Water carried into subduction zones with the down-going plate and subsequently released by dehydration reactions at depth affects the composition of the mantle wedge, triggers partial melting and affects subduction zone seismicity. Partially serpentinized peridotite may be a significant reservoir for water in the subducted plate, the mantle wedge and the overriding plate. Here we develop a model that relates the degree of serpentinization and water content of partially serpentinized peridotites to their seismic P-wave velocities. In partially-serpentinized ultramafic rocks, a 1% decrease in P-wave velocity corresponds to a 2.4% increase in serpentine content, and a 0.3% (0.18 moles/m3) increase in H2O content (up to a maximum of 13%). Where there is evidence of serpentinization, mantle serpentine content is typically ∼15%, corresponding to 4–5 wt% H2O (6–10 moles/m3).
 Water released by dehydration reactions in subduction zones is thought to cause partial melting in the mantle wedge, to alter the petrology of the mantle wedge and the overlying plate, and to affect subduction zone seismicity [Bostock et al., 2002; Iwamori, 1998; Kamiya and Kobayashi, 2000; Peacock, 1990, 1993; Peacock and Hyndman, 1999]. To understand these processes, it is necessary to know how much water is present and in what form. Much H2O is carried into subduction zones by hydrous mineral phases in the oceanic crust, and partially-serpentinized peridotite is a likely constituent of the overlying mantle wedge [Guillot et al., 2001; Peacock, 1993; Seno et al., 2001]. It has also been proposed that the upper mantle of the downgoing plate is partially serpentinized to a depth of 40 km [Peacock, 2001; Seno et al., 2001]. Thus, serpentine is an important hydrous phase in subduction zones. Because these regions cannot be directly sampled, the presence of partially-serpentinized peridotites and the degree of serpentinization must be inferred from their affects on the seismic properties of the rocks, which can also be used to estimate their H2O contents. The objective of this analysis is to establish an empirical model that relates H2O contents of partially-serpentinized peridotites to their seismic velocities.
 Previous studies have established a strong linear relationship between seismic velocities and the degree of serpentinization in ultramafic rocks [Carlson and Miller, 1997; Christensen, 1972; Horen et al., 1996]. It follows that linear relationship can be established between seismic velocities and bound water contents in these rocks because serpentine is by far the dominant hydrous phase in partially-serpentinized ultramafic rocks, and all three serpentine minerals (lizardite, chrisotile and antigorite) contain 13% water by weight (∼18 moles/m3). The approach we have taken is to develop a model that relates water contents to seismic velocities as a function of pressure and temperature by first establishing statistical relationships between water contents and P-wave velocities measured at pressures ranging from 200 to 1000 MPa, then developing corrections for the effects of temperature and higher pressures.
 The weight fraction of water (wh) can be estimated from the density and serpentine content of each sample
where, ws is the weight fraction of water in serpentine (0.13), αs is the volume fraction of serpentine in the sample, ρs is the density of serpentine (2485 kg m−3) and ρb is the measured bulk density of the sample. This estimate neglects water bound in other phases, such as phyllosilicates, which contain about 12% H2O, and amphiboles, which contain about 2.2% H2O. However, neither of the mineral phases is likely to be abundant in mantle rocks.
3.1. Statistical Relationships
 The relationships between the volume fraction of serpentine, computed water contents and seismic P-wave velocities, measured at 25 °C and at confining pressures ranging from 200 to 1000 MPa [Christensen, 1966], are illustrated in Figure 1. Additional data from Christensen  and Miller and Christensen  are included for comparison. At each pressure we observe a strong linear relationship between water contents and P-wave velocities. For example, the best fitting linear equation for velocities measured at 1000 MPa is
with a coefficient of determination R2 = 0.98, a standard error of the estimate S = 0.56 wt%, and N = 11. The intercept values, for 100% serpentine (H2O content = 13%) and 0% serpentine (0% H2O), are 4.91 and 8.34 km/sec, respectively. Similarly, for Vp measured at 200 MPa, the best-fitting equation is
 Depths of interest to this analysis reach 100 km or more, where pressures exceed 3000 MPa. Experimental evidence indicates that the upper limit of the serpentine stability field is near 500°C [Janecky and Seyfried, 1986]. Hence, a useful model relating water contents to seismic velocities must take into account the effects of temperature and pressure.
 For pressures up to 1000 MPa, the dependence of H2O content on seismic velocities at various pressures can be estimated directly from Christensen's  data. The best-fitting relationships between H2O contents and seismic velocities at pressures of 200, 400, 600, 800 and 1000 MPa are illustrated in Figure 1a. There is a slight but systematic increase of intercept values from 29.5 wt% at 200 MPa to 32 wt% at 1000 MPa, but there is no significant pressure dependence of the slopes, which are typically −3.7 ± 0.2 wt%/km/s. The misfit (s.e.) is near 0.6 wt% H2O in all cases.
 To estimate velocities at higher pressures, it is necessary to extrapolate from the published data. The variation of velocity with effective pressure in rocks is well approximated by a simple power law relation [Carlson and Gangi, 1985; Gangi, 1978]; moreover, at high pressures
where P is effective pressure, Vo is the velocity at P = 0, and m is a dimensionless exponent. To model the effect of pressure on seismic velocities in serpentinized ultramafic rocks, we made fits of (3) to average P-wave velocities measured at elevated pressures in dunite, pyroxenite and serpentinite reported by [Christensen, 1996]. The best-fitting exponents are m = 0.013 ± 0.002, 0.0156 ± 0.001 and 0.034 ± 0.001 for dunite, pyroxenite and serpentinite, respectively. Based on these results, we take m ∼ 0.015 ± 0.001 for a peridotite composed of approximately 2/3 olivine and 1/3 pyroxene. We then have, for fresh peridotite
and for pure serpentinite
at pressures greater than 200 MPa.
3.3. Effect of Temperature
 Little is known about the dependence of seismic velocities on temperature in ultramafic rocks (as opposed to minerals). Christensen  reported experimental values of P-wave velocities measured at 200 MPa over a range of temperatures, and found ∂Vp/∂T = −0.68 × 10−3 and −0.56 × 10−3 km/s/°C for serpentinite and pyroxenite, respectively for temperatures between 25 and 300°C. Accordingly, for partially-serpentinized ultramafic rocks, we take ∂Vp/∂T ∼ −0.60 × 10−3 km/s/°C. We applied this correction to the best fitting relationship between water contents and P-wave velocities measured at 200 MPa (2b) to approximate the variation of water content with P-wave velocity at temperatures of 100, 200, 300, 400, and 500°C, as shown in Figure 1b.
3.4. Compressibility and Thermal Expansion
 Owing to the compressibility of the samples, we might expect pressure to affect the water content of the rocks. However, because the mass remains constant, the water content is not significantly affected by a change of volume; e.g.,
A similar argument applies to the influence of thermal expansion.
3.5. Computing H2O% From V(P, T)
 The foregoing relationships between water content, serpentine content, and P-wave velocity as a function of pressure and temperature can be combined to estimate the water content of partially-serpentinized peridotite from a measured (in situ) P-wave velocity under known conditions of temperature and pressure
where PO = 200 MPa, To = 25°C, wo = (30 ± 2)%, ∂w/∂V = −3.7 ± 0.3 %-sec/km, ∂V/∂T ∼ −0.6 × 10−3 km/s/°C, and m ∼ 0.015.
 There is, however, an easier and more practical way to estimate the bound water content of partially-serpentinized peridotites from their P-wave velocities. Because the variation of water content with velocity is linear and essentially the same under all conditions of temperature and pressure, complete alteration of an initially unaltered peridotite leads to a reduction of the P-wave velocity of 3.4 km/s (about 42%) over a wide range of temperatures and pressures (see Figure 1). Thus a handy approximation is
(or ∼−0.18 moles/m3 ΔV(%)) where ΔV is the difference between the observed velocity and the velocity in unaltered peridotite under the appropriate conditions of temperature and pressure. These simple approximations are particularly useful for interpreting tomography models, in which the seismic structure is expressed in terms of ΔV, the percentage difference between the in situ velocity and a standard model.
4. Applications of the Model
 Serpentinization is not the only potential cause of anomalously low seismic velocities in mantle rocks. Low velocity zones that occur in the crust and mantle wedge beneath volcanic arcs, where temperatures may be expected to be high, have been interpreted as evidence of partial melt zones [Zhao, 2001]. Where temperatures are sufficiently low, however, serpentinization is likely. Here we offer several examples. To facilitate the discussion, we include Figure 2, which shows the variation of seismic velocities, Vp/Vs ratios and bulk density with serpentinization and water content at pressures and temperatures appropriate to the mantle wedge or upper mantle of the subducted plate, 1000 MPa and 400 °C. The following estimates of the degree of serpentinization and water content are based on equation (7) and/or Figure 2.
Graeber and Asch  observed a 25-km-thick zone beneath the Chile margin with P-wave velocities ranging from 7.7–8.3 km/s, and Vp/Vs ratios of 1.75–1.84, which they interpret as consisting of partially hydrated mantle rocks. These properties are consistent with serpentinite contents of 0 to 12%; the corresponding range of water contents is 0 to 3.6% (0 to 5 moles/m3).
Kamiya and Kobayashi  have argued in favor of partial serpentinzation in the mantle wedge beneath central Japan, based on a seismic tomography model which shows P- and S-wave velocities of 6.9 and 3.4 km/sec, respectively. These properties are consistent with 30–40% serpentinization and H2O contents of 4–5 wt% (6–7 moles/m3).
 Similarly, citing a seismic tomography model by Zhao et al. , Seno et al.  note the existence of a significant low velocity anomaly (−6%) within the subducted slab beneath the Kii Peninsula. Our model (equation (7)) suggests that the degree of serpentinization is ∼15% and that the water content is ∼2% by weight (∼3 moles/m3). Seno et al. also suggest that velocities of 6.7–7.2 km/sec within the subducted plate beneath the Kanto District of Japan are evidence of serpentinization within the subducted plate. The cited P-wave velocities are consistent with 25–40% serpentinization and H2O content in the range 3–6% (4–8 moles/m3) within the subducted slab.
 Finally, velocities as low as 7.16 km/s have been reported in the mantle wedge at depths near 40 km beneath the Cascadia margin [Rondenay et al., 2001]. According to our model, this P-wave velocity corresponds to about 30% serpentinization, in good agreement with a previous estimate by [Bostock et al., 2002], and to a water content of ∼4 wt% (5–6 moles/m3).
 We conclude that, where there is evidence of serpentinization, the mantle is typically about 15% serpentinized, with a bound water content of 4–6%, or 6–10 moles/m3.