Journal of Petrology Advance Access originally published online on December 13, 2006
Journal of Petrology 2007 48(3):459-493; doi:10.1093/petrology/egl068
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Phase Equilibria Constraints on the Chemical and Physical Evolution of the Campanian Ignimbrite
1Department of Earth Science and Institute for Crustal Studies, University of California, Santa Barbara, Santa Barbara, CA 93106, USA
2Department of Geological Sciences, Central Washington University, Ellensburg, WA 98926, USA
3956 National Center, US Geological Survey, Reston, VA 20192, USA
4Dipartimento di Scienze Della Terra, Università Di Napoli Federico II, 80134 Napoli, Italy
RECEIVED FEBRUARY 22, 2006; ACCEPTED OCTOBER 27, 2006
 |
ABSTRACT |
|---|
TOP
ABSTRACT
INTRODUCTIONThe Campanian Ignimbrite is a > 200 km3 trachyte–phonolite pyroclastic deposit that erupted at 39·3 ± 0·1 ka within the Campi Flegrei west of Naples, Italy. Here we test the hypothesis that Campanian Ignimbrite magma was derived by isobaric crystal fractionation of a parental basaltic trachyandesitic melt that reacted and came into local equilibrium with small amounts (5–10 wt%) of crustal rock (skarns and foid-syenites) during crystallization. Comparison of observed crystal and magma compositions with results of phase equilibria assimilation–fractionation simulations (MELTS) is generally very good. Oxygen fugacity was approximately buffered along QFM + 1 (where QFM is the quartz–fayalite–magnetite buffer) during isobaric fractionation at 0·15 GPa (
6 km depth). The parental melt, reconstructed from melt inclusion and host clinopyroxene compositions, is found to be basaltic trachyandesite liquid (51·1 wt% SiO2, 9·3 wt% MgO, 3 wt% H2O). A significant feature of phase equilibria simulations is the existence of a pseudo-invariant temperature, 
883 °C, at which the fraction of melt remaining in the system decreases abruptly from 0·5 to < 0·1. Crystallization at the pseudo-invariant point leads to abrupt changes in the composition, properties (density, dissolved water content), and physical state (viscosity, volume fraction fluid) of melt and magma. A dramatic decrease in melt viscosity (from 1700 Pa s to 200 Pa s), coupled with a change in the volume fraction of water in magma (from 0·1 to 0·8) and a dramatic decrease in melt and magma density acted as a destabilizing eruption trigger. Thermal models suggest a timescale of 200 kyr from the beginning of fractionation until eruption, leading to an apparent rate of evolved magma generation of about 10–3 km3/year. In situ crystallization and crystal settling in density-stratified regions, as well as in convectively mixed, less evolved subjacent magma, operate rapidly enough to match this apparent volumetric rate of evolved magma production. KEY WORDS: assimilation; Campanian Ignimbrite; fractional crystallization; magma dynamics; phase equilibria
|
INTRODUCTION |
|---|
Variations in physical and chemical properties of intermediate to silicic magmas are the result of a complex array of processes that occur in the source and in crustal-level magma reservoirs. In the case of erupted magmas, additional complications may be introduced during magma withdrawal from compositionally heterogeneous bodies. Documenting, and where possible quantifying, magma chamber processes leads to a description of how and why physical and chemical gradients form in magma bodies and also informs the debate about the origin and evolution of continental crust. The timescales over which intermediate to silicic magmas form, evolve and amalgamate are also critical, particularly because such chronological information may improve eruption assessment and mitigation efforts. Among the most exciting endeavors in the current study of evolved magmatic systems is integration of these two fundamental topics; that is, documenting the duration of discrete but definable processes that contribute to the formation, evolution, and, where relevant, eruption of these systems. A challenging goal of such an approach includes realistic representation of the physical and chemical nature of magma bodies in space and time.
The wide range of processes that affect the formation and evolution of evolved magmatic systems complicates realization of such a goal. Crystal–liquid separation and partial melting appear to be important mechanisms for generating chemical and physical diversity (e.g. Bacon & Druitt, 1988
; DeSilva & Wolff, 1995; Bohrson & Reid, 1997; Bryan et al., 2002; Costa & Singer, 2002), but careful studies over the last 25 years have revealed that a number of other mechanisms are also viable. Among these are incorporation of crustal materials, both melt and solids (Duffield et al., 1995; Bindemann & Valley, 2001; Reagan et al., 2003; Bacon & Lowenstern, 2005; Zellmer et al., 2005), magma recharge/magma mixing or mingling (e.g. Bryan et al., 2002; Nakagawa et al., 2002; Sumner & Wolff, 2003), remobilization of magmatic ‘mush’ zones (e.g. Murphy et al., 2000; Bachmann & Bergantz, 2003; Wilson et al., 2005), and inclusion of cumulates (e.g. Reubi & Nicholls, 2005; Zellmer et al., 2005). A critical observation is that in many cases, two or more of these processes may act simultaneously, thus complicating efforts to define and quantify magmatic evolution. Even when the chemical effects of a process can be reliably identified, a comprehensive physical description of the process often remains elusive. A good example of this is the important, but still incompletely understood, mechanism of crystal–liquid separation. Gas-driven filter pressing (Sisson & Bacon, 1999), Stokes’ or hindered settling (Bachmann & Bergantz, 2004), compaction (Bachmann & Bergantz, 2004), and in situ marginal porous mush crystallization (McBirney et al., 1985; Trial & Spera, 1990) are plausible physical mechanisms that have been proposed and quantitatively examined for separation of melt and crystals. The complicated nature of magmatic systems suggests that each of these mechanisms may play a role, but detailed application of governing physical principles is not yet routine in the study of magma bodies. Further, physical descriptions of processes such as crustal assimilation and magma recharge and mixing are also incomplete. Although deciphering the processes that lead to chemical evolution in magma bodies remains a challenge, an associated and equally significant undertaking is documentation of the physical properties of magma (melt + crystals + supercritical fluid) at each stage of evolution. The physical state of magma is intricately linked not only to the processes that affect it but also to its ultimate fate as intrusive or extrusive rock.
Implicit in the pursuit of understanding magmatic evolution is deciphering the chronology of events as magma forms and evolves. Particularly in the last 15 years, insightful applications of radiogenic isotope systematics (e.g. Christensen & DePaolo, 1993; Reid et al., 1997; Davies & Halliday, 1998; Charlier et al., 2003; Reagan et al., 2003), diffusion (e.g. Zellmer et al., 1999; Costa et al., 2003; Morgan et al., 2004), and magma transport and kinetic phenomena (e.g. Mangan, 1990; Spera et al., 1995; Higgins, 1996; Bergantz, 1999) have enhanced understanding of magmatic timescales, which have been shown to vary by several orders of magnitude (< 101 to 106 years). Improved precision and spatial resolution in studies involving radiogenic isotopes (e.g. Reid et al., 1997; Lowenstern et al., 2000), and the benefit of larger, faster computers and better algorithms that allow more detailed dynamical models of magma transport have provided constraints that were lacking even a decade ago. Along with the significant analytical and computational advances, however, come challenges of interpretation. The question of what magmatic timescales mean in the context of the mass, chemical, and thermal evolution of magma bodies is central to these studies, but the answers are not straightforward. In the case of isotope data, application of a particular system to address timing may be complicated by open-system processes that may be very difficult to recognize. In addition, studies based on diffusion and magma transport are limited by imperfect knowledge of material properties and associated kinetic factors that affect the rates of, for example, elemental diffusion or heat flow. A particularly promising approach for documenting timescales of processes involves integrated studies that attempt to document the timescales of magma chamber processes by independent means.
In this paper, we examine processes that gave rise to the Campanian Ignimbrite, a 39·28 ka (De Vivo et al., 2001) trachytic to phonolitic ignimbrite erupted near Naples, Italy. By applying constraints from phase equilibria embodied within the MELTS thermodynamic model (Ghiorso & Sack, 1995), we document the major element evolution and associated changes in physical properties of this magmatic system, where isobaric multiphase liquid–crystal separation is a dominant process and wall-rock assimilation plays a secondary role. A key aspect of our results includes identification of a pseudo-invariant temperature along the liquid line of descent. That is, at 883°C, the system undergoes marked changes in crystallinity, melt composition including volatile content, viscosity, and density. The behavior of the system at this point has fundamental implications regarding the origin of the Campanian Ignimbrite and may provide a triggering mechanism for this ignimbrite and possibly other volatile-rich, low-pressure pyroclastic style eruptions. Because MELTS tracks the enthalpy of the system along the liquid line of descent, application of a simple heat transport model provides time constraints on the duration of magmatic activity, including estimates of the durations of crystallization for individual mineral phases. Such temporal context provides one approach for evaluating the meaning of timescale information derived from other types of studies.
|
GEOLOGICAL SETTING |
|---|
The Campi Flegrei Volcanic Field covers an area of
2000 km2 west of Naples, Italy, and forms part of the Campanian Plain, a region located within a graben of Mesozoic carbonates (Fig. 1a). Formation of the graben as a consequence of extension along the western margin of the Apennine Mountains led to the opening of the Tyrrhenian Sea in the Pliocene–Quaternary (Rosi & Sbrana, 1987; Acocella et al., 1999). Eruption of the Campanian Ignimbrite (CI) at 39·28 ± 0·11 ka (De Vivo et al., 2001) is regarded as the dominant event in the history of the Campi Flegrei Volcanic Field and, via its influence on northern hemisphere and perhaps global climate, may have played a role in human evolution, particularly the replacement of the Neanderthals by modern Homo sapiens sapiens (Fedele et al., 2003). In addition to the CI, explosive eruptions have been documented in the Campanian Plain at >300 ka, 205 ka, 184 ka, 157 ka, and 18 ka (De Vivo et al., 2001; Rolandi et al., 2003).
|
The CI is a grey, phenocryst-poor, unwelded to partially welded trachytic–phonolitic ignimbrite with a minimum bulk volume of
310 km3 (200 km3 Dense Rock Equivalent; DRE) and initial areal distribution of 30 000 km2 (Rolandi et al., 2003). There are four distinct pyroclastic flow units (Rosi et al., 1996; De Vivo et al., 2001) overlying a volumetrically subordinate basal fallout deposit (Fisher et al., 1993; Rosi et al., 1996; De Vivo et al., 2001). In some sections of the CI, the Gray Tuff lies beneath a Lithic Breccia that grades upward into a weakly stratified Yellow Tuff and/or an incoherent pyroclastic flow deposit (Upper Incoherent Tuff). The Yellow Tuff has been modified through secondary mineralization, notably by zeolites. The Lithic Breccia is discontinuously exposed around the Campanian Plain, and in some locations, shows evidence of proximal depositional characteristics (De Vivo et al., 2001). A schematic cross-section of the CI is presented in Fig. 1b; detailed stratigraphic sections of localities from which the samples studied as part of this investigation were taken have been given by De Vivo et al. (2001).
Debate persists about the location of vents of the CI. Some workers advocate eruptions from quasilinear fissures controlled by regional fault systems (e.g. Di Girolamo, 1968, 1970; Milia et al., 2000, 2003; De Vivo et al., 2001; Rolandi et al., 2003). In particular, work by De Vivo et al. (2001) has identified proximal depositional characteristics and an elongate pattern in the distribution of the Lithic Breccia. An alternative hypothesis posits an association with a caldera ring fracture system located within the Campi Flegrei Volcanic Field (e.g. Rosi & Sbrana, 1987; Fisher et al., 1993; Rosi et al., 1999; Signorelli et al., 1999) or elsewhere in the Campanian Plain (e.g. Scandone et al., 1991; Orsi et al., 1996).
|
ANALYTICAL METHODS |
|---|
Major and minor elements were determined on fused glass disks by wavelength-dispersive X-ray fluorescence spectrometry (WD-XRF) using the method described by Taggart et al. (1987
) at the US Geological Survey laboratories, Denver, CO and at Activation Laboratories, Ancaster, Ontario, Canada. Before grinding, bulk tuff and pumice samples were examined and any alteration was removed; the sample was then washed in deionized water. Grinding and powdering were done with either mild steel or alumina disks. Major element oxides were determined in representative aliquots by WD-XRF after fusion with lithium metaborate–tetraborate. The titration technique used for FeO determination was that of Peck (1964). CO2 was determined coulometrically (Engleman et al., 1985). Total H2O was determined by Karl–Fischer titration (Jackson et al., 1987). H2O– was calculated by weight loss after heating at 110°C for a minimum time of 1 h (Shapiro, 1975). H2O+ is thus calculated from the difference between total H2O and H2O–. Loss on ignition (LOI) is the weight loss after heating over the temperature range 925–1050°C. Total sulfur reported as SO3 was determined by the combustion–IR spectroscopy method described by Kirschenbaum (1983). Fluorine was determined by the selective-ion electrode method of Kirschenbaum (1988), and chlorine was determined by either selective-ion electrode (Aruscavage & Campbell, 1983) or by instrumental neutron activation analysis (Hoffman, 1992).
Quantitative electron microprobe analyses of major and minor elements in feldspar were obtained by WD-XRF, using a JEOL JXA-8900 five-spectrometer, fully automated electron microprobe analyzer. Analyses were made at 15 keV accelerating voltage and 20 nA probe current measured with a Faraday cup; counting times on both peak and background varied from 20 to 120 s, with a 5 µm probe spot. The analyses were corrected for electron beam matrix effects, instrumental drift and deadtime using a Phi-Rho-Z algorithm (CITZAF; Armstrong, 1995) as supplied with the JEOL JXA-8900 electron microprobe. Relative accuracy of the analyses, based upon comparison between measured and published compositions of standard reference materials, is 1–2% for oxide concentrations >1 wt% and 5–10% for oxide concentrations < 1 wt%. Elements analyzed as oxides and their detection limits (wt%) at 3
are as follows: MgO (0·01), Al2O3 (0·03), SiO2 (0·03), CaO (0·02), FeO (total) (0·03), Na2O (0·03), K2O (0·01), BaO (0·04), SrO (0·04), and P2O5 (0·02).
|
PETROGRAPHY AND MAJOR ELEMENT COMPOSITIONS |
|---|
Our study is based upon 77 stratigraphically oriented pumice or bulk-rock (pumice + matrix) samples that cover a wide geographical area and include examples from each of the four ignimbrite units and the basal pumice (Fig. 1a and b). The locations and volcanological framework of these samples have been presented by De Vivo et al. (2001
) and Bohrson et al. (2006). Here, we present an overview of the petrography and describe the major element composition of the Gray Tuff, Lithic Breccia, Incoherent Tuff, and Basal Pumice. Because samples of the Yellow Tuff have undergone secondary alteration, which is obvious in hand sample, they are not discussed here. We also utilize data from several published studies on single and bulk pumice, glass, and melt inclusions (Civetta et al., 1997; Signorelli et al., 1999; Pappalardo et al., 2002a, 2002b; Webster et al., 2003).
Petrography and mineral compositions
The crystallinity of CI ash-flow tuff and Plinian fall (Basal Pumice) is low (average crystallinity 3 vol.%; maximum crystallinity 10 vol.%). The typical phenocryst assemblage includes alkali feldspar, with lesser plagioclase and sparse to trace clinopyroxene, spinel, apatite, and biotite. Based on analyses of Civetta et al. (1997), some clinopyroxene exhibits modest compositional zoning, whereas spinel and biotite are nearly homogeneous. Based on several hundred microprobe analyses of feldspar (Table 1; see also Electronic Appendix 1, available for downloading at http://www.petrology.oxfordjournals.org), the Or component of alkali feldspar ranges from 44 to 88. Alkali feldspar compositions exhibit a bimodal distribution with a sub-maximum near Or85–90 and a larger peak at Or55–65 (Fig. 2a). In some cases, crystals are modestly zoned from higher Or cores to lower Or rims. Plagioclase typically exhibits the most intracrystal variation. An content varies from 20 to 90 and, based on available data, plagioclase exhibits trimodality with local maxima at An80–90 and An50, and an absolute modal maximum at An30 (Fig. 2b). In some cases, sieved-textured calcic cores are surrounded by more potassic rims. Mean linear (equivalent size) dimensions for clinopyroxene, spinel, apatite, biotite, alkali feldspar and plagioclase are 0·60, 0·30, 0·09, 0·75, 1·56 and 0·50 mm, respectively, based on analysis of several hundred crystals.
|
|
Petrochemistry
Previous studies (e.g. Civetta et al., 1997
; Signorelli et al., 1999; Pappalardo et al., 2002a, 2002b) show that CI pumice and bulk-rock samples are dominantly trachyte and phonolite (Fig. 3). Also plotted in Fig. 3 are CI glass and pumice data from Civetta et al. (1997) and basal pumice glass and melt inclusion data from Signorelli et al. (1999). Because the crystallinity of CI ash-flow tuff and basal pumice fall is low, whole-rock, pumice and glass compositions are compositionally similar. Figure 3 also illustrates the loss of alkalis during zeolitization of the Yellow Tuff facies. Because these samples clearly show effects of non-magmatic processes, samples of the Yellow Tuff facies will not be considered further in this study (marked with an open circle in Fig. 3).
|
Also included in Fig. 3 are data for clinopyroxene-hosted melt inclusions (MI) studied by Webster et al. (2003
) (marked with + in the figure). These MI data were collected from a subset of samples presented in the present study. Webster et al. (2003) identified two groups of MI: high-MgO and low-MgO. The high-MgO MI data extend the compositional range of the CI dataset to basaltic trachyandesite and have been interpreted by Webster et al. (2003) as possibly representing parental or at least close to parental magma. These MI data will be described in more detail in the next section, where we use them to reconstruct a parental melt that is then used as a starting point in phase equilibria calculations.
Selected major oxide trends of CI data on MgO variation diagrams are presented in Fig. 4a–f, and representative major element data are presented in Table 2. All major element data are presented in Electronic Appendix 2 (available at http://www.petrology.oxfordjournals.org). For most oxides, pumice and bulk samples are characterized by relatively coherent oxide–oxide trends, and no systematic differences in these two sample types are evident. SiO2, Na2O, and MnO (not shown) vs MgO form relatively tight negatively correlated arrays, whereas in general, CaO, P2O5 (not shown), FeO, and K2O are positively correlated arrays. Al2O3 vs MgO forms a relatively tight cluster that lacks negative or positive correlation. In most cases, glass and melt inclusion data from the basal pumice (Signorelli et al., 1999) extend the major element compositional ranges compared with pumice and bulk-rock, consistent with compositional zonation defined by differences between basal pumice fall deposits and later-erupted pyroclastic flow deposits. Discussion of the computed trends in Fig. 4 is reserved for a later section of this paper. Although the thermodynamic calculations do self-consistently determine H2O solubility, the volatile content of CI pumice and whole-rock samples measured today provides little constraint on volatiles at depth. All comparisons between predicted and observed compositions are consequently made on an anhydrous basis unless stated explicitly.
|
|
|
PHASE EQUILIBRIA CONSTRAINTS: MELTS MODELING |
|---|
Phase equilibria constraints on major element changes in magma during crystallization form a fundamental starting point in the discussion of magma compositional evolution. MELTS is a rigorous thermodynamic model of crystal–liquid equilibria that uses experimentally derived data on the compositions of coexisting solid and magmatic silicate liquid phases at specified temperatures, pressures, and oxygen fugacities to calibrate models for the compositional dependences of thermodynamic potentials for mineral and silicate liquid phases. As heat is extracted and the temperature drops in a system, phase identities, compositions, and proportions are calculated. The MELTS algorithm is based on classical equilibrium thermodynamics and has been extensively reviewed elsewhere (Ghiorso & Sack, 1995
; Asimow & Ghiorso, 1998). Below we provide a brief summary of points that are relevant to the present study.
Application of MELTS to natural systems
From a practical perspective, using MELTS to model the evolution path of cooling magma requires specification of the initial state of a system and constraints under which evolution proceeds. Initial conditions define the starting temperature, temperature step, pressure, and a parental composition including an initial water concentration. In this study we have chosen the liquidus temperature as the starting temperature; MELTS computes a liquidus temperature based on the specified initial liquid composition and pressure. An end temperature is selected by comparison of MELTS results and observed data. A system ferric/ferrous ratio must be defined, either from FeO/Fe2O3 analyses or from total Fe, in which case selection of an oxygen fugacity distributes iron in the liquid and solids between FeO and Fe2 O3 according to the specified temperature and pressure. The constraints specify a reaction path in which the system is closed or open to mass transfer within standing magma (fractional or equilibrium crystallization) or from wall-rock to standing magma (assimilation). Pressure may be held constant or may vary along some P–T path. An oxygen fugacity constraint path may be defined. These and a number of additional issues must be considered when using MELTS to explore the consequences of magmatic evolution in natural systems.
Thermodynamic properties database
Compared with data from our investigation of some seven natural explosive volcanic systems, MELTS effectively predicts oxide concentrations for liquids during fractional crystallization, from the liquidus down to low melt fraction (0·03–0·04). However, predicted CaO and K2O concentrations are systematically displaced from observed data by up to 3 wt%. The discrepancy for CaO can probably be attributed to understabilization of clinopyroxene (M. Ghiorso, personal communication, 2006). The problem with K2O may be due to a lack of experimental data to calibrate the thermodynamic model that describes alkali feldspar–liquid equilibria (M. Ghiorso, personal communication, 2006). The calibration of the activity of the K-component in the liquid may give a value that is too high, resulting in overprediction of the stability of alkali feldspar that is manifest in a reduction of liquid K2O concentrations.
Lack of accounting for CO2
MELTS currently does not have a thermodynamic model for CO2 solubility. However, the solubility of CO2 in melt is minimal (Holloway, 1976) and we show through MELTS-based consideration of carbonate assimilation (discussed below) that carbonate assimilation probably was not an important factor in the petrogenesis of the CI.
Application of MELTS to modelling of the CI
In this study, all MELTS models are based on minimization of Gibbs free energy along an isobaric path of decreasing temperature and decreasing system enthalpy. We have imposed model constraints regarding the liquid oxidation state. Only the total Fe has been used as an input; FeO and Fe2O3 quantities in evolving liquid reflect ferric/ferrous ratios relative to the quartz–fayalite–magnetite (QFM) buffer. The mode of crystallization is fractional crystallization, whereby crystals, once crystallized, are chemically isolated (no further reaction allowed) from silicate liquid. We performed MELTS equilibrium crystallization calculations and found the results to be significantly off mark compared with CI geochemical data. The low average crystal content of the CI supports the occurrence of fractional crystallization. Chemical fractionation does not presuppose that crystals are physically removed from magma, although this may indeed occur (see below). It is obvious but still critical to note that isobaric removal of heat is the process driving fractional crystallization, a fact that facilitates construction of a timescale for petrological evolution (see below) that can be related to independent timescales derived from other constraints.
To document the pre-eruptive state of the CI magmatic system, we performed 110 MELTS crystal fractionation calculations for a range of potential parental compositions, pressures (0·1–0·5 GPa), oxygen fugacities (QFM – 1 to QFM + 3), and initial water concentrations (1, 2 and 3 wt%). The goal of these calculations is to develop bounds for the intensive thermodynamic parameters of pressure, oxygen fugacity and initial dissolved water content for a range of possible parental compositions that lead to the compositionally evolved CI liquid array depicted in Fig. 4. Each isobaric computation begins above the liquidus temperature and continues along a closed-system fractionation (CSF) path. Crystallization simulations were terminated at a minimum fraction of melt (fm) equal to 0·05. The ‘best-case’ set of intensive variables for a given starting composition was chosen based on comparison of MELTS predictions with observed pumice, glass and bulk-rock major element compositions, and comparison of predicted and observed phenocryst compositions. MELTS also predicts mineral proportions, but we have not used these as a criterion for model evaluation because of likely complications introduced by physical processes in the magma reservoir (e.g. separation of crystals and melt within the evolving magma body and during eruption) that may physically separate crystals from melt as well as each other.
Below, we present the results of sensitivity tests for a selected parental composition to demonstrate the effects of varying initial water content, pressure, and oxygen fugacity. We then present and discuss the best-case results.
|
ESTIMATION OF CAMPANIAN IGNIMBRITE PARENTAL MAGMA COMPOSITION |
|---|
The search for a representative CI parental melt composition is aided by the presence of MI found within clinopyroxene phenocrysts (Webster et al., 2003
). Webster and co-workers found a group of high-MgO (7 wt%) MIs with 53 wt% average SiO2 (Fig. 3) that represent a first approximation to the parental CI melt (Webster et al., 2003). A better approximation of the parental melt composition comes from recognition that the MIs within clinopyroxene phenocrysts are evidently saturated with respect to pyroxene and hence are related to the CI parental melt by reaction with their clinopyroxene host (i.e. clinopyroxene removal). As demonstrated by Watson (1976), MIs within multiple cotectic phases (e.g. plagioclase, olivine and clinopyroxene) can be related to a unique parental melt through a simple graphical construction. Unfortunately, the procedure developed by Watson could not be implemented here because MIs analyzed by Webster et al. (2003) are found solely in clinopyroxene. An attempt to reconstruct the parental composition following the method of Kress & Ghiorso (2004) failed to converge adequately essentially for the same reason; trapped MIs within multiple phases are needed to robustly reconstruct the parental liquid. Although, in principle, MIs within a single phase could be used, small uncertainties in activity–composition relations in monoclinic pyroxenes are too large to generate robust solutions. Similarly, detailed experimental data on the temperature and composition dependence of diffusivity of relevant species (Van Orman et al., 2001) necessary to implement the method proposed by Cottrell et al. (2002) are, unfortunately, not currently available. In light of these issues, we have had to invoke a less exact procedure.
Our method of parental melt reconstruction is based on two assumptions. The first is that the incompatible trace element concentration of the parental magma is lower than any other liquid related to it by crystal fractionation. Fortunately, trace element abundances have been measured in the high-MgO MIs (Webster et al., 2003), so these can be ordered by their incompatible trace element concentrations. MI sample VE1#3 has incompatible trace element concentrations that are generally lower than those of other high-MgO MIs from minimally altered CI units (i.e. Gray Tuff, Lithic Breccia, and Incoherent Tuff). In particular, concentrations of trace elements incompatible in clinopyroxene, including Ce, Th, Sm, and Nb, are lower in VE1#3 than in other high-MgO MIs by factors of 1·5–2. The second assumption is that a simple pyroxene-addition method can account for the effects of post-entrapment MI crystallization. The composition of clinopyroxene added to an analyzed MI corresponds to that of the MI host crystal taken from Webster et al. (2003). The reconstructed parental melt (RPM) is based on the mass balance relation
|
| (1) |
represents the mass fraction of host clinopyroxene added to the observed MI and 
represents the oxide composition vector of host clinopyroxene phenocrysts (cpx), MI and RPM, respectively. We used a range of values to construct an array of reasonable parental melt starting compositions (e.g. for VE1#3, = 0·1, 0·2, 0·3, and 0·4). A value of zero corresponds to the assumption that an MI is unmodified by reaction with its clinopyroxene host. Using this procedure, we performed approximately 15 MELTS simulations using as starting data the high-MgO MIs with low incompatible trace element concentrations (VE1#3 was the lowest of the group). For each MI studied, a range of reconstructed parental compositions was generated to test the quantitative effects of variable clinopyroxene addition (i.e. different ). Isobaric fractional crystallization simulations were then carried out for fixed values of oxygen fugacity, dissolved H2O concentration, and pressure for each RPM composition. Higher values result in higher proportions of magnesian phases crystallizing along the liquid line of descent, but major element paths are generally similar for values in the range 0·15–0·4. A parent composition based on = 0·2 best describes the observed major element data. Figure 5 compares the results of closed-system isobaric fractionation simulations based on two distinct parental melt compositions: unmodified MI VE1#3 with = 0 and VE1#3 with = 0·2. Both numerical simulations were carried out under the following conditions: pressure of 0·15 GPa, initial water content of 3 wt%, and oxygen fugacity fixed at the QFM + 1 buffer. The liquid line of descent based on VE1#3 with = 0·2 corresponds more closely to observed CI glass compositions than the results based on the unmodified VE1#3 composition. In both cases, there is a ‘jump’ or compositional discontinuity in the predicted sequence of liquids. This phenomenon is discussed in detail below. For now, the important point is that unmodified VE1#3 composition does not produce evolved liquids that fit observations (shaded field in Fig. 5). For example, in the VE1#3 ( = 0) case at low MgO values, the trend and the field for the CI data do not overlap (Fig. 5a). Al2O3 for the VE1#3 ( = 0) simulation also completely misses the CI field (Fig. 5b). This comparison clearly illustrates that a recalculated parental composition provides a better estimate of the parental magma for the CI. Although VE1#3 with = 0·2 provides the best correspondence to the CI data, similar results are found for other high-MgO MIs with values in the range 0·15–0·4. Based on these and additional numerical simulations of isobaric CSF, we conclude that the composition giving rise to fractionation trends most like those exhibited by the eruptive products of the CI is VE1#3 with = 0·2. For the remainder of this paper, this is referred to as the RPM. Table 3 presents the major element compositions of MI VE1#3, its host clinopyroxene, and the RPM of VE1#3 with = 0·2 (anhydrous).
|
|
|
PHASE EQUILIBRIA CONSTRAINTS ON THE CHEMICAL EVOLUTION OF THE CAMPANIAN IGNIMBRITE |
|---|
Results of closed-system fractional (CSF) crystallization MELTS simulations
In this section, we summarize results of numerous MELTS simulations of CSF to illustrate the effects of varying the initial H2O content, oxygen fugacity, and pressure on the calculated liquid line of descent and on phase relations. This is followed by a detailed comparison of observed and predicted liquid and solid (clinopyroxene, spinel, alkali feldspar, and plagioclase) phase compositions. The conclusion is that a liquid of RPM composition plus 3 wt% H2O undergoing closed-system isobaric fractionation at
6 km depth at between the QFM and QFM + 1 oxygen buffer conditions provides a good first-order model for derivation of the CI magma.
Effects of varying initial water concentration
The influence of initial water concentration on the fractional crystallization path of the CI was examined via isobaric (0·15 GPa) fractional crystallization simulations based on a parent melt of RPM and oxygen fugacity defined by the QFM + 1 buffer for 1, 2, and 3 wt% initial H2O. The results, presented in Fig. 6a and b, show that for 1 and 2 wt% initial H2O, SiO2 vs MgO trends do not intersect the data field for the CI; at low MgO, predicted SiO2 is much lower than observed. In contrast, at the lowest MgO values, Al2O3 greatly exceeds observed concentrations. A key difference between the 1 wt% H2O and the 3 wt% H2O case concerns the behavior of feldspar at lower melt fractions. An initial water concentration of 1 wt% H2O leads to stabilization of both feldspar phases at higher temperatures, with crystallization of plagioclase beginning at a higher temperature (1010°C) than that of alkali feldspar (965°C). In contrast, for the 3 wt% case, both feldspars saturate at the same temperature (883°C), and the abundance of both feldspar phases increases progressively with decreasing temperature. At the lowest melt fraction (fm = 0·05), the total proportion of feldspar is lower (0·37) in the 1 wt% case than in the 3 wt% H2O case (0·40) and, as a consequence, the liquid is characterized by greater Al2O3 enrichment and SiO2 depletion (Fig. 6a and b). In the 1 wt% H2O case, a rhombohedral oxide phase (ilmenite) appears at low melt fraction, which is inconsistent with the observed phase assemblage. Although not shown in Fig. 6, for cases in which initial dissolved water contents range from 4 to 6 wt%, calculated differentiation trends and the predicted phase assemblage do not agree with observed features of the CI. We therefore conclude that dissolved H2O contents around 3 wt% are most realistic for the initial dissolved water content of the RPM. The dramatic difference in behavior of the liquid line of descent at low melt fraction clearly shows that water contents around 3 wt% are required to best match CI data.
|
Effects of varying pressure
To explore effects of changing pressure, we compared results produced at fixed oxygen fugacity (QFM + 1) and 0·10, 0·15, 0·30, and 0·50 GPa. Results for the latter three pressures are presented in Fig. 7. An initial water concentration of 3 wt% was chosen, except in the case of the 0·10 GPa model, in which 3 wt% water supersaturates the system at the mineral liquidus. An initial water concentration of 2·6 wt% was chosen to set the system at water saturation at this low pressure.
|
The 0·30 and 0·50 GPa cases yield different phase assemblages at lower melt fraction when compared with the lower pressure cases of 0·10 and 0·15 GPa. The 0·30 and 0·50 GPa cases have leucite as a fractionating phase (Fig. 7b and c), a phase never observed in the CI. Compared with the 0·15 GPa case, where alkali feldspar and plagioclase saturate at
883°C (Fig. 7a), at 0·30 and 0·50 GPa the onset of feldspar crystallization occurs at lower temperatures (845°C in both cases), and with increasing pressure the mass fraction of feldspar crystallized in the first temperature increment of feldspar crystallization decreases (Fig. 7). As a consequence, compared with the 0·15 GPa case, isobaric fractionation at higher pressure yields SiO2 concentrations that are lower and Al2O3 concentrations that are substantially higher than the CI. An additional factor leading to SiO2 depletion at higher pressures is an increase in the calculated ratio of alkali feldspar to plagioclase in the first increment of feldspar crystallization (2·3 at 0·15 GPa, 5·5 at 0·30 GPa, and 148·2 at 0·50 GPa). Although liquid compositions are similar for the 0·10 and 0·15 GPa cases, the concentration of SiO2 is lower in the 0·10 GPa case because the proportion of feldspar over the first increment of feldspar crystallization is 2% lower.
The misfit of oxide data, coupled with the mismatch between observed and predicted phases, eliminates the possibility that the CI magma evolved at pressures of 0·30 GPa or higher. Results for the 0·10 GPa case are very similar to those for the 0·15 GPa case. However, at low melt fraction, the concentration of SiO2 is lower in the 0·10 GPa case and therefore, the 0·15 GPa case more closely approximates the range of SiO2 observed in the CI. We conclude that the major phase of crystal fractionation to generate CI composition melt took place at relatively low pressure (0·15 GPa) corresponding to a depth of about 6 km.
Effects of varying oxygen fugacity
With pressure and initial water concentration values fixed at 0·15 GPa and 3 wt%, respectively, we performed additional calculations to examine the effects of varying oxygen fugacity (QFM – 1, QFM, QFM + 1, QFM + 2, and QFM + 3). The results illustrate the strong influence of oxygen fugacity on the relative stability of Fe2+- and Fe3+-bearing phases. Olivine is the liquidus phase in all computations described in this study, but with oxygen fugacity constrained to follow the QFM – 1 buffer curve, crystallization of fayalitic olivine is also predicted at low melt fraction and makes up 7·5% of the end (fm = 0·05) phase assemblage. In comparison, at QFM + 1 and QFM + 3, olivine is stable only at high melt fraction (fm > 0·9). The decreasing abundance of olivine with increasing f O2 leads to stabilization of clinopyroxene and a slightly higher final clinopyroxene content (44% at QFM + 3, 42% at QFM – 1). More oxidizing conditions naturally lead to stabilization of spinel at higher temperatures and slight increases in the mass proportion of spinel [0·01 at QFM – 1 to 0·03 at QFM + 3 of the end (fm = 0·05) phase assemblages]. Differences in spinel stability lead to differences in SiO2 and FeO concentrations; Fig. 8a and b summarizes these differences for all of the five oxygen fugacity cases. In the QFM + 3 case, the relatively high abundance of spinel and onset of spinel crystallization at higher temperature leads to corresponding shifts in melt SiO2 and FeO contents. Maximum SiO2 concentrations are 1·5 wt% higher than in the QFM + 1 case, whereas the maximum difference in FeO contents is 3 wt%, with the QFM + 3 having the lower value. In contrast, the lower abundance of spinel in the QFM – 1 case leads to predictions of melt SiO2 concentrations that are lower than those of the 0·15 GPa case by a minimum of 1·5–2 wt%; in addition, at the lowest melt fraction (fm = 0·05), FeO is higher by >1 wt%.
|
The choice of oxygen buffer for the best case, which we define as between QFM and QFM + 1, is most influenced by the behaviors of SiO2 and FeO; the poor correlation between observed and model trends for QFM – 1, QFM + 2 and QFM + 3 demonstrates that these oxygen buffers are not relevant. We conclude that the fugacity of oxygen during isobaric fractionation between QFM and QFM + 1 gives the closest agreement between the numerical simulation and CI compositions.
Best-case closed-system fractional crystallization model
Based on the comparisons described above, the most plausible parental starting composition involves the anhydrous composition given in Table 3 plus 3 wt% H2O. This starting composition undergoes fractional crystallization along the QFM to QFM + 1 oxygen buffer at 0·15 GPa. For simplicity, the following discussion is based on the QFM + 1 results. Here we provide a detailed comparison between this particular case, hereafter referred to as CSF, and observations.
Mineral identities and abundances and the temperature at which water saturates for CSF along the liquid line of descent are summarized in Fig. 7a. Olivine is the liquidus phase (Tliq = 1235°C) followed by clinopyroxene, H2O, spinel and apatite at 1162°C, 1127°C, 1078°C and 1018°C, respectively. Significantly, at 883°C, alkali feldspar, plagioclase, and biotite appear simultaneously and crystallize along with apatite, spinel and clinopyroxene. At and below 883°C, crystallization is dominated by the growth of alkali feldspar and, to a lesser extent, plagioclase. A small modal amount of olivine (5% by mass) is predicted to crystallize during the first 100°C of cooling. With the exception of olivine (not reported in the CI), the CSF predictions match observed phases (alkali feldspar, plagioclase, clinopyroxene, biotite, spinel, and apatite) very well. Olivine crystals are not found as phenocrysts in the CI perhaps because, as the first phase to crystallize from the parental melt of low viscosity, settling has been especially efficient. Calculated liquid compositions of the CSF define trends that reflect removal of these mineral phases and are plotted in Fig. 4. Because the oxygen buffer is interpreted to lie between QFM and QFM + 1, Fig. 4 shows results for both buffers. Concentrations of SiO2, K2O, Na2O, and Al2O3 initially increase with decreasing MgO, reflecting olivine fractionation, and continue to increase as clinopyroxene, spinel, and apatite join the fractionating phase assemblage. The increase in SiO2 concentration becomes more gradual when clinopyroxene replaces olivine, but then increases sharply when spinel appears. CaO contents rise until clinopyroxene appears, then decrease sharply with further cooling. FeO decreases slightly in the early stages of crystallization, but decreases markedly with spinel crystallization. Whereas results using the QFM buffer more closely track observed FeO, the QFM + 1 buffer yields SiO2 that extends to higher values, and therefore, more closely approximates the observed range of SiO2.
Perhaps the most striking feature of the CSF calculation results is the abrupt change in melt composition (i.e. a compositional gap) at 883°C (Fig. 4). Within a degree or so of this temperature, identified here as the pseudo-invariant temperature, Tinv , melt simultaneously saturates in alkali feldspar, plagioclase and biotite, leading to a dramatic decrease in fm, from 0·46 to 0·09 (Fig. 7a). That is, a major increment of crystallization occurs essentially at a single temperature. At this pseudo-invariant point, the specific enthalpy of the system changes in response to the ‘wave’ of crystallization. Major element shifts in melt composition at Tinv are 2 wt% for Na2O and Al2O3, 1·5 wt% for K2O, 1 wt% for SiO2 and CaO, and 0·5 wt% for H2O (Fig. 4). For T < Tinv, SiO2 and Na2O concentrations in the melt continue to increase as a result of feldspar fractionation and the increase in Na2O becomes particularly steep. K2O, Al2O3, FeO and CaO decrease at T < Tinv. Compositional changes at Tinv are associated with physical changes [e.g. in the density and viscosity of water-saturated melt and magma (melt + supercritical fluid)] that have important dynamical consequences (discussed below).
As shown in Fig. 4, predicted CSF trends in melt composition agree fairly well with pumice or bulk-rock, and glass data for most oxides. Because the clinopyroxene-hosted MIs of Webster et al. (2003) have reacted with