This work describes measurements of the heterogeneous depositional nucleation of ice on micrometer-sized (NH4)2SO4 (AS), maleic acid (C4H4O4) (MA), and mixed AS/MA particles and on thin NH4NO3 (AN) films. The critical saturation ratio (Scrit) required for heterogeneous nucleation was found to be similar and temperature-dependent for all solids, ranging from Scrit = 1.42 ± 0.04 at 190 K to Scrit = 1.04 ± 0.05 at 240 K. Deliquescence of the solids was not observed in any experiment; ice nucleation always occurred below the deliquescence relative humidity. The observed saturation ratios for ice nucleation were all substantially lower than the saturation ratios required for homogeneous nucleation. These results suggest that vapor deposition of ice may compete with deliquescence for crystalline organic and inorganic solids below the eutectic temperature.
 Atmospheric aerosol particles influence climate directly by absorbing and scattering radiation and indirectly by influencing cloud formation and cloud optical properties. While there remains a significant level of uncertainty associated with both effects, the indirect effect is currently one of the largest sources of uncertainty in global climate models [Houghton et al., 2001]. In fact, the uncertainty associated with the indirect effect alone is as large as the total radiative forcing by CO2 [Houghton et al., 2001]. Thus the effect of aerosol particles on cloud properties must be further constrained if an accurate prediction of future climate change is to be made.
 As an air parcel containing aerosol rises through the atmosphere, it experiences decreasing temperature and, consequently, increasing relative humidity. Liquid aerosol particles supercool and may nucleate ice homogeneously. Similarly, solid aerosol particles may homogeneously nucleate ice if they first deliquesce. However, below the eutectic temperature deliquescence requires that the particles be supersaturated with respect to ice [e.g., Xu et al., 1998]. Therefore solid particles may also nucleate ice heterogeneously from the vapor phase below the eutectic temperature, setting up a competition between deliquescence and heterogeneous nucleation.
 Tropospheric aerosol is a complex mixture of organic and inorganic components. Single particle mass spectrometer data has shown that one of the most common particle types contains sulfate internally mixed with organics [Murphy et al., 1998]. In particular, carboxylic acids were noted in the mass spectra. Other data has shown that sulfate aerosol may be neutralized to varying degrees by ammonia [Talbot et al., 1998]. In regions where agricultural emissions mix with urban air, ammonium nitrate (AN) is also often internally mixed with sulfate and organics [Lee et al., 2002]. Furthermore, in some regions ammonium nitrate can be the dominant inorganic salt in the aerosol phase [Watson et al., 1998].
 The homogeneous nucleation of ice from AN, ammonium sulfate (AS) and H2SO4 aerosols has been extensively studied by a variety of methods and requires critical saturation ratios (Scrit) greater than ∼1.4 [Bertram et al., 2000; Chelf and Martin, 2001; Chen et al., 2000; Cziczo and Abbatt, 2001; Hung et al., 2002; Koop et al., 2000; Prenni et al., 2001b]. In contrast, there are only a few studies of the heterogeneous nucleation of ice from binary AS/H2O systems [Chen et al., 2000; Zuberi et al., 2001]. Furthermore, very little data exists for heterogeneous ice nucleation on organics [Prenni et al., 2001a] and none exists for internal mixtures of AS with organics or for AN. Numerous studies have shown that solid inclusions or multiple inorganic components in an aqueous particle can induce crystallization of inorganic salts at a relative humidity (RH) above the efflorescence RH of the pure aqueous aerosol [Hung et al., 2003; Martin et al., 2001a, 2001b; Onasch et al., 2000; Schlenker et al., 2004]. In addition, recent research has shown that organic components can also effect an increase in the efflorescence RH of mixed particles [Badger et al., 2005]. Therefore internally mixed particles may be found in the solid state under typical atmospheric RH conditions. In this manuscript, the heterogeneous depositional nucleation of ice on micrometer-sized AS, maleic acid (MA), 1:1 (by wt) AS/MA particles and on thin films of AN over the temperature range 190–240 K is examined.
2. Experimental Setup
2.1. Vacuum Chamber
 The vapor deposition of ice on micrometer-sized particles was studied using a vacuum chamber equipped with FTIR-Reflection Absorption Spectroscopy (FTIR-RAS). This instrument has been previously described in detail [Hudson et al., 2001]. Briefly, the infrared beam from a Nicolet Magna 550 spectrometer passes into the vacuum chamber where it is reflected from a gold substrate at a grazing angle of ∼84° from the surface normal. The beam exits the chamber and is detected with a mercury/cadmium telluride–A (MCT-A) detector. The temperature of the substrate is controlled by resistively heating against an attached differentially pumped liquid-nitrogen cryostat. Temperature is monitored with thermocouples attached to the backside of the substrate. A Teflon seal surrounds the substrate and separates the chamber from the cryostat, ensuring that the gold is the only cold surface exposed to the chamber atmosphere. Leak valves are used to introduce water vapor into the chamber. Vapor is drawn from a bulb containing HPLC-grade water subjected to several freeze-pump-thaw cycles. Water vapor partial pressures were read directly from a Baratron capacitance manometer as no other gases were present at significant levels in the vacuum chamber. The Baratron used for experiments in which T < 240 K has a range of 1 × 10−6 to 0.1 Torr and an accuracy of 0.08% of the reading (Baratron 690A). The Baratron used in the experiments at 240 K has the same accuracy, but a range of 1 × 10−6 to 1.0 Torr (Baratron 390).
2.2. Sample Preparation
 Particulate samples are prepared by spray depositing the desired aerosol onto the gold substrate. Aerosol is generated by atomizing an aqueous solution of the desired compound with a TSI constant-output atomizer. Samples are also deposited onto a Si wafer under identical conditions and analyzed with SEM to investigate particle size and morphology. As shown in Figure 1, agglomeration leads to a range of particle sizes, but most are spherical and 1–10 μm in diameter. The term particles will be used henceforth to describe this aggregate sample. Figure 1 also shows that all three types of spray-deposited particles contain numerous defects and appear to be polycrystalline. Furthermore, the particles have significant surface structure, increasing their surface area above that of a smooth particle of the same diameter. Finally, since we see only one particle type in the SEM image of the mixed particles (Figure 1b), we assume that mixed aerosol particles remain internally mixed upon crystallization. We note that regions of pure AS or pure MA may still exist within one particle.
 AN films are prepared in situ from its gas phase precursors. This process is described in detail in a previous publication [Shilling and Tolbert, 2004]. Briefly, HNO3 is deposited on the substrate at 160 K, neutralized with NH3, and warmed to 260 K. The resultant AN films are typically 30–50 nm thick. Because the initial HNO3 layer is deposited at low temperature and high HNO3 pressure, we believe the AN film completely covers the gold substrate. The infrared spectrum of a typical AN film produced in this manner is shown in Figure 2 (curve A). This spectrum agrees well with literature spectra of crystalline AN [Koch et al., 1996]. We note that in situ preparation of the AN films from gas phase precursors in a high vacuum chamber produces a sample that should be free of impurities.
2.3. Experimental Procedure
 Before beginning each experiment, the chamber pressure is reduced to less than 10−6 Torr to ensure that the solids are effloresced. Because the particles and films are dry at the beginning of the experiment, they should follow the deliquescence branch of the hysteresis curve. IR spectra are collected every 2.5 s using the first spectrum in the series as a background. Water partial pressure is then incrementally increased while monitoring for changes in the IR spectra. The size of these pressure steps is limited by the precision of the leak valve used. This precision is pressure-dependent, but is always less than 0.03 in terms of the ice saturation ratio (Sice), with somewhat smaller pressure increments possible at lower temperatures. The high angle of incidence used in the reflection technique combined and physical properties of the spray-deposited samples (i.e., a distribution of particles with diameters on the order of the wavelength of the IR light) caused scattering to dominate the IR spectra for experiments conducted on these samples. This necessitated the use of extinction features attributed to a change in the scattering properties of the system to detect ice nucleation during the particle experiments. Because the vapor-deposited AN films are flat on the scale of IR light, scattering was not a problem during these experiments. Therefore the growth of features in the OH stretching region of water ice (3600–3000 cm−1) was used to detect nucleation in these experiments. A decrease in the water partial pressure coincident with spectral changes in the IR at the ice nucleation point was observed in all experiments.
 An ice frost point calibration, in which the water vapor partial pressure is adjusted until no changes are seen in the IR spectrum of water ice, was performed at the end of each experiment. Temperature was calibrated by referencing the measured ice vapor pressure at the frost point to literature data [Marti and Mauersberger, 1993]. Within the accuracy of the thermocouple reading (±1.5 K), the calibrated temperature agreed with the temperature read directly from the thermocouple. The critical saturation ratio for nucleation (Scrit) was calculated by taking the ratio of the observed nucleation pressure to the frost point pressure. Identical experiments were performed on the blank gold substrate to ensure that ice nucleated on the particles and not on the partially covered gold substrate.
Figure 3 illustrates the experimental procedure used to determine the Scrit values reported in this work. The ice saturation ratio (Sice) is plotted on the left axis (solid trace) as a function of time during an experiment on pure AS particles. Sice is determined by taking the ratio of the measured water partial pressure to the ice frost point pressure determined at the end of the experiment. The right axis (dashed trace) shows the IR scattering as a function of time. The IR scattering was determined by integrating the area under an extinction peak; in this case the region 1349–1276 cm−1 was used. During this experiment at 240 K, the water pressure is incrementally increased, as seen by an increase in Sice. Scattering remains low and constant until ∼1200 s at which point it abruptly increases, indicating nucleation of ice has occurred. Nucleation is also indicated by the relaxation of Sice toward a value of unity (i.e., the ice vapor pressure). Although not shown in this experiment, we are able to maintain stable water pressures above the ice saturation point but below Scrit. Several different integration areas were tested for their sensitivity to scattering. Identical Scrit values are obtained regardless of which integrated area is used.
3. Results and Discussion
3.1. Supersaturations for Heterogeneous Nucleation
 The Scrit values determined in this work for the heterogeneous nucleation of ice on AS, MA, and mixed AS/MA particles, AN films, and the blank gold substrate are plotted as a function of temperature in Figure 4. Each point on the graph is the average of at least six experiments performed at that temperature. Error bars are derived by performing a t test at the 95% confidence limit on each set of experiments. In most cases, the Scrit value for nucleation on the solid salt or organic is significantly below the value for nucleation on the gold. This indicates that nucleation preferentially occurs on the deposited film or particles. It can be seen that the Scrit values required for heterogeneous nucleation of ice on all the solids tested are less than the supersaturations typically required for homogeneous nucleation from AS or H2SO4, which range from Scrit ∼ 1.7 to Scrit ∼ 1.4 [Bertram et al., 2000; Chelf and Martin, 2001; Chen et al., 2000; Cziczo and Abbatt, 2001; Hung et al., 2002; Koop et al., 2000; Prenni et al., 2001b]. As a result, crystalline AS, AN, MS, and mixed organic/inorganic particles in the atmosphere may nucleate ice at lower saturation ratios than deliquesced particles.
3.2. Temperature Dependence of S Values
 As seen in Figure 4, all of the Scrit values display a strong temperature dependence. Classical heterogeneous nucleation theory predicts that Scrit is strongly driven by the compatibility between the substrate and the nuclei. Therefore classical nucleation theory, described in section 3.6, should be able to predict the temperature dependence of the critical saturation ratio. For the data set presented here, this was not the case. The cause of this discrepancy is unclear, but there are several possibilities.
 First, it is possible that the surface tension of the ice/air interface varies strongly as a function of temperature. No data on the temperature dependence of σ for the ice/air interface exists in the literature. Using a constant value for σ across the entire temperature range, as was done in this case, would then underestimate the temperature dependence of Scrit. However, calculations show that σ would need to increase by a factor of five from 240 to 190 K to explain the data. This is physically unrealistic.
 Alternatively, the slow growth rates of the ice at lower temperature may affect our ability to detect nucleation, resulting in an overestimation of the nucleation point. However, similar trends in Scrit values with temperature have been observed for the heterogeneous nucleation of ice on sulfuric acid tetrahydrate (SAT) in our laboratory [Fortin et al., 2003]. While a similar detection scheme was used in these two studies, the temperature range and hence ice growth rates were different. Therefore it seems likely that the observed temperature trends Scrit are, at least in part, real. The failure of classical nucleation theory to predict a temperature dependence of Scrit suggests that it is inadequate in describing depositional nucleation.
3.3. Deliquescence of Substrates
Figure 5 shows the relative humidity with respect to liquid water at the time of nucleation on pure AS particles as a function of temperature. RH was calculated by taking the ratio of the observed nucleation pressure to the saturation vapor pressure of pure water calculated using the Wexler  expression. As before, error bars are derived by performing a t test at the 95% confidence limit on the data. Also shown on the graph is the theoretical deliquescence relative humidity (DRH) line for pure AS obtained by linear extrapolation of the Clegg et al.  model, the ice saturation line, and the data of Braban et al.  for reference. The few literature reports of subeutectic deliquescence suggest that this extrapolation accurately reflects the low-temperature DRH of AS [Braban et al., 2001; Fortin et al., 2002]. As shown in Figure 5, ice deposition occurred at a relative humidity significantly below the theoretical deliquescence relative humidity in all experiments. In addition, no changes were observed in the liquid water region of the IR prior to ice nucleation. In light of these two observations, we conclude that deliquescence did not occur during experiments conducted on the spray-deposited aerosol substrates. However, owing to the poor quality of the IR spectra during particle experiments, we cannot confirm that deliquesce followed by immediate ice nucleation did not occur.
 High-quality IR spectra were obtained for experiments conducted on the AN films. Spectra of the AN film immediately before (curve A) and after (curve B) ice nucleation are shown in Figure 2. As seen in curve A, there are no signs of liquid water present in the spectrum just prior to ice nucleation. Instead, the AN film shown in curve A is in excellent agreement with literature spectra of effloresced, crystalline AN films [Koch et al., 1996; Shilling and Tolbert, 2004]. As in our previous report, there are no changes in the liquid water region of the AN spectrum as RH is increased until ice nucleation occurs [Shilling and Tolbert, 2004]. The spectrum shown in curve B of Figure 2 is in excellent agreement with literature spectra of crystalline ice under similar conditions [Horn et al., 1995; Zondlo et al., 1997]. Thus, for the case of nucleation of ice on the AN films, we can conclude that deliquescence followed by immediate homogenous nucleation did not occur. These results suggest that, for polycrystalline particles, ice nucleation is favored over deliquescence below the eutectic point.
3.4. Nucleation Selectivity
 To illustrate the selectivity of the ice nucleation process, photographs of the gold surface at 240 K are shown in Figure 6 before ice nucleation on the particle-coated substrate (Figure 6a), after ice nucleation on the blank gold substrate (Figure 6b) and after ice nucleation on the AS-coated substrate (Figure 6c). As seen in Figures 6b and 6c, ice was visibly observed to nucleate on machining lines at the edge of the gold substrate during the blank experiments and on the substrate itself during the particle experiments. Furthermore, in all cases, the ice formed distinct islands rather than a smooth film, indicating that nucleation was highly selective. By taking the ratio of the number of ice islands that were visibly observed to the number of particles on our substrate as determined from the SEM images, we estimate that 1 in 105 particles nucleate ice at 240 K. At lower temperatures and higher Scrit values, nucleation was less selective and we were unable to count distinct ice islands. Therefore we cannot estimate the fraction of particles that act as IN at lower temperatures.
3.5. Literature Comparison
 The observation of ice nucleation occurring in favor of deliquescence is not surprising if one considers that liquid droplets are supersaturated with respect to ice below the eutectic [Xu et al., 1998]. However, IR observations of subeutectic deliquescence of ammonium sulfate do exist in the literature [Braban et al., 2001; Fortin et al., 2002]. The observation of ice nucleation at Scrit values of close to unity also contradicts the work of Chen et al.  who concluded that effloresced AS particles were poor heterogeneous ice nuclei. However, Zuberi et al.  observe heterogeneous nucleation of ice from aqueous solutions of AS on microcrystalline AS inclusions at Scrit values close to 1. In the same study, Zuberi et al. noted that solutions containing a single crystalline AS inclusion froze at temperatures closer to the homogeneous freezing temperature.
 The work of Fortin et al.  was conducted on AS films, rather than on aerosols. SEM images revealed that this film contained relatively few defects and had a smooth surface [Fortin, 2002]. As seen in Figure 1, the spray-deposited aerosol particles of this work contain numerous defects and extensive surface structure. Ice nucleation is likely to be initiated at these defect sites on the aerosols. Therefore the discrepancy between this work and that of Fortin et al.  is likely to be caused by the differences in the crystal structure and morphology of the AS. Furthermore, Fortin et al.  observed ice nucleation in favor of deliquescence in about half of their experiments. Observations of ice nucleation in favor of deliquescence were more common at the lower temperatures studied in that work, in accord with our observations. Fortin et al.  attribute the observation of both deliquescence and ice nucleation under identical conditions to the stochastic nature of the nucleation process.
 In the Braban et al.  and Chen et al.  studies, AS particles were generated by atomization, as they were in this work, so it is less likely that the different results were caused by differences in the crystal structure or morphology of the AS particles. Chen et al.  studied monodispersed, 0.2 μm diameter particles, while in this work the particles were polydispersed and 1–10 μm in diameter. This difference in particle size may contribute to the differences in the results. Heterogeneous nucleation theory, as discussed below, suggests that changing the particle diameter from 10 μm to 0.2 μm does not significantly affect Scrit. However, literature data suggests that there is a considerable size effect for particles in the hundreds of nanometer size range [Martin et al., 2001a; Onasch et al., 2000]. Studies investigating the heterogeneous crystallization of ammonium sulfate in droplets containing insoluble mineral cores found that the crystallization RH decreased as inclusion size decreased, indicating nucleation on smaller particles is less efficient [Martin et al., 2001a; Onasch et al., 2000]. Therefore particle size may be more important than is predicted by classical heterogeneous nucleation theory.
 Furthermore, we hypothesize that our experimental method provided an increased sensitivity to nucleation over the Chen et al.  method, allowing us to see a smaller number of particles nucleating at lower Scrit values. Indeed, Chen et al. report their threshold for detection of nucleation is 1 particle in 102–103, while we can detect 1 nucleation event in 105. Furthermore, the residence time of the particle in the Chen et al.  instrument is only ∼10 s, while the dry particles in this work were exposed to water vapor for times on the order of 10 min before nucleation was observed. Since nucleation is a stochastic event, the difference in the experimental timescales may also have contributed to the conflict of results.
 The temperature overlap between this work and that of Braban et al.  is small and Braban et al. do not see changes in the IR at the lowest temperatures reported in their work. Instead, they observe a pressure drop in their chamber, which they attribute to deliquescence followed by nucleation of ice. We observe a pressure drop coincident with changing IR features. Owing to the differences in the geometry of the observations (transmission versus reflectance) we are likely to be more sensitive to changes in the IR. Furthermore, in this work, IR spectra cannot be used to definitively determine that deliquescence followed by immediate freezing did not occur for the AS data. The fact that the RH could not be raised above the extrapolated theoretical DRH curve certainly suggests that deliquescence did not occur, but extrapolation of this curve may not be valid at the temperatures studied in this work.
3.6. Determination of the Contact Parameter and Matching Function
 As mentioned in section 3.1, the Scrit values required for heterogeneous depositional nucleation are significantly below those required for homogeneous nucleation. This suggests that cirrus clouds may nucleate at lower RHs than previously believed if solid nuclei are available. Also, cirrus clouds produced by heterogeneous nucleation would likely have optical properties different from those formed by homogeneous nucleation. Ultimately, cloud models must be used to determine the effects of heterogeneous nucleation on cloud properties. To this end, we have analyzed the data using heterogeneous nucleation theory as described by equations 9-24, 9-27, and 9-28 of Pruppacher and Klett  to determine the contact parameter (m) between ice and our spray-deposited particles [Pruppacher and Klett, 1980].
 We find that the contact parameter values are independent of the substrate, ranging from 0.94 at 190 K to 0.99 at 240K. The average value for this contact parameter for all substrates over the entire temperature range is 0.97 ± 0.01. The calculated m values are all close to unity, indicating that AS, maleic acid, and AS/maleic acid particles generated by atomization are highly efficient ice nuclei. These results suggest that an excellent lattice match exists between ice and all three substrates. However, we have calculated the disregistry (δ) between single crystals of ice and the pure substrates using equation (1)
where an is the lattice constant of the substrate unit cell [Donnay and Ondik, 1973a, 1973b] and a is the lattice constant of the ice unit cell [Pruppacher and Klett, 1980]. The resulting disregistries between ice and the pure substrate lattices are all greater than 3% and are typically tens of percent. For reference, Turnbull and Vonnegut report that nuclei form coherently on a substrate only for δ ≤ 1.5% [Turnbull and Vonnegut, 1952]. Therefore we believe that the particles' morphology rather than their chemical identity drives the nucleation. Specifically, nucleation is initiated on a small number of active sites rather than on the “bulk” substrate. Unfortunately, the nature of these active sites remains unknown.
 In summary, we find that heterogeneous nucleation of ice on dry polycrystalline AN films and AS, mixed AS/MA, and pure MA particles occurs preferentially over deliquescence for the temperature range studied in this work. The critical saturation ratio (Scrit) required for heterogeneous nucleation was found to be similar and temperature-dependent for all solids, ranging from Scrit = 1.42 ± 0.04 at 190 K to Scrit = 1.04 ± 0.05 at 240 K. We have also estimated that only 1 in 105 particles nucleates an ice germ at 240 K.
 A recent paper reports a high percentage of sulfate and salts in cirrus ice residue at temperatures below 233 K, which is attributed to homogeneous ice nucleation [Twohy and Poellot, 2005]. Our results indicate that crystalline sulfate, nitrate, and organic particles could also be effective heterogeneous ice nuclei at these temperatures. Furthermore, Twohy and Poellot find that heterogeneous nucleation becomes more important at warmer temperatures. We also observe that heterogeneous nucleation is more efficient at warmer temperatures on the crystalline surfaces studied in this work.
 A number of studies have now shown that mixed inorganic and organic/inorganic aerosol exist in the solid phase over a wider range of RH conditions than previously believed [Badger et al., 2005; Martin et al., 2001a; Onasch et al., 2000; Schlenker et al., 2004]. In light of these studies, aerosol reaching the upper troposphere may provide heterogeneous nucleation centers for cirrus cloud formation. In the atmosphere, nucleation is influenced by complex dynamics and must be evaluated carefully with a cloud model to determine the end effect of heterogeneous nucleation on cloud microphysical parameters. Such an evaluation is beyond the scope of this work; however, our work suggests that heterogeneous nucleation may play a role in determining these microphysical parameters.
 This work was supported by the Biological and Environmental Research Program, U.S. Department of Energy under grant DE-FG03-01ER63096 and by NASA-SASS under grant SA98-0005. J.E.S and T.J.F were supported by NASA ESS fellowships NGT5-30422 and NGT5-301136, respectively.