Competing models for arc magmatism make different predictions for the thickness and composition of arc lower crust [e.g., DeBari and Sleep, 1991]. Information on the composition of arc lower crust is also needed to estimate its long-term stability [Jull and Kelemen, 2001; Behn and Kelemen, 2006]. To reconcile the average “andesitic” composition of continental crust with primitive island arc compositions, many models call for foundering of dense mafic-ultramafic cumulates into the underlying mantle [e.g., Arndt and Goldstein, 1989; Kay and Kay, 1993].
 However, constraining the composition of the island arc lower crust and distinguishing high-velocity lower crust from upper mantle rocks is difficult because (1) lower crustal arc sections are poorly represented in obducted sections [Kelemen et al., 2003a and references therein]; (2) the primary constraints on the lower crust and upper mantle in many active arcs are xenoliths and P-wave velocities (VP). It is unclear how representative the former may be, and the latter cannot uniquely distinguish between the effects of composition, temperature and melt. For example, VP of 7.x km/s beneath the Izu-Bonin-Marianas arc are interpreted to represent hot mantle, possibly with melt [Suyehiro et al., 1996] or ultramafic cumulates [Kodaira et al., 2007]. Even in the absence of elevated temperatures and/or melt, VP cannot be used to differentiate between different possible lower crustal compositions [e.g., between garnet bearing and plagioclase-free compositions, Behn and Kelemen, 2003; Müntener and Ulmer, 2006] and/or serpentinized peridotite [e.g., Lizarralde et al., 2002].
 Ambiguity in constraining the composition of the deep parts of island arcs with seismic velocities can be reduced by incorporating information on S-wave velocity (VS) and VP/VS ratios [e.g., Christensen, 1996]. Here, we combine an analysis of sparse S-wave data from the central Aleutian arc and petrologic modeling to better constrain the composition of the lower crust.
1.1 Existing Constraints on Compositions in the Central Aleutian Arc
 Aleutian volcanic rocks exhibit a spectrum of compositions (high-Al basalts, high-Mg basalts, and andesites) and fractionation trends (calc-alkaline and tholeiitic); this compositional diversity has been attributed to variations in fractionation depth, state of stress in the overriding plate, differences in parental magma compositions, and water content [Kay et al., 1982; Myers, 1988; Singer and Myers, 1990; Miller et al., 1992; Sisson and Grove, 1993a; Kelemen et al., 2003b; Zimmer et al., 2010]. These models make different predictions for lower crustal composition. For example, one explanation for the abundance of high-Al basalts is the crystallization of a thick sequence of pyroxenite at depth (possibly due to the presence of water), which would enrich the remaining liquid in Al [Sisson and Grove, 1993a]. The mineral assemblages of lower crustal rocks may also be modified following crystallization by metamorphism, particularly the formation of garnet [Behn and Kelemen, 2006]. The only direct information on the Aleutian lower crust comes from limited xenoliths, many of which are (olivine-) clinopyroxenites [Conrad et al., 1983; DeBari et al., 1987; Yogodzinski and Kelemen, 2007], but it is not clear how representative these are.
 Existing active-source seismic data from the Central Aleutians acquired in 1994 with the R/V Maurice Ewing and onshore/offshore seismometers (Figure 1) indicate relatively high VP in the lower crust of the Aleutian arc [Holbrook et al., 1999; Lizarralde et al., 2002; Shillington et al., 2004; Van Avendonk et al., 2004]. For the lower crust of the oceanic island arc, these range from ~7.0–7.1 km/s directly beneath the active arc [Holbrook et al., 1999] to 7.3–7.6 km/s slightly trenchward [Shillington et al., 2004]. There is a sharp step in velocity at the top of the lower crust (~0.4 km/s), and along-arc variations in lower crustal velocity appear to correlate to variations in lava composition [Shillington et al., 2004]. These characteristics were attributed to mafic/ultramafic cumulates and/or garnet granulites in the lower crust [Shillington et al., 2004], but VP alone cannot distinguish between different possible lower crustal compositions and other explanations, such as partial melt in the subarc mantle or serpentinized mantle in the forearc mantle wedge.
1.2 Analysis of S-wave Arrivals
 To constrain the VP/VS ratio of the deep Aleutian arc crust, we performed a very simple analysis of sparse converted S-wave arrival times from the arc-parallel wide-angle seismic profile acquired in 1994. Seismometers on the Aleutian Islands recorded shots from the 8000 in3 airgun array of the R/V Maurice Ewing, which steamed south of the islands (Figure 1). Thus, the majority of P- and S-wave ray paths in this experiment sampled the arc crust trenchward of the active arc, but were still within the arc platform [Shillington et al., 2004; Van Avendonk et al., 2004].
 We focus our analysis on arrivals from four stations where converted S-wave reflections and refractions were observed at large enough shot-receiver offsets to sample much of the crust (Figure 1). Arrivals occur over source-receiver offsets of 20–180 km and have apparent velocities from ~3 to 4.2 km/s (Figure 2). Consistent with the observation of distinct P-wave reflections and refractions from three laterally continuous layers, we identify three crustal S-wave refractions with distinct apparent velocities; intracrustal and Moho S-wave reflections are also observed (Figure 2 and auxiliary material). Upper crustal arrivals have comparatively 3-D paths due to the experiment geometry, but the longer ray paths of lower crustal refractions and Moho reflections approximately fall in the 2-D plane along the arc platform (Figure 1). Our analysis included 2306 picks; they have large uncertainties (~150–500 ms) because they occur in the coda of the P-wave arrivals. Raytracing tests suggest that P-to-S conversions occurred at the seafloor or at the top of basement beneath a thin veneer of sediments.
 S-wave arrivals were previously identified in this data set by Fliedner and Klemperer , who used travel times in independent 3-D P- and S-wave tomographic inversions. We argue that the paucity of S-wave observations and large uncertainties in travel time picks favor an alternate, simpler analysis approach. We searched for the best-fitting, constant VP/VS ratio for each layer. An S-wave model was calculated from the P-wave model for each of a range of VP/VS ratios. We traced rays through each model in 3-D to produce predicted arrivals times for reflections and refractions, which were used to calculate a RMS misfit. Starting with the upper crust and working down, we found the best-fitting constant VP/VS ratio for each layer. A fixed delay of 1.8 s was used to account for structure beneath the stations; a similar approach was used for the P-wave modeling [Van Avendonk et al., 2004].
1.3 Results of S-wave Modeling
 This approach yielded ranges of best-fitting constant VP/VS for the upper, middle and lower crust along the central Aleutian island arc. Here we focus on results for the lower crust. The RMS misfit curve for S-wave refractions within the lower crust and reflections off the base of the lower crust (i.e., the Moho) shows a clear minimum at a VP/VS of 1.70 (Figure 3). Given the large uncertainties associated with travel time picks of these sparse data and the simple approach taken here, models with VP/VS between ~1.65 and 1.75 are considered acceptable. However, the apparent velocities of the refractions, alone, indicate a higher VP/VS (~1.75). Additionally, the average lower crustal VP/VS based on regional earthquakes indicates a VP/VS of ~1.74–1.77 [Abers, 1994], and higher lower crustal VP/VS are implied in the lower crust directly beneath the active arc by receiver functions at stations along the arc (H. A. Janiszewski et al., 2013, submitted). Thus, we favor the upper end of our acceptable range (1.7–1.75).
1.4 Interpretation of VP/VS
 The new VP/VS results presented here combined with the VP model along the same profile [Shillington et al., 2004; Van Avendonk et al., 2004] provide unique new constraints on island arc lower crust. Below we discuss different possible explanations for our observations.
 Although the range of permissible average VP/VS ratios from our study is large, it immediately excludes many possible explanations for 7.x km/s P-wave velocities in the lower crust and/or upper mantle. If P-wave velocities of 7.3–7.6 km/s were caused by serpentinization of the mantle wedge approaching the forearc, we would expect relatively high VP/VS [e.g., Christensen, 2004, Figure 4]. Likewise, high temperatures and the presence of melt would also increase VP/VS [e.g., Faul and Jackson, 2005]. Anisotropy can also influence the estimation of VP/VS [Hacker and Abers, 2012]. However, for the ray paths in this study and possible mineral assemblages in the lower crust, we infer that anisotropy is unlikely to completely account for the observed low VP/VS.
 In general, the dominant compositional control on VP/VS variations in the crust is silica content; higher silica rocks are generally associated with lower VP/VS [Christensen, 1996, Figure 4]. However, in mafic and ultramafic rocks with low SiO2, other minerals begin to play a role in controlling the velocity characteristics. There are several possible constituent minerals that could be present in the Aleutian lower crust that would result in a relatively low VP/VS (<1.75).
 Pyroxenite can have VP/VS ranging from ~1.68 to 1.85 (Figure 4), depending on the composition of the pyroxenite (orthopyroxene has a lower VP/VS than clinopyroxene) [Behn and Kelemen, 2006]. Many xenoliths from the Aleutians are (olivine-) clinopyroxenites [Conrad et al., 1983; Conrad and Kay, 1984; DeBari et al., 1987; Yogodzinski and Kelemen, 2007]. The estimated VP of these compositions based on Hacker and Abers  (~7.5–7.8 km/s) is at the upper end of the VP range for the lower crust from Shillington et al.  (7.3–7.6 km/s), but the VP/VS ratio (~1.77–1.79) is higher than the values presented here (Figure 4). Thus, another composition must be present in addition to (or instead of) clinopyroxenite.
 Orthopyroxene has a lower VP/VS ratio and could be present due to the breakdown of olivine plus plagioclase to form clinopyroxene, orthopyroxene, and spinel [Kushiro and Yoder, 1966]. Alternatively, metasomatism of olivine-rich rocks by silicious fluids can form orthopyroxene at temperatures above serpentinite stability but below the solidus (~700–1000°C) [Wagner et al., 2008]. Orthopyroxenite could fit our observed VP and VP/VS (Figure 4); however, orthopyroxene is not observed in any of the lower crustal or upper mantle xenoliths from the Aleutians [Conrad et al., 1983; DeBari et al., 1987]. Therefore, although orthopyroxene may be present, we find it unlikely that it forms in sufficient abundances to explain the observed VP/VS ratios.
 Another possible contribution to low VP/VS is the presence of quartz. Quartz is common in felsic and intermediate arc rocks. Its presence in more mafic rocks could occur due to fluxing of silicious material from the slab [Rossi et al., 2006]. Alternatively, the metamorphic reaction of enstatite and plagioclase forms garnet, clinopyroxene and quartz [Kushiro and Yoder, 1966]. The abundance of quartz in the deep Aleutian crust is unknown; Conrad et al.  reported that a gabbroic xenolith from Adak contains quartz. It is also observed in deep crustal rocks from the obducted Kohistan arc [Yamamoto, 1993; Jagoutz and Schmidt, 2012], but is not observed in lower crustal gabbronorites in the Talkeetna section [Kelemen et al., 2003a; Behn and Kelemen, 2006]. The elastic properties of quartz change dramatically with the transition from alpha to beta quartz; alpha quartz has a much lower VP/VS (~1.4) than beta quartz (~1.7) [e.g., Ohno et al., 2006]. The profound effect of the alpha-beta quartz transition is illustrated in Figure 4, which shows VP and VP/VS calculated using Perple_X [Connolly, 2005] for rocks from obducted arc sections in Talkeetna and Kohistan at 0.8 GPa (see auxiliary material). Calculations at 750°C lie within the alpha quartz stability field, and rocks with higher SiO2 trend toward low VP and low VP/VS ratios (Figure 4a). By contrast, velocities calculated at 900°C lie within the beta quartz stability field, and rocks with higher SiO2 trend toward low VP and high VP/VS (Figure 4b). Our rays sample the lower crust trenchward of the active arc line, where colder temperatures are expected, making the stability of alpha quartz more plausible [Shen et al., 1993].
 The sensitivity of the expected mineral assemblages arising from different bulk compositions as a function of temperature and pressure was assessed by examining several possible lower crustal compositions derived from obducted arc sections using Perple_X (see auxiliary material). To satisfy the high VP in the Aleutian lower crust, the presence of quartz, which has low VP, would need to be balanced by other components with higher VP, such as garnet. The pressure-temperature window in which both phases are stable is either nonexistent or very narrow and confined to conditions only present in the lowermost Aleutian crust (Figure S7). Consequently, we conclude that alpha quartz could contribute to the observed velocity properties of some parts of the crust, but cannot be the sole explanation for the low VP/VS ratios over the entire Aleutian lower crust.
 Based on the factors discussed above, it does not appear that a single composition can fully explain the VP and VP/VS of the Aleutian lower crust, but rather a combination of rock types is required. We favor the interpretation that there is abundant (olivine-) clinopyroxenite in the Aleutian lower crust, consistent with Aleutian xenoliths. These compositions have VP that fall within the upper end of the range of VP observed in the lower crust here, but their estimated VP/VS ratios are above the observed range (Figure 4). This requires that other compositions with lower VP and VP/VS must also be present to account for the combination of high VP and low VP/VS. Specifically, we favor mixtures that include compositions with a small amount (<5 wt %) of alpha quartz, such as rocks with ~50–65 wt % SiO2 (Figure 4). Mixtures with ~30–50% alpha-quartz bearing gabbro (VP = 7.1 km/s and VP/VS = 1.72) and ~50–70% clinopyroxenite (VP = 7.6 km/s and VP/VS = 1.775) could account for our observations.