Several cubic kilometers of Paleozoic graphite-bearing argillitic country rocks are present as lithic fragments in Bishop Tuff ignimbrite and fallout. The lithics were entrained by the 650 km3 of rhyolite magma that vented during the 5- to 6-day-long, caldera-forming eruption at Long Valley, California. The caldera is floored by a 350 km2 roof plate that collapsed during the eruption and consists in large part of the Paleozoic strata that provided the abundant hornfelsed metapelitic lithic clasts in the tuff. Graphite has been identified by Raman spectroscopy, electron-dispersive spectroscopy, and X-ray diffraction as an irregularly dispersed component in the small fraction of Bishop Tuff pumice that is dark-colored. Carbon concentration has been determined in pumice, lithics, and wall rocks. Values of δ13C range from –21‰ to –29‰ Vienna Peedee Belemnite (VPDB) for pumice, lithics, and argillitic wall rocks, reflecting the biogenic origin of the reduced carbon in oxygen-limited black Paleozoic marine mudrocks. Carbonate contents, measured separately, are negligible in fresh pumice and lithics. Microprobe analyses of titanomagnetite-ilmenite pairs show that oxygen-fugacity values of numerous batches of postcaldera Early Rhyolite (750–640 ka; ~100 km3) are up to one log unit more reduced than those of the temperature–oxygen fugacity (T-fO2) array of the Bishop Tuff (767 ka), despite similar major-element compositions and Fe-Ti–oxide temperature ranges. All of the many batches of Early Rhyolite, which erupted episodically over an interval of ~125,000 years, yield the reduced fO2 values, indicating that reaction with graphite lowered magmatic fO2 after the caldera-forming eruption but before the first eruption of Early Rhyolite. It is inferred that reaction of postcaldera rhyolite magma with the reduced carbon in a great mass of subsided roof rocks lowered its fO2. It is suggested that comparable effects could have attended caldera collapse of other magma chambers hosted in continental sedimentary rocks.

The Bishop Tuff is the rhyolitic pyroclastic product of one of the world’s greatest Quaternary caldera-forming eruptions. The eruption lasted 5–6 days and resulted in 2–3 km subsidence of a 350 km2 roof plate, thereby creating Long Valley caldera. The magmatic and structural consequences of abrupt caldera collapse have long been of broad interest (Clough et al., 1909; Williams, 1941; Lipman et al., 1984; Acocella, 2007). In this contribution, we report recognition of a previously undocumented process—reaction with graphite from collapsing argillitic roof rocks and its profound impact on the redox state of the rhyolitic magma that subsequently erupted in numerous postcaldera batches.

Proximate results of the collapse of continental ash-flow calderas (Lipman, 1997) include (1) dispersal of wall-rock lithic fragments throughout the pyroclastic deposits erupted (Eichelberger and Koch, 1979; Hildreth and Mahood, 1986; Hildreth and Wilson, 2007); (2) syneruptive slumping of breccia sheets and slabs of wall rock into the accumulating intracaldera tuff (Lipman, 1976; Suemnicht et al., 2006); (3) rapid growth of intracaldera lakes (Bethke and Hay, 2000; Bacon et al., 2002; Hildreth and Fierstein, 2016a); and (4) reaction of collapsed roof rocks with the magma left behind—the focus of the contribution at hand.

Assimilation, reaction, or melting of country rocks at the roof or walls of rhyolitic magma chambers were discussed by Hildreth et al. (1991), Duffield and Ruiz (1992), Gansecki et al. (1996), Wolff et al. (1999), and Bindeman and Valley (2000), but no definitive example is known to have demonstrated that such contamination took place in direct response to caldera collapse. Compositional effects resulting from reaction with collapsed roof rocks can be hard to separate from those caused by precaldera roof-zone assimilation, syncollapse convective disturbance of a zoned magma reservoir, recharge from below by new magma batches, or postcaldera roof-zone assimilation or melting. Among the clearest examples of roof-zone effects are assimilation of hydrothermally altered roof rocks that impart to the magma reduced values of δ18O (Bacon et al., 1989; Bindeman et al., 2008), but in these cases the assimilation was either a precaldera process (as at Crater Lake, Oregon) or is identifiable only in postcaldera lavas that erupted >105 years after caldera collapse (as at Yellowstone).

Here we identify assimilation of an unusual contaminant, metasedimentary graphite, which was sparsely and unevenly engulfed by the rhyolitic magma that erupted during the caldera-forming event and which, after massive roof-rock collapse, profoundly modified the redox state of the magma chamber that subsequently produced voluminous postcaldera rhyolite lavas and tuffs.

The Bishop Tuff consists of ignimbrite and fallout that represents ~650 km3 of rhyolitic magma that erupted during collapse of Long Valley caldera (Fig. 1) at ca. 767 ka. Owing to good preservation, exposure, and accessibility, extensive stratigraphic and petrologic data, and to fascinating inferences from those data about eruptive and pre-eruptive processes, the Bishop Tuff has attracted global attention as a paradigmatic model of a caldera-forming eruptive unit. Following pioneering studies by Gilbert (1938), Sheridan (1970), and Hildreth (1977, 1979, 1981), there have been (through 2017) more than 80 published articles concerning aspects of the Bishop Tuff. Most of these are cited in papers by Wilson and Hildreth (1997), Hildreth and Wilson (2007), Chamberlain et al. (2014a, 2014b, 2015), Evans et al. (2016), and Hildreth (2017a). One thing not previously investigated in detail is the presence, addressed here, of sparsely distributed carbon-bearing pumice in the Bishop Tuff.

The present study was an outgrowth of our multi-year investigation (Hildreth et al., 2017) of the Early Rhyolite, the 100 km3 postcaldera assemblage that erupted in as many as 40 batches over an interval as long as ~125,000 years after caldera collapse and that was subsequently uplifted structurally to form Long Valley’s resurgent dome. Only at a late stage in preparing the Early Rhyolite paper did our growing appreciation for the importance of carbon in lowering magmatic oxygen fugacity of the postcaldera rhyolites prompt us to determine concentrations and isotope ratios of carbon and the presence of graphite in some Bishop Tuff pumice clasts, as now discussed in detail here.

The compositional zoning, complex eruptive sequence, sectorial stratigraphy, pumice- and lithic-clast variations, and petrology of the Bishop Tuff were presented in detail by Wilson and Hildreth (1997) and Hildreth and Wilson (2007). These features and our model for incremental growth of the magma body were reviewed by Hildreth (2017a).

The caldera-forming eruption continued for 5–6 days (Wilson and Hildreth, 1997) and resulted in 2–3 km subsidence of a 350 km2 roof plate (Hildreth, 2017b), which consisted of Paleozoic marine metasedimentary strata and Mesozoic granitoid rocks. Paleozoic strata of the Mount Morrison pendant (Fig. 1) make up most of the south and northwest walls of the caldera, and they form the buried floor of the western half of the caldera (see Map A of Sheet 2 of Bailey, 1989). The steeply dipping pendant rocks, Cambrian through Permian in age, include siliceous hornfels and quartzite, but they are dominated by black argillite, pelitic hornfels, argillaceous limestone, black chert, and slate (Rinehart and Ross, 1964; Greene and Stevens, 2002). Although the strata are deformed and contact metamorphosed in the hornblende-hornfels facies (Morgan and Rankin, 1972; Russell and Nokleberg, 1977; Ferry et al., 2001), sedimentary structures are widely preserved (Stevens and Greene, 1999). Sparse fossils are pelagic or planktonic, including conodonts, radiolaria, and graphitic graptolites, suggesting deposition in marine oxygen-deficient environments poor in scavengers. The black argillites, metapelites, and hornfelsed metasiltstones contributed much of the 1–4 vol% of lithic fragments (Fig. 2) ubiquitously distributed throughout the Bishop Tuff (Hildreth and Mahood, 1986; Wilson and Hildreth, 1997).

Proportions of a dozen pumice types were determined by clast counts at ~110 outcrops that represent all emplacement packages of the Bishop Tuff (figures 2 and 6, and table 1, of Hildreth and Wilson, 2007). Most of the pumice (93%–99%) is compositionally zoned high-silica rhyolite (74%–78% SiO2; 75–198 ppm Rb; 2–614 ppm Ba), which is white internally, though commonly oxidized externally during cooling of the ignimbrite, and it ranges in phenocryst content from ~1% to 25% by weight (figure 7 of Hildreth and Wilson, 2007). The first two-thirds of the eruptive sequence were dominated by crystal-poor pumice, but the proportion of crystal-rich pumice (having 12%–25% phenocrysts) increased systematically from <5% early in the eruption to ~90% in the final emplacement packages.

A small fraction of the pumice, averaging less than 1% of the pumice at most outcrops, is dark colored (black, dark gray, or pale smoky gray) and is poor in phenocryts (0–6 wt%). Of 26 dark pumice clasts chemically analyzed, 21 contain 71.5%–75.5% SiO2, three have 67.0%–70.6%, and two have 75.7%–76.1% (appendix 4 of Hildreth and Wilson, 2007). Most contain commingled blebs, streaks, and bands (Fig. 3) of normal white pumice (which we tried to exclude during sample preparation for analysis), but nearly all dark pumice clasts are crystal-poor rhyolite with 83–190 ppm Rb and 259–670 ppm Ba (with outliers at 850, 1310, and 1347 ppm Ba; Fig. 4). Dark pumice has been observed in nearly all packages of the ignimbrite, including Ig1Ea (the first emplaced) and Ig2N (the last). It is most conspicuous in packages Ig1Eb, Ig2Eb, and Ig2NWb of Hildreth and Wilson (2007). The mingling of dark and white pumice (Fig. 3) shows that the rhyolite magmas that produced them coexisted as liquids in the erupting magma reservoir.

Swirly pumice, as described by Hildreth and Wilson (2007), is intrinsically pale to medium gray but is commonly oxidized yellow, pale orange, or tan, only rarely pink. It has a lineated microvesicular fabric, which is typically wavy, twisted, or contorted and, less commonly, sheared into a planar or undulating microfoliation. Clast cores are often inflated by late growth of coarse equant vesicles (10–30 mm). Some clasts have denser rinds 1–3 mm thick, which are typically fissured or incipiently bread-crusted. Among the lowest-density clasts in the Bishop Tuff, swirly pumices tend to be locally enriched by rafting into pumice concentration zones. In contrast to such low-density swirly pumice, some dark pumice exhibits densities and fabrics similar to the predominant normal white pumice. Swirly pumice is roughly three times as abundant as dark pumice and makes up 1%–5% of the pumice clasts in some emplacement packages. Swirly pumice is not as phenocryst-poor as the dark pumice; it has a measured range from <1 wt% to 9 wt% crystals. The 71 swirly pumice clasts analyzed range from 71.5%–76.8% SiO2, 82–196 ppm Rb, and 41–729 ppm Ba. As with the dark pumice population, no correlation is recognized between composition and texture or crystal content. Abundant chemical data for dark and swirly pumice types are given in appendix 4 and table 2 of Hildreth and Wilson (2007), and both types are illustrated in their figure 4.

First defined by Bailey et al. (1976), the earliest postcaldera eruptive units are phenocryst-poor rhyolite lavas and tuffs that were investigated in detail by Hildreth et al. (2017). Drill holes show the Early Rhyolite (ER) to extend beneath younger units across most of the caldera, but it is exposed principally on the 10-km-wide resurgent dome (and Lookout Mountain), where all or most of its vents were located and where it is as thick as 622 m. We defined and mapped 14 exposed ER units (Hildreth et al., 2017), and geothermal drill holes penetrated as many as 22 more (altered) lava flows intercalated with voluminous rhyolite tuffs in the subsurface. The total ER volume is estimated to represent ~100 km3 of rhyolite magma, about one-sixth as much as the Bishop Tuff (BT). Ten dated units gave 40Ar/39Ar ages that range from ca. 750–640 ka (Hildreth et al., 2017). How closely the first ER batch followed the 767 ± 2 ka caldera-forming eruption has not been adequately determined, owing to uncertainties of 22–37 k.y. in its ca. 750 ka ages (table 3 of Hildreth et al., 2017).

As much as 80% of the ER volume is pyroclastic and, as observed in drill cores and surface exposures, most of that is virtually aphyric. Quantitative heavy-liquid separations of 27 glassy samples of ER lavas yielded 0–2.5 wt% crystals, mostly <1 wt% (table 2 of Hildreth et al., 2017). Many lavas have sparse plagioclase, orthopyroxene, and ilmenite, each species generally euhedral to subhedral. Some lavas have still sparser biotite, rare tiny grains of titanomagnetite, and sparse inclusions of apatite and sulfide. Traces of Pleistocene zircon were identified in heavy-liquid separates of four ER obsidians and rare Mesozoic zircons in five ER obsidians (J. Vazquez, U.S. Geological Survey sensitive high-resolution ion microprobe (SHRIMP) laboratory, 2016, written commun.). Quartz, sanidine, clinopyroxene, and amphibole are absent.

The ER phenocryst suite contrasts drastically with that of the BT, especially with its late-erupted packages, which are rich in crystal-rich pumice that contains 10–20 wt% quartz and sanidine along with ten minor phases (Hildreth and Wilson, 2007; Chamberlain et al., 2015). Processes such as decompression, crystal resorption, fluid ascent, heating by non-rhyolitic recharge batches, and ascent of new batches of rhyolite melt from a deep reservoir of granitic crystal mush could have promoted the great change in phenocryst abundance and speciation during postcollapse reorganization of the ER magma reservoir, were discussed in Hildreth et al. (2017).

Chemical composition of the ER is remarkably uniform, especially in that it erupted as numerous batches over an interval more than 100,000 years long. For 83 fresh obsidian samples, the full range of SiO2 content is 74.3%–75.0%, of Al2O3 13.3%–13.6%, of K2O only 5.12%–5.26%, of Rb 132–147 ppm, and of Th 15–18 ppm. Titanium, Sr, Ba, and Zr have somewhat wider ranges, thus permitting chemical discrimination among look-alike eruptive units (table 1 and figures 4, 5, and 12 of Hildreth et al., 2017).

Figure 4 compares the concentration ranges of the abruptly erupted, compositionally zoned BT with the ranges shown by 108 fresh ER obsidians and felsites, which erupted intermittently over a protracted postcaldera interval (ca. 750–640 ka) at the site of what then became Long Valley’s structurally resurgent dome. Early Rhyolite concentration ranges are far narrower than those of the continuously zoned BT, and for most elements they are intermediate along the BT arrays, rather than extending to higher or lower concentrations. Exceptions are the ER ranges of Ba, Zr, Hf, and Eu/Eu* (as well as the isotope ratios of Nd and Pb), which all extend beyond BT arrays. The BT arrays represent dense continua, not bell curves, of >300 samples of the predominant white pumice. Abundant data for the minor variants, dark and swirly pumices, also fall within the BT arrays for most elements but extend to higher concentrations of Ba, Sr, TiO2, FeO*, and CaO, as indicated in the right-side column of Figure 4.

Another noteworthy distinction between BT and ER is the absence of xenoliths in ER lavas. Fragments of intracaldera BT and rounded cobbles of Sierran basement rocks (probably entrained from intracaldera pre-BT Sherwin Till) are sparsely present in ER pyroclastic deposits, but, unlike wall-rock lithics found within BT pumice, no lithic fragments (other than dense cognate ER) are found that had been engulfed in ER magma. The disposal of collapse-related roof-rock lithics that foundered syneruptively into the magma chamber is addressed in the concluding discussion, below.

Small amounts of noncarbonate biogenic carbon (δ13C = –25‰ VPDB) had been reported in BT pumice by Hildreth (1985), principally in the sparse variant pumice types that were later characterized as “dark or swirly” (by Hildreth and Wilson, 2007). Carbonate content, determined separately, was negligible (Table 1), so most of the carbon in the pumice was inferred to be graphite ingested from the foundered metasedimentary roof rocks. Recent work on the 100 km3 postcaldera ER revealed that oxygen fugacity values of the ER, as determined by analysis of coexisting titanomagnetite-ilmenite pairs, are as much as one log unit more reduced than those of the Bishop Tuff (Waters and Lange, 2014, 2016; Hildreth et al., 2017). This discovery prompted renewed investigation of the reduced carbon, as now reported here.


For the current investigation, we selected dark and swirly pumice clasts from all BT sectors and from early, intermediate, and late stages of the 5- to 6-day eruptive sequence. We also chose a typical hornfelsed metapelitic lithic clast from early in the BT sequence and argillite samples from the Ordovician Convict Lake Formation, where it strikes into the south wall of Long Valley caldera.

Well-trimmed interiors of individual pumice clasts chosen for their fresh appearance were analyzed for total C content (Table 1). Pumice was collected with care to avoid caliche or visible groundwater effects.

The 1984–1985 data were generated at U.S. Geological Survey (USGS) laboratories (Menlo Park; Paul Klock, analyst). Total carbon was measured using a Leco WR-12 carbon analyzer with an induction furnace. Trace or negligible amounts of carbonate were measured by acid digestion, and noncarbonate carbon was calculated by difference (Jackson et al., 1987). This older data set is less precise but is included in Table 1, in part because that reconnaissance effort provided impetus for the modern investigation.

Present-day samples were measured for carbon content by the Stable Isotope Core Laboratory at Washington State University (WSU; Total carbon was determined for ~100 mg dried pulverized samples by combustion at 980 °C and conversion to CO2 with an element analyzer (ECS 4010, Costech Analytical) coupled to a continuous-flow isotope-ratio mass spectrometer (Thermo Finnigan Delta Plus XP). For recent measurements, the mass spectrometer was the detector for total carbon and carbon-isotope ratios. Separate powder aliquots were triply washed with 3N phosphoric acid over a 3-day period to remove any carbonates. Little or no fizzing was observed, and total weight losses were small following acid washing. On comparing acid-washed samples to untreated samples (Table 1), however, reduction in total carbon content was observed in most samples. Lowering of carbon content by acid washing is thought to have been caused by loss of fines, as commonly observed during repeated acid treatment and rinsing. Trace amounts of nitrogen were also detected in a few samples, as similarly measured after combustion and conversion to N2 gas. Contents of pumice samples ranged from 0.001 to 0.046 wt% N, perhaps reflecting slight surface contamination by plant-derived dust or microbial colonization.


Dark-gray to black pumice clasts analyzed at WSU (n = 5) gave 0.059–0.992 wt% total carbon, and those analyzed at the USGS (n = 7) gave 0.02–0.35 wt%. Swirly pumice clasts are generally poorer in carbon, yielding 0.063–0.152 wt% at WSU (n = 3) and 0.02–0.08 wt% at the USGS (n = 2). Analyzed for comparison with the minor pumice variants were a white high-silica rhyolite pumice that erupted early in the Bishop Tuff sequence (B-619D; 0.078 wt% C) and a pale-gray pumice, which was also emplaced early in the eruption (B-624; 0.096 wt% C). A white aphyric pumice clast from the early postcaldera rhyolite (M-952) is still poorer in total carbon (0.058 wt%).

Of the dozen ignimbrite emplacement packages defined by Wilson and Hildreth (1997), the earliest (Ig1Ea) has the largest number of lithic fragments per unit area (Fig. 2), and they consist predominantly of hornfelsed metapelite, metasiltstone, and argillite, which were torn from the Cambrian to Permian strata that collapsed into Long Valley caldera during eruption of the Bishop Tuff (Fig. 1). All the packages contain such black metasedimentary lithic fragments, but in later emplacement packages, they are diluted by suites of basaltic and rhyolitic lithics torn from other segments of the collapsing ring fault (Hildreth and Mahood, 1986). A metapelite lithic (B-706) from Ig1Ea yielded 0.678 wt% C, a carbon content greater than all but one of the pumice clasts analyzed. We also sampled Ordovician black argillite from the Convict Lake Formation (Greene and Stevens, 2002), in place on the north wall of Convict Lake, where near-vertical Lower Paleozoic strata strike into the nearby caldera ring fault. The graptolite-bearing black argillite (B-707-a and B-707-b) contains 2.53–2.89 wt% C (Table 1), an order of magnitude more carbon-rich than most of the Bishop Tuff dark pumice.

Carbon stable-isotope results (Table 1) are reported in ‰ relative to Vienna Peedee belemnite (VPDB). Five dark pumices range widely in δ13C, from –21.7‰ to –29.1‰, whereas two swirly pumices gave –25.0‰ and –25.2‰. White and pale-gray BT pumice clasts yield δ13C values of –25.8‰ and –26.2‰. A postcaldera ER white pumice gave a relatively heavy δ13C value of –21.7‰. Values of δ13C are –27.5‰ to –25.7‰VPDB for in situ Ordovician argillite and –21.0‰ for a metapelitic lithic fragment entrained in Ig1Ea.

All the C-isotope values are biogenic, with little hint of dilution by carbon from marine carbonate (δ13C near zero) or volcanic sources (δ13C = –4 to –8). Contamination of the pumice by authigenic soil carbonate (including caliche or calcrete), which typically has δ13C values between +1 and –10 (Cerling, 1984), appears to be negligible. No fizzing was observed during acid wash. The carbon present in the Paleozoic wall rocks apparently originated in marine mudrocks. Disseminated sediment derived from marine algae (δ13C typically –20 to –22) would have been an important component of the Cambrian and Ordovician strata. Mosses and aquatic plants present by the Late Ordovician were augmented in the Silurian by photosynthetic C3 vascular land plants, for which δ13C was typically –25‰ to –27‰ (Strauss and Peters-Kottig, 2003). Silurian, Devonian, and Upper Paleozoic marine strata are also components of the Morrison pendant sequence that collapsed into the caldera (Greene and Stevens, 2002).

Carbonaceous material derived from the primary organic matter present in sediments, principally in mudrocks, is transformed to graphite during metamorphism. At low temperatures, polymerized multicomponent kerogens form, and, with increasing grade, loss of H, O, N, and S is accompanied by progressive structural ordering of crystalline graphite (Buseck and Huang, 1985; Beyssac et al., 2002). In the medium-grade hornblende-hornfels facies characteristic of the Mount Morrison pendant (450–600 °C at upper-crustal pressures of ~1.5–2.5 kb; Ferry et al., 2001), conversion to well-crystallized graphite is expected to have been advanced, though not necessarily complete. Contact metamorphism of the Paleozoic mudrocks here took place during intrusion of extensive granodiorite plutons of both Late Triassic and Late Cretaceous age (Bateman, 1992).

Experimental Methods

We used both single-crystal and bulk-powder crystallographic methods in order to determine the form of organic carbon present in the pumices. Analysis by micro-Raman spectroscopy allows for focused analyses on individual crystals in a rock sample or crystallites within a prepared powder (Fig. 5), providing direct measurements of graphite crystals and other minerals. X-ray diffraction provides true bulk mineralogical analysis of a powdered sample, from which the presence or absence of a mineral phase is observed.

Mineralogical analyses were performed on lithic-clast sample B-706, wall-rock sample B-707-b, and on five dark-gray pumice samples: B-701, B-702, B-704a, B-704b, and B-705. Fragments of each rock sample were powdered by hand in an agate mortar, and portions of these powders were analyzed using micro-Raman spectroscopy and X-ray diffraction (XRD), in addition to analyzing rock fragments of wall-rock samples by micro-Raman spectroscopy. Powdered samples are representative of bulk-rock mineralogy and included quartz and feldspar crystals and glass, as well as other minor minerals.

Laser Raman spectroscopic measurements were obtained at room temperature using a DXR Raman microscope spectrometer. All spectra were taken using a 532 nm diode-pumped, solid-state (DPSS) laser with a laser power of 10 mW (but 5 mW for sample B-706 and 2 mW for B-705, because these were fine powders that would burn under higher power), a filter of 532 nm, a grating of 900 lines/mm, and a resolution of 5.5–8.3 cm–1 at 100× magnification. The instrument was calibrated using a polystyrene standard. The spectrograph aperture was set to a 25 μm or 50 μm pinhole, dependent upon how many counts were measured, and the measurement range was 30–3579 cm–1. Automatic baseline subtraction was set to a polynomial order of 6 with 50 iterations. Samples were mounted on glass slides, as either pressed powder pellets or as rock fragments, and viewed under reflected light. Data were acquired using OMNIC Atlμs imaging software.

A Rigaku Multiflex X-ray diffractometer with Cu-Kα radiation (wavelength 1.54059 Å) with an accelerating voltage of 40.0 kV and a filament current of 20.0 mA measured X-ray diffraction patterns of randomly oriented powder mounts at room temperature. Powders were scanned from 2° 2θ to 70° 2θ with a step size of 0.01° 2θ and a count time of 0.6 seconds per degree. Peaks, d spacing, and intensity were determined graphically using the program Theta.XRD version 1.1.6 (Whitehouse, 2015).

Results: Micro-Raman Spectroscopy

The covalent C-C bonds in graphite have diagnostic vibrational energies measurable as peaked bands by micro-Raman spectroscopy. Three prominent bands are the G band at 1582 cm–1, the D band at ~1350 cm–1 (which indicates disorder in the graphite structure), and the 2D (or G′) band at ~2700 cm–1 (Hodkiewicz, 2010). Manufactured graphite from mechanical pencil lead was measured as an internal standard on the DXR micro-Raman using the same conditions as the Bishop Tuff pumice and wall-rock samples. Diagnostic G, D, and 2D bands were located at 1582 cm–1, 1343 cm–1, and 2715 cm–1, respectively (Fig. 6). The internal standard provides a good baseline to compare to analyses of the wall-rock samples (B-706 and B-707-b) and Bishop Tuff pumice samples (B-701, B-704-a, and B-705) (Fig. 6).

Samples B-706 and B-707-b (a lithic clast of black hornfels and Sierran basement argillite, respectively; Table 1) were analyzed as rock fragments to measure graphite in situ; sample B-707-b was further analyzed as a pressed powder pellet. Dark-looking areas that often included bright white, highly reflective flecks (Fig. 5) were selected for analysis. Each point analysis, less than one micron in diameter, was run for ~45 seconds (15 exposures for 3 seconds each). A minimum of five analyses in each sample have well-defined typical graphite D, G, and 2D bands at 1350–1359 cm–1, 1579–1593 cm–1, and ~2718 cm–1, respectively (Fig. 6). The sample was changed after at least five diagnostic graphite analyses. An additional seven analyses of the two wall-rock samples are amorphous in character with much broader peaks and larger D bands (disorder-induced Raman band) at ~1350–1359 cm–1. Twelve other analyses were mixtures of graphite and other minerals with additional lower Raman shift peaks.

Pumice samples B-701, B-704a, and B-705 were analyzed as pressed powder pellets created with a hand press with a 7 mm die in order to maximize the measurable surface by eliminating vesicles. Laser conditions of 15 exposures for 3 seconds each resulted in slightly better quality measurements than 45 exposures for 1 second each and avoided burning the sample; 512 background exposures were collected.

Out of ~70 ~0.7 μm point analyses of sample B-701, over a dozen analyses had characteristic G, D, and 2D bands of graphite, two were amorphous in character, and many mixed spectra also included lower Raman shift peaks indicative of other minerals, typically quartz or hematite. Of 98 point analyses of sample B-704-a (Fig. 5), eight had characteristic G and D graphite bands, four of which are wide bands with high D-band intensity, suggesting a more amorphous character (Kagi et al., 1994; Cuesta et al., 1998). Forty-five analyses of sample B-705 included five analyses with characteristic G, D, and 2D graphite bands in addition to an amorphous spectrum (Fig. 6; Supplemental Item A1).

Crystallite size and laser conditions may contribute to the quality of results obtained through micro-Raman analysis, because powders were prepared by hand and were not sieved to uniform size. High background noise in analyses of samples B-704-a and B-705 was caused by adjustments of the aperture and laser power (down to 2 mW) to avoid burning the graphite. Nonetheless, graphite is certainly present in both Bishop wall-rock samples (B-706 and B-707-b) and in three Bishop pumice samples (B-701, B-704-a, and B-705; Fig. 6).

Metamorphic Temperatures from Raman Spectra for Graphite

Graphite has been used as a geothermometer in metasediments (Beyssac et al., 2002). Using areas under the peaks of measured Raman spectra for graphite, peak metamorphic temperature can be estimated from the degree of organization of graphite crystals, as demonstrated by Beyssac et al. (2002). They correlate the R2 value, which represents the relative area of the D band, to temperature in their equation 1. We measured Raman spectra peak positions, intensities, and areas, and we used that equation to calculate a maximum metamorphic temperature of 613 °C for the wall-rock argillite samples (Supplemental Item A [see footnote 1]). Maximum temperatures calculated for graphite crystals from a lithic fragment within the Bishop Tuff and from the dark-gray BT pumice samples are lower—479 °C and 582 °C, respectively. Although still within the range anticipated for the hornblende-hornfels facies (Ferry et al., 2001), these values may not represent peak metamorphic or crystallization temperatures, but, instead, they could reflect disorder in the graphite induced by volcanic processes during incorporation into the BT.

Results: X-Ray Diffraction

The X-ray diffraction (XRD) profiles of quartz and graphite have nearly overlapping diagnostic peaks; quartz diffracts so strongly that the graphite peak is often difficult to resolve. Analyses of bulk rock powders by XRD nonetheless confirm the presence of graphite (as well as quartz) in dark-gray Bishop Tuff pumice samples and in the argillitic country rocks, because secondary peaks of both minerals are recognizable in most patterns.

A missing graphite peak at 44.56° 2θ (2.03Å d spacing) is not surprising, because the small crystallites of graphite in these powders are likely tabular in form, as seen in micro-Raman spectroscopy (Fig. 5), which could result in a preferred orientation of the sample to be flat lying, thus excluding the (101) crystal face that corresponds to 2.03Å d spacing. Albite peaks are also present in all samples, as is a broad bump signifying amorphous glass in the pumice. No other carbon-bearing phases were observed in the XRD profiles, confirming the virtual absence of carbonate, as reported above in discussion of Table 1.

Supplemental Item B (see footnote 1) lists the observed 2θ angle, calculated d spacing, and normalized intensity of peaks measured by X-ray diffraction for selected samples, and it includes diagnostic 2θ angles, d spacings, and peak intensities of quartz and graphite from standard references (Wyckoff, 1963; Trucano and Chen, 1975; Anthony et al., 2001).

EDS Methods

Electron dispersive spectroscopy (EDS) analysis was undertaken to confirm the presence of graphitic carbon in the glass phase of BT pumice and to examine further its spatially variable distribution. Samples of the lithic clast (B-706), three pumices (B-701, B-702, and B-704-a), and the argillite (B-707-b) were mounted onto one-half-inch aluminum stubs using double-sided copper (Cu) tape. They were then coated with a ~10 nm layer of Au-Pd for conductivity, which does not interfere with detection of light elements. Coated samples were analyzed in a Tescan VEGA3 scanning electron microscope (SEM) using Oxford AZtec software and a 50 mm X-MaxN EDS detector. Electron dispersive spectroscopy analyses were performed in both spot and area mode, with an accelerating voltage of 10 kV and a current of 15 nA and live count times of 60 s.

EDS Results

All of the samples show a wide range of apparent C contents (Fig. 7). The argillite (sample 707-b) shows the highest range of C contents, which was expected. The C signal is roughly an order of magnitude stronger than seen in the pumices and is present and fairly consistent throughout the sample. Numerous analyses of the lithic and the pumices indicate that the C signal ranges from a significant peak to none at all. The estimated C concentration in spots and areas containing significant C in the pumice and lithic samples is ~5–20 wt%. The variability of C peak heights in all of the samples suggests that the presence (or absence) of C is real and not just a background or interference phenomenon. Both types of analyses (point and area) show a wide range in C peak heights, which suggests that the size and type of area sampled are not alone influencing its apparent C content. As illustrated in the Raman image (Fig. 5), graphite in BT pumice is highly heterogeneous in its distribution.

Titanomagnetite-ilmenite pairs separated from several Early Rhyolite (ER) samples were analyzed by electron microprobe (appendix E of Hildreth et al., 2017), yielding temperatures in the range 752–844 °C and oxygen-fugacity values of ΔNNO = –0.32 to –0.65 (Fig. 8), as calculated by the method of Ghiorso and Evans (2008) for pairs that passed the Mn-Mg exchange equilibrium test of Bacon and Hirschmann (1988).

Similar ranges of T and fO2 were determined for ER samples by Waters and Lange (2014, 2016), who stressed that fO2 values for the ER are more reduced than those of the Bishop Tuff (BT), in spite of considerable overlap in Fe-Ti–oxide temperature (~700–820 °C for the BT, as recalculated by Ghiorso and Evans, 2008). Note that the recalibration by Ghiorso and Evans (2008) used the nickel–nickel oxide (NNO) oxygen buffer curve of O’Neill and Pownceby (1993), which renders ΔNNO values for the BT array ~0.5 log units more reduced than portrayed in figure 15 of Hildreth and Wilson (2007); that figure employed the NNO curve of Huebner and Sato (1970). As also plotted in figures 23 and 29 of Ghiorso and Evans (2008), the BT array (representing ~105 pairs) extends smoothly in ΔNNO values from –0.6 at ~700 °C to +0.6 at ~820 °C, crossing the NNO curve at ~760 °C (Fig. 8). The positive slope of the BT T-fO2 array (and its implicit positive T-αTiO2 slope) were discussed by Evans et al. (2016). Our determinations for ER pairs plot 0.5–1.0 log units below (more reduced than) the BT array (Fig. 8).

Because Fe-Ti–oxide temperature ranges overlap (Fig. 8), and concentration ranges of all major elements in the ER are within those of the zoned BT (Fig. 4), what can account for the postcaldera drop in fO2 of otherwise similar rhyolitic magmas that erupted sequentially at the same place? Waters and Lange (2014) reasonably suggested that a change in basalt conceivably parental to the rhyolites—from precaldera dominance of basalts influenced by subduction-modified lithosphere to postcaldera input of basalt from a more reduced asthenospheric source—might have accounted for the postcaldera drop in fO2 of Long Valley rhyolite. Neither chemical nor isotopic data (Hildreth et al., 2017), however, provide evidence for an asthenosphere-derived basaltic contribution (e.g., 87Sr/86Sr < 0.7045, εNd >3, TiO2 >2.0, and lack of Nb-Ta deficiency) to Long Valley magmas. See Rogers et al. (1995), Cousens (1996), Bailey (2004), Hildreth et al. (2014), and Hildreth and Fierstein (2016b) for data on all the mafic lavas within and near Long Valley.

A more likely process that could have lowered fO2 in postcaldera rhyolites is reaction with graphite from the metasedimentary roof rocks that foundered into the magma chamber during caldera collapse (Figs. 1 and 2). Shattered roof rocks that collapsed into the magma body need not have melted for some fraction of their reduced carbon to have reacted with the rhyolitic magma, producing CO2 and lowering the oxygen content and Fe3+/Fe2+ of the affected magma. Graphite has exceptionally high thermal stability and will not melt at magmatic temperatures. However, it will oxidize to CO2 at 700 °C or higher.

The graphite so erratically dispersed in a small fraction of the BT pumices (as shown in Fig. 5 and Table 1) was introduced during (or soon before?) the eruption and thus had little time to react with the magma. Values of fO2 for the carbon-bearing dark and swirly BT pumice samples were not reduced below those of the whole BT array (Fig. 8). The postcaldera ER batches that erupted from the reorganized magma reservoir, from soon after to as long as 125,000 years later, however, had ample time for reaction to have consumed graphite and reduced fO2 of the rhyolite magma.

All nonwelded BT pumice is hydrated, most samples yielding loss on ignition (LOI) values between 2 and 4 wt% (Fig. 8 of Hildreth and Wilson, 2007), so there was opportunity for secular groundwater contamination. However, the negligible carbonate contents of most pumices and the similar δ13C values of untreated and acid-washed splits (Table 1) show that the elevated carbon contents of dark and swirly pumices predated groundwater effects and surface exposure. The irregular distribution and widely varied concentration of unreacted graphite, conspicuous in only 1%–2% of the BT pumice (mainly in dark pumice, less in swirly), suggest that it was introduced either during or slightly before the caldera-forming eruption and only very locally—inferentially near contacts of metasedimentary roof rocks with the magma chamber.

As noted by Hildreth et al. (2017), Chamberlain et al. (2014a) had reported Mesozoic zircons to be more common in swirly pumice than in the predominant white BT pumice. (No data were obtained for zircon, if there is any, in dark BT pumice.) Compositional contrasts between Mesozoic and Pleistocene zircons in the BT led them to infer that the former were roof-rock xenocrysts, probably from Mesozoic granitoids, as Reid and Coath (2000) had previously inferred. Lack of rim overgrowths on the Mesozoic zircons suggests late-stage incorporation in the BT magma, potentially syneruptively. Magmatic reaction with Paleozoic metapelites from the chamber roof is similarly likely for the graphitic component in the BT rhyolite. Interaction of the postcaldera rhyolite magma with shattered graphite-bearing roof rocks (and the roof plate itself) that foundered during the caldera-forming collapse evidently reduced the fO2 of the subsequently erupted ER.

Additional impacts of the collapse on the remaining magma, including other possible effects of roof-rock contamination, were discussed at length by Hildreth et al. (2017) and can be summarized here with respect to Figure 4. The limited and intermediate ranges of most elements in the ER compared to the extended ranges of the zoned BT (Fig. 4) were attributed to postcollapse convective homogenization of the ER magma rather than to contamination. Slightly elevated values of Ba, Zr, Hf, and Eu/Eu* could reflect interaction with granitoid roof rocks (or, alternatively, modest recharge by more alkalic silicic magma). The shift in isotope ratios of Nd and Pb (Fig. 4) to slightly more upper-crustal values could reflect interaction with Mesozoic granitoids, but data are inadequate to verify this. The change from phenocryst-rich BT to nearly aphyric ER, which completely lacks the quartz and sanidine so abundant in the BT, could reflect thermal or volatile-component effects (Hildreth et al., 2017) but is unlikely to have been promoted by reaction with cold roof rocks.

An estimate of the amount of graphite-bearing metasedimentary rock ingested by the residual magma left behind after the caldera-forming eruption of the BT can be obtained by calculating the amount of Fe2O3 and FeO in the last-erupted part of the BT and in the residual magma that subsequently erupted as postcaldera ER.

Equation 7 of Kress and Carmichael (1991) is an extended empirical equation that combines thermodynamic equilibrium for iron oxidation in silicate melts with an empirical Fe-redox equilibrium equation at one atmosphere. Using temperature, pressure, and oxygen fugacity values for a given melt composition along with the constants given in their table 7 associated with that equation, one can calculate the ratio of the mole fraction of Fe2O3 to the mole fraction of FeO under those conditions and then determine wt% Fe2O3 and wt% FeO from the measured wt% FeO*. (The value for To we used, which was otherwise missing from the table, was 1673 K (1400 °C), provided to the authors by Tom Sisson (USGS, September 2016, personal commun.). We thus calculated the FeO and Fe2O3 contents of the late BT and the ER, based on an estimated pressure of 2 kilobars, a medial temperature of 795 °C, and corresponding values (for that temperature; Fig. 8) of log fO2 (–13.2 for late-erupted BT and –14.25 for the ER). We used the average major-element composition of the late BT to represent compositions of both the late-erupted BT itself and of the rhyolitic magma that remained in the chamber at the end of the caldera-forming eruption. This magma was subsequently reduced by reaction with graphite from the collapsed roof rocks and erupted episodically over the next ~125,000 years as batches of ER. Using wt% of Fe2O3 and FeO for oxidized BT and reduced ER as calculated from equation 7 of Kress and Carmichael (1991), the resulting difference in total wt% represents oxygen lost as CO2 gas by reduction of the magma.

If carbon reduces the magma so that oxygen is lost as CO2, then an average of ~50 ppm carbon in the residual magma is needed to reduce the magma left behind by one log unit fO2, as observed in the ~100 km3 of magma later erupted as the ER. If all the oxygen were instead lost as CO, then about twice that amount, ~101 ppm carbon, would have reacted with the magma volume erupted. Details of these calculation procedures are expanded in Supplemental Item C (see footnote 1).

If the Paleozoic metasedimentary rocks contained an average of 2.5 wt% carbon (Table 1), then a maximum of only ~0.21 km3 of that graphite-bearing metasediment needed to have reacted with the ~100 km3 of ER magma that later erupted. Proportionally, that represents a much smaller fraction than the lithic fraction actually observed in the BT (Fig. 2).

Such estimates assume that the reducing reaction takes place in a wholly liquid state, because we have used average bulk compositions to represent liquid compositions. However, the amount of oxygen lost is almost identical (or slightly lower) if we assume that all of the reduction occurs in the iron-bearing minerals instead of the melt. Using the compositions of titanomagnetite-ilmenite pairs that record similar temperatures but a log unit difference in fO2—from the late-erupted part of the BT and from the ER (respectively in appendix 3 of Hildreth and Wilson, 2007 and appendix E of Hildreth et al., 2017), calculation of respective wt% of Fe2O3 and FeO in the oxide minerals indicates that ~56 ppm carbon is needed to account for the oxygen lost as CO2 during reduction of those minerals (Supplemental Item C [see footnote 1]). The estimates ignore several complexities in the reduction reactions, including the coexistence of trace amounts of other Fe- and Ti-bearing minerals (hypersthene and biotite in ER) and minor differences in bulk FeO* between the BT and ER samples considered. Nonetheless, the contrasting approaches, based on bulk compositions and on mineral compositions, yield carbon-oxygen budget estimates that are acceptably similar.

Geologic uncertainties limit these nominal calculations to rough approximations. But it seems clear that only a small fraction of graphite-bearing country-rock contaminant is required to have reacted with the residual magma to reduce its fO2 by one log unit.

It can be inferred from the great height of the caldera walls, from the 2–3 km depth of roof-plate subsidence, and from the 1–4 wt% lithic clasts in the BT (Fig. 2) that tens of cubic kilometers of country rocks actually collapsed into the residual magma chamber, which itself was probably much larger than the 100 km3 of ER that eventually erupted—the volume necessarily used for our calculation.

It can also be inferred that reaction with graphite took place throughout the entire eruptible volume of the residual postcaldera magma reservoir, after caldera collapse accompanied eruption of the unreduced BT but before any of the reduced ER erupted. All eruptive batches of ER, which were released episodically over a ~125 k.y. interval following the 767 ka caldera-forming eruption, yield Fe-Ti–oxide values of T and fO2 that plot close to a common curve (Fig. 8) that roughly parallels NNO and falls about one log unit more reduced than the BT array. We conclude that graphite in the caldera’s metasedimentary country rocks (Figs. 1 and 2) provided the carbon that reduced the postcaldera residual magma.

Heterogeneous distribution of graphite in dark BT pumice and relative reduction of fO2 in the postcollapse ER rhyolites are now well established. Aspects of the physical processes involved in promoting graphite-rhyolite reaction, however, remain uncertain and invite further discussion. Several observations provide constraints:

  • (1) A small fraction of the BT magma (principally that which produced the dark pumice) had ingested roof-rock graphite before or during the caldera-forming eruption, but its residence time was insufficient to modify fO2 in the BT magma affected.

  • (2) An enormous mass of roof rocks subsided syneruptively into the partially evacuating magma chamber, principally as the roof plate but also as many cubic kilometers of shattered lithic fragments torn from the ring-fault zone and observed ubiquitously in fall deposits, ignimbrite outflow sheets, and intracaldera tuff.

  • (3) All ER batches, from first to last, that yielded T-fO2 data from Fe-Ti–oxide pairs, exhibit reduced fO2 values well below the NNO curve (Fig. 8). This persistence suggests that the graphite reaction took place soon after the caldera-forming eruption and established relatively reduced magmatic conditions that endured for at least 125,000 years.

  • (4) Mesozoic zircons in BT fallout (Reid and Coath, 2000) in swirly BT pumice (Chamberlain et al., 2014a) and in several ER obsidians suggest assimilative partial melting of granitoid roof rocks.

  • (5) Graphite itself would not melt at rhyolitic temperatures but would oxidize to CO2 at ~700 °C or higher.

  • (6) The metasedimentary rocks that dominate the BT lithic suites are not shales but, rather, amphibolite-grade hornfels and metasiltstone that consist of quartz, calcite, wollastonite, diopside, feldspar, and trace amounts of several other phases (Ferry et al., 2001). Conditions under which such dry rocks might melt are uncertain.

  • (7) Xenoliths and xenocrysts are rare or absent in the many ER magma batches erupted.

  • (8) Nd- and Pb-isotope data for ER (Fig. 4) show small shifts toward more upper-crustal values than those of the BT (Hildreth et al., 2017).

  • (9) On the contrary, abundant Sr-isotope determinations for ER are completely overlapped by the BT array (Fig. 4), suggesting little or no contamination by roof-rock Sr. This is noteworthy because abundant 87Sr/86Sr determinations for metasedimentary rocks of the Morrison pendant (Kistler and Peterman, 1973; Goff et al., 1991) range from 0.709 to 0.725, well outside the narrow 0.70655–0.7068 range of 22 ER samples (Hildreth et al., 2017, and references therein).

So, can it be that the graphite-bearing metasedimentary roof rocks that foundered into the collapsing magma chamber did not melt? Because such mudrocks generally have much higher ranges of δ18O than rhyolites, oxygen-isotope data can provide a critical insight. McConnell et al. (1997) gave δ18O values of 14.16‰ and 19.52‰ for hornfelsed metasedimentary samples from the Morrison pendant just south of the caldera wall, and Ferry et al. (2001) gave δ18O ranges for Morrison pendant quartz of 13.7‰–19.6‰ in sandstone and 13.7‰–15.8‰ in hornfels.

For Bishop Tuff quartz, the phase least likely to have experienced posteruptive isotopic modification, Holt and Taylor (1991) gave δ18O values of 8.30‰–9.04‰, McConnell et al. (1997) gave a value of 7.90 ± 0.19‰, and Bindeman and Valley (2002) made ~90 δ18O determinations for quartz phenocrysts from most BT emplacement packages (of Hildreth and Wilson, 2007), including: 30 quartz crystals from fall units (7.72‰–8.69‰); six from Ig1 ignimbrite packages (8.16‰–8.33‰); and 56 determinations for quartz from ignimbrite packages Ig2 (7.84‰–8.59‰). They stated an analytical precision of 0.1‰ or better. The last set included three determinations of quartz (8.35‰–8.47‰) from a dark pumice atop Aeolian Buttes, the same outcrop where our dark pumice sample B-146 was taken (Table 1 and Fig. 3).

For the Early Rhyolite (ER), which lacks quartz, Smith and Suemnicht (1991) gave δ18O values of 6.7‰–8.2‰ for six surface samples of fresh obsidian, averaging 7.4‰. McConnell et al. (1997) gave 8.38 ± 0.68‰ for an ER surface obsidian. And for two ER surface obsidians, Bindeman and Valley (2002) gave δ18O values of 8.09‰ and 8.11‰. The δ18O ranges of BT and ER thus show extensive overlap and provide no evidence for mixing of rhyolite with high-18O metasedimentary basement rocks in response to caldera collapse.

Small postcaldera shifts in Pb- and Nd-isotope ratios (Fig. 4) and the free Mesozoic zircon crystals sparsely found in both BT and ER are permissive of contributions from roof-rock granitoids (Fig. 1), presumably by partial melting. Lack of any significant difference in Sr- and O-isotope ratios between BT and ER, however, weighs heavily against extensive contamination by foundered hornfelsed metasedimentary rocks, whether by melting, disaggregation, or exchange. It appears likely that graphite in the hornfels reacted with the 750–850 °C rhyolitic magma to yield CO2 but that the bulk hornfels did not itself melt.

With no evidence for melting of the metasedimentary roof rocks (other than the sparse dark BT pumice itself), and with no xenoliths in the ER magmas, how were fragments of collapsed material disposed of? (1) One possibility is that little or no such shattered material actually entered the magma chamber, that lithic fragments produced along subsiding ring faults were completely entrained upward by the erupting BT, and that the graphite that reacted with the postcollapse ER magma was limited to the metasedimentary rocks of the 350 km2 roof plate. (2) A second conceivable process is that foundered roof fragments sank below the upper levels of the ER chamber from which the successive batches of ER magma were tapped. If the 750 ka age of the earliest ER batch were accurate, a nominal 17,000 years after the 767 ka caldera formation might be long enough for such settling. (3) Our favored scenario, as advanced in Hildreth et al. (2017), is that the crystal-poor ER melts separated from a pluton-scale reservoir of granitic crystal mush, which was the principal remainder of the BT chamber, and they ascended to the roof zone of the reorganized reservoir where they concentrated directly against the roof plate. Such buoyant rhyolitic melt ascent would leave below any crystal and lithic residue, would favor eruption of only nearly aphyric melts (as observed), would enhance contact with the granitic and metasedimentary rocks of the roof plate, and would help promote the intracaldera structural resurgence that took place late in ER time (Hildreth et al., 2017).

Uncertainties about mechanisms and timing remain, but the role of metasedimentary graphite in modifying the redox state of postcaldera magma is well substantiated, at least for Long Valley caldera. There may be 100 Cenozoic calderas in the Great Basin of Utah and Nevada (Best et al., 2013, and numerous references therein), and in the eastern two-thirds of that region most of the silicic magma chambers they overlay were hosted by Paleozoic metasedimentary and sedimentary rocks. Those calderas provide ample opportunities for geochemical investigation of the varied effects of caldera collapse. To the tools provided by radioisotopic and stable isotopes and by whole-rock and mineral analysis, the search for fO2 anomalies and for traces of graphite can now be added.

Thanks to Andrea Foster for access to the Micro-Raman Laboratory (U.S. Geological Survey, Menlo Park, California) and guidance on measuring graphite peaks; to Diane Moore for advice on X-ray diffraction measurements; to Leslie Hayden for the EDS determinations; and to Tom Sisson for suggesting and guiding the redox budget calculation. Becky Lange impressed us with the importance of reduced fO2 in the Early Rhyolite, and Judy Fierstein contributed to field mapping the Early Rhyolite in detail. Helpful comments by Ilya Bindeman, Katy Chamberlain, David Damby, Shan de Silva, Allen Glazner, Stephanie Grocke, Jocelyn McPhie, and James G. Moore substantially improved the manuscript. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

1Supplemental Items. Item A: Micro-Raman analyses of graphite in Bishop samples. Item B: Representative XRD analyses of powdered Bishop samples. Item C: Amount of carbon needed to reduce the residual magma, calculated based on wt% Fe2O3 and FeO (equation 7, Kress and Carmichael, 1991). Please visit or the full-text article on to view the Supplemental Items.
Science Editor: Shanaka de Silva
Gold Open Access: This paper is published under the terms of the CC-BY-NC license.

Supplementary data