The ~5 km3, 4.54–4.09 Ma Caspana ignimbrite of the Altiplano-Puna volcanic complex (APVC) of the Central Andes records the eruption of an andesite and two distinct rhyolitic magmas. It provides a unique opportunity to investigate the production of silicic magmas in a continental arc flare-up, where small volumes of magma rarely survive homogenization into the regional magmatic system that is dominated by supereruptions of monotonous dacitic ignimbrites.

The fall deposit and thin flow unit that record the first stage of the eruption (Phase 1) tapped a crystal-poor peraluminous rhyolite. The petrological and geochemical characteristics of Phase 1 are best explained by partial melting of or reheating and melt extraction from a granodioritic intrusion. Phase 2 of the eruption records the emplacement of a more extensive flow unit with a crystal-poor, fayalite-bearing rhyolite and a porphyritic to glomeroporphyritic andesite containing abundant plagioclase-orthopyroxene-Fe-Ti oxide (norite) glomerocrysts. The isotopic composition of Phase 2 is significantly more “crustal” than Phase 1, indicating a separate petrogenetic path. The mineral assemblage of the noritic glomerocrysts and the observed trend between andesite and Phase 2 rhyolite are reproduced by rhyolite-MELTS–based models.

Pressure-temperature-water (P-T-H2O) estimates indicate that the main (Phase 2) reservoir resided between 400 and 200 MPa, with the andesite recording the deeper pressures and a temperature range of 920–1060 °C. Rhyolite phase equilibria predict an estimated temperature of ~775 °C and ~5 wt% H2O. Pressures derived from phase equilibria indicate that the rhyolite was extracted directly from the noritic cumulate at ~340 MPa and stored at slightly shallower pressures (200–300 MPa) prior to eruption. The rhyolite-MELTS models reveal that latent-heat buffering during the extraction and storage process results in a shallow liquidus during the extensive crystallization that produced a noritic cumulate in equilibrium with a rhyodacitic residual liquid. Spikes in latent heat facilitated the segregation of the residual liquid, creating the pre-eruptive compositional gap of ~16 wt% SiO2 between the andesite and the Phase 2 rhyolite.

Unlike typical Altiplano-Puna volcanic complex (APVC) magmas, low fO2 conditions in the andesite promoted co-crystallization of orthopyroxene and ilmenite in lieu of clinopyroxene and magnetite. This resulted in relatively high Fe concentrations in the rhyodacite and Phase 2 rhyolite. Combined with the co-crystallization of plagioclase, this low oxidation state forced high Fe2+/Mg and Fe/Ca in the Phase 2 rhyolite, which promoted fayalite stability. The dominance of low Fe3+/FeTot and Fe-Ti oxide equilibria indicates low fO2 (ΔFMQ 0 − ΔFMQ − 1) conditions in the rhyolite were inherited from the andesite.

We propose that the serendipitous location on the periphery of the regional thermal anomaly of the Altiplano-Puna magma body (APMB) permitted the small-volume magma reservoir that fed the Caspana ignimbrite eruption to retain its heterogeneous character. This resulted in the record of rhyolitic liquids with disparate origins that evaded assimilation into the large dacite supereruption-feeding APMB. As such, the Caspana ignimbrite provides a unique window into the multi scale processes that build longlived continental silicic magma systems.

Discerning the development of large continental silicic magma systems is central to understanding the origin and evolution of the continental crust. Many such magma systems are dominated by “monotonous intermediate” bulk compositions (Hildreth, 1981; Best et al., 2016). These dacites or quartz latites consist of rhyolitic melts with high crystal contents (35%–60%) and are commonly understood to be the inevitable products of longlived thermochemical and thermomechanical histories that produce buffered, homogenized compositions (de Silva and Gregg, 2014; Caricchi and Blundy, 2015; Best et al., 2016). Powered and maintained by the often invisible hand of mafic recharge (e.g., Hildreth, 1981), it is clear that such long-lived systems have episodic and incremental histories (Coleman et al., 2004; de Silva and Gosnold, 2007; Lipman and Bachmann, 2015) that may be blurred and homogenized if the magmatic flux is high enough to promote homogenization over heterogeneity, particularly during flare-up conditions in continental arcs (e.g., de Silva et al., 2006; Huber et al., 2009; Best et al., 2016).

The extensive ignimbrite plateau of the Altiplano-Puna volcanic complex of the Central Andes (APVC; de Silva, 1989a) is the surface expression of protracted, focused volcanism that was generated during a period of high mantle flux (de Silva et al., 2006). This archetypal ignimbrite flare-up fostered the geophysically and petrochronologically imaged residual “batholith” known as the Altiplano-Puna magma body (APMB; Chmielowski et al., 1999; de Silva and Gosnold, 2007; Kern et al., 2016; Pritchard et al., 2018). Bulk compositions outside of 66–69 wt% SiO2 on the APVC typically make up a few percent of the total erupted magma (de Silva, 1989b; Lindsay et al., 2001b; Schmitt et al., 2001; de Silva et al., 2006; Grocke et al., 2017a), but they provide valuable insights into the behavior of their magmatic reservoirs and the magmatic history of the APVC as a whole. Rhyolites in the APVC are dominantly derived by crystallization of parental magmas that are represented by less felsic compositions in their eruptive sequences, and they have geochemical compositions that are dominated by assimilated continental crust. These mechanisms of evolved melt production typically cause the APVC rhyolites to have steeper rare-earth element (REE) patterns than their parental magmas (i.e., clinopyroxene and amphibole fractionation) and “crustal” isotopic signatures. These geochemical characteristics provide valuable insight into the variety of melts that ultimately accumulate into and segregate from this large continental silicic magmatic complex.

Strongly contrasting the typical APVC ignimbrite is the ~5 km3 Caspana ignimbrite, which crops out near the periphery of the APVC (Fig. 1; de Silva, 1991). The ignimbrite is notably heterogeneous with two distinct rhyolites and an andesite found in outcrop, connoting a physical storage condition that is unlike the typical “monotonous intermediate” reservoirs that evacuated the large-volume, crystal-rich dacites (de Silva and Wolff, 1995; Huber et al., 2012; de Silva and Gregg, 2014; Black and Andrews, 2020). Petrologically, the Caspana system defies the oxidized state that is typical of APVC dacites and, indeed, arc magmas in general (Kelley and Cottrell, 2009; Burns et al., 2020), because it was appropriate for a fayalite rhyolite and an andesite-bearing noritic glomerocrysts. These unique features prompted the petrologic and geochemical study presented here; the study captures the magmatic processes and physical storage conditions of the Caspana magmatic reservoir. The reconstruction of the Caspana system and its compositional gaps provides a new lens with which to investigate rarely preserved processes contributing to the massive APVC eruptions.

2.1 Geologic Background and Prior Work

Ignimbrites in the APVC are generally large-volume, monotonous dacites to rhyodacites that formed by the combination of crystal fractionation and crustal contamination (i.e., assimilation and fractional crystallization [AFC]) (de Silva, 1989a, 1989b; de Silva and Francis, 1989; Kay et al., 2010; Grocke et al., 2017a). In the Neogene, between ca. 25–10 Ma, a relative flattening of the subduction angle occurred as the aseismic Juan Fernandez ridge was progressively subducted southward beneath the South American plate. This was followed by subsequent rollback on the subducting Nazca plate, resulting in arc-scale delamination of subcontinental lithospheric mantle (SCLM) and the ignition of a Central Andes–wide ignimbrite flare-up (Kay and Coira, 2009; Freymuth et al., 2015; Best et al., 2016; de Silva and Kay, 2018). In the ~21° to ~24°S segment of the arc, a crustal-scale magmatic complex led to the development of an incrementally constructed regional batholith (de Silva and Gosnold, 2007; Salisbury et al., 2011; Kern et al., 2016), which is now detected as a seismic low-velocity zone known as the Altiplano-Puna magma body (APMB; Chmielowski et al., 1999; Prezzi et al., 2009; Ward et al., 2014; Pritchard et al., 2018). With an estimated depth range of 10–30 km and volume of >500,000 km3, the APMB is interpreted as the parental source of the voluminous supereruptions of the APVC that ultimately erupted from upper-crustal silicic magma chambers. Explosive activity during the Neogene ignimbrite flare-up (de Silva et al., 2006; Kern et al., 2016) occurred in distinct pulses with peak episodes at ca. 8, 6, and 4 Ma. Since 4 Ma, volcanism in the APVC region appears to have returned to steady-state (i.e., background) activity (Burns et al., 2015; Tierney et al., 2016).

The APVC ignimbrites record a time of prodigious crustal magmatism when batholithic volumes of monotonous crustal magmas were the norm. In this context, the small-volume (~5 km3) Caspana ignimbrite with its strongly heterogeneous character stands out, particularly since it erupted during the last peak of the flare-up (Kern et al., 2016). Prior case studies have found that many of the highsilica magmas that erupted onto the APVC are created dominantly by fractionation from the large-volume dacites or andesites in the region (de Silva, 1991; Lindsay et al., 2001a; Schmitt et al., 2001; Grocke et al., 2017a). Isotopic compositions of these rhyolites and their parental magmas record significant crustal assimilation, with a 50:50 mix of mantle and regional basement compositions generally agreed upon (de Silva, 1989a; Aitcheson et al., 1995; Mamani et al., 2008, 2010; Kay et al., 2010). Importantly, the crystal cargo in all of these magmas records a high oxidation state and unchanging ƒO2 within a given magmatic lineage (Grocke et al., 2016; Burns et al., 2020). For illustrative purposes and to emphasize the differences in rhyolite petrogenesis, the compositions of the low-silica rhyolites found in the Tara ignimbrite (Grocke et al., 2017a) and the high-silica rhyolites of the Alota ignimbrite (Salisbury et al., 2011; Kaiser, 2014) will be shown with the Caspana geochemical data.

2.2 The Caspana Ignimbrite

The Caspana ignimbrite crops out in the Toconce-Caspana area of N. Chile (de Silva, 1989b; de Silva, 1991; Figs. 1 and 2), and the age of the eruption is bracketed stratigraphically between the 4.09 Ma Puripicar ignimbrite and the 4.54 Ma Linzor I ignimbrite. Its source vent(s) is/are thought to be buried beneath the younger Toconce and Leon volcanoes. de Silva (1991) first described the bimodal andesitic and rhyolitic juvenile clasts found in the ignimbrite, defining a large compositional gap. On the basis of reconnaissance bulk and mineral chemistry, an origin of the rhyolite by fractional crystallization of the andesite was proposed to have led to a small bimodal, zoned magma chamber.

We have resampled and reexamined the same exposures and sections introduced in de Silva (1991). The northern outcrops above the community of Toconce contain a rhyolitic plinian fallout of nearly aphyric pumice clasts with occasional phenocrysts of plagioclase visible in hand specimen (Section B, Fig. 2). There is a fine ash deposit on top of the fallout; this deposit is in turn overlain by a distinct ~10–40 cm flow unit that contains equally aphyric rhyolite. This sequence is collectively referred to as Phase 1. Above this sequence lie several meters of massive ignimbrite containing the rhyolitic and andesitic pumice clasts described by de Silva (1991). These pumice clasts are referred to herein as Phase 2. At the clast-rich flow front, rhyolite and andesite pumices are largely mixed together with only hints of any internal stratigraphy (Section A, Fig. 2) and overlie a basal ash that is equivalent to the basal plinian (Section B, Fig. 2). Rhyolitic pumice in Phase 2 are distinct from those in Phase 1, because they have relatively higher crystallinity (~3%–5%) and are substantially less fragile in hand sample. Phenocrysts in the Phase 2 rhyolite include plagioclase, biotite, and occasional yellow-green to amber-colored olivine. Andesitic pumice in the Phase 2 ignimbrite has variable crystallinity from sample to sample that ranges from 20%–45%. Hand samples of andesite pumice clasts contain phenocrysts of plagioclase, orthopyroxene, and FeTi oxides.

3.1 X-Ray Fluorescence and Inductively Coupled Plasma–Mass Spectrometry (ICP-MS)

Whole-rock samples with the smallest amount of visible oxidation possible were selected for analysis. Those that contained oxidized surfaces were sawed or chipped off at Oregon State University (OSU) to expose the innermost fresh face possible. Samples were crushed using steel plates in the jaw crusher and ground to a fine powder at OSU. The samples were then processed at Washington State University (WSU) using a ThermoARL AdvantXP for X-ray fluorescence (XRF) analysis of major and some trace elements (high field strength elements [HFSEs]) following the method of Johnson et al. (1999). REE and remaining trace elements were measured by ICP-MS on an Agilent 7700 Q-ICP-MS. Loss on ignition (LOI) is high for some samples. A subset of samples was re-run after washing in 1 molar HCl at OSU and then sonicated and rinsed three to five times. The process was repeated until bubbles forming from reaction were no longer present; usually on the first wash. Results from XRF analysis on the digested samples show no systematic variation of CaO content from sample to sample (i.e., some are higher, some are lower) in the andesite, and there is no change in rhyolite compositions. Onesigma (1σ) error bars of analysis from the tables of Johnson et al. (1999) are shown on major-element graphs. For the whole-rock error on trace elements, the quoted analytical precision of U.S. Geological Survey (USGS) standards is 5% relative standard deviation (RSD) for the REE and less than 10% for all other trace elements. Repeat samples from four XRF runs from the WSU lab over the past few years have better precision than these reported values. Representative data from the Caspana ignimbrite are given in Table 1, with the full data set in Table S11. Data measured by XRF from de Silva (1991) are denoted in tables where presented.

3.2 Thermal Ionization Mass Spectrometry (TIMS)

Six HCl-digested samples (as in section 3.1) were analyzed for 87Sr/86Sr, 143Nd/144Nd, 208Pb/204Pb, 207Pb/204Pb, and 206Pb/204Pb analysis by thermal ionization mass spectrometry (TIMS) at New Mexico State University using the analytical methods highlighted in Ramos (1992). Analytical uncertainty is 0.000012, 0.001, 0.001, 0.002 for 87Sr/86Sr, 208Pb/204Pb, 207Pb/204Pb, and 206Pb/204Pb (National Bureau of Standards [NBS] 987 standard) and 0.00001 for 143Nd/144Nd (La Jolla standard). Representative data are given in Table 1, with the full data set in Table S1 (see footnote 1).

3.3 Electron Probe Micro-Analysis (EPMA)

EPMA analysis was carried out in the Stanford Microchemical Analysis Facility (MAF) at Stanford University on a JEOL JXA-8230 SuperProbe. Major- and minorelement abundances in the silicate minerals (i.e., feldspar, pyroxene, fayalite, and biotite) were analyzed using an accelerating voltage of 15 keV, a 20 nA probe current, and a 3 µm spot size. On-peak count times ranged from 10 to 60 s and were optimized to achieve the desired counting statistics. Majorelement concentrations in the matrix glasses were measured using conditions similar to those used for the silicate minerals, except the spot size was increased to 10 µm to minimize alkali migration. In addition, Na migration was monitored during analyses, and time-dependent intensity, corrections were applied when applicable. In order to gain detailed information on minor and volatile element abundances (P, Fe, Mn, Ti, Cl, and S) in the glasses, a second set of measurements were made at the same location(s) at higher probe currents and longer count times. This significantly decreases analytical uncertainties, reduces detection limit, and increases precision. Oxide phases were analyzed using a 20 keV accelerating voltage, 20 nA probe current, and focused 1 µm spot. On-peak count times ranged from 20 to 60 s. Full data sets are available in Tables S2–S7 (footnote 1).

3.4 Laser Ablation–Inductively Coupled Mass Spectrometry (LA-ICPMS)

LA-ICPMS was conducted at the Keck Collaboratory at OSU for trace-element concentrations of silicate glasses that were mounted in epoxy and cleaned in an ultrasonic bath in ethanol and then deionized water. Methodology closely follows that outlined in Kent and Ungerer (2006). 43Ca measured on BCR-2G was used as an internal standard, and ATHO-G was run after every 15 analyses to check for consistency (e.g., instrument drift and clean lines). To reduce surface contamination, samples were initially ablated for ~3 seconds with a 160 μm spot and given a brief washout period before measuring. Count times, dwell times, and the background interval for each analysis (taken prior to ablating) are 30, 0.01, and 12 seconds respectively. A 30 second washout time was used after each ablation period. The average 1σ values of all analyses are shown on trace-element plots, and errors are propagated where trace-element ratios are shown. Full data sets are available in Tables S2–S7 (footnote 1).

4.1 Petrography

4.1.1 Andesite Pumice

Andesite pumice clasts in the Caspana ignimbrite are moderately crystalline (25–45 vol%), with a phase assemblage consisting of plagioclase (75%–80%), enstatite (15%–20%), and oxides (1%– 5%) in a groundmass of well-vesiculated glass with ellipsoidal vesicles (~50%) (Table 2). Plagioclase is the dominant crystalline phase and occurs in a range of sizes (~100 µm–1.5 mm) with an average size of ~1 mm. Texturally, plagioclase defines a continuum of textures ranging from clear, concentrically zoned crystals with sharp rims, to crystals that are pervasively sieved (Fig. 3). There are two texturally distinct populations of enstatite in the andesite that can be easily differentiated by their crystal shapes and mineral inclusions (P1 and P2). P1 crystals range in size from ~0.5–1.5 mm, are more rounded, and have significantly more Fe-Ti oxide inclusions than P2 crystals. P2 crystals are roughly similar in size but are euhedral and contain minimal Fe-Ti oxide inclusions. The andesite contains both ilmenite and magnetite. Ilmenite is far more abundant than magnetite and can occur as both a phenocryst and micro pheno cryst within the groundmass and within enstatite. In contrast, magnetite only occurs as micro pheno crysts typically in enstatite or within glomerocrysts and is always exsolved. There are also two distinct types of glomerocrysts present in the andesite. One type (G1) is strictly plagioclase, whereas the other type (G2) is composed of plagio clase (Pl) + orthopyroxene (Opx) + ilmenite (Ilm) + magnetite (Mag). G1 glomerocrysts are more abundant and contain large, tabular, concentrically zoned plagioclase. G2 glomerocrysts are dominantly orthopyroxene in the presence of more lath-like plagioclase than those found in G1 glomerocrysts. Ilmenite crystals are generally larger in G2 glomerocrysts, and orthopyroxene can be heavily rounded at the edges. Amphibole is rare in the andesite pumice clasts, occurring as a single phenocryst and as a single inclusion in orthopyroxene. The andesite pumice clasts also contain quartzofeldspathic xenoliths.

4.1.2 Rhyolite Pumice

There are two types of rhyolite pumice in the Caspana ignimbrite (herein referred to as Phase 1 and Phase 2), and both are crystal poor (<1%–5% vol% crystals). Phase 1 is the less crystalline of the two (±1%). Phase 1 pumice in the plinian fallout can be entirely aphyric or contain <1% crystals by volume. Pumice that occurs in the flow unit just above the fallout has a crystallinity of ~1%. Feldspar is the most abundant mineral (~95%) in the Phase 1 rhyolite and frequently displays sharp rims and distinct, somewhat infrequent zoning boundaries. Microphenocrysts of oxides can be observed in thin section and mineral separates as well as muscovite and accessory zircon and titanite. Phase 1 rhyolite also contains small quartzofeldspathic xenoliths that contain quartz, feldspar, amphibole, oxides, ± pyroxene, and anhedral, micaceous material that bears semblance to restite. The quartzo feld spathic xenoliths are similar to those in the andesite. As previously stated, Phase 2 pumice clasts are more crystalrich (3%–5%), containing plagioclase (80%), fayalite (5%), biotite (10%), and ilmenite (3%) with accessory apatite and zircon. Phenocrystsize allanite can be observed in thin section as well. Magnetite is present but rare (<1%), and quartz was found in mineral separates.

4.2 Whole-Rock Major and Trace Elements

Bulk-rock analyses of the Caspana pumice (Tables 1 and S1, see footnote 1) show that the system has clear calc-alkaline affinities. The three pumice types define three distinct compositional groups along a high-K calc-alkaline trend (Fig. 4) with a large compositional gap between 60 and 74 wt% SiO2. Rhyodacitic glass from the andesite (66%–68% SiO2) generally lies on a distinct trend between andesite and Phase 2 composition in both major- and trace-element space (Figs. 4 and 5). The rhyodacite defines the termination of the trend of andesite pumice samples in FeO relative to MgO (Fig. 5C). The FeO contents of the rhyodacite also exceed the rest of the APVC bulk rock composition samples that have a comparable amount of MgO content.

The Phase 1 rhyolite displays slightly lower SiO2 and lower alkali concentrations than the Phase 2 rhyolite. Using Shand’s index, the Phase 1 rhyolites are strongly peraluminous (A/CNK ~1.25, A/NK ~1.59), whereas the Phase 2 rhyolite and andesite are metaluminous to slightly peraluminous (Table 1), similar to other APVC magmas. By normalizing FeO with other major elements (i.e., CaO and MgO; herein referred to as Fe-indices), the Phase 2 rhyolite pumice has considerably higher Fe-indices relative to the Phase 1 rhyolite and the rest of the APVC rhyolites (Fig. 5). The only other rhyolite that we know of that approaches these Fe-indices is the extremely evolved Alota-Juvina rhyolite (Salisbury et al., 2011; Kaiser, 2014). Phase 1 glasses (~73–74 wt% SiO2) have Fe-indices (0.75–0.81 FeO/[FeO + MgO]); average 0.79) that are lower than the andesite (0.77–0.85 FeO/[FeO + MgO]); average 0.81) and always lower than the Phase 2 glasses (0.95–0.98 FeO/[FeO + MgO]). The Fe-indices of Phase 1 pumice clasts and glass are commonly observed in APVC rhyolites and rhyodacites.

Trace-element concentrations for the Caspana pumice clasts and glasses presented in Figures 5 and 6 reveal that Rb concentrations in the andesite are significantly lower than in the rhyolites (47–113 ppm). Chondrite-normalized trace-element diagrams show typical arc affinity for all three pumice populations (i.e., Nb-Ta trough, negative Pb anomaly, and enriched large ion lithophile element [LILE]) (Fig. 6), and all three have relatively high light rareearth element to heavy rare-earth element (LREE/HREE) ratios (8.9–14.9). The andesite pumice clasts display either a flat or positive Eu anomaly (Fig. 6). Both rhyolites show negative Eu anomalies, but the anomaly is significantly more pronounced in the Phase 2 pumice clasts. The Phase 2 pumice and the Alota-Juvina have comparable Eu anomalies and LREE/HREE ratios (Fig. 6B). The Eu and Sr concentrations of Phase 1 glasses are within error of the Phase 1 pumice (as is Ba), whereas the Phase 2 glasses are depleted relative to their host pumice (Fig. 5). Phase 2 pumice and rhyodacite glass both have higher concentrations of REE and Y than the Phase 1 pumice. The Phase 2 rhyolite pumice has similar to slightly lower Nb concentration than the rhyodacite, and the two overlap one another in Y (Figs. 5E and 5F).

The Phase 1 rhyolite has the largest Sc depletion and Dy/Dy* anomaly of all other compositions from the Caspana system (Figs. 6 and 7) and is more typical of APVC rhyolites. Interestingly, the crystallinity (mostly feldspar) and the amount of restite material in the Phase 1 rhyolite thin sections decrease down section, but the expected systematic change in trace-element concentrations is not captured by matrix glass nor bulk rock data (i.e., Sr and Eu; Fig. 5). Furthermore, zircon is readily found in the Phase 1 thin sections (and mineral separates), but Zr and Hf concentrations of matrix glass seem to indicate that the Zr saturation temperature was not attained for any prolonged time scale in the collected pumice clasts (Fig. 7).

4.3 Whole-Rock Isotopes

Broadly, the isotope ratios measured in the Caspana pumice are consistent with other ignimbrites in the APVC (Fig. 8) (Lindsay et al., 2001a; Godoy et al., 2014, 2017; Grocke et al., 2017a). These isotopic ratios lie along the trend of a simple AFC model (DePaolo, 1981) between mantle-derived basalt and the most evolved samples within the Sierra de Moreno (SdM) meta morphic complex, which has been proposed as the local basement for this region of the arc (Godoy et al., 2014, 2017). Andesite pumice clasts have 87Sr/86Sr and 143Nd/144Nd ratios that are indistinguishable from those measured in the Phase 2 rhyolite pumice clasts (0.7112 and 0.5121 versus 0.7113 and 0.5121, respectively) (Table 1; Fig. 8A; Table S1, see footnote 1). The compositions are also close to one another in Pb isotope space (207Pb/204Pb–208Pb/204Pb–206Pb/204Pb) (Table 1; Fig. 8; Table S1).

Compared to Phase 2, the Phase 1 rhyolite is significantly less radiogenic in 87Sr/86Sr (0.7081–0.7082) and variable in 143Nd/144Nd (0.5121–0.5123). Phase 1 rhyolite samples are within analytical error of one another in Pb isotopic composition (Fig. 8C). These pumice clasts have slightly lower 208Pb/204Pb and 207Pb/204Pb ratios and slightly higher 206Pb/204Pb ratios than the Phase 2 rhyolite and andesite pumice but are well within the fields defined by other APVC ignimbrites.

4.5 Phase Chemistry

4.5.1 Plagioclase

Plagioclase phenocrysts in the andesite have a moderately broad range of An (An86−76) and little correlation between composition and texture (Figs. 3B, 3C, 3E, and 9; Table 3; Table S3, see footnote 1). The dominant mode in the distribution of all pheno-crysts is at ~An82 (Fig. 9C). This peak is defined by non-zoned phenocryst compositions, normally zoned cores, and reversely zoned rims. Kernel density estimates (KDEs) of core and rim An content from normally and reversely zoned phenocrysts are effectively mirrored distributions, though it should be noted that some normally zoned cores lay at low An content. The FeO concentrations in andesite phenocryst cores and rims have a mode at ~0.24 wt%, though FeO on normally zoned pheno-cryst rims can be skewed up to ~0.35 wt% FeO. Like the phenocrysts in the andesite, G1 glomero-crysts have a tight distribution at ~An82 with a slight left skewness (Fig. 9A). FeO concentrations of G1 plagioclase are non-zoned and low (FeO 0.22–0.26 wt%). G2 plagioclase defines the range of An and FeO contents (An88–76; FeO 0.23–0.53) (Fig. 9B). The distributions of An contents on G2 plagioclase rims are offset to lower An values than their cores and overlap with reversely zoned phenocryst cores, normally zoned phenocryst rims, and the subset of normally zoned phenocryst cores that are present at lower values of An. FeO of the G2 cores (mean ~0.29 wt%) is higher than that of G1 plagioclase and the phenocrysts, has a broader distribution, and is offset slightly to the right.

The Phase 2 rhyolite has two distinct groupings of plagioclase, with two subpopulations in the lower An content group (Fig. 9D). Excluding the high An plagioclase group in the Phase 2 rhyolite that is similar to plagioclase in the andesite, Welch’s two-sample t-test shows that the two apparent subpopulations above and below An35 are indeed two separate populations (p~0.003 and d.f.~10). The first subpopulation has anorthite contents ranging from An43−30 and FeO concentrations from 0.12 to 0.19 wt%. The second subpopulation of plagioclase in the Phase 2 rhyolite has slightly lower An contents but overlapping or higher FeO concentrations (An33−25; FeO: 0.09–0.35 wt%, respectively). The crystals within this latter subpopulation are occasionally microphenocrysts but are more commonly cores to large concentrically zoned crystals (Fig. 3F). Rim compositions of the concentrically zoned crystals overlap the first subpopulation. The other distinct type of plagioclase has high anorthite contents that overlap the andesite compositions. This group of plagioclases have similar FeO compositions to the andesite in the core (avg. 0.20 wt%), but rim FeO concentrations overlap the two low An plagioclases from the Phase 2 rhyolite (~0.14 wt%).

Plagioclases from the Phase 1 rhyolite have An contents slightly higher than those observed in the Phase 2 rhyolite (An44−32; average An39) (Fig. 9E). Most are non-zoned, although normal zoning is also found (Fig. 3; Table S3, see footnote 1).

4.5.2 Orthopyroxene

There are two populations of orthopyroxene (opx) phenocrysts, P1 and P2, present in the andesite pumice (see section 4.1 for details). Data are presented in Tables 4 and S4. Both types of opx plot in the enstatite field (En65−49). However, P1 opx phenocrysts are euhedral (P2 opx pheno crysts are subhedral) and have lower MgO, CaO, and Al2O3 concentrations than P2 opx (Fig. 10). Ortho-pyroxene from glomerocrysts (G2 only) are also enstatite (Fig. 3A). Some orthopyroxenes from the glomerocrysts are compositionally similar to P2 phenocrysts. However, most G2 orthopyroxenes have distinctly higher Mg# with the same Al2O3 concentration of P1 phenocrysts and are thus intermediate between P1 and P2 phenocrysts. There are also orthopyroxene cores of high Al2O3 (~1.7–3 wt%), low CaO (0.7–1 wt%), with equivalent Mg# in glomerocrysts that are omitted in Figure 10 for clarity but will be discussed below. A key observation is that P1 and P2 rims appear to converge on the intermediate compositions recorded in G2 opx glomerocrysts (Fig. 10).

4.5.3 Oxides

Ilmenite and magnetite compositional data are presented in Tables 5 and S5 (footnote 1). In the Caspana Phase 2 andesite, ilmenite is far more abundant than magnetite and occurs as both phenocrysts and microphenocrysts, whereas magnetite occurs strictly as microphenocrysts. There are no clear compositional distinctions between ilmenite phenocrysts and micro pheno crysts. However, ilmenites define two compositionally distinct groups easily differentiated by FeO (reduced with the algorithm of Stormer, 1983), TiO2, and V2O3 concentrations (Fig. S1, see footnote 1). The two groups have different TiO2 and V2O3 concentrations at a given FeO, defining two roughly linear arrays in FeO-TiO2 space. Ilmenite inclusions found in the orthopyroxene belong to both the high- and low-Ti group.

Similar to the andesite, the Phase 2 rhyolite has far more ilmenite than magnetite. Compared to the andesite, these ilmenites are either slightly lower than or equivalent to the TiO2 concentrations of the low-Ti group in the andesite (TiO2 48.3–49.7, average 49.1 wt%) but have substantially more FeO (41.8–43.2, average 42.7 wt%). Magnetites in the Phase 2 rhyolites are typically exsolved. The one un-exsolved magnetite found in Phase 2 has notably high Ti content of ~16%–17% (Usp51). This Ti-magnetite has the highest Ti composition of Ti-magnetite in rhyolite that we have found within the APVC, though we acknowledge that it is only a single grain. Phase 1 is different in that only magnetite was found and contains far less Ti content at Usp16.

4.5.4 Fayalite

The Phase 2 rhyolite is an anomaly from the other lithologies in the Caspana ignimbrite because it contains fayalite (Fig. 3H; Tables 6 and S6, see footnote 1). Fayalite is homogeneous at ~Fa89 (Fe90Mg8Mn2Ca<<1), and we did not observe a textural or chemical relationship between fayalite and other phases in the unit (i.e., overgrowths of opx).

4.5.5 Biotite

Like fayalite, biotite can only be found in the Phase 2 pumice (Fig. 3G). Biotite compositional data are presented in Tables 7 and S7 (footnote 1). Data recalculations were done assuming 22 oxygens and are well into the annite-siderophyllite solid solution field based on the classification of Deer et al. (1992). Caspana biotites are homogeneous with notably high Fe# (~77.5) and TiO2 contents (5–6 wt%). The Fe concentrations are significant when compared to APVC ignimbrites and lavas (Fig. S2). The only unit in the APVC with comparable Fe# biotite is the poorly known 5.23 Ma Alota ignimbrite (Salisbury et al., 2011), which has comparable Fe-indices (Fig. 5). Biotite in the Phase 2 rhyolite has the highest Ti content of all biotite found in APVC ignimbrites (Fig. S2). As pointed out above, the Ti-magnetite in these pumice clasts exhibits the same relationship.

4.6 P-T-H2O-fO2 Constraints

A variety of experimentally and theoretically calibrated phase equilibria models were used to constrain a suite of intensive parameters for the Caspana ignimbrite magmas. These are summarized in Table 8.

4.6.1 Andesite

Storage pressures for the andesite were calculated using the rhyolite-MELTS–based (r-MELTS) plagioclase, pyroxene (± oxides) geobarometer (Harmon et al., 2018) under a range of water contents (4–10 wt%) and oxidation states (Δfayalite-magnetite-quartz [FMQ] − ΔFMQ − 1) using rhyodacite matrix glass as input composition. There are two combinations of pressure, fO2, and H2O content that result in co-saturation of plagioclase and orthopyroxene with acceptably low residual temperatures (8 °C; Table 8). The first is 400–450 MPa (average 430 MPa), occurring at or close to water saturation (8–10 wt%) and at ΔFMQ − ΔFMQ − 0.5. Under these conditions, the crystallization sequence is ilmenite > magnetite > plagioclase + orthopyroxene. The second is approximately normally distributed between 415 and 315 MPa (average 366 Mpa) at undersaturated conditions (4–6 wt% H2O) and at or below the FMQ buffer. The typical crystallization sequence is magnetite > plagioclase + orthopyroxene > ilmenite, but ilmenite occasionally joins magnetite before the equilibrium pair (pl + opx) depending on the glass composition that is used.

Equilibrium temperatures and water contents for the andesite were estimated using Equations 24a and 25b of Putirka (2008). To assure plagioclaseliquid equilibrium, only graphic values between 0.05 and 0.15 for plagioclase rims and rhyodacite matrix glass were used for modeling purposes. Equilibrium temperature and water contents range from 915–956 °C (average 933 °C; 36 °C standard error of the estimate [SEE]) and 6.2–5.0 wt% H2O (average 5.5; 1.1 wt% SEE) when using an input pressure of 430 MPa. Changing the pressure to 350 MPa has a negligible effect on the output temperature and water contents (4 °C and 0.01 wt%, respectively). We also estimated water contents using the more precise Waters and Lange (2015) plagioclase-liquid hygrometer (SEE 0.3 wt%) to verify results of the Putirka (2008) hygrometer. For model inputs, we assumed a pressure of 430 MPa, which is derived from r-MELTS barometry, and temperature from the Putirka (2008) plagioclaseliquid model. Although the Waters and Lange (2015) hygrometer is more accurate than the Putirka (2008) hygrometer and has a smaller SEE, we point out that it slightly underestimates water content compared to direct measurement (see Ulmer et al., 2018). Output water contents are 4.0–5.1 wt% H2O (average 4.3 wt%), and the estimated equilibrium anorthite composition (~An83) agrees with observed compositions (Figs. 9B9D). This equilibrium plagioclase composition is also predicted by r-MELTS models (An81; see below). Interestingly, the water contents for the andesite agree with inferences for water content in mafic arc magmas globally (Kelley and Cottrell 2009; Plank et al., 2013).

The viability of the plagioclase-liquid temperature estimates was tested using the orthopyroxene-liquid thermobarometer of Putirka (2008; Equation 28a and Equation 29b). Temperatures were modeled assuming a pressure of 430 MPa and water contents ranging from 3.5–6.5 wt%. Input compositions included rhyodacite matrix glass for the liquid component and orthopyroxene rims. Temperatures were calculated using Equation 28a, because it yields the highest R2 and lowest SEE when modeling hydrous and lower T systems (i.e., <~1100 °C). Equation 29b was used for independently calculating equilibrium pressures because it relies on the enstatite-ferrosilite ([Fe,Mg]2Si2O6) component rather than the jadeite (NaAlSi2O6) component. This is selected, because in the Caspana system, the Na concentrations in the orthopyroxene approach the analytical detection limits resulting in unacceptably high uncertainties, and glass may be altered. Orthopyroxene-glass equilibrium was verified assuming a graphic of 0.27± 0.3 (Roeder and Emslie; 1970) and is displayed graphically following the methods of Rhodes et al. (1979; Fig. S3, see footnote 1). The orthopyroxenemelt thermobarometer yields average temperatures of 910–941 °C (range of 888–950 °C; SEE: 39 °C) and vary accordingly with the input water contents of 6.5–3.5 wt% H2O. This range of water contents causes temperature to change less than the SEE of the model and is within error of the Putirka (2008) plagioclase-liquid model. The corresponding equilibrium pressure that is output from the Opx-liquid model agrees with the r-MELTS geobarometer with averages of 390–490 MPa (range of 350–550 MPa; SEE: 260MPa), which vary with the input water contents above. Changing the input pressure to 350 MPa causes the pressure and temperature outputs to change by <0.01 MPa and 5 °C. While these pressure ranges are far outside of acceptable constraints, it is important to note that the averages are in agreement with the (arguably) more accurate r-MELTS geobarometer.

4.6.2 Phase 2 Rhyolite

Matrix glass and bulk-rock compositions from Phase 2 pumice were input into the r-MELTS geo-barometer (Gualda and Ghiorso, 2014, 2015) to estimate storage and extraction depth, respectively (Gualda et al., 2019). We exercise caution in presenting these results due to the potential for glass alteration affecting the pressure estimation (section 3.1) (Pamukcu et al., 2015). Extraction pressures calculated using bulk-rock compositions are ~330 MPa using 4 wt% H2O and an oxidation state of ΔFMQ as input, which is just shy of the andesite storage pressure under the same petrologic conditions (~360 MPa). When using matrix glass as input in order to estimate storage pressure of the Phase 2 rhyolite, r-MELTS predicts the observed assemblage at slightly lower pressure (275–222 MPa; average 235 MPa) using the same input water contents and at ΔFMQ − ΔFMQ – 1. Only one glass composition predicts the equilibrium assemblage under fluid-saturated conditions; the output pressure from the model is 200 MPa. Other wise, increasing the water content puts sanidine on the liquidus for both matrix glass and bulk-rock compositions, which is not observed in the Phase 2 rhyolite.

Equilibrium temperatures for the Phase 2 rhyolite were calculated using the olivine-liquid model of Putirka (2008). Equilibrium between olivine rims and matrix glasses were verified visually using the method of Rhodes et al. (1979). Using an input pressure of 250 MPa and water contents between 4 and 6 wt% H2O, calculated temperatures range from 787–805 °C (average 796 °C; SEE 29 °C). Changing the input pressure has a negligible effect on output temperature.

Equilibrium temperatures for the Phase 2 rhyolite were also estimated using compositions of coexisting magnetiteilmenite pairs. As stated above, we were only able to find a single magnetite grain that was not exsolved. Additionally, we acknowledge that the magnetite ilmenite “pairs” are not touching but simply coexisting and thus may not be in equilibrium. However, we attempted to verify this whenever possible by using the equilibrium test of Bacon and Hirschmann (1988) for all possible pairs. We therefore view the temperature and oxidation state calculated by two-oxide equilibria with great speculation. However, when integrated with the petrology of the Phase 2 rhyolite (ferrous-rich assemblage) and independent temperature estimates below, we also believe the results of oxythermometry presented here are informative. Using the method of Ghiorso and Evans (2008) yields average temperatures of 774 °C (range of 744–806 °C) and aTi = 0.43. Using the recalculations of Stormer (1983) and Andersen and Lindsley (1985) gives slightly lower average T (747 °C). This latter method does not coincide well with other estimates, and it is derived from a model that is calibrated on experiments that were conducted at irrelevant T-fO2 conditions and is therefore not considered further. Zr saturation temperatures (Boehnke et al., 2013) of the Phase 2 glass lie within the temperatures estimated by the above methodology (780 °C). Zr saturation temperatures estimated by the Watson and Harrison (1983) calibration are slightly higher (804 °C).

These temperature estimates for the Phase 2 rhyolite are consistent with other fayalite-bearing rhyolites and other high-silica rhyolites with anomalously high temperatures (Warshaw and Smith, 1988; Deering et al., 2010; Ghiorso and Gualda, 2013; Wolff et al., 2015). Also significant is the low aTi, which is consistent with particularly high temperature felsic melts crystallizing ilmenite as the dominant oxide (Ghiorso and Gualda, 2013; Schiller and Finger, 2019). Importantly, the fayalite rhyolites studied throughout the literature lie on or below the FMQ buffer with moderate to high-water content at fluid saturated conditions (Mahood, 1981; Novak and Mahood, 1986; MacDonald et al., 1987; Warshaw and Smith, 1988; Chesner, 1998; Portnyagin et al., 2012). For this purpose, water contents were estimated using a plagioclase-glass hygrometer (Waters and Lange, 2015). This allows us to assess the potential dependence of a generally ferrous iron assemblage on water content and the inherent implications for explosive rhyolite volcanism. Using input temperature of 770–800 °C as input to the Waters and Lange (2015) plagioclase-liquid hygrometer returns average water contents ranging from 4.9–5.7 wt% (average 5.29 wt%; Table 8).

4.6.3 Phase 1 Rhyolite

Temperature, pressure, and water contents of the Phase 1 rhyolite were estimated using the plagioclase-liquid method of Putirka (2008) using matrix glass and plagioclase rims. All rims and liquid combinations are within the equilibrium exchange window of graphic (0.05–0.15). The temperature and water contents are constrained between 834 and 850 °C (average 845 °C; SEE: 36 °C) and 4.8–4.9 wt% H2O (average 4.9 wt%; SEE: 1.1 wt%). The output pressure is always 210 MPa, but the SEE (247 MPa) could put the magma on the surface or in the mid-crust and is thus presented here with caution. The Waters and Lange (2015) hygrometer returns water contents of 4.4–5.1 wt% H2O (average 4.7 wt%; SEE: 0.3 wt%) using the range of temperatures output from the Putirka (2008) model. Varying the pressure input by 200 MPa results in a mere difference of ~0.1 wt%.

Zr saturation temperatures (Boehnke et al., 2013) of the glass are lower than temperatures estimated from plagioclase (808–829 °C; Table 8), implying the magma should not be crystallizing abundant zircon because this overlaps or is lower than equilibrium plagioclase-glass temperatures. The Watson and Harrison (1983) calibration inherently predicts higher Zr saturation temperatures (829–846 °C) but this estimate also overlaps with plagioclase-glass temperature.

4.6.4 Oxygen Fugacity

Phase assemblages preserved in eruptive units at Caspana imply multiple oxidation environments; understanding how oxidation changes throughout the eruption of the Caspana ignimbrite is paramount to understanding the physical and compositional evolution of the magmatic system (Table 2; Fig. 3I). Therefore, understanding how oxidation changes throughout the eruption of the Caspana ignimbrite is paramount to understanding the physical and compositional evolution of the magmatic system.

Although there are no readily accessible mineral-mineral or mineral-melt systems in the Caspana andesite that allow for fO2 to be estimated directly, the phase assemblage preserved in the andesite (plagioclase + orthopyroxene + FeTi oxide dominated) is consistent with experiments conducted on under saturated andesites with moderate to high H2O (2–5 wt%) and fugacity at ~ΔFMQ (i.e., this assemblage will only crystallize at or below FMQ; Eggler, 1972; Blatter and Carmichael, 2001). This assemblage is also predicted by r-MELTS. Importantly, these oxygen fugacities are lower than those estimated for most of the other systems in the APVC (Lindsay et al., 2001a; Schmitt et al., 2001; Folkes et al., 2011; Grocke et al., 2017b).

Oxygen fugacities for the Phase 2 rhyolite were calculated on coexisting magnetite-ilmenite pairs using the two-oxide method described above. The algorithm of Ghiorso and Evans (2008) yields an fO2 for the Phase 2 rhyolite of approximately one log unit beneath the ΔFMQ buffer (−1.03).

We cannot place direct constraints on the oxygen fugacity of Phase 1 in order to identify if this part of the system shared the same fO2 environment as Phase 2. However, the plagioclase phenocrysts in the Phase 1 rhyolite have Fe/Al ratios (Fe3+ substitutes for Al) that are notably higher compared to the Phase 2 rhyolite even though the FeOTot content is comparable between the two rhyolites (Fig. 5). These data suggest that the Fe content of the plagioclase crystals is potentially controlled by the oxygen fugacity of the system and not strictly on the melt composition (Tepley et al., 2013). In fact, Phase 1 plagioclase phenocrysts have Fe/Al equivalent to those in the andesite and other silicic magmas that have erupted in the APVC (Watts et al., 1999; Schmitt et al., 2001; Folkes et al., 2011; Grocke et al., 2017a, 2017b). Invoking observations by the correlation of these data to the observations of prior workers and the inferred phase equilibria leads us to believe that Phase 1 likely had an oxidation state at least a log unit higher than ΔFMQ, typical of APVC magmas with comparable Fe-indices (Fig. 5).

Given these constraints, it should be noted that the presence of ilmenite > magnetite and orthopyroxene >> clinopyroxene is the opposite of what is generally found in APVC intermediate magmas (Table 2; de Silva and Francis, 1989; Lindsay et al., 2001a; Folkes et al., 2011; Burns et al., 2015; Grocke et al., 2017a; Kaiser et al., 2017), as is the presence of fayalite in rhyolite (Fig. 3H). This implies significantly different petrologic conditions for the Caspana system, namely fO2 and H2O (Kelley and Cottrell, 2009; Grocke et al., 2016; Burns et al., 2020).

Stratigraphic changes combined with bulkrock analyses and mineral chemistry reveal that the Caspana ignimbrite eruption evacuated the most heterogeneous collection of magmas in any single known eruption from the APVC. This integrated data set suggests that heterogeneous magmatic systems can develop in proximity to a larger regional system under flare-up conditions that tend to promote homogeneity (e.g., de Silva et al., 2006). In this discussion, the relationship between the Phase 1 and Phase 2 magmas and the large compositional gaps are examined with special interest paid to the only known occurrence of fayalite in the APVC. Based on this examination, we then draw comparative relationships between the Caspana reservoir and its resident magmas with the rest of the APVC.

5.1 Production of the Phase 2 Rhyolite by Closed System Crystallization of Andesite

It is clear from the isotopic ratios that the Caspana andesite and Phase 2 rhyolite are related by nearly closed system fractionation (Fig. 8 and Fig. S4, see footnote 1). This is supported by high–Fe-indices in the Phase 2 rhyolite resulting from the fractionation of high-An plagioclase and enstatite from the andesite (Fig. 5). The elevated HREEs in the Phase 2 rhyolite (Fig. 6) also reflect an assemblage that is absent of clinopyroxene and/or abundant amphibole due to the low graphic of these minerals.

To model the relationship between the andesite and Phase 2 rhyolite, the Excel®-based software Magma Chamber Simulator (MCS; Bohrson et al., 2014, 2020) along with its rhyolite-MELTS major- (Ghiorso and Sack, 1995; Gualda et al., 2012) and trace- (Spera et al., 2007) element engines were used with relevant partition coefficients from the literature (Table S9, see footnote 1). For the purposes of modeling, wall-rock assimilation is not considered because of the isotopic concordance of Phase 2. Based on the extraction and storage pressures estimated for the andesite and rhyolite, the isobaric MCS was run from 400 to 200 MPa using a 100 MPa step in order to deal with the vertically extensive reservoir that is suggested by the phase equilibria. The chemical evolution at the deepest pressure (~400 MPa; section 4.6.1) uses the composition of andesite pumice. The input composition at the rhyolite extraction pressure (~300 MPa; section 4.6.2) is the liquid composition at 920 °C, which is the equilibrium temperature of the G2 glomerocrysts and plagioclase (Table 8). While the error on the orthopyroxene-liquid temperature estimate is obviously large, the choice is supported by the fact that the composition turned out to coincide with the composition of rhyodacite glass when analyzing observed versus modeled values. The composition for the final step is chosen such that the composition is appropriate for modeling the final stages of liquid evolution with r-MELTS (i.e., version 1.1 instead of 1.2) and at the maximum temperature of the olivine-liquid thermometer (800 °C + 29 °C SEE). The composition at this temperature is rhyolite.

The best-fit MCS models show that the major- and trace-element trends between the andesite and Phase 2 rhyolite can be produced via crystal fractionation (Figs. 11 and S5, see footnote 1) with initial water contents of 3–4 wt% H2O and ΔFMQ – 1 (Table 8) in the andesite. Higher and lower input water contents create a poor fit for the trend defined by pumice and glass at the estimated temperatures (Fig. 12), as does higher fO2. Significantly, rhyolite-MELTS predicts the crystallization sequence plagioclase ± magnetite > plagioclase + magnetite + ilmenite + orthopyroxene > plagioclase + orthopyroxene + ilmenite (Table 9) in modal proportions that matched the observed G1 and G2 glomerocrysts. Interpreting the glomerocrysts as disaggregated remnants of cumulates (de Silva, 1989c; Ellis et al., 2014) leads to the apparent, near perfect closed-system fractionation between the Phase 2 lithologies that is readily reproduced with the simple lever principle in major-element space (Figs. 4 and 5). While this is not a scientific breakthrough in any regard, observing it in the natural world should invoke some degree of bewilderment.

The thermodynamics (Table 8), phase chemistry, and cumulus textures (Fig. 3) record the progressive cooling in the more mafic part of the system. At the highest pressures, plagioclase is predicted to be first on the liquidus at ~1050–1100 °C, followed by orthopyroxene and magnetite saturation at ~1000 °C (Fig. 13). These orthopyroxenes have high-Al due to high pressure and are thus represented by the P1 phenocrysts that were co-crystallizing with plagioclase (Fig. 10). This co-crystallizing assemblage remained saturated all the way through the crystallization sequence in this hydrous, low-fO2 environment, driving up the Fe-indices that lead to fayalite stability (Figs. 5 and 10). Ilmenite then saturates at the same temperature estimated by equilibrium temperature of orthopyroxene (~930 °C).

In the second, lower-pressure step of the thermodynamic models, the crystallization sequence is similar, except that magnetite is first on the liquidus (Fig. 12), and the various phases saturate at temperatures within error of what is estimated by phase equilibria (Fig. 12; Table 8). The stability of this assemblage through the crystallization sequence allowed the large glomero crysts (Fig. 3A) to grow in the upper reaches of the reservoir as cooling progressed. The low Al and En contents of orthopyroxene in G2 glomerocrysts and P2 phenocrysts (Fig. 10) that record lower temperature and pressure than the P1 phenocrysts (Martel et al. 1999) provide further support for this interpretation of the glomerocrysts. The dominant equilibrium state during liquid evolution and formation of the noritic cumulate is clearly recorded by the convergence of P1 and P2 phenocryst rims toward the glomerocryst compositions (Fig. 10). Plagioclase in the G2 glomerocrysts provides further support that the glomerocrysts were critical in the final stages of liquid evolution (Fig. 9B). An content of these plagioclase rims are offset to low-An values, similar to reversely zoned pheno cryst cores and the subset of normally zoned phenocrysts (Figs. 9B and 9C). The more lath-like texture of G2 plagio clase compared to their counterparts (Figs. 3A, 3B, and 3E) also suggests that they were the plagioclase growing where thermal gradients were highest (e.g., de Silva, 1989c) and when orthopyroxene was already saturated in the advanced stages of crystallization (Figs. 11 and 12).

The switch from magnetite to ilmenite as the dominant oxide is recorded in the FeO content in the rims of the G2 glomerocrysts and reversely zoned pheno crysts (Figs. 3I, 9B, and 9C). As these plagioclase crystals grew, Fe3+ became progressively more available and oxidation state increased with changing pressure. That is, Fe3+/Fetot was probably controlled by a decrease in pressure (Kress and Carmichael, 1991) and the crystallizing assemblage (Cottrell and Kelley, 2011; and references therein), though it was probably only a local process given that the andesite and the Phase 2 rhyolite share a low oxidation state.

The dominance of ilmenite during advanced stages of crystallization is also reflected in the ilmenite chemistry (Fig. S1). Ilmenite crystals become progressively more enriched in TiO2 and V2O3 as the glomerocrysts continued to grow. Following Buddington and Lindsley (1964) and Ghiorso and Evans (2008), we interpret the low-TiO2 and V2O3 ilmenite to be the result of crystallization in the presence of Ti-magnetite (now exsolved), which is predicted by r-MELTS to be first on the liquidus (Table 9). Transition from magnetite to ilmenite dominance also agrees with the high aTi recorded in the mineral chemistry of the residual, Phase 2 rhyolite (i.e., biotite and Ti-magnetite). We propose this to be a more consistent interpretation than an alternative where ilmenite is introduced by recharging magma.

It is clear that in the Caspana system the liquid line of descent (LLD) associated with fractionation of rhyolite was controlled by crystallization of FeTi oxides (e.g., Toplis and Carroll, 1996; Morse, 2011). This data set, however, does not explain the lack of a compositional continuum in either bulk rock (Figs. 46 and 11) or mineral chemistry (Figs. 9 and 10) that would be inevitable during the physical process that would accompany fractional crystallization sensu stricto. In fact, it requires some other physical mechanism to explain the apparent shallow liquidus in Phase 2 (de Silva, 1991; and references therein). It is worthwhile to point out that when orthopyroxene and ilmenite saturate at 300 MPa, the liquid composition is rhyodacite to rhyolite with the initial input water contents of 3–4 wt%.

5.2 In Situ Crystallization in Gabbronoritic Mush Controls Caspana’s Shallow Liquidus

Compositional gaps have been identified in volcanic systems in virtually every tectonic environment with fractionation proposed as a dominant mechanism for their presence (Daly, 1925; Brophy, 1991; Dufek and Bachmann, 2010). The compositional gap of ~16% SiO2 between the andesite and Phase 2 rhyolite bulk compositions is one of the more extreme within global compilations by Brophy (1991) and Dufek and Bachmann (2010) and warrants consideration. A comparison of the modeled LLDs (Figure 12) naturally shows that the liquidus temperatures are negatively correlated with initial water content (Fig. 12). The result of this is a decrease in the temperature interval required for crystallization from andesite to rhyolite (Fig. 12 and Fig. S7, see footnote 1). The low fO2 of this system specifically would increase plagioclase stability as opposed to amphibole at liquidus temperatures (Martel et al., 1999), validating the comparisons of r-MELTS models in the absence of amphibole with high water content in the melt. Overall then, it seems that high water content will serve to lower the liquidus temperature while also decreasing the temperature interval required for fractionation in both low (in this study) and high (Grove and Donnelly-Nolan, 1986; Brophy et al., 2011) fO2 systems.

de Silva (1991) suggested that a shallow liquidus between andesite and rhyolite was responsible for the compositional gap present in the Caspana system. Similar conclusions have been made for the well-established gap at the Medicine Lake system where rhyolite is directly related to its gabbroic cumulate (Grove and Donnelly-Nolan, 1986; Brophy et al., 2011). The lack of a compositional continuum is, however, difficult to reconcile by classic models such as ideal convection and growth of a solidification front (Bachmann and Bergantz, 2004; and references therein), especially when considering the locked rheological state of the crystal mush represented by the G2 glomerocrysts. Instead, compositional gaps have been hypothesized to develop by interstitial melt extraction driven by crystal compaction in systems with intermediate to nearly rheologically locked systems (~40%–70%; Bachmann and Bergantz, 2004; Dufek and Bachmann, 2010). Eruptible volumes of evolved interstitial melt are created within these porous media so long as the latent heat of crystallization can dominate the heat budget, forcing a low thermal gradient that promotes prolonged crystallization of stable mineral assemblage (Morse, 2011; Dufek and Bachmann, 2010). Indeed, textural observations suggest that significant crystallization events, particularly from FeTi oxides, can temporarily halt textural maturity in Pl + Pyx-dominated mushes and allow for efficient, rapid melt extraction of evolved interstitial melt (Holness et al., 2007, 2011). The effect of latentheat buffering in the Caspana system is tested with r-MELTS using the same inputs as the MCS models described in section 5.1 and utilizing the method described in detail by Sliwinski et al. (2015).

In the 400 MPa step, a latent-heat spike is brought on by orthopyroxene saturation, but the latent:sensible heat is <1 and thus below the threshold where crystallization is no longer favored and in situ melt would be released (e.g., Morse, 2011) (Fig. 13). The crystallinity is also not above the threshold where the probability of melt extraction by channelization and compaction is non-negligible (40%) (Bachmann and Bergantz, 2004; Dufek and Bachmann, 2010). At 300 MPa, the latent-heat spike is brought on by orthopyroxene and, even though minor, ilmenite contributes some buffering during the fractional latent-heat dissipation of orthopyroxene. The temperatures at which ilmenite and orthopyroxene saturation occur—at 300 MPa—are within a few degrees of ilmenite saturation at 400 MPa, collectively pushing the latent:sensible heat >1 and putting the crystallinity within range of probable melt extraction from a “mushy” system with local channel development (e.g., Dufek and Bachmann, 2010).

The liquid composition at the crystallinity where latent heat released by crystallization favors melt extraction is rhyodacitic to rhyolitic and quickly evolves to the Phase 2 rhyolite composition (Fig. 11). The paucity of continuous mineral and bulk-rock compositions in the Caspana system is well-explained by in situ crystallization in a thermally buffered system and subsequent interstitial melt extraction. Importantly, the control that FeTi oxides exert on the system is significant for both the LLD and the heat budget (Figs. 12 and 13), which is consistent with observed plutonic equivalents of the gabbronoritic glomerocrysts (Holness et al., 2007, 2011). With respect to the expansive APVC and its many ignimbrites, this mechanism has obvious implications for efficient, rapid rhyolite production from either small-volume, short-lived systems (e.g., Schmitt et al., 2011) or large-volume, long-lived systems that are subjected to recharge (Lindsay et al., 2001a; Folkes et al., 2013; Grocke et al., 2017b).

5.3 Petrologic Conditions Explain Fayalite Rhyolite in the APVC

The occurrence of fayalite in rhyolitic magmas has been documented for well over a century (Iddings, 1885); however, the conditions that lead to its saturation and the co-crystallizing mineral assemblages appear to be quite varied (Bacon et al., 1981; Mahood, 1981; Novak and Mahood, 1986; MacDonald et al., 1987; Warshaw and Smith, 1988; Jónasson, 1994; Lowenstern et al., 1997; Chesner, 1998; Portnyagin et al., 2012; Holness et al., 2019; Rooyakkers et al., 2021). Warshaw and Smith (1988) originally proposed that fayalite is stable due to cations influencing the oxidation state of the melt. Specifically, FeO/CaO correlates negatively with Fe3+/Fe2+ due to increased amounts of alkaline Earth metals disrupting the melt structure and oxidizing multivalent ions. Fe2+/Mg must also be high, as the Mg component of the melt must be sufficiently low to allow an Fe2+-enriched mineral (fayalite) to be stable instead of an Mg-rich mineral (orthopyroxene). This latter inference is supported by textural evidence that shows fayalite develops orthopyroxene overgrowths during pronounced perturbations (i.e., recharge) to a volcanic system (Portnyagin et al., 2012; Troch et al., 2017; Chiaro, 2019). Both of these cation ratios are proxied by the Fe-indices in Figure 5. As far as we know, none of the other known ignimbrites containing high-silica rhyolites in the APVC (Toconao, Alota-Juvina, Talabre, Carcoté; Lindsay et al., 2001a; Salisbury et al., 2011) are known to have fayalite in them except for the Caspana Phase 2 rhyolite. The Phase 2 rhyolite also has the highest Fe-indices that we know of in the APVC (Fig. 5) and relatively flat REE patterns at high overall concentrations (Fig. 6). This provides us with an opportunity to investigate the petrology of the Caspana andesite and Phase 2 rhyolite with respect to the plethora of well-studied APVC ignimbrites that are derived from the APMB.

Rhyolite-MELTS modeling supports the crystallization of high-An plagioclase and enstatite in the “low” oxidation state (fO2 ≤ ΔFMQ) from a parental andesite (Eggler, 1972; Blatter and Carmichael, 2001) as the cause of the high Fe-indices in the Phase 2 rhyolite (Fig. 5). Additionally, we have shown that the relative amounts of magnetite and ilmenite in the MCS runs are controlled by low fO2 with little correlation of H2O (Fig. 12; File S7, see footnote 1), and the pressures and water contents of the Phase 2 rhyolite (Table 8) indicate that it was saturated (Newman and Lowenstern, 2002; their fig. 2). These constraints demonstrate that the reduced mineral assemblage of the Caspana system (Table 2) and the appearance of fayalite in high-Si rhyolite (Fig. 3H) is apparently not controlled by H2O, but solely low fO2. In fact, increasing the water content would only serve to increase the olivine stability field and promote fayalite stability (Portnyagin et al., 2012; and references therein).

The presence of hydrous, ferrous-rich phases (annite and allanite) provides further evidence for a high-water, low-fO2 environment (Tables 2 and 8; Fig. 3G; Fig. S2, see footnote 1). The nearly linear REE trends defined by the andesite pumice, rhyodacite glass, and the Phase 2 pumice are consistent with fractionation of Pl + Opx + Ilm, with the exception of La and Ce (Fig. 13), which can likely be attributed to crystallization of allanite. Indeed, the abnormally large size of allanite (for APVC ignimbrites) is likely the result of early saturation brought on by the low fO2 (Table 2, Fig. 3; Vlach and Gualda, 2007). Given that the phase has no correlation with REE concentration of host rocks (Vlach and Gualda, 2007), it may be that the low oxidation state promoted allanite stability due to a high Ce3+/Ce4+. The positive or flat Eu anomaly in the andesite pumice samples (Fig. 6) also implies a residual liquid with a rather steep negative Eu anomaly, as observed in the Phase 2 rhyolite. These characteristics are rare or unobserved in APVC magmas, adding more support to the interpretation that the oxidation state at least partially controlled REE partitioning while plagioclase was on the liquidus (i.e., high Eu2+/Eutot).

While providing decent first-order assessments of the system, the parental assemblage in the andesite and melt structure still does not explain why the Phase 2 rhyolite and andesite are carrying reduced assemblages in an arc setting where magmas should be oxidized by either mantle source properties (Kelley and Cottrell, 2009) or crystallization during stalling in the lower crust (Ulmer et al., 2018). This leaves three possibilities for the reduced state of the magma: (1) crystallization-induced reduction from the parental basalt; (2) degassing-induced reduction (Kelley and Cottrell, 2012); and (3) a source that has an oxygen fugacity lower than expected. The Caspana andesite is currently one of the most mafic andesites measured in the APVC (Fig. 4) and should be crystallizing abundant magnetite and have an oxidation state significantly higher than it does (Blatter et al., 2013; Ulmer et al., 2018; Burns et al., 2020) at these pressures and the observed degree of evolution from parental basalt. Thus, significant crystallization of magnetite causing reduction in the andesite can be ruled out. It has also been shown that the magnitude of reduction that can occur during degassing of most APVC magmas is too small to explain the oxidation state of the Caspana system (Grocke et al., 2016), and models suggest that that fO2 is too high for graphite stability. Rather, multiple studies have found that fO2 remains the same within a given magmatic lineage in the APVC (Grocke et al., 2016; Burns et al., 2020). The Phase 2 rhyolite and the gabbronorite cumulate formed during its fractionation bear obvious semblance to the relatively reduced fayalite rhyolites and gabbro-forming basalts discussed by Frost and Frost (1997) that are produced by mixing of primary mantle melts and partially melted, igneous rocks in the lower crust during periods of high heat advection. This model is consistent with the crustal foundering and the development of lower-crustal melting-assimilation-storage- and homogenization (MASH) zones (Hildreth and Moorbath, 1988) thought to have taken place during the Neogene ignimbrite flare-up in the APVC prior to adiabatic ascent into the upper crust (Kay and Coira, 2009; Burns et al., 2020). It is beyond the scope of this paper to directly address the lower crust and mantle, but inferential evidence suggests that the fO2 of the Caspana system was governed by its mantle source, or, its lineage has a significant contribution from partial melts of mafic igneous rocks in the lower crust. Given the well-established evidence for extensive assimilation of crust that effectively filters out mantle source characteristics in the Central Andes (Davidson et al., 1991; Kay et al., 2010), we prefer the latter alternative.

5.4 Recycling Intermediate-Silicic Compositions Produces the Phase 1 Rhyolite

The outlier isotopic composition, peraluminous character (Table 1; Fig. 8), and excursions in major- and trace-element space (Fig. 5) show that the Phase 1 rhyolite cannot be genetically related to the andesite and Phase 2 rhyolite. The FeO content of the Phase 1 glass is also appreciably lower than the rhyodacitic glass from the Phase 2 andesite and within error of the Phase 2 rhyolite glass, indicating that the high Fe/Al ratio of Phase 1 plagio clase (Table 3) is the result of a high oxidation state rather than the Fe content in the melt (Toplis and Carroll, 1996; Tepley et al., 2013). The Phase 1 rhyolite is similar to the peraluminous, garnet-bearing Coyaguayma rhyolites of Caffe et al. (2012) but has some important differences. The Coyaguayma rhyolites, apparently derived from ~30% contamination of metasedimentary rocks into dacitic melt, have Pb isotopic compositions slightly more radiogenic than the Phase 2 rhyolite and are more peraluminous (A/CNK > 1.3; Table 1). These strongly peraluminous (SP) magmas are also substantially more fractionated (Sr ~55 ppm and Ba ~65 ppm) and have notable differences in isotopic composition and mineral assemblages (i.e., sillimanite and garnet). Given these differences, it is unlikely that the Phase 2 rhyolite is an extension of these SP rhyolites that resided and isobarically cooled in the midcrust while assimilating metasedimentary rocks prior to eruption. Another potential source is partial melting of the low-grade metasedimentary rocks that are occasionally found as xenoliths in the APVC east calderas (Ort et al., 1996; Kay et al., 2010; Caffe et al., 2012), which could explain the Nd isotopic disequilibria (Fig. 9; Ayres and Harris, 1997; Wolf et al., 2019), but the Sr isotopic composition of Phase 1 is far too low. It has also been proposed that the rapid fractionation of andesitic liquid when intruded into the base of the previously emplaced magma reservoir may produce some of the rhyolitic magmas observed in the APVC (Schmitt et al., 2001). However, the only evidence for mafic recharge is the andesite that fractionated to form the Phase 2 rhyolite (Figs. 5, 10, and 11). This model is not applicable to the Caspana system, because our data indicate that fractionation of the andesite was not rapid and instead occurred over multiple kilometers during the ascent and subsequent cooling of the andesite. This model also relies on bulk density and viscosity contrasts that have been shown insufficient to explain an eruptive sequence that is initiated by a recharging magma unless the intruding melt was already less dense than resident mafic magma when recharge occurred (Carrara et al., 2020).

In the light of the weaknesses and inconsistencies of the alternatives above, we propose a partial melting origin for the Phase 1 rhyolite. Experimental results have found that peraluminous rhyolitic melts similar to the Phase 1 rhyolite can be formed by melting of granodiorites (Patiño Douce, 1997) or amphibolites and gneisses (Patiño Douce and Beard, 1995). Plagioclase in the Phase 1 rhyolite occasionally have “veins” of high-An content (Fig. 3D) that are indicative of unmixing during prolonged cooling below the solidus (e.g., Alling, 1932), providing direct textural evidence for some relation to a slow cooling igneous body. Equilibrium temperatures between glass and the rims of these plagioclase are higher than the Zr saturation temperatures (Table 8) indicating that zircon would not be stable prior to eruption. Combining these temperature constraints with the increase of Zr-Hf concentration of glass relative to juvenile pumice clasts (Fig. 8), clearly shows that the zircon in thin section was being introduced back into the melt prior to eruption. A parallel argument can be made for plagioclase, given that Sr and Eu concentrations of the glass that are within error of bulk-rock compositions (Fig. 5D). These data and lines of evidence suggest that the Phase 1 rhyolite can be formed by partial melting of granodiorite or cumulate melting. These processes will be explored further later in this section.

We build on the interpretation that the uppermost crust in the APVC has a large component of granodioritic intrusions (de Silva et al., 1994; Tierney et al., 2016) like that recorded in co-magmatic xenoliths at the Pastos Grandes caldera to the north of the Caspana outcrops (Watts et al., 1999; Kaiser et al., 2017). Some remnant cumulate mush within the upper crust is also reasonable to consider. Various groups of Ordovician granitoids with similar geochemistry to the Pastos Grandes ignimbrite (PGI) also contribute to the local basement (Lucassen et al., 1999, 2001). Cogenetic xenoliths and pumice clasts found in the PGI thus have the geochemical composition that can explain the petrogenesis of the Phase 1 rhyolite by either of the proposed methods. Notably, these xenoliths broadly have the composition of low-silica rhyolites and rhyodacites in the APVC (Fig. 11).

Least squares residual models (LSQ) using the open-source version of the program Igpet (Stormer and Nicholls, 1978; Carr and Gazel, 2017) were used to test the viability of the petrogenetic models (Tables S10 and S11, see footnote 1). Results of the LSQ for both fractional crystallization and batch melting are the same since the test is simply to derive the fit of one composition from another based on mass balance, except for the amount of melt remaining (F). For the melting case, F becomes the amount of protolith left after melting, and 1-F is the amount of melt produced. Least squares results are satisfactory for the granodiorite (F = 0.353, R2 = 0.236) in major-element space (Table S10, see footnote 1). To reconcile the complexities associated with phase changes during partial melting, the incongruent dynamic melting (IDM) model of Zou and Reid (2001) was employed to model changes in traceelement concentration. For the case of cumulate melting and prolonged melt presence, the same LSQ model is used, and the cumulate material is partially melted using the fractional melting equation (Shaw, 1970; Wolff et al., 2020). This cumulate melt is then mixed with the pumice clasts from the PGI using the common mixing equation, as these would represent the liquid dominated portion of the reservoir. Parameters, partition coefficients appropriate for rhyolitic melts, and a full explanation of these models are given in Table S10 (footnote 1).

In both models, Fe-Mg minerals, feldspars, and quartz (granodiorite) undergo melting, and in the IDM case, these minerals react to form residual titanite + clinopyroxene and partial melt. The depletion in Fe-indices in the glass can be easily reconciled considering that biotite would be introduced to the melt concomitantly with quartz, and progressive melting would eventually introduce clinopyroxene (cpx) and more Mg-rich hydrous minerals (e.g., amphibole). Evidence for cpx or amphibole at the source is indicated clearly by the severe Sc depletion (Fig. 6) and low Dy/Dy* values (Fig. 7), as these would behave compatibly in the protolith. The results of the trace-element models (Fig. 11) show that either mechanism proposed here can explain the geochemistry of the Phase 1 rhyolite in trace-element space. Thus, re-melting remnant plutonic protolith from previous eruptions onto the APVC, or, melting of a cumulate material in the presence of felsic melt can produce the Phase 1 rhyolite. Again, this model is presented here as our current best effort, to be tested if better constraints on the upper crust in this area become available.

5.5 The Architecture of the Caspana Reservoir

The emerging evidence of a system that consisted of a variety of magmas instead of voluminous monotonous intermediates that typify the APVC solicits some discussion on its relevance. At the edge of the APMB, beneath the Caspana area, seismic velocity models speed up (Ward et al., 2014), indicating that the crust is less thermally softened and riven with melts as it is in the rest of the APVC where storage and homogenization in large dacitic reservoirs are promoted (de Silva and Gosnold, 2007; de Silva and Gregg, 2014). The cooler conditions on the periphery of the APMB would therefore limit the prolific assimilation and homogenization that characterizes most other APVC felsic magmas.

While broadly consistent with regional compositions, the isotopic signatures of Phase 1 and Phase 2 also support that these magmas remained discrete during storage (Fig. 9). Kay et al. (2010) pointed out that there is general agreement that isotopic diversity is generated in the low-to mid-crust where DSr is ≤1 (e.g., de Silva et al., 2006). AFC calculations based on DePaolo (1981) and Aitcheson and Forrest (1994) suggest that 50%–60% assimilation of a Sierra de Moreno gneiss (following Godoy et al., 2017) into a primitive mantle-derived basalt (Davidson et al., 1991; Mamani et al., 2010; van Alderwerelt et al., 2021) are required to account for the isotopic composition of Phase 2 (Table S8, see footnote 1). Meanwhile, the Sr and Nd isotopic compositions of Phase 1 are less evolved and more similar to the lavas that have erupted from local edifices that typify the modern arc (Fig. 1). The Phase 1 magma was able to maintain relatively low isotopic compositions by eluding the significantly contaminated APMB (González-Maurel et al., 2019). In effect, the locale of the Caspana system at the edge of the APVC/APMB facilitated its heterogeneity and ability to host magmas of different lineage in discrete batches. The physical storage conditions that limited any homogenization are described herein.

The norite cumulate represented by the G1 and G2 glomerocrysts and the Phase 2 rhyolite is parallel to a variety of cumulates that have been shown to be responsible for the fractionation of high-silica rhyolites (Ellis and Wolff, 2012; Ellis et al., 2013, 2014; Troch et al., 2017), in agreement with observations of plutonic rocks that show the crystallization of cumulates is an efficient mechanism for the production of felsic magma (Tavazzani et al., 2020). For the case of the Caspana system and its regional context, the small-volume reservoir would crystallize faster than its typical APVC counterparts (e.g., de Silva and Wolff, 1995), producing a dense cumulate at the margins that may have limited introduction of assimilated material, allowing the system to remain closed (Fig. 14). Latent heat produced by crystallization of the Pl + Opx + FeTi oxides was high enough to either induce partial melting of country rock (e.g., Grove et al., 1997) or remobilize a discrete batch of remnant dacitic magma from prior eruptions (e.g., Godoy et al., 2019). This process of dense crystallization synchronously produced rhyolites by reheating (Phase 1) and crystallization (Phase 2). As ilmenite came on the liquidus with orthopyroxene, latent-heat buffering allowed an eruptible volume of Phase 2 rhyolite to be accumulated at the roof after extraction from the cumulate. The crystal-poor rhyolites have a rheology that does not require pre-eruptive homogenization like the crystal-rich, large-volume APVC dacites (Huber et al., 2009, 2012), which allowed each rhyolite to retain its primary geochemical signature. However, some plagioclase phenocrysts found in Phase 2 rhyolite pumice clasts clearly grew under the thermodynamic conditions appropriate to Phase 1, and some of the Phase 2 microlites are intermediary (Fig. 15). These plagioclase crystals are the product of cryptic mixing either before eruption and/or periodically on long time scales (de Silva et al., 2008).

The dense cumulate framework and rhyodacitic liquid that make up the andesitic portion of the chamber does, however, require thorough disaggregation to be erupted. The mirrored distributions of normally and reversely zoned plagioclase phenocrysts in the andesite (Fig. 9C) attest to mixing of similar mafic liquid prior to eruption, as do high Mg-Al rims on P1 orthopyroxene (Fig. 10). Meanwhile, the An contents of cores and rims on normally zoned plagioclase phenocrysts in the andesite coincide with the cores and rims on G2 plagioclase from glomerocrysts (Fig. 8), indicating prolonged cooling and in situ crystallization of the fractionates in the upper portion of the reservoir as discussed in section 5.1. Presumably, the shearing and perturbation exerted on the mush during the recharge event that is indicated by the data helped induce eruption, as is commonly found in silicic magma.

The ~5 km3, 4.56–4.09 Ma Caspana ignimbrite of the Altiplano-Puna volcanic complex (APVC) of the Central Andes records the eruption of an andesite and two distinct rhyolitic magmas from a vertically heterogeneous reservoir established between 400 and 200 MPa.

The first erupted magma (Phase 1) was a crystal-poor peraluminous (A/CNK >1.2) rhyolite that was produced by either partial melting of granodiorite or melt extraction from a granodiorite mush in the upper reaches of the reservoir. The subsequent, main stage of the eruption (Phase 2) tapped a crystal-poor slightly peraluminous, fayalite-bearing rhyolite and a low-fO2 (≤ ΔFMQ), crystal-rich andesite that exhibits more “crustal” isotopic characteristics than Phase 1. Rhyolite-MELTS–based models indicate that one of the rhyolites (the Phase 2 rhyolite) is derived from the andesite by extensive crystallization of an assemblage represented by the abundant noritic glomerocrysts in the andesite.

The large (16 wt% SiO2) compositional gap recorded in the main Phase 2 reservoir is proposed to be the result of pre-eruptive segregation of the rhyodacitic residual melts from a gabbroic (norite) mush. Latent-heat buffering produced during in situ crystallization of silicic melt within the gabbronoritic cumulate imposed a shallow liquidus, causing efficient production of rhyodacitic melt. Spikes in latent heat facilitated the segregation of this rhyodacitic residual liquid that further fractionated to produce high-SiO2 rhyolite. The hydrous, low-fO2 conditions promoted ilmenite stability instead of magnetite, causing an enrichment of FeO relative to the rest of the APVC during in situ crystallization. These petrologic conditions caused high Fe2+/Mg and Fe/Ca by keeping high-An plagioclase (~An83) and enstatite (Mg# ~58) leading to saturation of fayalite in the Phase 2 rhyolite (Warshaw and Smith, 1988)—unique in the APVC. Estimates of pressure and water contents suggest that the Phase 2 rhyolite was saturated and that fayalite stability has no dependence on, and is probably enhanced by, high water content (Toplis and Carroll, 1996; Portnyagin et al., 2012).

The Caspana ignimbrite records km3-scale compositional heterogeneity of diverse origin and a singular magmatic evolution within a regional magmatic complex that is dominated by monotony. This rare record is attributed to the development of the Caspana magma system on the cooler periphery of the regional upper-crustal magmatic reservoir—the Altiplano-Puna magma body—where the Caspana magmas fortuitously escaped being mixed into the APMB and the APVC magmatic system. As such, this small-volume ignimbrite provides a unique window into the multiscale processes that build large silicic magma systems.

The foundation of this work was conducted partly under the auspices of National Science Foundation (NSF) grants EAR 0710545, EAR 0838536, and EAR 0908324. Reviews from Francesca Forni, Lorenzo Tavazzani, and Brad Pitcher greatly strengthened the quality of this manuscript. CL acknowledges support from the NSF grant DUE:IUSE 1600403 and the NSF summer REU program at Oregon State University (NSF OCE-1758000). This work was completed while supported by an Oregon State University COEAS Graduate Teaching Assistantship. CL would like to thank Jordan Lubbers for his invaluable support and mentoring. Jason Kaiser, Frank Tepley, Sarah Lapinski, and Michael Iannucillo are thanked for fieldwork assistance. Chris Russo and Jordan Lubbers were incredibly helpful in gathering data for LA-ICP-MS. We would also like to thank Cerise Burns for her comments on data visualization. Stanford Earth Microchemical Analysis Facility (MAF) is part of the National Science Foundation National Nanotechnology Coordinated Infrastructure (NNCI).

1Supplemental Material. Full dataset and model parameters. Please visit to access the supplemental material, and contact with any questions.
Science Editor: Andrea Hampel
Guest Associate Editor: Guilherme Gualda
Gold Open Access: This paper is published under the terms of the CC-BY-NC license.