Quartz, mica, and amphibole exsolution from majoritic garnet reveals ultra-deep sediment subduction, Appalachian orogen

Samples

We study ~1050°C garnet-sillimanite-K-feldspar-biotite-quartz gneiss from the Brimfield Schist HPG-UHT field area (, ), located within one of several westward-dipping thrust fault-bound slices in the southern CMT. The rocks are highly aluminous (up to ~40 vol% garnet and ~30 vol% Al ₂SiO ₅) and are inferred to have metasedimentary, likely metapelitic, protoliths. Sillimanite pseudomorphs after kyanite are widespread in the field area (), and some samples preserve matrix kyanite. Metamorphism occurred during the Acadian and Neoacadian orogenies ~360 to ~420 Ma (see Materials and Methods for geologic history). The gneiss crops out ~35 km west of the Avalonia-Composite Laurentia subduction suture (fig. S1) as blocks of tens of meters in size within quartzofeldspathic gneiss and amphibolite as part of a ramp-flat sequence which contains ultramafic mélange including hornblende-spinel cumulate blocks in a chloritic, amphibole-bearing matrix (fig. S2). The field area has a complex history involving granulite (~750°C) and amphibolite facies (~575°C) overprints that produced a number of post-peak phases including cordierite and spinel (). The rocks of this study were almost certainly affected by these events, but the macroscopic mineralogical changes appear to have been relatively minor, primarily the pseudomorphic conversion of kyanite to sillimanite, alteration of matrix rutile to ilmenite, and alteration of garnet to biotite (or other hydrous phases) and rutile to ilmenite along cracks in garnet. The limited retrogression likely reflects limited post-peak fluid or melt infiltration. However, preservation of microdiamond and coesite is extremely unlikely given the multistage metamorphic history of the samples.

Garnet contains widespread needle- and plate-shaped lamellae with SPO parallel to <111> _garnet ( Figs. 2 and 3). Up to three concentric zones distinguished by different lamellae assemblages are present, an extraordinary and hitherto unreported texture ( Fig. 2A). The core zone has lamellae of rutile, ilmenite, apatite, and quartz ( Fig. 3, B to E), as well as talc and nonlamellar decrepitated inclusions that we infer to be crystallized fluid or hydrous melt (carbonate ± mica ± oxides ± quartz ± graphite) ( Figs. 2, B and C, and 3H). The mantle zone has lamellae of quartz, decrepitated mica + quartz ± rutile composites ( Fig. 3, F and G), minor rutile, and rare apatite and ilmenite. The outer ring zone has lamellae of rutile, ilmenite, apatite, quartz, orthoamphibole, clinoamphibole (hornblende), talc, Na-phlogopite (aspidolite), and phlogopite ( Fig. 3, A and I, and tables S2 and S3), as well as decrepitated inclusions similar to those in cores. Some outer ring lamellae comprise exotic combinations of amphibole, Na-phlogopite, phlogopite, and quartz ± apatite and rutile ( Fig. 3, A and I). Quartz is typically the dominant lamellae phase in all zones ( Fig. 2, B and C). The core lamellae assemblage (including decrepitated inclusions) is only shared with garnet in diamondiferous eclogite xenoliths from Yakutia (), whereas the mantle and outer ring assemblages are unique. Representative estimates of quartz lamellae content for cores range between ~0.4 and ~1.0 vol% (Materials and Methods).

Coesite and microdiamond are unlikely to have survived the multistage metamorphic history of the Brimfield Schist. This is probably why neither of these UHP phases have been found in our Raman investigations thus far (Materials and Methods). However, relics of coesite and diamond may remain. For example, some quartz lamellae have monoclinic shapes (fig. S4, A and B), suggesting that they precipitated as coesite [e.g., ()]. Furthermore, equant to cuboid graphite crystals that potentially represent former microdiamond are found in nonlamellar inclusions in the garnet that we infer to be former fluid or hydrous melt inclusions; these now contain mostly hydrous and/or carbonate phases (fig. S4C).

CORs and the precipitation hypothesis

We gathered COR data to test the lamellae precipitation hypothesis. Strongly developed apatite, ilmenite, and rutile COR have been demonstrated for these garnets (). Here, we present the first large COR dataset for quartz in garnet from any locality ( n = 268). These new data show that quartz has the same dominant COR as ilmenite: a-axis quartz//<111> _garnet (± 5°) ( Fig. 2D, fig. S3, and table S1) (). For quartz, this COR aligns rows of Si atoms with the garnet octahedral sites where Si ^VI substitutes in majoritic garnet. Other quartz CORs are also present, including c-axis in {111} _garnet (± 5°) and c-axis//<100> _garnet (± 15°), with one a-axis//<100> _garnet (± 15°) (table S1). Quartz rarely aligns c-axis//<111> _garnet, a COR common for apatite and ilmenite and rare for rutile. Overall, 72% of quartz lamellae have COR, comparable to results for rutile (). It is possible that decrepitation of fluid or hydrous melt inclusions created dislocations, facilitating precipitation () by heterogeneous nucleation or disrupting the garnet structure post precipitation. These processes could lower COR prevalence. Furthermore, as pointed out above, some quartz lamellae may have precipitated as coesite (fig. S4). These would have formed COR as coesite and not have α-quartz COR upon inversion.

Notably, electron probe microanalyzer (EPMA) spot analyses of garnet show very little TiO ₂, and commonly it is below detection limits (table S2). This is quite unusual for garnet in high-grade rocks [e.g., ()] and strongly suggests that Ti was nearly quantitatively removed from the garnet structure by precipitation of rutile- and ilmenite-bearing lamellae.

We conclude that lamellae are exsolved precipitates based on several lines of evidence. Lamellae components are plausibly soluble in the host garnet as constrained by experimental and natural samples (, ). Moreover, lamellae SPO is largely restricted to <111>, which connects the octahedral site in garnet where crucial lamellae nutrients substitute. Preservation of COR indicates that lamellae formed preferred relationships with the host, and shared COR between different lamellae minerals in the same garnet is robust evidence for precipitation (). Particularly, quartz lamellae share a dominant COR with ilmenite, demonstrating COR controls at the crystal system level (). This consistency would not be expected if lamellae were the product of, for example, external emplacement.

Some non-exsolution hypotheses for producing lamellae have been proposed, including overgrowth of lamellae by garnet, oriented lamellae attachment to garnet growth faces, various forms of epitaxial intergrowths, and cleaving of garnet followed by infiltration of fluid or melt from which lamellae grew (, ). However, despite more than a half century of experimental observations and more than two centuries of studies of natural garnet, we are not aware of examples of oriented lamellae attached to or protruding from garnet growth faces, severely complicating any sort of oriented attachment or garnet-lamellae cogrowth hypothesis. Furthermore, hypotheses involving cleavage are precluded by the fact that garnet lacks cleavage. Detailed discussions of these various lamellae formation hypotheses are given in () and ().

Alternatively, if lamellae were recrystallization products of preexisting inclusions, they would be expected to have roughly equant or spherical shapes, which minimize surface area-to-volume ratios and interfacial free energy. This is in strong contrast to the observed highly anisotropic needle and plate morphologies. Needles and plates are hallmarks of precipitation because they minimize lattice strains with the host (), such as in Widmanstätten patterns ().

Last, we note that the quartz lamellae described here are unrelated to the somewhat irregular, elongated quartz grains found in garnets with textural sector zonation formed by epitactic, highly overstepped growth (). This quartz is elongated roughly perpendicular to growth faces and has no quartz-garnet COR and no SPO parallel to <111> _garnet (, ).

Majoritic garnet and precursor garnet chemistry

If lamellae precipitated from garnet, then precursor garnet composition may be calculated (, , , ) using reintegrated chemical analyses of garnet and lamellae (Materials and Methods). However, because some lamellae phases such as rutile and quartz cannot precipitate stoichiometrically from garnet, reintegration does not produce a stoichiometric garnet formula for some lamellae suites. This complication requires a new approach to study the viability of reintegration, especially for quartz lamellae, which would demonstrate UHP majoritic garnet if successfully reintegrated.

Majoritic garnet compositions vary by rock type. Experimental and natural samples show that garnets in ultramafic bulk compositions are characterized by the Maj substitution, garnets in metasedimentary bulk compositions are characterized by Na-Maj, and eclogites are characterized by either or a combination (, ) ( Fig. 4a). Plotting Si + Ti + P against Al + Cr + Fe ³⁺ + M ²⁺ shows that the trend defined by metasedimentary majoritic garnets is distinct from the ultramafic trend because Na-Maj does not balance Si ^VI with M ^VI. The observed trend's −1.93 slope is close to the expected −2.0 for the Na-Maj substitution but is slightly larger indicating some Maj component ( Fig. 4A).

EPMA measurements of garnet + lamellae using wide beam grids of varying scales in each garnet zone (Materials and Methods) show M ^VI-deficient compositions, consistent with the metasedimentary trend ( Fig. 4A and Table 1). When Na is included, however, grid results fall close to but below stoichiometric majorite ( Fig. 4B), because they have underfilled dodecahedral sites lacking significant Na as well as Al ^IV or overfilled octahedral sites ( Table 1). These results are similar to those for quartz lamellae-bearing eclogite from Yakutia, which has Na-deficient calculated precursor compositions, whereas unexsolved garnets from the same kimberlites show high Na ₂ O contents (fig. S5 and Materials and Methods) (). Furthermore, loss of Na associated with rutile and apatite precipitation has been documented in diamondiferous UHP garnet from the Saxonian Erzgebirge [Fig. 7 of ()]. Areas in garnet lacking precipitates have elevated Na, Ti, and P contents, whereas precipitate-bearing areas are depleted in these elements. Evidently, then, Na becomes uncoupled to Ti and P during precipitation and diffuses rapidly enough to leave the garnet structure (). These examples demonstrate that UHP garnet breakdown was accompanied by open-system precipitation (OSP) () involving Na loss and that previous precursor calculations may have produced nonstoichiometric results because open-system behavior was unaccounted for.

Consequently, we present a method for reconstructing garnet precursor compositions using mass balance based on the observed majorite substitution vectors Na-Maj (Na ^VIII + Si ^VI = M ^VIII + Al ^VI) and Maj (M ^VI + Si ^VI = 2Al ^VI) ( Fig. 4A). This requires a geochemical reference frame [e.g., ()]. M ^VIII cations are clearly unsuitable. For example, Mg will have a characteristic length scale of several millimeters for 10 ⁶ years at 1000°C (), as evidenced by completely flat Mg profiles (fig. S6). More work is needed to characterize Al diffusion but, using Cr ³⁺ as a proxy, the diffusion coefficient D_Al is likely to be only 0.5 to 1.5 log ₁₀ units slower than M ^VIII cations (). Aluminum profiles in garnet are flat and, thus, qualitatively consistent with appreciable Al diffusivity (fig. S6). Sodium diffusion coefficients are not well known, but natural garnet records substantial mobility when lamellae precipitate (), and the measured Na contents in garnet for this study fall within a narrow range (table S2). In sharp contrast, Si ^IV diffusion in nesosilicates is extremely slow. For forsterite, is 10 ⁴ to 10 ⁵ times smaller than for garnet M ^VIII cations (). In our rocks, radial Si ^IV growth zonation is retained in a mapped garnet preserving all three lamellae zones (fig. S6), indicating that Si ^IV has extremely low mobility, even at the >1000°C conditions of metamorphism. Phosphorus and Si largely covary on the tetrahedral sites, and the amount of variation is at most a few tenths of a weight % (wt %) (fig. S6 and table S2). We note that slow diffusion of majorite component also entails sluggish diffusion of coupled Al (). However, once quartz- and Ti-bearing lamellae begin to form, the remaining garnet composition becomes more aluminous and, therefore, D_Al is likely to increase, as exemplified by documented cases of pyroxene exsolution (, , , , ).

Thus, we use Si as the immobile reference frame, positing that Si ^IV remained in tetrahedral sites and Si ^VI deposited locally as quartz lamellae or as silica in other silicate lamellae (e.g., amphiboles) such that Si measured by EPMA grids represents total Si ^IV + Si ^VI in precursor garnet. Reconstruction combined the Na-maj (−2.0) and Maj (−1.0) substitutions proportionately to produce the −1.93 slope defined by experimental results on metasedimentary bulk compositions ( Fig. 4A). The Na-maj component is adjusted by iteratively adding increments of 0.0001 mol of Na and subtracting 0.0001 mol each of Mg and Al, as set by the exchange vector, for each iteration step. Similarly, the Maj component is adjusted by iteratively adding 0.0001 mol of Mg and subtracting 0.0002 mol of Al. The garnet formula is then renormalized to 12 oxygens before the next iteration step. The iterations proceed until the dodecahedral site is filled at 3.0. Furthermore, the octahedral site should sum to 2.0 when the iteration is finished, providing a check on the robustness of the calculation ( Table 1). The choice of M ²⁺ cation to add or subtract (Mg in this case) is arbitrary.

As the main substitution vectors ( Na-Maj and Maj) are known ( Fig. 4A), the primary question surrounding calculation applicability is the Si geochemical reference frame. Tests easily show that reconstructions using Na or Al reference frames fail to converge, producing spurious negative site occupancies. Reconstructions using an M ²⁺ reference frame for substitutions dominated by Maj produce similar, unacceptable results. Reconstructions using an M ²⁺ reference frame can converge if the substitution mechanism is dominantly Na-Maj. However, the use of a M ²⁺ reference frame is precluded by the high diffusional mobility of these cations as discussed above. In sharp contrast, the Si reference frame produces the garnet site occupancies of 3.0 and 2.0 for dodecahedral and octahedral sites, respectively, and compositions fully consistent with the coupled majorite substitutions.

Reconstructed formulae require modest Al and Mg mass loss (mean −0.68 ± 0.48 wt % 2σ) and Na mass gain (mean +0.44 ± 0.31 wt % 2σ) ( Table 1). The reconstructed compositions fall on the expected majorite trend when Na is included ( Fig. 4B), demonstrating that the amount of Na restored balances the majoritic substitution preserved by lamellae. Consequently, the majoritic garnet signal is retained by silicate precipitation and may be recovered. Because the ratio of Na-maj and Maj substitutions is approximate and we have not accounted for other possible substitutions, such as those including K, the formulas should be regarded as semiquantitative. Nonetheless, they firmly establish the operation of majorite-type substitutions consistent with lithological trends ( Fig. 4A).

Although we consider open-system behavior as a means to produce quartz lamellae, note that there is no cause for Si to diffuse into garnet (e.g., from the matrix) during lamellae formation. Garnet Si content increases with pressure (, , ), and exsolution processes occur during retrogression. This excludes the possibility of Si diffusing into the garnet structure from the matrix to form lamellae because, to do so, the garnet would have to become (more) majoritic. Similar arguments obtain for titanium and phosphorus. Substitutions that invoke substantial Al ^IV to force Si off the tetrahedral site to make a silica polymorph are not documented and, in any case, are inapplicable because the garnets have little or no tetrahedral Al (table S2).

We applied a recent majorite barometer () to estimate formation pressures for each Zone using reconstructed formulae ( Table 1). Pressure estimates range from ~6 GPa for core and outer ring zones to ~7 to 8.5 GPa for mantle zones.

Majoritic lamellae

Precursor majoritic garnet is also fingerprinted by talc, amphibole, and mica lamellae. Each of these minerals may be formed from majorite components and water ± SiO ₂

Water could be supplied by garnet, which can hold up to ~1000 parts per million of H ₂O at UHP conditions (). As these model reactions show, amphibole and mica lamellae (e.g., Fig. 3, A, F, G, and I) could have formed entirely from majoritic garnet components. Some precursor garnet K content, as demonstrated in metasedimentary bulk compositions (), would yield potassic mica. Notably, the assemblages preserved in composite silicate lamellae (e.g., aspidolite + phlogopite + anthophyllite + quartz + rutile + apatite; Fig. 3A) do not correspond to known matrix assemblages in rocks or melt compositions of which we are aware. However, their unique nature can be accounted for by majorite breakdown reactions. We also distinguish mantle zone micaceous lamellae produced by model reactions such as Eq. 5 from needles formed by melt elongation () because the former rarely contain central rutile and therefore lack an elongation mechanism and, critically, are inconsistent with melt compositions. We speculate that the decrepitation features present around the micaceous lamellae ( Fig. 3F) formed when mica precipitates underwent partial melting during exhumation and then locally ruptured the garnet during further decompression [e.g., ()].

The best-fit slope of −1.93 for the metasedimentary trend on Fig. 4A is somewhat greater than the ideal 2.0 for the Na-Maj substitution, indicating some (~7%) Maj component. As shown in Eqs. 1 and 4 above, Maj or the calcium analog ( Ca-Maj) is required to form amphibole lamellae. Talc formation would also require breakdown of Maj component ( Eq. 2). The talc- and amphibole-bearing lamellae are the least common ones observed, and this can be accounted for by the dominance of the Na-Maj substitution relative to Maj or Ca-Maj. Hornblende, anthophyllite, and gedrite are grossly inconsistent with the mineral assemblages expected in aluminous metasedimentary gneisses but are fully compatible with the breakdown of majoritic components in garnet. As the amphiboles can precipitate stoichiometrically, or very nearly so, from garnet via the Maj and Ca-Maj substitutions, they need not have been involved in any OSP processes.

Garnet phosphorus content

Garnet phosphorus content increases at elevated P-T conditions (, ). Measured (not reconstructed) P ₂O ₅ in garnet EPMA spot analyses reaches 0.32 wt % in garnet mantle zones commensurate with the highest majorite contents (table S2). Experimental and natural garnets ( n = 32) containing 0.24 to 0.36 wt % P ₂O ₅ all equilibrated at ≥2.5 GPa (mean 6.3 GPa, 5.2 to 7.7 GPa 2σ; fig. S8). The quartz lamellae-bearing rocks are phosphorus poor (0.03 to 0.04 wt % bulk rock P ₂O ₅; Materials and Methods) and, thus, were likely apatite undersaturated with garnet, kyanite, and K-feldspar all competing for phosphorus (). Measured garnet phosphorus contents are therefore lower than would be expected for a phosphate-saturated rock yet still are consistent with equilibration at UHP conditions. We speculate that garnet phosphorus contents would have been considerably higher if the rocks were more phosphorus rich.

High phosphorus contents have been found in non-UHP, spessartine-rich garnets from pegmatites and their orthogneiss host rocks (). These garnets have very low MgO (~0 to 1.5 wt %) and up to 40 wt % MnO (fig. S9), as well as distinctive very low CaO (<0.35 wt %) and little Na ₂O. The substitutions producing elevated phosphorus contents in these garnets are not well constrained, although Al ^IV is present and may be involved, in addition to vacancies or some unidentified light element (). The key point is that none of the substitutions that produce elevated phosphorus at HP or UHP conditions () have been identified in these non-UHP garnets. Moreover, the rocks have high bulk rock P ₂O ₅ (up to 1.5 wt %) and can contain Li, Fe, Mn, and Ca phosphates; topaz; lepidolite; amblygonite; beryl; columbite; and/or considerable tourmaline. These bulk compositions are clearly not comparable to the aluminous gneisses we study.

In summary, the main characteristics of the pegmatite and (meta)igneous spessartine garnets are: very low Mg and Ca, very high Mn, ubiquitous presence of Al ^IV, and common association with exotic pegmatitic minerals. These characteristics are grossly at variance with the garnets of our study and UHP garnets generally.

Pseudosection results

Pseudosection results for a sample with plentiful quartz lamellae ( Fig. 2B) show that corundum, spinel, and/or cordierite stability is predicted below ~3 GPa at relevant ~1050°C temperatures (fig. S7 and Materials and Methods). These minerals are well known from UHT and/or HPG rocks at the locality (-), so their absence in the rocks of this study suggests matrix equilibration at UHP conditions in excess of about 3 GPa. This pressure result holds for a wide range of H ₂O activities and even at minimum is still ~2.6 GPa (fig. S7). Furthermore, quartz is not predicted in the matrix assemblage over most of the investigated P-T space for this sample and is absent at UHP conditions, further invalidating any possibility of quartz lamellae representing overgrowths by garnet. The pseudosection does predict a small amount (<1 vol%) of jadeite-rich clinopyroxene at UHP conditions, which is not observed. This is likely due to the small amount and/or the strong propensity of jadeite to rapidly revert to sodic feldspar components during retrogression.

Regional relations and P-T history

The rocks of this study are part of the Brimfield Schist, a unit within the CMT that is bounded by westward-dipping thrust faults (fig. S1). Granulite facies metamorphic conditions for the CMT and neighboring units are well documented (-), and UHT metapelitic gneisses (, ), as well as a silica-undersaturated HPG (), have recently been discovered within the Brimfield Schist. Ternary feldspar reintegration thermometry yields an average temperature of ~1050°C for the HPG (), while Zr-in-rutile thermometry and ternary feldspar reintegration thermometry give average temperatures of ~1000°C for the UHT metapelitic gneisses (, ). Sillimanite is the primary aluminosilicate phase found in the Brimfield Schist, although rare relict kyanite is present in the rocks of this study, and sillimanite pseudomorphs after kyanite are well known throughout the Brimfield Schist (, , ). Brimfield Schist zircon ages do not show a significant peri-Gondwanan signature and record metamorphism between ~360 and ~420 Ma (, ).

Evidence for partial melting is preserved in the field as variably developed leucosomes in the gneiss and as leucosomes bordering the blocks in their quartzofeldspathic gneiss and amphibolite host unit (fig. S2). The timing of melting is unconstrained but does not affect our results, as the UHP diagnosis is independent of bulk rock water content (fig. S7).

Several lines of evidence support a prolonged retrograde history for the Brimfield Schist. Granulite facies metamorphism at 700° to 800°C and 0.5 to 0.6 GPa texturally overprints the UHT metapelitic gneisses (). Cross-cutting sillimanite signifies granulite facies conditions in the HPG (). A further kyanite zone overprint is known from cross-cutting kyanite-garnet-quartz-carbonate veins () and texturally late kyanite (, ).

Samples, petrography, and imaging

Rock samples of garnet-sillimanite-biotite gneiss were collected from the field site in northern Connecticut, USA (41.873164°N and 72.275133°W). Petrographic thin and thick sections were made at Yale University. Polishing steps used SiC grit, diamond paste, and lastly, 0.05 μm colloidal silica suspension. Lamellae selection for electron backscatter diffraction (EBSD) analysis followed the procedure in (), with lamellae with clear needle or plate forms and an orientation along <111> _garnet chosen. Lamellae locations were recorded. Vertically integrated thin section images were made using a Leica DMC2900 camera attachment and LASv.4.10 software by manually focusing through the depth of the thin section. The sections for 58A-8 and 374A-1 in Fig. 2 are 45 and 200 μm thick, respectively. All sample numbers begin with the prefix JAQ, which is omitted in figures and tables for brevity.

Bulk rock chemistry

Rock bulk chemistry for the pseudosection (sample 58-A) is given in fig. S7. Both x-ray fluorescence analyses for major and minor elements and inductively coupled plasma mass spectrometry analyses for trace elements were performed by SGS Mineral Services. The P ₂O ₅ contents for representative samples 58A and 334C are 0.04 and 0.03 wt %, respectively.

Pseudosection analysis

Pseudosection calculations (fig. S7) were done in the system Na ₂O─CaO─K ₂O─FeO─MgO─Al ₂O ₃─SiO ₂─H ₂O─TiO ₂─O ₂ (NCKFMASHTO) using the Theriak-Domino software package () together with the thermodynamic database of (). Activity models follow the usage of () except for clinopyroxene, which is from (). The rocks contain graphite and pyrrhotite, and garnet has little or no Fe ³⁺ detectable by EPMA (see "Mineral chemistry" section). Consequently, molar Fe ³⁺/(Fe ²⁺ + Fe ³⁺) was set to the low value of 0.01. Increasing this to as much as 0.1 has little impact on the phase relations relevant for our study. Mineral abbreviations used in pseudosections and figure captions follow ().

Mineral chemistry

Quantitative mineral chemistry analyses and backscattered electron imaging were done using a JEOL JXA-8530F field emission gun EPMA in the Yale University Department of Geology and Geophysics. Garnet unknowns and standards were coated in the same carbon coat application to standardize coat thickness. Kakanui garnet was used as a garnet standard for Si, Al, Ca, and Mg because of its well-characterized ternary composition (). Spessartine was used as the Mn standard and almandine as the Fe standard.

Wavelength-dispersive spectroscopy was used in all runs with natural and synthetic standards and off-peak background corrections. Operating conditions were 15 kV of accelerating voltage and 10.8 mm of working distance with a focused beam. Beam current was 5 nA for apatite, 10 nA for micas and amphiboles, and 20 nA for garnet. Wide beam grids used 25 μm diameter beam spots spaced 25 μm apart. The garnet Si map of fig. S6 used 15 kV of accelerating voltage, 75 nA of beam current, and 200 ms of dwell time with a 1-μm spot size. The garnet P and Al maps of fig. S6 used 15 kV of accelerating voltage, 300 nA of beam current, and 200 ms of dwell time with a 10-μm spot size.

The average structural formula for the garnet analyses sums to eight cations, which implies that garnet Fe ³⁺ is minimal (table S2). It would likely have been higher if the samples had reached greater depths (, ).

Raman spectroscopy

Raman spectra were taken using a HORIBA-Jobin Yvon HR-800 Raman microscope at Yale University. The instrument uses a 50-mW green laser (532 nm) with a grating of 1800 lines/mm and was calibrated using a Si wafer.

EBSD analyses

EBSD analyses followed the methods in (). Analyses were performed using a Phillips XL30 Environmental Scanning Electron Microscope at Yale University and Oxford Instruments HKL Channel 5 Flamenco software. Analytical conditions were the following: 1 × 10 ⁻⁴ Pa of minimum vacuum strength, 20 mm of working distance, 70° sample tilt, 15 kV of accelerating voltage, ~2 nA of beam current, and beam spot size of 5 (allowing for analysis of lamellae with a diameter of~ ≥1 to 2 μm). Before analyses, backgrounds were acquired in the polygranular rock matrix with 150 to 200 ms timing per frame and 2 × 2 binning. Multiple spot analyses of host garnet were collected for each grain to ensure that garnet hosts were single crystals. Band detection limits were 5 (min) and 6 (max), and only analyses with a mean angular deviation of <1 were considered.

Quartz lamellae volume estimation

Vertically integrated images of thin- and thick-section core zones were taken with the Leica DMC2900 camera attachment and LASv.4.10 software. These images have a scale calibrated to the microscope objective used while acquiring the image. Thin sections cut perpendicular to garnet <111> were selected. Using measuring tools in the Leica Application Suite software, quartz needles were measured for diameter and length, which were used to calculate the volume of each quartz needle assuming a cylindrical shape. Because the sections are viewed along <111>, lamellae are foreshortened because of their dip angles of 35.26°. To correct for this, we multiplied measured lamellae length by 1/cos(35.26°). The vertically integrated images do not show the length of lamellae parallel to the <111> coming out of the image, and consequently, SiO ₂ estimates are minima because this direction should account for ~25% of all lamellae. The measured quartz mode for sample 58A-8 (30,000 μm ²) is 0.46 vol%, and the corrected is 0.77 vol%. The measured quartz mode for sample 374B-1 (71,780 μm ²) is 0.25 vol%, and the corrected is 0.31 vol%. These increase to 1.03 and 0.41 vol% if the additional inferred ~25% of the lamellae parallel to <111> are included.

COR categories

Determination of COR types followed the methods in (). The primary search criterion for COR was small angular deviations from major garnet structural features (e.g., <111> and {111}). The proportion of quartz with a prominent COR for rutile, c-axis within a 28.5° ± 2.5° cone around <111> _garnet, is also reported (table S1). Reporting results of the same COR search for each mineral allows for better standardization of COR selection methodology. The most common CORs for each investigated lamellae mineral are shown in fig. S3.

We found only α-quartz-garnet COR, which indicates that most of the SiO ₂ precipitated as α-quartz. It is possible that some of this inverted to β-quartz and then back to α-quartz during the retrograde history. It has been shown, however, that in this case neither the a-axis nor the c-axis change position during the α → β and β → α transformations ().

Precursor majorite reconstruction

Garnet + lamellae compositions were measured by EPMA wide beam grids with a spot size of 25 μm. Wide beam grid results are given in Table 1. Precursor compositions were calculated from these grids using a MATLAB code available from the authors by request. Results are identical if the Na ₂Mg[Si ₂]Si ₃O ₁₂ end-member () is used to represent Na-maj, and exchange vectors are adjusted accordingly.

The precursor reconstruction method was tested using exsolved and unexsolved eclogite garnet examples from Yakutia. Sample UV662/11 garnet illustrated in Fig. 4 (A and B) in () contains needle-shaped quartz lamellae, as well as coesite and monoclinic quartz paramorphs, after coesite. The lamellae of our study appear texturally identical, so UV662/11 is an excellent basis for comparison. The reintegrated garnet + lamellae composition () is Na deficient relative to unexsolved majoritic garnet (). This is confirmed in fig. S5A, which shows that the garnet plots near the Na-maj trend. In fig. S5B, which includes Na content, the garnet falls off the general majorite trend due primarily to the Na deficiency. The reconstruction produces majoritic garnet with perfect site occupancies and only requires small changes in mass. All of these traits are comparable to the garnets in our study.

Figure 4 data sources

Compositional data for natural and synthetic majoritic garnets plotted in Fig. 4 were gathered from the literature from the following references: (, -). Data from () with wt % totals >102% or <98% were excluded.