Abstract– Particles from comet 81P/Wild 2 were captured with silica aerogel during the flyby Stardust mission. A significant part of the collection was damaged during the impact at hypervelocity in the aerogel. In this study, we conducted impact experiments into aerogel of olivine and pyroxene powder using a light-gas gun in similar conditions as that of the comet Wild 2 particles collection. The shot samples were investigated using transmission electron microscopy to characterize their microstructure. Both olivine and pyroxene samples show evidence of thermal alteration due to friction with the aerogel. All the grains have rounded edges after collection, whereas their shape was angular in the initial shot powder set. This is probably associated with mass loss of particles. The rims of the grains are clearly melted and mixed with aerogel. The core of olivine grains is fairly well preserved, but some grains contain dislocations in glide configuration. We interpret these dislocations as generated by the thermal stresses that have emerged due to the high temperature gradients between the core and the rim of the grains. Most of the pyroxene grains have been fully melted. Their high silica concentration reflects a strong impregnation with melted aerogel. The preferential melting of pyroxene compared with olivine is due to a difference in melting temperatures of 300°. This melting point difference probably induces a bias in the measurements of the ratio olivine/pyroxene in the Wild 2 comet. The proportion of pyroxene was probably higher on Wild 2 than expected from the samples collected into aerogel.
Grains from the comet 81P/Wild 2 were trapped in silica aerogel cells by the Stardust spacecraft (Brownlee et al. 2006). The particles impacted the collector with a relative velocity close to 6.1 km.s−1 (Hörz et al. 2006). The deceleration of the particles in the aerogel took place over typically a one centimeter length for incident grain sizes around 50 μm (see Burchell et al. 2008 for details). During the deceleration, the temperature conditions were extreme. Most of the grains have experienced various degrees of modification induced by the capture, including fragmentation, mass loss due to friction with the aerogel, heating, and mixing with melted aerogel (e.g., Zolensky et al. 2006; Leroux et al. 2008a).
In the Stardust aerogel, the shape of the tracks, as well as the dispersion of individual components along the track, suggests that most Wild 2 dust were either (1) well consolidated materials with some weakly bound fine components which gave type A, carrot-shaped tracks, or (2) loose aggregates with some large subcomponents which gave type B, tracks with an initial bulbous cavity and some emerging finer styli roughly aligned with the impact direction (Hörz et al. 2006; Zolensky et al. 2006; Burchell et al. 2008). In both cases, the larger components have penetrated more deeply into the aerogel, and are found as terminal particles. These terminal particles are typically coarse-grained crystalline material (over 1 μm) and are found relatively well preserved from thermal capture modification. They include CAI-like particles and chondrule-like objects (Zolensky et al. 2006; Leroux et al. 2008b; Nakamura et al. 2008; Simon et al. 2008; Chi et al. 2009; Jacob et al. 2009). A dense glassy rim is frequently observed around a crystalline core (Zolensky et al. 2006). The core is generally well preserved, permitting reliable microstructural information to be obtained. Nevertheless, some grains contain dislocations (Schmitz and Brenker 2008; Tomeoka et al. 2008; Jacob et al. 2009) and it remains unclear whether these dislocations result from a shock event prior to the collection or generated during the collection itself.
Although the large and coarse-grained terminal particles survived the aerogel capture relatively well, the fine-grained material found in the walls of tracks has been severely modified. The material encased in the walls shows evidence for strong thermal modifications in addition to intermixing with melted aerogel (Leroux et al. 2008a); this renders the understanding of the origin of some significant proportion of cometary grains difficult. In summary, because of the extreme conditions due to the collection, the microstructure of the grains can be modified, leading to possible misinterpretation if the capture effects are not well understood.
Tsou et al. (1988) and Zolensky et al. (1989) reported the first use of silica aerogel to collect small objects moving at high velocity (see Burchell et al.  for a review of the use of aerogel as a dust collector in space). Aerogel silica was chosen for its transparency, low density, composition simplicity, and well-controlled manufacturing. Zolensky et al. (1989) shot various projectiles (olivine, pyrrhotite, beads of silicate glass, and polymers) ranging in size from one to several hundred micrometers. Some experiments were also performed with aggregates of minerals such as olivine, pyroxene, spinels, graphite, pyrrhotite, calcite, smectites, and serpentine assembled in an epoxy resin to form spheres of about a millimeter in diameter. Melting of material was occasionally observed, associated with a loss of volatile elements. It was concluded that for these relatively large grains, capture in silica aerogel enabled a relatively good preservation of the grains. Subsequently, Bunch et al. (1991) noticed a reduction of the grain size during their aerogel capture experiments. This was attributed to the projectile acceleration resulting in collisions between projectiles during the flight time to the target. They also reported the presence of a ring of dense material around the grains, attributed to the melting of the particle and the surrounding aerogel. Soon after, Barrett et al. (1992) showed that a fraction of the projectile was deposited along the track; this gives an explanation for the observed mass loss during the collection. They also showed that some mineral phases experienced significant changes during the impact, depending on the nature of phases, the density of aerogel, and the impact velocity. During the preparation of the Stardust mission, more detailed studies of the collected particles were performed, including transmission electron microscopy (TEM) investigations. Zolensky et al. (1994) did experiments with olivine, enstatite, and pyrrhotite. They observed that olivine was one of the most resistant materials. Nevertheless, some grain rims showed strong interaction with aerogel. The authors also reported volatilization of sulfur during impacts of pyrrhotite grains. More recently, Hörz et al. (1998) and Burchell et al. (1999a, 2001) confirmed that the mass reduction during the impact can be up to half of the initial mass. Burchell et al. (1999a) showed that olivine of 100 μm in diameter captured at 5.3 km.s−1 in an aerogel density of 92.5 kg.m−3 kept 60–80% of their initial mass in the terminal grain, decreasing to 20% for olivine of about 300 μm in diameter. They then suggested that the collection in aerogel was more efficient for smaller particles. The research that followed focused on determining the range of temperature and pressure suffered by the particles. Noguchi et al. (2007) conducted experimental shots using targets of aerogel with a density up to 30 kg.m−3 with hydrated minerals as projectiles. The microstructure of projectiles captured at 6.18 km.s−1 indicated that the temperature exceeded at least 500 °C for about a microsecond. They observed the formation of a vesiculated glass rim attached by silica with a thickness of about 0.5–1 μm. Burchell et al. (2009a) showed that stainless steel projectiles of about 500 μm in diameter captured in an 30 kg.m−3 density aerogel at a speed of 5 km.s−1, revealed partial melting indicating that the samples were heated locally at least at 1400 °C. Hörz et al. (2009) have also made shots in aerogel (density 20 kg.m−3) of 60 μm alumina beads (Al2O3) with a speed of 6 km.s−1. The ablation and fusion of the samples indicated a peak temperatures above 2054 °C with approximately 25% of mass lost. These very high temperatures do not affect the entire volume of the grains, but only a peripheral zone.
However, very few studies have been undertaken to understand the possible microstructure modifications induced by the collection at high speed in the aerogel. The present work concerns transmission electron microscopy observations of experimentally fired samples. It is aimed at gaining an understanding of the heating processes that occurred during the capture of the Wild 2 grains and of the subsequent microstructure modifications. The final goal is a better knowledge of the temperature history experienced by the particles during capture and of the related induced mechanical constraints. This is to determine whether the microstructure observed in the Stardust samples is probably primordial (i.e., originated precollection) or not.
Samples and Experimental Procedures
Olivine and pyroxene were chosen because they are the major mineral components of Wild 2 (Zolensky et al. 2006). Olivine (from San Carlos, Fo91) and pyroxene (from an orthopyroxenite of unknown origin, En93) single crystals from a terrestrial origin were selected because they are almost free of crystal defects (such as dislocations, planar defects, or subgrains boundaries). The initial microstructure was checked with samples prepared for the TEM using conventional ion-milling techniques (5 keV, 1 mA at 15°). The final grain size of the crushed samples was less than 100 μm for both mineral phases (Fig. 1). The powder was divided into two parts; one was kept to investigate the consequences of crushing and slicing by ultramicrotomy and the other was impacted in a density gradient silica aerogel similar to that flown on the Stardust mission. Experimental shots were performed at the University of Kent using a two-stage light-gas gun (Burchell et al. 1999b). The particle velocity was about 6.05 km.s−1 for enstatite and 6.49 km.s−1 for olivine. These are mean values in each shot, with the speed of each individual grain in the shot within typically ±4% of the mean value. The impact generated deceleration tracks were comparable to that observed in the Stardust aerogel (Fig. 2).
The TEM preparation of the fired samples was done using a compression method. First, a bloc of aerogel containing the terminal particle was isolated using a razor blade. The bloc was then compressed between two glass slides. The prepared samples were embedded in an epoxy resin “Embed 812” (see Zolensky et al.  for details). A set of powder that was not shot was also prepared using this resin. After curing of the epoxy, the specimens were sectioned by ultramicrotomy using the same conditions as used for the Stardust sample preparation (Zolensky et al. 2008). The thickness of the slices varied from 80 to 120 nm. The samples were deposited on a standard formvar/carbon coated copper grid.
The samples were studied by analytical TEM using a Tecnai G2-20 twin (LaB6 filament, 200 kV) and a Philips CM30 (LaB6 filament, 300 kV) both equipped with an energy dispersive X-ray spectrometer (EDS). Grain microstructure was studied using bright- and dark-field imaging in conventional TEM mode and using bright and annular dark-field detectors in scanning mode (STEM). The two microscopes are equipped with a precession module from the Nanomegas Company. Chemical compositions were measured using EDS. Calculations of element concentrations were carried out using calibrated k-factors and thin film matrix correction procedures. The k-factors for the major elements were determined using standard specimens according to the parameter-less method of Van Cappellen (1990). Elemental distributions were obtained by EDS X-ray intensity maps, using spectral imaging wherein each pixel of a spectrum image contains a full EDS spectrum (see Leroux et al. [2008a] for details). The absorption correction was based on electro-neutrality of minerals or glasses (Van Cappellen and Doukhan 1994).
The samples were studied before crushing (preparation by ion-milling) and after crushing (preparation by ultramicrotomy) to have a reference for chemical composition and microstructure. Concerning the microstructure, both the olivine and pyroxene were originally almost free of crystal defects. Some dislocations were detected with a density of less than one defect per 25 μm2. Some clinoenstatite lamellae were observed in the pyroxene orthoenstatite matrix, with a modal abundance of less than 5%.
Ultramicrotomy sectioning significantly changes the general aspect of the grains. We observed the typical “shard-like” aspect for both olivine and pyroxene (formation described by Zolensky et al. 2008). Locally, the pyroxene grains exhibit a lamellar microstructure that appears damaged compared with that of the surrounding shards. Diffraction patterns (not shown) reveal that the lamellar laths are roughly parallel to common (100) planes, but are misoriented by a few degrees with respect to one another. This kind of microstructure has already been reported by Jacob et al. (2009) and has been properly attributed to the ultramicrotomy preparation procedure.
The deceleration tracks of olivine grains have type A shapes (carrot like, Fig. 2a). Three terminal particles obtained from three different tracks were studied. Figure 3 shows the largest slice of each particle after the ultramicrotomy sectioning. The sizes range from approximately 2 μm to 15 μm. The particles impacted in the aerogel have rounded corners, whereas their shape was angular before impact, after the grinding (Fig. 1a). Their shape thus reveals ablation during the interaction with the aerogel. The whole samples are made up of a crystal core surrounded by one rim.
The rim surrounding the biggest particle (diameter >10 μm, Fig. 3b) is made up of two distinguished parts (Fig. 4a). The first one is a mixture of amorphous and crystalline material, whereas the second one is fully amorphous. Within the first part, the orientation of the crystalline part is identical to that of the adjacent inner crystal (Fig. 5). The average composition of this area is 67 mole% of SiO2, 31 mole% of MgO, and 2 mole% of FeO.
The interfaces between these three parts are very sharp. The interface between the crystal core and the first part of the rim (amorphous + crystal) is named “interface #1,” and the one between the first and the second part of the rim (full amorphous material) is named “interface #2.” A concentration profile across the two interfaces is shown in Fig. 4b. The interface #1 displays a strong peak of SiO2 combined with a low content of iron and magnesium. This composition is associated with beads (diameter about 20 nm) as observed in bright-field on the TEM images (Fig. 5). The ratio Mg/(Mg+Fe) at this interface is about 0.81–0.85. The opposite interface (interface #2) shows a decrease of SiO2 content, whereas the MgO and FeO contents locally increase. The ratio Mg/(Mg+Fe) at this second interface is about 0.58–0.61. The average compositions of the different parts of the rim are given in Table 1. They reveal an amount of impregnation of silica by about 70% in the mixed crystal/amorphous area, and about 95% within the amorphous area.
Table 1. Representative EDS composition in wt% of the samples characterized in this study.
Crystalline and amorphous coexisting area
Looking at the diffraction contrast on the bright-field TEM images of the interfaces (Fig. 4a), we also notice a darker contrast of the amorphous + crystal area at interface #2 than at interface #1. This reveals a gradient of the crystalline ratio. The two other olivine particles have a diameter less than 5 μm (Fig. 3a and 3c). They are composed of a fully crystalline core surrounded by a fully amorphous rim without mixed crystal/amorphous area.
We show samples of impacted olivine that contain dislocations in Fig. 6. They are in contrast with the diffraction vector g = 004 and out of contrast with g = 400, which suggests a Burgers vector c = . The orientation of the lines is compatible with this c direction, and thus a screw character is deduced for the dislocations. The dislocations have the same characteristics as that reported for Wild 2 samples (Jacob et al. 2009). They were not observed within the original (non-shot) samples. One grain contains a series of curved dislocations with Burgers vector c (Fig. 7). The curvature indicates the proximity of a dislocation source. Measurement of the length of the line enables us to deduce the order of magnitude of the stress, considering the equilibrium between the applied stress and the line tension of the dislocation considered as part of a circular loop. We used the basic flow stress equation for dislocations τ∼ 2 μb l−1 (e.g., Hirth and Lothe 1982), where μ is the shear modulus of forsterite (70 GPa; Durinck 2005), b is the Burgers vector (0.599 nm for c dislocations), and l is the source length (here approximately 60 nm). The deduced stress for the formation of the dislocation loops is around 1.5 GPa.
The deceleration tracks generated by the deceleration into aerogel of pyroxene grains have mainly type B shapes (bulb shape with one or various terminal particles) and some have type A. A typical type B shape track is shown in Fig. 2b. Particles obtained from three different tracks have been characterized (Fig. 8). The grain size ranges from less than 1 μm to approximately 12 μm. The largest particle has a preserved crystalline part (Fig. 8b), whereas the others are fully amorphous. The compression method used for the preparation of the samples allows seeing the material in interaction with its capture medium. Figure 9 shows that a number of grains have been dispersed in the aerogel. This phenomenon was not observed for the olivine samples. The fully amorphous material displays some vesicles, and is quite comparable to that of the thermally modified particles observed in the Stardust samples (Leroux et al. 2008a, 2009). However, no metal beads are found in the samples, and composition maps show that iron is still present in the amorphous matrix, probably as FeO (Fig. 10). Given the known initial composition, measuring the average composition of the amorphous grains allows us to infer the degree of impregnation of aerogel. It extends from 76% to 62%, with an average at 67% of SiO2. This is compatible with the measurements conducted for the Stardust samples (Stodolna et al. 2009). Representative compositions of the amorphous material are given in the Table 1. Concerning the crystalline component, the single grain used was too thick to be properly characterized using TEM. Nevertheless, the ortho-clino lamellae structure is still present with a ortho/clino modal abundance <10%, and dislocations were not observed. Furthermore, the interfaces between the crystalline and the amorphous parts were sharp and did not present any intermediate mixed area as observed for olivine.
Olivine and Pyroxene Behavior During Deceleration in Aerogel
Deceleration tracks in aerogel clearly display different behaviors depending on the nature of the mineral impacted (olivine or pyroxene). As shown by Fig. 2, tracks generated by olivine are type A (carrot-shaped), whereas enstatite generates type B tracks (tracks with bulbs). In addition, several styli frequently emerge from the bulb cavity for pyroxene, whereas for olivine, only one terminal particle is observed. This difference in behavior is probably due to a difference in mechanical behavior. Indeed, the pyroxene is a mineral that cleaves easily, whereas the olivine does not. Some models (e.g., Trigo-Rodríguez et al. 2008) then suggest that the fragmented particle starts to shed material in a lateral direction due to build up of a bow shock in front of the incident particle; the fragmentation of the pyroxene grains during the impact in aerogel thus causes an enlargement of the tracks by lateral dispersion of the fragments, leading to the formation of type B tracks.
At the scale of the grains, those that were used for these experimental shots initially had an irregular shape associated with the powder milling. However, within the aerogel, the studied grains (especially olivine) have rounded edges as number of particles collected during the Stardust mission. The rounding results from abrasion and/or ablation during the deceleration stage and friction with the aerogel (Burchell et al. 2006, 2009a; Hörz et al. 2009). A large number of previous studies have already reported mass loss for high velocity experiments (Bunch et al. 1991; Barrett et al. 1992; Hörz et al. 1998, 2009; Burchell et al. 1999a, 2001). Burchell et al. (2009b) observed a significant mass loss of material on experiments with organic microparticles collected into aerogel for impact speeds above 2 km.s−1. Based on the degree of mass loss, they estimate that at least 35% of the incident kinetic energy was coupled back into the particles as heat during their aerogel capture. In our experiments, thermal ablation is clearly illustrated by the presence of an amorphous rim around the particles, showing that the surface temperature exceeded the melting temperature of the grains. The high amount of rounded fragments observed for the pyroxene experiment suggests that main primary particles have lost droplets of liquid during the deceleration.
A major result of our study concerns the observed differences between olivine and pyroxene grains. While olivine is fairly well preserved, pyroxene is strongly damaged by the collection process. Indeed, most of the pyroxene particles were melted and mixed with aerogel. This shows that the temperature reached during the deceleration (including that in the core of the incident grains) exceeded the melting temperature of pyroxene (approximately 1550 °C). The melting temperature of olivine is 300° higher (approximately 1850 °C), and this appears sufficient for a better preservation of the particle core in its original state. The thickness of the rims is about 250 nm for olivine, suggesting that initial particles with a size lower than 0.5 μm in diameter would not survive the thermal event associated with the impact. For pyroxene particles, those with a diameter up to 4 μm in diameter are completely melted. Only the biggest particles were partly preserved, which were surrounded by a thick silica-rich amorphous rim (approximately 2 μm on Fig. 8b). Furthermore, as pyroxene is less mechanically resistant than olivine, it has been probably more finely fragmented than olivine and thus more melted and mixed with aerogel during the high-temperature stage of collection in the aerogel.
If we consider the implications of our work for the Stardust samples, then our results show that the melting temperature difference between olivine and pyroxene probably induced a bias in the olivine/pyroxene ratio within the Stardust samples. The measurements from Zolensky et al. (2006, 2008) concerning the Stardust terminal particles show a predominance of olivine. On the contrary, Leroux et al. (2010) measured an olivine/pyroxene ratio close to one in craters from aluminum foils. This is consistent with our study that shows that the olivine/pyroxene ratio is likely to be overestimated in the aerogel Stardust samples due to the difference in melting temperature of these two minerals.
Formation of the Mixed Crystalline/Amorphous Area in Olivine Samples
The olivine samples show some interfaces between the crystal and the silica-rich amorphous phase including a rim composed of a mixture of crystal and glass (Figs. 4 and 5). Selected area electron diffraction patterns show that within this mixed crystalline/amorphous area, the crystalline part is monocrystalline and has the same crystallographic orientation as the neighboring crystalline core (Fig. 5). We interpret this area as a consequence of the incongruent melting of olivine. The surface of the grain was brought at high temperature into the regime of coexistence of liquid and solid. The compositions at the two interfaces (#1 and #2) permit an estimation of the peak temperature reached at the surface of the grain based on the binary diagram of olivine (Fig. 11). The liquid is enriched in iron and the crystal is enriched in magnesium (Mg/(Mg+Fe), approximately 0.81–0.85 at interface #1 and Mg/(Mg+Fe) approximately 0.58–0.61 at interface #2), which is consistent with the olivine binary phase diagram. We expect that the equilibrium was reached because the compositions at the two interfaces correspond to the same temperature of about 1750 °C. Furthermore, the diffusion coefficient of cations is about 10−10 m2.s−1 in a silica-rich liquid (Roskosz et al. (2008) and references therein). The high temperature duration is estimated about 0.1 ms (Roskosz et al. ), giving a mean diffusion distance of about 500 nm. The distance between the two interfaces is lower than this value, which confirms that thermodynamical equilibrium could have been reached. Note that for this interpretation we use the 1 atm pressure binary diagram for the olivine, even though the local vapor pressure around the particles is not yet established.
To explain the formation of this area, we propose the following sequence. A rim of olivine is melted and mixed with the surrounded aerogel. The liquid is thus enriched in silica. Considering that the subsequent cooling comes from the outside of the grain, the forsterite is the first phase to crystallize. Silica glass is formed simultaneously to preserve the balance of elements. The remaining liquid then contains relatively more and more silica. At the end of the crystallization, the remaining silica forms droplets at interface #1 to minimize surface energy (Fig. 5). This sequence could also explain the gradient of crystallinity observed in Fig. 4a, higher outside the rim. The newly formed crystal orientation is about the same as the initial crystalline forsterite, because it has probably not completely melted, and the crystallization of forsterite occurred preferentially on remaining nuclei, thus preserving the initial crystallographic orientation.
The Amorphous Material
The observed enstatite particles are largely dominated by amorphous material, showing that the grains did not survive unaltered under the extreme condition of the collection. The amorphous material is strongly enriched in silica (typically 70–80%). This clearly indicates melting of the particles and mixing with melted aerogel. This configuration is frequently observed in the Stardust samples (e.g., Leroux et al. 2008a). In the Wild 2 samples, the silica-rich glassy matrix embeds a large number of small Fe-Ni-S inclusions. The size of the Fe-Ni-S beads extends from a few nanometers to one hundred nanometers in rare cases. Leroux et al. (2009) suggested that a significant part of the Fe-droplets were formed by reduction of melted ferromagnesian silicates by C-bearing material. The amorphous areas within our experimentally shot samples do not contain any metallic beads. The FeO originally present in the enstatite has not been reduced, despite the aerogel used in these experiments containing carbon at a level similar to that used in the Stardust collection, estimated to ≈2.4 atom% by Gallien et al. (2008). This means that a reduction process to account for the presence of numerous Fe-Ni-S beads in the melted Stardust samples, as suggested by Leroux et al. (2009), would require a contribution of additional carbonaceous materials such as organics, or that these beads originated solely from thermally decomposed iron-sulfides, scattered within molten aerogel as Fe-S beads, as proposed by Ishii et al. (2008).
The Crystalline Material
In addition to the thermal effects, indications of mechanical deformation have been identified in olivine through the formation of dislocations. They are dominantly screw dislocations, oriented along the  direction and curved at the edge, suggesting dislocation source activity. This configuration is the typical signature of shocked olivine (Leroux 2001). Similar dislocations were observed in Stardust terminal particles (Schmitz and Brenker 2008; Tomeoka et al. 2008; Jacob et al. 2009). Their occurrence in the experimentally fired olivine suggests that the dislocations observed in Wild 2 olivine grains could have been induced by the hypervelocity impact. The dislocation curvature radius is associated with a shear stress value around 1.5 GPa. Trigo-Rodríguez et al. (2008) estimated that the peak pressure associated with the deceleration into the aerogel could have reached a mean maximum value of 300 MPa (with an absolute peak value estimated at 900 MPa); this is quite low compared with that required to explain the dislocation microstructure observed here. However, stress concentration probably occurred during the capture. Another cause for dislocation formation could be the thermal stresses associated with the high temperature gradients at grain surfaces during the rapid heating and cooling of the particles.
We performed a TEM mineralogical study using olivine and pyroxene powder captured in silica aerogel under conditions similar to the Stardust flyby mission. The samples display clear evidence of thermal effects. The corners of the grains have been rounded due to thermal abrasion, probably associated with mass loss. For olivine, the surface temperature reached at least 1750 °C as revealed by a rim, 250 nm thick, which consists of a mixture of crystalline and amorphous material. The core of the olivine grains appears to be well-preserved, but dislocations in glide configuration were identified, suggesting that thermal stresses reached about 1.5 GPa. For pyroxene, most samples have been fully melted. The composition of the impact melt is strongly enriched in silica showing that impregnation with melted aerogel occurred. The preferential fusion of pyroxene compared with olivine is due to a melting temperature 300° lower. This difference in melting temperature probably induces a bias in the measurements of the ratio olivine/pyroxene in the Wild 2 comet. The proportion of pyroxene was probably higher on Wild 2 than expected via the Stardust sample collected into aerogel.
Acknowledgments–– We thank J. F. Dhenin and A. Addad for their continuous assistance with the TEM facility. We also thank D. Joswiak and T. Zega for the thorough and constructive reviews which improved the quality of the manuscript, as well as H. Ishii (associate editor) for helpful comments and her editorial management. We appreciate support from Centre National d’Etudes Spatiales (CNES) and the electron microscope facility by European FEDER and region Nord-Pas-de-Calais. M. J. Burchell’s research is funded by STFC (UK).