U SERIES DISEQUILIBRIA: INSIGHTS INTO MANTLE MELTING AND THE TIMESCALES OF MAGMA DIFFERENTIATION

Transcription

1 U SERIES DISEQUILIBRIA: INSIGHTS INTO MANTLE MELTING AND THE TIMESCALES OF MAGMA DIFFERENTIATION David W. Peate Department of Geoscience University of Iowa Iowa City, Iowa, USA Chris J. Hawkesworth Department of Earth Sciences University of Bristol Bristol, UK Received 22 March 2004; revised 18 December 2004; accepted 17 January 2005; published 31 March [1] Several U series nuclides have half-lives ( 230 Th, 76 kyr; 231 Pa, 33 kyr; and 226 Ra, 1.6 kyr) comparable to timescales of magmatic processes. We review the basic principles of extracting time information from U series nuclides and summarize variations in ( 230 Th/ 238 U), ( 226 Ra/ 230 Th), and ( 231 Pa/ 235 U) observed in magmas from mid-ocean ridges, within-plate settings, and subduction zones to contrast melt generation processes in different tectonic settings. U series disequilibria on melt and crystal phases of igneous rocks can provide temporal information on different stages in the magmatic history (melting duration, melt transport rates, magmatic crustal residence times, and timing of crystal growth) and potentially provide clues about the nature and mineralogy of mantle sources, mantle upwelling rates and porosity, fluid influences, and mechanisms of melt generation and transport. The subject is beginning to take a genuinely integrated approach to developing physically realistic quantitative models that offer increasingly exciting opportunities in the study of magmatic processes. Citation: Peate, D. W., and C. J. Hawkesworth (2005), U series disequilibria: Insights into mantle melting and the timescales of magma differentiation, Rev. Geophys., 43,, doi: /2004rg INTRODUCTION [2] In order to develop quantitative models for magmatic processes it is necessary to have constraints on the likely rates and timescales of processes such as melting, differentiation, and crystallization, and yet such information has been difficult to estimate from the geological record. However, the increasingly widespread application of U series isotopes, driven by advances in analytical methods in the late 1980s, has revolutionized our understanding of melt generation and crystallization processes and their timescales. Unlike many radioactive nuclides that decay directly to a stable daughter nuclide (e.g., 87 Rb! 87 Sr, 147 Sm! 143 Nd, and 176 Lu! 176 Hf ), the long-lived isotopes of uranium and thorium ( 238 U, 235 U, and 232 Th) decay ultimately to isotopes of Pb ( 206 Pb, 207 Pb, and 208 Pb, respectively) via a series of intermediate unstable nuclides that have half-lives ranging from 250 kyr (thousand years) to microseconds (Figure 1). Attention has focused mainly on the decay products of 238 U and 235 U, the so-called U series nuclides, because several of these intermediate nuclides have half-lives (t 1/2 ) ( 230 Th, t 1/2 76 kyr; 231 Pa, t 1/2 33 kyr; 226 Ra, and t 1/2 1.6 kyr; see Table 1) that are comparable with the inferred timescales of magmatic processes. [3] The utility of these U series nuclides stems from the unique properties of the decay chains. If the system is undisturbed, then it is said to be in secular equilibrium in which the activity of all nuclides in the decay chain are equal. The activity of a nuclide is the rate of radioactive decay of that nuclide in the sample. Typical measurement units for activity are decays per minute per gram (dpm g 1 ) or Bequerels per gram (Bq g 1 or decays per second per gram). The activity of a nuclide is simply the product of the number of atoms of the nuclide and the decay constant l, where l = ln(2)/t 1/2.As the parent nuclide at the top of each decay chain has a much longer half-life than all the intermediate nuclides, the activity or flux of radioactive decay from this parent nuclide remains essentially constant over timescales relevant to the study of the shorter-lived intermediate nuclides. Because different elements have distinct chemical properties, it is possible for magmatic processes to change the relative abundances of the different U series nuclides from the proportions expected for secular equilibrium, such that the activities of the nuclides in the decay chain are no longer equal. This elemental fractionation leads to disequilibrium between parent and daughter nuclides within the chain, and the return to secular equilibrium is governed by the half-life of the shorterlived nuclide. It is this process of reestablishing secular equilibrium that allows timescales of magmatic processes to be determined. Disequilibrium between a parent and daughter nuclide pair is detectable up to about five half- Copyright 2005 by the American Geophysical Union /05/2004RG000154$15.00 Reviews of Geophysics, 43, / of43 Paper number 2004RG000154

2 Figure 1. Periodic table summary of parts of the 238 U, 235 U, and 232 Th decay chains. Nuclides of interest are outlined in black with the half-lives given below. The very short-lived intermediate nuclides of Pa and Th are indicated as shaded. The dotted arrows indicate the decay of 226 Ra, 231 Pa, and 228 Ra to 206 Pb, 207 Pb, and 208 Pb, respectively, via a series of very short-lived intermediate nuclides that are omitted for clarity. lives of the shorter-lived nuclide, depending on the initial extent of disequilibrium and the precision of the measurements. For example, if 230 Th- 238 U disequilibrium is measured in a sample (i.e., activity of 230 Th does not equal activity of 238 U), then this indicates that an event that chemically fractionated U from Th took place less than 380 kyr ago as the half-life of 230 Th is 76 kyr. [4] Processes such as partial melting, crystal growth, and fractional crystallization can all potentially produce disequilibrium between U series nuclides with different chemical properties. Therefore we should, in principle, be able to use these nuclides to constrain transport rates of melts from mantle sources, the residence time of magma in crustal reservoirs, and the timing of crystal growth in addition to estimating eruption ages. During partial melting, disequilibria between U series nuclides will depend critically on the source mineralogy, the rate of melting, and the degree of interaction with the surrounding mantle as the melt percolates upward. Therefore U series disequilibria data on primitive magmas can potentially also provide important information about the nature and mineralogy of the mantle source, the porosity during melting, and plausible mechanisms of melt generation and transport in the mantle. [5] The 230 Th- 231 Pa- 226 Ra isotopes are the most widely used in the study of igneous rocks, but there is also considerable potential to exploit isotopes with even shorter half-lives, including 228 Ra (t 1/ years), 210 Pb (t 1/2 22 years), and 210 Po (t 1/2 138 days), particularly to look at timescales of magma degassing and the timing of very recent crystal growth [see Condomines et al., 2003, and references therein]. This paper aims to provide an accessible and up-todate review for a broad audience of the applications of U series disequilibria to magmatic processes, focusing on the 2of Th- 238 U, 226 Ra- 230 Th, and 231 Pa- 235 U schemes. In the decade or so since the last published general reviews [e.g., Gill and Condomines, 1992; Gill et al., 1992; Macdougall, 1995], significant advances have been made, particularly through detailed comprehensive regional studies combining 230 Th- 226 Ra- 231 Pa data on otherwise well-characterized samples and more advanced modeling incorporating new ideas and observations from other areas of geophysics and geochemistry. [6] This paper is structured in four parts: (1) a summary of the basic principles, concepts, and assumptions behind U series disequilibria applications; (2) a summary of the degree and sense of disequilibria observed between 230 Th- 238 U, 226 Ra- 230 Th, and 231 Pa- 235 U in young igneous rocks from different tectonic environments; (3) a discussion of how disequilibria are used to constrain timescales of magma differentiation and crystal growth; and (4) a discussion of the controlling factors in generating the observed disequilibria in magmas from different tectonic settings, together with implications for melt generation models. More detailed recent reviews on these individual TABLE 1. Half-Lives and Decay Constants for U Series Nuclides of Interest a Nuclide Half-Life, years Decay Constant (l), year Th ± ± U ± ± U ± ± U 245,250 ± ± Th 75,690 ± ± Pa 32,760 ± ± Ra 1,599 ± ± a Recommended values are from Bourdon et al. [2003a].

3 Figure 2. Disequilibrium within each of the parentdaughter nuclide pairs ( 226 Ra- 230 Th, 231 Pa- 235 U, and 230 Th- 238 U) returning to secular equilibrium (i.e., activity ratio equal to 1) over a timescale governed by the half-life of the daughter nuclide (i.e., 226 Ra, 231 Pa, and 230 Th, respectively). Figure 2 also emphasizes that U series disequilibrium methods are only applicable to geologically very young samples. Examples are given for the temporal evolution of disequilibria in samples with initial daughter excesses of 20% and initial daughter deficits of 20%. topics are available [see Bourdon et al., 2003c; Rudnick, 2003]. 2. BACKGROUND TO U SERIES DISEQUILIBRIA IN IGNEOUS ROCKS 2.1. Systematics and Conventions [7] Isotope ratios of the U series nuclides are usually expressed as activity ratios, and by convention they are enclosed in parentheses, e.g., ( 230 Th/ 238 U). Activity ratios are easily related to the atomic ratio by the ratio of the respective decay constants, i.e., ( 230 Th/ 238 U) activity = [ 230 Th/ 238 U] atomic [l 230 /l 238 ]. At secular equilibrium the activities of all nuclides in a particular decay chain are equal, and so ( 230 Th/ 238 U) = 1, ( 226 Ra/ 230 Th) = 1, ( 231 Pa/ 235 U) = 1, and ( 234 U/ 238 U) = 1. As fresh igneous rocks are both expected and usually observed to have ( 234 U/ 238 U) = 1, then 238 U can essentially be treated as the parent of 230 Th. It is important to remember that for all these U series nuclide pairs both the parent and daughter nuclides are unstable and will undergo radioactive decay at different rates depending on their respective half-lives. [8] If a sample has measured disequilibrium between a particular parent-daughter pair because of some chemical process that affected the two elements to different extents, it is said to have an excess of the nuclide with the higher activity. For example, a mid-ocean ridge basalt with ( 230 Th/ 238 U) = 1.20 has a 20% 230 Th excess. In this case, there is an insufficient quantity of the parent nuclide 238 Uto maintain this amount of excess daughter 230 Th through radioactive decay, and the excess 230 Th is therefore said to be unsupported. This excess will decay away at a rate governed by the 230 Th half-life until the amount of 230 Th is equal to that being produced by radioactive decay from the 238 U in the sample. 3of43 [9] On the other hand, a subduction zone lava with ( 230 Th/ 238 U) = 0.80 has a 20% deficit of the daughter nuclide, and the amount of 230 Th is less than that being produced from decay of 238 U. In this case the amount of 230 Th will gradually increase at a rate determined by the 230 Th half-life until the amount of 230 Th in the sample is equal to that being produced by 238 U decay. [10] As it takes about five half-lives of the shorter-lived nuclide for secular equilibrium to be restored between a parent and daughter nuclide pair, this means that the different U series nuclide pairs provide information on different timescales (Figure 2): The 230 Th- 238 U system records U-Th fractionation within the last 380 kyr, the 226 Ra- 230 Th system records Ra-Th fractionation within the last 8 kyr, and the 231 Pa- 235 U system records Pa-U fractionation within the last 165 kyr. Note that igneous geochemists use the word fractionation in a broad sense to cover the effects of any process that changes the relative abundances of different elements or isotopes. The magnitude of 230 Th- 238 U, 226 Ra- 230 Th, and 231 Pa- 235 U disequilibria measured in a sample today will be governed by several factors: (1) the extent of chemical fractionation between parent and daughter nuclides (e.g., partial melting, source mineralogy, crystallization, diffusion, melt immiscibility, and melt-matrix interaction); (2) the elapse of time since the fractionation event during which radioactive decay will occur (e.g., melt transport time from mantle source regions to the crust, residence time in crustal magma reservoirs, and time since lava eruption); and (3) surface weathering or hydrothermal alteration processes. [11] It should now be apparent that U series disequilibria methods can essentially be applied only to geologically very young samples (Figure 2): For example, samples older that about 380 ka should have ( 238 U/ 230 Th) = 1, as secular equilibrium will have been reestablished. Furthermore, the eruption ages of samples need to be well known in order to be able to correct for any posteruptive radioactive decay. For example, if it is known only that a sample was erupted sometime in the last 8000 years that is fine for the 238 U- 230 Th system because this is a short time period compared to the 76 kyr 230 Th half-life. However, for the 226 Ra- 230 Th system, which will return to secular equilibrium over a similar period of about 8000 years, it is clear that the eruption ages would need to be known more precisely Analytical Developments [12] The most significant analytical advance (see reviews by Chen et al. [1992] and Goldstein and Stirling [2003]) has been the development of thermal ionization mass spectrometric (TIMS) techniques for the measurement of 234 U [Chen et al., 1986], 230 Th [Edwards et al., 1987; Goldstein et al., 1989; Palacz et al., 1992], 226 Ra [Cohen and O Nions, 1991; Volpe et al., 1991; Chabaux et al., 1994], and 231 Pa [Pickett et al., 1994; Bourdon et al., 1999a]. Earlier studies used a spectrometry, which counts the number of a particles emitted by radioactive decay from a sample in a given time. For nuclides such as 230 Th that have relatively slow decay rates and low abundances in rocks,

4 long counting times on the order of tens of weeks are needed to achieve acceptable counting statistics. Mass spectrometry, on the other hand, counts the number of ions rather than waiting for the radioactive decay of the nuclides, thus achieving far more favorable counting statistics for the nuclides of interest here ( 234 U, 235 U, 238 U, 230 Th, 232 Th, 226 Ra, and 231 Pa). The 2s precisions of 0.1 2% can be obtained by mass spectrometer measurements of less than 2 hours duration, compared with 2 10% for much longer a spectrometry measurements, and sample size requirements for mass spectrometry are also significantly less than for a spectrometry (e.g., U-Th, mg versus mg, and Pa-Ra, pg versus pg) [Goldstein and Stirling, 2003]. This reduction in sample size has been especially advantageous for studies looking at magma chamber processes and geochronology by analysis of separated mineral phases from rocks. [13] Mass spectrometric measurement of the isotopic composition of thorium poses several particular challenges. First, most igneous rocks have Th/U ratios of 3 ± 1, which is equivalent to an atomic 230 Th/ 232 Th ratio of (6 ± 3) Thus for silicate samples, there is a huge contrast (5 orders of magnitude) in beam intensity between the major 232 Th peak and the minor 230 Th peak. Collisions with residual molecules within the mass spectrometer cause ion scattering, leading to a low mass tail from the large 232 Th peak that partly obscures the small 230 Th peak. An energy filter is needed to reduce this background from scattered ions so as to improve the detectability of the 230 Th peak. Second, thorium is a difficult element to ionize thermally, with less than 1% of the sample ionized. Therefore the recent development of multicollector inductively coupled plasma mass spectrometers, which are much more efficient at ionizing Th and other elements, offer much potential for the future, particularly in further reducing sample size requirements and improving measurement precision. Plasma-based techniques have been developed for the measurement of 230 Th/ 232 Th and 234 U/ 238 U [Luo et al., 1997; Turner et al., 2001b; Shen et al., 2002], 226 Ra [Pietruszka et al., 2002], and 231 Pa [Regelous et al., 2004] Robust Indicators of Magmatic Processes? [14] If U series disequilibria are to be used to provide information about magmatic processes, it is important to confirm that the observed disequilibria are primary and magmatic in origin and that they do not result from posteruptive alteration through interactions with seawater or groundwater or from shallow level contamination with sedimentary or hydrothermal materials. Most concern has centered on mid-ocean ridge basalts (MORB), where the potential for contamination is high given the relatively low initial elemental concentrations and the availability of materials potentially enriched in various U series nuclides. Palagonite rinds and crystalline pillow lava interiors have elevated uranium contents relative to unaltered glass samples, indicating pervasive uptake of uranium from seawater [e.g., Macdougall et al., 1979]. Fe-Mn oxides and hydrothermal precipitates that often coat MORB glasses contain significant amounts of unsupported 230 Th, 231 Pa, and 234 U scavenged from seawater, and inclusion of even small amounts of such material will significantly influence the measured disequilibria. Therefore careful handpicking of glass chips free of any visible alteration and Fe-Mn oxide coatings, together with chemical leaching to remove material in cracks, is critical for analysis of MORB glasses. Bourdon et al. [2000a] used the abundance of 10 Be to place limits on the contribution of metalliferous sediments to the measured U series disequilibria in handpicked and leached MORB glasses. Beryllium 10 is a radioactive nuclide (t 1/2 1.5 Myr) that is only produced at significant rates through cosmogenic processes in the atmosphere. It is highly enriched in the metalliferous sediments found near the ridge axes, but it can be assumed to be absent in the pristine mantle-derived glasses, and thus any 10 Be found in the MORB glass must result from near-surface sedimentary contamination. The 10 Be data demonstrate that sedimentary contamination can only account for <1% of the measured excess 230 Th and 231 Pa in carefully selected and leached samples. [15] Although 226 Ra is not strongly concentrated in Fe-Mn oxides, it is likely to be highly enriched in hydrothermal Ba-rich phases such as barite, which is a ubiquitous mineral in the ridge environment [Volpe and Goldstein, 1993]. It is more difficult to assess how the measured 226 Ra- 230 Th disequilibria in individual samples might have been influenced by assimilation of even minor amounts of barite, although one would expect elevated Ba contents relative to other highly incompatible elements such as Rb or Th. The near constancy of Ba/Rb for both mid-ocean ridge basalts and ocean island basalts and the limited variations of Ba/Th with 226 Ra excesses suggest that barite assimilation has a minimal influence on the composition of ridge basalts [Lundstrom et al., 1999; Elliott and Spiegelman, 2003]. [16] The 234 U/ 238 U ratio of a sample can provide a useful monitor of alteration by seawater or groundwaters. At magmatic temperatures, 234 U is not expected to be fractionated from 238 U, and so fresh igneous rocks should have ( 234 U/ 238 U) = 1, whereas most natural waters show disequilibrium between 234 U and 238 U, principally because of alpha recoil effects. The physical process of alpha decay by a 238 U atom results in recoil of the daughter nuclide 234 Th (which rapidly decays to 234 U) because of conservation of momentum, so that it is displaced from its original location. It is either ejected directly into the surrounding fluid phase or sits in a now damaged site in the crystal where it is more susceptible to mobilization by fluids. Seawater is well mixed with respect to uranium and has ( 234 U/ 238 U) of [Chen et al., 1986]. [17] The effects of seawater alteration are illustrated in Figure 3 using two different sample suites: (1) submarine glass and crystalline pillow interior powders from a single MORB sample from the Reykjanes Ridge [Peate et al., 2001a; D. Peate, unpublished data, 1996] and (2) subaerial arc lavas from a single coastal outcrop on Miyakejima volcano, Japan [Yokoyama et al., 2003]. In each case the 4of43

5 Figure 3. ( 234 U/ 238 U) versus ( 232 Th/ 238 U) diagram showing the effects of seawater contamination on coastal subaerial arc lavas (Miyakejima volcano, Japan [Yokoyama et al., 2003]) and submarine mid-ocean ridge basalts (Reykjanes Ridge [Peate et al., 2001a; D. Peate, unpublished data, 1996). Altered samples (shaded symbols) lie on a linear mixing line between fresh samples with ( 234 U/ 238 U) = 1 (solid symbols) and seawater with ( 234 U/ 238 U) = and ( 232 Th/ 238 U) = 0 (solid diamond). Note that even minor seawater contamination that elevates ( 234 U/ 238 U) in a sample by just 1% will produce a corresponding decrease of almost 7% in the Th/U ratio relative to the pristine sample, and it will have an effect of similar magnitude on the 230 Th- 238 U disequilibrium [Yokoyama et al., 2003]. variably altered samples lie on a linear mixing line between pristine samples with ( 234 U/ 238 U) = 1 and seawater, and an elevation in ( 234 U/ 238 U) of just 1% corresponds to almost 7% difference in ( 232 Th/ 238 U) from the true pristine value. This observation highlights why it is critical to measure ( 234 U/ 238 U) to a high precision 5% [e.g., Sims et al., 2002a; Yokoyama et al., 2003] on all subaerial samples from coastal regions and on submarine samples in order to screen for altered samples. However, ( 234 U/ 238 U) data are often not published along with the other disequilibria data ( 226 Ra- 230 Th- 238 U and 231 Pa- 235 U), and it is notable that less than 40% of the data compiled for the global review of disequilibria in igneous rocks discussed in section 3 have accompanying ( 234 U/ 238 U) data Principal Mechanisms to Produce 238 U- 230 Th, 226 Ra- 230 Th, and 231 Pa- 235 U Disequilibria Trace Element Partitioning [18] We can define trace elements as those elements that have low abundances in igneous rocks, i.e., less than about 0.1% or 1000 ppm (parts per million) by weight. This definition includes all of the U series nuclides of interest to us (i.e., U, Th, Ra, and Pa). The concept of the partition coefficient (or distribution coefficient) is an important tool to describe quantitatively the behavior of trace elements in magmatic systems. The partition coefficient simply measures the relative partitioning of a trace element between two phases. The Nernst partition coefficient (D) is used to quantify the partitioning of a trace element at equilibrium, usually between a mineral and a melt phase. It is given by the concentration of the element in the mineral divided by the concentration of the element in the melt. For example, if a clinopyroxene phenocryst contains 1 ppm Th and the glassy matrix of the host lava contains 20 ppm Th, then D Th clinopyroxene/melt = When more than one mineral phase is present, such as during mantle melting (olivine + orthopyroxene + clinopyroxene + garnet or spinel), a bulk partition coefficient must be calculated to describe partitioning of a trace element between the melt and the bulk solid. This is done by weighting the mineral/ melt partition coefficients of the individual mineral phases by their proportions in the bulk solid. [19] If an element has a mineral/melt partition coefficient of less than 1, this means that the element prefers to be in the melt rather than in that particular mineral, and it is said to be incompatible in the mineral. Conversely, a partition coefficient greater than 1 implies that the element preferentially partitions into the mineral relative to the melt, and it is said to be compatible in the mineral. Partition coefficients are a measure of the extent to which a trace element can be accommodated in the crystal structure of a particular mineral, which will be governed by the ionic charge and radius of the incorporated element and the nature of the crystallographic sites within the mineral. The values of specific mineral/melt partition coefficients have to be determined experimentally, and they vary as a function of temperature, pressure, and the composition of the mineral. However, crystal lattice strain models are being developed to allow extrapolation of the experimental results to other temperatures and pressures and compositions and even to predict values for other elements (see Blundy and Wood [2003] for a comprehensive review of partitioning data for U series nuclides) Crystallization [20] As a magma cools, different mineral phases nucleate and begin to grow. The different minerals will incorporate or exclude different trace elements to different degrees, depending on the value of the relevant mineral/melt partition coefficient. If one element is preferentially incorporated into a certain mineral phase relative to another element, then removal of this mineral phase from the magma during fractional crystallization will change the ratio of the two elements in the magma, i.e., lead to fractionation of the two elements. The extent to which this elemental ratio is changed will depend on the ratio of the partition coefficents of the two elements, the absolute values of the partition coefficients, and the amount of the mineral removed from the magma. [21] U and Th are highly incompatible elements in most common crystallizing minerals (olivine, pyroxene, feldspar, amphibole, mica, and Fe-Ti oxides) with partition coefficients <0.05 [Blundy and Wood, 2003]. Thus these common phases will all have very low concentrations of U and Th relative to the host magma. The low U and Th concentrations 5of43

6 in the crystalling phases mean that not much U or Th will be extracted from the magma as the crystals settle out during magmatic differentiation. Therefore closed system fractional crystallization does not usually result in significant variations in the U/Th ratios of related whole rock samples, irrespective of any differences in the relative partition coefficients for U and Th. Fractional crystallization is therefore also unlikely to be a significant cause of 238 U- 230 Th disequilibrium in basaltic magmas. However, 238 U- 230 Th disequilibrium can be produced in the crystallizing minerals because small differences in partition coefficients will determine the relative extents to which 238 U and 230 Th are incorporated into the crystals from the melt, even if these amounts are small compared to the abundances in the melt. [22] Crystallization of small amounts of accessory phases with very high U and/or Th contents (e.g., zircon, apatite, sphene, chevkinite, allanite, and monazite), on the other hand, can cause significant variations in U/Th ratios in more differentiated magmas. For example, phonolitic glasses from the 13 ka Laacher See zoned eruption in Germany show a 12% variation in U/Th that Bourdon et al. [1994] interpreted as being due to rapid crystallization and removal of apatite and sphene, which are accessory phases with high-th and high-u contents and low U/Th. [23] Radium partitions into feldspar crystals more readily than Th (D Th 0.003), and it becomes a compatible element (i.e., D Ra > 1) for many alkali feldspar compositions [Blundy and Wood, 2003]. Feldspar crystals (especially alkali feldspar) will develop elevated 226 Ra/ 230 Th values compared to the magma, and thus their removal during differentiation should lower 226 Ra/ 230 Th values in the magma. Fractional crystallization of feldspar is therefore a potential cause of 226 Ra- 230 Th disequilibrium in more evolved magmas. The effect of fractional crystallization on 231 Pa- 235 U disequilibrium is uncertain because of a lack of experimentally determined partition coefficient data and a lack of direct measurements of 231 Pa- 235 U in separated minerals. Both Pa and U are likely to be highly incompatible in most common crystallizing minerals [Blundy and Wood, 2003], and so minimal effect on 231 Pa- 235 U disequilibrium during fractional crystallization is expected. However, Pa might be partitioned strongly into oxides (e.g., ilmenite and rutile) and zircon [Blundy and Wood, 2003], and so these phases might influence 231 Pa- 235 U disequilibrium during differentiation of more evolved magmas Melt Generation and Transport [24] The melting process is probably the principal mechanism that produces 238 U- 230 Th, 226 Ra- 230 Th, and 231 Pa- 235 U disequilibrium in magmas. Disequilibrium during melting originates through differences in the chemical behavior of the various U series nuclides, as reflected primarily in different partition coefficients. A common assumption is made that the source rocks are in secular equilibrium, which will be the case unless there has been any recent chemical modification of the source. Therefore the initial activity ratios of the source will be known, i.e., ( 238 U/ 230 Th) = 1, ( 226 Ra/ 230 Th) = 1, and ( 231 Pa/ 235 U) = 1, which is not the case for trace element ratios. This assumption might not hold in certain situations: for example, if melting in subduction zones is triggered by recent addition of a fluid rich in U and Ra but not Th and Pa. [25] The simplest melting models treat the U series nuclides like any other trace elements and attribute any disequilibrium between U series nuclides simply to net elemental fractionation as controlled by differences in partitioning between melt and residual solid. They include a range of different physical models of melt generation from batch melting (where all the melt produced stays in equilibrium with the residual solids until it is extracted instantaneously to the surface) to fractional melting (where infinitesimal amounts of melt are extracted instantaneously as soon as they are produced), but the common feature is that they are time-independent. A characteristic of any equilibrium melting model is that significant trace element fractionation will only occur when the degree of melting is similar to or less than the bulk partition coefficients. For example, bulk partition coefficients for U and Th in typical peridotitic mantle will be 10 3, and those for Ra and Pa are likely to be at least a factor of 10 smaller [Blundy and Wood, 2003]. In this case the degree of melting would have to be extremely small, on the order of a percent or less, in order to produce significant disequilibrium between the U series nuclides. [26] Melting takes place over a finite timescale that is likely to be similar in magnitude to the half-lives of 230 Th, 231 Pa, and 226 Ra. Thus the duration of melt generation and melt extraction will also play a critical role in determining the degrees of disequilibrium between these nuclides that are preserved in a magma exiting the melting region. This will be influenced by the physical dynamics of the melt generation process in terms of parameters such as melting rate, porosity, melt velocity, solid upwelling rate, and the length of the melting column. Melting models that explicitly take into account the radioactive ingrowth and decay of the daughter nuclides during melt production are often referred to as ingrowth models. [27] In these models, movement of melt relative to the solid during the melting process (i.e., two-phase flow [McKenzie, 1985]) will lead to disequilibrium between the U series nuclides because of differences in the residence time of parent and daughter nuclides within the melting region. To explain the principle of daughter ingrowth in simple terms, consider the slow melting of a garnet peridotite mantle source. For this composition, Th is more incompatible than U (i.e., D Th < D U ), and so the 230 Th that was initially in secular equilibrium with 238 U in the source will be preferentially partitioned into the first melt formed, relative to 238 U. This leaves the residual solid deficient in 230 Th (i.e., ( 230 Th/ 238 U) <1), and additional 230 Th will begin to accumulate (or ingrow) from decay of the 238 Uinthe solid as the system attempts to reestablish secular equilibrium. If the melt is moving faster than the solid matrix, then U and Th will have different residence times within the melting region because of their different bulk partition coefficients. The less incompatible element (U) will spend 6of43

7 Figure 4. Examples of how ( 230 Th/ 238 U), ( 226 Ra/ 230 Th), and ( 231 Pa/ 235 U) activity ratios vary for a given set of parameters for an equilibrium porous flow model in a midocean ridge setting in terms of porosity and mantle upwelling velocity [from Spiegelman, 2000]. In principle, it is possible to determine unique values for porosity and upwelling velocity that are consistent with different disequilibria measured in a single sample, but it is important to realize that these results are model-dependent (both in terms of the exact melting model used and in the choice of partition coefficients). more time in the melting column within the solid matrix compared to the more incompatible element (Th) that is preferentially partitioned into the faster moving melt. Further melting will continue to preferentially add the 230 Th ingrown from 238 U in the source at a faster rate than 230 Th decay in the melt, because the melt travels faster than the matrix, thus enhancing the 230 Th- 238 U disequilibrium in the final melt. As D Th > D Ra and D U > D Pa for mantle compositions, then daughter ingrowth can also enhance 226 Ra- 230 Th and 231 Pa- 235 U disequilibria in mantle melts. The main difference between the various ingrowth melting models is the extent to which the melt equilibrates with the solid during melt extraction: the end-member models being continuous equilibration [e.g., Spiegelman and Elliott, 1993] and chemical isolation [e.g., McKenzie, 1985; Williams and Gill, 1989] (see section 5.1). [28] As the different disequilibria pairs ( 230 Th- 238 U, 226 Ra- 230 Th, and 231 Pa- 235 U) respond differently to the various parameters of melting models because of significant differences in bulk partition coefficients and half-life, the potential of the U series method will clearly be maximized if all three pairs are measured on the same sample. Figure 4 shows an example [from Spiegelman, 2000] of how measurements of ( 230 Th/ 238 U), ( 226 Ra/ 230 Th), and ( 231 Pa/ 235 U) from a single mid-ocean ridge basalt sample could potentially be used to constrain values for porosity and upwelling rate for a specific melting model (in this case the equilibrium porous flow model of Spiegelman and Elliott [1993]). In general, the extents of 230 Th- 238 U and 231 Pa- 235 U disequilibrium are controlled mainly by the melting rate, while the extent of 226 Ra- 230 Th disequilibrium is controlled mainly by the mantle porosity. The extent of 230 Th- 238 U disequilibrium is also strongly influenced by the depth of melting due to differences in relative partitioning of U and Th in the different mineral assembleges that are stable at different depths in the mantle. In the shallow mantle (<1 GPa), clinopyroxene is the main host phase for U and Th, and it has D Th > D U. However, at higher pressures where aluminous clinopyroxene or garnet are stable, the sense of U-Th fractionation changes, such that Th is more incompatible than U (i.e., D Th < D U ). This contrasts with the behavior of 231 Pa- 235 U disequilibria, as Pa is thought to be more incompatible than U throughout any mantle melting column. [29] It is important to be aware that the absolute values for mantle porosities and melting rates inferred from U series disequilibria observations are critically dependent both on the specific choice of assumed melting model as well as on the values chosen for the bulk partition coefficients for Th, U, Pa, and Ra between melt and mantle. The values for mineral/melt partition coefficients can vary as a function of temperature and pressure, as well as mineral composition, and so the bulk partition coefficient for an element will vary both with the major element composition of the mantle source and also with height within the melting column Diffusion [30] Most melting models assume that local chemical equilibrium is maintained between the solid phases and the melt, allowing trace element behavior to be described through the use of equilibrium partition coefficients. During melting the transfer of trace elements between solid phases and melt will ultimately be controlled by solid-state diffusion. If the solid-state diffusion rate for a particular element is very small, then there might not be sufficient time for chemical equilibrium to be established prior to the melt being extracted. Experimentally determined distribution coefficients for U and Th in clinopyroxene [Van Orman et al., 1998] indicate that diffusion of these large, highly charged ions is very sluggish (10 21 m 2 s 1 at 1200 C), such that chemical equilibrium might not be achieved during melting. This will increase their effective partition coefficients [e.g., Iwamori, 1993], such that they remain in the solid phase longer during the melting process than is predicted for simple chemical equilibrium. Several models have been developed to account for the potential effects of diffusion control on the U series nuclides during melting [Qin, 1992, 1993; Iwamori, 1994; Van Orman et al., 1998, 2002]. [31] Our ability to assess the potential for differences in solid-state diffusion rates to produce disequilibrium between the U series nuclides is rather limited by the general lack of experimentally determined distribution coefficients for U, Th, Ra, and Pa in most minerals. 7of43

8 Figure 5. U-Th equiline, or isochron, diagram illustrating the kind of age information that can be obtained for three different sets of samples. Samples that are older than 5 times the 230 Th half-life, i.e., 380 kyr, have ( 238 U) = ( 230 Th) and plot on the equiline with ( 230 Th/ 232 Th) ratios that reflect their U/Th ratios (group 1). Samples younger than 380 kyr can also plot on the equiline provided that the processes that formed these samples did not involve any fractionation of U and Th. If samples plot off the equiline, then this indicates an event that fractionated U and Th happened less than 380 kyr ago. If samples define an isochron (group 2), their age can be calculated as the slope equal to 1 e lt, where l is the 230 Th decay constant and t is the age (note that the slope of an isochron is constrained to values between 0 (t =0) and 1 (t = 1)). The assumptions are that at time t all these samples had the same Th isotope ratio, and different U/Th ratios, and that they would therefore have plotted on a horizontal line. If samples plot to the right of the equiline, they are said to have excess 238 U, and if they plot to the left, they have excess 230 Th. Samples with excess 238 U and excess 230 Th migrate vertically toward the equiline at a rate that depends on the half-life of 230 Th, and hence the slope of the isochron equals 1 e lt. The samples in group 3 have variable ( 230 Th/ 232 Th) and almost constant ( 238 U/ 230 Th), and so the time needed to move from the sample with the greatest to the least ( 230 Th/ 238 U) can also be calculated from the decay equation [e.g., Bourdon et al., 2003b]. Van Orman et al. [2001] have developed a crystal lattice elastic strain model that allows diffusion coefficients to be estimated based on the ionic radius and charge of the diffusing ion. This model predicts, for example, that Ra should have a diffusion coefficient in clinopyroxene (10 17 m 2 s 1 at 1200 C) that is 3 orders of magnitude greater than that of either Th or U. This marked difference has some intriguing implications for the development of 226 Ra- 230 Th disequilibrium in magmas [e.g., Saal and Van Orman, 2004; Feineman and DePaolo, 2004]. Saal and Van Orman [2004] have suggested that the 226 Ra excesses 8of43 observed in oceanic basalts might result from the diffusive interaction of magmas percolating through older cumulates containing clinopyroxene (and/or plagioclase) within the crust or at the crust-mantle transition zone. The faster diffusion rate of 226 Ra compared to 230 Th and 238 U suggests that the 226 Ra being produced in the crystal by decay of 230 Th could readily migrate out of the clinopyroxene crystal into the percolating melt, thus increasing the ( 226 Ra/ 230 Th) of the melt Extracting Time Information Graphically From U Series Disequilibria [32] For the 238 U- 230 Th system, time information is usually obtained with reference to the isochron, or equiline, diagram in which the parent ( 238 U) and the daughter ( 230 Th) nuclides are normalized to an element or a longlived isotope ( 232 Th) not involved in the decay scheme (Figure 5), with all isotope ratios expressed as activity ratios. The horizontal axis is equivalent to the elemental U/Th ratio, and the vertical axis is the Th isotopic composition, 230 Th/ 232 Th. All samples in secular equilibrium, where the activities of the parent and daughter nuclides are equal, i.e., ( 238 U)/( 232 Th) = ( 230 Th)/( 232 Th), will therefore plot on the 1:1 line (called the equiline) on this isochron diagram [e.g., Allègre and Condomines, 1976]. Secular equilibrium can be disturbed by any chemical process that changes the relative abundances of the parent and daughter nuclides, in this case U and Th. When that process took place is determined by the difference between the initial and the present-day Th isotope ratios. The initial Th isotope ratio ( 230 Th/ 232 Th) is that of the magma at the time of the event being studied, and it may be significantly different from the Th isotope ratios of magmas at the time of eruption due to subsequent radioactive decay or ingrowth of 230 Th. In contrast, other radiogenic isotope systems such as 87 Sr/ 86 Sr would remain essentially invariant over such short timescales. The great strength of these U series isotopes is that their isotope ratios change in response to radioactive decay within the timescales of magmatic processes, and age information is often obtained in one of two ways. [33] 1. If samples with different U/Th ratios plot on a positive linear array (Figure 5), this array can potentially represent an isochron, where the slope of the isochron will correspond to an age (slope equal to 1 e lt, where l is the 230 Th decay constant and t is the age, note that the slope of an isochron is constrained to values between 0 (t = 0) and 1(t = 1)). This age represents the time since the different U/Th ratios were established, and it will be geologically meaningful if the samples all had the same initial ( 230 Th/ 232 Th) and remained as closed systems for U and Th subsequently. The samples with different U/Th ratios may be either suites of comagmatic whole rock lava samples or mineral separates from an individual rock. Mixing processes can also produce samples that fall on a straight line, because the denominator ( 232 Th) is the same for both axes, but in this case the slope of the line will not give a meaningful age. Thus it is important to

9 establish that mixing between unrelated end-members has not occurred. One potential indicator of mixing is significant compositional variations of long-lived radiogenic isotopes such as 87 Sr/ 86 Sr that correlate with U/Th ratios. [34] 2. Some comagmatic igneous rocks have similar U/Th ratios but different Th isotope ratios, so that they plot in near-vertical arrays on the U-Th equiline diagram (Figure 5). In primitive rocks (i.e., that have experienced minimal fractional crystallization) such vertical arrays may reflect dynamic melting processes [e.g., McKenzie, 1985], but in rocks related by fractional crystallization they may reflect the time taken for magma differentiation to occur. If we assume that the group 3 samples shown on Figure 5 are related through different extents of fractional crystallization, then the parental magma would have the high initial ( 230 Th/ 232 Th) value, and the more evolved samples would have lower ( 230 Th/ 232 Th) values because of radioactive decay of 230 Th as time elapses during the process of magma differentiation. [35] An analogous diagram to the U-Th equiline diagram has been proposed for the 226 Ra- 230 Th system using Ba as the normalizing element as Ra does not have any long-lived isotopes [e.g., Volpe and Hammond, 1991; Reagan et al., 1992]. The Ba contents are essentially being used as a means to estimate the amount of Ra initially present in a melt or mineral. However, this diagram is more difficult to interpret because the underlying assumption that Ba is chemically similar to Ra is not strictly true [e.g., Cooper et al., 2001]. The 5% difference in ionic radius (Ra Å, Ba Å: eightfold coordination) means that they will not partition identically into different silicate minerals [e.g., Blundy and Wood, 2003], and the implications of this are discussed in more detail in section For the 231 Pa- 235 U system, Bourdon et al. [1999b] have suggested using Nb as the normalizing element for Pa. Despite their similar ionic charge and broadly similar geochemical behavior the difference in ionic radius (Pa Å, Nb Å: sixfold coordination) will also lead to different crystal partitioning behavior [Blundy and Wood, 2003]. 3. U SERIES DISEQUILIBRIA IN IGNEOUS ROCKS FROM DIFFERENT TECTONIC SETTINGS 3.1. Data Compilation [36] There are now sufficient high-quality data available in the literature to do an integrated review of global 230 Th- 238 U, 226 Ra- 230 Th, and 231 Pa- 235 U disequilibria in igneous rocks. This allows us to establish the observed range in values for these disequilibria in nature and also to highlight differences in disequilibria found in magmas generated in different tectonic environments. The results of this new compilation are summarized in Figures 6 and 7. We have restricted the data to mass spectrometer analyses, except for some of the 226 Ra data, and we have concentrated on analyses of samples that appear to have been little affected by alteration and shallow level processes. For simplicity, samples are divided into the three tectonic 9of43 environments that represent the principal melt generation regimes on the planet: (1) mid-ocean ridges, (2) withinplate settings (ocean island basalts and continental intraplate lavas), and (3) subduction zones. Figure 6 shows a series of histograms of ( 230 Th/ 238 U), ( 226 Ra/ 230 Th), and ( 231 Pa/ 235 U) activity ratios for volcanic rocks from different tectonic environments (data sources are given in Figure 6 caption). Figure 7a shows the U-Th equiline diagram (a plot of the activities of 230 Th versus 238 U, both normalized to the activity of the long-lived 232 Th isotope), which highlights the extent of element U-Th fractionations. Figures 7b and 7c show the covariations between ( 230 Th/ 238 U) disequilibria and ( 231 Pa/ 235 U) and ( 226 Ra/ 230 Th) disequilibria, respectively. [37] All data are plotted as initial ratios calculated back to the eruption age, except for the submarine MORB and back arc basin samples for which the eruption ages are generally unknown. The short t 1/2 of 226 Ra (1.6 kyr) means that it is reasonable to assume that any MORB sample with 226 Ra- 230 Th disequilibrium is younger than 8000 years and that any correction for posteruptive decay of 230 Th or 231 Pa would be negligible. Those MORB samples with 226 Ra- 230 Th disequilibrium are distinguished on the ( 230 Th/ 238 U) and ( 231 Pa/ 235 U) histograms in Figure 6. Furthermore, for the MORB samples plotted on Figures 6 and 7, the ( 226 Ra/ 230 Th) values only represent minimum estimates for the ( 226 Ra/ 230 Th) disequilibria on eruption because of the unknown extent of posteruptive decay (except for a few samples of known age collected by submersible from the East Pacific Rise [Sims et al., 2002a]). [38] The most extreme disequilibria values of any igneous rocks are found in historic natro-carbonatite samples from Oldoinyo Lengai volcano, East Africa [Williams et al., 1986; Pyle et al., 1991; Pickett and Murrell, 1997]. These rare carbonate-rich melts have ( 230 Th/ 238 U) of , ( 226 Ra/ 230 Th) of 40 80, and ( 231 Pa/ 235 U) of These data are consistent with models in which the natrocarbonatite magma forms by immiscibility from a silicate nephelinite magma, where Ra and U are strongly partitioned preferentially into the natro-carbonatite magma relative to Th and Pa. These lavas are further unique in being the only lava samples where 228 Ra- 232 Th disequilibrium has been convincingly measured, with ( 228 Ra/ 232 Th) of 27. Given the short half-life of 228 Ra (5.75 years), such large disequilibria indicate a very short time between Ra-Th fractionation and lava eruption: 7 18 years for the A.D. lavas [Williams et al., 1986] and years for the 1988 A.D. lavas [Pyle et al., 1991] The 230 Th- 238 U Disequilibria [39] Mid-ocean ridges and within-plate settings are both characterized predominantly by lavas with 230 Th excesses (Figure 6). Within-plate lavas show a greater range in ( 230 Th/ 238 U) than MORB lavas, with values generally between 1.0 and 1.6, although a few basalts from the western United States do have slight 238 U excesses [Reid and Ramos, 1996]. Most MORB lavas have ( 230 Th/ 238 U) values between 1.0 and 1.3. MORB samples with

10 Figure 6. Histograms of measured ( 230 Th/ 238 U), ( 226 Ra/ 230 Th), and ( 231 Pa/ 235 U) disequilibria in volcanic rocks from different tectonic settings: mid-ocean ridge basalts (MORB); within-plate, ocean island basalts and continental within-plate lavas; and ARCS, subduction zone lavas. Dotted line indicates secular equilibrium, i.e., samples with ( 230 Th) = ( 238 U), ( 226 Ra) = ( 230 Th), or ( 231 Pa) = ( 235 U). All data are obtained by thermal or plasma ionization mass spectrometry, except some ( 226 Ra/ 230 Th) data. Data sources are McDermott and Hawkesworth [1991], Goldstein et al. [1992], Williams et al. [1992], Cohen and O Nions [1993], Goldstein et al. [1993], Volpe and Goldstein [1993], Reagan et al. [1994], Asmerom and Edwards [1995], Lundstrom et al. [1995], Reid [1995], Bourdon et al. [1996a, 1996b], Cohen et al. [1996], Reid and Ramos [1996], Sigmarsson [1996], Turner et al. [1996], Elliott et al. [1997], Huang et al. [1997], Pickett and Murrell [1997], Regelous et al. [1997], Turner et al. [1997a, 1997b], Widom et al. [1997], Bourdon et al. [1998], Clark et al. [1998], Claude-Ivanaj et al. [1998], Lundstrom et al. [1998b], Sigmarsson et al. [1998a], Turner et al. [1998], Asmerom [1999], Chabaux et al. [1999], Lundstrom et al. [1999], Sims et al. [1999], Thomas et al. [1999], Turner et al. [1999], Vigier et al. [1999], Bourdon et al. [2000a, 2000b], Sturm et al. [2000], Turner et al. [2000a, 2000b], Claude-Ivanaj et al. [2001], Cooper et al. [2001], Peate et al. [2001a, 2001b], Pietruszka et al. [2001], Turner et al. [2001a], Turner and Foden [2001], Cooper et al. [2002], Sims et al. [2002a], Thomas et al. [2002], Cooper et al. [2003], Dosseto et al. [2003], Fretzdorff et al. [2003], George et al. [2003], Kokfelt et al. [2003], Lundstrom et al. [2003], Stracke et al. [2003], Turner et al. [2003b], Yokoyama et al. [2003], Zou et al. [2003], and Tepley et al. [2004]. Samples that appear to have been affected by alteration or shallow level contamination processes have been filtered out. All data are plotted as initial ratios calculated back to the eruption age, except for submarine mid-ocean ridge and back arc basin lavas for which eruption ages are generally unknown. For these settings, samples with measured 226 Ra- 230 Th disequilibrium are distinguished on the ( 230 Th/ 238 U) and ( 231 Pa/ 235 U) histograms as black. These samples are inferred to have been erupted within the last 8000 years, and thus there will have been minimal radioactive decay of 230 Th and 231 Pa since eruption. 10 of 43

11 Figure 7. Global variations in U series disequilibria in igneous rocks from different tectonic settings: (a) ( 238 U/ 232 Th) versus ( 230 Th/ 232 Th), (b) ( 230 Th/ 238 U) versus ( 231 Pa/ 235 U), and (c) ( 230 Th/ 238 U) versus ( 226 Ra/ 230 Th). Data sources are as for Figure 6. Black dotted lines on Figure 7a indicate different extents of 230 Th and 238 U excesses in percent: For example, 50% 230 Th excess (to the left of the equiline) indicates ( 230 Th/ 238 U) of 1.5, whereas 50% 238 U excess (to the right of the equiline) indicates ( 238 U/ 230 Th) of 1.5. Shaded dotted lines indicate secular equilibrium. 11 of 43 ( 230 Th/ 238 U) < 1 are rare, but a few examples are known from the Mid-Atlantic Ridge at N [Bourdon et al., 1996b], the Kolbeinsey Ridge [Sims et al., 2002b], and the Garrett Transform [Tepley et al., 2004]. In contrast, many subduction zone lavas have ( 230 Th/ 238 U) < 1 [e.g., Gill and Williams, 1990], with values as low as 0.4, although about a quarter of the samples compiled for Figures 6 and 7 have ( 230 Th/ 238 U) > 1 similar to MORB and within-plate lavas. The horizontal axis on Figure 7a is ( 238 U/ 232 Th), which is directly proportional to the elemental U/Th atomic ratio in the sample. Subduction zone lavas clearly show the widest range in ( 238 U/ 232 Th) from 0.4 to 3.4, with the highest values generally found in the more trace element-depleted (i.e., lower Th content) samples [e.g., McDermott and Hawkesworth, 1991]. MORB lavas have ( 238 U/ 232 Th) between 0.9 and 1.6. Within-plate lavas generally have lower ( 238 U/ 232 Th) values than MORB, between 0.4 and 1.2, with the lowest values ( ) found in highly enriched continental potassic lavas (Gaussberg, Antarctica [Williams et al., 1992]; Nyamuragira, East Africa [Pickett and Murrell, 1997]; Wudalianchi, China [Zou et al., 2003]; and Tibet [Cooper et al., 2002]) The 231 Pa- 235 U Disequilibria [40] With the exception of some depleted arc tholeiite lavas from the Tonga-Kermadec island arc [Bourdon et al., 1999b] and the Oldoinyo Lengai carbonatite lavas [Pickett and Murrell, 1997] all lava samples show 231 Pa excesses irrespective of their tectonic setting. MORB lavas generally have the greatest observed excesses, with ( 231 Pa/ 235 U) values from 1.9 to 4.0. Within-plate lavas, including both ocean island basalts and continental-intraplate lavas, usually have lower ( 231 Pa/ 235 U) values from 1.1 to 2.4. Subduction zone lavas have even lower ( 231 Pa/ 235 U) values from 0.8 to 1.7, with the lowest values found in lavas with the lowest ( 230 Th/ 238 U) values (Figure 7b). Kick em Jenny volcano in the Lesser Antilles is an exception, with the highest ( 231 Pa/ 235 U) value of 2.15 and yet one of the lowest ( 230 Th/ 238 U) values of 0.6 [Pickett and Murrell, 1997]. Subduction zone lavas and within-plate lavas form broad positive arrays on Figure 7b, with MORB lavas displaced to higher ( 231 Pa/ 235 U) The 226 Ra- 230 Th Disequilibria [41] In general, mafic igneous rocks with ( 226 Ra/ 230 Th) < 1 are very rare. Subduction zone lavas show the greatest range in ( 226 Ra/ 230 Th) values from 1 to almost 7, with the highest values mainly associated with the lowest ( 230 Th/ 238 U) values (Figure 7c). MORB lavas mostly have ( 226 Ra/ 230 Th) values between 1 and 3, with a few exceptional samples having values up to 4.2. Withinplate mafic lavas have a relatively restricted range in ( 226 Ra/ 230 Th), with values from 0.8 to 1.6, and it is only a few oceanic island basalts (OIB) samples from the Azores and Samoa [Claude-Ivanaj et al., 2001; Bourdon and Sims, 2003] that have ( 226 Ra/ 230 Th) < 1. Evolved lavas (e.g., phonolites from Mount Erebus, Antarctica [Reagan et al., 1992]; trachytes from Azores [Widom et al., 1992]; and

12 dacites from Santorini, Greece [Zellmer et al., 2000]) often have significant 226 Ra deficits, e.g., ( 226 Ra/ 230 Th) , as a result of extensive fractional crystallization of feldspar, which preferentially removes Ra from the melt relative to Th. [42] As most magmas have experienced some degree of differentiation, we look first at the timescales of magma differentiation and crystallization within the crust, which are potentially linked to the eruptive behavior of volcanoes. Next we address melt generation processes at mid-ocean ridges and ocean islands and then finish with the processes of fluid transfer from the downgoing slab and melt generation in the more complex setting of subduction-related magmatism. grains, and it is difficult to determine the textural relations of the individual crystals. Exceptions include accessory minerals with high-u and/or high-th contents such as zircon and allanite for which in situ 238 U- 230 Th ages can be determined by ion microprobe [e.g., Reid et al., 1997; Charlier et al., 2003; Vazquez and Reid, 2004]. The general principles of how U series disequilibria are used to establish timescales of magma chamber processes and to date young volcanic rocks are reviewed by Condomines et al. [2003] and Reid [2003]. Hawkesworth et al. [2000, 2004] provide broader overviews that integrate the U series results with other methods of determining the timescales of magmatic processes and then discuss the wider implications of such information. 4. TIMESCALES OF MAGMATIC DIFFERENTIATION AND CRYSTAL GROWTH 4.1. Background [43] Differentiation of mantle-derived magmas occurs principally by fractional crystallization, where early formed crystals are removed from the magma, thereby changing its composition [Bowen, 1928]. The time required for such crystal-liquid separation to occur will depend on the density contrast between the crystals and the liquid, the viscosity of the liquid, the size of the crystals, and the dynamics of the magma chamber system. The causes and timescales of crystallization vary: (1) Crystallization in response to cooling will depend on the size of the magma body, the magma replenishment rate, and the thermal structure of the crust [e.g., Annen and Sparks, 2002]. (2) Crystallization in response to degassing and decompression may be fast (days or weeks), and there is increasing evidence that it is too fast, and occurs too close to the time of crystallization of the host rock, for crystals formed in this way to be involved in fractional crystallization [e.g., Zellmer et al., 2003]. Thus magma differentiation processes may be largely thermally controlled. Progressive crystallization will result in the buildup of volatiles in the melt, and the sudden release of these dissolved volatiles may, in turn, trigger explosive eruptions. Thus there is considerable interest in the links between crystallization and the timing and style of volcanic eruptions and also in the primary controls on magma differentiation. [44] U series disequilibria are just one way to determine the timescales of magmatic processes; others include relative chronometers such as crystal size distributions and compositional profiles across crystals that may have been modified by diffusion [e.g., Cashman, 1990; Davidson and Tepley, 1997; Zellmer et al., 1999; Morgan et al., 2004]. The latter provide information on how long crystals have been at magmatic temperatures rather than when that might have been, and age information is obtained on individual crystals so that the age profiles of crystal populations can be determined for individual rocks. U series disequilibria yield absolute ages, but in most cases they are determined on mineral separates that may comprise several tens to hundreds of crystal 12 of Magmatic Differentiation Rates and Crustal Residence Times From Whole Rock Samples Magma Residence Times and Crustal Transit Times [45] Early attempts to infer magma residence times from U series nuclides developed models for steady state magma chamber systems, in which the eruptive volcanic output effectively balances the influx of new magma batches so that the chamber volume remains essentially constant. These models have been applied to persistently active centers such as Hawaii, Etna, Stromboli, and Reunion [Pyle, 1992; Albarède, 1993; Condomines, 1994; Hughes and Hawkesworth, 1999; Condomines et al., 2003]. The extent of isotopic disequilibrium in the erupted lavas from a steady state reservoir will be roughly constant and governed by the balance between higher values due to input of fresh magma to the chamber and lower values due to radioactive decay during residence of the mixed magma in the chamber prior to eruption. Provided that the initial degree of isotopic disequilibrium in the input magma is known, often from some flank eruptive unit that bypassed the chamber, an average magma residence time can be determined. Residence times of tens to hundreds of years are typically obtained, with 10% of the volume of the magma chamber inferred to be erupted annually. Such magma residence times may or may not be accompanied by significant magma differentiation, and so different approaches are required to distinguish rates of magma differentiation from the time taken for magma to traverse the crust. [46] If the degree of isotopic disequilibrium of a magma when it leaves the melting zone can be inferred, then its isotope composition at the time of eruption can be used to constrain how long it took to travel through the crust. The simplest case is for magmas erupted with significant 226 Ra excesses, as these imply crustal residence times of less than 8000 years, provided that the 226 Ra- 230 Th disequilibrium was established by mantle melting processes. The same arguments can be made for samples with 230 Th- 238 U disequilibria, but the longer half-life of 230 Th means that it constrains the crustal residence ages to be less than 380,000 years. This provides much less insight given the much shorter timescales for variability within the plumbing

13 Figure 8. Variations of ( 226 Ra/ 230 Th) with Th content (in ppm), an index of differentiation, in three suites of comagmatic lavas: (1) Asal rift, Afar [Vigier et al., 1999]; (2) Surtsey-Heimaey, Iceland [Sigmarsson, 1996]; (3) Miyakejima volcano, Japan [Yokoyama et al., 2003]. ( 230 Th/ 238 U) and ( 87 Sr/ 86 Sr) are both constant within the Afar and Iceland suites, indicating that the samples within each suite are simply related by different degrees of differentiation from similar parental magmas. The slope of the arrays will therefore depend on the rate of differentiation relative to the decay of 226 Ra and on the degree of Ra-Th fractionation during fractional crystallization. However, ( 230 Th/ 238 U) is not constant for the historic samples from Miyakejima but increases from to as ( 226 Ra/ 230 Th) decreases, and this is consistent with control by magma mixing rather than simple closed system differentiation. systems beneath many volcanoes inferred from stratigraphic records Estimates of Magma Differentiation Rates From ( 226 Ra/ 230 Th) [47] A widely used approach to determine rates of magma differentiation is to investigate how the magnitude of U series disequilibria, from which age information can be obtained, varies with indices of differentiation [e.g., Hawkesworth et al., 2000]. Figure 8 shows ( 226 Ra/ 230 Th) activity ratios plotted against Th contents as an index of differentiation for three magmatic suites (Asal rift, Surtsey- Heimaey, and Miyakejima). As Th is normally an incompatible element during crystallization, then Th contents will increase in the magma as differentiation proceeds and progressively more crystals are removed from the magma. The samples all have ( 226 Ra/ 230 Th) > 1 and will evolve to secular equilibrium, i.e., ( 226 Ra/ 230 Th) = 1, by radioactive decay. ( 226 Ra/ 230 Th), and hence the degree of 226 Ra- 230 Th isotope disequilibrium, decreases with increasing Th contents in these volcanic suites, suggesting that the more evolved magmas are older than the less evolved magmas. However, interpretations are complicated by the fact that Ra/Th ratios in magmas can potentially be changed by progressive removal of feldspar crystals during magma differentiation. The slope of the arrays will therefore depend both on the rate of differentiation relative to the decay of 226 Ra and on the relative bulk partition coefficients for Ra and Th during fractional crystallization. The latter will be governed by the modal proportion and composition of feldspar in the assemblage of crystals being removed from the magma. [48] 1. For the Asal rift lavas the 30% fractional crystallization of a plagioclase-dominated assemblage necessary to explain the major and trace element variations cannot explain all of the observed range in ( 226 Ra/ 230 Th), which therefore must also reflect different magma residence times, even though the samples were all erupted within a one week period in 1978 A.D. [Vigier et al., 1999]. Several magma batches must have been injected into one or more reservoirs beneath the rift at different times prior to eruption and then differentiated for between 1000 and 2000 years, depending on whether they evolved as independent closed systems or as an open system zoned magma chamber replenished by each new batch of primitive magma. [49] 2. In contrast, Sigmarsson [1996] argued that plagioclase fractionation alone could explain the differences in ( 226 Ra/ 230 Th) between the 1963 and 1967 A.D. Surtsey alkali basalts and the 1973 A.D. Heimaey hawaiitemugearite lavas in the Vestmannaeyjar volcanic system, southern Iceland, and variations of the short-lived 210 Pb nuclide (t 1/2 22 years) instead indicated rapid differentiation from alkali basalt to hawaiites and mugearites in only 10 years or so. He proposed a model in which a small volume of alkali basalt magma was injected into a deep (15 20 km) reservoir in relatively cold crust beneath Heimaey at the same time as similar magmas were erupted at Surtsey. The relatively cold crust led to rapid cooling and differentiation to hawaiites and mugearites, and the consequent buildup of volatiles led to the 1973 A.D. eruption. [50] 3. For the historic lavas from Miyakejima volcano, Japan [Yokoyama et al., 2003], the variations in ( 226 Ra/ 230 Th) correlate with ( 230 Th/ 238 U) and thus reflect magma mixing processes rather than timescales of differentiation Estimates of Magma Differentiation Rates From ( 238 U/ 230 Th) [51] The longer-lived 238 U- 230 Th system has been used to determine the timescales of closed system differentiation from mafic parental magmas to felsic evolved magmas for several different magmatic suites [e.g., Reagan et al., 1992; Widom et al., 1992; Bourdon et al., 1994; Bohrson and Reid, 1998; Hawkesworth et al., 2000]. In each case it is assumed that the 230 Th- 238 U disequilibrium measured in an erupted mafic sample is the same as that of the mafic magma prior to its differentiation into a more evolved magma. The difference in initial ( 230 Th/ 232 Th) between the mafic magma and a differentiated magma with similar U/Th in a closed system simply represents radioactive decay, thus allowing the duration of the magma differentiation to be calculated. For example, Bourdon et al. [1994] estimated a differentiation time of 100 kyr for the Laacher See phonolite (Germany) to evolve from a possible parental basanite magma (Figure 9), and Widom et al. [1992] 13 of 43

14 Figure 9. Example of different timescales of magma differentiation, the circa 13 ka Laacher See zoned phonolite eruption, Eifel, Germany [Bourdon et al., 1994]. These data are consistent with a two-stage model in which differentiation of a parental basanitic magma in a deep crustal reservoir takes 100 kyr and then the resulting mafic phonolite magma is emplaced into a shallow level magma chamber where it differentiates to more evolved compositions over a period of <20 kyr. The composition of the initial parental basanitic magma (open circle) was estimated from the ( 230 Th/ 232 Th) ratio of an olivine xenocryst in the phonolite and the ( 238 U/ 232 Th) of a local older basanite lava. The difference in ( 230 Th/ 232 Th) between the parental basanite and the mafic phonolite can be explained by radioactive decay during 100 kyr of magma differentiation, with no change in U/Th of the magma. Subsequent differentiation within the shallow magma chamber leads to increasing ( 238 U/ 232 Th) in phonolitic pumice glass samples (shaded circles) because of progressive removal of a mineral assemblage containing accessory phases, apatite, and sphene, with high Th and U contents and low U/Th ratios (see Figure 10b). These phonolitic glasses define an isochron with an age of 14.3 ± 6.5 ka that is within error of the known eruption age as determined from 238 U- 230 Th mineral isochrons, 14 C dating, and 40 Ar/ 39 Ar dating. Figure 9 is modified from Bourdon et al. [1994], with permission from Elsevier. inferred that it took 80 kyr to produce trachytic magmas from a parental alkali basalt at Agua de Pão, Sao Miguel, in the Azores. [52] The basanite to phonolite sequence on Tenerife (Canary Islands) can be modeled by 80% largely closed system fractional crystallization, but this differentiation occurred in distinct stages at different crustal levels: Mafic and intermediate lavas fractionated in a deep crustal reservoir at 6 9 kbar and then subsequently differentiated to evolved phonolitic compositions in a shallow chamber at kbar [Ablay et al., 1998]. Variations of initial ( 230 Th/ 238 U) values with Zr, an index of differentiation, suggest that the extensive fractionation in the deep crust took much longer than that at shallow levels [Thomas, 1998; Hawkesworth et al., 2000]. While the decrease in ( 230 Th/ 238 U) with differentiation indicates an overall differentiation time of 230 kyr, whole rock 226 Ra- 230 Th data suggest that the final feldspar-dominated differentiation to the most evolved phonolites took place within a thousand years or so before eruption. However, Lundstrom et al. [2003] measured significant 226 Ra excesses in three lavas from Tenerife that spanned a large compositional range from 9 to 2 wt % MgO. This would require a much shorter differentiation time than suggested by the 230 Th- 238 U system, perhaps implying that the 230 Th- 238 U disequilibria were affected by open system processes. One difficulty with this is that these lavas show a striking increase and then decrease in Ba with increasing differentiation, as reflected in Zr contents, and thus there seems to have been no simple mixing between high- and low-mgo magmas that might have a crustal origin. [53] Assimilation of basement rocks during differentiation in the crust is always a possibility in both continental and oceanic environments, and it will tend to give lower 238 U- 230 Th disequilibrium in the evolved magma as the assimilated material is likely to be old enough to be in secular equilibrium and lie on the equiline. Therefore, in such open systems the timescales for differentiation will be overestimated because the reduction in measured disequilibrium in the evolved magmas is due to the combination of radioactive decay and addition of material in secular equilibrium. Bohrson and Reid [1998] showed that rhyolites on Socorro Island, Mexico, are related to trachytes by fractional crystallization, but systematic variations in 87 Sr/ 86 Sr indicated that differentiation was accompanied by assimilation of seawater-altered silicic basement rocks, and thus the 40 kyr differentiation time indicated by the difference in ( 230 Th/ 232 Th) is a maximum estimate. If the assimilants are relatively young plutons or crystal mushes, perhaps related to an earlier phase of the same overall magmatic system, then this assimilation might be difficult to recognize with conventional geochemical monitors such as Sr and Pb isotopes because not enough time has elapsed for the isotope compositions to respond to changes in Rb/Sr, U/Pb, and Th/Pb ratios. However, if this crust has a markedly different U/Th value, it will develop a distinctive ( 230 Th/ 232 Th) value over just a few hundred thousand years. This cryptic assimilation has been proposed to explain the 232 Th- 230 Th- 238 U variations in some continental arc magmas from the Cascades and the Aleutians [George et al., 2003; Reagan et al., 2003]. [54] It is important to bear in mind that not all silicic magmas are formed by extensive fractional crystallization from a mafic parental magma. They can also be produced by partial melting of crustal material, and U series disequilibria can sometimes help to differentiate between these two mechanisms [e.g., Sigmarsson et al., 1991]. For example, the trachytes on Socorro Island (Mexico) cannot be related to contemporaneous basaltic magmas through closed system fractional crystallization as they, in fact, have greater 230 Th- 238 U disequilibrium at a given ( 238 U/ 232 Th) than the basalts. Bohrson and Reid [1998] proposed a model to generate the trachytes by partial melting of basaltic base- 14 of 43

15 ment that could explain both the U series disequilibria data as well as the major and trace element variations Rapid Magma Differentiation Timescales in Evolved Magmatic Systems [55] Shorter differentiation timescales are often associated with the development of zonation in silicic magma chambers. About 80% fractional crystallization is necessary to explain the compositional range of phonolitic glasses from the circa 13 ka Laacher See zoned eruption, and yet they define a 238 U- 230 Th isochron on Figure 9 with an age of 14.3 ± 6.5 ka. This age is within error of the eruption age and therefore requires that the differentiation within the phonolitic magma occurred rapidly and shortly before eruption. Detailed modeling gives a maximum timescale of kyr for the development of this compositional zonation, consistent with thermal and fluid dynamical models for the Laacher See magma chamber [Bourdon et al., 1994]. The extreme chemical zonation preserved in trachytic eruptive units from Agua de Pão in the Azores requires 70% fractional crystallization dominated by alkali feldspar, and Widom et al. [1992] estimated that this zonation took less than 4600 years to develop, based on ( 226 Ra/ 230 Th) data and stratigraphic constraints. [56] Rogers et al. [2004] analyzed 238 U- 230 Th- 226 Ra isotope ratios in a stratigraphic sequence of aphyric trachytic lava flows from Longonot volcano, Kenya, that were erupted in the last 6000 years. These lavas record progressive closed system fractional crystallization, dominated by alkali feldspar in which Ra and Ba are highly compatible. The ( 230 Th/ 238 U) disequilibria indicate that fractionation of U/Th took place within the last 10,000 years, as the samples all fall on a broadly horizontal trend (i.e., constant ( 230 Th/ 232 Th)) on an equiline diagram. Rogers et al. [2004] developed a model that combined radioactive decay and fractional crystallization to estimate rates of differentiation from the ( 226 Ra/ 230 Th) data. Differentiation rates for evolution from hawaiite to trachyte were about year 1, with faster rates for differentiation within the trachytes of approximately year 1. The ( 226 Ra/ 230 Th) data on whole rock and alkali feldspar separates indicated that phenocryst formation continued almost up to the time of eruption Timing of Crystallization From Mineral Separates Discordance Between Time of Crystal Growth and Time of Lava Eruption [57] The incentive behind early studies of 238 U- 230 Th disequilibria in mineral separates was as a means of determining the eruption age of young volcanic rocks using an isochron approach: Different mineral phases with a range of U/Th ratios, separated from an individual sample, should define an isochron age on the equiline diagram (Figure 5). Results were mixed, as 238 U- 230 Th isochron ages were often much older than independently known eruption ages. It was soon realized that such discordant data actually provided useful information about the timing of crystal growth, which did not necessarily coincide with the eruption age. [58] Analysis of fine-grained groundmass crystals that grew after eruption during rapid surface cooling of a lava flow should provide an unequivocal determination of the eruption age, and Peate et al. [1996] determined a 230 Th- 238 U isochron age of 156±29 ka on groundmass magnetite separates (grain size of 20 mm) and whole rock samples from a sequence of lavas from Albuquerque, New Mexico, that record a Quaternary geomagnetic excursion (Figure 10a). However, most studies have analyzed phenocryst phases that likely began to grow within the magma chamber or plumbing system prior to lava eruption. [59] For samples where 230 Th- 238 U dating of phenocrysts gives the same age as independently determined eruption ages, this implies that there was a very short time (relative to the 230 Th half-life) between crystal growth and eruption. For the Laacher See zoned eruption, three pumice samples from the crystal-rich mafic phonolitic units give internal 230 Th- 238 U mineral isochrons with a mean age of 13 ± 3 ka (Figure 10b), identical to the eruption age, indicating a maximum residence time for the crystals in the magma chamber of 1 2 kyr [Bourdon et al., 1994] Timescales of Crystal Residence and Growth From 226 Ra/ 230 Th Disequilibria [60] For those cases where 230 Th- 238 U mineral isochrons give ages within error of the known eruption age (e.g., Mount St. Helens andesite (Figure 11a) [Volpe and Hammond, 1991]), this implies that crystal growth must have happened less than a few thousand years before eruption. As these timescales are of a similar magnitude to the 226 Ra half-life (1599 years), the 226 Ra- 230 Th system can potentially provide better estimates of the crystal residence and growth times if the samples are young enough. Obtaining age information using the 226 Ra- 230 Th system on separated components of a rock is not as easy as for the 230 Th- 238 U system, because interpretation of data on a Ba-normalized 226 Ra- 230 Th isochron plot, as alluded to earlier, is not straightforward. Volpe and Hammond [1991] showed that for andesite samples from Mount St. Helens the groundmass and whole rock analyses always plotted above a reference line through the mineral phases (Figure 11b), which suggested that some open system behavior had taken place. However, the main assumption behind the construction of a diagram like Figure 11b, that Ba and Ra have identical geochemical behavior, is probably incorrect [e.g., Cooper et al., 2001; Blundy and Wood, 2003]. [61] Cooper et al. [2001] devised an elegant method to determine crystal residence ages from 226 Ra- 230 Th-Ba data that takes into account the different mineral/melt partitioning of Ba and Ra. They used an elastic strain partitioning model [Blundy and Wood, 2003] to estimate partition coefficients for Ba and Ra in different mineral phases under conditions relevant to the magma under investigation. Differences in partition coefficients mean that the mineral phases will have different initial ( 226 Ra)/Ba ratios. They plotted ( 226 Ra)/Ba versus time evolution curves for the groundmass and for melts in equilibrium with each analyzed crystal phase (using the estimated partition 15 of 43

16 coefficients). If the curves intersect, then this defines the time when the minerals could have crystallized from the host magma. It is important to correct for the effects of impurities in the analyzed bulk mineral separates (adhering glass and melt inclusions) by mass balance calculations using the difference between Th and Ba contents in the bulk mineral separates and those in the pure minerals as determined by in situ ion probe measurements. For the Mount St. Helens andesite example shown in Figure 11c, this approach gives a residence time of ka for plagioclase and ka for pyroxene prior to eruption [Cooper and Reid, 2003]. [62] Cooper et al. [2001] used this technique to show that 226 Ra- 230 Th-Ba data from the 1955 A.D. east rift eruption at Kilauea volcano (Hawaii) are consistent with cocrystallization of plagioclase and pyroxene from a melt represented by the groundmass with a mean age of 1000 ± 400 years. Another example is provided by Zellmer et al. [2000], who studied the Kameni dacites (46 A.D. to 1950 A.D.) on Santorini (Aegean arc) and determined an essentially zero age 230 Th- 238 U mineral isochron (18 ± 18 ka) for a sample of the 1950 A.D. lava. Modeling of 226 Ra- 230 Th-Ba data in whole rock Kameni dacites indicated that plagioclase fractionation occurred less than 1000 years prior to each individual eruption, which is consistent with the short (maximum of a few hundred years) plagioclase crystal residence times calculated from Sr diffusion profiles [Zellmer et al., 1999] and crystal size distribution studies [Higgins, 1996] Interpretations of Discordant Ages [63] There are numerous examples in the literature where 230 Th- 238 U mineral isochrons give well-defined ages that are older than the known eruption age: (1) Volpe [1992] obtained ages of 27 ± 18 ka and 28 ± 10 ka from two Hotlum andesite samples from Mount Shasta (Cascades, United States) that were erupted less than 4000 years ago; (2) Volpe and Hammond [1991] measured old ages of 34 ± 16 ka in a basalt lava and 27 ± 12 ka in an andesite lava from the Castle Creek eruptive period ( ka) at Mount St. Helens (Cascades, United States); and (3) Heath et al. [1998] analyzed four separate lava samples younger than 4000 years old from Soufriere volcano (St. Vincent, Lesser Antilles) that all yielded old mineral isochron ages of between 46 and 77 ka. [64] Minerals with old ages may be xenocrysts incorporated by the magma during ascent to the surface and as such 16 of 43 Figure 10. (a) The 230 Th- 238 U isochron defined by whole rock and groundmass magnetite (20 mm crystals) from a suite of related basaltic lavas from the Albuquerque volcanic field, New Mexico, United States, that records an eruption age of 156 ± 29 ka [Peate et al., 1996]. (b) Internal mineral isochrons for pumice samples from the main part of the zoned Laacher See phonolite eruption (solid circles) indicating a crystallization age of 13 ± 3 ka, indistinguishable from the eruption age and the isochron defined by glass samples (see Figure 9). Separates from the most evolved, crystal-poor samples give an older and less precise age of 30 ka, suggesting incorporation of crystals from older cumulates [Bourdon et al., 1994]. (c) Pyroxene and magnetite separates from an 190 ka basanite lava flow (Fornicher Kopf, Eifel, Germany) lying on the equiline, whereas the groundmass has 5% excess 230 Th. Tie lines between either mineral phase and the groundmass have slopes greater than the equiline and thus give indeterminate ages. This implies that these mineral phases are old (>380 ka) and not in equilibrium with the host groundmass and are likely to be xenocrysts perhaps inherited from an earlier magmatic episode [Peate et al., 2001c]. Abbreviations are gnd, groundmass; gl, glass; wr, whole rock; mt, magnetite; pyx, pyroxene; ap, apatite; sp, sphene; am, amphibole; and ha, haüyne.

17 Figure 11. Crystallization ages of a Mount St. Helens andesite (0.4 ka, Cascades, United States). (a) The 230 Th- 238 U equiline diagram (modified from Volpe and Hammond [1991], with permission from Elsevier). (b) Banormalized 226 Ra- 230 Th isochron diagram (modified from Volpe and Hammond [1991], with permission from Elsevier). The whole rock does not lie on the pl-pyx tie line, which suggested open system behavior to Volpe and Hammond [1991], but it is more likely a consequence of the different partitioning of Ra and Ba between these minerals and melt [Cooper and Reid, 2003]. (c) The 226 Ra/Ba time evolution diagram [Cooper and Reid, 2003], showing that the plagioclase crystallized from the host melt at 3 ka (shaded area). However, the timing of pyroxene growth is less well constrained ( ka) (modified from Cooper and Reid [2003], with permission from Elsevier). provide little insight into magma residence times or the timescales of differentiation. Alternatively, they may be minerals that crystallized from earlier magma batches and so constrain the length of time over which a particular magma system has been active. It is possible that the old ages reflect a significant residence time of the crystals in the magma prior to eruption. However, at least for the examples given above, closed system residence times >10 kyr can be ruled out because the magmas all show 226 Ra- 230 Th disequilibrium that would have returned to secular equilibrium after only 8 kyr, and thus some type of open system behavior is required. [65] Analyses of different mineral phases do not always lie on simple linear trends on the 230 Th- 238 U isochron equiline diagram, suggesting that each phase might have crystallized at different times. For example, Peate et al. [2001c] analyzed several different phases separated from two samples of an 190 ka basanitic lava from the Eifel volcanic field, Germany. The groundmass samples have ( 230 Th/ 238 U) of ± 0.015, whereas both the magnetite and pyroxene separates lie within error of the equiline (i.e., 230 Th in equilibrium with 238 U) at lower ( 238 U/ 232 Th), so that a tie line between either phase and the groundmass has a slope greater than the equiline and thus is not a valid isochron (Figure 10c). This indicates that the magnetite and pyroxene crystals are older than the groundmass and are not in chemical equilibrium with it and thus represent xenocrysts rather than phenocrysts. Returning to the example of the zoned Laacher See eruption [Bourdon et al., 1994], in contrast to the crystal-rich mafic phonolitic units that give internal 230 Th- 238 U mineral isochrons identical to the 13 ka eruption age, glass and mineral separates from the crystalpoor felsic phonolite units show more scatter, with a poorly defined isochron age of 30 ka (Figure 10b). This apparent older age is interpreted as being due to inheritance of crystals from an earlier episode of differentiation within the magma chamber. [66] A good example of nonconcordant 230 Th- 238 U and 226 Ra- 230 Th age information and how this can give fresh insights into the complex history of crystals is provided by studies of historic lavas and cogenetic cumulate xenoliths from Soufriere volcano (St. Vincent, Lesser Antilles [Heath et al., 1998; Turner et al., 2003c]). Data on mineral separates from a 1979 A.D. lava and from a cumulate xenolith lie on a common array on the 238 U- 230 Th equiline diagram (Figure 12a) that corresponds to an age of 47 ± 10 ka, and yet all phases also preserve 226 Ra- 230 Th disequilibria. However, using the approach of Cooper et al. [2001], 226 Ra/Ba evolution curves for all minerals from the lava and cumulate plot below that for the host lava (Figure 12b), which demonstrates that none of the minerals grew in equilibrium with the present groundmass. The crystal size distribution of plagioclase grains for the host lava [Turner et al., 2003c] indicates that some process has caused a relative increase in the number of large crystals and suggests a mixture between two crystal populations that have grown at different rates in different environments. These data together suggest that the minerals are 17 of 43

18 Figure 12. (a) The 230 Th- 238 U equiline diagram for separated phases from a lava (STV 354) and a cumulate xenolith (WI 1A 18) from the 1979 A.D. eruption of Soufriere volcano (St. Vincent, Lesser Antilles) that define a common trend corresponding to an age of 47 (+12, 9) ka [Heath et al., 1998; Turner et al., 2003c]. (b) The 226 Ra/Ba versus time evolution diagram for STV 354 whole rock and calculated liquids in equilibrium with mineral separates from STV 354 (solid curves) and WI 1A 18 (shaded curves) [Turner et al., 2003c]. Abbreviations are cpx, clinopyroxene; pl, plagioclase; gm, groundmass; and ol, olivine. Figure 12 is modified from Turner et al. [2003c], with permission from Elsevier. zoned in terms of composition and age, and the 226 Ra excesses are due to young overgrowths on old cumulate crystals that were entrained by recent injection of a new magma batch. The 230 Th- 238 U apparent age indicates a significant residence time for crystals in the cumulates beneath the Soufriere volcano. [67] Similar examples of discrepancies between 230 Th- 238 U and 226 Ra- 230 Th mineral ages have been explained either by young growth of crystal rims on older crystal cores (e.g., Tonga arc lavas [Turner et al., 2003c]) or by mixtures of young and old crystals (e.g., Castle Creek lavas at Mount St. Helens [Cooper and Reid, 2003]). Reagan et al. [1992] presented data showing nonconcordant ages for the 226 Ra- 230 Th and 228 Ra- 232 Th systems for anorthoclase crystal growth in phonolitic magmas erupted at Mount Erebus (Antarctica) that can also be explained by younger overgrowths on crystals. The 226 Ra- 230 Th-Ba data indicated an average age of 2400 years for anorthoclase growth, which is inferred to have occurred within a shallow reservoir system as melt inclusions within the crystals all have very low volatile contents. Disequilibrium between 228 Ra and 232 Th in the anorthoclase separates was attributed to growth of thin rims on the crystals within 30 years of eruption, which was probably caused by cooling associated with intrusion into the lava lake present at the summit Effects of Crystallization of Th- and/or U-Rich Accessory Phases on 238 U- 230 Th Disequilibria [68] The crystallization of accessory phases with very high U and/or Th contents (e.g., zircon, apatite, sphene, chevkinite, allanite, and monazite) and low modal abundance, particularly in felsic magmas, provides an additional means of gaining fresh insights into the evolution of magmatic systems. Most major phenocryst phases often contain inclusions of these accessory phases that will dominate the U and Th budget of the analyzed mineral separate. Their presence can sometimes lead to analytical problems as many accessory phases are difficult to dissolve completely with conventional acid digestion techniques [Condomines et al., 2003]. A mineral isochron determined on the major phases may, in fact, be a mixing line controlled by varying proportions of one or more accessory phase(s), so that the isochron age is only giving the crystallization age of the accessory phase(s). For example, the 238 U- 230 Th mineral isochron of 25 ± 10 ka for a comendite lava from Olkaria, Kenya, is apparently governed by inclusions of chevkinite, a phase with very high Th contents (1.5 wt % ThO 2 ) and very low ( 238 U/ 232 Th) (0.06) [Heumann and Davis, 2002]. [69] It is more difficult to establish how the timing of crystallization of such accessory phases fits into the crystallization history of the major fractionating phases and hence with the magma differentiation process. Growth of these phases can occur at different times because they are not controlled by phase equilibria, as is the case for the major phenocryst phases, but by saturation in the melt of a specific essential structural component (e.g., Zr, zircon; P, apatite; and Th, monazite), which depends on melt composition and temperature. Accessory phases can also crystallize as a result of local saturation adjacent to a growing crystal of one of the major phenocryst phases [e.g., Bacon, 1989], and thus its growth may be contemporaneous with that of the major phenocryst phase rather than reflecting the timing of saturation in the overall magma reservoir. [70] Accessory phases can also show evidence for multiple growth stages, as demonstrated by Charlier and Zellmer [2000], who analyzed three different size fractions of zircons from the 26.5 ka Oruanui eruption in the Taupo 18 of 43

19 Figure 13. The 230 Th- 238 U equiline diagram showing different zircon-whole rock model ages for different zirconsize fractions from the 26.5 ka Oruanui rhyolite eruption from the Taupo volcanic zone, New Zealand. Ages are calculated as two-point model ages relative to the whole rock analysis. Data are age-corrected back to the time of eruption, and so all model ages indicate time prior to the eruption. The 230 Th- 238 U data and the size distribution of zircons can be used to constrain a mixing model between old cores and younger rims. Figure 13 is modified from Charlier and Zellmer [2000], with permission from Elsevier. volcanic zone (New Zealand). They found that the zirconwhole rock 238 U- 230 Th data gave preeruption ages that were different in each size fraction, with the older ages given by the larger zircons (<63 mm, 5.5 ± 0.8 ka; mm, 9.7 ± 1.7 ka; and mm, 12.3 ± 0.8 ka) (see Figure 13). These data can be explained by a continuous zircon growth model over a period of 90 kyr, but cathodoluminescence images showed that the zircon crystals typically have euhedral cores and rims, and the data can be modeled instead by mixing an older population of zircons (27 kyr at the time of eruption) with a young zircon rim overgrowth that crystallized shortly before eruption Evolution of Major Rhyolitic Systems From 238 U- 230 Th Disequilibria in Accessory Phases [71] A key issue in understanding the volcanic hazards associated with large rhyolitic magma systems is the length of time that magma resides at shallow levels in the buildup to major eruptions (>100 km 3 magma). Several studies have demonstrated the potential of U series disequilibria in accessory phases to provide important time constraints in major rhyolitic systems: Long Valley caldera, California, United States [Reid et al., 1997; Heumann et al., 2002]; Toba caldera, Indonesia [Vazquez and Reid, 2004]; Yellowstone caldera, Wyoming, United States [Vazquez and Reid, 2002]; and Taupo volcanic zone, New Zealand [Charlier and Zellmer, 2000; Charlier et al., 2003]. [72] Postcaldera rhyolites at Long Valley have eruption ages of 110 ka based on 40 Ar/ 39 Ar dating of sanidine crystals. However, Heumann et al. [2002] showed that glasses from these units define a Rb-Sr isochron age of 257 ± 39 ka (Figure 14a) that is interpreted as reflecting a feldspar fractionation event 150 kyr before eruption. (Note that extensive feldspar fractionation can produce very large Rb/Sr ratios in high-silica magmas that are extreme enough to produce resolvable variations in 87 Sr/ 86 Sr over a period of a thousand years or so from the decay of 87 Rb, despite its long half-life of years.) The 238 U- 230 Th data on mineral separates from one of the postcaldera units (Deer Mountain) define a linear trend corresponding to an age of 236 ± 1 ka (Figure 14b). As all the major phases contain inclusions of zircon (high U/Th) and allanite (low U/Th), this array represents a mixing line between zircon and allanite populations. It only has age significance if zircon and allanite crystallized contemporaneously, and the observation that the glass analysis plots below the array indicates that this cannot be true. Instead, the data are consistent with two separate and short (1 kyr) episodes of accessory phase crystallization, with zircon growth at 250 ± 3 ka, followed by allanite growth at 187 ± 9 ka. [73] Reid et al. [1997] developed an ion probe technique to measure in situ 230 Th- 238 U model ages of zircons in rhyolites from Long Valley. Although individual model ages are not very precise, this method allows many crystals to be dated and thus provides an age spectrum of a crystal population. For the Deer Mountain rhyolite, except for a few young zircon ages close to that of the eruption, most zircon ages were within error of a weighted mean age of 226 ± 18 ka (1s), consistent with the thermal ionization mass spectrometry data of Heumann et al. [2002] (Figure 14c). A sample from a much younger rhyolite (Inyo dome, 0.6 ka) from the same region had a similar zircon age population with a weighted mean age of 229 ± 22 ka (1s). This is consistent with other evidence that suggests that this eruption tapped the same magma body as the Deer Mountain rhyolite 100 kyr earlier. Zircon saturation temperatures indicate that the magmas cooled to below 815 C more than 200 kyr ago, but there is some uncertainty over whether the magmas solidified then and were subsequently remelted or whether they have been crystal mushes since these zircons crystallized. 19 of Overview of Differentiation and Crystallization Timescales From U Series Disequilibria Links Between U Series Residence Times and Magma Composition and Eruption Rate? [74] Hawkesworth et al. [2000] investigated how the ages of crystal separates in recent volcanic rocks at the time of eruption varied with an index of differentiation (Figure 15a). Molar Si + Al was chosen as the index of differentiation because these are framework-forming elements in the melt and there is a marked increase in the viscosity of liquids at Si + Al = 66 [e.g., McBirney, 1984]. It was striking that most of the old preeruption mineral ages are from rocks with relatively high Si + Al values. Thus the simplest observation is the intuitive one that the likelihood of erupted magmas containing old crystals is much greater in the more evolved and hence

20 higher-viscosity liquids. Moreover, simple cooling and crystal-settling models imply that primitive basalts affected solely by cooling in a chamber will contain a small proportion of near-liquidus crystals, whereas more evolved magmas will have an increasing proportion of older crystals inherited from earlier stages of the magma s history. [75] While the significance of old preeruption ages may be different in different centers, it is striking that there is a broad negative array between those ages and the average eruption rates (Figure 15b). A marked exception is the Auckland volcanic field where very small amounts of magma have been rapidly erupted from mantle depths, which is where any differentiation took place [Rout et al., 1993]. Nonetheless, in general, magmas with high eruption rates tend to spend less time in the crust and vice versa, and magmas that spend longer in the crust also tend to be more evolved. Mid-ocean ridge basalts may take weeks to a few thousand years to traverse the crust [Sigmarsson, 1996], and it appears that volcanoes that erupt semicontinuously have relatively small volumes of evolved magma types. In contrast, evolved magma types tend to be more common at volcanoes where magmas spend longer in the crust, the eruptions are more episodic, and the volcanic center as been active for 10 5 years. In western North America, for example, there is typically kyr of magmatic activity at any center before caldera-related rhyolites are erupted [Lipman, 1984; Reagan et al., 2003]. Moreover, models of the thermal evolution of magmatic provinces suggests that significant amounts of crustal melting will also take place after years at reasonable emplacement rates of mafic magma [Annen and Sparks, 2002]. Another inference of the broad array in Figure 15b is that it constrains the rate of eruption, and hence presumably the melt generation rate, of mafic magmas that traverse the crust very rapidly, which may, in turn, inform models of magma transport through the crust. Steady state models for magmatic systems predict a negative relation between magma residence times in a magma chamber (or in the crust) and either the influx or outflux of magma, linked by the volume of magma in the chamber (or in the crust) [e.g., Pyle, 1992]. Some of the time estimates in Figure 15b are made on the basis of steady state modeling from Pyle [1992], but for the others, there need be no simple link between the estimated magma timescales and magma residence times. Nonetheless, it is interesting that steady state modeling of the negative array in Figure 15b would suggest 20 of 43 Figure 14. Long Valley magmatic system, California, United States. (a) Rb-Sr isochron diagram for postcaldera glasses (eruption age 110 ka, from 40 Ar/ 39 Ar sanidine ages) of an age 150 kyr older than the eruption age (modified from Heumann et al. [2002], with permission from Elsevier). (b) The 230 Th- 238 U isochron diagram for separated phases from the 110 ka postcaldera Deer Mountain rhyolite [Heumann et al., 2002]. Amphibolebiotite-sanidine-zircon define a linear array corresponding to an age of 236 ± 1 ka, but the glass and whole rock samples plot below this array, close to the equiline. The inset diagram shows a reduced scale to include the zircon point with its high U/Th ratio. Figure 14b is modified from Heumann et al. [2002], with permission from Elsevier. (c) ( 238 U/ 232 Th) versus model 230 Th- 238 U ages obtained by in situ ion probe measurements of zircons extracted from a sample of the Deer Mountain rhyolite [Reid et al., 1997]. Solid circles (with 1s error bars) show zircon data used to calculate a weighted mean age of 226±18 ka (shown by shaded rectangle). Three zircons (shaded circles) have model ages similar to the eruption age. Figure 14c is modified from Reid et al. [1997], with permission from Elsevier.

21 magma volumes in the crust of km 3, consistent with other estimates Links Between U Series Disequilibria Ages and Earlier Volcanic Episodes? [76] An important question is whether the U series disequilibria data can aid our understanding of the behavior of volcanoes. Several studies have highlighted intriguing coincidences between time information obtained from U series disequilibria data (e.g., preeruptive crystal ages and whole rock magma differentiation ages) and specific episodes in the history of a volcanic system known from surface stratigraphic studies (e.g., earlier eruptive episodes and times when major changes in eruptive behavior occurred). These include the following: (1) The Castle Creek eruptive period 2000 years ago coincided with a major shift in the eruptive behavior of the Mount St. Helens volcano, from dominantly explosive dacitic volcanism to dominantly effusive andesitic and basaltic lavas, and Cooper and Reid [2003] were intrigued that many of the plagioclase crystals entrained by younger lavas were apparently formed during this interval, based on U series disequilibria data. (2) Hawkesworth et al. [2000] noted that the 230 kyr differentiation timescale from basanite to phonolite magma at Tenerife, inferred from variations in ( 230 Th/ 238 U) in whole rock samples, is broadly similar to the periodicity of the different eruptive cycles there over the last 1.6 Myr. (3) Turner et al. [2003b] compared 226 Ra/ 230 Th data from two adjacent and compositionally similar volcanoes in Indonesia in the Sunda arc. Sangeang Api lavas and cumulates indicated residence times of 2000 years in a relatively small 10 km 3 magma chamber, whereas a lava sample from the cataclysmic 1815 A.D. eruption (100 km 3 ) from Tambora indicated a longer crustal residence time of 5000 years that is identical to the time elapsed since the preceding major eruption at Tambora. Such speculative links to specific events in a volcano s history clearly merit further investigation at other volcanoes with well-constrained stratigraphic records Summary [77] It is clear that U series disequilibria data on melt and crystal phases of igneous rocks can provide valuable temporal information on different stages in the magmatic history of a particular volcanic center. The groundmass of different related samples can be used to estimate the timescale for a parental magma to differentiate to more evolved compositions, and where the effects of magma mixing and crustal assimilation appear to have been minimal, estimates range from 1000 years [e.g., Vigier et al., 1999] to 230,000 years [e.g., Hawkesworth et al., 2000]. In general, the ages indicate slower differentiation rates at the higher temperatures in the deep crust and more rapid differentiation at cooler shallow crustal levels. Crystals found in igneous rocks can provide a complex record of magmatic processes, preserved as both textural and compositional variations, and yet it is often not clear how these variations are related to the evolutionary history of the host 21 of 43 Figure 15. (a) Preeruption ages of various recent magmatic suites as derived from U series disequilibria versus an index of differentiation, in this case molar Si + Al of the bulk rock [Hawkesworth et al., 2000]. Most old 238 U- 230 Th mineral isochrons have been obtained from rocks with relatively high Si + Al contents and higher viscosities. By contrast, studies of bulk rock variations of 238 U- 230 Th in suites of comagmatic but compositionally diverse samples suggest that differentiation from mafic parental magmas to more intermediate compositions can take kyr. Similar studies, principally using whole rock 226 Ra- 230 Th data, suggest that the inferred differentiation times for more evolved magmas are significantly shorter, perhaps tens to hundreds of years. Overall, these observations are consistent with slower rates of differentiation in more mafic magmas at greater depths and much faster rates of differentiation in more evolved magmas at shallower depths. (b) Variation of eruption rates [from Crisp, 1984] with preeruption ages estimated for a wide range of different magmatic suites (see Hawkesworth et al. [2004] for details). Figure 15 is taken from Hawkesworth et al. [2004], with permission from Elsevier.

22 magma. Petrographic observations often show clear evidence for the complex multistage growth history of crystals, with overgrowths of young rims on older inherited cores or xenocrysts perhaps related to an earlier stage in the magmatic system and patterns of zoning and resorption indicating fluctuating pressure-temperature-f O2 conditions during crystal growth over a time period that is long relative to the nuclide half-life. It is important to realize that U series analysis of mineral separates will tend to average such effects, and in addition, the U, Th, and Ra budgets of bulk mineral separates might be dominated by inclusions of trapped melt or accessory phases. The recent study by Vazquez and Reid [2004], who measured trace element and 230 Th- 238 U variations in situ in zoned allanite crystals from the voluminous 75 ka Toba Tuff, represents an important advance because the trace element variations can be linked to specific stages in the differentiation of the magma at the time given by 230 Th- 238 U data. In conclusion, U series disequilibria data have provided important constraints on the timing of crystal growth and residence time in magmas. 5. TIMESCALES AND PROCESSES OF MELT GENERATION AND TRANSPORT 5.1. Background to Melt Generation Processes in Upwelling Mantle [78] Melt generation processes are best constrained under mid-ocean ridges, where partial melting occurs in response to decompression in the upper mantle as the plates move away from the spreading center and new crust is generated in their place. The application of trace element data greatly increased the sensitivity of partial melting and fractional crystallization models [e.g., Gast, 1968], and initially at least, these models assumed bulk equilibrium between the melt and the residual matrix until the magma was extracted (i.e., batch melting). U and Th are highly incompatible elements in the upper mantle, and so U/Th ratios in the melt should be similar to those in their source [O Nions and McKenzie, 1993; Elliott, 1997] at the large degrees of partial melting (10%) inferred for MORB [e.g., Klein and Langmuir, 1987; McKenzie and Bickle, 1988]. Thus the expectation was that 238 U and 230 Th should be close to secular equilibrium in MORB magmas. [79] However, the demonstration of significant 238 U- 230 Th isotope disequilibria in MORB [Condomines et al., 1981; Newman et al., 1983] highlighted that such models might be overly simplistic [e.g., McKenzie, 1985]. Element fractionation is only efficient when the degree of melting is similar to or less than the bulk distribution coefficient between melt and mantle minerals, which for U and Th is <0.01. Thus, to obtain 238 U- 230 Th disequilibria, the degrees of melting must be small (<1%), and so dynamic melting models were developed in which small degree melts migrated through the melt zone and were then mixed together with larger degree melts before melt extraction [e.g., Williams and Gill, 1989]. In such models the degree of U series isotope disequilibria depends both on the porosity in the melt generation zone, and hence the fraction of melt in equilibrium with the solid, and on the melting rate relative to the half-life of the daughter product. Thus the duration of the melting process is important, and as melting rate depends on potential temperature and the upwelling rate, there is a direct link to mantle dynamics. In the relatively simple tectonic setting of mid-ocean ridges it is possible to estimate the solid mantle upwelling rate directly from the plate separation or spreading rate. [80] In the dynamic melting model [McKenzie, 1985], as melt is generated it remains in equilibrium with the residual mantle until a threshold porosity is reached, at which point any melt in excess of this porosity value is instantaneously extracted to the surface in chemically isolated channels without further interaction with residual mantle (Figure 16a). This dynamic melting model may now be usefully regarded as one end-member of a number of open system ingrowth models [Williams and Gill, 1989; Spiegelman and Elliott, 1993] in which excess daughter products over that in the source are produced by ingrowth from the parent because of their different residence times within the melting column. For example, peridotite residual after extraction of a small degree melt will have an elevated U/Th ratio and plot to the right of the equiline, and subsequent radioactive decay will result in higher 230 Th/ 232 Th ratios that may then be mobilized by further melting (Figure 16c). For equilibrium dynamic melting, significant radioactive disequilibrium can therefore be generated by ingrowth provided that the melting rate is slow relative to the decay rate of the short-lived daughter isotope (e.g., slow upwelling mantle or a long melting column so that the duration of melting is significant). For these ingrowth models, ( 226 Ra/ 230 Th) is controlled mostly by the porosity of the melt region (which controls the velocity of the melt relative to the residual mantle), whereas ( 230 Th/ 238 U) and ( 231 Pa/ 235 U) are controlled more by the melting rate (which is related to the residual mantle upwelling velocity). In practice, the inferred porosities are small relative to those estimated from modeling of trace element abundances in abyssal peridotites and melt inclusions [e.g., Johnson et al., 1990; Slater et al., 2001]. One explanation may be that the melt is in equilibrium with the crystal surfaces, and as crystal-liquid interactions are sensitive to the diffusion of elements within the crystals [Qin, 1992, 1993], this will change the effective partition coefficients and allow larger inferred porosities. [81] The other end-member ingrowth model involves melt generation and continuous chemical equilibrium between melt and residual mantle during melt percolation and transport (e.g., equilibrium or chromatographic or reactive porous flow model (Figure 16b) [Spiegelman and Elliott, 1993]). This equilibrium porous flow affects the relative transport velocities and hence the overall residence times of the parent and daughter nuclides in the melt column. In single porosity models the melt velocity 22 of 43

23 depends on the solid upwelling velocity, the degree of melting, and the porosity at the same height. The porosities are still small (0.2% for grain sizes of 3 mm), but these can produce significant U series disequilibrium provided that the parent spends longer in the system than the daughter [Spiegelman and Kelemen, 2003]. Spiegelman and Elliott [1993] also discussed other possible mechanisms that might produce disequilibria as a result of differential transport velocities of parent and daughter nuclides, such as different degrees of surface adsorption. Hiraga et al. [2004] presented a thermodynamic model for equilibrium grain boundary segregation that predicts that grain boundaries should preferentially host incompatible elements such as U, Th, and Ra, thus providing another potential mechanism for their selective enrichment in melts that, as yet, has not been considered in melting models. [82] For a given set of model parameters, equilibrium porous flow models will generate larger disequilibria than dynamic melting models because the difference in residence time in the melting column between parent and daughter is maximized [Spiegelman and Elliott, 1993]. The major difference between dynamic melting models and equilibrium porous flow models is where the U series disequilibria are set up within the melting column. In dynamic melting the U series disequilibria are generated during the onset of melting near the bottom of the melting column, and so preservation of 226 Ra excesses in magmas at the surface requires very rapid rates of melt transport. In equilibrium porous flow models the U series disequilibria are created throughout the melting column, and this 23 of 43 Figure 16. Cartoons to illustrate the main features of the two end-member ingrowth melting models [from Spiegelman and Elliott [1993]. (a) Dynamic melting [McKenzie, 1985], where each instantaneous melt increment is extracted, transported instantaneously in chemical isolation, and then mixed with the other instantaneous melts produced at other depths in the melting column prior to eruption. Incompatible elements are effectively removed from the melting region during the initial stages of melting (where the degree of melting is similar to the bulk partition coefficients of the nuclides of interest), and so nearly all of the daughter ingrowth takes place at the base of the melting column and requires rapid extraction to be preserved in an eruption at the surface. (b) Equilibrium porous flow [Spiegelman and Elliott, 1993], where the melt interacts with residual matrix mantle during transport to the surface. If the daughter nuclide is more incompatible than the parent nuclide, it will have a faster effective velocity through the melt column, and so ingrowth of additional daughter nuclides will take place because the parent nuclide spends more time in the melting column than the daughter nuclide. (c) The principle of nuclide ingrowth is illustrated by the U-Th equiline diagram [Spiegelman and Elliott, 1993], which shows how the activities of 238 U and 230 Th change in both the melt and the residual solid mantle as the degree of melting increases from 0% (base of melting column) to 25% (top of melting column). Elemental fractionation of U and Th is only significant at low degrees of melting. If D U > D Th, as in this example, then the solid becomes enriched in 238 U, and as this decays, the solid and any subsequently formed melt will have higher 230 Th/ 232 Th, thus producing ingrowth of 230 Th. The key parameters that will influence the disequilibria values for ( 230 Th/ 238 U), ( 226 Ra/ 230 Th), and ( 231 Pa/ 235 U) are source mineralogy, partition coefficients within the melting column, matrix porosity, mantle upwelling velocity, melt transport rate, melt-mantle interaction during transport, and presence of volatiles.

24 Figure 17. (a) ( 230 Th/ 238 U) versus axial ridge depth, showing a broad negative correlation of decreasing 230 Th excesses with increasing water depth (updated from Bourdon et al. [1996b] by Lundstrom [2003]). (b) Slope of segment-scale linear data trends on the 230 Th- 238 U equiline diagram versus half-spreading rates of the ridge segment [Lundstrom et al., 1998a; Lundstrom, 2003]. Abbreviations are as follows: AAD, Australian-Antarctic discordance; EPR, East Pacific Rise; FAZAR, 33 N 40 N MAR; JdF, Juan de Fuca; KR, Kolbeinsey Ridge; MAR, Mid-Atlantic Ridge; and RR, Reykjanes Ridge. Figure 17 is taken from Lundstrom [2004], with permission from Mineralogical Society of America. lessens any requirement for rapid melt ascent to explain large 226 Ra excesses. The latest models for melt generation beneath ridges have focused on hybrid models that combine aspects of these two end-member models. In these two-porosity models, melts migrate in channels that are fed by porous flow on the grain scale in the country rock [Spiegelman and Elliott, 1993; Iwamori, 1994; Kelemen et al., 1997; Lundstrom, 2000; Jull et al., 2002]. Such models increase the transport rate of 230 Th from depth and allow isotope disequilibria to be generated at different depths Mid-Ocean Ridge Basalts Observations [83] Obtaining comprehensive U series disequilibria data on suites of MORB glasses is not straightforward, not least because the eruption ages of most samples are unknown. This means that the 226 Ra- 230 Th disequilibria is used primarily to confirm young (<8 kyr) ages for lava flows rather than in constraining melting processes. Furthermore, most sample collection is done by dredging, and sample suites tend to be regional in coverage rather than focusing on local segment-scale variations, although use of submersibles can improve sample density and help to target fresh-looking flows [e.g., Sims et al., 2002a]. Samples also tend to have lower trace element abundances compared to lavas from other tectonic settings, and there is greater potential for seawater alteration and contamination by Fe-Mn crusts (see section 2.3), which all conspire to make it analytically more challenging to obtain high-quality U series disequilibria data. [84] The 230 Th- 238 U disequilibria data can be used as a tool for dating mid-ocean ridge basalts where eruption ages are otherwise difficult to determine [e.g., Goldstein et al., 1992]. Young MORB lavas (essentially those with 226 Ra- 230 Th disequilibrium, i.e., <8000 years) from local ridge segments, in general, form subhorizontal arrays on the U-Th equiline diagram (Figure 18a) [e.g., Lundstrom et al., 1998a]. Provided that the melt generation process in the region has remained uniform over the last 100,000 years or so, such arrays can be used to estimate the initial 230 Th/ 232 Th of any MORB sample in the region from its measured U/Th ratio. Off-axis samples generally have lower than expected 230 Th/ 232 Th, and the simplest explanation is that this difference reflects posteruption radioactive decay of 230 Th and can therefore be used to calculate a model eruption age for each sample. [85] There are five global-scale observations from U series disequilibria measurements of MORB lavas that need to be discussed and explained: (1) the almost ubiquitous 230 Th excesses (Figures 6 and 7); (2) the elevated ( 231 Pa/ 235 U) compared with within-plate and subduction zone magmas (Figure 7b) [e.g., Pickett and Murrell, 1997]; (3) the broad correlation of ( 230 Th/ 238 U) with axial ridge depth (Figure 17a) [e.g., Bourdon et al., 1996b]; (4) the apparent correlation with spreading rate of the slope of data trends from different ridge segments on the 238 U- 230 Th equiline diagram (Figure 17b) [e.g., Lundstrom et al., 1998a]; and (5) the negative correlation between 226 Ra excess and 230 Th excess (Figure 18c) [Kelemen et al., 1997; Sims et al., 2002a]. Recent reviews by Lundstrom [2003] and Elliott and Spiegelman [2003] provide extensive details of the observed variations of U series disequilibria in MORB samples and the potential constraints such data can provide on the nature of the melt generation and transport processes beneath ridges Constraints on Source Mineralogy and Depth of Melting [86] The 230 Th excesses found in most MORB magmas imply that Th is more incompatible than U during mantle melting, i.e., bulk D Th /D U < 1, so that U is preferentially retained in the residual solid relative to Th. Partition coefficient data for U and Th in mantle minerals (summarized by Blundy and Wood [2003]) indicate that the 230 Th 24 of 43

25 Figure 18. The best studied mid-ocean-ridge segment is the East Pacific Rise at 8 10 N (bounded by the Siqueiros Transform to the south and the Clipperton Transform to the north), which has an abundance of samples with known young ages (<8 ka) [Goldstein et al., 1993; Volpe and Goldstein, 1993; Lundstrom et al., 1999; Sims et al., 2002a]. (a) The 230 Th- 238 U equiline diagram, showing a broad linear correlation. (b) Th content versus ( 238 U/ 232 Th), showing that depleted samples with low Th have high ( 238 U/ 232 Th) and low ( 230 Th/ 238 U). (c) ( 226 Ra/ 230 Th) versus ( 230 Th/ 238 U), showing an inverse correlation. (d) ( 231 Pa/ 230 Th) versus ( 230 Th/ 238 U), showing little correlation. excesses require small degree melting in the presence of garnet [Beattie, 1993; LaTourrette et al., 1993] or highpressure aluminous clinopyroxene [Landwehr et al., 2001], as these phases both have D Th /D U < 1 and D U of U and Th have extremely low mineralmelt partition coefficients in olivine and orthopyroxene with D U of , at least an order of magnitude smaller than for garnet or clinopyroxene, giving them little leverage over U-Th fractionation, and so neither olivine nor orthopyroxene have major roles to play in governing 230 Th- 238 U disequilibria. Clinopyroxene is a special case as the relative partitioning of U and Th is very sensitive to composition, which varies as a function of temperature and pressure. At shallow mantle pressures, D Th /D U > 1, but with increasing depth the clinopyroxenes become more aluminous with D Th /D U < 1 at depths greater than 50 km [Landwehr et al., 2001]. Garnet is stable at depths >75 90 km [Robinson et al., 1998; Sims et al., 2002a]. Thus 230 Th excesses may be produced during melting at pressures greater than 1.5 GPa (i.e., >50 km), irrespective of the presence of garnet. However, garnet is much more efficient at generating 230 Th excesses because D Th /D U is much less in garnet than in clinopyroxene [Landwehr et al., 2001]. Although partition coefficients for Ra and Pa have not been measured directly for most mantle minerals, it seems reasonable to assume that 25 of 43 they are very small, such that D U > D Pa and D Th > D Ra for both garnet peridotite and spinel peridotite [e.g., Blundy and Wood, 2003]. This is consistent with the observation that MORB lavas have ( 231 Pa/ 235 U) > 1 and ( 226 Ra/ 230 Th) > 1. [87] Although rare, a few examples of MORB lavas with ( 230 Th/ 238 U) < 1 have been found [e.g., Bourdon et al., 1996b; Sims et al., 2002b; Tepley et al., 2004]. Tepley et al. [2004] showed that the measured disequilibria in depleted samples from the Garrett Transform with 238 U excesses (accompanied by significant 231 Pa and 226 Ra excesses) and low U and Th abundances could be modeled by shallow melting of depleted spinel lherzolite. Furthermore, Lundstrom [2000] noted that MORB samples with 238 U excesses generally have major element compositions similar to experimental melts of spinel lherzolite at 1 GPa. [88] As shown on Figure 7b, MORB magmas have greater 231 Pa excesses than both within-plate magmas and subduction-related magmas. Most within-plate magmas are generated beneath a relatively thick lithosphere that restricts the height of the melt column so that most of the melting takes place in the garnet peridotite stability zone. MORB melting, on the other hand, can continue to shallower levels within the spinel peridotite stability zone, and this longer melting column allows greater ingrowth of

26 231 Pa [Lundstrom, 2003]. It is notable that the within-plate magmas that plot closest to the MORB field are from the Asal rift and Iceland [Pickett and Murrell, 1997; Vigier et al., 1999], both in extensional tectonic settings more similar to a ridge environment. The 231 Pa and 230 Th excesses in MORB can be explained by shallower melting of mantle with a higher clinopyroxene:garnet ratio than most ocean island basalts [Bourdon et al., 1998; Bourdon and Sims, 2003] Influence of Variations in Mantle Temperature, Spreading Rate, and Source Heterogeneity [89] Global correlations between major element compositions of MORB and axial depth of the ridge have been explained by along-axis mantle temperature variations [Klein and Langmuir, 1987], and this model can also account for the broad negative correlation between ( 230 Th/ 238 U) and axial depth [Bourdon et al., 1996b]. Higher mantle temperatures mean that melting initiates at a greater depth, resulting in a longer melt column and more melt generation, which, in turn, forms a thicker crust and a shallow depth for the ridge. Bourdon et al. [1996b] quantitatively explained the increase in 230 Th excess with increasing depth of melt initiation within the garnet stability field, using an equilibrium porous flow model. Nonetheless, there can be a wide range in ( 230 Th/ 238 U) within particular area and depth ranges (e.g., Juan de Fuca samples have a range in ( 230 Th/ 238 U) from 1.1 to 1.4 at constant depth of 2200 m), and this is usually interpreted as primarily the result of local source heterogeneities [Bourdon et al., 1996b; Lundstrom et al., 1998a]. [90] MORB samples from particular regions or segments nearly always show a marked correlation between ( 230 Th/ 238 U) and U/Th. A good example is provided by data from the well-studied 8 10 N region of the East Pacific Rise (EPR) (Figure 18a). In detail, the slope of the trends vary (Figure 17b), and Lundstrom et al. [1998a] argued that they are correlated with the half-spreading rate of the particular ridge. In contrast, Elliott and Spiegelman [2003] argued that these covariations are not significant. Additional studies where there are better constraints on the slopes of local data sets are required to resolve this issue, particularly for those ridge segments for which only older, a-counting data are presently available. Furthermore, the primary mechanism behind the generation of these linear trends is still the subject of much debate. Melt generation beneath ridges is a polybaric process with melt produced from a range of depths within a variably depleted mantle and then mixed prior to eruption. The linear trends are consistent with mixing between two melts derived from different depths in the melting column, but the ultimate origin of the end-member melts is not clear. Lundstrom et al. [1998a] interpreted these local trends as being due to binary mixing of melts derived from enriched and depleted sources within a heterogeneous mantle upwelling at the same rate but melted at different depths, whereas Sims et al. [2002a] and Elliott and Spiegelman [2003] argued that complexities of the melt generation, segregation, transport, and mixing process are capable of generating melts from a homogeneous source with a range in U/Th values. [91] Several ridge segments display significant local variations in U/Th coupled with limited variations in the compositions of long-lived radiogenic isotopes such as 87 Sr/ 86 Sr, 143 Nd/ 144 Nd, 176 Hf/ 177 Hf, and 208 Pb/ 204 Pb (e.g., 9 10 N EPR [Sims et al., 2002a] and Kolbeinsey Ridge [Sims et al., 2002b]), and Sims et al. [2002a, 2002b] therefore argued that the mantle source must essentially be homogeneous in both areas and that polybaric melting and progressive source depletion rather than source heterogeneity produced the observed variations in U/Th. However, the significant diversity in trace element and radiogenic isotope compositions of off-axis lavas [Niu and Batiza, 1997; Niu et al., 2002] suggests that the mantle in the EPR area is heterogeneous on a length scale smaller than ridge segments. Lundstrom [2003] has also argued that mixing curves between the depleted and enriched endmembers are strongly hyperbolic such that limited variations in, for example, 87 Sr/ 86 Sr that are linked to large variations in U/Th can still be consistent with source heterogeneities. Clearly, additional comprehensive data sets on suites of local MORB samples that combine long-lived radiogenic isotope measurements and trace element analyses with U series disequilibria data are required to ascertain the relative roles of source heterogeneity and melting processes in generating local variations in U/Th. Application of high-precision Pb isotope measurements should be particularly beneficial as they offer greater resolution of subtle compositional effects due to mixing of melts from heterogeneous sources [Galer et al., 1999; Thirlwall et al., 2004] Constraints From ( 226 Ra/ 230 Th) Disequilibria [92] As mentioned in section 5.1, determining the depth of generation of the observed 226 Ra- 230 Th disequilibria in MORB can place important constraints on the nature of the melt migration process (rapid melt transport through conduits or slow melt percolation during equilibrium porous flow) because in dynamic melting models all disequilibria are generated at the base of the melting column, whereas in equilibrium porous flow models different disequilibria can be generated at different levels within the melt column. Samples with known eruption ages from the East Pacific Rise (8 10 N) and the Juan de Fuca Ridge show an inverse correlation between 226 Ra excess and 230 Th excess (Figure 18c). This observation is inconsistent with a purely dynamic melting model, which would predict a positive correlation between 226 Ra excess and 230 Th excess [Sims et al., 2002a]. On the other hand, the observations that the major and trace element compositions of MORB melts are not in equilibrium with the shallow upper mantle [Kelemen et al., 1997] and that abyssal peridotites represent residues from near-fractional melting processes [Johnson et al., 1990] together indicate that purely equilibrium porous flow models cannot be the only process involved in the generation and transport of MORB melts. [93] Models have been developed that combine different porosity regimes in the mantle melting column [Kelemen et 26 of 43

27 al., 1997; Lundstrom, 2000; Lundstrom et al., 2000; Jull et al., 2002; Sims et al., 2002a]. In these two-porosity models the anticorrelation between 226 Ra excess and 230 Th excess is produced by mixing of slow moving melt transported through reactive, low-porosity mantle with relatively fast moving melt transported within unreactive, high-porosity channels, again emphasizing the role of melt mixing in the generation of MORB magmas. One problem with these models has been the difficulty of deriving sets of parameters that can simultaneously produce the observed U series disequilibria and still match the extreme elemental depletions observed in abyssal peridotites. An important caveat to this discussion is that these interpretations all assume that the 226 Ra- 230 Th disequilibrium results from the melting process. Saal and Van Orman [2004] recently suggested that 226 Ra excesses might result from diffusive interaction between melts and shallow cumulates. More data are required from other mid-ocean ridge segments to test the robustness of the ( 230 Th/ 238 U)-( 226 Ra/ 230 Th) anticorrelation observed at the Juan de Fuca Ridge and East Pacific Rise. [94] Fluid dynamical models for reactive flow in a deformable permeable matrix [Aharonov et al., 1995; Spiegelman et al., 2001] have shown that a coalescing network of high-porosity melt channels surrounded by extremely low porosity regions will develop in the mantle melt column, consistent with field observations of reactive dunite channels in the mantle sections of ophiolites [e.g., Kelemen et al., 2000]. Modeling of the chemical consequences of such full reactive transport models indicates that significant chemical diversity will be produced in melts of a homogeneous source, consistent with the observed variability in melt inclusions trapped in MORB olivine crystals [Spiegelman and Kelemen, 2003]. More importantly, preliminary investigations [Elliott and Spiegelman, 2003] indicate that these models are capable of at least qualitatively reproducing some of the critical U series disequilibria observations, namely, the linear arrays of ( 238 U/ 232 Th) versus ( 230 Th/ 232 Th) and the negative correlation of ( 230 Th/ 238 U) with ( 226 Ra/ 230 Th), from melting of a homogeneous source, and they clearly merit further attention Within-Plate Magmatism Background [95] In this section we discuss both oceanic island basalts and continental intraplate lavas. Within-plate magmas are compositionally more heterogeneous than MORB, both on a global scale and on a local scale. This is particularly apparent both from variations in radiogenic isotopes (Sr-Nd-Pb-Hf-Os) and elemental concentrations and indicates a wider diversity of potential mantle source materials and degrees of melting [e.g., Zindler and Hart, 1986; Hawkesworth et al., 1990; Hofmann, 1997]. Physical details of the melt generation process are less certain than at mid-ocean ridges, and it is not as easy to constrain uniquely the degrees of melting or to ascertain the links to geodynamical parameters such as mantle upwelling velocity. The presence of the overlying lithosphere adds additional complexity to the melt generation process by controlling the height of the melting column and also by acting as a potential source of melts. Bourdon and Sims [2003] provide a comprehensive review on the constraints that U series disequilibria can provide on the nature of melt generation processes in within-plate settings Relative Roles of Elemental Fractionation and Nuclide Ingrowth on Disequilibria [96] As emphasized by Elliott [1997], a key issue for understanding U series disequilibrium in within-plate basalts (especially OIB) is determining the relative effects of net elemental fractionation versus nuclide ingrowth. The high degrees of melting (10%) inferred for the generation of MORB magmas and the observed large 230 Th and 231 Pa excesses are consistent with an ingrowth model and slow melting rates. However, many within-plate lavas have compositional features (enriched incompatible trace element patterns and silica undersaturation) that indicate much lower degrees of melting [Green and Ringwood, 1967; Gast, 1968] where net elemental fractionation has a greater potential to influence U series disequilibrium. [97] If 230 Th- 238 U disequilibrium is established by net elemental fractionation via a batch melting process, then the measured ( 230 Th/ 232 Th) in a magma should be equivalent to the ( 238 U/ 232 Th) and hence a robust estimate of the U/Th of the mantle source, provided melt production and extraction is rapid. If the melting rate is slow, then ingrowth of 230 Th in the residual mantle becomes significant, causing the ( 230 Th/ 232 Th) in the magma to be greater than in the source. Thus, if ingrowth is significant in controlling 230 Th- 238 U disequilibrium in most within-plate basalts, then the U/Th of the magma should be a better estimate of the source U/Th. If variations in the U/Th ratios of mantle sources are long-lived and related to elemental fractionations of other elements such as Rb-Sr, then correlations between 87 Sr/ 86 Sr and an estimate of source U/Th would be expected. Recent compilations of OIB data by Condomines and Sigmarsson [2000] and Bourdon and Sims [2003] show better correlations between 87 Sr/ 86 Sr and ( 230 Th/ 232 Th) than with ( 238 U/ 232 Th), and this has been used to infer that for many OIB lavas, ( 230 Th/ 232 Th) is a better estimate of source U/Th and that net elemental fractionation of U and Th can be important. However, the diverse range in radiogenic isotope compositions shown by OIB magmas requires at least four types of distinct mantle source components [e.g., Hofmann, 1997], and it is not clear that any robust correlation between 87 Sr/ 86 Sr and either ( 230 Th/ 232 Th) or ( 238 U/ 232 Th) would be expected. [98] Sims et al. [1995, 1999] showed that the extent of 230 Th- 238 U and 231 Pa- 235 U disequilibrium in Hawaiian lavas (Figures 19a and 19d) was related to the degree of melting inferred from major and trace element systematics, with small 230 Th and 231 Pa excesses (2 6% and 10 15%, respectively) in tholeiites (large degree melts) and larger 230 Th and 231 Pa excesses (15 30% and 78%, respectively) in alkali basalts (small degree melts). Thus the 230 Th- 238 U and 231 Pa- 235 U disequilibrium and rare earth element variations could be explained by a batch melting model of a garnet-bearing source (to create the 230 Th excesses). 27 of 43

28 Figure 19. Variations of U series disequilibria with distance from inferred center of hot spot upwellings at Hawaii and Iceland: (a) ( 230 Th/ 238 U), (b) ( 238 U/ 232 Th), (c) ( 226 Ra/ 230 Th), and (d) ( 231 Pa/ 235 Th). Data sources for Hawaii are Cohen and O Nions, [1993], Cohen et al. [1996], Sims et al. [1999], and Pietruszka et al. [2001]; data sources for Iceland are Peate et al. [2001a] and Kokfelt et al. [2003]. However, the significant 226 Ra excesses in the same lavas (Figure 19c) cannot be explained by such a model and require some ingrowth of 226 Ra during melting [Elliott, 1997; Sims et al., 1999] Source Heterogeneities [99] One issue in understanding the compositional diversity of OIB lavas is whether this is related to lithological variations within the mantle source materials. Different lithologies would have different mineral compositions and hence different bulk partition coefficients for trace elements that might have a significant influence on the extent of U series disequilibria developed in suites of OIB lavas. In particular, there has been much discussion over the possible role for garnet pyroxenite or eclogite veins (approximately 70% 50% clinopyroxene and 30% 50% garnet) in the source of OIB lavas. These lithologies are observed in mantle xenolith suites and in peridotite massifs, hosted in a peridotitic matrix (approximately 50% olivine, 20% orthopyroxene, 20% clinopyroxene, and 10% garnet or spinel), and they are thought to represent remnants or melts of recycled oceanic crust [e.g., Hirschmann and Stolper, 1996; Sigmarsson et al., 1998a; Stracke et al., 1999]. [100] It is an important issue because the ubiquitous 230 Th excesses in OIB lavas are generally assumed to be a garnet signature, indicative of deep melting within the garnet peridotite stability field. However, garnet-bearing pyroxenites or eclogites are stable at shallower pressures than garnet peridotite, and if they are responsible for the 230 Th excesses, then this lessens the estimates of the depth of initiation of melting that can be incorporated into geodynamic models. Different authors have reached different conclusions about the relative importance of pyroxenites in controlling 230 Th- 238 U disequilibria in OIB [e.g., 28 of 43 Sigmarsson et al., 1998a; Stracke et al., 1999]. The mineral compositions of garnet and clinopyroxene in pyroxenitic versus peridotitic lithologies are quite different, and the compositional dependence of mineral/melt partition coefficients should thus lead to different partitioning behavior for U and Th, as well as other trace elements such as Lu and Hf [e.g., Stracke et al., 1999; Pertermann et al., 2004]. Therefore, provided we have the relevant partitioning data and comprehensive trace element and U series disequilibria analyses from a suite of OIB lavas, this should be a resolvable issue Constraints on Variations in Mantle Upwelling Rates Near Hot Spots [101] Iceland is a ridge-centered hot spot, and yet the rift tholeiites have lower than expected 230 Th excesses based on the global MORB correlation between 230 Th excess and ridge depth (Figure 17a). These tholeiites are generated by large degrees of melting broadly similar to MORB magmas, and so nuclide ingrowth is likely to be important in controlling 230 Th- 238 U disequilibrium. The lower than expected 230 Th excesses suggest faster melting rates, and this is most simply explained as active mantle upwelling (i.e., upwelling faster than expected from passive plate separation), consistent with models for a hot, buoyant mantle plume beneath most of Iceland [Bourdon et al., 1996b; Peate et al., 2001a; Kokfelt et al., 2003]. Sleep [1990] and Ribe [1996] used geophysical observations to determine the buoyancy flux beneath many oceanic hot spots. There is a direct link between buoyancy flux and mantle upwelling velocity such that the inferred global range in buoyancy flux equates to a threefold to fourfold variation in upwelling velocity, which can explain the range in observed 230 Th and 231 Pa excesses in ocean island basalts [e.g., Bourdon et al., 1998]. The broad negative correlation

29 between average ( 230 Th/ 238 U) at an oceanic hot spot and buoyancy flux was noted by Chabaux and Allègre [1994] and can be explained by variations in upwelling rate, with the low ( 230 Th/ 238 U) values typical of tholeiites from the high-buoyancy Hawaiian hot spot consistent with limited 230 Th ingrowth during fast mantle upwelling. [102] Within individual hot spot locations, there are often significant variations in U series disequilibria that will have been influenced by several factors, including compositional heterogeneities within the source, different extents of melting, and variable upwelling rates. In several cases the U series disequilibria vary systematically with distance from the inferred hot spot center (e.g., Hawaii [Sims et al., 1999], Iceland [Kokfelt et al., 2003], and the Canaries [Lundstrom et al., 2003]) (Figure 19), which is consistent with models for radial variations in mantle upwelling rate above a mantle plume: The lowest ( 230 Th/ 238 U) and ( 231 Pa/ 235 U) values are found in lavas from the fast upwelling central axis, with higher values found in the slower upwelling periphery allowing more time for ingrowth. At least in the case of Iceland the increase in 230 Th excesses with increasing distance from the inferred plume axis is found in samples with similar U/Th ratios (Figure 19b), thus minimizing the potential effects of source heterogeneities that might otherwise influence the magnitude of 230 Th- 238 U disequilibria through variations in melting rates (i.e., more/less productive mantle sources) or mineralogy. Differences in 230 Th excesses for a given U/Th ratio between samples from the periphery of Iceland and the adjacent Reykjanes Ridge can be interpreted as indicating that the whole of Iceland is influenced by active upwelling [Peate et al., 2001a]. Although the absolute values for the solid upwelling velocity beneath the different Hawaiian islands inferred from the U series disequilibrium data are highly sensitive to the choice of melting model and partition coefficients, Sims et al. [1999] showed that there has to be at least an order of magnitude decrease in upwelling velocity from the tholeiitic systems to the alkali basalt systems. These spatial variations in upwelling are broadly consistent with predictions from geodynamic models for a plume-like upwelling beneath Hawaii Influence of Lithospheric Mantle [103] Melt generation in within-plate tectonic settings is not necessarily restricted to upwelling asthenospheric mantle, and melting within the overlying lithosphere is also possible in both oceanic and continental environments. Hydrous phases such as amphibole and phlogopite can be present at intermediate pressures within the oceanic and continental lithosphere following metasomatic enrichment episodes, and their presence is known both from xenolith samples and also indirectly from trace element signatures in certain magmas. These minerals will preferentially incorporate Ba (and thus Ra too) relative to Th, and if present at the few percent level in a mantle source, this will lead to reduced ( 226 Ra/ 230 Th) or even 226 Ra deficits in magmas (e.g., Grande Comore [Claude-Ivanaj et al., 1998], Samoa [Bourdon and Sims, 2003], and the Azores [Claude-Ivanaj et al., 2001]). [104] In continental within-plate settings the lithospheric mantle is likely to have a more significant role in influencing the compositions of the erupted magmas. Large lateral variations in thickness will potentially limit the extent to which the underlying asthenospheric mantle can decompress and melt, if at all. Metasomatic fluid and melt enrichments of the continental lithosphere will produce significant mineralogical and compositional heterogeneities, and its long stabilization history (1 3 Ga) means that trace element enrichments will potentially develop significant radiogenic isotopic compositions distinct from the asthenosphere [Hawkesworth et al., 1990]. Trace element and radiogenic isotope data can therefore be used to assess whether a particular sample was likely to have been derived from the asthenosphere or from enriched metasomatized lithospheric mantle. [105] This approach has been used for within-plate magmas from the western United States [e.g., Perry et al., 1987; Kempton et al., 1991]. Asmerom [1999] and Asmerom et al. [2000] found that low 143 Nd/ 144 Nd lavas (inferred to be lithospheric melts) had 230 Th- 238 U equilibrium, whereas high 143 Nd/ 144 Nd lavas (inferred to be asthenospheric melts) had large 230 Th excesses of 10 40% (Figure 20a), but both types had significant 231 Pa excesses (Figure 20b). The large 231 Pa excesses rule out the possibility that the 230 Th- 238 U equilibrium values in the lithospheric melts are simply the result of radioactive decay during slow magma transport and long crustal residence. Modeling indicated that the combination of 231 Pa excesses and 230 Th- 238 U equilibrium require a spinel peridotite source, consistent with the relatively shallow depth to the base of the lithosphere around the margins of the Colorado Plateau where the lithospheric melts are found. It should be noted that equilibrium 230 Th- 238 U values are not a distinguishing characteristic of all lithosphere-derived melts, as low- 143 Nd/ 144 Nd samples from other western United States locations can have 230 Th excesses (Figure 20a). A good example are lavas from SW Utah [Reid and Ramos, 1996], where the lithosphere is at least 100 km thick [Wang et al., 2002], thick enough for garnet peridotite to be stable in the lower 20 or so km, thus enabling 230 Th excesses to be generated. It is also possible, though, for samples with relatively low 143 Nd/ 144 Nd to have 230 Th excesses simply as a result of mixing between a lithospheric melt with even lower 143 Nd/ 144 Nd and 230 Th- 238 U equilibrium (generated in the spinel-peridotite stability field) and an asthenospheric melt with a large 230 Th excess. [106] Highly potassic magmas are generally interpreted as melts of metasomatized regions of the lithospheric mantle. Data from three examples (Tibet [Cooper et al., 2002], Wudalianchi, China [Zou et al., 2003], and Gaussberg, Antarctica [Williams et al., 1992]) all show very low 143 Nd/ 144 Nd and significant 230 Th excesses of between 7 and 60% (Figure 20a). For the Gaussberg lamproites and the Wudalianchi potassic basalts the small within-suite variations in 230 Th- 238 U disequilibria correlate with radiogenic isotope compositions indicating a role for source heterogeneity, likely to be the result of variable extents of 29 of 43

30 Figure 20. (a) ( 230 Th/ 238 U) versus 143 Nd/ 144 Nd for within-plate lava suites: Gaussberg lamproites, Antarctica [Williams et al., 1992], Tibet trachyandesites [Cooper et al., 2002], Wudalianchi potassic basalts, China [Zou et al., 2003], and Auckland alkali basalts, New Zealand [Huang et al., 1997]. For the continental basalts from the western United States, open circles are lithospheric melts from Reid [1995] and Reid and Ramos [1996], while the data from Asmerom and Edwards [1995], Asmerom [1999], and Asmerom et al. [2000] are divided into lithospheric melts (shaded circles) and asthenospheric melts (solid circles). (b) ( 230 Th/ 238 U) versus ( 231 Pa/ 235 U), showing the distinction between asthenosphere-derived (high 143 Nd/ 144 Nd) magmas with large 230 Th excesses (solid circles) and lithosphere-derived (low 143 Nd/ 144 Nd) magmas with small or zero 230 Th excesses (shaded circles) from the western United States [Asmerom, 1999; Asmerom et al., 2000]. Field for oceanic island basalts (OIB) and MORB samples is shown for reference (data sources as for Figure 6). metasomatism of a garnet peridotite source. Major and trace element data on the Wudalianchi potassic basalts indicate that the source is a phlogopite-bearing garnet peridotite, and in this region the lithosphere is known to be 120 km. Zou et al. [2003] preferred a model in which 30 of 43 these basalts were derived by 5 7% melting of a slow upwelling mantle source, consistent with evidence for some lithospheric extension in the area, although very small degree melting ( %) of a static source was also permissible (i.e., using a time-independent melting model) and highlights the ambiguities still present in such modeling. Thus the degree of U series disequilibria in continental within-plate lavas depends on many factors, including the tectonic setting (thinned rift zone versus thick lithosphere), the location of melting (asthenosphere versus lithosphere), the source mineralogy (garnet-bearing versus spinel-bearing mantle, which reflects the depth of melting), and source compositional heterogeneities Subduction Zone Magmatism Background to Melt Generation at Subduction Zones [107] Studies of elemental and isotopic variations in subduction zone lavas over the last few decades, combined with constraints from experimental petrology, have lead to a broad consensus as to the dominant source components involved in the genesis of subduction zone magmas (see reviews by Hawkesworth et al. [1993] and Pearce and Peate [1995]). Melt generation takes place predominantly in the mantle wedge that overlies the subducting slab, as a result of the lowering of the mantle peridotite solidus by addition of water-rich fluids. These fluids are released by dehydration reactions in the slab, primarily from the subducting basaltic crust but potentially from serpentinized mantle as well. Many subduction zone magmas contain a significant additional contribution from subducted sediments, and there is some evidence that the sediment component may be added to the mantle wedge as a melt before the fluid addition event that leads to arc magma generation [e.g., Reagan et al., 1994; Elliott et al., 1997; Turner and Hawkesworth, 1997]. The resulting magmas may subsequently interact to some extent with the overlying crust during ascent to the surface [e.g., Davidson, 1996]. [108] Subduction zones probably represent the most complex melting environment and, unlike at mid-ocean ridges, the primary controls on melting are poorly established. Melt generation will be influenced by the amount of fluid introduced from the downgoing slab, the thermal structure of the mantle wedge, and decompression in the mantle wedge. Significant compositional variations both along individual arcs and globally between arcs are also partly controlled by variations in the amount and lithology of sediments being subducted [e.g., Plank and Langmuir, 1993] and in the extent of depletion of the mantle wedge due to prior melt extraction in the rear arc region [e.g., Woodhead et al., 1993]. A recent study by England et al. [2004] argues that the location of arc volcanoes is critically dependent on the thermal structure of the mantle wedge rather than on the depth to the subducting slab. [109] Fluids clearly play a major role in generating subduction zone magmas, and it is likely that they will have a significant influence on the extent and sense of disequilibria between U series nuclides in these magmas.

31 Figure 21. (a) The 238 U- 230 Th equiline diagram for historic Mariana arc lavas [Elliott et al., 1997] plus sketched model interpretation. (b) The 238 U- 230 Th equiline diagram for Tonga arc lavas (shaded symbols) and Vala Fa back arc lavas (open symbols) [Turner et al., 1997a; Peate et al., 2001b]. Arc samples with low SiO 2 (<55 wt %) and the Valu Fa back arc samples all have similar 143 Nd/ 144 Nd, indicating a similar mantle source but with variable addition of a U-rich fluid, and they define an isochron age of 50 ka. (c) ( 230 Th/ 238 U) versus Ba/Th for Tonga and Marianas arc lavas. (d) ( 226 Ra/ 230 Th) versus Ba/Th for Tonga and Marianas arc lavas. This is because of the contrasting chemical behavior of the different U series nuclides in the presence of fluids. U and Ra should behave like large ion lithophile elements (e.g., Ba, Rb, K, and Sr) and be readily mobilized in oxidized aqueous-rich fluids, whereas Th and Pa should behave like the relatively immobile high field strength elements (e.g., Zr, Nb, and Ti) [e.g., Brenan et al., 1995; Keppler, 1996]. Thus there is great potential to use 238 U- 230 Th, 226 Ra- 230 Th, and 231 Pa- 235 U disequilibria to provide constraints on the timing of fluid addition to the mantle wedge source beneath arcs. Turner et al. [2000c, 2003a] provide comprehensive recent reviews of the origin and interpretation of U series disequilibria in the subduction zone environment. A critical aspect in the interpretation of U series disequilibria data in subduction-related rocks is distinguishing the relative contributions of elemental fractionation caused by fluid addition from that caused by the partial melting process The 238 U Excesses in Subduction Zone Magmas and Links to Slab Fluid Addition [110] A key feature of the subduction zone environment is that many lavas have significant 238 U excesses [e.g., Newman et al., 1984; Gill and Williams, 1990; Hawkesworth et al., 1997a], and many of these lavas also have much higher U/Th ratios than MORB or within-plate lavas (Figure 7a). Observations of 238 U-excess in other tectonic settings are very rare and only of small magnitude, with the exception of carbonatites (Figures 6 and 7). In subduction zone lavas the highest 238 U excesses are found in samples with the lowest Th contents [e.g., Condomines and Sigmarsson, 1993]. These samples are mainly from trace element-depleted intraoceanic arcs such as Tonga and the Marianas (Figures 21a and 21b), and they also tend to have the strongest trace element evidence for addition of a slab-derived fluid (e.g., high Ba/Th (Figure 21c)). The mobility of uranium in fluids is governed by the redox environment, as it is only fluid mobile under oxidizing conditions where it can exist as U 6+ rather than U 4+, unlike Th which only exists in the relatively immobile Th 4+ state. Compositional data on subduction zone peridotite samples show that conditions in the subarc mantle are markedly more oxidizing than those in oceanic or ancient cratonic mantle, indicating that the infiltrating slabderived fluids are oxidizing enough for U 6+ to be the dominant uranium species [e.g., Parkinson and Arculus, 1999]. Thus the simplest explanation of the elevated U/Th and Ba/Th ratios in many subduction zone lavas is the addition of a U-Ba-rich slab fluid to the mantle wedge source. The observation that many subduction zone samples have large 238 U excesses indicates that this fluid addition was, on average, a relatively recent event (<380 ka). Along-arc correlations between plate convergence rate and ( 230 Th/ 238 U) in the Alaska-Aleutian arc further suggest that 238 U excesses are linked to the relative size of the fluid flux [George et al., 2003]. 31 of 43

32 [111] Lavas erupted at active spreading centers situated behind the main volcanic arc front within back arc basins show a diverse range of compositions. They include samples showing significant input from slab-derived fluids as well as samples virtually indistinguishable from normal mid-ocean ridge basalts. Thus these back arc lavas provide a means to assess the relative influences of mid-ocean-ridge style decompression melting processes and addition of slab-derived fluids on 230 Th- 238 U disequilibria. In the two examples that have been studied in detail (Lau Basin [Peate et al., 2001b] and Scotia Basin [Fretzdorff et al., 2003]), lavas with 230 Th excesses and lavas with 238 U excesses have both been found (Figure 6). In both cases the samples with the 238 U excesses show the clearest trace element evidence for enrichment in a slab-derived fluid component, and they tend to be found closest to the arc front. For the Lau Basin the transition from samples with 238 U excesses to those with 230 Th excesses takes place at about 250 km behind the trench. The Lau Basin samples with 230 Th excesses and trace element compositions broadly similar to mid-ocean-ridge basalts still have enriched water contents relative to typical mid-ocean-ridge basalts. [112] Protactinium should behave similarly to high-fieldstrength elements such as Zr and Nb and be relatively immobile in slab-derived fluids relative to uranium. Thus, by analogy with the 230 Th- 238 U system, excesses of 235 U over 231 Pa should be expected in typical arc lavas. It is somewhat surprising therefore that most subduction zone lavas in fact have ( 231 Pa/ 235 U) > 1 (Figures 6 and 7) [Pickett and Murrell, 1997; Bourdon et al., 1999b; Thomas et al., 2002; Dosseto et al., 2003; Regelous et al., 2003]. In contrast to fluid addition, partial melting should generate 231 Pa excesses because D Pa < D U for mantle sources. The observation of ( 231 Pa/ 235 U) > 1 for most subduction zone lavas suggests that partial melting has a greater effect than fluid addition in controlling the sense and magnitude of 231 Pa- 235 U disequilibria. The only known exceptions are some extremely trace element-depleted lavas from the Tonga arc. Bourdon et al. [1999b] argued that the highly depleted mantle wedge in this region (because of prior melt extraction in the back arc [Ewart and Hawkesworth, 1987]) contains only a small proportion of clinopyroxene and that this makes it difficult to fractionate Pa from U during melting The 230 Th Excesses in Subduction Zone Magmas and Links to Melting Processes [113] It is notable that a significant proportion of subduction zone magmas (20%) have 230 Th excesses similar in magnitude to values observed in many mid-ocean ridge and within-plate lavas (Figures 6 and 7). These samples tend to be found in particular tectonic settings within subduction zones, such as in rear arc regions (e.g., Kamchatka and Sunda) and areas of thick crust (e.g., Central America, Alaska, and the Andes), and to be associated with intra-arc rifts (e.g., Kamchatka and Vanuatu). The process responsible for producing these 230 Th excesses is not necessarily the same in each case, but it likely involves some meltingrelated fractionation of Th and U. The three main options are the following Decompression/Dynamic Melting [114] Excesses of 230 Th are an almost ubiquitous signature of decompression melting in mid-ocean ridge and within-plate settings because of ingrowth of 230 Th as melt and matrix migrate through the melting region at different velocities (see section 5.1 and Figure 15). Various lines of evidence have been used to suggest that a component of decompression melting can be important in the generation of arc magmas [Pearce and Peate, 1995] (e.g., global interarc correlations of major elements with lithospheric thickness [Plank and Langmuir, 1988], theoretical modeling of melting of hydrated peridotite [Hirschmann et al., 1999], and the existence of low-h 2 O basalts in some arc front volcanoes [Sisson and Bronto, 1998]). Decompression melting is likely to be responsible for the 230 Th excesses observed in some lavas from extensional environments such as back arc basins (e.g., Lau Basin [Peate et al., 2001b] and Scotia Basin [Fretzdorff et al., 2003]) and some intra-arc rifts (e.g., Aoba in the Vanuatu arc [Turner et al., 1999]). Turner et al. [2003a] argued that the association of 230 Th excesses with lavas erupted through thick continental crust rather than thin oceanic crust was consistent with models in which the thickness of the overlying lithosphere influences the mantle dynamics and partial melting within the wedge [e.g., Plank and Langmuir, 1988], and George et al. [2003] suggested that dynamic melting processes produced the 230 Th excesses in certain continental Alaskan lavas. Thomas et al. [2002] showed that 230 Th excesses can also be produced by small degrees of melting using a flux melting model, in which the extent of fluid addition drives the melting process Slab Melting [115] Thermal modeling indicates that melting of the subducting slab is possible in the specific case where the subducting oceanic crust is young and hot [Peacock et al., 1994]. At depths >65 km the basaltic crust will be converted to eclogite (garnet + clinopyroxene), and so slab melts should be generated in the presence of residual garnet. This will give rise to melts with 230 Th excesses and trace element features such as high La/Yb and Sr/Y. Addition of a slabderived melt has been suggested to explain the 230 Th excesses and trace element features of lavas in the austral Andes in southern Chile [Sigmarsson et al., 1998b] and in the central Kamchatkan depression [Dosseto et al., 2003] Crustal Assimilation [116] Bourdon et al. [2000b] showed that crustal assimilation can, at least in the case of Parinacota volcano in Chile, lead to 230 Th excesses if the crustal material is added as a melt from a garnet-bearing residue. This will potentially only affect lavas passing through very thick continental crust as garnet is only stable at depths >30 km The 230 Th/ 232 Th Variations in Subduction Zone Lavas: Influence of Subducted Sediments [117] The Th budget of subduction zone lavas is largely dominated by the addition of sedimentary material because mantle rocks have much lower Th contents than continental- 32 of 43

33 Figure 22. (a) The 238 U- 230 Th equiline diagram for samples from the Sunda arc [Turner and Foden, 2001; Turner et al., 2003b], Kamchatka arc [Turner et al., 1998; Dosseto et al., 2003], and the Nicaraguan part of the Central American arc [Thomas et al., 2002]. (b) The 238 U- 230 Th equiline diagram for lavas from Mijakejima volcano, Izu arc [Yokoyama et al., 2003]. derived sediments [e.g., Plank and Langmuir, 1993; Elliott et al., 1997; Hawkesworth et al., 1997a]. Simple mass balance calculations show that addition of just 1% of typical subducted sediment [Plank and Langmuir, 1998] will contribute 80% 98% of the Th in a mantle wedge source [e.g., Hawkesworth et al., 1997b]. Thus the Th isotope composition of a mantle wedge source will be largely governed by the 230 Th/ 232 Th of the added sediment, which can be estimated directly from the U/Th ratio of the subducted sediment, assuming that the sediment is in secular equilibrium and that it is added by bulk mixing. However, there are several lines of evidence that suggest that the sediment material might be transferred as a melt, which can change the U/Th value of the added sedimentary component. Hawkesworth et al. [1997a] showed that arc lavas globally display a broad negative correlation between Th/Ce and 143 Nd/ 144 Nd consistent with variable addition of sediment to a MORB-like mantle wedge. However, samples with the highest sediment contribution (i.e., those with the lowest 143 Nd/ 144 Nd) had Th/Ce values higher than most estimated subducted sediment compositions, which suggested that this elemental ratio had been modified by partial melting of the sediment. In the Marianas arc the samples with the highest sediment contribution lie near the equiline on Figure 21a, with ( 230 Th/ 232 Th) 1.0, and yet the bulk composition of the subducting sediments is estimated to have ( 230 Th/ 232 Th) 0.6. Elliott et al. [1997] argued that the ( 238 U/ 232 Th) value of the sediment was modified from 0.6 to 1.0 by partial melting but that this sediment melt was then added to the mantle wedge at least 380 kyr prior to the fluid addition event so that the sediment-modified wedge had time to ingrow 230 Th so as to be close to secular equilibrium (see Figure 21a). [118] Samples from sediment-dominated arcs such as the Sunda arc (as inferred from trace element and radiogenic isotope data) tend to have low U/Th and 238 U- 230 Th disequilibria values close to the equiline (Figure 22a). This is because addition of a slab-derived fluid has little influence on 238 U- 230 Th disequilibria because of the high Th and U contents of the sediment-modified mantle wedge [e.g., Hawkesworth et al., 1997a]. Most subducted sediments, with the exception of carbonates and U-rich hemipelagic deposits, have U/Th values lower than average MORB mantle. Lavas from some subduction zones, notably parts of Central America (Nicaragua) and Kamchatka have high ( 230 Th/ 232 Th) values of , greater than observed in any MORB samples (Figure 22a). For Kamchatka the preferred model is for an old (>380 ka) fluid enrichment episode, which produced high U/Th ratios that through subsequent radioactive decay has lead to the observed high ( 230 Th/ 232 Th) values [Turner et al., 1998; Dosseto et al., 2003]. However, in the case of Nicaragua, high ( 230 Th/ 232 Th) values are consistent with the addition of sedimentary material with high U/Th (and thus high ( 230 Th/ 232 Th)) together with some effect also from an old fluid-enrichment episode [McDermott and Hawkesworth, 1991; Reagan et al., 1994; Thomas et al., 2002] Systematic 238 U- 230 Th Disequilibria Variations Within Individual Island Arcs [119] In detail, samples from individual subduction zones, mainly those from trace element depleted island arcs, often define broad linear arrays on the 238 U- 230 Th equiline diagram. If these trends are interpreted as isochrons, they have apparent ages ranging from 10 to 200 kyr prior to eruption (summarized byturner et al. [2000c]). The clearest example is shown by data from the oceanic Marianas arc that define an age of 30 ka (Figure 21a) [Elliott et al., 1997]. Data from the Tonga arc and back arc system show a more scattered trend (Figure 21b) with a slope corresponding to an age of 50 ka [Turner et al., 1997a; Peate et al., 2001b]. There has been considerable debate as to the age significance of these pseudo isochrons, and some of the various interpretations are sketched on Figure 21a. [120] Earlier studies tended to interpret the slope of the array as an isochron, for which the age would represent the time elapsed since a discrete episode of fluid addition 33 of 43

34 to the mantle wedge source. This interpretation requires several assumptions: (1) The fluid contains no Th, so variable fluid addition initially produces a horizontal array on the equiline diagram; (2) the mantle wedge has the same 230 Th/ 232 Th throughout the arc; and (3) melting processes do not modify the 238 U- 230 Th systematics. Another possible interpretation for the arrays is that the fluid does contain some Th, so the initial fluid addition may produce a sloped rather than a horizontal array. Similarly, if the mantle wedge has variable U/Th, addition of a constant flux of fluid containing just U will have a greater effect on the more depleted parts of the wedge with higher initial U/Th, thereby producing a sloped data array. Several recent studies have also argued that melting processes beneath arcs can produce sloped arrays on the 238 U- 230 Th equiline diagram [e.g., Elliott et al., 2001; Thomas et al., 2002; Bourdon et al., 2003a; George et al., 2003], which will be discussed further in section [121] It is important to realize that most studies have been regional in coverage with only one or two samples from individual volcanoes within an arc. Yokoyama et al. [2003] have shown, through a detailed study of a single arc volcano (Miyakejima volcano, Izu arc), that the 238 U- 230 Th systematics can be locally very complex. The earliest lavas erupted between 25 and 10 ka define a linear trend with an apparent age of 22 ka, and the lavas erupted between 7 and 4 ka define a different linear trend with an apparent age of 12 ka (Figure 22b). Both lines intersect the equiline at a common ( 230 Th/ 232 Th) value of 1.3, which, together with the uniform 143 Nd/ 144 Nd ratios in these samples, suggests that the mantle wedge beneath Miyakejima was compositionally homogeneous prior to fluid addition. The more differentiated samples erupted in the last few thousand years define an array that is steeper than the equiline and indicative of magma mixing, consistent with petrological and other compositional data. Yokoyama et al. [2003] argued that it was difficult to explain the two trends in the least differentiated lavas by a flux-melting-type model [e.g., Thomas et al., 2002], and instead, they explained the data in terms of the episodic events of fluid addition triggering melt generation in the mantle wedge on a timescale of several thousands of years The 226 Ra- 230 Th Disequilibria and Implications for Magma Ascent Rates [122] Many subduction zone lavas have significant 226 Ra excesses (Figure 7c) [e.g., Gill and Williams, 1990; Turner et al., 2001a], and the question is whether these 226 Ra excesses result from melting processes or fluid addition. The highest ( 226 Ra/ 230 Th) values are generally found in the least differentiated lavas, indicating that the 226 Ra- 230 Th disequilibrium has a mantle origin. Turner and Hawkesworth [1997] suggested that the 226 Ra excesses were generated by partial melting and in that way decoupled from the 230 Th- 238 U disequilibrium caused by addition of a slab fluid. However, the degree of 226 Ra- 230 Th disequilibrium in some arc lavas, with ( 226 Ra/ 230 Th) up to 7, is far greater than in MORB and OIB, which generally have ( 226 Ra/ 230 Th) < 3, and generating such extreme values by decompression melting alone would require extremely small porosities (<0.0001% [Turner et al., 2001a]). Furthermore, within individual arcs, 226 Ra excesses generally show a reasonable correlation with Ba/Th (Figure 21d), which suggests a link with addition of a slab-derived fluid [Turner et al., 2001a]. Bourdon et al. [2003a] also noted that the positive correlation of ( 226 Ra/ 230 Th) with Sr/Th shown by samples from both the Kamchatka arc and the Tonga-Kermadec cannot be due to melting because Sr is more compatible than Th during mantle melting (unlike for Ba, which is more incompatible than Th). Instead, the trends can be explained by the fluid mobility of Sr and Ra relative to Th. A further link between a slab-derived signature and 226 Ra excess is provided by the positive correlation between ( 226 Ra/ 230 Th) and 10 Be/ 9 Be in lavas from southern Chile [Sigmarsson et al., 2002], as 10 Be is unambiguously derived from subducting sediments. [123] The principal 230 Th- 238 U and 226 Ra- 230 Th fractionation appears to occur when the oxidizing fluid is released from the dehydrating slab. We are therefore faced with an apparent discrepancy between the timescales for fluid transfer as indicated by these two disequilibria pairs. The arrays on the 230 Th- 238 U equiline diagram can be interpreted as the result of a fluid addition event kyr ago, but if 226 Ra excesses were produced during the same event, they would have already decayed back to equilibrium. The observed 226 Ra excesses therefore suggest a very recent fluid addition to the mantle wedge within the last few thousand years. If this interpretation of the 226 Ra excesses is correct, that they were generated as the fluids were released from the subducted slab, then this has significant implications for the mechanism of fluid transfer and ascent rate of magmas. The fluid has to be transferred from the slab to the melting zone within the mantle wedge, and then the resulting melt has to be transferred from 80 to 100 km depth to be erupted at the Earth s surface within just a few thousand years, at most, of the fluid release event. This implies rapid fluid transfer, probably via hydraulic fracturing [Davies, 1999], rapid melt segregation and ascent by channeled flow, followed by minimal residence in magma chambers within the crust [Clark et al., 1998; Turner et al., 2000b; Turner et al., 2001a; Sigmarsson et al., 2002; Yokoyama et al., 2003]. [124] Feineman and DePaolo [2003] argue that minerals in the mantle wedge might not be in secular equilibrium for 226 Ra- 230 Th. The 226 Ra generated by decay of 230 Th can readily diffuse out of clinopyroxene and potentially into phlogopite if present, as Ra is compatible in phlogopite. If the added slab fluid is stabilized in the wedge by growth of phlogopite, then as the mantle flow brings this material to the melting zone over a finite time (perhaps tens of kiloyears), the 238 U excesses would decrease because of radioactive decay, while the phlogopite would maintain a steady state elevated ( 226 Ra/ 230 Th) by diffusive reequilibration. Feineman and DePaolo [2003] argue that this model lessens the constraints for a rapid fluid transfer from the slab as previously suggested based on 226 Ra/ 230 Th 34 of 43

35 [e.g., Turner et al., 2001a], although it still implies rapid melt ascent rates (1 km yr 1 ) from the initial site of melting through the mantle wedge and crust to the surface. [125] The question remains as to how the 230 Th- 238 U and 226 Ra- 230 Th data can be reconciled. Several studies have shown reasonable correlations between ( 230 Th/ 238 U) and ( 226 Ra/ 230 Th) for particular subduction zone suites [Reagan et al., 1994; Chabaux et al., 1999; Sigmarsson et al., 2002], and these authors suggested that both U and Ra were added at the same time to the wedge by a single slab fluid less than a few thousand years ago, such that the arrays on the U-Th equiline diagrams are essentially mixing trends between the fluid and the wedge compositions. However, Bourdon et al. [2003a] have argued that this would imply unrealistically high fluid partitioning values for Th and hence that a significant proportion of the 230 Th in arc lavas must come from ingrowth from excess 238 U over a time period >10 kyr. Experimental work by Schmidt and Poli [1998] indicates that slab dehydration will probably result in the multistep or continuous release of fluids over a significant range of depths. Turner et al. [2000b] modeled a simple two-stage addition approximation in an attempt to explain data from the Tonga arc. The initial fluid addition will remove virtually all the 238 U and 226 Ra but leave 230 Th and 232 Th behind in the slab. Further dehydration of the slab will release liquids that contain 226 Ra produced from decay of the 230 Th remaining in the slab but no U. The 226 Ra excesses produced by the first fluid addition event will have decayed away by the time of the subsequent fluid additions, so Turner et al. [2000b] argued that the 238 U excesses reflected the timing of the initial fluid addition, whereas the 226 Ra excesses reflected the last increment of fluid addition Subduction Zone Melt Generation Models [126] Compared to mid-ocean ridges and even withinplate setting, the development of realistic quantitative melting models for subduction zones is still in its infancy, largely because of the complexity of the melt generation process. There are uncertainties about the details of the mechanisms that must be incorporated into a melting model (e.g., dynamics of mantle flow in wedge, how the fluid is transferred from the slab and how it migrates within the wedge to the melting zone, and dynamic melting versus fluid-fluxed melting), and many of the critical parameters are poorly constrained (e.g., thermal structure of mantle wedge and element partitioning in fluids). Nevertheless, initial exploratory attempts are being made to model details of melting in arcs as constrained by U series disequilibria [Thomas et al., 2002; Bourdon et al., 2003a; George et al., 2003; Yokoyama et al., 2003]. [127] Bourdon et al. [2003a] argued that equilibrium porous flow melting models [e.g., Spiegelman and Elliott, 1993], in which there is extensive reequilibration between melt and residual mantle throughout the melting column, are unlikely in the arc setting for several reasons: (1) The fast melt migration times inferred from the 226 Ra data leave little time for melt equilibration. (2) The elevated 187 Os/ 188 Os isotope compositions of some arc lavas, if confirmed as a slab signature, would not survive if there was significant melt equilibration with the mantle wedge. (3) Explaining the ( 231 Pa/ 235 U) and ( 226 Ra/ 230 Th) data of arc lavas by melting of a fluid-modified source using a equilibrium porous flow model requires very high melting rates, and by implication very fast upwelling velocities (200 cm yr 1 ), compared to geologically more reasonable rates using a dynamic melting model. Therefore initial models have investigated the consequences of dynamic melting [e.g., McKenzie, 1985; Williams and Gill, 1989] taking place either subsequent to fluid addition [Elliott et al., 2001; Bourdon et al., 2003a; George et al., 2003] or contemporaneous with fluid addition (fluxed ingrowth melting [Thomas et al., 2002; Bourdon et al., 2003a]), and these two scenarios are illustrated in Figure Dynamic Melting Following Fluid Addition [128] Bourdon et al. [2003a] considered the situation in which fluid addition and partial melting are separate events, perhaps with a finite time interval between them. Assuming reasonable values for partition coefficients, porosity, and fluid-induced disequilibria, the modeling achieved values for ( 231 Pa/ 235 U) and ( 226 Ra/ 230 Th) typical of arc lavas for melting rates on the order of kg m 3 yr 1.At larger melting rates, there is not sufficient time to ingrow 231 Pa in the residual mantle, and at smaller melting rates the initial 226 Ra excess from fluid addition will have decayed away. The dynamic melting process will also produce linear arrays on the U-Th equiline diagram reminiscent of an isochron [see also Elliott et al., 2001; George et al., 2003]. Slower melting rates produce steeper arrays (and hence older isochron ages ) than faster melting rates, and the slopes are also influenced by the choice of partition coefficients for U and Th that will reflect the mineral mode and extent of depletion of the mantle wedge Fluxed Ingrowth Melting [129] In the model of Thomas et al. [2002], partial melting occurs as an immediate response to continued fluid addition. Fresh hot mantle is dragged through the zone of slab fluid addition because of corner flow in the mantle wedge, and it is assumed that the transfer of fluid from the slab, the partial melting, and the melt transport are all essentially instantaneous. Flow of the solid mantle, on the other hand, takes a long enough time ( years) to pass through the melting region to allow ingrowth of 231 Pa and 230 Th from U remaining in the residues of melt extraction. Arc magmas are then formed by integrating melts from different parcels of mantle that have experienced different amounts of fluid addition and different extents of melting. This model also produces linear arrays on the U-Th equiline diagram, as a consequence of mixing of melts that have experienced fluid addition and melting over a range of time intervals. [130] Although these two models are physically quite different, parameters can be chosen such that either model can generally fit the observed U series disequilibria in magmas from different subduction zones [Bourdon et al., 2003a]. A critical issue remains the extent of mobility of Th and Pa in fluids, and other constraints may come from a better understanding of the thermal and dynamical regime 35 of 43

36 Figure 23. Sketch cross sections to illustrate the principal features of two melting models for subduction zone magmas: (a) dynamic melting following fluid addition and (b) fluxed ingrowth melting (Figures 23a and 23b taken from Turner et al. [2003a], with permission from Mineralogical Society of America.) (c and d) Predictions for U series disequilibria variations for model shown in Figure 23a as various parameters such as melting rate are varied (assumes D Pa = , D U = 0.003, D Th = 0.002, and D Ra = ) (Figures 23c and 23d taken from Bourdon et al. [2003a]). In Figure 23c the solid lines represent models for different melting rates, at a constant porosity of 3%, while the dashed curve is a model for a melt depleted mantle wedge with D U = and D Th = In Figure 23d the curves illustrate how ( 226 Ra/ 230 Th) and ( 231 Pa/ 235 U) are predicted to vary as a function of melting rate for two different matrix porosity values: Solid curves assume an initial ( 231 Pa/ 235 U) = 0.8 (i.e., prior addition of a U-bearing fluid), while the dashed curves assume an initial ( 231 Pa/ 235 U) = 0.8. (e and f) Predictions for U series disequilibria variations for model shown in Figure 23b as various parameters are varied (assumes D Pa = , D U = , D Th = , D Ra = , melting rate = 0.35 g m 3 yr 1, and critical porosity = 1%) (Figures 23e and 23f taken from Thomas et al. [2002], with permission from Elsevier). Shaded lines denote melting of an enriched mantle (with U = 0.05 ppm and Th = ppm). Solid lines denote melting of a depleted mantle (with U = 0.03 ppm and Th = ppm); tick marks indicate the average extent of melting in percent. within the mantle wedge as the two models also differ in the timescales of melting and the size of the inferred melting region Final Comments on U Series Disequilibria and Melting Processes [131] Interpretation of U series disequilibria data on mantle-derived magmas has been one of the driving forces leading the development of more realistic models for melt 36 of 43 generation and transport in the Earth. Observed variations of U series disequilibria in lavas clearly preserve important information about several physical parameters of the melt generation and transport process (e.g., melting rate, porosity, solid upwelling rate, melt ascent rate, and extent of melt-solid equilibrium during melt ascent). However, any quantitative interpretation of U series disequilibrium data is highly dependent on the exact choice of melting model, as well as the absolute values for partition coefficients