Low water contents in garnet of orogenic peridotite: clues for an abyssal or mantle-wedge origin?

Garnet-bearing ultramafic rocks (GBU) occurring as lenses within high-grade gneiss and granulite units represent valuable witnesses of geodynamic processes because they record the emplacement of mantle material into the continental crust and delineate suture zones within collisional belts (e.g., Coleman, 1971). The emplacement of GBU within crustal rocks is attributed to orogeny during which upper-mantle slices are intercalated in deeply subducted continental crust. To understand crust–mantle interaction during orogeny and to reconstruct orogenic processes, it is vital to unravel the setting of GBU before they were emplaced in the continental crust. For this, different possibilities were suggested: (i) GBU may represent pieces from the mantle wedge that were emplaced in the underlying slab during subduction or exhumation (e.g., Brueckner, 1998; Scambelluri et al., 2006; Brueckner et al., 2010), (ii) other peridotites were part of the asthenospheric mantle just before they came in contact with a subducting plate (Medaris et al., 1990; Schmädicke et al., 2010), (iii) and yet another possibility is that GBU originated as abyssal peridotite that were subducted together with eclogite (Evans & Trommsdorff, 1978; Evans et al., 1979).

In the Erzgebirge, both GBU and common, metabasaltic eclogite are present. Common eclogite occurs in three high-pressure (HP) units that record different pressure–temperature (P–T) peak conditions, one of which is an ultra-high pressure (UHP) unit. The ultramafic rocks are restricted to the latter unit, which also contains coesite eclogite and diamond-bearing gneiss (Schmädicke, 1991, 1994; Nasdala & Massonne, 2000; Massonne, 2001). Garnet peridotite is the dominating type of ultramafic rocks, which includes volumetrically small layers and lenses of garnet pyroxenite and garnetite (Schmädicke & Evans, 1997). The Erzgebirge GBU were interpreted as fragments from the mantle wedge that were picked up by the eclogite-bearing UHP unit during subduction as the unit was near its maximum depth of burial (Schmädicke & Evans, 1997). This suggestion was based on identical P–T peak conditions of GBU and coesite eclogite and the observation that GBU are absent in the other two, less deeply subducted HP units with quartz eclogite.

In this study, GBU from Erzgebirge and two additional samples from Alpe Arami (Lepontine Alps) are investigated: (i) to obtain information on water in garnet for orogenic peridotite and (ii) to find out if such data may provide clues with respect to petrogenesis. Hitherto, little is known about water in orogenic GBU (compared to xenoliths) and available data are mainly restricted to olivine and pyroxene (Peslier, 2010; Yu et al., 2019). Unfortunately, due to strong serpentinization of the EG samples, only garnet can be used in the present context. However, garnet is a suitable proxy for water in the peak assemblage because, (i) it is present in all UHP rocks of the Erzgebirge, (ii) it is well preserved and forms relatively large grains, and (iii) data on garnet from EG eclogite are already known (Gose & Schmädicke, 2018). Thus, garnet is an ideal mineral to compare the different rock types. The available data show that garnet in both quartz and coesite eclogite is water-deficient, containing less water compared to its water storage capacity deduced from experimental studies. The authors inferred that most of the water generated by dehydration reactions during subduction obviously left the slab and, most probably, migrated to the mantle wedge. In the case that the Erzgebirge GBU were derived from the lowermost part of the wedge and incorporated into the subducting slab at peak metamorphic conditions (Schmädicke & Evans, 1997), they should have incorporated this water as structural OH (colloquially “water”) in their nominally anhydrous minerals (NAMs). If true, peridotitic garnet is expected to contain distinctly more water than garnet of associated, strongly dehydrated coesite eclogite. With the present study, we try to find out whether garnet possibly gives some clues in this context.

The majority of the investigated samples are derived from the Erzgebirge, Germany. The Erzgebirge represents a rare example of UHP metamorphism linked to Variscan orogeny, with peak metamorphic ages of about 360 Ma (Schmädicke et al., 1995, 2018). The Erzgebirge is made up of an 80 km by 40 km large, oval-shaped northeast–southwest trending crystalline complex (Fig. 1) that is located at the northern margin of the Bohemian Massif. The crystalline complex consists of a monotonous gneiss–migmatite unit, devoid of eclogite-facies relics, that is overlain by three HP units (Units 1, 2, and 3; Fig. 1) with eclogite-facies rocks (e.g., Klemd & Schmädicke, 1994; Schmädicke, 1994; Schmädicke et al., 1995). All HP units are composed mainly of high-grade quartzofeldspatic gneiss in which eclogite lenses with MORB composition occur as intercalations.

The eclogite-facies peak conditions increase systematically from Unit 3 (600–650 °C, 20–22 kbar), to Unit 2 (670–730 °C, 24–26 kbar), and to Unit 1 (840–920 °C, ≥30 kbar; Schmädicke et al., 1992; Schmädicke, 1994). The HP Units 2 and 3 contain quartz eclogite whereas Unit 1 experienced UHP conditions enabling the formation of coesite in eclogite (Schmädicke, 1991, 1994; Massonne, 2001) and diamond in felsic rocks (Nasdala & Massonne, 2000; Stöckhert et al., 2001). Apart from diamond, symplectites of sodian diopside plus albite are present in felsic rocks (Schmädicke et al., 1992) indicating that the latter are co-facial with meta-basaltic eclogite and suggesting that both rock types experienced a common metamorphic history.

Mantle-derived, garnet-bearing ultramafic rocks are rare in the Erzgebirge and solely occur in the UHP unit (Schmädicke & Evans, 1997). The ultramafic rocks are present as elongate bodies within felsic gneiss, with a length from few hundred meters up to 1–2 km. The GBU lenses consist of variably serpentinized (up to 100 vol%) garnet peridotite that includes rare, 1–15 cm thick interlayers of garnet pyroxenite (Mathé, 1990; Schmädicke & Evans, 1997). The GBU lenses occur in the vicinity (several hundred meters apart) of lenses of coesite eclogite but not in direct contact with them. Thermobarometric investigations of Erzgebirge GBUs provided P–T peak conditions of ca. 900 °C and 33–36 kbar that are very similar to those of coesite eclogite occurring in the same unit (Schmädicke & Evans, 1997).

Two additional samples of garnet peridotite were selected from Alpe Arami in the Central Swiss Alps. The rocks belong to the Adula–Cima Lunga unit, which is part of Penninic nappe pile (e.g., Schmid et al., 1996). At Alpe Arami, garnet peridotite occurs in the core of a 1 × 0.4 km body of chlorite peridotite that is hosted by migmatitic gneiss (O’Hara & Mercy, 1966; Möckel, 1969). Garnet peridotite and metabasaltic eclogite of crustal origin are spatially related, as in the Erzgebirge, but in contrast to the latter both rock types may occur in direct contact. The P–T estimates for Alpe Arami peridotite of 32 kbar and 840 °C (Nimis & Trommsdorff, 2001) are very similar to Erzgebirge peridotite. However, it should be mentioned that there is no consensus concerning the P–T peak for this locality, including considerably higher estimates (i.e., P ≥ 50 kbar, T = 1100–1200 °C; Brenker & Brey, 1997; Paquin & Altherr, 2001; Hermann et al., 2005; Green et al., 2010). Hence, the derivation of the protoliths of Alpe Arami peridotite is uncertain whereas those from Cima di Gagnone, another occurrence of GBU in the Adula–Cima Lunga unit, are thought to represent metamorphosed equivalents of serpentinized ocean-floor peridotite (Evans & Trommsdorff, 1978; Evans et al., 1979; Trommsdorff et al., 2000).

The investigated Erzgebirge samples were derived from five localities: Ansprung, Zöblitz, Oberlochmühle, Niederlochmühle, and Lochmühle (Fig. 1; Table 1); three of them (except the latter two) were already subject of a previous study (Schmädicke & Evans, 1997) which includes a more detailed outcrop description. Peridotite is exposed in all locations, pyroxenite only in the Ansprung quarry. The following description of thin sections focuses on the primary mineral assemblages and the initial post-peak re-equilibration, pre-dating low-grade metamorphism and serpentinization.

All samples of garnet peridotite consist of the primary mineral assemblage garnet (grt) + olivine (ol) + orthopyroxene (opx) + clinopyroxene (cpx), which defines a granoblastic texture (Table 1). Despite strong serpentinization, grain contacts between the primary minerals are preserved in a few cases in most thin sections. Primary, dark-brown spinel was only found in peridotite from Oberlochmühle but its presence was also reported for peridotite from Zöblitz and Ansprung (Mathé, 1990). Spinel is extremely rare and occurs solely as inclusion in garnet and, thus, was previously interpreted as a relic of a pre-peak assemblage (Schmädicke & Evans, 1997). Very rare, pargasitic amphibole occurs in a few samples and, if present, is invariably intergrown with garnet and shares equilibrium grain boundaries with this mineral. The textures led to the interpretation that calcic amphibole grew mainly at the expense of garnet during post-peak re-equilibration (Schmädicke & Evans, 1997). Kelyphitic breakdown, generally being a ubiquitous feature of peridotitic garnet, is not observed in the Erzgebirge rocks.

The original grain sizes deduced for the primary minerals typically range from <1 to 10 mm, but garnet in samples from Zöblitz and Ansprung forms porphyroblasts of up to 15 mm diameter. The pyroxenes and olivine are affected by strong serpentinization; relics occur in clusters of smaller grains (≤1 mm; typically 0.1–0.2 mm) in the serpentine matrix, but judging from the parallel extinction of groups of such small individuals they obviously represent fragments of larger grains, which were separated by serpentinization along internal fractures. The size of these fractured grains is similar to or smaller than the garnet grain size.

Garnet pyroxenite from Ansprung has a granoblastic texture and is coarse-grained (grain sizes: grt > cpx >> opx). Garnet, clinopyroxene, and orthopyroxene form a well-preserved equilibrium assemblage. Garnet grains (1–15 mm in diameter) rarely host inclusions of clinopyroxene and opaque minerals. About 50% of the clinopyroxene grains show numerous exsolution lamellae (up to 3 μm width) of orthopyroxene. Inclusions of orthopyroxene and opaque minerals are very rare. Pargasitic amphibole was described to occur in textural equilibrium with grt, cpx, and opx (Schmädicke & Evans, 1997) but this mineral was not found in the samples investigated here.

Garnet peridotite from Alpe Arami consists of the primary four-phase assemblage garnet–olivine–orthopyroxene–clinopyroxene (Table 1). The rocks have a porphyroclastic texture. The overall grain size is similar to that of Erzgebirge samples. Serpentinization is very weak (5 vol%) so that grain contacts between primary minerals are well preserved. Spinel was not detected in our samples but rare inclusions of brown spinel were described previously (Nimis & Trommsdorff, 2001). Garnet is rimmed by dark-brown kelyphite, which is absent in the Erzgebirge samples.

Doubly polished, self-supporting rock slices with a thickness between 0.2 and 0.5 mm, depending on grain size and garnet crystal quality, were prepared for almost all samples. Only in the case of samples with weak coherence, grain separates were prepared (Table 1) for which at least 10 garnet grains were extracted per sample, embedded in a slice of epoxy resin and doubly polished. These specimens were used for all analytical tasks that were accomplished in the order, (1) infrared (IR) spectroscopy, (2) electron-microprobe analysis, and (3) inductively-coupled plasma mass spectrometry (ICP-MS). Performing the non-destructive IR spectroscopy first and the most destructive analysis (ICP) last allows one to collect all analytical data from identical garnet grains.

The major- and minor-element composition was determined with a JEOL 8200 electron microprobe, equipped with five wavelength-dispersive spectrometers, at conditions of 15 kV, 15 nA, and a counting time of 20–40 s. For calibration, silicate and oxide standards were used. At least three garnet grains were analyzed in each sample, and at least four spot analyses were collected for each grain. The obtained uncertainty was ≤2% relative. Selected analyses, representative for each sample, are given in Table S1 (in the Supplementary Material linked to this article and freely available online at https://pubs.geoscienceworld.org/eurjmin).

The trace element concentrations were analyzed using an UP193FX laser ablation unit (New Wave Research) connected to an Agilent 7500i quadrupole ICP mass spectrometer. Argon was used as plasma and cooling gas (14.9 L/min), auxiliary gas (0.9 L/min) and carrier gas (1.1 L/min) while He was utilized as secondary carrier gas (0.65 L/min). Spot diameter and repetition rate were 30–50 μm and 20 Hz, respectively, by applying an irradiance of 0.56 GW/cm² and a fluence of 2.8 J/cm². The background and mineral ablation times were 20 and 25 s, respectively. The NIST SRM 612 standard was used for external and sample SiO₂ content for internal calibration. The reference material BCR-2G (USGS) was used as secondary standard to evaluate reproducibility (mostly <5%) and accuracy (mostly <8%). Data evaluation was carried out with GLITTER (Van Achterbergh et al., 2000). For each sample 4–10 (mostly seven) analyses were collected. Average analyses for each sample are given in Table 2.

The water concentration (given as wt.ppm H₂O) was determined by Fourier transform IR spectroscopy using a Vertex 70 spectrometer, equipped with a Hyperion 3000 microscope and an MCT detector. The IR spectra were collected in transmittance mode with non-polarized IR radiation by averaging over 64 scans in the 550–7500 cm⁻¹ wavenumber range using an instrumental resolution of 2 cm⁻¹. Spectra were taken from clear and transparent crystal volumes, free of inclusions, cracks, and alteration products. To restrict the probed material to such volumes, a square aperture of 30 × 30–100 × 100 μm² (in most cases ca. 50 × 50 μm²) was applied. If possible, at least ten spectra were collected for each crystal from different grain portions. The integral absorbance was determined by fitting the spectra between 3000 and 3700 cm⁻¹. The amount of water in the garnet structure (= structural water [SW]) was determined from the absorption of bands with maxima in the wavenumber range of 3460–3670 cm⁻¹. For baseline correction, deconvolution, spectra fitting, and quantification of water the method utilized for eclogitic garnet (Schmädicke & Gose, 2017; Gose & Schmädicke, 2018) was adopted here. Water contents were calculated by applying the mineral-specific molar absorption coefficient of Bell et al. (1995) that was determined for a garnet with 4.6 wt% CaO, 19.5 wt% MgO, and 11.0 wt% FeO that is very close to the composition of the target garnets. The total uncertainty of the contents of structural water (Table 3) is estimated to be 20–30% in most cases.

Peridotitic and pyroxenitic garnet from Erzgebirge samples has high MgO contents (ca. 18.0–21.5 wt%; Fig. 2) and, accordingly, is rich in pyrope component (66–75 mol%) with considerably lower almandine (13–19 mol%) and grossular contents (4–9 mol%). Grains from the same sample have homogeneous composition and retrograde zoning is negligible, even at the outermost grain margin. Peridotitic garnet has distinctly higher Cr₂O₃ contents (sample average: ca. 1.0–2.5 wt%) compared to pyroxenitic garnet (<0.3 wt%; Fig. 3). Chromium is positively correlated to Ca (r = 0.91; Fig. 3) and Ca also to Sc (r = 0.99; not shown). The TiO₂ contents are low in all garnets (sample averages: 0.1–0.45 wt%) and do not depend on the rock type (Fig. 2). Titanium is positively correlated to Mg (r = 0.73; Fig. 2), and negative correlations are observed for the element pairs Ti–Mn (r = −0.81; Fig. 2), Ti–Cr, and Fe–Cr (−0.44 and −0.61, respectively). Sodium is one of the most abundant trace elements with sample averages between 150 and 620 ppm. Sample averages of P contents are between 110 and 244 ppm. Garnet from pyroxenite has higher Na and P concentrations (445–620 ppm and 218–244 ppm, respectively) than peridotitic garnet (Na, 150–450 ppm; P, 110–140 ppm). Phosphorus is positively correlated to Co (r = 0.83; Fig. 3) and negatively to Sc (r = −0.90; Fig. 3), Ca (r = −0.96), Zr (r = 0.64), and Ga (r = 0.80). Furthermore, a positive correlation is found for the element pairs Na–Zr (r = 0.76) and Na–Hf (r = 0.76; Fig. 3).

Garnet from Alpe Arami peridotite is slightly zoned, with somewhat higher Fe concentration at the rim. The major-element composition is similar to that of peridotitic Erzgebirge garnet (Table S1; Fig. 2). The same applies to most trace elements such as Na, P, Zr, and Hf (Fig. 3) whereas other trace elements (e.g., Sc) have concentrations more similar to Erzgebirge pyroxenite.

Peridotitic and pyroxenitic garnets from Erzgebirge and Alpe Arami show basically the same rare earth element (REE) concentrations. The normalized patterns (Fig. 4) show a steep increase from La to Sm and a flat one from Sm to Lu. The heavy (H) REE are concentrated by a factor of about two compared to primitive mantle.

Almost all samples reveal IR absorption bands due to structural water in garnet. Garnet grains from Erzgebirge (EG) and Alpe Arami (AA) samples show the same type of IR absorption pattern. The most typical IR absorption bands are present at 3570 ± 10 cm⁻¹ (dominant band), in the wavenumber range of 3600–3630 cm⁻¹, and at 3650 ± 10 cm⁻¹ (Fig. 5). Additional bands at lower wavenumbers (i.e., <3460 cm⁻¹) occur in about half of the probed garnet domains, predominantly in the EG samples. An example of the variability of spectra in single is shown in Fig. 5a. The spectra for peridotitic garnet (Fig. 5a and b) are very similar to those from pyroxenitic garnet (Fig. 5c).

The IR bands at >3500 cm⁻¹ are attributed to structural water (SW) in garnet. In analogy to former studies on garnet in Erzgebirge eclogite (Schmädicke & Gose, 2017; Gose & Schmädicke, 2018), IR bands centered at 3650 ± 10 cm⁻¹ are designated here as “type I”, those in the 3570–3630 cm⁻¹ range as “type II”. The half width of recorded SW bands (ca. 45 cm⁻¹ for type I, 50–70 cm⁻¹ for type II) is very similar compared to garnet of EG eclogite from the same unit (ca. 45 cm⁻¹ and 60–95 cm⁻¹, respectively). Bands centered at wavenumbers <3460 cm⁻¹ are two to three times broader, with a half width of 150–185 cm⁻¹, being similar to corresponding bands in EG eclogite (i.e., 150–160 cm⁻¹). Again, in analogy to the study on eclogite, these bands are ascribed to molecular water (MW) and designated as “type M” (Fig. 5). This inference is based on several observations such as, (i) the large band width and (ii) the fact that SW bands at this wavenumber are unknown for pyropic garnet, (iii) in contrast to the SW bands, the appearance of the MW band is irregular in the sample set as well as on grain scale, which can best be reconciled with inhomogeneously distributed inclusions with liquid water, (iv) in addition, the water concentrations derived by including MW bands are unrealistically high.

The type-I band is a typical feature of mantle garnet and present peridotite, pyroxenite, and mantle eclogite (e.g., Schmädicke et al., 2015). The band was also found in less pyropic garnet of crustal eclogite (Schmädicke & Gose, 2017; Gose & Schmädicke, 2018) and generated by HP experiments in natural, impure pyrope (e.g.Lu & Keppler, 1997) but not in end-member pyrope (Withers et al., 1998). Though the particular OH substitution mechanism is subject to discussion, the band is unequivocally ascribed to structural water in garnet. Bands in the range of 3585–3630 cm⁻¹ were described from a great variety of compositions: synthetic end-member pyrope (Geiger et al., 1991; Withers et al., 1998; Geiger et al., 2000), pyrope-rich garnet (Mookherjee & Karato, 2010), natural grossular-rich garnet (e.g., Beran et al., 1993), and natural grossular (Rossman & Aines, 1991) and ascribed to the hydrogarnet substitution, i.e. tetrahedral $(OH)44-$ replacing $SiO44-$ tetrahedra.

The band at 3570 ± 10 cm⁻¹ was found in synthetic Ti-doped pyrope (Geiger et al., 2000) and regarded as typical feature of garnet with ≥1 wt% TiO₂ (e.g., Geiger et al., 2000; Schmädicke et al., 2015). However, this band is also present in Ti-poor peridotitic garnet (Cima di Gagnone; Padrón-Navarta & Hermann, 2017) and in eclogitic garnet with less than 0.1% TiO₂ but absent in crystals with the same amount of TiO₂ (Gose & Schmädicke, 2018). Therefore, the latter authors suggested that the band may also be caused by the hydrogarnet substitution. They argued that the band at 3570 ± 10 cm⁻¹ is very close to the wavenumber range of the “hydrogarnet band” (3585–3630 cm⁻¹), and the latter is known for changing its position along with garnet composition (e.g., Geiger et al., 1991; Rossman & Aines, 1991). Due to the possibility that all bands in the 3570–3630 cm⁻¹ range could be related to the same substitution, they were summarized as “type-II” bands (Gose & Schmädicke, 2018), an approach adopted also in the present study.

Apart from SW and MW bands, some samples yielded spectra with unusually large band intensities indicative of secondary hydrous minerals such as amphibole and/or chlorite (cf.Skogby et al., 1990; Yang et al., 2018). For peridotite samples 151, NLM, and OM 13-1 only a few (1–3) of such spectra were recorded, but for pyroxenite Zö-Z4 they are very common. In this sample only a single garnet domain without such bands was detected (see Fig. S1 in Supplementary Material).

The content of structural water in Erzgebirge garnet deduced from SW band-types I and II is relatively low and does not depend on the rock type (i.e., peridotite or pyroxenite). The H₂O contents, including all analyzed domains (SW total; Table 3), range between 0 and 115 ppm and the sample averages are between 3 and 68 ppm. If the above band assignment is ignored and MW bands are included in the calculation of structural H₂O, the contents scatter in a large range from 0 to 269 ppm. The latter exceeds concentrations reported from natural pyropic garnet and it is also higher compared to values obtained in high-pressure experiments for H₂O-saturated conditions (see below), supporting the treatment of the broad MW band as molecular water.

The lowest concentrations of structural water are present in the two peridotites from Alpe Arami (sample averages: 10 and 13 ppm) and in Erzgebirge peridotite Zö 13-9 (sample average: 3 ppm). All three samples are characterized by particularly low contents of molecular water (sample averages: <5 ppm; Table 3). Higher contents are restricted to samples or garnet domains, respectively, that additionally host considerable amounts of molecular water as indicated by a strong type-M band. Comparing samples from different localities, the average contents of structural water show moderate variation (Table 3). In comparison, garnet grains from the same sample reveal similar or even larger variability with respect to both structural and molecular water. The contents of the latter may scatter over two orders of magnitude within a single sample, pointing to inhomogeneously distributed, sub-microscopic fluid inclusions with liquid water. Even a single garnet crystal may host domains being free of molecular water occurring next to grain portions with a few hundred ppm molecular water.

Hence, the above findings have the implication that part of structural water is, most probably, secondary (see Sect. 6). The data reveal that the presence of molecular water enhances the amount of structural water (Table 3). It is also obvious that the two types of structural water, corresponding to SW I and II bands, are differently related to molecular water. Including analyses from all analyzed domains, having molecular water of up to ca. 300 ppm, the intensity of band-type II is positively correlated with that of band-type M (r = 0.78; Fig. 6). In contrast, no correlation exists for band-types I and M (r = 0.16; Fig. 6). This finding is identical to former results on garnet of Erzgebirge eclogite (Schmädicke & Gose, 2017; Gose & Schmädicke, 2018) for which it was demonstrated that only garnet domains without molecular water (no fluid inclusions) are suitable to determine the primary content of structural water.

Accordingly, if only IR absorption spectra with very small or without type-M bands are considered, the content of structural water is less variable. In domains with ≤20 ppm molecular water (Table 3), the sample averages for structural water (SW I + II) range from 3 to 68 ppm, and in six (from 10) samples they are ≤24 ppm. The content of structural water due to band-type I (0–11 ppm) is much smaller compared to that attributed to band-type II (0–54 ppm). The same becomes obvious from the frequency distribution of water contents (Fig. 7). For Erzgebirge samples, the contents of structural water are 0–115 ppm if all domains (up to 300 ppm MW) are included. In contrast, domains with <10 ppm molecular water contain less than 60 ppm structural water, with a median of only a few ppm (Fig. 7). These domains have the same or even lower water contents than garnet from Alpe Arami that invariably hosts very little molecular water (<10 ppm; Table 3 and Fig. 7).

The content of structural water is related to garnet composition. Using sample averages of EG samples (including domains with <100 ppm MW), a positive correlation is observed for water and the contents of MnO (r = 0.59), Y (r = 0.68), and HREE (e.g., Er: r = 0.70) and a negative one for H₂O and MgO (r = −0.72), TiO₂ (r = −0.80), and V (r = −0.74; Fig. 8). A correlation between H₂O and CaO, as reported for garnet from EG eclogite, does not exist (r = 0.16). Adding the Alpe Arami samples to the dataset, elemental correlations are absent (Fig. 8). Considering domains with less molecular water (<50 ppm), the correlations are the same or slightly stronger (e.g., MgO–H₂O: r = −0.77, TiO₂–H₂O: r = −0.82). In contrast, if domains with considerably more molecular water (up to 300 ppm) are included, the mentioned elemental correlations are no longer present (e.g., MgO–H₂O, r = 0.33; TiO₂–H₂O, r = 0.10). Instead, a positive correlation arises between structural H₂O and Li (r = 0.95; Fig. 9) for EG garnet which, in turn, is absent if only domains with <100 ppm MW are considered (r = 0.09).

As demonstrated above, the content of structural water tends to be enhanced in garnet domains with fluid inclusions (containing molecular water). This correlation renders it unlikely that molecular water originated from dissolved structural water, in which case the relation should be opposite to the observed one. Accordingly, molecular water should have come from an external source. Incorporation of this water in the form of fluid inclusions obviously resulted in an increase of structural water in garnet volumes around such inclusions. The inference that part of the structural H₂O is secondary is also supported by the observation that it only correlates with major and immobile trace elements in MW-poor garnet domains (see Sect. 5.3). In contrast, in MW-rich domains (up to 300 ppm) only a correlation between structural H₂O and Li (Fig. 9) was found. Because Li is fluid-mobile and an indicator for post-peak fluid influx (e.g., Paquin & Altherr, 2002; Scambelluri et al., 2006), its correlation with (part of) structural water is a further indication that part of the latter is most probably secondary.

As a consequence, the primary content of structural water can only be inferred from domains with very little or no molecular water bands. For Erzgebirge samples, this content is only 3–68 ppm (sample averages), but even these low amounts may be overestimated in some cases. Judging from xenolithic garnet in which the type-II band is absent in domains without molecular water (Schmädicke et al., 2015), the OH substitution in garnet that causes band II seems to be governed solely by molecular water. This being the case, all structural H₂O related to band-type II should be secondary.

In conclusion, it is possible that molecular water also influenced garnet volumes without type-M bands, simply by fluid influx occurring in neighboring domains (Gose & Schmädicke, 2018). If true, the primary content of structural H₂O could be smaller than the above given maximum value of 68 ppm for EG samples. The frequency distribution for structural H₂O revealing a distinct maximum at only a few ppm (Fig. 7) points in the same direction. In addition, the high variability of structural and molecular water even in single garnet crystals from EG samples is indicative of grain-scale disequilibrium due to secondary processes and testifies to incorporation of a considerable amount of structural water due to post-peak fluid infiltration.

Experimental data help to further constrain the genetic linkage of the two types of structural water and to corroborate the above reasoning. In natural, impure pyrope the bands at 3650 cm⁻¹ (our type I) and 3600–3630 cm⁻¹ (our type II) differently respond to pressure (Lu & Keppler, 1997). Although the total H₂O solubility positively correlates with pressure, only the intensity of band I grows with pressure whereas the size of band II decreases (Lu & Keppler, 1997). The authors suggested that the OH incorporation mechanism behind band II (i.e., the hydrogarnet substitution; e.g., Geiger et al., 1991) is not favored by high pressure, in accordance with the substitution-induced lattice increase. This agrees with our finding of a correlation of band-type II (but not type I) with molecular water (Fig. 6) and supports the interpretation that most or even all of the structural water inducing band-type II is secondary. Evidently, fluid influx during decompression gave rise to fluid inclusions and led to the uptake of much more structural water than present in the garnet lattice at metamorphic peak conditions. If all structural water, inducing band II, is secondary, the amount of primary water in garnet could be as low as 0–11 in Erzgebirge GBU and 0–1 ppm in Alpe Arami peridotite.

The deduced incorporation of secondary structural water in garnet has two important implications: (i) Garnet was originally water-deficient and hosted less water than possible at the P–T conditions given, simply because garnet saturated with structural water would be unable to take up additional water in its crystal structure, (ii) the genetic link between molecular water, present in fluid inclusions, and structural water related to the hydrogarnet substitution (band II) point to fluid influx during decompression.

Having inferred that part of the structural water in garnet from Erzgebirge samples is of secondary origin, the question remains if the content specified here as “primary” indeed reflects the original amount of water at peak metamorphism or if it was reduced by water loss. Water loss may occur during uplift simply because the incorporation of structural water is pressure-enhanced in most mantle NAMs (e.g., Kohlstedt et al., 1996; Lu & Keppler, 1997; Withers et al., 1998) and the diffusivity of hydrogen in those minerals, depending on the mineral species and its composition, is generally high (e.g., Stalder & Skogby, 2003; Blanchard & Ingrin, 2004; Demouchy & Mackwell, 2006; Koch-Müller et al., 2007). However, it was also shown that hydrogen diffusion can be as slow as that of other cations, depending on the OH substitution type and oxygen fugacity (e.g., Padrón-Navarta et al., 2014; Reynes et al., 2018). In dehydration experiments using grossular and spessartine (Reynes et al., 2018), hydrogen filling Si vacancies was preserved in crystal cores even at 1050 °C whereas bands attributed to iron-oxidation-related point defects readily disappeared, particularly at high oxygen fugacity. Hence, H bound by the hydrogarnet substitution should be preserved during decompression. This applies, at least, to band type II in our samples (see above) whereas the nature of band type I is uncertain.

Given that tectonic exhumation is much slower than the upward transport of xenoliths, water loss should preferably affect the NAMs of orogenic peridotite with the consequence that their water content should be much lower than that of xenolithic equivalents. Surprisingly, this is not the case; the H₂O content of xenolithic garnet (Bell & Rossman, 1992; Snyder et al., 1995; Matsyuk et al., 1998; Grant et al., 2007; Katayama et al., 2011; Schmädicke et al., 2015) is similar or even lower compared to our samples. The same finding was reported for eclogite (crustal type versus xenolith; Schmädicke & Gose, 2017). The authors argued that crustal eclogite not only records lower peak temperature (≤900 °C in most cases) than xenoliths (≥1000 °C) but also does progressively cool during exhumation so that its average exhumation temperature is several hundred degrees lower compared to xenoliths. Accordingly, the authors suggested that crustal eclogite could even be capable to better preserve original water contents from mantle depths than xenolithic eclogite. This conclusion fits well with the data of the present study, which may be interpreted in the same way, i.e., orogenic peridotite may actually lose less water during exhumation than rapidly uplifted xenoliths.

The experimental data by Lu & Keppler (1997) on Mg-rich, natural garnet from Dora-Maira are useful to evaluate the probability of this hypothesis. Although Dora-Maira garnet (Chopin et al., 1991) contains less Fe and Ca than our samples, both are pyropic and, most importantly, they share the same types of OH substitution as indicated by identical IR absorption bands. According to the experimental results, the maximum amount of water storable in garnet is identical for 30, 25, and 20 kbar (i.e., 50 ppm). A reduction of the storage capacity is restricted to pressures of ≤15 kbar. Nonetheless, at 15 kbar garnet is still able to host 75% of the original amount of water incorporated at 30 kbar. As a consequence, garnet in peridotite with some 30 kbar peak pressure should not dehydrate during the first third or even half of the decompression path. The P–T path for Erzgebirge UHP rocks shows considerable cooling during decompression and indicates a temperature of about 600 °C at 15 kbar and 500 °C at 10 kbar (Schmädicke, 1994; Schmädicke & Müller, 2000). At these low temperatures, diffusional water loss from mm- to cm-sized grains is limited.

Moreover, the observed correlation between primary structural water (excluding domains with molecular water) and the concentrations of major (e.g., Mg and Ti) and minor elements (e.g., REE) in garnet is difficult to explain if water loss was significant. An additional, and probably the most convincing argument against severe water loss, is the uptake of secondary structural water during decompression. As pointed out above, all of our samples were originally water-deficient (and many still are) and hosted less water than possible at peak conditions, because otherwise they would not have been able to incorporate secondary structural water. By the same token, if we assume that decompressional water loss initially occurred during exhumation, how can it happen that garnet incorporates secondary structural water during continued exhumation under conditions that are less favorable (i.e., lower P) compared to the previous stage of supposed water loss? Apart from these arguments, the observed correlation between structural H₂O and garnet composition (Fig. 8) cannot be reconciled with water loss either, particularly because the correlation includes relatively immobile elements such as Ti and HREE. Thus, the only justifiable conclusion that can be reconciled with all observations is that the H₂O contents of garnet domains without molecular water are close to the primary concentrations attained at peak pressure and that these low amounts indicate water deficiency.

The studied samples of Erzgebirge peridotite and pyroxenite are characterized by low contents of structural water in garnet (i.e., SW I + II, 3–68 ppm; SW I only, 0–11 ppm), which are similar to those obtained for Alpe Arami peridotite but lower than those of associated coesite eclogite (i.e., 50–180 ppm; Gose & Schmädicke, 2018). Given that the incorporation of water in garnet is favored by its Ca content (Rossman & Aines, 1991; Beran & Libowitzky, 2006), the difference in H₂O between Erzgebirge GBU and eclogite may simply be a function of garnet composition. In any case, the primary H₂O content inferred for garnet in Erzgebirge GBU (i.e., 0–11 ppm) must be lower than its storage capacity, because otherwise it would not have been able to incorporate additional (secondary) structural water. This reasoning is corroborated by experimental data showing that pyropic garnet may incorporate more water (i.e., at 20–30 kbar: 50 ppm; Lu & Keppler, 1997; 140–430 ppm; Withers et al., 1998) than our samples. Hence, the primary H₂O contents of both Erzgebirge and Alpe Arami GBU testify to severe water-deficiency. Because our results are not compatible with significant water loss of garnet, the low H₂O concentrations are interpreted as a primary feature attained more or less at peak metamorphic conditions.

The incorporation of secondary water in Erzgebirge GBU is attributed to exhumation-related fluid influx, which was already proposed for EG eclogite by Gose & Schmädicke (2018). Since the GBU- and eclogite-bearing UHP unit was exhumed during ongoing subduction (Gose & Schmädicke, 2018), dehydration reactions in footwall unit(s) are the most likely source of water. Inhomogeneous distribution of secondary water (both structural and molecular) at grain scale points to local interaction with fluid and to grain-scale disequilibrium. The finding that the major- and trace-element composition (except Li) of garnet was obviously not modified suggests that melt was probably not involved (e.g., Spandler et al., 2004) and further indicates influx of almost pure H₂O. Most probably, fluid influx occurred during the same, early stage of exhumation as in EG coesite eclogite (Gose & Schmädicke, 2018), at which the P–T conditions were still beyond the stability fields of most hydrous minerals (ca. 25–30 kbar, >800 °C). If true, the content of structural H₂O in garnet domains with fluid inclusions should be equal to the H₂O solubility at these conditions. This amount, however, is variable because equilibrium is not attained, but one might argue that the maximum content (i.e., 100 ppm in four of five EG peridotite samples) is closest to equilibrium conditions. This amount is somewhat higher compared to the result of HP experiments on Dora-Maira garnet (i.e., 50 ppm; Lu & Keppler, 1997) but, taking the difference in Ca and Fe content (see above) into account, the contents well agree.

In the following, we explore whether the new results can be used to shed more light on the petrogenesis of the Erzgebirge GBU, which were interpreted as fragments of the mantle wedge by Schmädicke & Evans (1997). However, due to the unknown P–T evolution prior to peak metamorphism, an ocean-floor origin cannot be excluded. Is it possible that the newly obtained data provide information useful in this context? In particular, does the water content of garnet that is present at peak metamorphic conditions depend on the type of protolith? Unfortunately, relevant literature data on water in orogenic garnet peridotite for which the protolith setting is known are extremely rare. Presumably, the mantle wedge should generally be rich in water, due to more or less continuous supply by successive dehydration reactions taking place in the underlying subducting slab. This assumption is confirmed by water contents in lavas and xenoliths and indicated indirectly by viscosity data suggesting water contents close to water saturation for the mantle above the Farallon slab (Dixon et al., 2004). Other direct evidence from xenoliths shows that water distribution in the mantle wedge is heterogeneous (e.g., Peslier et al., 2002, 2012; Soustelle et al., 2010; 2013; Hao et al., 2016). Of course, different portions of the wedge are expected to contain different amounts of water, depending on the amounts supplied by the dehydrating slab and extracted in the course of partial melting. In addition, at depths of <100 km, a significant amount of water may fractionate into calcic amphibole. If re-fertilization results in elevated LILE contents in the mantle wedge, enabling formation of phlogopite, water can be stored in a hydrous mineral also at depths exceeding the stability field of calcic amphibole (ca. >100 km; e.g., Pawley & Holloway, 1993). Though the content and partitioning of water in NAMs coexisting with hydrous minerals is not well known, data from EG quartz eclogite (hydrous minerals stable) versus coesite eclogite (no hydrous minerals) suggest that only little water is incorporated in NAMs if they coexist with prograde hydrous minerals (Gose & Schmädicke, 2018).

If the investigated GBU were originally part of the mantle wedge, they should be derived from its lowermost part, which is directly overlying the subducting slab and, thus, should be the most hydrated portion of the wedge. Accordingly, the NAMs in such rocks are expected to be rich in water, especially if calcic amphibole is not part of the peak assemblage as is the case in the investigated EG samples. If this assumption is correct, the low water concentrations of our samples render it unlikely that they were originally part of the mantle wedge.

Alternatively, our samples could have formed as abyssal peridotite. Unfortunately, information on HP abyssal peridotite is very rare. Garnet peridotite from Cima di Gagnone is – to our knowledge – the only example of HP metamorphic abyssal peridotite (Evans & Trommsdorff, 1978; Evans et al., 1979) for which the content of water in garnet was determined (Padrón-Navarta & Hermann, 2017). The Cima di Gagnone rocks belong to the same unit as the Alpe Arami samples, and their water content in garnet (6 ppm) is equally low as that of the Erzgebirge and Alpe Arami GBU samples. Although it remains unknown if such low contents are a characteristic feature of HP abyssal peridotite, there are two possibilities that would plausibly explain the data. First, the protoliths were not affected by ocean-floor metamorphism and the dry low-pressure mineral assemblage was metastably preserved during subduction and subsequently transformed to a dry HP assemblage. Second, garnet failed to incorporate the water liberated by dehydration of variably serpentinized ocean-floor peridotite. The first option can be excluded, at least, for Gagnone peridotite, which was formed from serpentinite (Evans & Trommsdorff, 1978; Scambelluri et al., 2014). The second possibility was inferred for EG coesite eclogite, which, prior to peak metamorphism, hosted a considerable amount of prograde zoisite and calcic amphibole (up to 10 vol%; Gose & Schmädicke, 2018). However, garnet and omphacite were unable to incorporate the water released by decomposition of hydrous minerals due to UHP metamorphim (Gose & Schmädicke, 2018). The authors suggested that the failure of NAMs to re-equilibrate with H₂O could be due to fast escape of water from the reaction site. If true, HP metamorphism of water-rich protoliths may result in dry HP metamorphic equivalents.

In conclusion, if the low water content of garnet in Gagnone peridotite (Padrón-Navarta & Hermann, 2017) is a typical feature of an abyssal protolith setting, the studied examples of Erzgebirge and Alpe Arami GBU may also be derived from the ocean floor. Admittedly, since this hypothesis is solely based on H₂O concentrations in garnet it remains currently speculative and is difficult to prove. Nonetheless, the similarity of EG and AA samples with Gagnone peridotite (Scambelluri et al., 2014) is not only restricted to H₂O in garnet but extends to its major- and trace-element composition. The samples from all three localities have almost identical normalized REE patterns that are devoid of signs associated with melt-mediated metasomatic overprint, such as medium-REE humps or sinusoidal curves (e.g.,Simon et al., 2007; Gazel et al., 2011) – features which are to be expected if rocks resided in the mantle wedge at greater depth for an extended period of time providing for high re-equilibration temperature and eventually interaction with silicate melts. Thus, until proven otherwise, we tentatively suggest that the Erzgebirge garnet-bearing ultramafic rocks initially formed in an abyssal setting at the ocean floor.

Christian Chopin has made fundamental contributions to our present knowledge of the interplay between metamorphism, petrology and mountain building. It is our great pleasure to contribute our work to the special issue published by the European Journal of Mineralogy in honour of Christian Chopin’s achievements in the past decades. Christian, who is held in highest regard by us, invited E.S. to her first international conference at the EUG in Strasbourg in 1991. Being an invited speaker at the – then – groundbreaking session on high-pressure metamorphism had a severe influence on her scientific future and was one of the key events that shaped her career and research interests. Christian – many thanks for that and all the very best for your future!

Thoughtful and constructive reviews by Jörg Hermann (Bern) and José Padrón-Navarta (Montpellier) are gratefully acknowledged; their suggestion helped to strengthen this contribution. Thomas Armbruster (Bern) is thanked for his great effort with editorial handling. We are also indebted to Roland Stalder (Innsbruck) for providing the analytical facilities (IR) and to Helene Brätz (Erlangen) for help with the ICP analyses. Funding by Deutsche Forschungsgemeinschaft (grant Schm1939/9-1) is gratefully acknowledged.