A SEM map has been established with a Jeol JSM 840 A (Paris-6 University) in order to select melt inclusions in several type-II chondrules in the Semarkona chondrite (LL3). Fig. 1 shows an example of a SEM image from a selected chondrule. Major elements of bulk chondrules and of all phases have been analysed with an electron microprobe CAMECA SX100, Paris-6 University (15 keV, 4 nA). A San Carlos olivine and synthetic glasses (BHVO-2) were analysed for the major elements with the electron microprobe. These analyses were used to calibrate the NanoSIMS elemental yields.

Melt inclusions are very small (diameter less than 15 μm), so that for an analysis of concentration gradients a nanometric spatial resolution is needed at the mineral/MI interface. The primary beam has a 150-nm diameter with O⁻ and sputters a depth around 20 nm. Analyses were made with a 1-pA intensity beam, with a mass resolution around 6000. At such a mass resolution, isobaric interferences are negligible. The major isotope species of Mg, Si, Ca, Cr, and Fe were analysed. Examples of NanoSIMS profiles obtained are shown in Fig. 2. The elemental counting rate (expressed in counts/pixels through the electron multipliers) can be affected by the matrix effect, because different elements have different capabilities to be ionised, depending on their sputtered host phase. Such so-called ‘matrix effects’ were calibrated by using electron microprobe analysis on the San Carlos olivine and BHVO-2 glass.

The analyzed chondrules are type-II chondrules, which are the most common chondrule type in ordinary chondrites (around 80% of the falls). Type-II chondrules are selected because they are supposed to form in a closed system, whereas type-I chondrules show evidences for the formation of an open system. We also selected the type-II chondrules of the Semarkona chondrite (LL3), which have olivine porphyric textures. These chondrules have apparent diameters between 600 and 1600 μm. The olivine crystals’ sizes range from 30–50 to 300–500 μm. Melt inclusions have spherical to ovoid shapes varying between 1 to 15 μm in diameter. All MI analysed are glassy phases free from either mechanically entrapped or daughter minerals and apparently not devitrified (confirmed by Raman microspectroscopy). Depending on the cut path, on the polished section, they may show a bubble (near 10–15% total MI volume) that can be considered as a shrinkage one formed by volume change of the glass during cooling. These petrographic observations are generally considered as indications of their primary origins and fast cooling [9]. An example of a type-II chondrule from the Semarkona meteorite is given in Fig. 1.

Tables 1–3 display the results of microprobe analyses of bulk chondrules, of olivines (that were analysed close to MIs), and of melt inclusions, respectively. The composition of those melt inclusions could not crystallize olivine, and hence they are not direct samples of the original melt. Instead, they can be a sample of a residual glass, affected by the crystallization of olivine. To test this idea, we made a simple calculation of the original melt: we graphically estimated the olivine volume percentage (60% for chondrule 15), and calculated a composition made of 60% of olivine and 40% of melt inclusion. The result is shown in Table 4 and is very similar to the bulk chondrule composition. This pleads in favour of a residual melt captured during olivine growth.

NanoSIMS profiles between the host crystals and melt inclusions do not show a sharp boundary layer, but a smooth transition zone (1 to 2 μm wide). The concentrations profiles smoothly expand between two plateaus standing for the olivine and the glass composition (Fig. 2). Note that a few profiles (like in Fig. 2d) have more complex transition zones with bumps and valleys for some elements at the edges of those transition zones – hereafter referred to as profiles B in contrast with simpler ones, as profiles A. The width of the transition zone is more or less the same for ²⁸Si, ²⁴Mg and ⁵⁶Fe (1 to 2 μm), but it is always different for ⁴⁰Ca (1.5 to 2.5 μm). The shapes of these profiles can be attributed to three different processes (i.e. analytical mixing, effect of a constitutional supercooling or diffusion between olivine and the melt inclusion, for example). These processes are now discussed.

Smooth transition zones could result from a too low spatial resolution, causing an overlap of olivine and crystals under the NanoSIMS primary beam ion spot. The variations in concentration along the profiles are indeed close to a two-end-member mixing model (Fig. 3a). However, numerical simulations (not reported here) show that, contrary to measured values ranging between 150 and 200, the primary ion spot size must be ≥1 μm in order to reproduce the observed transition zone (Fig. 3b). In addition, a mixing model caused by such a narrowed primary ion spot would not explain the difference in the transition zones between Ca and other elements. Hence, the transition zone cannot be entirely explained by an overlap of glass and olivine under the primary beam.

Constitutional supercooling is a solidification process: when solidification appears, elements will be partitioned between liquid and solid, hence creating a variation at the melt interface composition, subsequently changing the liquidus temperature at the interface. If the liquidus temperature gradient becomes higher than the temperature gradient in the melt, it yields a supercooled zone at the interface. The compositional gradient created by this constitutional supercooling could resemble that analysed by the NanoSIMS. However, it remains difficult to explain the different sizes of the transition zone for the different elements.

A different explanation could be found in diffusion processes. Indeed, the shape of the transition zones resembles the diffusion profile between two different compositions. In this situation, the boundary layer between olivine and glass would remain chemically unchanged during the diffusion. In other terms, in diffusion, the boundary layer should be envisaged as a physical object implying that olivine's stoichiometry is conserved in the diffusion zone. This is clearly not the case since the SiO₂ content drops down to 43 wt.% in the olivine (Fig. 4, thermodynamical parameters from [2,7]), while FeO does not compensate for MgO. On the contrary, in the NanoSIMS composition profiles, FeO follows MgO concentration from olivine to glass.

Another process could explain the observed parallel FeO and MgO concentrations: the dissolution of olivine in the silicate melt. Olivine dissolution experiments have been done by Zhang et al. [14] in silicate melts at high pressure (Fig. 5). Their results show composition profiles with the same shape than our composition profiles in the melt, with a different scale. In their experiments, Zhang et al. [14] showed that this dissolution was controlled by diffusion, and hence the extent of the profile is related to the heating time: the shorter the heating time is, the shorter the profile size will be. Therefore, we can imagine that olivine found in type-II chondrules dissolved in the liquid of the melt inclusion. In such a situation, the residual liquid has a composition incompatible with olivine, which results in the dissolution of the crystal at the interface. If dissolution is controlled by diffusion, the dissolution process must have been very short (seconds to minutes) when considering the small size of the transition zone.