02

I

0' 2800

2750 2700 2650 2600

2550

2500

Wavenumbers (cm-1)

Figure 6. Spectra in the 0-D stretch mode region for D?O nanocrystals of cubic ice at 80 K: (a) absorbance spectrum; (b) difference spectrum from subtraction of spectrum of annealed (145 K) nanocrystals from spectrum of unannealed 80 K nanocrystals.

the slab. This simulation, in agreement with an earlier study of hexagonal ice by Goes: shows that a regular "cleaved" ice surface does undergo a pronounced relaxation and reconstruction. This relaxation was modeled by simulated heating of a perfect crystalline structure to 200 K, followed by recooling to 27 K. The effect of relaxation can be appreciated by comparing parts a and b of Figure 1; the relaxation was associated with local distortion of the hexagonal lateral order of the top layer, apparently driven by the energy minimization resulting from formation of additional H-bonds. A one-third reduction in the number of three-coordinated molecules with d-H and d - 0 occurred as numerous surface H20 molecules achieved formation of four-coordinated distorted, or s-4, tetrahedra. A minor spread in the vertical positions of the molecules of the fully four-coordinated surface layer and of the second and third subsurface bilayers was also effected by the relaxation. Experimental and Simulated Surface Spectra of Cubic Ice. Examination of the high-frequency region of the 0-H (0-D) stretch-mode band complex of the cubic ice nanocrystals has revealed only a single obvious surface-molecule absorption feature, a weak relatively narrow band of the three-coordinated molecules with a dangling-H(D) at 3692(2725) cm-'. This band (Figure 6a) is downshifted by only 4(3) cm-' from the corresponding position noted for microporous amorphous ice. However, the existence of other absorption bands caused by the surface molecules of cubic ice becomes apparent if the spectrum for a nanocrystalline sample, obtained after annealing at a temperature above 120 K, is subtracted from the spectrum of the same sample obtained prior to annealing. Such a difference spectrum (Figure 6b) displays us positive bunds the extra absorption by the more numerous surface molecules of the unannealed ice, particularly in the region above 2500 cm-'. (Though most of the intensity of the bulk-mode bands is canceled in the difference spectrum, negative bands dominate the bulk-ice absorption region between 2500 and 2300 cm-' (see also Figure 14). This is because the annealing process generates additional bulk through the annihilation of small nanocrystals and the growth of larger nanocrystals within the network of suspended ice particles.) From the 80 K difference spectrum of Figure 6b, a minimum of four positive (surface) bands are apparent with frequencies centered near 2725, 2640, 2580, and 2500 cm-' (with the position of the latter band difficult to place because of the rapid rise of bulk intensity below 2500 cm-I). The richness of the difference spectra between small and larger nanocrystals makes clear the abundance of spectroscopic information available for the surface-localized modes of ice and the possibility of characterizing ice-adsorbate interactions through the different responses of the various absorption bands. However, in the absence of an understanding of the origin of

D

I

I

I

I

I

I

I

2750

2700

2630

26W

2¶90

2500

2450

I DO

Wavenumber (cm-I)

Figure 7. Experimental (a) and simulated (b) spectra of the surfacelocalized modes of HOD ice in the 0-D stretch region. Spectrum b relates to a cubice ice model of 100% HOD, while spectrum a is from a sample containing 32% HOD, 4% DzO, and 64% HzO.

these features, the spectra would be of minor value. Significant progess in assigning the features of Figure 6b has been made through a comparison of the simulated spectra of the modes of bulk ice with spectra of modes "localized" near the ~urface.~ A simulated difference spectrum for HDO cubic ice compared with the corresponding experimental difference spectrum is presented in Figure 7. As explained in ref 5, this difference spectrum is estimated as the difference between the simulated absorption intensity per molecule for the surface layer and the similar absorption intensity for the bulk. Such simulated spectra, similar in major aspects to the experimental difference spectra (Figures 6b and 7a), have been divided into components identified with modes of the three-coordinated molecules with d-D and with d-0, and the s-4 molecule^.^ From this division, the experimental bands of D20 and HDO have been identified as labeled in the figures. For D2O ice (Figure 6) the 2725 cm-I band is identified as the out-of-phase stretch of the three-coordinated molecules with a d-D, while the corresponding in-phase vibration, with a large amplitude of the bound 0 - D group, is downshifted nearly 400 cm-' to -2350 cm-I. Since the strongest simulated band in the 2600 cm-' region is caused by the d - 0 molecules, the observed 2640 cm-' band is assigned to the (out-of-phase) stretch of that subclass of molecules. The simulated spectra also indicate that the fully coordinated but distorted s-4 molecules are a prime source of intensity in the 2550-2300 cm-' range. For this reason, we have assigned the 2580 cm-' band of Figure 6b to the s-4 (out-of-phase) vibrations. The positive band nearhelow 2500 cm-' in the experimental spectrum is attributed to d - 0 and s-4 molecules with the d - 0 (in-phase) contribution perhaps dominant. The simulated spectra further indicate that the surface-localized modes, with the exception of the 2725(3692) cm-' d-D(H) vibration, are delocalized over several molecules. Bending Modes of Surface Water Molecules. Relatively little attention has been directed to the infrared bands of the surface-localized bending modes of ice. Nevertheless, features

J. Phys. Chem., Vol. 99, No. 45, 1995 16541

Feature Article

0 2

0 1

0 0

n

-0 1

3 02-

2760

2740

2720

2700 2680

2660 2640

2620

2600

wavenumbers

Figure 9. Infrared doublet of the d-D of D20 ice, bare and with the adsorbates Ar, Nz,CO2, CO, and SO2 (from ref 15).

O

response will be examined in the discussion that follows or in section VII. Reponse of the Surface-Localized Modes to Adsorbates. When a microporous amorphous ice film covered with an ultrathin amorphous overlayer of a potential adsorbate is warmed to near the crystallization temperature of the overlayer, the molecules usually become sufficiently mobile to penetrate the ice channels and micropores. Alternatively, the ice can be observed under an equilibrium overpressure of the adsorbate gas at a somewhat higher temperature. In either instance, molecular diffusion is accompanied by adsorption on the walls of the channels and micropores, so that the various surfacelocalized modes experience sizable frequency shifts. Though the difference spectra in Figure 8A, for amorphous ice samples having different surface-to-volume ratios, reveal only the d-H(D) band (2728 cm-') as a distinct surface feature, the positions of other surface modes can be highlighted by the use of adsorbates. That several of the surface modes are significantly affected by an adsorbate is obvious from difference spectra obtained by subtracting the spectrum of the ice(D20) saturated with adsorbate from that for identical conditions without adsorbate. Results for Hz, CH4, Nz, and CO are given in Figure 8B, where positive features are attributed to unshifted bands and, generally speaking, the intensity minima are near the new positions of surface-localized modes of the adsorbatesaturated ice. a. Adsorbate-Znduced Shifts. Regardless of the adsorbate used, three maxima occur in the difference spectra of Figure 8B, at 2728, 2650, and -2590 cm-I, at slightly higher frequencies than the positions of the three highest frequency surface-localized modes of crystalline ice (see section V). By contrast, the minima, indicative of the shifted mode frequencies, are unique to the particular adsorbate. For example, the minimum from the shifted 2650 cm-' band of the d - 0 varies from 2635 cm-' for H2 to 2620 cm-' for CO, an indication of a stronger interaction of CO than H2 with the d - 0 surface sites (as expected from ab initio results for the corresponding dimeric c ~ m p l e x e s ) . The ~ ~ observed shifts of the d-H(D) mode are particularly unique to the different adsorbate molecules. Example d-D shifts displayed for several small molecules in Figure 9 are 7 (CF41, 10 ( H A 12 (Ad, 19 (Nd, 19 (CH4), 34 (COz), 38 (NO), 44 (CO), 68 (SOz), 74 (acetylene), 76 (ethylene), and 78 (benzene) cm-' (with the NO value that of adsorbed dimeric NO)^).^^ Similar but somewhat smaller shifts of the d-D of crystalline ice have also been observed for most of these same molecules.

I

-0.01

ZIOO

m 2 m 2400 Wavenumber (cm-I)

2700

2330

22W

Figure 8. Infrared difference spectra of microporous amorphous ice with a subtract factor of 1.000: (A) Spectra for ice at Tminus spectra for ice annealed at T 20 for (a) T = 12 K, (b) T = 32 K, (c) T = 52 K, and (d) T = 7 2 K. The bottom curve is the composit of the other four. The ice was deposited at 12 K, all spectra were scanned at 12 K, and a-d are on the same adsorbance scale. (B) Spectra for bare ice minus spectra for ice saturated with the adsorbates (a) H2 at 12 K and (b) C&, (c) NZ,and (d) CO at 80 K. The bottom curve is for CO adsorbed on ice at the minimum uptake T of 28 K (from ref 5).

+

at 1650 (H20), 1403 (HDO), and 1211 (D20) cm-I, prominent in the spectra of the ice nanocrystals and with a width (-16 cm-I) and intensity comparable to those of the stretch mode of the d-H(D) molecules, have been assigned to three-coordinated water molecules with dangling-H(D).' I Broader bands near 1700 (HzO), 1470 (HDO), and 1230 (D20) cm-' have been tentatively attributed to the bending mode( s) of surface d-O/ s-4 molecules. Because the surface-localized bending modes are relatively insensitive to the presence of adsorbates, they are not highlighted in the discussions that follow.

VI. Adsorbates on Amorphous Ice For amorphous ice the bands of the surface modes, other than of the d-H(D), are not resolved in the difference spectra that compare samples with varying surface-to-volume ratios (Figure 8A).5 For this reason insight to adsorbate-ice interactions has been derived primarily from observations of the d-H(D) band. Data available for that band include shift magnitudes, enhancement factors, and (potentially) interaction energies from sitespecific optical isotherms. However, the surface-adsorbate interaction can also be monitored through the response of the fundamental bands of the adsorbate. This may include band positions relative to other phases of the adsorbate, induced intensity of otherwise infrared inactive modes, the splitting of modes due to multiple-site effects, the splitting of strongly infrared active vibrations of an adsorbate into transverse (T) and longitudinal (L) modes through coupling of the transition dipoles, and the evolution of new bands from chemical reaction with the ice surface (as is the case with adsorbed which are not covered in this article). Examples of each type of

16542 J, Phys. Chem., Vol. 99, No. 45, 1995 It is interesting to compare these observed shifts with computed shifts for the dimer complexes with water. For example, we have computed H20 symmetric/antisymmetric stretch-mode ~ , j5/ shifts, using high-level ab initio methods, of r v,

E

-

.'t

f 0.0 ,I

4.0

0.0

40.0

80.0

0.9

1

p.,

0.5

'

0.5

I

;

1

0.0 40.0

K

'. \

\

'. -._. 0.0 10.0 l n g * n y Cm.1

0

m.0

-.-.

0.0

4.0

0.0

.

80.0

409

Figure 11. Rotational Raman spectra of Hz adsorbed on the microporous amorphous ice surface. Zero frequency was set at the free rotor 2 band: pure para value. Intensity in relative units. (a) J = 0 3; -, adsorbate; -, experimental: ---, calculated. (b) J = 1 experimental for ortho/para mixture with '50% ortho: calculated for 8% p a r d 9 2 8 ortho. (c) Calculated J 0 2: pure para adsorbate; -, spectrum of Hz molecules within 3.5 A from dangling atoms: - - -, spectrum of Hz at least 4 A away from dangling atoms. (d) Calculated J =0 2 : -, pure para; - - -, 8% pard92% ortho mixture.

-

-

-

-*-e-,

-

These results suggest that the surface of nanocrystalline ice particles may reflect the influence of the interior crystalline structure. Further, the peak positions of the ortho- and paraHZ on the crystalline ice surface are stable with time. For microporous amorphous ice the peak maxima of both the orthoand the para-Hz, originally fairly close to those of the crystalline ice samples, drift to significantly lower frequencies (-5 cm-') over a matter of a few hours. This shift is consistent with the H? slowly assuming lower energy, more tightly bound positions and confirms that the microporous amorphous ice surface is more complex and less uniform than that of cubic ice. 6. H2 as a Quantum Rotator on the Ice Sulface. The rotational Raman spectrum of H2, adsorbed on the rough and disordered surface of microporous amorphous ice at 12 K, has been observed and modeled using the PIMC technique and the (H20)450 simulated amorphous ice cluster.33 The experimental Raman spectra for adsorbed hydrogen are distinctly different for ortho- vs para-H2 (Figure 1l), with the ortho giving a smooth broad band centered at 594 cm-I while the para-H2 band consists of a sharper feature near 345 cm-' with a pronounced shoulder that extends above 375 cm-I. The PIMC modeling of the spectrum, that closely reproduces the band shapes without use of adjustable parameters, relates the extra para-Hz band structure and the breadth of the ortho band to removal of the IJM) state degeneracies of the different J manifolds caused by H? interaction with the surface. This modeling revealed the sensitivity of the rotational spectrum to the relative anisotropy of the interaction potential for different sites on the ice surface. For example, the rotational Raman spectrum is informative of the distribution of molecules between sites near and well removed from the d-H, with the spread of IJM) states greater for the former (Figure 1IC). For this reason it is not surprising that the unique structure of the para-HZ rotational band collapses into a single band for H2 adsorbed on amorphous methanol (for which d-H groups are apparently missing as no d-H infrared band is readily observable).54 c. Two-Site Binding of Adsorbed CO and Acetylene. As has been noted, ice surfaces are rich in different adsorbate bonding possibilities. Electrostatichydrogen bonding is possible to both

positive surface centers (dangling-H atoms) and negative centers (dangling-0 atoms). In addition, substantial van der Waals interactions may develop between polarizable adsorbate molecules and the surface (in particular involving s-4 water molecules). Ab initio calculations on dimers of water with small molecules such as H2, N2, and C035indicate that the attractive component of the interaction in the dimers is dominated by electrostatic interactions between anisotropic charge distributions of the monomers (all three diatomics have substantial quadrupole moments; CO has in addition a small dipole). The van der Waals interaction is substantial but smaller in size at the dimer potential minima. However, as discussed in ref 12, the relative importance of the different interaction components may be altered significantly on the surface. This is since the electrostatic interactions are directional, i.e. require a specific orientation of an adsorbate molecule with respect to HzO. This orientation can be assumed easily in a dimer, but an adsorbate molecule on the surface has several water neighbors and cannot assume a favorable orientation with respect to all of them. Thus on the surface the electrostatic interactions can largely cancel each other, while the always attractive and less directional van der Waals interactions add up and may dominate the adsorbate-surface bonding, particularly for adsorbates in surface cracks and molecular-sized indentations. This argument ignores however cooperative effects. As mentioned above, cooperative effects seem to enhance significantly partial positive charges on dangling-H atoms, thus enhancing their capacity for electrostatic bonding of adsorbate. In addition, the structural features of the ice surface are expected to influence adsorbate bonding mechanisms. As seen in Figure 1, the ice surface abounds in rings of water molecules, and the centers of these rings provide very favorable sites for strong van der Waals bonding. Figure Id shows a simulated layer of CF4 adsorbate, which fills such ring sites. One may note dangling-H atoms protruding between CF4 molecules; these atoms, which are not particularly attractive to the hydrophobic CF4 adsorbate, provide binding sites for adsorbates with strongly anisotropic charge distributions, such as CO. Dangling-0 atoms may provide an additional set of electrostatic bonding sites. Moreover as shown by Kroes and Clary, polarizable adsorbates located inside ring sites, which are also proton donors, may interact strongly with dangling-0 atoms neighboring the ring.55 Experiments show that a given adsorbate molecule may assume more than one bonding configuration on the surface. Two examples of molecules that, as ice-surface adsorbates, do show distinct doublets for individual vibrational modes are carbon monoxideI3 and with the splitting in each case interpreted in terms of different interaction strengths and bonding structures with the d-H and the surface oxygen sites. In the case of acetylene the distinction is clearly between sites at which the C2H2 acts as a proton acceptor (d-H site) and as a proton donor (oxygen site). A strong proton-acceptor interaction is revealed in the large shift (95 cm-I) and enhancement of the d-H mode band of amorphous ice.lba A strong protondonor interaction with the surface oxygens is expected on the basis of both matrix-isolation data56 and ab initio results for the CzHyH20 c o m p l e ~ , ~an~ interaction ~.~' that is also expected to activate a transition dipole for the C2H2 triple-bond stretch mode. Therefore, it is not surprising that the triple-bond mode appears as a Raman doublet (1963 and 1957 cm-') while a single activated infrared band (1957 cm-I) is observed.'6a The assignment of the 1957 cm-' component to acetylene complexed as a proton donor with the ice surface, as suggested by the infrared activity, was affirmed by an ab initio prediction that

Devlin and Buch

16544 J. Phys. Chem., Vol. 99, No. 45, 1995

of CF4 (curve a of Figure 12) is as expected. The CO data provide a graphic example of how competitive adsorption at the polar or less polar surface sites can be used to differentiate sites spectroscopically.

0.ooe

4

0.006

VII. Adsorbates on Cubic Ice Surfaces

0

’ 0.0044 c

-o.ooo+,

,

-0.002

-o.ooj /\ 1

-0.0064 2 160

1

I

2140

2120

WAVENUMBERS

Figure 12. Comparison of the stretching-mode infrared absorption of CO adsorbed in pure DzO microporous amorphous ice at 42 K (a) with that of CO adsorbed in ice with coadsorbed CFI at 58 K (b) and with that of CO adsorbed following preadsorption of ethylene oxide (c).

the triple-bond stretch is downshifted 7 cm-I by a dimeric proton-donor interaction with H20. The surprising strength of the acceptor action of the acetylene is the most remarkable aspect of the surface interaction. Acetylene does not complex as a proton acceptor with HzO in an inert matrix,56and the computed complexation energy is less than 1 k c a l / m ~ l . ’ ~Therefore, ~ . ~ ~ the much stronger acceptor action with ice surfaces, estimated as 3.7 kcallmol, hints at a major enhancement of the polarity of the surface d-H groups through the cooperative interaction with neighbor water molecules. The identification of the nature of the sites occupied by CO on an amorphous ice surface has made use of a competition between coadsorbates having different strengths of interaction with the polar and less polar parts of the surface. CO clearly adsorbs at two quite different classes of sites, since the stretch mode of the CO appears as a doublet (2136 and 2152 cm-’) with an intensity ratio of -2:l (Figure l2).I3 Our ab initio results for the water-CO complex indicate that the interaction strength of the negative carbon end of the CO with the water 0-H is over twice that for binding of CO to the oxygen of water in a T-shaped structure (and that the CO stretch-mode frequency is greater by 9 cm-’ in the former On the other hand, an exceptionally small shift of the surface d-H band for ice saturated with CF417 suggests that CF4 avoids the d-H, although it is held tightly by the rest of the surface. This combination of properties implies that CF4 should compete quite well with CO for the less polar part of the ice surface but not for the d-H sites. Accordingly, when coadsorbed with CF4 in “equal” amounts, the CO should preferentially occupy the d-H sites. The curves of Figure 12 show that, with CF4 present, the lower frequency CO band at 2136 cm-’ is markedly weakened relative to the 2152 cm-’ band. This is a result that is anticipated by the CF4 distribution on a relaxed cubic ice surface which, as shown in Figure Id, apparently leaves dangling-H protruding between the adsorbed CFI molecules. Along with the ab initio results, this strongly suggests that the 2152 cm-’ band is produced by CO at the d-H sites. Since, as noted in section 111, we find fewer d-H than oxygen sites on the surface of the simulated amorphous cluster, the relatively large intensity of the 2136 cm-’ band in the absence

The surface of cubic ice is distinctly different from that of microporous amorphous ice. This is clear from the different band positions of the surface-localized modes, with the d-H and d - 0 frequencies downshifted about 4 cm-’ in the Crystalline case. Further, the shift of the d-H mode of the crystalline surface induced by complexation with a small adsorbate molecule is typically only -80% that observed for the micropore surfaces. Even more obvious is the marked difference in the positions and shapes of infrared bands of small molecules such as H2 adsorbed on the crystalline and the nanoporous amorphous surfaces (Figure 10). However, just as there is no single surface fully representative of the amorphous phase, since the surface varies substantially with the annealing temperature, the nature of the surface of the crystalline phase is dependent on the temperature at which surface reconstruction occurs (see section 111). The influence of formation conditions on the effective surface structure of ice nanocrystals is of immediate interest and will be considered in a later section on adsorbed CF4. Response to Adsorbates of the Surface-Localized Modes of Ice. The surface-localized modes of crystalline ice are observed to respond to the presence of a nonreactive adsorbate in much the same manner as described for microporous amorphous ice (section VI). However, because of reduced band widths of the d - 0 and s-4 modes, more detail is apparent in a comparison of the crystalline ice spectra before and after the adsorption of molecules such as Hz, N?, CO, and CF4. The difference spectra of Figure 13, which were obtained by subtracting spectra of large nanocrystals covered with adsorbate from spectra of smaller nanocrystals similarly covered (using the same ice deposit), show some sensitivity of band positions to the nature of the adsorbate. For example, just as the shift of the d-D band (2725 cm-’) is small for H2, adsorbed H2 has only a marginal effect on the positions of the “maxima” of the d - 0 and s-4 bands near 2640 and 2580 cm-I. By contrast, adsorbed CO clearly causes a significant shift of each of these three bands, while CF4 appears to shift the d - 0 band more strongly than either the d-D or s-4 bands. There is a potential for use of the relative magnitude of these shifts to help characterize the nature of the interactions of the various molecules with the different types of icy surface sites. This potential is more obvious for ethylene, a molecule that interacts much more strongly with the surface sites than do the adsorbates of Figure 13. The spectrum for adsorbed ethylene, compared with that of the bare ice and of ice with CO as the adsorbate, is shown in Figure 14.52 It is clear that both the d-D and d - 0 bands (and perhaps the s-4 band) are very strongly shifted by the surface interaction with ethylene, with the d - 0 shift more than double that caused by CO. In the long term a more important use of the response of surface-localized modes to the presence of adsorbate molecules may be the quantitative determination of site-specific heats of adsorption. For example, the degree of saturation of the d-D as a function of CO pressure at 96 K is shown in Figure 15. Similar spectra have been used to determine the d-H site-specific heat of adsorption for Hr, N2, and CO, yielding values of -0.64, -2.08, and -2.81 kcal/mol, respe~tively,’~ values that, on the average, exceed the corresponding computed dimer binding energies (0.52, 1.15, and 1.76 kcal/m01)~~by 50%. No quantitative values have been determined for the corresponding

Feature Article a

J. Phys. Chem., Vol. 99, No. 45, 1995 16545 0.03

10

0.025

8 0.02

ow

m

2 0.015

::

2

0.01

0.005

-0 10

0 2760 al c

2740

2720 2700 Wavenumbers

-0 20

2660

2680

Figure 15. Example of infrared spectra of the d-D mode of cubic ice with varying CO pressures, as used to determine the adsorption isotherms and site-specific heat of adsorption: CO pressures were, in

Torr at 96.0 K, (a) 0.0, (b) 24.2, (c) 50.8, (d) 97.4, and (e) 300.

o.onl W

I

,

2760

2720

OIO-.

3 0 001-

I ,

,

/

2680 2640 2600 UAVENUMBER

1

,

2560

,

,

2520

Figure 13. Infrared difference spectra: All spectra are of ice nanocrystals annealed at 100 K minus spectra for the same ice sample annealed at 145 K, with a subtract factor of 1.000. (A) Top spectrum is for bare nanocrystals (as in Figure 5b), while spectra b-e are for nanocrystals saturated with N2, CO, CF4, and ortho-Hz. For each adsorbate difference spectrum, the two spectra were each obtained with the sample under the same gas pressure at 85 K (except for H2, for which the scans were at 25 K). (B) Spectra differ from those of part A only from expansion along both axes (from ref 5 ) .

0 000

E

~

2BW

~

2750

t

2700

~

26M

2600

,

2550

2500

l

2450

!

2400

2350

l

2300

l

22SO

HAVENWEER

Figure 16. Difference spectra from infrared spectra for bare ice nanocrystals minus spectra of nanocrystals saturated with adsorbate, with a subtract factor of 1.OOO. The adsorbate and scanning temperatures were (a) 20 K for ortho-Hz, (b) 80 K for Nz, and (c) 80 K for

co.

0 15 0.12

I

P)

g 0.09

\

-u

e

4a 0.06 0.03 0' 2800

I 2750

2700

2650

2600

2550

2500

Wavenumber (cm-I)

Figure 14. Difference spectra as in Figure 13, except the adsorbates

2750

2mo

2650 WVENUMBER

2600

2550

are CO and C2H4.

Figure 17. Difference spectra obtained from (a) bare ice minus ortho-

adsorption heats for the oxygen sites, as measured using the shift of the 3560/2540 cm-' band, but data such as for ethylene in Figure 14 demonstrate that possibility. Though small effects on the surface-localized modes are difficult to characterize using spectra such as presented in Figure 13, a sensitive indication of weak molecule-surface interactions can be obtained from difference spectra that compare the bare ice with the same ice sample covered with the adsorbate. The results for spectra of adsorbed H2. N2, and CO, subtracted from the bare ice spectrum, are shown in Figure 16. As noted earlier for amorphous ice, above 2500 cm-' the positive bands correspond to the unshifted surface-localized modes (2725,2640, and 2580 cm-I) while the negative features approximate the adsorbate-shiftedband positions (which can be determined more accurately from spectra such as in Figures 13 and 14). A particularly interesting use of these features for adsorbed H2 can be discerned from Figure 17. Here, the top curve shows

a difference spectrum similar to curve a of Figure 16, while the middle curve reflects the effect of an incremental uptake of H2 for a pressure increase near the saturation value and the bottom curve denotes the loss of H2 for a sample held at 10 torr and 16 K during the conversion of ortho- to para-H2. The curves are standardized to give a common amplitude for the d-D band near 2720 cm-l. Comparison of the amplitudes of features of the middle and top curves shows that, as expected from computed relative interaction strengths, the last H2 adsorbed as a result of increasing pressure is near the weakly interacting d-D positions. A similar comparison with the bottom curve shows that the H2 loss during the ortho-to-para conversion is substantial near the anisotropic d-H and d - 0 positions but trivial at the s-4 sites, as expected from the PIMC results37 referred to in section VI.

H2 on ice as in Figure 16a, (b) ice near saturation with ortho-H2 minus ice saturated with ortho-H2, and (c) ice saturated with para-H2 minus ice saturated with ortho-H2.

l

l

Devlin and Buch

16546 J. Phys. Chem., Vol. 99, No. 45, 1995 Moreover, the latter comparison establishes that the surfacelocalized infrared absorbances at 2640 and 2580 cm-’ are associated with different sites on the ice surface, an affirmation of their assignment to d - 0 and s-4 sites, respectively, as deduced from simulated ~ p e c t r a . ~ Cubic Ice Surface Effects on Adsorbate Molecule Modes. The influence of the interaction with the ice surface on the vibrational modes of nonreactive adsorbates has been examined in some detail for H2, C2H2, CO, and CF4. The results for H2 were summarized in the context of results for microporous amorphous ice in section VI and Figure 10. Results for the acetylene interaction with the ciystalline ice surface have also been used primarily to support the more extensive data obtained for organic molecules interacting with amorphous ice, as noted in section VI. The latter has been valuable, since even small unsaturated organic molecules resist diffusion into the amorphous ice micropores, necessitating the simultaneous codeposition of the organics with the amorphous ice in attempts to observe the “surface” interactions. In most respects the response of CO to adsorption on the crystalline ice surface also mirrors that noted for the amorphous surface in Section VI and Figure 12. Again, CO occupies two different classes of sites so that the CO stretch mode appears as a doublet at 2152 and 2137 cm-I. Further, adsorption at the d-H site, the source of the 2152 cm-’ band, is preferred in competition with less polar molecules such as Nz and CF4. In the latter case, very little CO uptake at the oxygen sites of the cubic ice surface is observed until the d-H sites are first saturated.” T-L Splitting of CF4: A Special Probe of Surface Structure. Theoretical models of normal modes of surface adsorbates often presume that the vibrational excitations can be treated as quasilocalized and i n ~ o h e r e n t .This ~ ~ view may be appropriate in some instances, but results such as those of Ewing et al. for multilayers of CO and COz on NaCl,59 which clearly show that long-range polarized modes give rise to reasonably narrow transverse (T) and longitudinal (L) bands of the dipole-active stretching modes, are indicative of coherent coupling of molecular modes with vibrational excitations spread over the sample. That these infrared measurements, made at off-normal incidence from adsorbate multilayers on singlecrystal planes, reveal the usually inactive longitudinal mode is understood in terms of the Berreman effect, Le. absorption at LO-mode frequencies of radiation that has an electric field component perpendicular to a thin film surface.60 Ultrathin adsorbate layers on small particles, such as the nanocrystals of ice considered in the present study, are also probed spectroscopically using radiation largely incident off the surface normal, so that strong longitudinal (L) as well as transverse (T) mode bands should be observed for adsorbate vibrations that couple to give T and L modes having significantly different frequencies.m,61It has been recognized for some time that the various disordered condensed-phases of CF4 are marked by vibrational excitations of the v3 mode that resemble a strongly split T-L pair.62 It follows that observation of a similar T-L splitting is expected for multilayer adsorbed CF4 regardless of the assumed structure. We are unaware of any previous reports of T-L splittings observed for a monolayer of any molecular adsorbate on a dielectric surface, so expectations for monolayer or submonolayer amounts of CF4 adsorbed on ice were uncertain. However, spectra such as those in Figure 18a-c have made it apparent that distinct features of the T (-1245 cm-I) and L (-1320 cm-I) modes are observable for monolayer (and submonolayer) amounts of adsorbed CF4. Further, since the magnitude of the

1340

1300

1260

1220

YViNUnsiR

Figure 18. Infrared spectra of the v1 mode of CFJ adsorbed on microporous amorphous ice (a) and on ice nanocrystals annealed at temperatures of (b) 83 and (c) 140 K. Curve d is the simulated spectrum of CFI at near monolayer coverage of the relaxed surface of Figure lc,d. Samples b and c and simulated conditions were 0.06 Torr of CF, at 83 K. (Here, x marks the Evan’s holes of a CFJ overtone not modeled in curve d.) (e) Calculated spectrum of a hexagonal crystalline monolayer of CFJ.

splitting of T-L doublets is dependent on the concentration of molecular oscillators, it was predictable that such splittings could be used to study dilution and phase separation of CF4 coadsorbed with a second adsorbate.” One of the reasons that CF4 gives such large T-L splittings, even for orientationally disordered phases (including the liquid and the adsorbed phases), stems from the threefold degenerate nature of the v3 vibration which projects an exceptionally large magnitude of the transition dipole independent of direction. As a result of the strong intermolecular dipole-dipole coupling, the CF stretch excitations of the adsorbate layer are collective. Figure 18d shows a simulated spectrum of u3 modes of a monolayer of CF4 adsorbed on a model ice surface shown in Figure lc,d. The spectrum reveals two peaks, in accord with experiment; the peaks are due to collective in-phase excitations of adsorbate molecules. The high-frequency peak corresponds to delocalized longitudinal excitations, with the dipole oscillation perpendicular to the adsorbate film, while the low-frequency peak corresponds to delocalized transverse modes with dipole oscillations in the adsorbate layer.43 The side-by-side parallel relationship of the neighbor transition dipoles of the longitudinal mode generates a repulsion such that the “monolayer” longitudinal frequency is calculated to be -78 cm-I greater than that for the transverse pair. This compares favorably with the -75 cm-’ T-L splitting observed for a “monolayer” of CF4 on ice (Figure 18c). Simulated CF4 spectra have also shown that the overall structure of the CF stretch band is a sensitive probe of the nature of the underlying ice surface. Moreover these studies support the suggestion based on simulations of ice surface spectra5 that ice surfaces (at least the ones generated under the present experimental conditions) display a considerable amount of disorder. Since this is true for nanocrystals whose infrared spectrum indicates a well-crystallized interior, it appears that in these nanocrystals the ice surface is much more disordered than the bulk. Figure 18e shows a simulated spectrum of a hypothetical monolayer of CF4 of uniform crystalline (hexagonal) structure. The TO peak is very narrow and twice higher

Feature Article than the LO peak, and the absorption intensity between the two peaks is zero, in some contrast with the experimental spectra. Thus, the simulations suggest that the observed broadening of the TO peak (see Figure 18b,c) is due to lateral disorder in the distribution of the CF4 adsorbate, which reflects in turn the lateral disorder of the underlying ice surface. Figures lb,c show our models for such laterally disordered surfaces, generated as described in section 111. Additional evidence for nanocrystal surface disorder comes from simulations of pressure dependence of CF, spectra. While the true physical process shaping the nanocrystal surfaces is complex, lateral disordering of ice surfaces is expected to be driven by enhanced freedom of motion of surface molecules (with respect to the bulk) and their attempt to saturate their hydrogen bond c ~ o r d i n a t i o n . ~ . ~ . ~ ~ Intensity enhancement in the intermediate frequency region, between the LO and TO peak positions, is a signature of vertical disorder, Le. of surface roughness. This feature is particularly prominent in Figure l8a, which shows a spectrum of CFq on amorphous ice, which is characterized by a very rough surface. Annealing of ice nanocrystals at 140 K seems to increase the smoothness and perhaps the hexagonal regularity of the surface of nanocrystals formed during rapid cooling to 80 K, as the intensity intermediate to T and L is mostly lost and the transverse band becomes sharper (compare Figure 18b and c). Tentative interpretation of this annealing effect is that rapid crystallization at about 200 K generates particles with rough surfaces; and the nanocrystals are then cooled on a time scale that inhibits restructuring. Subsequent annealing then produces smoother surfaces representative of the annealing temperature (140 K). We do not know at present whether the lateral surface disorder, indicated by simulations of CF4 spectra for our smoothest nanocrystals (see Figures 1 and 18c,d), is a general characteristic of icy surfaces at thermodynamic equilibrium or a result of metastability induced by the experimental method of preparation. Though interpretation of the variations in the CF4 v3 band system are presently uncertain, it is noteworthy that the comparative experimental and simulated band structures represent a useful probe of the surface that will prove informative of the details of the ice surface structure and the effect of formation conditions on that structure.

VIII. Summary Remarks The combined spectroscopic and computational studies of the ice surface have shown that, as for bulk ice, one must be very conscious of the history and conditions of the ice in question. In some respects all surfaces of ice at low pressures are the same, as all are characterized by three basic molecular groups that we have labeled d-H, d-0, and s-4 and all give spectroscopic features and interact with adsorbates in a manner indicative of these molecular types. But in more general terms the ice surface depends on many factors such as the formation temperature, the annealing or relaxation temperature, the crystalline or amorphous nature of the underlying bulk ice, exposure to adsorbates, and exposure to reactive vapors. For these reasons, the future advance of knowledge of icy surfaces will depend on the ability to characterize the surface for each preparative method more than any other factor.

Acknowledgment. Research support by the U.S.National Science Foundation, the US.-Israel Binational Science Foundation, the Israel Science Foundation, and the Petroleum Research Fund of the American Chemical Society is gratefully acknowledged. The authors are also grateful to past and present research collaborators. In particular we thank Brad Rowland,

J. Phys. Chem., Vol. 99, No. 45, 1995 16547

Lance Delzeit, Tova Feldmann, Joanna Sadlej, Maria Szczesniak, and Marek Wojcik, for their productive collaborative work.

References and Notes (1) (a) Faraday, M. Philos. Mag. 1859, 17, 162. (b) Petrenko, V. F. The Surface of Ice. Special Report 94-22, U.S.A m y Corps of Engineers, August, 1994. (2) Jorgensen, W. L. J. Chem. Phys. 1982, 77, 4156. (3) (a) Rice, S. A.; Bergen, M. S.; Belch, A. C.; Nielson, G. J. Phys. Chem. 1983, 87, 4295. (b) Reimers, J. R.; Watts, R. 0. Chem. Phys. Lett. 1983, 94, 222. (c) Wojcik, M. J.; Buch, V.; Devlin, J. P. J. Chem. Phys. 1993, 99, 2332. (4) Kroes, G.-J. Surf: Sci. 1992, 275, 365. ( 5 ) Rowland, B.; Kadagathur, N. S.; Devlin, J. P.; Buch, V.; Feldmann, T.; Wojcik, M. J. J. Chem. Phys. 1995, 102, 8328. (6) (a) Buch, V. J. Chem. Phys. 1990, 93, 2631. (b) Zhang, Q.; Buch, V. J. Chem. Phys. 1990, 92, 5004. (c) Buch, V. J. Chem. Phys. 1992, 96, 3814. (7) Rowland, B.; Devlin, J. P. J. Chem. Phys. 1991, 94, 812. (8) Rowland, B.; Fisher, M.; Devlin, J. P. J. Chem. Phys. 1991, 95, 1378. (9) Buch, V.; Devlin, J. P. J. Chem. Phys. 1991, 94, 4091. (10) Callen, B. W.; Griffiths, K.; Norton, P. R. Surf: Sci. Letr. 1992, 261, L44. (11) Rowland, B.; Fisher, M.; Devlin, J. P. J. Phys. Chem. 1993, 97, 2485. (12) Hixson, H. G.; Wojcik, M. J.; Devlin, M. S.; Devlin, J. P.; Buch, V. J. Chem. Phys. 1992, 97, 753. (13) Devlin, J. P. J. Phys. Chem. 1992, 96, 6185. (14) Horn, A. B.; Chesters. M. A.: McCoustra, M. R. S.; Sodeau, J. R. J. Chem. SOC.,Faraday Trans. 1992, 88, 1077. (15) Devlin, J. P.; Silva, S. C.; Rowland, B.; Buch, V. In Hydrogen Bond Networks; Bellissent-Funel, M.-C., Dore, J. C., Eds.; Kluwer Academic Publishers: Dordrecht, The Netherlands, 1994. (16) (a) Silva, S. C.; Devlin, J. P. J. Phys. Chem. 1994, 98, 10847. (b) Sandford, S. A.; Allamandola, L. J.; Tielens, A. G. G. M.; Valero, G. J. Astrophys. J. 1988, 329, 498. (17) Rowland, B.; Kadagathur, N. S.; Devlin, J. P. J. Chem. Phys. 1995, 102, 13. (18) Schaff, J. E.; Roberts, J. T. J. Phys. Chem. 1994, 98, 6900. (19) Rowland, B.; Devlin, J. P. To be submitted. (20) See, for example: (a) Page, R. H.; Vernon, M. F.; Shen, Y. R.; Lee, Y. T. Chem. Phys. Left. 1987, 141, 1. (b) Pribble, R. N.; Zwier, T. S. Science 1994, 265, 75. (c) Buck, U. Private communication. (21) See, for example: (a) Makarova. M. A,; Ojo, A. F.; Karim, K.; Hunger, M.; Dwyer, J. J. Phys. Chem. 1994, 98, 3619. (b) Kiricsi, I.; Flego, C.; Pazzuconi, G.; Parker, W. 0.;Millini, R.; Perigo, C.; Bellussi, G. J. Phys. Chem. 1994,98,4627. (c) Florian, J.; Kubelkova, L. J. Phys. Chem. 1994, 98, 8734. (22) Tolbert, M.; Middlebrook, A. M. J. Geophys. Res. 1990,95,22423. (23) (a) Koehler, B. J.; Middlebrook, A. M.; McNeil, L. S.; Tolbert, M. A. J. Geophys. Res. 1993,88, 10563. (b) Delzeit, L.; Rowland, B.; Devlin, J. P. J. Phys. Chem. 1993, 97, 10312. (24) Horn, A. B.; Koch, T.; Chesters, M. A,; McCoustra, M. R. S.; Sodeau, J. R. J. Phys. Chem. 1994, 98, 946. (25) (a) Tielens, A. G. G . M.; Allamandola, L. J. In Physical Processes in Interstellar Clouds; Morfill, G. E., Scholet, M., Eds.; Reidel: Dordrecht, 1987. (b) Buch, V. In Molecular Astrophysics; Harquist, T. W., Ed.; Cambridge University: Cambridge, 1990. (26) Hagens, W.; Tielens, A. G. G. M.; Greenberg, J. M. Astron. Astrophys. 1983, 117, 389. (27) Sandford, S. A.; Allamandola, L. J.; Geballe, T. R. Science 1993, 262, 400. (28) (a) Geiss, J. Asrron. Astrophys. 1987, 187, 859. (b) Whipple, F. L. Astron. Astrophys. 1987, 187, 852. (29) Gross, G. W. J. Geophys. Res. 1982, 87, C9, 7170. (30) See, for example: Thiel, P. A.; Madey, T. E. Surf: Sci. Rep. 1987, 7, 211. (31) Mayer, E.; Pletzer, K. B. Nature 1986, 319, 298. (32) (a) Ritzhaupt, G.; Smyrl, N.; Devlin, J. P. J. Chem. Phys. 1976, 64,435. (b) Hardin, A.; Harvey, K. B. Spectrochim. Acta 1973, A29, 1139. (33) (a) Buch, V.; Silva, S . C.; Devlin, J. P. J. Chem. Phys. 1993,99, 2265. (b) Li, P. C.; Devlin, J. P. J. Chem. Phys. 1973, 59, 547. (34) Ewing, G. E.; Sheng, D. T. J. Phys. Chem. 1988, 92, 4062. (35) (a) Sadlej, J.; Rowland, B.; Devlin, J. P.; Buch, V. J. Chem. Phys. 1995, 102.4804. (b) Zhang, Q.; Chenyang, L.; Ma, Y.; Fish, F.; Szczesniak, M. M.; Buch, V. J. Chem. Phys. 1992, 96, 6039. (36) Jeziorski, B.; Kolos, W. In Molecular Interactions; Ratajczak, H., Orville-Thomas, W. J., Eds.; Wiley: Chichester, 1982. (37) Buch, V.; Devlin, J. P. J. Chem. Phys. 1993, 98, 4195. (38) Chandler, D.; Wolynes, P. J. Chem. Phys. 1981, 74, 4078. (39) Buch, V.; Devlin, J. P. Astrophys. J . 1994, 431, L135. (40) Reimers, J. R.; Watts, R. 0. Chem. Phys. 1984, 85, 83.

16548 J. Phys. Chem., Vol. 99, No. 45, 1995 (41) Hermansson, K.; Knuts, S.; Lindgren, J. J . Chem. Phys. 1991, 87. 4295. (42) Jones, L. H.; Swanson, B. I. J. Phys. Chem. 1991, 95, 2701. (43) Buch. V.; Devlin, J. P.; Delzeit, L.; Blackledge, C. To be submitted. (44) Honegger, E.; Leutwyler, S. J. Chem. Phys. 1988, 88, 2582. (45) Engdahl, A.; Nelander, B. J. Chem. Phys. 1987, 86, 1819. (46) Jenniskens, P.; Blake, D. F. Science 1994, 265, 753. (47) Davy, J. G.; Somorjai, G. A. J. Chem. Phys. 1971, 55, 3624. (48) Johari. G. P.: Hallbrucker, A,: Mayer. E. J. Chem. Phys. 1991, 95, 2955. (49) Hallbrucker. A.; Mayer. E.; Johari, G. P. J . Phys. Ckem. 1989.93. 7151. ( 5 0 ) Bergren, M. S.; Schuh, D.; Sceats. M. G.; Rice. S. A. J. Chem. Phys. 1978, 69. 3477. (51) (a) Torchet. G.; Schwarta, P.; Farges, J.: Feraudy, M. F.; Raoult, B. J . Chem. Phys. 1983, 79, 6196. (bj Bartell, L. S.; Huang, J. J. Phys. Chem. 1994, 98, 1455. (52) From unpublished work of B. Rowland. N. S. Kadagathur, and J. P. Devlin.

Devlin and Buch (53) Maes. G.; Smets, J . J. Phys. Chem. 1993, 97. 1818. (54) Silva. S. C.: Devlin, J. P. Unpublished results. ( 5 5 ) Kroes, B. G.; Clary, D. C. J. Phys. Chem. 1992. 96, 7079. (56) Engdahl. A.; Nelander, B. Chem. Phys. Left. 1983, 78, 4063. (57) Frisch, M. J.: Del Bene, J. E.; Pople. J. A. J. Chem. Phys. 1983, 78, 4063. ( 5 8 ) Kuhnke, K.; Harris, A. L.: Chabal, Y . J.: Jakob, P.: Morin, M. J. Chem. Phys. 1994. 100, 6896. (59) (a) Chang, H.-C.; Richardson, H. H.: Ewing, G. E. J. Chem. Phys. 1988, 89. 7561. (bj Berg, 0.;Ewing. G. E. S u d Sci. 1989, 220, 198. (60) Berreman. D. W. Phys. R ~ L '1963. . 130, 2193. (61) Ovchinnikov, M. A.: Wight. C. A. J. Chem. Phys. 1995. 102, 67. (621 See, for example: Gilbert. M.; Drifford, M. J. Chem. Phys. 1977. 66, 2701, JP95 l375N