The Upper Crustal Evolution of a Large Silicic Magma Body: Evidence from Crystal-scale Rb^Sr Isotopic Heterogeneities in the Fish Canyon Magmatic System, Colorado

Batholith-sized bodies of crystal-rich magmatic ‘mush’ are widely inferred to represent the hidden sources of many large-volume high-silica rhyolite eruptive units. Occasionally these mush bodies are ejected along with their trapped interstitial liquid, forming the distinctive crystal-rich ignimbrites known as ‘monotonous intermedi-ates’. These ignimbrites are notable for their combination of high crystal contents (35^55%), dacitic bulk compositions with interstitial high-silica rhyolitic glass, and general lack of compositional zonation.The 5000 km 3 Fish CanyonTuff is an archetypal eruption deposit of this type, and is the largest known silicic eruption on Earth. Ejecta from the Fish Canyon magmatic system are notable for the limited compositional variation that they define on the basis of whole-rock chemistry, whereas (cid:1) of feldspars þ quartz while hornblende þ titanite þ biotite were crystallizing). Assimilation and blending of phenocrysts, antecrysts and xenocrysts reflects chamber-wide, low Reynolds number convection that occurred within the last (cid:1) 10 000 years before eruption.

Batholith-sized bodies of crystal-rich magmatic 'mush' are widely inferred to represent the hidden sources of many large-volume high-silica rhyolite eruptive units. Occasionally these mush bodies are ejected along with their trapped interstitial liquid, forming the distinctive crystal-rich ignimbrites known as 'monotonous intermediates' . These ignimbrites are notable for their combination of high crystal contents (35^55%), dacitic bulk compositions with interstitial high-silica rhyolitic glass, and general lack of compositional zonation.The 5000 km 3 Fish CanyonTuff is an archetypal eruption deposit of this type, and is the largest known silicic eruption on Earth. Ejecta from the Fish Canyon magmatic system are notable for the limited compositional variation that they define on the basis of whole-rock chemistry, whereas $ 45 vol. % crystals in a matrix of high-silica rhyolite glass together span a large range of mineral-scale isotopic variability (microns to millimetres). Rb/Sr isotopic analyses of single crystals (sanidine, plagioclase, biotite, hornblende, apatite, titanite) and sampling by micromilling of selected zones within glass plus sanidine and plagioclase crystals document widespread isotopic disequilibrium at many scales. High and variable 87 Sr/ 86 Sr i values for euhedral biotite grains cannot be explained by any model involving closed-system radiogenic ingrowth, and they are difficult to rationalize unless much of this radiogenic Sr has been introduced at a late stage via assimilation of local Proterozoic crust. Hornblende is the only phase that approaches isotopic equilibrium with the surrounding melt, but the melt (glass) was isotopically heterogeneous at the millimetre scale, and was therefore apparently contaminated with radiogenic Sr shortly prior to eruption. The other mineral phases (plagioclase, sanidine, titanite, and apatite) have significantly lower 87 Sr/ 86 Sr i values than whole-rock values (as much as À0Á0005). Such isotopic disequilibrium implies that feldspars, titanite and apatite are antecrysts that crystallized from less radiogenic melt compositions at earlier stages of magma evolution, whereas highly radiogenic biotite xenocrysts and the development of isotopic heterogeneity in matrix melt glass appear to coincide with the final stage of the evolution of the Fish Canyon magma body in the upper crust. Integrated petrographic and geochemical evidence is consistent with pre-eruptive thermal rejuvenation of a near-solidus mineral assemblage from $720 to 7608C (i.e. partial dissolution of feldspars þ quartz while hornblende þ titanite þ biotite were crystallizing). Assimilation and blending of phenocrysts, antecrysts and xenocrysts reflects chamber-wide, low Reynolds number convection that occurred within the last $10 000 years before eruption.

I N T RO D UC T I O N
In the varied spectrum of processes involved in continental silicic magmatism there is an important linkage between high-silica rhyolites and granites, with the former widely inferred to be generated as liquids expressed from underlying voluminous crystal mush that eventually forms granitoid plutons (e.g. Bachmann & Bergantz, 2004;Hildreth, 2004;Hildreth & Wilson, 2007, for reviews). Rhyolitic eruptives generally represent only a selected fraction of the liquids expressed, and these have usually undergone a subsequent prolonged history of further evolution that blurs the picture of their origins. Exposed granitic plutons have undergone long histories of near-solidus crystallization and subsolidus cooling that limit the amount of information obtainable about their generation (Davidson et al., 2007). Occasionally, pyroclastic eruptions disgorge deep-seated 'crystal mushes' (i.e. magma bodies with crystal contents 430^40 vol. %) from the mid-to upper-crustal reservoirs that are inferred to lie below high-silica rhyolite magma chambers. Such pyroclastic units, which constitute the 'monotonous intermediate' group of ignimbrites [summarized by Hildreth (1981)], provide windows into processes within the mush zone that are otherwise obscured.
The largest examples of monotonous intermediate ignimbrites have been erupted in continental settings during magmatic 'flare-ups' (e.g. Francis et al., 1989;de Silva, 1991;Lindsay et al., 2001;Maughan et al., 2002). These voluminous (500^5000 km 3 in volume) crystal-rich ignimbrites (35^50 vol. % crystals) commonly show little systematic variation in either elemental and isotopic compositions at the scale of whole-rock samples from first-to last-erupted magmas. Whole-rock and pumice compositions are typically dacitic (66^69 wt % SiO 2 ), but the interstitial glass is typically high-silica rhyolite ($76^77 wt % SiO 2 ) in composition. The magma bodies that fed the monotonous intermediate ignimbrites may represent an important reservoir for crystal-poor high-silica rhyolite magmas (Bachmann & Bergantz, 2004) that have, for whatever reason, undergone wholesale evacuation.
Multiple processes can be inferred to be occurring in crystal mush zones, all of which are ultimately fuelled by heat and mass transfer associated with mantle-derived mafic magmatism (Hildreth, 1981). Multi-stage modification of mantle-derived magmas may occur during ascent through the lithosphere via processes of fractional crystallization, magma mixing, contamination, assimilation of country rock and remobilization of earlier crystallized, largely or wholly solid material (e.g. Pitcher et al., 1985;Hildreth & Moorbath, 1988;Johnson et al., 1989;DePaolo et al., 1992;Pitcher, 1993). Such processes inevitably result in some degree of textural and/or isotopic disequlibrium, which subsequently may be partly homogenized during extraction of interstitial high-silica rhyolite melt, or slow solidification of the mush to form a pluton. Petrological studies of intermediate volcanic rocks (and related mineral-specific compositional relations) have often identified crystal^melt disequilibrium, implying that some phases did not grow from their host melts (e.g. Tsuchiyama, 1985;Dungan et al., 2001;Dungan & Davidson, 2004). Isotopic compositions determined by microsampling among and within diverse crystal components of magmatic rocks have been used to evaluate magmatic processes, as they permit 'fingerprinting' of multiple implicated components (e.g. Geist et al., 1988;Simonetti & Bell, 1993;Davidson & Tepley, 1997;Davidson et al., 2005a). Zonations in isotopic composition documented by coreto-rim traverses across single crystals, especially when these traverses are tied to major element zoning profiles and textural features such as resorption surfaces, provide a high-resolution temporal record of the isotopic evolution of a magma during the growth of the relevant crystal phases (e.g. Hora, 2003).
A high degree of mineral and melt isotopic disequilibrium is preserved down to the scale of individual crystals in the Fish Canyon magma ($68% SiO 2 ). This rapidly erupted and quenched, batholithic-scale crystal mush is the archetypical monotonous-intermediate ignimbrite and the most voluminous known example of this phenomenon. The Rb^Sr system is applied here to whole-rock samples, single crystals and microgram-sized mineral microsampled aliquots to illuminate petrological processes occurring in a batholith-scale crystal mush prior to eruption. A combination of sampling from thick sections by micromilling coupled with new refinements in analytical techniques (Charlier et al., 2006) allow Sr isotopic differences to be measured on a sub-millimetre scale to an external precision of 50 ppm. These data have been combined with diffusion modelling of trace-element profiles to document not only the isotopic diversity within a thin-section sized domain but also to estimate the timescale for assimilation events in the crystal mush that was parental to the Fish Canyon Tuff and co-magmatic units.
is the second of eight major Oligocene ignimbrites with sources in the central San Juan caldera cluster in Colorado (Lipman, 2000). The FCT is characterized by a uniform mineral assemblage, few systematic geochemical variations from early to late-erupted material, and a lack of evidence for compositional and/or thermal gradients in the parental magma body (Steven & Lipman, 1976;Whitney & Stormer, 1985;Lipman, 2000;. The same magma that fed the FCT was also tapped by an earlier low-energy pyroclastic eruption (Pagosa Peak Dacite) and by a later lava flow (the Nutras Creek Dacite: Lipman et al., 1997;Bachmann et al., 2000). These three compositionally identical products of what we refer to as the Fish Canyon magmatic system contrast with the diverse compositions and mineral assemblages of other large ignimbrites that were erupted from the central San Juan caldera cluster (Riciputi et al., 1995;Lipman, 2000). Fish Canyon juvenile material is dacitic (68 AE1wt % SiO 2 ) and crystal-rich ($45 vol. % crystals), and it contains a near-solidus mineral assemblage comprising plagioclase, K-feldspar, quartz, hornblende, biotite, titanite, Fe^Ti oxides, apatite, zircon, and (rare) pyrrhotite (in order of decreasing abundance). Interstitial glasses in Fish Canyon magma samples are microlite-free, high-silica rhyolite (76Á5^78 wt % SiO 2 , calculated volatile-free; Bachmann et al., 2005).
Studies of the mineralogy and petrology of the FCT (e.g. Whitney & Stormer, 1985; have shown that, despite large-scale homogeneity in whole-rock magma composition, there is abundant textural and compositional complexity at the crystal scale and that these complexities can be used to constrain the operation of magmatic processes. Quartz, sanidine and 30^50 vol. % of the plagioclase occur as resorbed, amoeboid grains. In contrast, hornblende (0Á5^3 mm long) and biotite (0Á5^5 mm across) are euhedral crystals with no physical evidence for reaction with host melt. Interstitial glass, representing the melt composition at the time of eruption, is preserved in multiple samples of the Pagosa Peak Dacite and FCT as brown pools within clasts, as well as individual shards up to a few millimetres long. These high-SiO 2 rhyolite glasses have unusually high contents of Ba (480^560 ppm: Bachmann et al., , 2005 and Sr (93^131ppm: Bachmann et al., 2005, and this study) for such otherwise-evolved compositions, consistent with the textural evidence that the feldspars were undergoing  Bachmann et al., 2007). The localities of the two samples used for micromilling (BFC187 holocrystalline xenolith and BFC191 intracaldera Fish Canyon Tuff) are marked. resorption prior to eruption. An isobaric, $458C temperature increase from $7158C to 7608C during the late evolution of the FCT magma is inferred from Al-zoning profiles in hornblendes (Bachmann & Dungan, 2002). 'Gas sparging' (upward percolation of a hot H 2 O-rich fluid phase, released from an underplated, less-evolved magma, through the largely solidified FCT crystal mush) has been proposed as a potential mechanism to induce the observed thermal rejuvenation of the magma (Bachmann & Bergantz, 2003.

A NA LY T I C A L M E T H O D S A N D S A M P L E S
We only summarize here the techniques used in this study, and the reader is referred to Charlier et al. (2006) for fuller details. Microsampling for Rb/Sr isotopic determinations was performed on 100^120 mm polished thick sections, using a New Wave Õ Micromill to generate either a powder slurry from a target area within a single crystal, or to excise a single whole grain from within thick sections. The amount of sample produced by milling (10^100 mg) is primarily dependent on the depth of penetration of the mill bit (and thus the section thickness), and the number of points sampled (see Charlier et al., 2006, fig. 2). Because Sr concentrations in plagioclase, sanidine and glass were all sufficiently high (4900, 4400 and 490 ppm, respectively) these materials were sampled directly using the micromill to produce a slurry. For phases with lower Sr concentrations (e.g. hornblende and biotite; 432 and 46 ppm), single crystals were excised from thick sections by milling around the crystal of interest, then using a dentistry pick to carefully prise the crystal from the section. For analyses of titanite and apatite, single crystals were picked by hand from roughly crushed and sieved material. Single crystal analyses of plagioclase and sanidine were also carried out to compare the 87 Sr/ 86 Sr variations found at the intra-grain level with those found at the inter-grain level.
Slurry samples from micromilling were collected with a micropipette, dried and weighed to AE0Á1 mg. Excised and picked individual crystals were cleaned first in methanol followed by water, dried and then weighed. After addition of the requisite amount of a high-purity 84 Sr^8 7 Rb spike, the samples were dissolved in 3Á5 ml Savillex capsules using concentrated HF þ HNO 3 and finally equilibrated with 2Á5M HCl. Quartz glass microcolumns filled with $0Á75 ml of Dowex AG 50W-X8 cation exchange resin were used for initial separation of Rb and Sr. The Sr eluate was collected and further purified using a 50 ml column filled with Sr-Spec Õ resin.
A Ta emitter solution, which enhances Sr ionization efficiency, was used to load nanogram quantities of Sr onto single outgassed Re filaments for thermal ionization mass spectrometry (TIMS) analysis. All Sr aliquots were analysed using a Finnigan Triton TIMS instrument in static collection mode with simultaneous monitoring of 85 Rb. The isotopic composition of Rb remaining in the Sr eluate was monitored prior to Sr ionization (to check for a possible spike contribution). Sr mass fractionation corrections used 88 Sr/ 86 Sr ¼ 8Á37521 and an exponential law relationship. Rb was run on a Neptune multicollector inductively coupled plasma mass spectrometer using admixed Zr to correct for instrumental bias. No blank corrections were applied because total procedure blanks were $3^10 pg for Sr and $10^25 pg for Rb, which are negligible compared with the sample loads of 1^10 ng for both elements. Long-term external reproducibility of Sr measurements of the NBS987 standard during the course of this study and of a similar load size to micromilled samples ($3 ng Sr) was 87 Sr/ 86 Sr ¼ 0Á710258 AE 36 2SD (50 ppm) and 84 Sr/ 86 Sr ¼ 0Á056489 AE 24 2SD (433 ppm), where n ¼ 288. The uncertainty of the 87 Rb/ 86 Sr ratio is $0Á25% using the measurement techniques detailed by Charlier et al. (2006).
Samples of Fish Canyon Tuff were chosen from the sample suite collected for the doctoral thesis work of Bachmann (2001), and the sampling locations are shown in Fig. 1. The selected samples showed little or no evidence of alteration or hydration. The microsampling and single crystal work was carried out on thick sections cut from two samples.
(1) As all samples of Fish Canyon magma are very similar, we chose a large welded pumice fragment from the intra-caldera facies of the Fish Canyon Tuff (Sample BFC191) because it contains all the relevant mineral phases in addition to particularly fresh interstitial glass in areas large enough to sample, and because this large juvenile clast (fiamme) warrants against non-magmatic contamination. Feldspars from two other sections of this sample were analysed to ensure that the full range of feldspar compositions had been covered.
(2) Sample BFC187 (co-magmatic holocrystalline xenolith from the same location as BFC191, with a TIMS U^Pb age from zircon indistinguishable from those of Fish Canyon Tuff zircons: Bachmann et al., 2007) was chosen for its coarse grain size and the probability that it crystallized at the margins of the magma chamber.
In addition to these samples, we analysed 24 whole-rock powders spanning the lithological diversity, eruptive order, and geographical range of the Fish Canyon Tuff and its precursor, the Pagosa Peak Dacite (Bachmann, 2001). Included in this whole-rock suite are two samples of coarse-grained co-magmatic xenoliths (CSGR14 and CSGR16) and two lithic fragments of amphibolitic basement (MLX Amph2 and Amph3) from the northern intra-caldera FCT.
Prior to microsampling, electron microprobe traverses across selected feldspar crystals were carried out using a Cameca SX50 at the University of Lausanne. Standard analytical conditions were used (voltage 15 kV, filament current 15 nA, beam diameter 1^2 mm), and care was taken to avoid Na loss by using a shorter counting time on Na. Only analyses with totals in the range 99^101% were accepted.

R E S U LT S
Sr isotopic data plus Rb and Sr concentrations are presented for whole-rock powders, single crystals and micromilled aliquots in Table 1 Renne et al., 1998), the Rb/Sr ratio of the sample, and the overall 2 uncertainty associated with these values. The last variable quadratically combines the 2 external standard reproducibility of the 87 Sr/ 86 Sr measurement, the within-run measurement uncertainties, the uncertainty on the age used for correction (28Á02 AE 0Á28 Ma: Renne et al., 1998), and the uncertainty on the 87 Rb/ 86 Sr ratio.

Whole-rock Rb^Sr data
The majority of the FCT and Pagosa Peak Dacite wholerock samples define ranges in 87 Sr/ 86 Sr i from 0Á70626 to 0Á70657 and 87 Rb/ 86 Sr from 0Á49 to 0Á93 (Fig. 2b). We stress that these ranges are restricted in comparison with the entire dataset. The two less radiogenic, low 87 Rb/ 86 Sr data points that plot to the left of the main group are: (1) a whole-rock sample of distal tuff (BFCElep), which is inferred to have lost some ash through winnowing during transport (Whitney & Stormer, 1985;; (2) an andesitic enclave from the FCT intracaldera facies (BFC138). Its low 87 Sr/ 86 Sr i value is consistent with its less evolved composition. The two most radiogenic whole-rock values from the Fish Canyon system are from co-magmatic xenoliths [CSGR14 (Fig. 2b) and BFC187 ( Fig. 2c)], both of which have been dated by U^Pb singlegrain zircon TIMS analyses (Bachmann et al., 2007) at $28Á0^28Á6 Ma (i.e. indistinguishable from zircon ages of the main FCT deposit). BFC187 is atypical in that it has by far the most radiogenic 87 Sr/ 86 Sr i value (0Á70759), but not an especially elevated value of 87 Rb/ 86 Sr (0Á70). MLX Amph2 and MLX Amph3 are accidental lithic fragments (not magmatic xenoliths) of the surrounding amphibolitic Precambrian basement within which the San Juan Tertiary magmas evolved (Riciputi et al., 1995;Smith et al., 1999;Wobus et al., 2001). They were sampled in the northern intracaldera FCT facies (same general area as BFC187 and BFC191) and have extremely radiogenic 87 Sr/ 86 Sr i values (0Á94235 and 0Á79182), not plotted for the sake of scale.

Single mineral Rb^Sr data
Values of 87 Sr/ 86 Sr i and 87 Rb/ 86 Sr determined on single grains of FCT apatite, titanite, hornblende, biotite, sanidine and plagioclase span very large ranges (0Á70578^0Á72942 and 0Á002^172Á5, respectively). The lowest 87 Rb/ 86 Sr values (50Á02) are from apatite and plagioclase (Fig. 2a), whereas the highest are from biotites (8Á01^172Á5: Fig. 2d and e). Large euhedral single apatites ($200^300 mm in length) with Sr concentrations of 436^685 ppm, and 0Á3^1Á4 ppm Rb, define a restricted range in 87 Sr/ 86 Sr i of 0Á70623^0Á70634 and are generally less radiogenic than plagioclase, which defines a larger range from 0Á70632 to 0Á70670. Single euhedral titanite crystals ($200 mm) also have variable 87 Sr/ 86 Sr i (0Á70578^0Á70644), and four of these yield the least radiogenic values recorded in this study. As Rb concentrations in titanite are similar to those in apatite (0Á5^1Á3 ppm), the higher 87 Rb/ 86 Sr values for titanite (0Á04^0Á12) reflect much lower Sr concentrations (30Á5^52Á9 ppm).
Single euhedral hornblende crystals extracted from glassy fiamme (Fig. 2b) define a restricted range in 87 Rb/ 86 Sr and 87 Sr/ 86 Sr (0Á15^0Á22 and 0Á70656^0Á70667), but have values that are distinct from, and more radiogenic than, both titanite and apatite. They are, however, within the range of 87 Sr/ 86 Sr i values, albeit towards the more radiogenic end of the plagioclase and whole-rock analyses.
Fish Canyon sanidine crystals are invariably resorbed and they occur as large amoeboid grains that resemble comparably resorbed quartz grains. The sanidines often contain small inclusions of plagioclase, although hornblende, biotite, oxides, apatite and titanite are occasionally present in minor quantities. These textural similarities render picking of single 'clean' sanidine grains very difficult, and it is probable that some of the grains contained minor mineral and/or glass inclusions. Such complexities may be the cause of variable Sr and Rb concentrations in the sanidine single-crystal population (576^933 ppm and 74^149 ppm, respectively), which may have led in turn to large ranges for 87 Rb/ 86 Sr (0Á30^0Á73). Values of 87 Sr/ 86 Sr i in sanidine (0Á70632^0Á70672), not surprisingly, encompass the ranges defined by plagioclase, hornblende, titanite and apatite data, with the exception of the four least radiogenic titanite analyses.
Glass analyses were performed by micromilling areas of crystal-free fresh brown glass directly from a single thick section (BFC191; Fig. 3). These aliquots are sufficiently large to yield 10^20 ng Sr for isotopic analysis (Fig. 2c). Concentrations of Rb and Sr are both high (161^180 ppm and 93^131ppm respectively), but with a greater range seen in the Sr concentration data. 87 Rb/ 86 Sr varies from 3Á1to 5Á4       (Fig. 2d and e). Those from the intra-caldera Fish Canyon Tuff (BFC191) have 87 Rb/ 86 Sr from 8Á6 to 63Á9 and 87 Sr/ 86 Sr i from 0Á70669 to 0Á70834. Rb concentrations in biotites from both BFC191 and BFC187 are similar (280^361 vs 244^411), but Sr concentrations generally extend to lower values in BFC187 biotites (15Á7^87Á2 vs 5Á8^26Á6). The least radiogenic of these biotite analyses is within the range of 87 Sr/ 86 Sr i of the low 87 Rb/ 86 Sr phases, whereas eight of these crystals plot within the range of 87 Sr/ 86 Sr i defined by glass. The remaining eight biotite analyses are much more radiogenic than the 87 Sr/ 86 Sr i values in glass. Even more extreme 87 Sr/ 86 Sr i isotopic disequilibrium and higher 87 Rb/ 86 Sr values were found in the biotite crystals from the co-magmatic xenolith BFC187, wherein the least radiogenic 87 Sr/ 86 Sr i value (0Á71052) is considerably higher than the most radiogenic BFC191 biotite.

Microsampling Rb^Sr data
To provide additional sub-millimetre constraints on the isotopic variations defined by single crystals, microsampling was used to define intra-crystal heterogeneity in several large, zoned plagioclase and sanidine grains. This technique has the added benefit of permitting us to avoid inclusions and cracks, thereby producing purer samples of the target minerals than those obtained by picking single grains.

Plagioclase
Plagioclase occurs as both euhedral and resorbed free grains in the glass matrix, and as inclusions, chiefly in sanidine. Crystal Plg-1 was chosen for micromilling ( Fig. 3a) because plagioclase, hornblende and glass could all be sampled from a small area, and the adjacent glass is optically homogeneous and unaltered. Four separate areas from core to rim in Plg-1 (aliquots 1^4), as well as two areas in the surrounding glass (Gls-1 and Gls-2) and a neighbouring single hornblende grain (Hbl-1) were sampled for Rb^Sr analysis.
The margin of Plg-1 is slightly resorbed. A highresolution electron probe traverse shows that the core composition is fairly uniform (An 28^30 ), apart from more calcic spikes up to An 55^60 near the rim (Fig. 3b). The latter may be the result of late but episodic crystallization from more mafic and/or water-rich melts than those that were present during core crystallization. Aliquots 1 and 2, from the inner portion of Plg-1 (Fig. 3c), have identical 87 Sr/ 86 Sr i (0Á70637 and 0Á70636) and similar 87 Rb/ 86 Sr (0Á029 and 0Á019). These 87 Sr/ 86 Sr i values are within the low end of the range of analyses of single plagioclase crystals (Fig. 2a). The two aliquots taken from near the grain margin (3 and 4) have different but higher 87 Rb/ 86 Sr (0Á044 and 0Á078) and similar 87 Sr/ 86 Sr i (0Á70628 and 0Á70625), which are lower than the core ratios (1 and 2) and lie outside the range of 87 Sr/ 86 Sr i values encompassed by the single-grain data. Sr concentrations in Plg-1 are within the range defined by single grains (920^1054 ppm for Plg-1 vs 932^1317 ppm for single grains), but Rb concentrations are higher (6Á2^29Á6 ppm for Plg-1 vs 1Á2^4Á4 ppm for single grains), suggesting that small inclusions of a Rb-rich phase (most probably glass) were sampled.
Spatially associated hornblende and glass samples from the area of the section shown in Fig. 3a have more radiogenic 87 Sr/ 86 Sr i than any Plg-1 samples. The 87 Sr/ 86 Sr i value for euhedral Hbl-1 is 0Á70664 (about 0Á003 higher than the highest Plg-1 value) and the two separate areas of glass, Gls-1 and Gls-2, are even more radiogenic (0Á70708 and 0Á70682, respectively). This example demonstrates that in addition to significant isotopic disequilibrium between crystalline phases and glass, isotopic heterogeneity occurs in glass on the scale of a few millimetres.
Two euhedral plagioclase grains (Plg-2 and Plg-3) from this sample were also investigated (selected from a thick section on the basis that their cut surface passes through the core of the crystals: Fig. 4a and c). Both are euhedral grains that display the typical rimward increase in An The plagioclase crystal Plg-1 was sampled in four areas from core to rim, and the glass sampled in two areas (Gls-1 and Gls-2). Sample Gls-1 was sampled by rastering a target area, whereas Gls-2 was sampled using multiple shallow spots. Hbl-1 was sampled after the image was taken by milling a trench around the crystal, followed by removal from the section using a dentistry pick.  content ($An 27^28 to $An 32^35 ; Fig. 4b and d) that was recognized by , but Plg-3 also has a high-An core region ($An 40 ; Fig. 4d). Three microsamples from core to rim were taken from each crystal (Fig. 4e). All analyses from these crystals plot within the field of single-grain plagioclase data, but significant isotopic heterogeneity exists within both crystals. Both crystal cores (analyses 1 and 4) have the same value of 87 Sr/ 86 Sr i (0Á70635 and 0Á70636), which in turn are indistinguishable from the compositions of the inner portions of Plg-1 (0Á70637 and 0Á70636). However, the intermediate areas in Plg-2 and Plg-3 (analyses 2 and 5) are significantly more radiogenic (0Á70656 and 0Á70646). Rim compositions (analyses 3 and 6) duplicate 87 Sr/ 86 Sr i values in the cores of Plg-2 and Plg-3.

Sanidine
The textural similarities of extensively resorbed sanidine and quartz often make it difficult to distinguish between them in thick sections. Sanidine occurs as large (2^5 mm), amoeboid grains and is often shattered as a result of decrepitation of melt inclusions during rapid syn-eruptive decompression. Mineral inclusions are common, with the most abundant being plagioclase, but biotite, oxides, titanite, and apatite are also present. Rounded pools containing glass that were connected to the host melt are ubiquitous. Broad compositional oscillations (0Á1^1mm in width) in sanidine are defined primarily by Ba content. Barium zoning is manifested as undulating bands, which locally truncate earlier zones as a consequence of multiple periods of growth and dissolution .
Two large, unshattered sanidine grains were chosen for micromilling on the basis of their well-defined zoning patterns. San-1 (Fig. 5a) is a large irregularly shaped single grain $4 mm across that contains a plagioclase inclusion (bright area in Fig. 5a). The outermost band was precipitated on a dissolution surface that truncates the inner bands (indicated by white arrows in Fig. 5a). The electron probe traverse (Fig. 5c) shows an increase in the Celsian (Cn) component from near the plagioclase inclusion up to the truncation (0Á5^1Á2 mol%), followed by a sharp jump to 2Á9 mol% across the boundary. Ab and Or mol% contents are fairly flat within the main part of the crystal, and show a correlated upward and downward trend, respectively, outside the dissolution surface.
Aliquots 1 (plagioclase inclusion) and 2^4 (from San-1) were sampled by milling grids of multiple shallow holes, whereas 5 and 6 are multiple shallow points along a curved course (Fig. 6b). The 87 Sr/ 86 Sr i value for the plagioclase inclusion (analysis 1; 0Á70638) is significantly less radiogenic than analyses of the host sanidine (0Á70654^0Á70670; Fig. 5d), but it is still within the range of values for single plagioclase grains (Fig. 2a). The remaining analyses (2^6) from this sanidine record significant internal 87 Sr/ 86 Sr i variations. Aliquots 2, 3 and 4 (inboard of the truncation) have effectively identical 87 Sr/ 86 Sr i values within analytical uncertainty (0Á706540 Á70658), whereas analyses 5 and 6, which lie on and outboard of the truncation, have almost identical, but more radiogenic 87 Sr/ 86 Sr i values (0Á70667 and 0Á70670). All the 87 Sr/ 86 Sr i values for San-1 lie within the range defined by single sanidine grains (Fig. 5d).
In the second sanidine crystal (San-2, Fig. 6), there is a slight overall increase in mol% Cn towards the margin of the crystal and a fairly flat profile for Ab and Or. Towards the margin of the crystal, Or decreases and this is mirrored by an increase in Ab (compare Fig. 5c). 87 Sr/ 86 Sr i values (Fig. 6d) are all within error of one another and they span the range 0Á70644^0Á70649. These values are all less radiogenic than those found for analyses of San-1, except for analysis 6, which just overlaps with the less radiogenic end of the San-1 data.

D I S C U S S I O N
The results of Sr isotopic determinations in single crystals of multiple phases from the Fish Canyon magma display considerable 87 Sr/ 86 Sr i diversity, ranging from 0Á70578 AE 0Á00004 (titanite Sp2) to 0Á72942 AE 0Á0001 (biotite 187Bi5). All mineral phases in the two micromilled samples, except hornblende, were not in isotopic equilibrium with their host melts, and even the glass is isotopically heterogeneous at the millimetre scale. We are aware that generalizing about chamber-wide processes using only two samples for such a huge caldera system may seem overreaching, but our point is not to show that heterogeneities will be identical everywhere, but that isotopic disequilibrium does occur in representative samples of the Fish Canyon magma body, and should therefore, be present everywhere (albeit to different degrees as suggested by the variable whole-rock 87 Sr/ 86 Sr i data).

The role of contamination
Ingrowth of 87 Sr/ 86 Sr during the residence time of the Fish Canyon magma ($0Á5 Ma: Bachmann et al., 2007) cannot explain the observed variations in 87 Sr/ 86 Sr i , even for high Rb/Sr phases (Fig. 7). Assimilation of the surrounding wall-rocks must, therefore, be invoked [see similar conclusions by Wolff et al. (1999) and Wolff & Ramos (2003) for the Bandelier Tuff].
Assimilation of Precambrian upper crust by the immense but not extremely hot Fish Canyon magma (58008C), is confirmed by the presence of zircon cores dated by secondary ionization mass spectrometry (SIMS) at $1Á7^1Á8 Ga (Lanphere & Baadsgaard, 2001). The timing and processes are elucidated by the microsampling isotopic database. We interpret the less radiogenic 87 Sr/ 86 Sr i of feldspars, titanite and apatite as evidence for derivation from partly to wholly solidified portions of the early Fish Canyon magmatic system and, following Hildreth (presentation at Penrose Conference on 'Longevity and Dynamics of Rhyolitic Magma Systems', 2001), we use the term 'antecrysts' for such crystals. Antecrysts are interpreted to pre-date the final assembly of the magma in the (holding) chamber from which it was erupted, and were mostly crystallized from earlier, less contaminated co-genetic magmas. These crystals have either grown very little (e.g. apatite and titanite) or have been resorbed (sanidine and some plagioclase) as a consequence of thermal rejuvination, which also may have contributed to the assimilation event that increased 87 Sr/ 86 Sr i in the melt (mechanical fracturing of the magma camber walls as a result of inflation, in addition to heating). Biotites are also out of isotopic equilibrium with the surrounding glass by virtue of possessing more radiogenic 87 Sr/ 86 Sr i than the glass. This suggests that biotite crystals are likely to have been in part inherited from the Precambrian crustal rocks. The euhedral biotite crystal morphologies are

Timescales of evolution of the FCTç constraints from diffusive re-equilibration
The presence of isotopic heterogeneity within single grains indicates that either crystals grew from a magma as its isotopic composition changed and/or crystals that had already grown (xenocrysts or antecrysts) have undergone partial isotopic exchange with a magma into which they had been entrained and immersed. In either case, an assessment of diffusive isotopic re-equilibration can be used to constrain the timescales over which isotopic heterogeneity formed and persisted in the pre-eruptive magmatic system. Our analysis ( Fig. 8; see also the supplementary electronic appendix, which is available for downloading at http://www.petrology.oxfordjournals.org) assumes a temperature of 7608C; the upper limit of temperatures determined by Bachmann & Dungan (2002) using hornblende thermometry, and that the rims of the mineral phases were in thermal and chemical equilibrium with their host liquids. Feldspars are excluded from consideration as they are characterized by disequilibrium textures indicating they were undergoing dissolution prior to eruption. Our calculations also conform to the ranges of typical grain sizes for each phase.
The presence of both titanite and apatite in the Fish Canyon magma places some limiting constraints on mineral residence times because of their vastly different Sr diffusivities (Fig. 8). The analysed Fish Canyon apatite grains define a narrow range of isotopic ratios that is consistent with relatively fast Sr diffusivity and the quasihomogeneous size of the analysed grains. Up to 60% re-equilibration of the apatite cores with the rim composition would occur in $10 000 years and this is likely to be the cause of the limited isotopic diversity. The population of euhedral titanite single crystals is isotopically more diverse and the extremely slow Sr diffusivity in this phase precludes significant re-equilibration on a timescale that is applicable to the Fish Canyon magma chamber. Our data do not discriminate between the presence of radiogenic overgrowth rims on titanite (i.e. variably heterogeneous individual grains) and an initially isotopically variable titanite population (i.e. sequential crystallization during magma evolution and assembly).
The extremely radiogenic biotite grains in xenolith BFC187 have been modelled as the products of partial re-equilibration between high 87 Sr/ 86 Sr i , low Sr metamorphic cores (xenocrysts), with low 87 Sr/ 86 Sr i , higher Sr rims that grew from the FCT magma. These crystals give re-equilibration timescales of the order of hundreds of years at 7608C ( Table 2). The 87 Sr/ 86 Sr i values for assimilated biotite crystals may have been even more radiogenic than the reported country-rock values ( 87 Sr/ 86 Sr i 40Á9), but calculated timescales are insensitive to this value because as the initial ratio increases, the Sr content decreases, thereby buffering the whole-grain re-equilibration time.
The implication of the isotopic diversity among Fish Canyon magmatic biotite crystals is that re-equilibration following introduction of highly radiogenic Sr into the host melt is unlikely to have occurred for more than a few thousand years. This is consistent with the preservation of isotopic heterogeneity in glass. The isotopic diversity in the biotite population is, given the 87 Sr/ 86 Sr i elevation of many crystals with respect to even the most radiogenic glass 87 Sr/ 86 Sr i , consistent with input from country rock material. The exact time of origin of the isotopic variations cannot be addressed more precisely with this method because of low resolution at longer timescales and uncertainty with respect to the initial isotopic compositions of potentially highly variable assimilated lithologies and the buffering liquids.   Figs 2^6 can realistically be accounted for by radiogenic ingrowth in the developing magma system. Á 87 Sr/ 86 Sr is the maximum difference in Sr isotope ratio that can be generated for a given Rb/Sr ratio (contours) over a given time. The maximum lifespan of the FCT system is considered to be $0Á5 Ma (Bachman et al., 2007). The ranges in isotope ratios for various phases are shown in the shaded bar to the right. From this it is clear that the range in isotopic ratios seen among most of the phases with Rb/Sr 51 (plag, san, ap, sph, hb and whole-rocks) requires much longer than 1Myr to generate, and even the highest Rb/Sr ratios recorded ($50 in the biotites) require nearly 0Á5 Myr to generate the ranges in 87 Sr/ 86 Sr observed. In short, even with very long-lived magma chambers and high Rb/Sr liquids, the observed isotopic heterogeneity cannot be realistically generated in situ, and open-system processes must be invoked.

Assimilation dynamics and magma chamber model
On the basis of Sr isotopic disequilibrium and diffusion modelling, assimilation must have occurred 'shortly' prior to the eruption in order for isotopic heterogeneities to persist within and among the different phases (particularly biotite), and for feldspars (plus titanite and apatite) not to grow significantly after the assimilation event. This leads to the conclusion that the assimilation event occurred in the upper crust after assembly of the Fish Canyon magma body during thermal rejuvenation. Shallow storage of the Fish Canyon magma at 2Á24 AE 0Á05 kbar (Bachmann & Dungan, 2002) and its relatively low temperature of 760 AE 208C would not appear to provide favourable conditions for large-scale assimilation had there not been an elevation of the thermal gradient below this magmatic focus as a consequence of several million years of high-flux magmatism in and around the central San Juan caldera cluster prior to 28 Ma.
As some of the Proterozoic wall-rocks are probably more radiogenic than the amphibolite xenolith reported in Table 1 ( 87 Sr/ 86 Sr i $0Á942), only a tiny amount of assimilation ($1%) would be necessary to raise the 87 Sr/ 86 Sr i from mantle-like ($0Á704) to Fish Canyon whole-rock values (0Á7065 AE 0Á0002). If assimilation occurred by stoping (Yoshinobu et al., 2003;Hawkins & Wiebe, 2004), thermally efficient reactive bulk assimilation (Beard et al., 2004(Beard et al., , 2005 could have played a critical role in digesting incorporated xenoliths. As mentioned above, the co-magmatic xenolith (BFC187) is the most contaminated sample in the dataset; it has the highest whole-rock 87 Sr/ 86 Sr i ratio and extremely radiogenic biotite crystals. The compositional, mineralogical and U^Pb zircon age similarities between this sample and the main Fish Canyon magma imply that this holocrystalline granodiorite fragment was part of the co-magmatic, fully crystalline rind of the Fish Canyon magma chamber. Its more contaminated character is readily explained by close spatial proximity to the Precambrian wall-rocks. As the holocrystalline nature of the rock requires that it cooled to a lower temperature than the main Fish Canyon magma body (57008C?), its minerals should have undergone less diffusive re-equilibration.
Our preferred model for the late evolution of the Fish Canyon magma body is the following.
(1) A dacitic magma chamber was assembled in the upper crust by sequential addition from below (including both silicic and mafic replenishment events). Temperature and crystallinity oscillated in space and time as new magmatic pulses arrived and were incorporated into the growing chamber, but the magma probably existed as a static (non-convecting) crystal mush (!50 vol. % crystals) for most of its lifetime, as this is the most stable thermal configuration (i.e. involving slow cooling by conduction: Koyaguchi & Kaneko, 1999;Bachmann & Bergantz, 2006). (2) Stoped country-rock blocks were detached from the roof and sides of the magma chamber during dynamic episodes, such as earthquakes and replenishment events, and became immersed in the crystal-rich magma. For country-rock blocks to sink into the magma, the crystallinity had to be below the critical rheological threshold of $50 vol. % crystals (Marsh, 1981;Vigneresse et al., 1996), which would impede significant stoping.
(3) The stoped blocks partially melted, disaggregated and reacted with the silicic magma. Partial digestion via thermally efficient reactive bulk assimilation would liberate radiogenic eutectic grain boundary melts and refractory solid material (i.e. zircons) from the xenoliths. We infer that solid anhydrous residues (pyroxene, oxides) were eliminated by reaction with the surrounding water-rich silicic melt (Beard et al., 2005) to create hydrous phases (e.g. hornblende, biotite). Hydrous minerals liberated from xenoliths were at most partly resorbed and may have regrown or overgrown as stable phases from the hybrid melt. Such partial transformations probably account for hornblende grains, and especially biotite, that have retained disequilibrium radiogenic signatures. As discussed by Beard et al. (2005), such a process will be texturally cryptic when it has evolved toward completion, and is most likely to be identified by mineral grains that preserve isotopic disequilibrium. (4) It is probable that stoping and reactive bulk assimilation were abetted by thermal rejuvenation of the Fish Canyon magma from $7208C to 7608C shortly prior to eruption (Bachmann & Dungan, 2002;Bachmann & Bergantz, 2003). It should be noted that the values are all less than 360 years (see Fig. 7).

JOURNAL OF PETROLOGY
Low Reynolds number convection permitted by partial remelting of the Fish Canyon crystal mush could then lead to partial mixing of assimilated components and to a hybrid, homogeneous magma composition at the decimetre scale, which, none the less, preserves a record of complexities at the millimetre to micron scale. We assume that the Fish Canyon magma body was thermally rejuvenated by multiple heating pulses (as suggested by both geochemical and physical observations; Bachmann & Bergantz, 2006). Therefore, it is possible that several assimilation events occurred in the course of the evolution of the Fish Canyon magma body. Our data require only that most of the assimilation took place after the growth of most feldspar, apatite and titanite crystals.

Rb/Sr, U/Pb and 40 Ar/ 39 Ar geochronology of Fish Canyon minerals
The Fish CanyonTuff is an important unit for the geochronological community because its sanidine, zircon and apatite are widely used as standards for 40 Ar/ 39 Ar and fission-track dating techniques. In light of the isotopic results presented in this paper, some caution in the use of these geochronological standards is indicated. The low closure temperature of the argon isotopic system in sanidine (54008C for millimetre-sized crystals: McDougall & Harrison, 1999), leads to the conclusion that 40 Ar/ 39 Ar dating of Fish Canyon sanidine should provide an accurate eruption age, apart from potential uncertainties in the relevant decay constants (28Á02 AE 0Á28 Ma: Renne et al., 1994Renne et al., , 1998Villeneuve et al., 1998;Kwon et al., 2002;Daze et al., 2003;Kuiper et al., 2004). Zircon has a closure temperature (Lee et al., 1997) above that which has been determined for the Fish Canyon magma (Bachmann & Dungan, 2002), and therefore we would expect this phase to show complex age patterns, as has been observed (Lanphere & Baadsgaard, 2001;Schmitz & Bowring, 2001;Schmitz et al., 2003;Bachmann et al., 2007). Rb/Sr isochron ages have also been proposed as providing useful age information for the Fish Canyon magmatic system (Lanphere & Baadsgaard, 2001), but the Sr-isotopic systematics we have determined suggest that this proposal may not have a sound basis. Figure 9 is an 87 Sr/ 86 Sr vs 87 Rb/ 86 Sr isochron plot with data for multi-crystal biotite separates of varying purity from the FCT (Lanphere & Baadsgaard, 2001), in which a spread in 87 Rb/ 86 Sr was achieved by sequentially purifying a bulk biotite separate with various separation techniques. Lanphere & Baadsgaard (2001) attributed the progressive increase in 87 Rb/ 86 Sr to the increasingly successful removal of low 87 Rb/ 86 Sr included phases (e.g. apatite). Two analyses of plagioclase and sanidine were then combined with their biotite data to better constrain the 87 Sr/ 86 Sr i of the isochron slope, and this generated an isochron with an age of 27Á38 AE 0Á13 Ma, with 87 Sr/ 86 Sr i ¼ 0Á70642 AE10 (n ¼ 34). The fit of this isochron is statistically impressive (MSWD ¼ 0Á40), because of the large quoted uncertainty of 0Á222ø on the 87 Sr/ 86 Sr ratios, but it assumes that all the minerals were simultaneous crystallization products of FCT melt (compare . We compared our data on single biotite grains from sample BFC191 (intra-caldera FCT) with those of Lanphere & Baadsgaard (2001) on bulk separates, and investigated the extent to which bulk mineral samples might mask the heterogeneities that we have shown to be present in the 87 Sr/ 86 Sr ratios of individual biotite grains in earlier sections (Fig. 9). Our new single-grain data fall within the range of 87 Rb/ 86 Sr defined by the previous study, but with significant scatter and with a shift to more radiogenic 87 Sr/ 86 Sr values at a given 87 Rb/ 86 Sr. A best-fit line to the single grain biotite data defines an age of 28Á7 AE1Á6 Ma (n ¼16), which is indistinguishable from the currently accepted eruption age. This regression line cannot be termed an isochron because the MSWD ¼ 2377 (Wendt & Carl, 1991). When the single-grain biotite data are age-corrected to 28Á02 Ma, they have a large 87 Sr/ 86 Sr external variability of 1Á38ø (2), which is far outside the $50 ppm (2) external precision attained from analyses of NBS 987 at similar load sizes (Charlier et al., 2006;compare Lanphere & Baadsgard, 2001). Our data indicate that at the single-grain scale, biotites preserve considerable radiogenic isotope diversity resulting from partial to complete isotopic re-equilibration of basement-derived grains with FCT melt shortly before eruption (see Fig. 2d). Extremely radiogenic single-grain biotite analyses from the Fig. 9. Measured 87 Sr/ 86 Sr vs 87 Rb/ 86 Sr isochron plot of bulk biotite separate analyses of varying purity from the Fish Canyon Tuff (Lanphere & Baadsgaard, 2001), and single grain biotite analyses from this study. Lanphere & Baadsgard (2001) isochron (dashed line); 27Á38 AE 0Á13 Ma, 87 Sr/ 86 Sr i ¼ 0Á70642 AE 9Á6e^5, MSWD ¼ 0Á40, n ¼ 34. Single biotite data (dotted line); 28Á7 AE1Á6 Ma, 87 Sr/ 86 Sr i ¼ 0Á7068 AE 5e^4, MSWD ¼ 2377, n ¼16. In both cases, the 2 uncertainties are smaller than the symbol size. Lines of best fit were generated using Isoplot (Ludwig, 2002) and 87 Rb ¼1Á42e À11 (Steiger & Ja« ger, 1977).
co-magmatic xenolith (BFC187) reinforce this general observation, which contravenes the fundamental tenet of isochron dating; that is, that 87 Sr/ 86 Sr at t ¼ 0 is constant among constituent phases (Davidson et al., 2005b 6). As demonstrated previously, preservation of such single grain-scale initial isotopic heterogeneities in rocks requires short durations for the implicated processes.

C O N C L U S I O N S
The eruptive products of the Fish Canyon magmatic system exhibit measurable and, in some cases, extreme Sr isotopic heterogeneity within individual crystals and between different mineral phases on the scale of a single thin section. The observed variations result from assimilation events that occurred during residence in and transport through the continental crust, but within 510 000 years of the eruption. Our Rb^Sr data and the U^Pb zircon age data of Lanphere & Baadsgaard (2001) require that the crustal contaminant was Precambrian in age [most probably granite and granodiorite lithologies exposed in several localities in Colorado (Sims et al., 2001)], and was in small quantity ($1%). In accord with their resorbed textures, feldspars are significantly less radiogenic than the rhyolitic melt preserved in the Fish Canyon magma, suggesting that they grew mostly before the assimilation of country rocks and that they were undergoing resorption (or not crystallizing) prior to eruption. In contrast, biotite is more radiogenic than the melt, implying the presence of inherited radiogenic strontium from the basement rocks. Apart from some rims on euhedral biotite crystals, euhedral hornblende crystals appear to be the only incontrovertible crystallization products of immediately pre-eruptive Fish Canyon magma, as they approach isotopic equilibrium with the melt. These complexities at the crystal scale preclude certain types of petrogenetic conclusions on the basis of whole-rock chemistry, and call for intense scrutiny of the petrological evolution of such rocks before their constituent mineral phases are used as geochronological standards.
The conjunction of whole-rock homogeneity and extreme compositional complexities recorded by the crystal cargo in the Fish Canyon magma suggest that it attained its bulk identity by large-scale, chamber-wide stirring and blending of magma batches with different histories shortly prior to eruption. As the Fish Canyon magma body represents a typical upper crustal building block (its composition is nearly identical to the average upper continental crust and its mineralogical characteristics are indistinguishable from those of many granodioritic batholiths; e.g. Bachmann et al., , 2005, we believe that its petrogenetic evolution can be generalized to many large upper crustal silicic magma bodies (including batholithic plutons; e.g. Lipman, 2007). The pre-eruptive chamberwide stirring required by our data is at odds with the view that large plutons are generated by amalgamation of many small magma aliquots over millions of years (Glazner et al., 2004). Although we do agree that large upper crustal magma bodies grow by incremental addition, their evolution may frequently involve a high-crystallinity 'big tank' stage at some point relatively late in their history.