P-wave velocities (Vp) have been measured in the laboratory and calculated using thermodynamic modeling for seven representative rock samples from the lower crust to mantle section of the Kohistan paleo-island arc. Lower crustal rocks comprise plagioclase-rich gabbro, garnet-bearing gabbro, and hornblendite; mantle rocks comprise garnetite, pyroxenite, websterite, and dunite. Measurements were performed at confining pressures up to 0.5 GPa and temperatures up to 1200°C. Vp were also calculated using rock major element chemistry with the Perple_X software package. Calculated Vp match closely the laboratory measurements. At depths representative for the arc root, Vp of upper mantle rocks vary from 7.7–8.1 km/s, whereas the lower crustal rocks have velocities between 6.9–7.5 km/s. P-wave anisotropy is small, with exceptions of sheared gabbros. Measured and calculated seismic properties are consistent with, and complement a growing database of published seismic properties from the Kohistan arc. In the light of such data, we discuss seismic imaging of present-day island arcs. Intermediate Vp (7.4–7.7 km/s) in arc roots can be explained by pyroxenites and garnet-bearing mafic rocks. Strong seismic reflectors may be related to garnetites (8.0–8.2 km/s).
 The geologic nature of seismic properties and seismic reflectors remains an important question to interpret seismic tomography and reflection profiles. Hypotheses for reflective boundaries formulated in the literature generally refer to lithological changes or to the presence of melts and/or pressurized fluids [e.g., Rudnick and Fountain, 1995; Christensen and Mooney, 1995]. Particular attention was paid to ductile shear zones because fabric-related variations in seismic anisotropy can also produce reflections within the same lithology [e.g., Fountain et al., 1984; Barruol and Mainprice, 1993; Rey et al., 1994; Khazanehdari et al., 1998]. At greater depth, the most conspicuous reflection is usually attributed to the Mohorovičić discontinuity (Moho), although petrological descriptions report a zone thicker than the seismic reflective level (see discussion in Rudnick and Fountain, ). In this study, we focus on the seismic structure of island arcs and its relationship to the petrological structure. These regions represent key locations for the creation and destruction of continental crust, due to the interaction of middle to lower crust with the upper mantle [Rudnick and Fountain, 1995]. Recent studies admit the importance of seismic velocity profiles of crust to mantle sections, in particular to clarify seismic imaging of active island arcs [i.e., Suyehiro et al., 1996; Holbrook et al., 1999; Kitamura et al., 2003; Kodaira et al., 2007a, 2007b; Takahashi et al., 2008; Calvert et al., 2008; Brown et al., 2009; Kono et al., 2009; Calvert, 2011; Calvert and McGeary, 2012]. On the one hand, it is imperative to know the physical and chemical signature of the different rocks and fluids of island arc systems. On the other hand, by assigning seismic velocities to specific rock compositions, it is possible to interpret seismic velocities in terms of geological conditions. There are still few systematic studies on sections of the mantle-crust transition zone because such exposures are rare [e.g., Fountain, 1976].
 Two ways to link mineral composition and petrological properties with the elastic properties of rocks are (1) through laboratory measurements, at relevant pressures and temperatures, of rocks collected from exposed sections or drill cores [i.e., Birch, 1960, 1961; Christensen, 1965; Kern and Richter, 1981] and (2) using the chemical and mineral composition of rocks to calculate their elastic properties with thermodynamic modeling [i.e., Connolly, 2005] and effective medium theory [i.e., Voigt, 1928; Reuss, 1929; Hill, 1952; Hashin and Shtrikman, 1962; Mainprice, 1990; Rudnick and Jackson, 1995; Mavko et al., 2009; Hacker and Abers, 2004]. Each of these approaches has been applied effectively to study the seismic structure of island arc systems. Examples of laboratory measurements include Chroston and Simmons  and Miller and Christensen  who performed detailed acoustic velocity measurements on rocks from the exposed Kohistan arc section in Pakistan, using the ultrasonic pulse transmission technique at pressures up to 1 GPa, but at room temperature. These measurements were later complemented by ultrasonic velocity measurements at elevated pressures (up to 1 GPa) as well as temperatures (up to 1000°C) on rocks from the same geological setting [Arbaret and Burg, 2003; Kono et al., 2004; Burlini et al., 2005; Kono et al., 2009]. Behn and Kelemen  used Perple_X on samples collected from the Talkeetna arc section in Alaska to calculate seismic properties of the arc lower crust. Tatsumi et al.  also used thermodynamic modeling to interpret the origin of seismic velocities observed in the Izu-Bonin-Mariana arc crust. Despite obvious potential advantages, combining laboratory measurements and thermodynamic modeling has not yet been attempted.
 For this reason, we collected samples across the well-exposed crust-mantle transition (CMZ) of the Kohistan fossil island arc [Burg et al., 1998, 2005] and measured the P-wave velocity (Vp) in the laboratory at temperatures up to 1200°C and up to 500 MPa confining pressure. Besides the petro-physical characterization of lithologies building the CMZ, we aimed to investigate the effect of temperature on the seismic velocity and reflection record expected at Moho depth. We determined the mineral composition, bulk chemistry, and a description of the microstructures for a complete characterization of the investigated samples. Seismic velocities were further calculated using major element compositions with the Perple_X software [Connolly, 2005]. Results from both measurements and calculation have been compiled to obtain, along with previously published data, a complete record of the seismic P-wave velocities and their pressure and temperature derivatives. We do not consider the possible influence of partial melting on seismic properties. Although the presence of melt beneath or in the arc root may have a significant influence on seismic velocities, the focus is on the contribution of rocks in the absence of melt. Using the Kohistan island arc as a type setting for the lower crustal to mantle transition, we discuss the geological origin of seismic velocity profiles and reflection records in present-day island arc settings.
2 Geological Setting
 The Kohistan Complex, in NW Pakistan, separates the Indian and Asian plates. It was developed as an island arc above a north-dipping subduction in the Tethys Ocean during the Mesozoic [Tahirkheli et al., 1979; Bard et al., 1980; Bard, 1983; Coward et al., 1987]. Its tectonic history involves igneous growth since 154 Ma (subduction initiation), intra-arc rifting at about 85 Ma, accretion to the Asian plate, to the north, and closure of the Tethys Ocean to the south at ca.55 Ma [Burg, 2011 and references therein]. The southern, lower boundary of the Kohistan Complex is a northward dipping fault zone, the Indus Suture, along which the Kohistan Arc Complex has been obducted over continental India (Figure 1). The “magmato-stratigraphic” build-up of the Kohistan Arc Complex is a primary feature that predates the India--Asia collision [e.g., Treloar et al., 1996; Searle et al., 1999]. The generally north-dipping Jijal Complex, in the direct hanging wall of the Indus Suture, includes the arc crust-mantle boundary with mantle peridotites, pyroxenites, garnetites, and hornblendites overlain by lower-crustal garnet-bearing granulite facies gabbros [Jan and Howie, 1981; Jan and Windley, 1990; Burg et al., 1998; Ringuette et al., 1999]. These metagabbros are in turn overlain by the so-called Southern Amphibolites, which mostly are strongly deformed, amphibolite facies metagabbros and metadiorites [Burg et al., 2005; Burg, 2011]. We first briefly describe the lithologies of this mantle-lower crust transition zone, from the bottom to the top, i.e., from south to north (Figure 1).
 The Jijal mantle section is dominated by dunites and websterites interlayered with subordinate pyroxenites and rare harzburgites [Jan and Howie, 1981; Jan and Windley, 1990; Miller et al., 1991; Burg et al., 1998; Figure 1]. The proportion of websterites increases up-section, whereas garnetites and hornblendites are particularly abundant at the transition with overlying garnet granulites. Trails of chromite and rotated clinopyroxenite dykes delineate the compositional layering parallel to the general mineral foliation. Dykes of green Cr-rich clinopyroxenites locally build an angular framework in the dunites, and flames of dunite are frequently found in websterites. These features are compatible with reactive magma transport in the refractory sub-arc mantle. Therefore, these ultramafic rocks are demonstrably part of the residual “arc mantle”, with clinopyroxene crystallization ages of 117 ± 7.0 Ma constrained by a Sm-Nd isochron constructed from several rocks from the Jijal ultramafic section [Dhuime et al., 2007].
2.2 Lower Crust
 The granulite facies, garnet-rich metagabbro is the deepest “crustal” unit of the Jijal Complex [Yamamoto1993; Ringuette et al. 1998]. Its bottom is marked by a sharp contact that dips gently to the north, parallel to the mantle “layering” (Figure 1). The metagabbro is intrusive into underlying hornblendites and garnetites of the mantle sequence and, in turn, has been intruded by hornblendites (Figure 1). The bottom boundary of the metagabbro is interpreted as the petrological arc Moho [Burg et al., 1998]. Xenoliths of peridotites found up to Patan (Figure 1) support the interpretation that the metagabbro has intruded mantle rocks and justify the geological description of a crust-mantle transition zone. The intrusion age of the granulite facies, garnet-gabbro has been estimated from 91 ± 6.3 Ma to 95.7 ± 2.7 Ma (Sm-Nd cooling ages based on pyroxene-garnet assemblages; Yamamoto and Nakamura ; Anczkiewicz and Vance ; Schaltegger et al. ). The upper contact of the granulite facies metagabbro marks the boundary with the Southern Amphibolites, which represent the “deep to mid-crustal” section of the arc [Bard, 1983; Treloar and Rex, 1990; Khan et al., 1993, Yoshino et al., 1998].
2.3 Lower Middle Crust
 At the bottom of the Southern Amphibolites, the 98.9 ± 0.4 Ma Sarangar gabbro (206Pb/238U zircon age; Schaltegger et al. ) crystallized under granulite facies conditions at 800°C and 0.8–1.1 GPa [Yoshino et al., 1998]. It is composed of garnet-poor to garnet-free gabbros showing the development of cm-scale mylonitic shear bands [Arbaret et al., 2000; Arbaret and Burg, 2003; Burg et al., 2005]. The stable metamorphic assemblage in shear zones is composed of fine-grained pargasitic to tchermakitic amphiboles, epidote, plagioclase, quartz, and poikiloblastic garnet [Burlini et al., 2005]. Partial melting generated quartz-plagioclase-garnet bearing segregation veins pointing to temperatures exceeding 750°C under pressures >0.8 GPa.
 The overlying sequence is mainly composed of amphibolite facies metagabbros, hornblende-gabbros, diorites, and subordinate tonalites (Figure 1). They have locally preserved igneous layering, have intruded each other without any identified logic, and have been intruded by small volumes of hornblendite pegmatoids and plagioclase-quartz, with or without amphibole pegmatite veins. The petrography and geochemistry of the plutonic rocks have been portrayed by Jan and Howie , Treloar and Rex, , Miller et al. , Yamamoto , and Yoshino et al., . All authors agree that they represent calc-alkaline magmas emplaced during the arc activity from a partially molten mantle source with mid-ocean ridge basalt-type isotopic characteristics [Schaltegger et al., 2002]. All published ages for the lower middle crust (excluding the Kohistan batholith) are distributed between 75 and 110 Ma [Burg, 2011].
3 Sample Description
 The samples examined in this study are listed in Table 1, which contains information on sample location, experimental conditions, and sample density. The chemical and the modal mineral compositions are summarized in Tables 2 and 3, respectively. Bulk rock chemical composition was determined by X-ray fluorescence, whereas the modal mineral composition was quantified by grain-counting in thin sections.
 The sample is mainly composed of garnet and clinopyroxene with minor orthopyroxene and amphibole. The grain size is coarse, ranging from 0.5 to 3 mm. Garnet boundaries are rectilinear to slightly curved and the structure is well equilibrated; no undulose extinction is observed for pyroxenes (Figure 2a). There is almost no sign of reaction or retrogression, except for some amphibole grains crystallized at the rim of clinopyroxene.
3.1.2 Coarse Grained Pyroxenite (Samples K19 and K11)
 These are massive, relatively coarse grained (0.5 to 5 mm) bright green pyroxenites. They both contain clino- and orthopyroxene and minor olivine. Sample K19 is partially serpentinized along grain boundaries, and sample K11 contains, in addition, some plagioclase partly altered to epidote. Clinopyroxenes show exsolution lamellae in both samples. The microstructures (Figures 2b and 2c) are thermally equilibrated (e.g., triple junctions with corner angles of 120°); the rocks are isotropic and both pyroxenes display weak undulose extinction in thin section.
3.1.3 Websterite (Sample K14)
 This sample is massive, almost black and isotropic in hand specimen. Its microstructure is very similar to that of the K19 pyroxenite, with less serpentine at grain boundaries (Figure 2d). The grain size is coarse (1 to 5 mm), with curved grain boundaries and triple junctions. The mineral composition consists of clino- and orthopyroxene and opaques. Olivine is sporadic. No late ductile or brittle deformation feature is visible in thin section, and the rock preserves a thermally equilibrated texture.
3.1.4 Dunite (Sample K15)
 The dunite is mainly composed of olivine. No alteration or serpentinization is visible. The grain size, 0.5 to >5mm, is relatively coarse. Triple junctions provide an indication of equilibrium microstructures. The larger grains show deformation kink bands, deformation lamellae, and undulose extinction representing late and weak solid-state deformation (Figure 2e).
3.2 Sarangar Gabbro
3.2.1 Unsheared to Sheared Gabbro (Samples PR90, GR90, and MY90)
 Samples PR90, GR90, and MY90 come from the hanging wall gradient of a gently NE-dipping, 0.5 m thick shear zone in the lower mid-crustal Sarangar Gabbro (Figures 1, 3a–3d; Table 1); the sense of shear is top-to-SW. Samples represent the undeformed protolith (PR90) and the central mylonite (MY90). Three mutually orthogonal cores (15 mm in diameter and 30 mm long) were drilled (i) parallel to the foliation and lineation (X axis), (ii) in the foliation plane but perpendicular to the lineation (Y axis), and (iii) normal to the foliation plane (Z axis) in order to determine the directional dependence of the seismic wave velocities with respect to the foliation and lineation. Note that images presented in Figures 3a–3d are not oriented with respect to the structural axes (X, Y, and Z), but are rather shown as indication of mineral composition and typical microstructure.
3.2.2 Garnet-Bearing Gabbro (Sample K7)
 The lower crustal, granulite facies gabbro is mainly composed of clino- and orthopyroxene, garnet, plagioclase, quartz, rutile, and minor oxides. The grain size ranges from 0.5 to 2 mm (Figure 3e). Equilibration under granulite facies conditions has occurred, with allotriomorphic quartz, plagioclase and pyroxenes, and subidiomorphic garnet; rutile is idiomorphic. Thin sections display a weak foliation but no evidence for cataclastic deformation and mineral reactions. Garnet and clinopyroxene have microcracks.
3.2.3 Hornblendite (sample K18)
 This rock is composed of more than 80% dark green hornblende (Figure 3f). Amphibole grains are coarse, often exceeding 5 mm. Other minerals are epidote, sphene, opaques, and seldom chlorite. Hornblende is subipidiomorphic, and epidote often occurs along amphibole boundaries, with allotriomorphic shapes. In thin section, there is no evidence for ductile or brittle deformation and for mineral reaction or disequilibrium. Within grain microcracks and a possible shape preferred orientation of epidote and hornblende have not been quantified.
4 Measurement Technique
4.1 Assembly and Calibration
 The velocity of longitudinal waves was measured using the ultrasonic pulse transmission technique on cylindrical samples placed in an internally heated gas medium apparatus (Figure 4a) [Birch, 1960; Christensen, 1965; Burlini et al., 2005; Ferri et al., 2007; Caricchi et al., 2008]. The piezo-crystals have a resonance frequency of 1 MHz. Since the transducers cannot withstand temperatures >200°C, a buffer rod of zirconia-ceramic was introduced between the sample and the tips of each transducer. The buffer rods introduce delays that must be subtracted from the experimental measurement of time of flight. Since buffer rods also change properties with pressure and temperature, a full calibration was required. For this purpose, a single crystal of sapphire cut parallel to the  axis was inserted in place of the sample. Its elastic constants and their pressure and temperature derivatives are from Gieske and Barsch . From these, we could calculate the calibration time (tc) at any pressure and temperature:
 where ta is the measured arrival time and Vsa are the elastic wave velocities of the sapphire at specified pressures and temperatures (for example 500 MPa, 1100°C), and ls is the length of the sapphire crystal. Pressurization effectively decreases the sample volume. Conversely, temperature causes sample expansion. Therefore, changes in sample length during the experiments are altogether negligible and not considerable for correction [Burlini et al., 2005]. More importantly, the sample length and the length of the sapphire crystal were kept as similar as possible to maintain a comparable thermal profile during experiments. The velocities (vr) of the rock samples of length (ls) are given by
 Our samples had typically ~30 mm length and 15 mm diameter. This can be compared with Kono et al. [2004, 2009] who used 6.0 mm long samples with a 5.7 mm diameter. We also performed measurements at lower frequencies than Kono et al. (0.1–1 MHz instead of 8 MHz), which permitted to avoid the critical problem of measuring an elastic wavelength of the same order or smaller than the grain-size of the rock.
4.2 Effect of Frequency in a Cylindrical Rod Assembly
 In a long cylindrical rod, where the length of the cylinder greatly exceeds its diameter (more than five times), the propagation mode of ultrasonic waves depends on their frequency [Birch, 1960]. Measurements were performed at 0.05–0.1 MHz and 1 MHz in order to investigate possible dispersion in the ultrasonic velocity. We attempted measurements at 3 MHz; but at this frequency, the ultrasonic wave was usually too attenuated to easily determine the first arrival in the waveform. Typical unfiltered waveforms of a single sapphire crystal at measurement frequencies are shown in Figure 4b. Arrival times were picked manually, because (1) the different frequencies we used resulted in rather different waveforms, which could be problematic for automatic picking routines, and (2) the number of traces to be picked was relatively small and generally first arrival could be clearly distinguished. The uncertainty described for each frequency is based mainly on the sharpness of the first break in the waveform (Figure 4b).
 Measurements at 0.05–0.1 MHz were affected by a large uncertainty (typically Δt = 1 µs), which means that the error is ~5% on a 30 mm long sample with a velocity of ~8.0 km/s. However, within this Δt, the first break in the ultrasonic signal can be clearly identified. In terms of velocity, this uncertainty amounts to about 0.4 km/s. The signal was treated with a 1.4 MHz Nyquist filter to remove a high frequency signal that becomes increasingly apparent at higher confining pressure. The origin of this signal is likely related to the natural vibration frequency of the transducers, which is 1 MHz.
 At 1 MHz, it was sometimes difficult to pick the first break in the waveform, because the beginning of the seismic signal is in part obscured by low amplitude oscillations (Figure 4b). Nevertheless, for most samples, it was possible to determine the P-wave velocity from the 1 MHz frequency measurements. At these conditions, the uncertainty was about 0.5 µs, which yields an error of nearly 2% for a velocity of ~8.0 km/s (e.g., ~0.2 km/s uncertainty). It is generally not possible to distinguish a difference between first arrival in 0.1 and 1 MHz generated ultrasonic waves, which indicates that mainly body/compressional waves (P) propagate through the assembly, despite the high length to diameter ratio in the column.
4.3 Temperature—Pressure Conditions
 Temperature was measured with two K-type thermocouples (with a precision of ±1 K) placed directly on the top and bottom of the samples. The temperature difference between the two thermocouples at 600°C was <5°C and <20°C at 1200°C. Confining pressure was measured with a manganin coil pressure transducer with a precision of about 1 MPa at 0.5 GPa. The elastic wave velocities were measured while increasing confining pressure by generally 0.05 GPa increments until about 0.5 GPa, and again while decreasing confining pressure by 0.05 GPa decrements. From these dual measurements along pressurization and depressurization paths, we could extract the P-wave pressure derivative (∂Vp/∂P). After the pressure cycle, both pressure and temperature were raised by steps of 0.1 GPa and 100°C until the maximum pressure, generally 500 MPa, was reached (at this pressure, the temperature was typically ~500°C); beyond that point, only temperature was raised in steps of 100°C until between 600 and 1200°C, depending on the experiment (Table 1). P-wave temperature derivates were obtained after velocities had been corrected for the effect of pressure (i.e., using the pressure derivative obtained during the first part of the experiment). The pressure release path to room conditions first followed decreasing temperature in steps of 100°C down to room temperature, keeping the pressure at 0.5 GPa, and then the pressure was released by steps of about 0.05 GPa. Using this specific PT path avoided extensive thermal cracking of the samples. Moreover, the two different paths, up P then T and down T then P, allowed a better estimation of the P and T derivatives.
5 Calculations of P-Wave Velocity
5.1 Voigt and Reuss Averages to Calculate Vp
 It is possible to calculate the seismic velocities, knowing the modal mineral composition, and the individual mineral elastic moduli and densities. One of the simplest, yet effective methods to predict seismic velocities in rocks, is using the Voigt and Reuss bounds. The Voigt bound [Voigt, 1928] is known as the iso-strain bound, representing the upper elastic boundary of the rock. It is calculated by
where M is the Voigt elastic modulus, fi is the volume fraction of phase i, and Mi is the elastic modulus of phase i (i.e., bulk and shear moduli). The Reuss bound [Reuss, 1929], also known as the iso-stress bound, represents the lower limit of the elastic modulus. It is calculated as
 From mineral compositions and respective volume fractions of each studied rock type (Table 3), the isotropic Vp was calculated as Vp = ([K + 4/3*μ]/ρ)1/2, where K is the bulk modulus, μ the shear modulus, and ρ is the density. The elastic moduli used in the calculation are from the compilation of Bass .
5.2 Vp and P Calculations with Perple_X
 The Perple_X software package and codes [Connolly and Kerrich, 2002; Connolly, 2005] allow calculating seismic wave velocities at any pressure and temperature using classical thermodynamics equations, databases, solution models, and knowing the bulk composition of the system considered. A fixed-composition phase diagram for the considered bulk composition must be constructed before calculating the corresponding seismic properties. The determination of the most stable mineral assemblage at any given P-T condition is realized with the free Gibbs energy minimization algorithm [Connolly and Kerrich, 2002; Connolly, 2009] for progressively increasing P-T resolution. The P-T grid (700–1000°C, 0.6–1.4 GPa) was subdivided into smaller PT cells, and the solution phases subdivided into fixed-composition pseudo-compounds. The final iteration computed the most stable assemblages on 237 × 237 cells.
 We used the thermodynamic database of Holland and Powell [1998, updated in November 2002]. Solution models for heterogeneous phases are from Powell and Holland  for pyroxenes and olivine, Newton et al.  for plagioclase, White et al.  for garnet, White et al.  for spinel, and White et al.  for ilmenite. The calculated output contains the modal proportion of each mineral phase and the composition of all stable solution phases for all cells in the resulting phase diagram (Figure 5a).
 The density (ρ) of each mineral phase was computed from the pressure derivative of the molar Gibbs energy (= molar volume) and their respective molar mass [see Connolly and Kerrich, 2002]. Aggregate density was then estimated by summing densities of single phases weighted by their volumetric proportions (Figure 5b).
 The adiabatic bulk modulus (Ks) was expressed as a function of first and second order derivatives of the free Gibbs energy [Connolly and Kerrich, 2002]:
where G is the molar free Gibbs energy, P the pressure, and T the temperature.
 The shear modulus μ was estimated from the computed bulk modulus and the Poisson ratio (σ) with the following relation:
 According to the compilation of Christensen , a value of 0.28 was retained for the Poisson ratio of mafic rocks (gabbro, mylonite, and granulitic gabbro) while 0.26 was used for pyroxenites and dunite.
 The P-wave velocity (Vp) was calculated according to the following expression:
 The seismic properties of the aggregate (Figure5c) were computed using a Voigt–Reuss–Hill averaging of velocities for all stable phases weighted by their respective volume proportions.
 Calculations were not performed for the hornblendite mainly because thermodynamic modeling of amphibole is complex, especially at high temperature [Dale et al., 2005]. A pure hornblendite assemblage as observed in the Jijal section is not stable in the models. Therefore, water was excluded for all calculations. Amphibole is present as a retrograde phase in some samples (<10 vol% for most samples but up to 40 vol% in the mylonitic gabbro). Its exclusion has no major consequence on calculated seismic properties, since amphibole-rich rocks have almost the same Vp values than anhydrous granulites at 1 GPa [Christensen, 1996].
 Calculations with the Holland and Powell  database generate non-physical (i.e., negative) values of the isobaric heat capacity near the Landau transition from α-quartz to β-quartz. This can lead to inconsistent seismic velocity of quartz and to a large decrease of the aggregate seismic velocity around the phase transition. In order to avoid this artifact, α-quartz was taken as the only stable SiO2 polymorph in the whole PT range. Quartz is only stable in low proportion (<10 vol%) in three fixed-composition phase diagrams (for the protolith gabbro, the mylonite, and the granulitic gabbro). Our simplification induces a deviation <0.05 km/s for Vp, which is negligible for the purpose of this study.
6.1 Room Temperature Measurements and Pressure Derivatives
 Experimentally-determined seismic velocities for mantle and crustal rocks are shown in Figures 6a, 6c, 6e and 6g and Figures 7a 7c, 7e, and 7g, respectively. The nonlinear part of the confining pressure-velocity curves at room temperature and at confining pressure ≤100 MPa is attributed to crack closure. The linear portion, at higher pressure, reflects the intrinsic seismic properties in the selected direction of the rock under consideration [Birch, 1961]. The velocity measured during pressurization is consistently lower than that measured during depressurization because some of the cracks and pores do not immediately reopen during depressurization, i.e., a hysteresis effect. According to Burke , measurements made during pressurization are not reproducible, whereas those during depressurization are reproducible within measurement error limits. Therefore, the most reliable experimental data are those obtained during decompression.
 Room pressure velocities were recalculated using a linear regression of the data measured between 0.15 and 0.5 GPa (Table 4). The slope of this linear regression represents the pressure derivative (∂Vp/∂P), and the intercept is referred to as Vp0 (i.e., Vp at zero pressure). Vp measurements at high confining pressure varied from 6.9 to 8.8 km/s in the mantle rocks, with the exception of the partially retrogressed pyroxenite (K11) whose Vp was about 6.6 km/s (Figure 7). At ~0.4 GPa, confining pressure samples from the lower and middle crust have Vp that ranges from ~7.0 km/s to 8.5 km/s. Pressure derivatives are within the range of the measurements made by previous authors on Kohistan crustal and mantle rocks (e.g., around 10−3–10−4 km s−1 MPa−1; Table 4; Miller and Christensen ).The uncertainty has nearly the same size as the symbol used in the Figures, at most ~0.4 km/s, and therefore not readily visible in the diagrams.
Table 4. Vp at Room Pressure and Temperature, and Seismic Pressure and Temperature Derivativesa
Perple_X (at 1.0 Gpa, 800°C)
Vp (Room P and T, km/s)
∂Vp/∂P (km s-1 Mpa-1)
∂Vp/∂T (km s-1 °C-1)
∂Vp/∂P* (km s-1 Mpa-1)
∂Vp/∂T* (km s-1 °C-1
The seismic derivates from the Perple_X thermodynamic modeling.
Sarangar Gabbro (PR90)
Mylonite (MY90: Z)
Mylonite (MY90: X)
Garnet Gabbro (K7)
Average lower crust
6.2 High Temperature Measurements and Temperature Derivatives
 Increasing temperature typically reduces the bulk seismic velocity in a rock. The reduction of elastic constants is more effective than the reduction of density, unless a phase transition or mineral reaction occurs. The dependence of Vp upon temperature is reported in Figures 6b, 6d, 6f, and 6h for the mantle rocks and in Figures 7b, 7d, 7f, and 7h for the lower and mid crustal rocks. Temperature derivatives are summarized in Table 4 and range from −2.32 × 10−3 to −0.32 × 10−3 km s−1°C−1 for the mantle rocks. For the crustal rocks, the temperature derivatives range from −1.93 × 10−3 to −0.68 × 10−3 km s−1°C−1 and are generally higher than those of mantle rocks.
6.3 Voigt and Reuss Bounds
 The results from Voigt and Reuss calculations, for room pressure and temperature conditions, are shown in Figure 8. Experimental results, extrapolated to room pressure and temperature conditions (Vp0), are additionally shown with respective uncertainties. The distance between Voigt upper and Reuss lower bounds depends on the number of phases that compose the rock and the respective elastic moduli. A large difference in the bounds arises when the rock is composed of several phases with strongly contrasting elastic moduli. This is particularly the case of the garnet bearing gabbro (K7). In contrast, the dunite (K15) and websterite (K14) display almost negligible differences in Voigt and Reuss Vp bounds. Laboratory measured Vp0 generally fall within or close to the Voigt-Reuss bounds; however, three specimens deviate from the predicted Vp0: (i) the hornblendite (K18) has predicted Vp ~0.5 km/s faster than measured values; (ii) the dunite (K15) has measured P-wave velocity ~0.3 km/s lower than predicted value, and (iii) the partially serpentinized specimen, K11, has measured velocity ~1 km/s slower than predicted.
6.4 Thermodynamic Modeling With Perplex and Vp Calculations
 An advantage of the thermodynamically-based Vp calculations is that they concern mineral assemblages for a given bulk composition under given pressure and temperature conditions. We can thus extract the physical properties of the rocks at their equilibrium condition in the arc root before exhumation or hydration. This is currently not possible with laboratory measurements for technical reasons. A drawback of modeled Vp is that the chemical system is simplified compared to nature so that calculated parameters likely deviate from actual sample properties. The phase diagram for all samples, excluding hornblendite, is presented in an auxiliary figure (online supporting information) together with values of Vp, density, pressure, and temperature derivatives of Vp over the whole P-T range (700–1000°C, 0.6–1.4 GPa).
 The P-T conditions at which the measured samples (re)crystallized are given in the literature. Maximum conditions estimated for the Sarangar protolithic gabbro crystallized at 800°C, 0.8 to 1.0 GPa [Yoshino et al., 1998; Burlini et al., 2005]. Garnet is not stable at 800°C and 0.8 GPa for the protolith gabbro (see supporting information), but it is present in minor amount in the PR sample. A pressure of 0.9 GPa, under which garnet becomes stable, is thus more relevant. Pressure estimations for the garnet-bearing granulitic gabbro K7 span the large interval 0.8 to 2.2 GPa [Bard, 1983; Yoshino et al., 1998; Ringuette et al., 1999; Zeilinger, 2002]. Pressure values (1.4 to 2.2 GPa) published by Yoshino et al.  and Ringuette et al.  are overestimated compared to the actual thickness of the Jijal section (~6 km) below the 0.9–1.0 GPa Sarangar gabbro, assuming that 3 km depth correspond to 0.1 GPa and knowing that no major tectonic discontinuity is mapped in the Jijal section. This barometric difference was explained by the preservation of igneous garnet crystallized in the mantle during magma ascent [Ringuette et al., 1999]. With reference to the Sarangar gabbro, the garnet granulite and the garnetite were at a pressure of 1.0–1.1 GPa, whereas the pyroxenite, the websterite, and the dunite equilibrated around 1.1–1.2 GPa. Temperature for the granulitic gabbro and the garnetite was fixed at 850°C according to calculations by Bard  and Yoshino et al. . The only temperature estimation (900°C) for the ultramafic samples is from Yoshino et al. . Whatever the petrological uncertainty, choosing different equilibrium conditions (i.e., 800–900°C and 1.0–1.4 GPa; see supporting information) does not significantly modify seismic velocities.
 Dunite (olivine-spinel), websterite (olivine-clinopyroxene-orthopyroxene), and pyroxenite (olivine-clinopyroxene-orthopyroxene) have mineral assemblages stable over the whole investigated P-T range. Dunite has the highest Vp at the P-T conditions corresponding to the lowermost Jijal section: 8.10 km/s for a density of 3270 kg/m3. Websterite and pyroxenite have similar properties: 7.74 km/s–3280 kg/m3 and 7.83 km/s–3270 kg/m3, respectively.
 The phase diagram for mafic rocks and garnetite shows several stability fields (Figure 5, and supporting information). The physical properties vary largely for a single bulk composition, and a strong increase in Vp and density appears to be related to an increase in garnet volumetric proportions. The modeled garnetite has the garnet-clinopyroxene-orthopyroxene assemblage stable over most of the considered P-T domain. It has high density and P-wave velocity at 850°C, 1.0 GPa: 8.21 km/s and 3560 kg/m3 (online supporting information). However, garnet is consumed to form plagioclase at P < 0.6 GPa and T < 950°C. As a consequence, Vp and density values drop to 7.4 km/s and 3175 kg/m3, respectively.
 The granulitic (K7), the undeformed (PR), and the mylonitic (MY) Sarangar gabbros are characterized by similar phase assemblage evolutions with changing P-T conditions. They show a high-T, low-P domain where garnet is not stable (PR and MY) and seldom present or absent (K7). The consumption of orthopyroxene to form garnet (see complete stoichiometric reaction in Yoshino et al. ) in response to P increase and/or slight decrease in T, increases Vp and density. For the granulitic gabbro, the P-T equilibration conditions (850°C, 1.0 GPa) correspond to a field where garnet is stable and abundant (plagioclase-garnet-clinopyroxene-quartz-rutile). The corresponding Vp and density are 7.36 km/s and 3150 kg/m3, respectively. The P-T conditions of the mylonitic and undeformed gabbros correspond to the plagioclase-garnet-orthopyroxene-clinopyroxene-ilmenite stability field, where volumetric proportions of garnet are increasing steadily with increasing pressure. Consequently, the physical properties, especially Vp, at 0.9 GPa: 7.08 km/s–3010 kg/m3 contrast with those at 1.0 GPa: 7.26 km/s and 3090 kg/m3.
6.5 Measured and Thermodynamically Derived P and T Derivatives
 Thermodynamic Vp are dependent on phase changes, at different P and T, whereas measurements are not. In the latter, kinetics of reactions are too slow to enhance phase changes at measurement timescales. A comparison of laboratory-derived velocities with those derived from thermodynamic modeling is useful to test the reliability of each method. Thermodynamically derived pressure derivatives are typically one order of magnitude higher than those obtained from measurements. This discrepancy may in part be explained by the microstructure of natural rocks, which is not included in modeled rock. Microfractures and grain-boundaries act as discontinuities for a seismic wave that reduces the bulk velocity and increases laboratory pressure derivatives, particularly at low pressures in laboratory measurements. When a rock is pressurized, the grain to grain contact increases and consequently velocity increases. This effect, which does not originate from the compressibility of the minerals themselves, can substantially influence the measured pressure derivative; it is not taken into consideration in the thermodynamic modeling. Measured and modeled temperature derivatives are typically between −10–3 and −10−4 km s−1 °C−1. Modeled temperature derivatives show less variability than measured derivates.
7 Comparison With Previously Published Laboratory Seismic Properties
7.1 Ultrasonic Velocities and P, T Derivatives
 The laboratory pressure and temperature derivatives presented in this study generally match previously published Vp measurements on Kohistan rocks. Temperature derivatives are comparable to those reported in Kono et al. [2004, 2009], although they are higher for some samples (Figure 9). The difference is in part attributed to differences in experimental setups. For example, Kono et al. [2004, 2009] invoked thermal cracking at experimental confining pressure ≤0.5 GPa in order to explain temperature derivatives that were in the order of 2 × 10−3 − 3 × 10−3 km s−1 °C−1. Thermal expansion and cracking is promoted when ratio temperature/confining pressure is >1°C/MPa [Fredrich and Wong, 1986]. Therefore, thermal cracking may have influenced the particularly high temperature derivative of garnetite (K16) and the somewhat high temperature derivatives of the garnet gabbro (K7), the hornblendite (K18), and the websterite (K14) with respect to other samples (Figure 9).
 Pressure and temperature derivatives for Vp of the Kohistan crustal and mantle rocks are within one order of magnitude of values reported in the literature for similar rocks [e.g., Christensen, 1979; Kern and Tubia, 1993; Kern et al., 1999, 2002; Wang et al., 2009]. The pressure and temperature derivatives listed in Table 4 and shown in Figure 9 will be used to estimate the seismic P-wave velocities at increased pressure and temperature (section 8). This procedure serves as a validation of laboratory results, as well as the thermodynamic modeling.
7.2 Acoustic Impedance
 Figure 10 compiles the published density and Vp of Kohistan rocks [Chrostonand Simmons 1989; Miller and Christensen1994]. The measurements were performed at 0.4–0.7 GPa confining pressure and room temperature. Acoustic impedances, Z (= Vp * ρ), are higher than 20 km s−1 g cm−3 for the Kohistan samples. Crustal gabbros and pyroxenite have Z = 20–22.5. Mantle lithologies (dunite, websterite, and garnetite) have Z ~30. The transition from crust to mantle should present a gradual increase in impedance.
7.3 Seismic Anisotropy
 The studied mantle rocks display no macroscopic anisotropy because they have recrystallized during pervasive reactions with arc-related fluids and magmas [e.g., Burg et al., 1998; Garrido et al., 2007]. The lack of textural anisotropy was confirmed by Kono et al.  after calculating the seismic anisotropy from electron backscatter diffraction measurements on plagioclase, orthopyroxene, clinopyroxene, and garnet. A weak seismic anisotropy seems to be a distinctive characteristic of arc mantles compared to continental and oceanic mantles, and this is likely attributable to magma/fluid/rock reactions in arcs.
 In contrast, some lower crustal rocks such as gabbros that contain shear zones display P-wave anisotropy from 5 to 7% [Arbaret and Burg, 2003; Burlini et al., 2005; Kono et al., 2009]. The intrinsic seismic anisotropy measured from the three measurement directions (X, Y, and Z) is defined as A = 100% * ([Vpmax − Vpmin]/Vpmean). Measured seismic anisotropy is smallest (<2%) in the PR90 isotropic gabbro and significantly greater in the mylonite (MY90; 5.6–6.4%). The crystallographic preferred orientation of plagioclase and amphibole likely causes the anisotropy in the mylonite [Barruol and Mainprice, 1993; Burlini et al., 2005; Tatham et al., 2008]. Burlini et al.  showed that the seismic anisotropy in the Kohistan shear zones is not strong enough to produce seismic reflections, even if shear zones are thick enough (i.e., mylonite thickness >¼ of the wavelength of the seismic waves, Sheriff and Geldart , 50 to 100 m for standard seismic surveys). Therefore, the wave velocity contrast of the arc “Moho” can be mainly attributed to the lithological contrast.
8 Insights Into the Seismic Velocity Structure and Reflectivity in Oceanic Arcs
 The main reason for conducting seismic surveys in active intra-oceanic island arcs is to elucidate their structure and composition. For this purpose, specific velocities tend to be translated into given rock compositions [i.e., Tatsumi et al., 2008]. The Kohistan paleo-island arc represents a rather thick arc, with a lower crustal arc root deeper than 30 km [Yoshino et al., 1998; Ringuette et al., 1999]. The lower crust of modern thick arcs, like the Aleutian and the northern Izu-Bonin arcs, can be up to 15 km thick [Miller and Christensen, 1994; Holbrook et al., 1999; Shillington et al., 2004; Kodaira et al., 2007a, 2007b; Calvert, 2011; Calvert and McGeary, 2012]. Shallow arc roots at depths <25 km are found in the southern Izu-Bonin-Mariana arc system, where the lower crust is >10 km thick [Takahashi et al., 2008]. Laboratory measurements and modeling results for the Kohistan arc can be used to gain insight into the lithological structures of the above-mentioned modern-day arc roots.
 We constructed a synthetic seismic profile of the lower crust-mantle transition zone based on our laboratory measurements and thermodynamic modeling, with pressure-temperature estimations adopted from section 6.4. This profile (Figure 11) shows Vp, density, and the seismic reflection coefficient (|R|). The velocity distribution in both the upper mantle and lower crustal sections yields three distinct domains: (1) 6.9–7.0 km/s (Sarangar gabbro), (2) 7.4–7.7 km/s (garnet granulite and pyroxenite), and (3) 8.0–8.1 km/s (dunite). The highest velocity (8.0–8.2 km/s) corresponds to the garnetite layer. The velocity profile across the Kohistan mantle-crust transition zone emphasizes the roles of plagioclase in lowering bulk velocities, and garnet and pyroxenites in increasing Vp from crust to residual mantle.
 The density profile (Figure 11b) was obtained from thermodynamic modeling, with values ranging from 3200 to 3600 kg/m3. Garnetite stands out with significantly higher density than other samples. Within the crustal section, garnet in the Jijal granulite is responsible for the density increase by about 150 kg/m3 with respect to the Sarangar gabbro.
 The seismic reflection coefficient was calculated as |R| = |(Z1 − Z0)/(Z1 + Z0)|, where Z1 and Z0 represent the acoustic impedance of two juxtaposed lithological units (i.e., Sarangar gabbro and Jijal granulites, Figure 11c). Strong seismic reflections are expected at the boundary from garnet granulites to garnetite (R ≥ 0.1), and at the boundary between garnetite and mantle websterite (R = 0.07–0.1). Weaker reflections can be expected at the transitions Sarangar gabbro to Jijal granulites and websterite to dunite. According to the criteria of Warner , |R| must be ≥0.1 in the deep crust to yield seismically visible reflections. This condition is fulfilled only for the garnetite, making this rock the likely source for visible seismic reflections in mantle-crust transition zones.
8.1 Role of Garnet and Mantle Pyroxenites on the Moho Transition Under Arcs
 Garnet has a strong influence on the physical properties of lower crust because of its high density and seismic velocities. According to petrological studies on Jijal samples, garnet granulites and garnetites are residues after dehydration or dehydration-melting of lower crustal hornblende-gabbros [Yamamoto and Yoshino, 1998; Garrido et al., 2006]. Local preservation of igneous structures in garnet granulites and very high-pressure crystallization of garnet in granulites alternatively point to an igneous origin for garnet in the same Jijal section [Burg et al., 1998; Ringuette et al., 1999]. Crystallization experiments of hydrous andesitic magmas at high pressure (0.8–1.2 GPa) produced garnet gabbros, pyroxenites, and hornblendites [Müntener and Ulmer, 2006; Alonso-Perez et al., 2009], whereas garnet crystallization in primitive basaltic arc magmas requires much higher pressures (≥1.5 GPa; Müntener and Ulmer ). Whether igneous or metamorphic in origin, garnet forms at pressures >0.8 GPa for bulk compositions corresponding to common mafic to intermediate arc rocks or magmas [Müntener and Ulmer, 2006; Ringuette et al., 1999: Figure 5]. Garnet is thus expected beneath island arcs mature enough to be thicker than ca 25 km, but not under the back-arc or fore-arc regions where crust is thinner [Stern, 2010].
 The Vp values measured and inferred across the Jijal mantle-crust transition are similar to those measured in thick active arcs (Figure 12). In both cases, there are two sharp seismic boundaries. The shallower is between P-wave velocities ≤7.0 km/s (garnet-free mafic rocks in Kohistan) and velocities of 7.4 to 7.5 km/s (granulitic, garnet-bearing gabbros in Kohistan). The deeper seismic boundary separates 7.4 to 7.5 km/s rocks from 7.7–8.1 km/s rocks (the mantle pyroxenites in the residual mantle peridotites of Kohistan). By contrast, the Moho below back-arc region is a unique seismic boundary characterized by the abrupt transition from gabbros (7.0–7.2 km/s) to mantle peridotite (8.0 km/s) [Takahashi et al., 2008].
Tatsumi et al.  explained the Vp increase, from 7.3 to 7.7 km/s, below the Izu-Bonin-Mariana arc as either a transition from garnet-poor to garnet-rich granulites, if a dehydration-melting model is considered, or from garnet-bearing gabbros to mantle harzburgite if igneous segregation is considered. In light of our results, this 25 km deep transition can be explained by garnet-bearing gabbros overlying a pyroxenite-rich mantle.
 The 100–200-m thick garnetite between lower crustal granulites and mantle pyroxenites of the Jijal section is a potential reflector (R ≥ 0.1). Reflectors in the crustal root of the thick Aleutian arc [Calvert and McGeary, 2012] can originate from garnet-rich rocks formed by fractional crystallization from an andesitic magma or representing residues after partial melting of hydrous gabbros.
 We combined thermodynamically-based calculations with laboratory measurements of P-wave velocities on representative samples from the crust-to-mantle transition zone of the Kohistan arc (Jijal section). Minor differences in the values of pressure and temperature derivatives of Vp are likely due to rock textures not implemented in calculation. The measured seismic properties extrapolated to natural arc root conditions are comparable to those calculated with thermodynamic models.
 Changes in P-wave velocities from lower crust to mantle are explained by both an increase in volumetric proportion of garnet within the mafic rocks and by the dominance of pyroxenites in the uppermost mantle. A mantle-crust transition zone is thus expected under mature arcs where crust is thick enough to form garnet. Reflectors may arise from garnetite lenses and layers (in case they are thick enough to fulfill the one quarter seismic wavelength criteria). The distribution and boundaries of Vp domains in the Kohistan sequence fit those deciphered in modern, active, and mature island arcs. By comparison with the Kohistan section, we suggest that the role of garnet-producing processes and the presence of pyroxenites have been underestimated in interpreting seismic signals in active margins.
 Field work was supported by the Swiss NF grant # 20–49372.96. Laboratory equipment and analyses were supported by the R'equip NF grant 2160–053289.98 and the ETH grant # 02150/41-2704.5. JB is funded by an Intra-European Marie Curie fellowship. R. Hofmann helped greatly in the maintenance of the rock deformation lab working at the ETH Zurich. F. Pirovino is thanked for the thin section preparation. Discussions with Jamie Connolly about seismic velocity calculation using Perple_X were very helpful. We gratefully acknowledge comments and suggestions from Erik Rybacki and Ian Jackson on an earlier version of this manuscript. We thank two anonymous reviewers for comments and suggestions that significantly improved the manuscript.