Influence of pressure and mineralogy on seismic velocities in oceanic gabbros: Implications for the composition and state of the lower oceanic crust



[1] We have analyzed the relationship between P wave velocities, measured at pressures of 40, 100, and 200 MPa, and modal mineralogy in oceanic gabbro samples from Ocean Drilling Program Holes 735B, 894G, and 923A. At all pressures, increasing velocities correspond to increasing pyroxene contents and decreasing alteration (phyllosilicate and amphibole content) but indicate little or no variation of olivine content, in part because the olivine compositions are in the range Fo65 to Fo73. A Voigt-Reuss-Hill inverse model reveals that the effective bulk density and elastic moduli of olivine in the gabbros are low, even relative to Fo73, possibly because the olivine grains contain ubiquitous networks of cracks. Even if the cracks are not present in situ, on average, gabbros with velocities typical of seismic layer 3 (6.7–7.0 km s−1) contain 5–15% alteration products, including 5–15% amphibole and 0–5% phyllosilicates. These results suggest that the lower crust is, on average, slightly to moderately altered.

1. Introduction

[2] The purpose of this study is to evaluate the influence of effective pressure, temperature, and mineralogy on the seismic properties of gabbroic rocks recovered by drilling from the oceanic crust. Because lower crustal rocks have been recovered by drilling in only a few places (notably Deep Sea Drilling Project/Ocean Drilling Program (DSDP/ODP) Sites 504, 735, 894, and 923 [Becker et al., 1988, 1992; Alt et al., 1993; Robinson and Von Herzen, 1989; Dick et al., 1999; Gillis et al., 1993; Cannat et al., 1995]), the oceanic crust is still best known by analogy with ophiolites [e.g., Salisbury and Christensen, 1978; Spudich and Orcutt, 1980; Christensen and Smewing, 1981] and by its seismic velocity structure (Figure 1). According to the ophiolite model, “normal” oceanic crust produced at fast spreading ridges and in the central regions of ridge segments at slow spreading ridges is composed of 1–1.5 km of extrusive mid-ocean ridge basalt (MORB), underlain by 1–1.5 km of sheeted diabase dikes that are, in turn, underlain by 4–5 km of gabbros [Coleman, 1977; Nicolas, 1989]. While ophiolites have three melt-derived lithologic components, the seismic structure of the crust is defined by two “layers”. Seismic layer 2 is characterized by very high vertical velocity gradients (∼1 s−1), with velocities commonly increasing from 3–4 km s−1 at the top of the crust to >6.5 km s−1 at an average depth near 2 km [e.g., White et al., 1992]. The lower crust (seismic layer 3) is characterized by lower-velocity gradients (0 to ∼0.1 s−1), and P wave velocities commonly in the range from 6.7 to 7.0 km s−1. Unraveling the relationship between the three-part lithologic structure of the crust and its two-layer seismic structure is an outstanding unresolved problem of marine geophysics.

Figure 1.

Examples of oceanic crustal seismic structure: NAT ESP 5 [NAT Study Group, 1985], ESP 2in [Minshull et al., 1991], and ESP 15 and ESP 18 [Morris et al., 1993]. Also shown are the downhole sonic log from DSDP/ODP Hole 504B [Alt et al., 1993] and velocity/depth profiles calculated from the average measured velocities in diabase samples from Hole 504B and gabbro samples from Holes 735B, 894G, and 923A (see text).

[3] Seismic velocities in the extrusive pile that caps the crust are known to be profoundly affected by cracks and voids in the formation [Houtz and Ewing, 1976; Hyndman and Drury, 1976; Wilkens et al., 1991; Grevemeyer and Weigel, 1997; Carlson, 1998]. In the underlying dikes, where the velocity gradient is lower (e.g., in DSDP Hole 504B, the gradient in the dike section is ∼0.5 s−1), it has been argued that velocities are controlled in large part by variations of modal mineralogy [e.g., Alt et al., 1993; Salisbury et al., 1996]. Christensen and Salisbury [1975] observed that chlorite is associated with low seismic velocities, while the presence of epidote signals higher velocities. In samples from ophiolites, velocities increase with increasing metamorphic grade, and the layer 2/3 seismic boundary is widely thought to be associated with a transition from greenschist to epidote-amphibolite grade mineral assemblages [e.g., Salisbury and Christensen, 1978; Christensen and Smewing, 1981; Detrick et al., 1994]. Similarly, gradually increasing velocities in the gabbro sections of ophiolites have been ascribed to decreasing degrees of alteration (because fresh gabbros have higher P wave velocities than gabbros altered to amphibolite grade [e.g., Christensen and Salisbury, 1975]) and to increasing olivine content [Salisbury and Christensen, 1978; Christensen and Smewing, 1981]. Increasing olivine content is also a favored cause of significant velocity gradients in seismic layer 3; for example, Collins et al. [1989] suggest that the range of velocities that is typical of seismic layer 3 (6.7–7.0 km s−1) can be explained by a 24% variation of the abundance of olivine.

[4] Given that (VOL > VCPX > VACT > VPLAG > VCHL), the influence of mineralogy on the seismic properties of lower crustal rocks would seem to be self-evident, but the seemingly obvious can prove to be wrong. The properties of diabase samples from DSDP/ODP Hole 504B offer a case in point. A large number of modal analyses of dike samples from Hole 504B show unequivocally that there is a systematic increase of amphibole content with depth in the dikes, and a corresponding decline in the abundance of chlorite [Alt et al., 1993]. These observations led Salisbury et al. [1996] to conclude that the in situ velocity gradient in the dikes (∼0.5 s−1) is caused by the combined effects of increasing amphibole and decreasing chlorite contents. They further suggested that velocities typical of seismic layer 3 are reached when chlorite disappears altogether. However, Carlson [2001a] computed Voigt-Reuss-Hill (VRH) average velocities from the modal mineralogy of the rocks and found that the VRH values agree well with velocities measured in the laboratory at 600 MPa, but both the measured and computed velocities are appreciably higher than the in situ velocities, and neither shows a variation of velocity with depth. Contrary to expectations, the variation of modal mineralogy in the dike rocks does not cause a variation of their seismic properties, probably because the effect of declining chlorite content is compensated by a corresponding increase of amphibole at the expense of pyroxene. The apparent relationships between modal mineralogy and the seismic properties of oceanic gabbros have not been similarly tested.

[5] The dependence of seismic velocities on mineralogy in gabbros can be investigated in several ways. Kelemen and Holbrook [1995] and Korenaga et al. [2002] have shown that P wave velocities in unaltered gabbros consisting of plagioclase, pyroxene, and olivine can be computed from their chemistry (SiO2 and MgO), but this approach is limited to unaltered rocks, whereas there is evidence that hydrothermal alteration has a significant influence on the seismic properties of gabbros [Salisbury and Christensen, 1978; Spudich and Orcutt, 1980; Christensen and Smewing, 1981]. Similarly, Hacker et al. [2003] demonstrated that seismic velocities in a variety of rocks can be quite accurately computed from the physical properties of their constituent minerals. They did not, however, compute the seismic properties of metagabbros as such. Instead, Hacker et al. computed the seismic properties of several “end-member” compositions: anhydrous diabase, olivine gabbro, gabbronorite, and the corresponding greenschist and amphibolite mineral assemblages. Noting that seismic velocities in the crust are higher than the computed velocities in the diabase and gabbro, but lower than the computed velocity in the amphibolite, Hacker et al. inferred the water content of the lower crust must be lower than that of the amphibolite, 1.3 wt %.

[6] There are several problems with computational approaches. One is that oceanic gabbros commonly contain alteration products, particularly clays, chlorite, and talc, that are not included in the anhydrous gabbro models used by Hacker et al. [2003], Kelemen and Holbrook [1995], or Korenaga et al. [2002]. Moreover, neither computational approach includes the influence of micro cracks that populate real rocks, and affect their properties. Other problems arise from the fact that the elastic properties of all of the major phases found in oceanic gabbros vary with chemical composition. Consequently, forward modeling of seismic velocities in oceanic gabbros is a good predictor of their average velocity but a poor predictor of the velocities in individual samples. This point is illustrated in Figure 2, which shows the agreement between the average measured velocities in a variety of rocks [Christensen and Mooney, 1995], and the velocities computed by Hacker et al. [2003]. Also shown are the measured velocities in the 64 gabbro samples used in this study (see below) versus velocities computed from their modal analyses using mineral properties from Hacker et al. The averages of the measured and predicted velocities are in excellent agreement, but because the scatter is as large as the comparatively small range of measured velocities, there is little correspondence between the measured and computed velocities in the individual samples. Clearly, velocities computed from the modal mineralogy cannot be used to assess the relationship between seismic velocity and composition in real oceanic gabbros. A realistic interpretation of crustal seismic velocities must be based on relationships between seismic velocities and modal mineralogy in real oceanic gabbros.

Figure 2.

Observed P wave velocities in oceanic gabbros versus Voigt-Reuss-Hill velocities calculated from modal analyses and physical properties from Hacker et al. [2003] (circles). Also shown are average velocities (diamonds) in a variety of rock types from Christensen and Mooney [1995] and velocities calculated by Hacker et al. [2003].

[7] Our approach is therefore empirical. We reason that if modal mineralogy affects the seismic properties of lower crustal rocks in situ, the same relationships must hold for laboratory samples recovered from the crust by drilling. Using velocities measured at elevated pressures [Miller and Christensen, 1997; Iturrino et al., 1991, 1996], and corresponding modal analyses [Miller and Christensen, 1997; Robinson and Von Herzen, 1989; Gillis et al., 1993], we have evaluated the relationships between mineral contents and seismic velocities in oceanic gabbros from ODP Holes 735B [Robinson and Von Herzen, 1989], 894G [Gillis et al., 1993], and 923A [Cannat et al., 1995]. We have also assessed the influence of temperature and effective pressure on the properties of these rocks.

[8] In brief, we find that there is a systematic increase of P wave velocities in oceanic gabbros with increasing pyroxene content; velocities also increase as the volume fraction of alteration products (phyllosilicates and amphiboles) decreases. Gabbros having velocities typical of the lower oceanic crust (6.7–7.0 km s−1) contain, on average, 5–15% alteration. An unexpected finding is that P wave velocities are not highly correlated with the abundance of olivine because the olivine in these rocks has low effective densities and elastic moduli. Correcting for the low moduli and density of olivine does not, however, alter the relationship between their seismic velocities and the degree of hydrous alteration.

2. The Data Set

[9] Our analysis requires measurements of seismic properties and modal mineralogy in the same materials. Seismic properties of gabbro samples, measured at elevated confining pressures, have been reported by Iturrino et al. [1991, 1996] and Miller and Christensen [1997]. Unfortunately, it has not been a common practice to report modal analyses of the samples in which P and S wave velocities have been measured. Only Miller and Christensen [1997] reported modal analyses in their study of seismic velocities in samples from ODP Hole 923A. We therefore searched the ODP initial reports for Legs 118 [Robinson and Von Herzen, 1989] and 147 [Gillis et al., 1993] to find modal analyses of samples for which the seismic properties were reported by Iturrino et al. [1991, 1996]. In some cases, samples used for velocity measurements and modal analyses are taken from the same interval of the drill core. In other cases, modal analyses are from adjacent parts of the core deemed close enough to have essentially the same compositions as the laboratory samples used for the velocity measurements. Our data set includes 64 samples, of which 44 include both P and S wave velocities. We did not include samples with velocities less then 6.6 or more than 7.3 km s−1 because there are so few of them. The data are summarized in Table 1.

Table 1. Summary of Velocities Measured at 200 MPa and Modal Analyses of Gabbros From ODP Holes 735B, 894G, and 923A
HoleSampleDepth, mPorosity, %Density, kg m−3Vp, km s−1Vs, km s−1OlivinePlagio-clasePyroxeneAmphibolePhyllosilicateAlteration
  • a

    The 894G Vp and Vsat 200 MPa were estimated from Vp (200) = Vp (100) + (0.073 ± 0.015) and Vs (200) = Vs (100) + (0.040 ± 0.020).

735B4D-2, 7–9h19.1 2.886.603.800.000.650.
735B85R-4, 9–11h476.1 2.936.663.800.000.630.
735B30R-5, 91–93h140.4 2.886.673.750.000.700.
923A15R-2, 88–91 1.512.836.673.640.140.820.
894Ga4R-2(71–75)V47.10.572.916.68 0.000.550.
923A2R-2, 55–58 1.652.836.713.690.050.700.
923A16R-2, 2–5 1.232.836.733.620.060.720.
735B12R-1, 36–39h39.8 2.996.77 0.000.500.410.060.000.06
735B33R-4, 129–131h159.3 2.906.793.880.000.430.000.400.170.57
923A14R-1, 55–58 0.572.846.803.740.280.690.
735B12R-3, 8–10h42.5 2.896.80 0.180.650.
735B76R-3, 50–52h397.5 2.986.81 0.000.450.370.160.000.16
923A16R-3, 42–45 0.472.856.833.760.060.800.
735B57R-2, 135–137h277.90.302.906.853.850.020.550.
735B26R-1, 62–64116.20.202.936.873.910.020.500.050.310.100.41
735B46R-2, 128–130h223.8 2.926.883.940.000.700.
923A12R-1, 120–124v 0.752.826.903.780.260.700.
735B42R-4, 62–64206.10.602.926.913.910.160.550.
735B58R-2, 33–35282.30.102.846.923.770.100.770.
923A14R-2, 6–9 0.982.856.953.720.060.750.
923A4R-1, 52–56 0.782.976.973.900.250.500.
923A10R-1, 106–109 0.622.936.973.860.090.640.
923A3R-1, 95–98 0.632.836.973.830.050.750.
735B85R-7, 17–19h490.2 2.886.98 0.160.650.
735B42R-2, 119–121h203.7 2.956.98 0.080.780.
735B84R-3, 14–16h465.2 2.956.98 0.070.630.
735B43R-4, 64–66h211.2 3.026.983.870.040.600.
735B31R-2, 120–122h146.2 2.936.993.870.010.460.
735B63R-6, 28–30313.30.302.927.003.900.030.500.110.370.030.40
894Ga12R-3(147–149)98.30.383.007.00 0.000.450.370.150.000.15
923A3R-2, 34–38 0.842.887.013.800.110.540.350.000.000.00
735B62R-3, 104–106h304.6 3.017.01 0.030.430.370.100.040.14
923A9R-2, 70–73 0.922.927.023.760.230.690.
923A13R-3, 50–53 0.952.947.023.920.130.650.
923A5R-2, 114–117 0.482.957.033.830.090.650.
894Ga12R-5(79–81)100.60.342.997.03 0.000.400.
923A10R-3, 44–47 0.422.927.043.810.090.600.
923A12R-2, 41–44 0.562.927.073.990.180.600.
735B13R-2, 55–58h46.6 2.907.07 0.020.680.
735B79R-7, 99–101h424.5 2.947.08 0.200.730.
923A8R-1, 107–110 0.592.777.083.760.180.750.
923A2R-2, 89–92 0.332.967.083.970.010.550.400.000.000.00
923A6R-1, 106–110 0.672.977.093.880.090.550.340.010.000.01
923A13R-1, 114–117 0.432.977.093.910.130.500.330.020.020.04
735B59R-3, 64–69h289.1 2.917.12 0.070.550.380.020.000.02
735B82R-2, 13–15h444.6 2.917.15 0.150.700.
735B23R-2, 34–36102.40.502.947.163.860.040.530.340.080.010.09
735B60R-1, 18–20h290.7 2.997.16 0.150.550.300.000.000.00
735B81R-7, 64–67h443.2 2.987.17 0.090.550.320.030.010.04
735B75R-6, 75–77h392.8 2.947.19 0.140.600.
923A7R-1, 81–84 0.642.997.203.950.090.600.
923A10R-1, 40–44 1.132.967.223.980.090.550.350.010.000.01
923A9R-1, 130–133 0.632.947.223.930.070.530.330.050.010.06
923A11R-2, 56–61 0.652.987.263.950.040.550.390.010.000.01
735B78R-4, 65–67h414.7 2.967.26 0.050.790.
735B83R-7, 104–106h462.6 3.107.28 0.400.
735B45R-2, 15–17h217.7 2.937.30 0.220.500.
735B69R-4, 138–140h346.90.202.977.324.050.080.550.
923A11R-1, 55–58v 0.523.007.334.170.070.600.310.020.000.02

2.1. Hole 735B Gabbros

[10] Hole 735B penetrated 500 m of gabbroic rocks beneath the Atlantis Bank on the Southwest Indian Ridge on ODP Leg 118 [Robinson and Von Herzen, 1989]. In contrast to the comparatively fresh rocks sampled at Site 923 [Cannat et al., 1995], the gabbros from Hole 735B are described as altered. Olivine is commonly present but rarely abundant. Serpentine is absent or so low in abundance that it was not reported. The dominant mineral phases are plagioclase and clinopyroxene, which are commonly recrystallized. The rocks are characterized by hydrothermal alteration at high temperatures. The most common alteration products are amphiboles (hornblende, tremolite, and actinolite). Other alteration phases are clays, chlorite, talc, opaque minerals, and occasional sphene, epidote, and carbonate. Apart from the Fe-Ti gabbros, olivine and plagioclase compositions are in the range Fo58–84 and An50–79, and the clinopyroxene compositions straddle the boundary between augite and diopside, with Mg numbers in the range 65–85 [Hébert et al., 1991].

[11] Iturrino et al. [1991] measured porosities, bulk densities and seismic velocities (at pressures to 200 MPa) in a suite of samples from the upper 500 m of ODP Hole 735B. A search of the site summary report [Robinson and Von Herzen, 1989] produced 34 modal analyses from parts of the core corresponding to the samples in which the seismic velocities were measured. Thus Table 1 includes P wave velocities and modal analyses for a total of 34 gabbro samples from Hole 735B, of which 15 include both P and S wave velocities. Fe-Ti gabbros, foliated gabbros, and mylonites are not included in the data set.

2.2. Hole 894G Gabbros

[12] Hole 894G recovered roughly 25 m of core from a 154-m section of gabbroic rocks at Hess Deep, in the equatorial Pacific [Gillis et al., 1993]. The sample suite is dominated by gabbronorite, with some gabbro and olivine gabbro. The Hole 894G gabbros were altered to greenschist facies mineral assemblages under static conditions, in contrast to the higher-temperature, synkinematic alteration observed at Site 735. The most common alteration product is amphibole, with some clays, chlorite, and talc; 40% alteration is not uncommon. Olivine compositions cluster near Fo89 and in the range Fo63 to Fo71, with a mean near Fo65; pyroxenes are typically Ca-rich augite; and plagioclase contents range from An50 to An70, with a median of about An60 [Pederson et al., 1996].

[13] Iturrino et al. [1996] reported porosities, bulk densities, and seismic velocities of samples from Hole 894G. Roughly half of the seismic velocities were measured at pressures to 600 MPa; the remainder were measured at pressures to 100 MPa. Gillis et al. [1993] reported just 10 modal analyses from parts of the core where samples were taken for the velocity measurements. Eight of these data sets are included in Table 1.

2.3. Hole 923A Gabbros

[14] Hole 923A was drilled in the MARK area of the Mid-Atlantic Ridge [Cannat et al., 1995]. The rocks recovered from Hole 923A consist of olivine gabbro, gabbro, and troctolite. The dominant primary minerals are plagioclase, clinopyroxene, and olivine. Total alteration is typically 10% or less, and the dominant alteration phases are amphiboles (tremolite and hornblende). Other phases (oxides, clay, chlorite, serpentine, and calcite) occur in small and variable amounts. On average, these samples are olivine gabbros; their physical properties and modal analyses, taken from Miller and Christensen [1997], are also summarized in Table 1. Plagioclase compositions range from An40 to An80, with a mean near An60, while olivine compositions range from Fo65 to Fo85, with a median value of about Fo73 [Casey, 1997; Ross and Elthon, 1997]; the average Fo number from 232 microprobe analyses of 68 grains made for this study is 73 ± 5.

[15] The average compositions of the gabbro samples from each site and the data set as a whole are summarized in Table 2. The gabbros from Holes 735B, 894G, and 923A exhibit a range of compositions. Alteration of the 923A gabbros has been described as “slight” [Cannat et al., 1995], while the rocks from Hole 735B are more altered [Robinson and Von Herzen, 1989], and alteration of the samples from Hole 894G is considerable [Gillis et al., 1993]. The average composition for the entire data set is 60% plagioclase, 20% pyroxene, 8.5% olivine, 8.7% amphibole, and 2% phyllosilicates (clays, chlorite, talc, etc.), and 1% opaque and other phases. The degree of alteration is highly variable. As shown in Figure 3, total hydrous alteration amounts to less than 5% in about half of these gabbros, and alteration in most of the remaining samples is in the range 5 to 30%.

Figure 3.

Distribution of total alteration (amphiboles + phyllosilicates) in gabbros from Hole 735B, 894G, and 923A used in this study (see Table 1).

Table 2. Average Modal Mineralogy of Gabbro Samples From Holes 735B, 894G, and 923A
HoleNOlivine, %Plagioclase, %Pyroxene, %Amphibole, %Phyllosilicate, %Alteration, %
735B318.1 ± 1.158.2 ± 1.518.8 ± 1.511.3 ± 1.42.5 ± 0.513.8 ± 1.7
894G86.3 ± 0.248.1 ± 0.423.5 ± 1.521.4 ± 1.04.3 ± 0.626.6 ± 1.2
923A2511.6 ± 0.963.9 ± 1.121.4 ± 1.51.4 ± 0.20.9 ± 0.12.3 ± 0.2
Total/mean648.5 ± 1.159.3 ± 1.420.4 ± 1.58.7 ± 1.32.1 ± 0.410.8 ± 1.6

[16] It has been suggested that the average composition of this suite of samples is not representative of the gabbros recovered by drilling because the data set does not include samples from the 1-km section of nearly unaltered gabbros cored in Hole 735B on ODP Leg 176 [Dick et al., 1999]. In fact, however, the average of 566 modal analyses from all three locations (including 241 and 216 from Hole 735B, Legs 118 and 176, respectively; 57 from Hole 894G; and 52 from Hole 923A) is 6.5% olivine, 57% plagioclase, 24% pyroxene, 9.3% amphiboles, and 2% phyllosilicates, in excellent agreement with the average composition of the samples in our data set (Table 2).

3. Influence of Temperature and Pressure on Seismic Velocities

[17] Christensen and Salisbury [1975] and Spudich and Orcutt [1980] noted the potential influence of effective pressure on the seismic structure of the oceanic crust. Pressure affects the properties of crystalline materials in two ways. The elastic moduli of single crystals show a modest increase, while the elastic moduli of rocks increase more dramatically with increasing pressure (particularly at pressures less than 200 to 400 MPa) owing to progressive stiffening of microcracks by compression [e.g., Kuster and Toksöz, 1974; Cheng and Toksöz, 1979; Carlson and Gangi, 1985; Gangi and Carlson, 1996]. The latter effect is certainly the larger of the two, and both depend on effective pressure.

[18] To evaluate the influence of pressure on seismic velocities in the oceanic crust, we must first estimate the in situ pressures. The effective pressure Pe is approximately equal to the difference between the confining pressure Pc and the pore fluid pressure Pp, i.e., PePcPp. At the top of the crust, Pp = Pc, and Pe = 0. The density of the crust is near 3000 kg m−3 [Carlson and Raskin, 1984], so the lithostatic pressure gradient is ∼30 MPa km−1, and the hydrostatic gradient is 10 MPa km−1. Where there is even a very sparse network of interconnected cracks, pore pressures are hydrostatic, and the effective pressure gradient is ∼20 MPa km−1, but if the cracks are closed at depth, the effective pressure will increase rapidly until it reaches the lithostatic pressure. Below the point where the cracks are entirely absent, the pressure gradient will be 30 MPa km−1. The effective pressure in the oceanic crust is near zero at the top of basement and increases at 20 MPa km−1 and reaches 140 MPa at the base of the crust if a network of cracks persists throughout the crust, or to 260 MPa at the Moho if cracks are absent near the base of the crust. In the latter case the pressure gradient will be variable, and the pressure gradient averaged over the entire crust will be ∼37 MPa km−1.

[19] Pressure acting on mineral grains does not appear to be a significant contributor to observed seismic velocity gradients. Few direct measurements of the pressure derivatives (∂V/∂P) of elastic wave velocities in the minerals that are most common in the crust (plagioclase, pyroxene, and amphibole) have been made because most studies of the elastic properties of minerals have focused on phases that occur in the mantle. Measured derivatives from studies of polycrystalline aggregates and single crystals have been compiled by Christensen [1982], Bass [1995], and Carlson [2001b]. The ratio ∂Vp/∂P is typically 5–10 × 10−3 km s−1 per 100 MPa, and the highest reported rate of change is 20.6 × 10−3 km s−1 per 100 MPa in a bronzite specimen [Frizillo and Barsch, 1972]. If the average pressure acting on the mineral grains increases at 20 to 37 MPa km−1, even the highest reported rate of change of Vp with pressure in a mineral will account for velocity gradients of only 0.004 to 0.008 s−1, a total increase of only 0.054 km s −1 over the 7-km thickness of the crust.

[20] Velocities in samples from Holes 735B, 894G, and 923A have been measured as a function of pressure by Iturrino et al. [1991], Iturrino et al. [1996], and Miller and Christensen [1997]. If the rocks recovered from these sites are representative of the rocks in situ, these data can be used to estimate the in situ variation of velocity with depth in oceanic gabbros. Seismic waves sample the properties of the crust on a scale of hundreds of meters or more [e.g., Spudich and Orcutt, 1980] and thus reflect the properties of the rocks averaged over very large volumes. Consequently, it is appropriate to consider the variation of average velocity with pressure, which is well approximated at pressures ranging from 10 to 200 MPa, by V(P) = Vo + b ln(P) [Carlson, 2001a]. For a given effective pressure gradient, this equation is readily converted to V(z) = Vo + b ln(z). This function has the advantage that the derivative (∂V/∂z = b/z) is easy to calculate. Model parameters obtained by fitting the average velocities in the gabbros (measured at pressures of 20, 40, 60, 80, 100 and 200 MPa), to depths (calculated by assuming a pressure gradient of 25 MPa km−1), are summarized in Table 3. Velocity gradients range from about 0.08 s−1 at 2 km to 0.02 s−1 at a depth of 7 km. These gradients are comparable to those that are characteristic of seismic layer 3, which range from near 0 to ∼0.1 s−1. Hence velocity gradients in lower oceanic crust can often be explained by compression of micro cracks in “intact” gabbros (i.e., laboratory samples that do not contain large-scale cracks) by increasing in situ pressure.

Table 3. Summary of Velocity-Depth Model Parameters Vp = Vo + b ln(z)
HoleLithologyVo, km s−1bS, km s−1R2
504Bdiabase6.345 ± 0.0070.179 ± 0.0080.010.992
735Bgabbro6.786 ± 0.0100.130 ± 0.0070.010.986
894Ggabbro6.653 ± 0.0050.129 ± 0.0040.0030.996
923Agabbro6.720 ± 0.0180.152 ± 0.0140.0170.967

[21] Temperature also affects the elastic properties of rocks. Velocities decrease with increasing temperature, so that the increase of temperature with depth tends to mitigate the influence of increasing effective pressure. At 200 MPa, ∂Vp/∂T ∼ 5 × 10−4 km s−1 °C [Christensen, 1979]. If the conductive temperature gradient in old crust is ∼20°C km−1, the contribution to the velocity gradient arising from the geothermal gradient is about −0.01 s−1. This gradient is lower than the velocity gradient caused by increasing effective pressure at the top of the lower crust, but comparable to the gradient caused by increasing pressure at the deeper levels of the crust. Thus the temperature gradient in old crust will reduce the overall velocity gradient by 0.01 s−1. Where the temperature gradient is higher, i.e., in young crust, the influence of temperature will be appreciably greater.

4. Comparison With Seismic Profiles

[22] The average gabbro velocity/depth profiles are compared to typical seismic velocity profiles in Figures 1 and 4. In the gabbro profiles, velocities increase from ∼6.8 km s−1 at 1 km, to ∼7 km s−1 at a depth of 8 km. In seismic profiles the layer 2/3 transition is defined by two criteria, velocities “typical” of seismic layer 3, and a reduction of the velocity gradient to ∼0.1 s−1 [e.g., White et al., 1992; Detrick et al., 1994]. A particularly striking feature of Figures 1 and 4 is that in almost every case, seismic profiles show a very rapid increase of velocity with depth until they reach the gabbro curves, at which point the velocity gradient decreases to values characteristic of layer 3. This fact suggests that the layer 2/3 transition marks the top of the intact gabbro sequence. Worthy of note in this regard is that measured velocities in intact diabase samples from Hole 504B (see Figure 1) are higher than the sonic log velocities but appreciably lower than velocities at the layer 2/3 transition in these profiles. Another significant point is that several of the seismic profiles shown in Figure 4a (notably ESP 1 [Lindwall, 1991], ESP 4S, and EPR 9N [Vera et al., 1990]) lie on or very near the gabbro curves; the EPR 9N profile is a nearly exact match to the gabbro curves from the layer 2/3 boundary at a depth of 1.5 km to the Moho at a depth of 6.5 km. In other cases, however, there are significant discrepancies between the in situ seismic velocities and the seismic properties of the average gabbro. In the NAT ESP 18 and ESP 1 profiles (Figure 4b), for example, the observed layer 3 velocities are 0.1 to 0.2 km s−1 higher than the velocities in the gabbros. In the Ngendei, ESP 2in, and ESP 15 profiles [Shearer and Orcutt, 1986; Minshull et al., 1991; Morris et al., 1993] (Figure 4c) the lower crustal velocity gradients differ from the gradient in the gabbro profiles. These cases suggest that variations of mineralogy, as well as pressure, are required to account for the range of seismic velocities in the lower oceanic crust, at least in some cases.

Figure 4.

Comparison of seismic profiles with velocity/depth profiles computed from the measured velocities in gabbros from Holes 735B, 894G, and 923A. (a) Cases in which there is good agreement between the seismic profiles and the gabbro profiles for ESP4S [Vera et al., 1990], ESP1 [Lindwall, 1991], and EPR 9N [Vera et al., 1990]. (b) Cases in which in situ lower crustal velocities exceed the average gabbro velocities for NAT ESP 18 [NAT Study Group, 1985], and ESP 1 [Vera et al., 1990]. (c) Cases in which in situ velocity gradients differ from gradients in the average gabbro profiles for ESP 2in [Minshull et al., 1991], ESP 15 [Morris et al., 1993], and Ngendei [Shearer and Orcutt, 1986].

5. Influence of Mineral Content on Seismic Velocities

[23] We expect the seismic properties of oceanic gabbros to exhibit some dependence on modal mineralogy because the minerals that comprise these rocks have a considerable range of densities and elastic moduli. The primary minerals in oceanic diabases and gabbros are plagioclase, pyroxene, and olivine. Amphiboles (hornblende, tremolite, and actinolite), phyllosilicates (talc, chlorite, and clays), and epidote occur as alteration products. The properties of some of these minerals are summarized in Table 4, and their P and S wave velocities are shown in Figure 5.

Figure 5.

Vp versus Vs in selected minerals and P and S wave velocities in gabbro samples from ODP Hole 735B [Iturrino et al., 1991], 894G [Iturrino et al., 1996] and 923A [Miller and Christensen, 1997] measured at 200 MPa. Also shown are reduced major axis fits to the mineral velocities and velocities in the gabbro samples measured at 40 and 200 MPa. (The RMA line is a type of best fit in which the slope is given by the ratio of the variances.)

Table 4. Physical Properties of Selected Minerals
MineralDensity, Mg m−3k, GPaμ, GPaVp, km s−1Vs, km s−1Source
Forsterite3.22127828.565.03Hacker et al. [2003]
Olivine3.31129798.424.89Verma [1960]
Fo903.36129788.344.82Abramson et al. [1997]
Fo733.54130727.994.50Voigt-Reuss-Hill average
Fayalite4.40137516.833.41Hacker et al. [2003]
Diopside3.27113657.814.45Hacker et al. [2003]
Diopside3.31112647.704.40Alekandrov et al. [1964]
Augite3.3294597.204.20Aleksandrov et al. [1964]
Hedenbergite3.65119617.414.09Hacker et al. [2003]
Hornblende3.1595467.003.80Aleksandrov and Ryzhova [1962]
Hornblende3.2594547.164.10Hacker et al. [2003]
Tremolite2.9885497.104.06Hacker et al. [2003]
Albite2.6254285.903.27Hacker et al. [2003]
An242.6463296.223.34Ryzhova [1964]
An532.6871336.573.53Ryzhova [1964]
An582.6875346.703.55Ryzhova [1964]
Anorthite2.7684376.953.66Hacker et al. [2003]
Epidote3.3396496.963.82Ryzhova et al. [1966]
Epidote3.5106617.334.18Hacker et al. [2003]
Antigorite2.5963185.822.65Hacker et al. [2003]
Chloritea3.1677276.003.00Christensen and Wilkens [1982]
Chloriteb3.0656345.423.32Hacker et al. [2003]
Talc2.7842235.082.85Hacker et al. [2003]
Talc2.7941.622.65.072.85Bailey and Holloway [2000] and Pawley et al. [1995]

[24] An interesting feature of Figure 5 is the fact that the mineral P and S wave velocities lie near a linear trend, or mixing line, indicated by the reduced major axis (RMA) line. The properties of gabbros composed of these minerals lie on or near the mixing trend, and different points on the trend represent samples with different compositions, but the converse is not true; no point on the trend represents a unique composition. Indeed, a wide range of possible (even unlikely) compositions will yield the same velocities. A gabbro composed entirely of olivine, pyroxene, and plagioclase could have the same P and S wave velocities as metagabbro that contains less plagioclase and pyroxene, but includes hydrous alteration products such as chlorite and hornblende.

[25] Also shown in Figure 5 are P and S wave velocities, measured at 200 MPa, in the oceanic gabbros for which both Vp and Vs have been reported (Table 1), and the RMA lines for gabbro velocities measured at both 40 and 200 MPa. There are three significant points to be drawn from this plot. One is that the velocities in the gabbro samples span the range of seismic velocities that are typical of the lower oceanic crust; another is that the gabbro data lie on the same trend as the minerals that comprise them, suggesting that the properties of these rocks are controlled largely by their mineralogy. We should thus expect velocities to increase with increasing olivine or pyroxene content, and decrease with increasing plagioclase or chlorite content, as inferred in previous studies [e.g., Christensen and Salisbury, 1975; Salisbury and Christensen, 1978; Christensen and Smewing, 1981; Spudich and Orcutt, 1980; Salisbury et al., 1996]. The third point is that the RMA lines for velocities measured at 40 and 200 MPa are not statistically distinguishable; as pressure increases, and the cracks in these rocks are compressed, the P and S wave velocities increase, but continue to lie on the same trend.

[26] Owing to the nonuniqueness of the relationship between composition and seismic properties, we cannot hope to infer unique compositions from seismic velocities. We can, however, seek to determine how the average modal mineralogy of the gabbros varies with their seismic velocities. For that purpose, we binned the P wave velocities and modal abundances listed in Table 1 in 0.1 km s−1 intervals beginning at 6.6 km s−1, then averaged both the P wave velocities and modal fractions over each bin. Because velocities in the Hole 894G samples were measured to only 100 MPa [Iturrino et al., 1996], we estimated the velocities at 200 MPa from the average difference between velocities measured at 100 MPa and those measured at 200 MPa in other studies [Iturrino et al., 1991; Miller and Christensen, 1997], as indicated in the footnote to the table. Five samples were not included in the analysis, three that have velocities lower than 6.6 km s−1 and two that have velocities higher than 7.3 km s−1, at 200 MPa. The phyllosilicates include clays, chlorite, and talc; total alteration is the sum of the amphibole and phyllosilicate concentrations. Though Table 1 includes only the velocities measured at 200 MPa, we are interested in the relationship between seismic velocities and modal mineralogy throughout the lower crust. For that reason, we carried out the analysis using velocities measured at 40 and 100 MPa as well.

[27] The results are summarized in Figure 6. Error bars indicate the standard errors (SE) of the means, not the standard deviations (SD) of the sample populations. The standard errors are typically 0.05 (5%) or less. There are several sources of scatter in the averages. One is the nonuniqueness of the relationship between seismic properties and modal mineralogy. Rocks that have different mineral contents have velocities in the same range. Another is the error inherent in the point counting procedure, which depends on the number of points counted. The standard error of each modal fraction αi is Si = √[αi(1 − αi)/Nc], where Nc is the total number of counts per sample (= 1000 or more). For Nc = 1000, values of α and their estimated uncertainties are 0.010 ± 0.003, 0.100 ± 0.010, and 0.200 ± 0.013. For α ≥ 0.300, the uncertainty is near ±0.015. In cases where the modal abundances have been estimated by inspection, the errors are appreciably greater. A significant source of scatter arises from the heterogeneity of the samples. The samples are commonly fairly coarse grained, and the pattern of alteration is patchy, even on the scale of the thin section, because alteration is concentrated around cracks. Thin sections are taken from core ends, or from adjacent parts of the drill core (within a few centimeters). Consequently, the mineral contents of the thin sections used for modal analyses are likely to differ from the composition the samples in which the velocities were measured.

Figure 6.

Relationships between modal mineral abundances and P wave velocities in gabbros from ODP Hole 735B, 894G, and 923A, measured at 40, 100, and 200 MPa. These data were obtained by binning velocities and modal analyses in 0.1 km s−1 intervals and averaging. Shading represents the range of velocities that are typical of the lower crust, and lines represent least squares fits to the data. Open squares in the 200 MPa plot indicate the theoretical relationship between seismic velocity and olivine (Fo90) content.

[28] Figure 6 shows the relationships between seismic P wave velocities and average modal mineralogy in the oceanic gabbro samples. Statistical correlations of P wave velocities measured at 40 and 200 MPa with modal mineral contents and between mineral concentrations, are summarized in Table 5. The best fit lines shown in Figure 6 are intended only to indicate the trends; we have not attempted to quantify the relationships between seismic velocities and mineral proportions. At 200 MPa, average pyroxene concentrations increase, while plagioclase contents tend to decrease with increasing P wave velocity. These trends are to be expected, given that plagioclase and pyroxene have markedly different seismic properties, and they are the most abundant phases in the gabbros. Phyllosilicate, amphibole, and total alteration decrease with increasing seismic velocity, as noted by Christensen and Salisbury [1975] and Christensen and Smewing [1981], among others. We also expect higher velocities to signal higher olivine contents. The theoretical dependence of velocity on Mg-rich olivine (Fo90) is indicated in Figure 6. However, at 200 MPa, there is, at most, a very small increase of olivine content with increasing velocity, and over the range of velocities between 6.6 and 7.2 km s−1, olivine contents range from 6 to 11%, with no statistically significant trend (Figure 6 and Table 5).

Table 5. Summary of Correlationsa
 40 MPa, Vp = 6.38–7.12, N = 6440 MPa, Vp = 6.67–7.12, N = 46
  • a

    Numbers in parentheses indicate the probability that the observed correlation occurs by chance.

Vp1      1      
OL0.151     0.091     
(0.24)      (0.57)      
PLAG−0.200.091    −0.07−0.011    
(0.12)(0.48)     (0.67)(>0.99)     
PYX0.38−0.37−0.421   0.24−0.43−0.381   
(<.01)(<.01)(<.01)    (0.12)(<0.01)(0.01)    
AMPH−0.23−0.40−0.55−0.231  −0.17−0.27−0.56−0.271  
(0.07)(<.01)(<.01)(0.07)   (0.27)(0.08)(<0.01)(0.08)   
Phyll.−0.23−0.14−0.30−0.420.541 −0.30−0.07−0.31−0.380.571 
(0.07)(0.27)(0.02)(<0.01)(<0.01)  (0.05)(0.67)(0.01)(0.01)(<0.01)  
(0.05)(<0.01)(0.24)(0.65)(<0.01)(0.24) (0.34)(0.04)(0.37)(0.44)(<0.01)(0.05) 
 200 MPa, Vp = 6.60–7.33, N = 64200 MPa, Vp = 6.7–7.2, N = 50
Vp1      1      
OL0.271     0.151     
(0.03)      (0.29)      
PLAG−0.250.091    −0.080.431    
(0.05)(0.48)     (0.58)(<0.01)     
PYX0.40−0.37−0.421   0.39−0.43−0.571   
(<0.01)(<0.01)(<0.01)    (<0.01)(<0.01)(<0.01)    
AMPH−0.28−0.40−0.55−0.231  −0.34−0.58−0.61−0.111  
(0.02)(<0.01)(<0.01)(0.07)   (0.01)(<0.01)(<0.01)(0.44)   
Phyll.−0.28−0.14−0.30−0.420.541 −0.38−0.22−0.29−0.400.611 
(0.02)(0.27)(0.02)(<0.01)(<0.01)  (<0.01)(0.12)(0.04)(<0.01)(<0.01)  
(0.01)(<0.01)(0.24)(0.65)(<0.01)(0.24) (0.19)(<0.01)(0.02)(0.11)(<0.01)(0.74) 

[29] These patterns persist at 100, and even at 40 MPa. This result is somewhat surprising because micro cracks have a considerable effect on seismic velocities in crystalline rocks [e.g., Carlson and Gangi, 1985]; between 40 and 200 MPa, average P wave velocities increase from 6.8 to 7.0 km s−1. One might therefore expect any correlation between seismic velocities and modal mineralogy to be masked by the influence of cracks, particularly at lower pressures, but that is not the case. At 40 MPa, the correlations between seismic velocity and plagioclase or olivine concentration are not quite statistically significant at the 95% confidence level, but the trends in the data are very much the same at 40 MPa as they are at 200 MPa. Apparently, the cracks that populate these oceanic gabbros are so similar from sample to sample that, even at comparatively low pressures, the dependence of seismic velocity on the mineral composition of the gabbros is evident. On average, for Vp = 6.7 km s−1 and P = 40 MPa (a pressure appropriate for the top of layer 3 at 2 km depth), the gabbros in this data set contain 55–65% plagioclase, 15–25% pyroxene, 5–10% olivine, 5–10% amphibole, and 5–15% total alteration (amphiboles + phyllosilicates). For Vp = 7.0 km s−1 and P = 200 MPa (corresponding to a depth near the base of the crust) the composition profile is essentially the same. Conversely, on average, gabbros that have velocities in excess of 7 km s−1 have higher pyroxene contents (20–30%), and comparatively low abundances of alteration products (<10%); phyllosilicates are essentially absent.

6. Olivine Anomaly

[30] As noted previously, increasing olivine content is widely cited to account for velocity gradients in the lower crust, [e.g., Salisbury and Christensen, 1978; Christensen and Smewing, 1981; Collins et al., 1989]. In these rocks, P wave velocities are most strongly correlated with pyroxene contents, and more strongly correlated with plagioclase contents and total alteration than they are with olivine contents (Figure 6 and Table 5). The absence of any meaningful correlation between the P wave velocities in these rocks and their olivine contents (Figure 6) is unexpected. Even the weak correlation between velocity and olivine content at 200 MPa (Table 5) arises from the properties of just two samples, 735B 4D-2 (7–9), which has a velocity of 6.60 km s−1 and contains no olivine, and 735B 83R-7 (104–106), which has a velocity of 7.28 km s−1 and contains 40% olivine. For velocities in the range 6.7 and 7.2 km s−1 there is no significant correlation of seismic velocity with the abundance of olivine at any pressure (see Table 5 and Figure 6).

[31] There are three possible explanations for the fact that olivine has such a small effect on seismic velocities: the abundance of olivine in these gabbros could covary with the abundance of other mineral phases in a way that confounds the expected influence of olivine on the seismic velocities, in the same way that increasing amphibole balances the effect of decreasing chlorite in the diabase samples from Hole 504B [Carlson, 2001a]. A second possibility is that the elastic moduli of olivine are lower than they are typically assumed to be because of their comparatively low magnesium contents; and a third is that the elastic moduli are affected by the networks of cracks that are a characteristic feature of the olivine grains.

[32] The expected increase of seismic velocity with increasing olivine content could be compensated by phases having lower elastic moduli (e.g., phyllosilicates, plagioclase, or amphibole) if and only if the olivine concentration has a sufficiently strong, positive correlation with the abundance those phases. Similarly, if the elastic moduli of the olivine phase are low, the influence of increasing olivine content could be masked by a corresponding decrease of pyroxene content. Otherwise, the velocity in the rock will necessarily increase with increasing olivine content.

[33] As indicated in Table 5, the correlations between olivine, amphibole, and phyllosilicate concentrations are negative. Thus rising olivine concentrations cannot be compensated by the abundance of amphibole or phyllosilicate minerals. The relationship between seismic velocities, olivine concentrations and plagioclase contents is less clear. Over the full range of velocities measured at 200 MPa, there is a statistically significant correlation between P wave velocities and the abundance of olivine, and no correlation between olivine and plagioclase contents, while over velocities ranging from 6.7 to 7.2 km s−1, there is no correlation between velocity and olivine content, but there is a significant positive correlation (r = 0.43) between the olivine and plagioclase contents. These correlations suggest that the effect of increasing olivine content on the seismic properties of these rocks could be compensated by a corresponding increase in the abundance of plagioclase. However, these relationships do not persist at lower pressures. At 40 MPa, there is neither a significant correlation of seismic velocities with olivine contents nor a significant correlation of olivine contents with plagioclase contents. Hence the low correlation between olivine contents and seismic velocities measured at 40 MPa cannot be explained by a corresponding increase in the abundance of plagioclase. Apparently, the low correlation between P wave velocities and olivine concentrations in oceanic gabbros cannot be explained by a compensating increase of plagioclase, amphibole, or phyllosilicate contents. However, in all cases, there is a comparatively strong negative correlation between the olivine and pyroxene contents of the gabbros (Table 5). If the properties of these two phases are comparable, the influence of olivine on the seismic properties of the gabbros may be masked, in part, by an exchange of olivine for pyroxene.

[34] The physical properties of olivine vary markedly with chemical composition. Though Hacker et al. [2003] used an olivine composition of about Fo78 to model the properties of olivine gabbro, olivine is usually assumed to be Mg-rich Fo90. As noted earlier, the average olivine composition in these gabbros is in the range Fo65–73, for which velocities are comparable to the velocities in pyroxene minerals (Table 4). Hence another possible explanation for the fact that seismic velocities in these rocks do not increase significantly with increasing olivine contents is that the Mg contents of olivine are lower than we assume them to be. To investigate this possibility, we solved for the effective elastic moduli of olivine, plagioclase, pyroxene, amphibole, phyllosilicates and “other” phases in the gabbros by inversion using the Voigt-Reuss-Hill model [Hill, 1952] relating elastic moduli to modal mineralogy. The inversion solves explicitly for the best fitting average elastic moduli of the different mineral phases. Owing to variations of mineral chemistry, there is considerable variation of physical properties of the different phases (see Table 4). These variations affect the uncertainties in the model parameters, but the best fitting parameters are not affected by compositional variations or by covariances among the minerals. Thus, if the weak observed correlation between the seismic velocities and olivine contents in the gabbros is an artifact of covariance among several variables, the best fitting olivine moduli obtained from the inversion will be consistent with the properties of olivine grains with compositions in the range Fo65–73 (Table 4). Conversely, if the effective density and elastic moduli of olivine in the gabbros are in fact lower than expected, the inversion will yield low best fitting values for the density and elastic moduli of olivine, while the best fitting densities and moduli of the other phases will agree with their expected values.

[35] Two theoretical models are commonly used to compute the elastic moduli of aggregates from the volume fractions and known densities and elastic properties of their mineral constituents, the average of the Hashin-Shtrikman (HS) bounds and the Voigt-Reuss-Hill (VRH) average [Hashin and Shtrikman, 1962; Hill, 1952]. For crystalline materials there is good agreement between the two models [Watt et al., 1976; Hacker et al., 2003]. Because the VRH model is considerably simpler, we have used it to estimate the effective isotropic elastic moduli of the principal phases in the gabbros by nonlinear least squares.

[36] The cost function is

display math

The VRH average is given by

display math

where MV and MR are the Voigt and Reuss average elastic moduli, respectively;

display math

αi is the volume fraction of the ith phase and Mi is its bulk modulus ki, or shear modulus μi. Because both the bulk and shear moduli depend on the shear wave velocity, only the 44 samples in which both velocities were measured (Table 1) were used in the inversion. We solved separately for the best fitting bulk and shear moduli; we also solved for the mineral densities by linear regression based on the volume average equation

display math

[37] Because the 44 samples we used for the inversion represent a subset of ∼70% of the whole data set, it is essential to establish that this subset of the data exhibits the same relationships between the measured velocities and modal mineralogy that are shown by the whole data set (Figure 6). We therefore repeated the binning and averaging procedure we applied to the full data set. The results are plotted in Figure 7, where we see that the pattern we observed previously persists; P wave velocities increase with increasing pyroxene content and decreasing alteration, but there is no significant increase of seismic velocity with increasing olivine content. That we observe this pattern in a subset of the data chosen essentially at random encourages us to believe that these relationships between measured P wave velocities and modal mineralogy are fairly robust.

Figure 7.

Relationships between modal mineral abundances and P wave velocities in gabbros from ODP Holes 735B, 894G, and 923A, showing trends based on 44 samples used for the Voigt-Reuss-Hill inversion to estimate best fitting mineral densities and elastic moduli for (a) raw data and (b) data “corrected” for low density and elastic moduli of olivine (see text). Shading represents the range of velocities that are typical of the lower crust, and lines represent least squares fits to the data.

[38] Returning to the inversion, we solved for the best fitting elastic moduli of six phases: olivine, pyroxene, plagioclase, amphibole, phyllosilicates, and “other” phases. P and S wave velocities were calculated from the best fitting densities and elastic moduli. We also estimated the P and S wave velocities directly by linear regression using the time-average equation and obtained the same results. The results are reported, with examples of published isotropic elastic moduli of olivine, pyroxene, plagioclase, and hornblende in Table 6 and are illustrated in Figure 8.

Figure 8.

P and S wave velocities computed from the best fitting densities and elastic moduli of olivine, pyroxene, plagioclase, and amphibole, compared with velocities computed from the isotropic elastic constants derived from single-crystal properties. Error bars are 1σ. Data and sources are given in Tables 4 and 6.

Table 6. Summary of Estimated Elastic Moduli
MineralPressure, MPak, GPam, GPar, Mg m−3Vp, km s−1Vs, km s−1Source
  • a

    Calculated using end-member properties from Hacker et al. [2003].

  • b

    Six-parameter fit, includes phyllosilicates and “other phases.”

Olivine40108 ± 1743 ± 93.05 ± 0.117.38 ± 0.483.76 ± 0.40 
100104 ± 1651 ± 103.05 ± 0.117.51 ± 0.474.10 ± 0.40 
200108 ± 1653 ± 103.05 ± 0.117.66 ± 0.464.17 ± 0.40 
200b105 ± 1757 ± 113.06 ± 0.117.69 ± 0.504.32 ± 0.42 
Fo90 128783.348.334.83Abramson et al. [1997]
Fo73 129723.547.974.50calculateda
Fo65 130693.637.824.36calculateda
Pyroxene40120 ± 956 ± 53.27 ± 0.057.70 ± 0.234.12 ± 0.19 
100121 ± 859 ± 53.27 ± 0.057.82 ± 0.224.25 ± 0.19 
200126 ± 860 ± 53.27 ± 0.057.94 ± 0.224.27 ± .019 
200b125 ± 963 ± 63.27 ± 0.057.98 ± 0.234.39 ± 0.20 
Clinopyroxene 117723.338.004.65Collins et al. [1989]
Diopside 112643.317.724.40Aleksandrov et al. [1964]
Amphibole4081 ± 744 ± 53.04 ± 0.086.76 ± 0.263.80 ± 0.22 
10083 ± 743 ± 53.04 ± 0.086.80 ± 0.253.77 ± 0.21 
20085 ± 744 ± 53.04 ± 0.086.86 ± 0.253.79 ± 0.21 
200b85 ± 1043 ± 63.04 ± 0.086.84 ± 0.333.76 ± 0.29 
Hornblende 87433.126.803.71Aleksandrov and Ryzhova [1962]
Hornblende 95463.157.053.82Aleksandrov and Ryzhova [1962]
Plagioclase4066 ± 336 ± 22.75 ± 0.036.42 ± 0.123.61 ± 0.11 
10071 ± 337 ± 22.75 ± 0.036.60 ± 0.123.64 ± 0.10 
20071 ± 337 ± 22.75 ± 0.036.64 ± 0.123.69 ± 0.10 
200b73 ± 435 ± 22.75 ± 0.036.60 ± 0.133.58 ± 0.11 
An53 71332.686.553.51Ryzhova [1964]
An56 67382.696.613.76Ryzhova [1964]
An58 75342.686.703.56Aleksandrov and Ryzhova [1962]
Phyllosilicates200b74 ± 2766 ± 272.98 ± 0.247.3 ± 1.14.7 ± 1.0 
Other phases200b52 ± 26125 ± 543.26 ± 0.388.2 ± 1.56.2 ± 1.4 

[39] The misfits to k and μ are 5.08 and 3.24 GPa, respectively. The effective elastic moduli of phyllosilicates and other phases are not well resolved, probably because their abundance in these rocks is so low (see Table 1). For that reason, we also made a four-parameter fit to solve for the best fitting moduli of olivine, pyroxene, plagioclase, and amphibole alone. These results are also included in Table 6. In this case, the misfits are 5.02 and 3.32 GPa, for k and μ respectively. The best fitting moduli of olivine, pyroxene, plagioclase and amphibole obtained from the six-parameter and four-parameter fits are statistically indistinguishable, but the fact that the six-parameter fit does not yield a smaller misfit than the four-parameter fit indicates that the best fitting values for the phyllosilicates and “other phases” obtained from the six-parameter fit are not meaningful; by including these two parameters in the model, we are fitting the model to, or “honoring”, the noise in the data. This problem is also evidenced by the fact that the uncertainties in the elastic moduli we obtained from the six-parameter fit are slightly higher than the uncertainties from the four-parameter fit (Table 6).

[40] Referring again to Table 6 and Figure 8, the best fitting densities, velocities, and effective elastic moduli of plagioclase, pyroxene, and amphibole are in excellent agreement (i.e., within one standard deviation) with the values reported in the literature. The estimated density and elastic moduli of olivine are lower than the elastic moduli of Fo65 or Fo73. However, the uncertainties in the olivine moduli are 15–20%, about twice as large as the errors in the estimated properties of plagioclase, pyroxene, and amphibole. If the errors were known to have Gaussian distributions, we could conclude that the best fitting moduli of olivine are not statistically distinguishable from the moduli of Fo65 and Fo73 at the 95% confidence level, but the distributions are not known. Thus an important question that arises here is whether the low best fitting properties of olivine are an artifact of the data or might have occurred by chance.

[41] Because the elastic properties of minerals vary with mineral composition, and there is a considerable variation of mineral compositions in the oceanic gabbros, it follows that those variations of composition will contribute to the uncertainties in the model parameters. The question is, how much? The properties of olivine are particularly sensitive to composition (Figure 8). Taking the most extreme case, in which the compositions are uniformly distributed between Fo0 and Fo100, we find that the average VRH velocity is 7.6 km s−1, with a standard deviation of 0.5. The olivine composition within any given sample from Hole 923A is essentially constant; variations occur from sample to sample. That being the case we can estimate the standard error (SE) of the mean velocity (i.e., the expected uncertainty in the best fitting velocities of olivine) for 44 samples, and find it to be ±∼0.1 km s−1. In view of the fact that this number likely to be a maximum, compositional variations are likely to account for at most a quarter of the variance of the best fitting model parameters.

[42] We also considered the possibility that the best fitting olivine values are poorly constrained and/or subject to systematic error because olivine is not abundant in these samples, but that is unlikely because average olivine content for this data set (8%) is comparable to the abundance of the amphibole (9%), for which the physical properties are well constrained. Though the range of values for olivine (0–28%) is somewhat smaller than the range for amphibole (0 to 40%), more than half of the samples (24 of 44) have amphibole concentrations less than 5%, while more than half the samples (27) have olivine concentrations greater than 5%.

[43] To explore the possibility that the high uncertainties and low estimated density and elastic moduli olivine occurred by chance, we conducted a numerical experiment. We first calculated the VRH bulk moduli of the samples we used for the inversion from their modal mineralogy, using mineral properties from Hacker et al. [2003]; we assumed the olivine to be Fo78, the plagioclase to be An50 and the amphibole to be hornblende. We used the properties of talc for the phyllosilicates and the properties of magnetite for the opaque minerals. To match the misfit we observed in the inversion, we added 3% Gaussian error to the modal mineral contents and 6% error to the calculated bulk modulus of each sample, then found the best fitting mineral bulk moduli by least squares, as described above, and repeated this procedure 100 times. Because the model errors are Gaussian, we expect the average of the best fitting parameters to be in close agreement with the input values; that is, the occurrence of a systematic errors would indicate a serious error in the method. In fact, both the misfits and the standard errors of the best fitting parameters are in good agreement with the inversion of the real data, suggesting that the numerical model mimics the real process well. From the inversion of the real data, we obtained a best fitting bulk modulus for olivine that is 1.5 times the standard error lower than the nearest reasonable value, that for Fo73 (see Table 6), while the values for plagioclase, pyroxene, and amphibole were within one SD of the mineral values. The numerical experiment produced only one comparable set of model parameters. These results indicate that the effective elastic moduli of the olivine in the samples are, in fact, low, even relative to the properties of Fo73.

[44] As noted above, the fact that olivine grains are preferentially cracked relative to the other minerals in the rock contributes to the low olivine density and elastic moduli. Most olivine grains observed in thin section are cracked; indeed, a network of cracks is one of the diagnostic features by which olivine is commonly recognized in thin section. Cracks are known to cause a large reduction of the elastic moduli of rocks [Kuster and Toksöz, 1974; Cheng and Toksöz, 1979; Carlson and Gangi, 1985; Gangi and Carlson, 1996]. Thus, unless the cracks occur only in thin section, and not in the samples, they will certainly affect the properties of the samples, and because the cracks are specifically associated with the olivine grains, their effects will be reflected in the estimated effective elastic moduli of olivine in the samples.

[45] Whether cracks populate olivine grains in the lower crust is not known. Because we use the properties of gabbro samples measured in the laboratory to interpret seismic velocities in the lower oceanic crust, it is important to determine how the seismic properties of the gabbros relate to their mineral contents if the olivine grains have “normal” densities and elastic moduli (i.e., if the olivine grains are uncracked in situ). To address that question, we corrected the VRH data set for the olivine density and moduli to the Fo73 values in the following way. The Voigt and Reuss average elastic moduli that we obtain by forward calculation are nearly the same (kv/kr = 1.01 ± 0.01 and μvr = 1.05 ± 0.01). That being the case, we can use the Voigt average to correct the moduli computed from the measured properties of the rocks

display math

where Mc is the corrected modulus of the rock, M is the measured modulus, MOL is the modulus of Fo73, and MOL′ is the modulus obtained from the VRH fit to the data (Table 6). Densities were corrected in the same way. By this method, we corrected the effective density and elastic moduli of olivine, but retained the effects of other minerals and cracks in the rocks that we cannot account for by forward modeling. We then calculated the corrected velocities and analyzed the velocities and modal analyses as before, by binning the data in 0.1 km s−1 intervals, and averaging over each bin.

[46] The results are illustrated, with those for the uncorrected data, in Figure 7. This point must be emphasized: correcting the properties of olivine to values appropriate for Fo73 makes no significant difference in the relationship between P wave velocities and the modal mineralogy of the oceanic gabbros. In the corrected data, samples with velocities below 6.9 km s−1 have average olivine contents less than 5%, while those with higher velocities have olivine contents near 10%. Aside from that, there is no meaningful difference between these two figures. In both cases, higher velocities indicate increasing pyroxene concentrations and declining concentrations of plagioclase and hydrous phases, particularly amphiboles; and in both cases velocities typical of the lower oceanic crust are associated with samples containing 10–30% hydrous alteration products, with a mean value near 15%. Hence, even if the olivine grains in the oceanic gabbro samples have the physical properties of Fo73, the relationship between P wave velocities and modal mineralogy would be essentially the same as that for the uncorrected data. In either case, gabbro samples that have velocities typical of seismic layer 3 (6.7–7.0 km s−1), are characterized, on average, by ∼15% hydrous phases, consisting mostly of amphibole, with a smaller fraction of phyllosilicates (clays, chlorite, or talc), and seismic velocity is not a good indicator of the abundance of olivine in the rocks.

7. Discussion and Conclusions

[47] The principal points to be drawn from our analysis of the oceanic gabbro samples from ODP Hole 735B, 894G, and 923A are the following:

[48] 1. Seismic P wave velocities in oceanic gabbro samples show a clear dependence on modal mineralogy. On average, higher velocities correspond to higher pyroxene contents and lower total alteration. In the range of P wave velocities that is typical of the lower oceanic crust (6.7–7.0 km s−1), the degree of alteration (phyllosilicate plus amphibole content) is 5 to 15% (Figure 6). The measured velocities bear essentially no relationship to olivine content, which ranges from 0 to 40%, with an average near 8%. These relationships hold for the full data set of 64 samples, over a range of pressures from 40 to 200 MPa, and for the subset of 44 samples we have used to solve for the best fitting elastic moduli (Figures 6 and 7), leading us to conclude that the correlations we observe are fairly robust.

[49] 2. The best fitting elastic moduli and densities of pyroxene, amphibole and plagioclase that we estimated by nonlinear least squares, are in quite good agreement with the isotropic elastic moduli of these phases. The best fitting values for olivine are low (Table 6 and Figure 8). The chemistry of the olivine, and the cracks that are characteristic of olivine in thin section may both contribute to the low moduli. Whether the cracks are present in situ is not known, but we note that in thin sections from the deepest part of Hole 735B (deepened in 1997 to 1500 m [Dick et al., 1999]) olivine grains show extensive networks of cracks. An example is shown in Figure 9. Owing to the low effective elastic moduli and density of olivine, the expected influence of olivine on seismic velocities is masked, in part, by a negative correlation between olivine content and the abundance of pyroxene in the rocks.

Figure 9.

Photomicrograph illustrating crack networks in an olivine grain in gabbro sample 176-735B-204R-7, 39 cm, recovered from a depth of 1500 m in ODP Hole 735B [Dick et al., 1999]. Note the absence of cracks in the surrounding material.

[50] 3. Correcting the moduli for the low effective density and moduli of olivine, and repeating the analysis, we find that the relationship between velocities and modal mineral contents is essentially unchanged (Figure 7).

[51] Turning to the implications of these findings for the composition and state of alteration of the lower oceanic crust (seismic layer 3), we emphasize two points. One is that in using the properties of gabbros sampled at ODP Sites 735, 894, and 923 to interpret the seismic structure of the oceanic crust, we assume that the lower crust is gabbroic in composition, and that the range of mineral compositions in the sample set spans the range of mineral compositions that occurs in situ, but we do not assume that the distribution of velocities and modal analyses of gabbros in the crust is identical to our data set. The second point to be made here is that we do not assume that the lower crust is homogeneous; both the primary composition and the state of alteration are known to be variable, even on the scale of a thin section. Seismic velocities reflect compositional variations averaged over volumes with dimensions of several hundred meters, at least. At smaller scales, the composition and state of alteration of the lower crust may be highly variable, and hydrothermal alteration is likely to be concentrated around faults and fractures or cracks on a wide range of scales.

[52] We note that the velocity/depth profiles estimated from the average velocities in the oceanic gabbros, measured as a function of pressure, are in very good agreement with most seismic profiles (Figures 1 and 4). Moreover, the high-velocity gradients that are typical of the upper crust (layer 2) commonly end where the seismic profiles approach the gabbro curves. The properties of the oceanic gabbros are consistent with the in situ seismic properties of the lower oceanic crust (Figures 1 and 3).

[53] The observed relationship between seismic velocities and modal mineralogy has particularly important implications for the composition of the lower oceanic crust. Samples that have velocities typical of the lower crust (6.7–7 km s−1) contain, on average, 5 to 15% alteration products (5–15% amphiboles and 1–3% phyllosilicates; see Figure 6). This observation implies that, on average, the oceanic crust is hydrothermally altered from the top of basement to the Moho, as implied or suggested by previous investigations [e.g., Salisbury and Christensen, 1978; Spudich and Orcutt, 1980; Hacker at al., 2003]. Whereas Hacker et al. [2003] suggest that the degree of alteration could be as high as 62%, our results show that total alteration of these rocks is much lower - probably less than 15% and certainly less than 27% (see below). We must emphasize the statistical nature of this finding. The average composition is the most probable composition of gabbros having velocities typical of the lower oceanic crust, based on the samples in the data set, but it is not unique.

[54] The fact that there is such good agreement between seismic profiles and the gabbro velocity-depth profiles (Figures 1 and 4) suggests that the properties of gabbros that have the average composition of the samples in our data set are also consistent with the seismic properties of the lower crust. On average, the samples are characterized by ∼11% total alteration (9% amphibole minerals and 2% phyllosilicates). These numbers are consistent with the percentages cited above. The gabbros from Hole 894G have velocities that are slightly lower than the velocities that typify most seismic profiles, and therefore may represent a lower velocity bound on the composition of the lower crust. The 894G gabbros contain 27% total hydrous alteration products (4% phyllosilicates and 27% amphiboles).

[55] At the other extreme, the average properties of the gabbros from Hole 923, which have a mean P wave velocity of 6.93 km s−1 and contain only 2.3% alteration (Table 2), also match the seismic properties of the lower crust (Figure 4). Thus, at least some nearly unaltered gabbros have seismic properties that are consistent with seismic velocities in the lower oceanic crust. Given that fresh gabbros should have velocities in excess of 7 km s−1, this observation begs for explanation. Something other than alteration must serve to lower the velocities in the rocks from Hole 923A. Relative to the rest of the rocks in our data set, those from 923A are at least 10% enriched in plagioclase and 43% enriched in olivine, both of which have lower elastic moduli and seismic velocities than pyroxene, if we (again) use the properties of Fo73 (Table 6 and Figure 8). We suspect that the relative enrichment in olivine and plagioclase causes these rocks to have lower seismic velocities than we might expect given their low degree of hydrous alteration. To test this hypothesis we estimated the velocity-depth profile for gabbros from Hole 923A after correcting for the low density and elastic moduli of olivine to the Fo73 values. The uncorrected and corrected curves are shown in Figure 10. The correction increases the velocities in the Hole 923A gabbros by ∼0.07 km s−1, making them consistent with the highest lower crustal velocities (e.g., ESP 18, Figure 4c). It follows that comparatively fresh gabbros may be the principal constituents of the lower oceanic crust if there is appreciable plagioclase, and the olivine has Mg numbers that are consistent with the Mg numbers of the gabbros that have been recovered by drilling.

Figure 10.

Comparison of seismic profiles with velocity/depth profiles computed from the measured velocities and velocities corrected for low olivine density and elastic moduli in gabbros from 923A. For comparison, seismic profiles ESP4S [Vera et al., 1990], ESP1 [Lindwall, 1991], and EPR 9N [Vera et al., 1990] are shown.


[56] This paper benefited from many fruitful discussions with Peter Kelemen and Julie Newman and from the constructive comments of three anonymous reviewers and the Associate Editor. This work was supported in part by NSF grant OCE-0221250.