3D X-ray tomographic analysis reveals how coesite is preserved in Muong Nong-type tektites

Muong Nong-type (MN) tektites are a layered type of tektite associated to the Australasian strewn field, the youngest (790 kyr) and largest on Earth. In some MN tektites, coesite is observed in association with relict quartz and silica glass within inclusions surrounded by a froth layer. The formation of coesite-bearing frothy inclusions is here investigated through a 3D textural multiscale analysis of the vesicles contained in a MN tektite sample, combined with compositional and spectroscopic data. The vesicle size distribution testifies to a post-shock decompression that induced melting and extensive vesiculation in the tektite melt. Compared to free vesicles, nucleated homogeneously in the tektite melt, froth vesicles nucleated heterogeneously on relict quartz surfaces at the margins of coesite-bearing inclusions. The rapid detachment of the froth vesicles and prompt reactivation of the nucleation site favoured the packing of vesicles and the formation of the froth structure. Vesicle relaxation time scales suggest that the vesiculation process lasted few seconds. The formation of the froth layer was instrumental for the preservation of coesite, promoting quenching of the inclusion core through the subtraction of heat during froth expansion, thereby physically insulating the inclusion until the final quench of the tektite melt.

. Major oxide composition of the glass of MP26 (normalized to 100 wt.% on anhydrous basis) and liquidus temperatures (Tliq) calculated at 0.1 GPa using Rhyolite-MELTS code (only for traverse 1).

Raman analyses of MN20 and P20
Raman spectra of SiO2 rich inclusions in two polished sections of samples MN20 and P20 (Fig. S2) were obtained in Lyon using a LabRAM HR800 Evolution spectrometer that has a confocal Czerny-Turner geometry and a laser source of 532 nm in wavelength. Each spectrum was acquired with a power of 10 mW, and 25 accumulations of 5 to 15 s. Gratings with 600 groove/mm were used in order to cover the frequency range 60 to 1300 cm -1 .

Figure S2. Backscattered electron images of samples MN20 (a) and P20 (b) indicating the spot location of Raman
analyses (c). The Raman spectra of coesite and quartz are reference spectra from RRUFF database.

Analysis of SR-µCT tomographic data
Reconstruction of SR-µCT tomographic data. The experimental parameters used for the tomographic scans are reported in Table S2. The reconstruction of the tomographic data was carried out by using the SYRMEP Tomo Project (STP) software suite [2] , applying pre-reconstruction filters for reducing ring artefacts caused by detector inhomogeneity. Prior to reconstruction, a single-distance phase-retrieval procedure [3] was applied to sample projections. This procedure allows to improve both the reliability of the further segmentation process, and related morphological and textural analyses, and to fully exploit the potential of phase-contrast imaging. Combining phase-retrieval with the Filtered Back-Projection algorithm [4] it is possible to obtain the 3D distribution of the refraction index of samples with constant composition, and thus characterized by a constant ratio γ = δ/β between the real and imaginary parts of the refractive index at a given X-ray energy. It was demonstrated that this kind of algorithm can also be employed on multiphase rock samples imaged with a filtered polychromatic synchrotron Xray beam [5] . In order to enhance the contrast between vesicles and rock matrix the y ratio was set at 50 and 29 for the 2.5 µm and 0.9 µm settings, respectively. Legend: XRM: X-ray microscopy, MCT: microfocus X-ray tomography, Bin: pixel binning of the detector.
3D image processing and analysis. The 3D image processing and analysis of MCT and SR-µCT data was performed using the Pore3D software library developed at Elettra [6][7] which allows extraction of quantitative microstructural and textural parameters of porous and multiphase systems. Segmentation and analysis of XRM data was performed using the free software Fiji [8] (Version 2.0.0-rc-69/1.52p). Examples of pore and inclusion segmentation are reported in Figs. S3 and S4, while results of the image analysis are reported in the supplementary Excel spreadsheets. The tools available in Pore3D and Fiji allowed a quantitative description of the morphology and topology of the components (vesicles and coesite-bearing inclusions), as well as the analysis of the bulk porosity. In order to perform the textural analysis, several volumes of interest (VOIs) were extracted from each sample (Table 1). Only VOIs of the order of or larger than 60 mm 3 can be considered representative of the grain heterogeneities and thus defined as Representative Elementary Volume (REV). After the extraction of the VOIs, XRM and SR-µCT data were filtered in order to remove noise and enhance edges by means of a 3D Bilateral Filter, which smooths images but preserves edges applying a non-linear combination of nearby image grey-level values. Image segmentation to obtain binary volumes containing only the objects of interest was performed by a manual 3D Otsu algorithm [9] on all volumes containing groundmass, vesicles and coesite-bearing inclusions. Outliers were then removed from the segmented images corresponding to the cut-off values listed in Table 1.
The Pore3D Basic Analysis module [6] was used to compute the vesicularity parameter (i.e., the fraction of vesicles in the selected VOI) and the topological characteristics, as described by the Minkowski functionals [10][11] such as: specific surface area (surface area of all vesicles divided by the total VOI), integral of mean curvature (index of the dominance of convex or concave shapes [12] , and Euler characteristic (index of connectivity of the object network [13][14] . The results of the Basic analysis are reported in Table S3. In particular, the Euler characteristic has the attributes of a topological order parameter describing the spatial connectivity where positive values typically consist of isolated objects dispersed in the matrix. Samples feature also positive values of the integral of the mean curvature, indicating a high fraction of convex surfaces [12,15] , as occurs for samples dominated by isolated spherical voids. After this computation, a filter was applied to the selected VOI in order to suppress blobs (connected components) connected to the image borders [6] . Then the Blob Analysis module of Pore3D was used to perform the analysis of each vesicle. This approach is based on the concept of maximal inscribed spheres. We then calculated the number of vesicles, their volume, the sphericity (ratio of the surface area of an equivalent sphere to the surface area of the object), aspect ratio (the ratio of the minimum and maximum axis of each vesicle) and diameter of the maximal inscribed sphere. The results of the Blob Analysis are also summarized in Table 1.
The 3D rendering of XRM data was done using the 3D Viewer plugin of Fiji while the volume (as reconstructed) and isosurface (segmented) renderings of MCT and SR-mCT data were obtained using both the commercial software VGStudio Max 2.0 (Volume Graphics, Germany) and the 3D Viewer plugin of Fiji.  Figure S3. Example of pores segmentation and extraction of a VOI from the MCT data on MP26a. Figure S4. Example of inclusions segmentation in a VOI from SR-µCT data on MP26b.