2.1. X-ray analysis calibration

It is difficult to adequately calibrate X-ray measurements where an impurity of size less than the excitation volume is surrounded by a medium of different conductivity. To calibrate X-ray analysis measurements, Reference Reid, Potts, Oates, Mulvaney and WolffReid and others (1992) froze aqueous standards into track-etched (Nuclepore) filters to simulate impurity-filled veins in ice. Constraining the acid in the filter holes prevented significant partitioning of the impurity phase during freezing. They found a linear relationship between elemental concentration and the X-ray count rate stimulated by the incident electron beam. Here the calibration was repeated, this time using sulphuric acid solutions without a surface coating to replicate conditions in this study. Acids of 0.5, 1.0 and 4.9 mol L^-1 (the eutectic concentration) were used, frozen into 0.4 μm pores in the polycarbonate membrane. The technique involved rapidly freezing the saturated filters between liquid-nitrogen-cooled copper blocks. The absence of a conductive coating allowed a large variation in the detected X-ray count rates; however, an approximately linear relationship was still seen between the integrated elemental count rate for sulphur and the acid concentration (Fig. 1a). The presence of a small carbon peak in the collected spectra indicated that the X-ray excitation volume extended beyond the circumference of the holes. The count rates were even more variable for the same acids frozen into larger pore sizes (2 and 8 μm), suggesting that impurity segregation was occurring in the frozen mixture.

Fig. 1. (a)The relationship between integrated count rate collected by X-ray analysis and concentration of sulphuric acid frozen into 0.4 μm holes (error bars give standard deviation of measurements). (b) Cut surface of frozen 0.5 mol L^-1 sulphuric acid (uncoated). Partitioning can be clearly seen between the bulk ice and the interstitial acidic components.

We make a crude estimate of the minimum quantity of sulphate detectable on an ice surface under the conditions in this study. The shape and volume of interaction of electrons within ice is not well established, particularly in the presence of impurities (Reference EchlinEchlin, 1992). Early work by Brombach, as well as Fuchs and Lindeman, suggested the conversion of the interaction volume from pear to pancake shape (400 nm wide, 100 nm deep with the beam accelerated by 5 kV) after several seconds of exposure on a pure ice surface (reviewed by Reference EchlinEchlin, 1992). In contrast to this, Reference Oates and PottsOates and Potts (1985) detected sulphur from below an ice layer 3 yum thick when it was stimulated with electrons accelerated by 15 kV; however, 90% of X-ray emission should occur within the top 2 μm. Based on these latter results, the volume of acid stimulated in a hole of diameter 0.4 μm is approximately 0.25 μm³; if the acid is at 0.5 mol L^-1 then the signal emitted originates from 1.3 × 10^-16 mol of sulphur. In this study, a signal around five times less than this was no longer distinguishable from a background spectrum. Given a linear relationship between count rate and impurity concentration (Fig. 1a), this indicates that the minimum detectable quantity of sulphur is about 3 × 10^-17 mol. We estimate the minimum detectable quantities for Na, Mg and Cl to be factors of 2.6, 1.0 and 1.5 greater, respectively, than the value for sulphur (Reference Reid, Potts, Oates, Mulvaney and WolffReid and others, 1992).

It is generally presumed that the frozen solution formed between rapidly cooled ice grains is at the eutectic composition regardless of the bulk ice concentration if curvature and substrate effects are negligible. This is expected for liquid components of the polar specimens in this study on freezing in liquid nitrogen. To test whether this is the case, sulphuric acid droplets of varying concentration were rapidly frozen on a copper block cooled by immersion in liquid nitrogen (-196°C). Flat surfaces were microtomed for the 0.5 (see Fig. 1b) and 1 mol L^-1 droplets, but the 4.9 mol L^-1 droplet (the eutectic concentration) was left uncut as a droplet since it would be liquid at the cutting temperature of -20°C. No sulphur is detectable from the surfaces of the ice grains, but it is clearly present in the interstitial regions surrounding the small rapidly formed grains. If the interstitial solution is at the eutectic composition in all cases, we expect similar count rates from each droplet. The net sulphur peak count-rate integrals of the measured spectra were 53±39, 102 ± 75 and 36 ± 9 counts per second for the 0.5, 1.0 and 4.9 mol L^-1 specimens, respectively (a minimum of five spectra were used for each measurement). Because of the effects of charging, and of geometry in the concentrated drop, the results are very variable and we can only say that the concentrations of the interstitial solution are high and do not rule out a uniform interstitial composition.

2.2. Interpreting etching

Understanding the features of an etched surface is important when ascertaining the origin of observed impurity. As shown by Reference BarnesBarnes (2003), the location of etching channels formed on a surface by rapid sublimation under vacuum conditions (etching) does not necessarily coincide with the location of the grain boundary on the surface. This is exemplified by the bubble cavity embedded in an etched surface shown in Figure 2. The grain-boundary grooves visible in the bubble cavity, a site where less sublimation occurred because of curvature, do not match the location of the surface etching channels at the top and bottom of the image. The location of the grain boundary on the cut surface is only seen when it is congruent with the relatively smooth base of the etching channel; in this case a ridge is noted, probably forming because of the coagulation of hydrated impurity salts concentrated at the boundary. Numerous other examples were noted during this study where, after etching, previously connected grooves at surface-cavity interfaces no longer matched.

Fig. 2. The intersection of the etching surface with the grain boundary grooves on the bubble cavity surface, marked by arrows (scale bar 200 μm). The ice taken from 140 m at Dome C had been cut and pre-etched for 1 day at-20°C before being etched for 10 min at -70°C in the SEM. Note the mismatch between the grain-boundary location (GB) and the etch channels (EC) on the cut surface; there is no sign of the grain boundary on the surface when the two do not match.

The effect is expected because of the difference between pre-etching and etching. The grain-boundary grooves that initially form under vapour saturation conditions during pre-etching balance the surface free energies at the interface. During sublimation under a vacuum (etching) the vapour pressure above the surface is no longer at or close to its saturation point. Consequently the observed etching channels develop because of the topological features present on the surface at the start of etching, not because of the presence of the higher-energy density at the grain boundary (an un-pre-etched surface produces no discernible etching channels because the grain-boundary grooves are so small).

Consequently, grain boundaries and triple junctions on an etched surface will only be distinguishable by surface fabric differences, unless they coincide with etching channels. We expect that, during etching, any impurity at a grain boundary perpendicular to the surface will coagulate as a filament (a slender, thread-like body) on the retreating surface; however, it could also coagulate to form discrete particles or evaporate from the surface if sufficiently volatile. Impurity at a grain boundary or vein that is not parallel with the direction of etching will be spread across the retreating surface, ending up as coagulated surface spots. Some data in previous studies using etching should be reevaluated in light of this.

3.1. Snow and firn

Figure 3 shows images of surface snow through to firn at the pore close-off depth. The complex faceted structures observed in snow crystals (Reference Rango, Wergin and ErbeRango and others, 1996a) and diamond dust (Reference KumaiKumai, 1976) rapidly sinter and densify to the observed smooth-surfaced structures similar to those observed by Rango and others (1996b). Grain-boundary grooves and triple junctions are apparent between the grain surfaces and at the necks between crystals. The dihedral contact angle was measured at grain boundaries between sintering snow crystals in instances where the bond between grains was perpendicular to the angle of observation, and the specimen had not been etched. Measurements of grain boundaries at 13 bonds from three different specimens (two from Halley, one from Dome C) gave a mean air-ice dihedral angle of 132 ±3°.

Fig. 3. Dome C snow and firn. Examples of the location of grain boundaries (GB) and triple junctions (TJ) marked. (a) Sintering surface snow, with a density of approximately 340 kgm^-3. No etching has been performed on this specimen. Grain-boundary grooves can be seen at the necks between grains. Small ice crystals suspended in the liquid nitrogen used to cool the ice have been deposited on the surface. (b) Firn from 23.3 m depth with an approximate density of 526 kgm^-3.The surface has been cut to reveal an un-etched section. (c) Firn from 95.8 m with a density of 820 kgm^-3 around the pore close-off depth. The surface has been cut, pre-etched for 1 day and then etched in the SEM for 10 min at -70ºC to reveal the location of the grain-boundary grooves more clearly.

3.2. Pore surface impurity

Soluble impurity was detected on un-etched cavity (vapour-ice) surfaces, both on snow grains and on bubble walls. Figure 4 shows examples of both of these features in Dome C snow and ice. It should be noted that impurity detected on an un-etched cavity surface was not common. The filament of NaCl associated with the rim of the bubble in Figure 4b has coagulated from the bubble surface during etching, and has a similar composition to other impurities noted on the sublimating surface. Coagulating spots of impurity were also seen on the rims of bubbles from other depths (Table 1).

Fig. 4. Surface impurity. (a) Surfaces of sintering surface snow grains from Dome C; the surfaces shown are un-etched (scale bar 500 μm) Traces of chlorine were detectable on some grain surfaces. Grain surface 1 (GS1) showed no detectable impurity. However, the adjacent grain surface (GS2) consistently showed traces of chlorine at the detection limit; if sodium was also present, it was below the detection limit. (b) Section of a bubble wall, from Dome C 243.50 m (scale bar 50 pm).The surface has been pre-etched for 1 day and then etched for 8 min at -70°C (~400 pm removed). The bubble surface filament (BSF) is a hydrated salt coagulated from the surface of the bubble at its rim during etching and has a width of ~ 2 μm. A grain-boundary filament (GBF) has also formed at the original location of the grain boundaries on the un-etched surface; the position of the filament triple junction (TJ) is marked. During etching, spots of impurity also coagulate on the surface (SES), while no impurity can be detected in the dark background areas (e.g B). (c) X-ray spectral data for (a) and (b), offset by multiples of 2 counts ^-1.

3.3. Impurity at grain boundaries

Occasionally impurity could be located at grain boundaries without etching (Reference Barnes, Mader, Rothlisberger, Udisti and WolffBarnes and others, 2002b; Reference BarnesBarnes, 2003). However the examples of impurity found at grain boundaries shown here were all revealed by the use of surface etching, since impurity is not usually detectable on a freshly cut or pre-etched surface. Caution must be exercised when interpreting any etched surface, since the process of sublimation can enhance the mobility of some coagulated impurities and obscure their original location, as discussed in section 2.2. Figure 5 shows examples of impurity filaments, coagulated filaments and impurity ridges originating from grain boundaries. Images from the etched necks of snow crystals at Halley (Fig. 5a and d) and Dome C (Fig. 5b and e) are seen. The etch rate from snow surfaces is likely to be generally lower than for flat cut surfaces because of the porous structure. The tortuous path of the filaments at bonds between neighbouring grains suggests an impurity component with some structural strength under compression. As the radius of the bond decreases during sublimation, the initially smooth band of coagulated impurity was observed to buckle as it sticks to the surface. The presence of Mg and S (Fig. 5c) in the filament on the uncut surface but not on the cut surface of the bond shown in Figure 5d may indicate the role of pre-etching in concentrating impurities on the specimen surface.

Fig. 5. Impurity at grain boundaries. (a) Neck between grains of sintering surface snow from Halley. The specimen has been etched for the equivalent of 6 min at -70°C. A grain-boundary filament (GBF) has developed at the neck during etching Its tortuous appearance suggests that it has buckled as the neck’s radius reduces with sublimation. No impurity was directly detectable in this filament. (b) A similar neck between grains of Dome C surface snow. The specimen has been etched for 4 min at-70°C. A grain-boundary filament has developed in the etching channel (EC) formed at the location of the grain-boundary groove. Note that the filament, which did not contain detectable impurity, is not continuous around the boundary, suggesting that the bond is only partially covered by an impurity layer. (c) X-ray spectra for many of the points marked with arrows in (d) and (e), offset by multiples of 2 counts ^-2. (d) A bond between grains of Halley surface snow that has been cut by the microtome, pre-etched for 1 day, and then etched in the SEM for the equivalent of 6 min at -70°C. A grain-boundary filament has developed at the etching channel on both the snow grain surface and the cut surface. The filament on the snow grain surface, which has split into two strands during etching, contained detectable magnesium and sulphur (GBF1) while no impurity was detectable in the filament on the cut surface, (e) Dome C ice from 243.30 m depth. The surface has been cut but not preetched, and then etched in the SEM for 10 min at -60°C (600 μm). A clear grain-boundary filament has been produced, apparently joined at triple junctions (TJ); it can be seen in the insets and has a diameter of approximately 400 nm. Spectra were taken from the filament (GBF2, GBF3) and also from the surface spots (GBS) that have formed by the coagulation of the grain-boundary filament during etching and the triple junction (TJS). Spectra were also collected from typical background surface etching spots (SES) and from the dark areas of the background (B).

The freshly cut, then etched surface shown in Figure 5e is from Dome C at 243.30 m, a depth that corresponds to an acidic sulphate layer (probably fallout from a volcanic eruption) with a meltwater concentration of about 5 μM (Table 1). Filaments and particles have coagulated on the surface without etching channels, since there was no pre-etching. The relatively large quantity of ice sublimed from the surface (600 μm) may have included a grain-boundary plane; its contents could be added to the bulk impurity already coagulated on the surface.

3.4. Impurity at triple junctions

For natural samples, impurity was not noted at triple junctions without some form of pre-etching or etching. Pre-etching the surface of a specimen after cutting was found to be the most reliable method for revealing impurity at triple junctions; the greater the duration of pre-etching, the greater the likelihood of detecting impurity; the implications of this are discussed in section 4.2. Even in ice containing high bulk concentrations of acid, impurity located at triple junctions was not always apparent after pre-etching. Triple junctions can be seen in Figure 6, prepared by a variety of methods and showing examples of triple junctions with and without impurity, and their appearance after etching. Impurity concentrated at triple junctions was normally observed only in ice that contained relatively high bulk concentrations of acid, except where etched filaments joined (e.g. Fig. 5e), as can be seen from Table 1.

Fig. 6. Triple junctions. (a) Dome C ice from 243.30 m, surface cut and pre-etched for 3 days. Although this ice contains a relatively high concentration of sulphate, no sulphur is detectable at the triple junction, although it is apparent on others on the specimen. The junction is also typical of those observed in background ice. (b) The triple junction seen in (a) after etching for 7 min at -80ºC. A grain-boundary filament is visible in the etching channel, although the impurity is below the detection limit, both at the triple junction and in the filament. (c)Trace sodium and chlorine are revealed at this triple junction by1min of etching, at -80ºC, of the surface of this Dome C ice from501m (the surface was not pre-etched). Note that grain-boundary filaments are visible close to the triple junction; they die out after around 15 m. (d) A triple junction containing only sulphur (TJ1) detected in Dome C ice from 243.30 m. The surface was pre-etched for 1day. The dark spot to the left of the triple junction is a hole that was burned by the electron beam. (e) Atriple junction containing sulphur and other insoluble impurities (TJ2), found on the same specimen as (d).The grain boundary (GB) in the vicinity of the triple junction was not significantly different from the background spectrum (B). (f) Triple junction observed in Dome C ice from 218.95 m, pre-etched for 7 days. A very clear filled triple-junction structure is apparent (TJ3) and the impurity appears to extend into the grain boundaries, although impurity is not detectable in these locations. (g) X-ray spectra for (d–f), offset by multiples of 2 counts s^-1.

3.5. Bulk impurity and dislocations

A significant quantity of impurity could be at defects within the ice lattice. Furthermore, impurity will be incorporated directly as inclusions. In addition to the spots of detectable impurity shown in Figures 3b and 4d, Figure 7a shows further examples of spots built up on the surface of ice from a Dome C acidic layer. The spots progressively coagulated and separated throughout the etching process. They show trace of chlorine that was likely to have been at defects in the lattice (dissolved) before etching. Sublimation of around 140 μm of ice has left coagulated hydrated impurity on the surface. Also shown are brighter spots or inclusions containing chlorine in greater quantities at the edge of etching channels. It is likely that these are formed by the advancing edges of the channels sweeping up the smaller spots.

Fig. 7. Bulk impurity and inclusions. (a) Dome C ice from 218.9.5 m, pre-etched for 7 days and then etched for 7 min at -70°C. Etch channels (EC) show original location of the grain boundaries; some contain filaments (GBF). Surface etching spots developed on surface (see inset) in which only chlorine was arguably detectable (SES1). Either inclusions or coagulated surface etch spots (INI) are located on the rim of the etching channels. (b) Dome C ice from 243.50 m, pre-etched for 10 days and etched for 5 min at -70°C. The surface imaged was located in the centre of a grain that was featureless before etching Disconnected filaments are revealed, which may be indicative of impurity located at dislocations (DF). A bulk inclusion has also been revealed (IN2), which has a similar composition to the surface etching spots (SES2). These can be compared with the dark background (B2). (c) X-ray spectra for (a) after 10 min at -70°Cand (b), offset by multiples of 2 counts s^-1.

The surface in Figure 7b was featureless before etching. The etching spots also contain detectable levels of sodium, in contrast to the previous example, and a bulk inclusion of similar composition is also seen. However, the most striking features are the long filaments seen on the surface that do not appear to be connected to any grain boundaries or triple junctions. These sodium-chloride-filled filaments could be residues of impurity originating from dislocations within the bulk of the grain.

4.1. Grain-vapour surfaces and grain boundaries

Filament ridges (i.e. a filament structure on top of a ridge generally observed at the base of etching channels) were noted at the majority of grain-boundary bonds in both Dome C and Halley snow samples after etching. We interpret this as indicating the presence of soluble impurity, although it was below the X-ray analysis detection limit in most cases. In sintering surface snow of 340 kg m^-3, there is a surface area of around 26 m² kg^-1 (Domine and others, 2002). Assuming a mean grain radius of 200 μm and a bond radius of 100 μm (Fig. 3a and other observations) we can calculate the approximate number of grains per kilogram to be [917 × 4/3 π (2 × 10^-4)³]^-1 ≈ 3 × 10⁷ kg^-1 and the mean bond area to be 3 × 10^-8 m². The grain coordination number is approximately 3.2 neighbours per grain (Reference Alley, Perepezko and BentleyAlley and others, 1986b), which gives 1.6 bond areas per grain. The grain-boundary area (i.e. the area of ice-ice interface) is then 1.6 m² kg^-1, a factor of 16 times less area than the air- ice interface.

It is illustrative to consider what proportion of this surface could be covered by impurity. For a liquid-like monolayer primarily composed of water molecules, there will be of the order of [6 × 10²³ (mole^-1) × 10⁶ (g m^-3)/ 18 (g mole^-1)]^2/3 ≈ 1 × 10¹⁹ molecules m^-2 of which, if we assume a concentration close to the eutectic, about 10% will be an impurity species. The monolayer would therefore require 1 × 10¹⁸ molecules of impurity m^-2(~1.7 yμmol m^-2). This value will in reality be lower because of the influence of the surface interactions in such a thin layer (Reference WettlauferWettlaufer, 1999), which will lower the equilibrium concentration below the eutectic. However, the surface impurity density for a liquid-like monolayer quoted above will be used as an example to estimate the coverage of grain boundaries. Typical bulk impurity concentrations at Dome C are around 1.5 μmol L^-1 for the sodium and sulphate species, both of which are highly insoluble in the lattice. If all this impurity formed a monolayer, it would cover 1.5 (μmol kg^-1)/ 1.7 (μmol m^-2) ≈ 0.9 m² kg^-1. This value is of a similar magnitude to the area of ice-ice interface (grain boundary) in the snow, but considerably less than the air-ice interface area, where only about 3% would be covered by any one species. In the case of Halley snow where sodium and chlorine concentrations are around 20 μmol L^-1, there is potential for a much higher proportion of the snow surface to be coated if aerosol inclusions are available in liquid form during the summer.

From these considerations we can surmise that, even if all impurity is on snow-grain surfaces, the impurity is almost certainly distributed heterogeneously, with coatings (possibly more than a monolayer thick) on some grains and absent on others (see Fig. 3a). The evidence suggesting that impurity is resident at many bonds therefore implies that it is energetically preferable for at least one of the impurity species to exist at a grain boundary as opposed to a grain surface. On the few occasions when impurity was detectable in snow and upper firn at filaments, sulphate ions were common to all examples, although invariably in conjunction with other ions (Table 1). The only location where sulphate ions were observed originating from the grain surface was on the rim of a bubble in ice from an acidic sulphate layer in which the concentration of sulphate was high enough to coat all available grain boundaries with at least a monolayer (an acidic bulk concentration of 5.2 μmol L^-1 could cover up to ~ 3 m² kg^-1 with a monolayer; the available grain boundary at this depth is around 2.2 m² kg^-1).

At other depths less rich in sulphate, a far smaller proportion of boundaries were filled, and elements detected in filaments tended to be sodium and chlorine, indicating their presence at grain boundaries (Table 1; Fig. 5e). Filaments containing detectable impurity were observed coagulating at the rims of bubbles during etching (Fig. 4b). This implies that the impurity originated from the bubble surfaces and was observed in specimens where bulk impurity concentrations were insufficient to coat all grain boundaries (at concentrations of the magnitude proposed), suggesting a possible affinity of sodium for grain-vapour interfaces in preference to grain boundaries. Whatever the case, it seems clear that impurity is again heterogeneously distributed at grain boundaries. The likelihood of a boundary being occupied is probably dependent in part on its relative lattice orientations. Closely aligned crystals are likely to exclude impurity from their interface, while the presence of an impurity layer between two lattices of different orientation may be favourable.

The concentration of impurity situated at a grain boundary can be inferred from observation of a filament on an etched surface, for example the filament seen at GBF2 in Figure 5e. The X-ray signal strength emitted from this filament is of a similar magnitude to that generated by the 0.5 mol L^-1 standard frozen into a 0.4 gm hole discussed in section 2.1. If 1 μm of the filament has been stimulated to produce this signal then we expect about 1 × 10^-16 mol of impurity per micrometre of filament, the radius of the filament is 200 nm, so the concentration of hydrated salt is ~ 1mol L^-1, which is of the same order as the eutectic composition. This filament was created by sublimation through up to 600 μm of grain boundary. If the whole of the boundary were perpendicular to the direction of sublimation and covered with a monolayer of impurity (at the concentration proposed for liquid-like properties) then we would expect a coagulation of 1 × 10^-6 (m) × 600 × 10^-6 (m) × 1.7 μmol m^-2 ≈ 1 × 10^-15 mol of impurity per micrometer of filament, which is an order of magnitude more than observed in this case. It is likely that the impurity, when present, is at a lower concentration than the eutectic because of attractive interfacial forces; the whole of the grain boundary is unlikely to have been perpendicular to the surface, and not all impurity will have necessarily coagulated in the observed filament. However, the calculation is consistent with a heterogeneous distribution of impurity at grain boundaries, with much of the boundary containing no impurity.

The reasoning above relies on the assumption that impurity at a grain boundary exists in a monolayer at concentrations close to the eutectic concentration, i.e. it is behaving in a liquid-like manner. It is quite possible that this may not be the case and that the ionic species could disperse to greater dilutions at the grain boundaries through diffusion. Grain-boundary diffusivities for ions in ice are not well known. However, images indicating incomplete coverage of grain boundaries (e.g. Fig. 5b) are suggestive of a liquid-like layer covering the maximum possible boundary area allowed by the liquidus relation and interfacial forces for impurity spread at a boundary; further dilution by diffusion would cause an apparent phase change (i.e. a reduction in impurity concentration to a level where liquid-like properties are no longer exhibited), hence locking the impurity in place.

4.2. Triple junctions

Impurity was generally only detected at triple junctions in specimens where a high concentration of sulphate ions was present. Pre-etching served to concentrate acid at the triple junctions, probably by the slow loss of mass from the surface during pre-etching. It is theoretically possible that impurity collected at triple junctions by pre-etching could have originated from the cut surface; however, this is unlikely, at least for sulphate ions. Sulphate ions were not generally observed in immobile surface etching spots, suggesting that they were not present in the bulk of grains. The dimensions of filled triple junctions seemed to be related to pre-etch time and bulk concentration, and the largest triple junctions observed on pre-etched surfaces had areas of several square microns. When compared with spectra from 4.9 mol L^-1 sulphuric acid frozen into 2 μm holes, the triple junction in Figure 6e produced a signal consistent with the entire electron interaction volume containing impurity at the eutectic concentration. However, the X-ray spectra collected from other triple junctions gave smaller signals, which were consistent with the production of X-rays close to the surface only. For example, the triple junction in Figure 6d, which has a surface area of ~ 1.4 μm², produced a signal strength that could be explained by X-ray emission from impurity at the eutectic composition extending to ~ 0.4 μm depth. When triple junctions were etched, no clear vein residues were revealed (often only grain-boundary filaments and etching spots were detectable (e.g. Fig. 6b)). Furthermore, impurity was never detectable at triple junctions on freshly cut surfaces, even in ice from the acidic layers, where pre-etched images show veins with areas of several square microns. This could, however, be an artefact of the cooling wave penetrating the specimen from the surface inwards during its immersion in liquid nitrogen; cooling specimens from the base up could test this possibility. It appears highly unlikely that veins have a diameter as large as the cross-sections observed on a pre-etched surface. It is more plausible that far narrower veins are the source for the impurity, which collects in the rotting triple junctions formed during pre-etching (Reference NyeNye, 1991). Without data to constrain the rate of sublimation during pre-etching, it is difficult to estimate the vein dimensions below the surface. A similar process may well take place when filaments are formed during pre-etching. It is less likely that surface spots have a similar source since they have not been observed before etching.

Although sulphate ions were found alone in some cases (e.g. Fig. 6d), indicating the presence of sulphuric acid, it was more usual for a combination of ions to be present, particularly sulphate and sodium, suggesting that the following reaction had occurred:

(1)

allowing the H⁺ and Cl^- ions to dissolve into the lattice and leaving a sodium sulphate residue at the veins mixed with any excess acid. This is consistent with the high lattice solubility of chloride ions compared to the other species present (Reference Petrenko and WhitworthPetrenko and Whitworth, 1999). Chloride ions were not noted at triple junctions in any of the sulphate peaks. Similar reactions are likely with the other cations and would also release the chloride ions into the lattice.

A relatively high concentration of sulphate in the bulk ice appears necessary for impurity to be detectable at triple junctions. This is consistent with either saturation of the surrounding ice lattice, forcing any excess to concentrate at triple junctions, or a saturation of the surrounding grain boundaries with an impurity layer, allowing the excess to concentrate in a vein. In this study, the lowest bulk concentration of sulphate ions for which filled triple junctions were detectable was 1.6 μmol L^-1. This could represent either an upper limit at which the surrounding ice (at -53°C and this particular grain-size) is saturated, or perhaps more probably the bulk concentration at which sufficient grain boundaries are saturated to allow the collection of ice at some triple junctions.

4.3. Interface energies

The observed boundary interface angles (vapour-ice-ice in section 3.1. and ice-impurity-ice in Fig. 6) have implications for boundary interface energies and hence impurity distribution. At a steady-state interface between a grain boundary and air, the ice-vapour surface free energies are balanced by the grain-boundary free energy (Reference HobbsHobbs, 1974, p. 436–441). Thus the dihedral angle of the grain boundary groove (which must form to achieve equilibrium) is indicative of the surface energies of the system. The dihedral contact angle found here of 132±3° is significantly less than the grain-boundary contact angle of 145° quoted by Hobbs for pure ice measured at 0°C. The value seems more consistent with those quoted for unclean ice, which contained many surface irregularities, particularly at the grain boundaries, where the dihedral angles were 130–148°.

The reduction in the dihedral angle implies a reduction in the ratio of ice-vapour to grain-boundary surface free energy (γ_iv/γ_ii) (Reference HobbsHobbs, 1974, p. 436–441), from 1.67 to 1.23. This is hard to reconcile with the earlier observation of impurity preferentially located at grain boundaries between snow bonds (section 4.1), which requires γ_ii to be reduced by the introduction of impurities. It may be explained by a significant reduction in γ_iv caused by the lower observation temperature (-20°C) and/or the addition of an impurity layer to the surface. Relevant interfacial energy information is not currently available to test this.

Figure 6e appears to show no sign of impurity in the grain boundaries surrounding the triple junction, although the salty composition (the following impurities were detected: Na, S, K and Ca) may indicate a solid precipitate. Etching the junction suggested the presence of trace impurity at the grain boundaries. Figure 6c shows sodium and chlorine extending a short distance from the grain boundary into the triple junction. Figure 6f, however, shows that the dihedral angle at the triple junction is zero or very close to zero. It is not clear whether this also characterizes the dihedral angle of the vein below the surface, as the vein cross-section is expected to widen like a funnel towards the surface (see section 4.2.).

At temperatures close to 0°C, liquid was not present at grain boundaries in studies of laboratory ice where dihedral angles around 32° were observed (Reference MaderMader, 1992), implying 1.9γ_i1 = γ_i1, where γ_i1 is the ice-liquid interface energy. The concentration of solute in the veins was low enough to have a negligible impact on γ_i1. At the temperatures typical in polar ice (e.g. -53°C at Dome C), the higher intergranular solute concentrations could be sufficient to reduce γ_i1 such that 2γ_i1 ≤ γ_i1, allowing distribution of a layer of solute at grain boundaries to be energetically favourable. This is also borne out by the widespread observation of filaments at grain boundaries, both in other work (e.g. Reference Cullen and BakerCullen and Baker, 2000) and here. Under our proposed scenario, if a liquid-like layer of solute is situated at a grain boundary, any excess will locate at the triple junction because of curvature effects. Different ionic species will have varying effects on γ_i1 or γ_i1, which might account for the apparent preference of salts for grain boundaries over veins. Note that at Dome C NaCl is present below its eutectic temperature and is therefore unlikely to behave like a liquid; however, it may still be able to reduce γ_i1, making its location at grain boundaries preferable to inclusion in the lattice. If it does not behave like a liquid, NaCl will have no significant energetic preference for triple junctions. Its significant presence at triple junctions in acidic ice (Fig. 6e and f) could be explained by diffusion from grain boundaries to the vein to fuel the reaction given by Equation (1).

The SEM techniques applied in this work could be used to test the impact of impurities on boundary interface energies in laboratory-grown ice prepared at different temperatures, presuming that sufficient time was available for the specimens to reach equilibrium.

4.4. Lattice

Immobile spots of impurities that have coagulated on the surface during etching appear to originate directly from impurity previously dissolved or included in the lattice (except in the relatively unlikely case where a grain-boundary layer is deposited on the surface). The most commonly detected impurity in these spots was chlorine, particularly in the specimens from the acidic sulphate layers. It is likely that the chlorine, when detected alone, was originally deposited in a salty form, which has subsequently been liberated by the reaction given by Equation (1).

As an example, we estimate the quantity of chloride dissolved in the bulk for Dome C ice at 218.95 m (Fig. 7a). Chloride is just above the detection limit in the surface etch spots after 10 min at -70°C, by which time approximately 200 μm of ice had been sublimed from the surface. The number of surface spots per unit of etched volume is 1.4 × 10¹³ m^-3, so assuming that a spot contains approximately 3 × 10^-17 mol of chloride (section 2.1), the bulk concentration is ~ 0.4 μmol L^-1; we estimate an uncertainty of at least 50% on this calculation. However, it is consistent with the majority of the chloride in the ice, which has a bulk concentration of 0.4 μmol L^-1, being dissolved in the lattice; it is considerably below the lattice solubility limit of ~ 600 μmol L^-1 for HCl at -53.5°C found by Thibert and Dominé (1997).

In other cases where impurity was detectable, a combination of sodium and chlorine was often noted, although generally in larger, less regularly spaced spots than those containing chlorine alone, perhaps indicating impurity inclusions disrupting the lattice. Non-acidic chloride ions could only be substituted for H₂O molecules if an associated cation was also included interstitially within the ice molecular cage to prevent charge separation. The amount dissolved is strongly dependent on the equilibrium partition co-efficient, or solubility limit of the lattice. The possible dislocations observed in Figure 7b could be the result of an amalgamation of inclusions formed to reduce the lattice free energy.

4.5. Arrangement summary

Our best assessment of the arrangement of impurities is as follows:

4.6. Comparison with previous work

Some of the observed differences in the distribution previously noted are linked to small differences in the technique of the different groups. In this work we have noted all the features observed in the previous studies as well as new variations of them. The features of impurity filaments and inclusions observed in studies on Greenland (GISP2) and Antarctic (Byrd core) ice (Reference Cullen and BakerCullen and Baker, 2000, 2001; Reference Baker and CullenBaker and Cullen, 2002) are seen in many specimens here when prepared by a similar method: pre-etching followed by etching. Our conclusion that these studies could have had difficulty locating triple junctions accurately on the specimen surfaces makes their findings regarding veins doubtful except in cases where filaments are joined at triple junctions. The sulphate-filled veins observed in acidic layers of H₂SO₄ seen in this work are consistent with studies of acidic Antarctic ice from Dolleman Island (Reference Mulvaney, Wolff and OatesMulvaney and others, 1988; Reference Wolff, Mulvaney and OatesWolff and others, 1988). However, the absence of detectable impurity in veins from less acidic ice casts some doubt on the concept of a fully connected network of impurity-filled veins in the ice sheet. Without the presence of impurities, vein diameters are negligibly small (~10^-10 m) except in the warmer regions of the ice sheet, such as near the base. A more likely model includes layers of connected veins at depths where sulphate ions are already covering the majority of neighbouring grain boundaries at least in a monolayer. Triple junctions at other less acidic depths would contain little more impurity than the surrounding grain boundaries.

The study of Antarctic ice by Reference Fukazawa, Sugiyama, Mae, Narita and HondohFukazawa and others (1998) found acidic SO₄ ^2-, HSO₄ ^- and NO₃ ^- ions present in veins at -20°C using Raman spectroscopy. They did not detect these species at grain boundaries, possibly because of the detection limit of the technique. The ice from both sites contained high concentrations of sulphuric acid (3.42 μmol L^-1 at Nansen and 14.35 μmol L^-1 at South Yamato), which is consistent with the results seen here. In the case of the South Yamato ice, where only 3% of the acid was estimated to reside at triple junctions, a vein dihedral angle close to zero (as observed here) might allow a significant volume of this acid to spread into the grain boundaries, in addition to a monolayer. They also observed acidic nitrate ions in veins, but these cannot be detected by the technique used here. The chloride ions apparently dissolved in the lattice are below the solubility limit found by Thibert in laboratory- grown ice (Thibert and Dominé, 1997).

4.7. Implications

The heterogeneous distribution of impurity observed in specimens of low impurity content is interpreted to imply an absence of connected impurity-filled pathways either at grain boundaries or at veins. At higher bulk concentrations the greater proportion of grain boundaries coated by impurity will increase the connectivity between grain boundaries displaying liquid-like properties due to impurity coating. If grain boundaries are covered or saturated with at least a monolayer of impurity locally then there will be increasing connection between filled veins. Hence a vertically connected network of impurity-filled veins is extremely unlikely in areas of an ice sheet with low temperature and background impurity levels such as Dome C. However, horizontally connected networks of impurity-filled vein remain quite feasible within sulphuric acid layers. The likelihood of connection in either veins or grain boundaries will be dependent on impurity content, grain-size and temperature. Impurity-coating the surfaces of bubbles should link unconnected veins within acidic layers.

The theory of electrical conduction through sulphuric- acid-filled veins (Reference Wolff and ParenWolff and Paren, 1984) remains applicable in cases where bulk concentration is high enough to saturate the surrounding grain boundaries and hence fill the veins. At lower bulk concentrations, another conductivity mechanism is required, perhaps by the mobilization of protonic defects from the heterogeneous patches of impurity at grain boundaries. The ionization defects are then free to travel either along grain boundaries or through the lattice. In regions where NaCl reacts with H₂SO₄ at triple junctions (Equation (1)), the release of H⁺ and Cl^- ions into the lattice should allow conduction via ionization defects in the lattice (Reference Petrenko and WhitworthPetrenko and Whitworth, 1999) to replace the liquid conduction lost from the vein.

This explanation of conduction is in good agreement with the work of Fujita (Reference FujitaFujita and others, 2002) who noticed a step increase in the conductivity of Dome F ice from sulphuric acid layers (10-20 μmol L^-1) as the temperature was increased above -81 °C. They explained this by the addition of liquid conduction to the total conductivity as the frozen aqueous mixture melted in the veins at around its eutectic temperature. When the experiment was repeated on ice with background impurity levels, the step change in conductivity was not seen, implying the absence of significant quantities of soluble impurity in veins.

Dust particles have been located at grain boundaries, in some cases visibly pinning them (Reference Barnes, Mader, Rothlisberger, Udisti and WolffBarnes and others, 2002b). The widespread observation of soluble impurity species at grain boundaries in this work suggests that the post-depositional production of CO₂ suggested by Reference Tschumi and StaufferTschumi and Stauffer (2000) is feasible, although determining the rate will be complicated by factors such as the grain-boundary diffusivity, since this will dictate the pace of acid reaching the dust.

Water molecules confined to sub-nanometre spaces such as those at grain boundaries are thought to adopt a structured arrangement (e.g. Reference Gallo, Rovere and SpohrGallo and others, 2000; Reference PollackPollack, 2001); this could reduce the diffusivity of aqueous ionic impurities confined at the boundaries. However, Reference Raviv, Klein and LauratRaviv and others (2001) found the viscosity of highly purified water within sub-nanometre films to be close to that of the bulk liquid, suggesting that the diffusivity of water is similarly unaffected by confinement. They emphasize, however, that the addition of hydrated ions to the layer could alter the behaviour of the film. If the diffusivities of aqueous ions are significantly reduced by confinement at grain boundaries, the shift of the chemical profile by the mechanism described by Rempel and others (2001b) would be slowed. Liquid displaying bulk properties would only be present in veins contained by regions of relatively high bulk concentration. Diffusion of solute under a temperature gradient would be limited by the diffusion of solute from veins into unfilled grain boundary. This could provide an alternative explanation for movement of solute noted in the upper layers of the Dome C ice core (Reference Barnes, Mader, Rothlisberger, Udisti and WolffBarnes and others, 2002a).