We present a unified analysis of physical properties of cataclastic fault rocks collected from surface exposures of the central Alpine Fault at Gaunt Creek and Waikukupa River, New Zealand. Friction experiments on fault gouge and intact samples of cataclasite were conducted at 30–33 MPa effective normal stress (σn′) using a double-direct shear configuration and controlled pore fluid pressure in a true triaxial pressure vessel. Samples from a scarp outcrop on the southwest bank of Gaunt Creek display (1) an increase in fault normal permeability (k = 7.45 × 10−20 m2 to k = 1.15 × 10−16 m2), (2) a transition from frictionally weak (μ = 0.44) fault gouge to frictionally strong (μ = 0.50–0.55) cataclasite, (3) a change in friction rate dependence (a-b) from solely velocity strengthening, to velocity strengthening and weakening, and (4) an increase in the rate of frictional healing with increasing distance from the footwall fluvioglacial gravels contact. At Gaunt Creek, alteration of the primary clay minerals chlorite and illite/muscovite to smectite, kaolinite, and goethite accompanies an increase in friction coefficient (μ = 0.31 to μ = 0.44) and fault-perpendicular permeability (k = 3.10 × 10−20 m2 to k = 7.45 × 10−20 m2). Comminution of frictionally strong (μ = 0.51–0.57) cataclasites forms weaker (μ = 0.31–0.50) foliated cataclasites and fault gouges with behaviors associated with aseismic creep at low strain rates. Combined with published evidence of large magnitude (Mw ∼ 8) surface ruptures on the Alpine Fault, petrological observations indicate that shear failure involved frictional sliding within previously formed, velocity-strengthening fault gouge.
 The Alpine Fault accommodates Pacific-Australian plate boundary convergence on a single northeast-southwest striking structure with a surface trace at least 800 km long and a cumulative offset of ∼470 km [Wellman, 1953]. The central segment of the Alpine Fault, between Haast River and Toaroha River, accommodates ∼70% of the 37 ± 2 mm/yr relative motion between the plate boundaries (Figure 1) [Sutherland et al., 2007]. Detailed field mapping within this segment shows that in the near surface, the Alpine Fault is serially partitioned into alternating oblique thrust and strike-slip segments with average strikes ranging from 010° to 050° and 070° to 090°, respectively [Norris and Cooper, 2007]. Along-strike dimensions of the segments (typically ≤3 km) correspond to the spacing of major rivers [Norris and Cooper, 1997], and the segmented structures likely merge into a single oblique structure at a depth comparable to the topographic relief (∼1000–1500 m) [Townend et al., 2009].
 Multiple lines of geological, seismological, and paleoseismological evidence indicate that accumulated elastic strain on the central segment is released every few hundred years in Mw 7.5–8 earthquakes with horizontal displacements up to 8 m and vertical displacements up to 1.5 m [e.g., Sutherland et al., 2007, and references therein]. Geodetic studies further indicate that strain release is entirely coseismic, with interseismic creep at or near zero from the surface to 13–18 km depth [Beavan et al., 2010]. Extremely rapid (∼5–10 mm/yr) exhumation along the central segment has exposed a sequence of mylonites, ultramylonites and cataclastic (brittle frictional) fault rocks in the uplifted hanging wall [Little et al., 2005]. Displacement on the shallowest part of this section of the Alpine Fault (<1 km depth) is localized in a narrow zone of fault gouge and cataclasite. Outcrops of shattered, hydrothermally altered mylonites that have been thrust over Quaternary gravels yield estimates of strike-slip displacement rates of ∼25 mm/yr and dip-slip displacement rates of <10 mm/yr [Norris and Cooper, 2007].
 Our study combines new petrological observations with the first laboratory measurements of hydrological and frictional properties of cataclastic fault rocks exposed on oblique thrust segments of the Alpine Fault at Gaunt Creek and Waikukupa River, South Island, New Zealand (Figure 1). X-ray diffraction (XRD) results quantify in situ and along-strike changes in the mineralogy of fault gouges. Numerical constraints on the frictional and hydrological properties of the fault gouges and cataclasites also inform a discussion of the mechanisms governing earthquake rupture nucleation and propagation on this seismically active crustal-scale structure.
2. Surface Exposure Fault Rocks
 Over the last 5–8 Ma, oblique slip on the Alpine Fault has exhumed deformed rocks from depths up to 35 km (Figure 1) [Little et al., 2005]. Along-strike, brittle-ductile fault rocks form in a near-homogeneous quartzofeldspathic Alpine Schist protolith containing small amounts of metabasite and metachert. The 1 km thick mylonite sequence become progressively deformed with increasing proximity to a 10 to 50 m thick damage zone containing fault rocks in sharp, striated, faulted contact with fluvioglacial gravels [Norris and Cooper, 2007]. In the hanging wall, pseudotachylytes (solidified friction melts) are preserved in several settings within the mylonite-cataclasite sequence [Toy et al., 2011]. The widespread occurrence of pseudotachylyte indicates that earthquakes nucleate and propagate dynamically within these rocks in the upper crustal seismogenic zone [Sibson and Toy, 2006]. Exposures in outcrop, trench, and borehole sections provide further evidence that earthquake ruptures propagate to the Earth's surface within a thin (<60 cm), moderately dipping (38–45°) plane of fine-grained fault rocks (Figure 2).
 In this study, we systematically sampled rocks forming the fault core and hanging wall of a prominent, well-documented surface outcrop of the Alpine Fault on the south bank of Gaunt Creek (hereafter referred to as the GC scarp exposure) (Figures 2a–2c) [Cooper and Norris, 1994; Warr and Cox, 2001]. We also collected fault gouge from a new exposure of the Alpine Fault located in an excavation adjacent to a terrace riser 360 m northeast of the GC scarp exposure (hereafter referred to as the GC terrace exposure) (Figure 2d).
 At Gaunt Creek, quartzofeldspathic mylonites are thrust over fluvioglacial gravels 14C dated at circa 12,650 years B.P. [Cooper and Norris, 1994]. Four general types of fault rock are found adjacent to and within the fault core; each rock type sampled contains microstructural evidence for strain localization and cataclasis sensu stricto (Figure 2b) [Sibson, 1977]. These cataclastic fault rocks are described below, following the fault rock nomenclature of Sibson . Unit 1 is a gray-brown cataclasite derived from Quaternary fluvioglacial gravels deposited on the Australian plate. Unit 1 varies in thickness from 0 to 40 mm, forms a gradational contact with the underlying clast-supported gravels, and contains ≥60% matrix grains (matrix grains are by definition <0.1 mm in diameter). Unit 1 is absent where cobbles or boulders form the contact with Units 2 or 3.
 Unit 2 is a discontinuous gray-brown, fine-grained fault gouge that overlies and forms a sharp contact with Unit 1. In the GC scarp exposure, the contact is oriented 059/42°SE with a prominent set of 0/059 striations overprinted by a faint 36/114 set. Unit 2 varies in thickness and color; where present, it is always less than 2 cm thick and contains ≥90% matrix grains. Visible grains include quartz, feldspar, carbonate, opaques and phyllosilicates set in a fine-grained matrix cemented with cryptocrystalline carbonate and amorphous iron oxide. Rare clasts include subrounded reworked fault gouge, metamorphic quartz and vein quartz. Unweathered Unit 2 gouge has a foliation defined by phyllosilicates, with uniform extinction and color viewed through the sensitive tint plate, indicating that this mineral aggregate has a crystallographic preferred orientation (Figure 2e). Unit 2 weathers rapidly in outcrop, becoming indurated, stained with iron oxides, and fractured by numerous empty tension cracks. This alteration makes sampling intact wafers impractical.
 Unit 3 is a gray incohesive fault gouge that weathers to a brown indurated material; it forms a sharp undulating contact with either Unit 1 or Unit 2 (Figures 2a–2d). Hand specimens and thin sections of Unit 3 display variations in grain size, with some samples containing >30% visible clasts >1 mm. Clast long axes are tilted toward the SE with respect to the fault plane, indicating a top-to-the-northwest shear sense (Figure 2b). In the GC scarp exposure, Unit 3 varies in thickness between 6 and 8 cm and occasionally forms irregular contacts with Unit 4 (Figure 2c). In the GC terrace exposure, Unit 3 reaches a maximum thickness of 60 cm (Figure 2d). Because it appears continuous along strike, Unit 3 is the fault core lithology sampled and tested for permeability and frictional properties in this study.
 Unit 4 is a pale green cataclasite that is foliated in places. Unit 4 cataclasite contains fractured, sometimes sericitized, feldspars and recrystallized quartz aggregates with undulose extinction and finely sutured grain boundaries, separated by a faint anastamosing foliation defined by fine-grained chlorite, illite-muscovite and calcite (Figure 3a). The cataclasite commonly contains altered pseudotachylyte and/or ultracataclasite [e.g., Toy et al., 2011].
 Near the contact with Unit 3, the lower 4–8 cm of Unit 4 contains fault plane-parallel layers of finely comminuted incohesive cataclasite up to 1 cm thick oriented 059/50°SE with striations plunging 44/117; this lithology is referred to as the Unit 4 foliated cataclasite (Figure 3b). Microstructures in Unit 4 provide evidence of multiple deformation mechanisms; each deformation mechanism would have dominated at different conditions of strain rate, temperature and pressure. Unit 4 is the hanging wall lithology sampled for permeability and frictional properties in this study.
 For comparison, we sampled Unit 3 fault gouge and Unit 4 cataclasite from a surface exposure of the Waikukupa Thrust located 25 km to the southeast of Gaunt Creek (Figure 1). The Waikukupa Thrust emplaces Pacific Plate mylonites and cataclasites over Australian Plate fluvioglacial gravels older than 40,000 years B.P. The Waikukupa Thrust was abandoned approximately 20,000 B.P. when slip was transferred to the Hare Mare imbricate thrust [Norris and Cooper, 1997]. Where sampled, Waikukupa Thrust Unit 3 is an 8 cm thick, reverse-graded fault gouge oriented 055/56°SE; the fault gouge forms a sharp contact with underlying footwall gravels with striations plunging 15/066 (Figure 2e). Clasts within the Unit 3 fault gouge include subangular to subrounded unaltered to retrogressed quartzofeldspathic and metabasic mylonite set in a fine-grained matrix of quartz, feldspar, albite, chloritized biotite and amphibole, sericitized feldspar, epidote, titanite, and calcite (Figure 3f). Unit 4 is a cataclasite with mylonite clasts that have been subject to in situ fragmentation, translation and rotation in a matrix of comminuted quartz, feldspar, illite-muscovite, chlorite, calcite and pyrite.
3. Analytical Methods
 For quantitative X-ray diffraction, approximately 1.5 g of each oven-dried sample was ground for 10 min in a McCrone micronizing mill under ethanol. The resulting slurries were oven-dried at 60°C, thoroughly mixed with an agate mortar and pestle, and lightly back pressed into stainless steel sample holders. XRD patterns were recorded in steps of 0.017°2θon a PANalytical X'Pert Pro Multipurpose Diffractometer using Fe-filtered Co Kαradiation, variable divergence slit, 1° antiscatter slit and fast X'Celerator Si strip detector. Quantitative analysis was performed using the commercial package SIROQUANT. Smectite identification was done using XRD patterns obtained from oriented slides of Mg-saturated, glycerolated <2 μm clay separates. Results are listed in Table 1.
Table 1. Summary of Surface Outcrop Alpine Fault Rock Mineralogy
 Large blocks of fault core gouges (Unit 3) and adjacent hanging wall cataclasites (Unit 4) were collected from the outcrop using a portable rock saw. These blocks were cut into wafers 6–8 mm thick, 54 mm wide, and 61 mm long; the wafer long axis was cut parallel to the fault shear direction as indicated by striation measurements. A total of 10 double-direct shear friction experiments were conducted on Gaunt Creek and Waikukupa River fault rocks (Table 2).
Table 2. Summary of Experiment Details and Resultsa
Effective normal stress (σn′) represents the combined effects of applied normal load, confining pressure (Pc) and pore pressure (Pa).
 We deformed fault rock samples in a servohydraulic controlled biaxial testing apparatus fitted with a pressure vessel [Samuelson et al., 2009; Ikari et al., 2011b]. We sheared all samples under saturated, drained conditions at room temperature. We held effective normal stress (σn′), confining pressure (Pc), and pore pressure (Pp) constant at ∼31 MPa, 25 MPa, and 10 MPa, respectively (Table 2). These variables are geologically reasonable if we assume hydrostatic pore fluid pressure, an effective normal stress appropriate for 1 km depth on an oblique thrust fault striking NW-SE, and a stress tensor withσ1 = 45 MPa, σv = σ2 = σ3 = 25 MPa. Along the Alpine Fault, the maximum compressive stress (σ1) is subhorizontal trending 120° and the least principal stress (σ3) is subhorizontal trending 030°, with some local variance between σv = σ2 and σv = σ3expected for a combination of strike-slip and reverse faulting [Leitner et al., 2001].
 Flow-through permeability tests were conducted prior to shearing intact wafers of fault rock. In each experiment, two wafers were placed between a three-piece steel block assembly fitted with porous, stainless steel frits and then jacketed (seeSamuelson et al.  for a complete description of the experimental apparatus and methods). In the pressure vessel, normal (σn), confining (Pc), and pore fluid (Pp) pressures were applied to the sample, resulting in a true triaxial stress state. Pore fluid (University Park, Pennsylvania, USA tap water with pH = 7 and total cation concentration [Ca2+ + Mg2+] = 2–4 mmol/L) was plumbed directly into the fault rock samples and allowed to equilibrate until steady state boundary conditions were reached. We imposed a differential fluid pressure to induce flow normal to the shear direction of the intact wafers. When flow rate reached steady state, permeability was calculated using Darcy's law.
 In the double-direct shear configuration, fault rock samples are sheared by driving the center block with a servo controlled vertical piston at a constant displacement rate to attain steady state frictional behavior (Figure 4a). At steady state, the coefficient of sliding friction μ was determined from the ratio of shear stress τ to effective normal stress σn′ assuming cohesion of zero. We then performed velocity step tests (Figure 4b) to determine the rate and state dependence of friction. Load point velocity was varied between 1 and 100 μm/s, and in some cases 300 μm/s, to measure the friction rate parameter (a-b) = Δμ/Δ ln V, where a is the direct effect, b is the evolution effect, Δμ is the change in steady state sliding friction, and V is the sliding velocity [e.g., Marone, 1998] (Figure 4b). We also carried out a series of slide-hold-slide tests to measure the frictional healing rate. In these tests, the load point was driven at 10 μm/s, held motionless for a prescribed time, t, between 1 and 300s and then driven at 10 μm/s again. Frictional healing Δμ is the difference in peak friction, following a hold, relative to the steady state sliding friction prior to the hold. The healing rate is the change in Δμ per factor of 10 change in hold time measured in seconds (Figure 4c).
 All permeability measurements were made on intact wafers of cataclastic fault rocks at effective normal stresses between 30 and 31 MPa. The fault-perpendicular permeability of Unit 3 fault gouges varies betweenk = 1.16 × 10−20 m2 and k = 7.45 × 10−20 m2 (Table 2); these values are typical for natural and experimental fault gouges [e.g., Faulkner and Rutter, 2000, 2001]. The fault-perpendicular permeability of Unit 4 is almost 3 orders of magnitude greater than Unit 3, increasing fromk = 1.41 × 10−18 m2 in the Gaunt Creek Unit 4 foliated cataclasite to k = 1.15 × 10−17 m2 in Gaunt Creek Unit 4 cataclasite and k = 1.23 × 10−16 m2 and k = 1.45 × 10−17 m2 in Waikukupa Thrust Unit 4 cataclasites. This increase in permeability corresponds with a decreasing abundance of phyllosilicate layers and an overall increase in mean grain size (Figures 2 and 3). These results are consistent with measurements on similar fault rocks collected from the Nojima Fault Zone [Lockner et al., 2000] and Median Tectonic Line in Japan [Wibberley and Shimamoto, 2003].
 Values of steady state sliding friction coefficients for fault core (Unit 3) and hanging wall cataclasites (Unit 4) range from μ = 0.31 to 0.57 (Table 2 and Figure 5a). The Unit 3 fault gouges are the weakest, with μ = 0.31 and μ = 0.44 for intact wafers of fault gouge collected from the GC terrace and GC scarp exposures, respectively. The friction coefficients of Waikukupa Thrust fault gouge, μ = 0.39 and μ = 0.41, lie within this range. Hanging wall cataclasites are stronger, with intact wafers of Unit 4 foliated cataclasite having a μ = 0.50. Intact wafers of Gaunt Creek and Waikukupa Thrust cataclasites have friction coefficients of μ = 0.51–0.55 and μ = 0.54–0.57, respectively. The powdered sample of Gaunt Creek Unit 3 has a higher friction coefficient, μ = 0.57, than its wafer equivalent, which is consistent with recent experiments showing that phyllosilicate rich rocks are generally stronger in powder form [Collettini et al., 2009; Ikari et al., 2011b].
5.3. Friction Rate Dependence
 We performed velocity step tests in six experiments to determine rate/state friction constitutive parameters (Figure 4b). This parameterization is useful for inferring whether shear will be stable or unstable under conditions of tectonic faulting [e.g., Marone, 1998]. The friction rate parameter, a-b, describes the variation in steady sliding friction with slip velocity. Materials are said to be velocity strengthening if a-b > 0; these materials tend to slide stably, accommodating strain aseismically. Materials exhibiting negative a-b values are termed velocity weakening; these materials may undergo a frictional instability that results in earthquake nucleation and seismic rupture propagation [Scholz, 2002].
 Results from velocity step tests are summarized in Figure 5b. Jacket failure precluded measuring the friction rate parameters and frictional healing rates of intact wafers of Unit 3 from the GC scarp exposure; a powdered sample of Unit 3 from that locality exhibited velocity-strengthening behavior (a-b = 0.00012–0.0016). Unit 3 fault gouges collected from the GC terrace exposure and Waikukupa Thrust displayed clear velocity-strengthening behavior (a-b = 0.0052–0.018) (Figures 4b and 5b). Gaunt Creek Unit 4 foliated cataclasite also exhibited velocity-strengthening behavior (a-b = 0.0019–0.016). These units also show strong positive a-b rate dependence, a behavior observed in other clay gouges [Ikari et al., 2011a].
 Intact wafers of Unit 4 cataclasites a range of a-b values, depending on the magnitude of the velocity perturbation. Values of a-b for the cataclasites range between –0.0016 and 0.00082. We observe no systematic relationship between the size of the velocity perturbation and the value of a-b in the cataclasites (Figure 5b). Other frictionally strong gouges (μ ≥ 0.5) commonly exhibit both velocity-strengthening and velocity-weakening behavior in velocity step experiments [Ikari et al., 2011a].
5.4. Frictional Healing
 Frictional strength varies with the real area of contact. Under stationary conditions, asperity contacts can heal with time, increasing the likelihood of unstable frictional sliding in some materials [Marone, 1998; Scholz, 2002]. Slide-hold-slide tests were performed to measure the rate of frictional healing (Figure 4c). As in the velocity stepping experiments, data on GC scarp exposure Unit 3 are from gouge powder, which healed at a faster rate than intact wafers of all lithologies (Figure 5c).
 Intact wafers of GC terrace exposure Unit 3 and Unit 4 foliated cataclasite weakened with hold time for all hold times between t = 3s and t = 100s (Figures 4c and 5c). The gouge layers have very low permeabilities, and consolidation during hold time may have resulted in elevated pore fluid pressures and thus lower friction coefficients [e.g., Byerlee, 1993; Faulkner and Rutter, 2001]. However, we observe no systematic correlation between hold time and frictional healing across the range of hold times imposed. The results of the GC terrace exposure Unit 3 experiment are discussed further below.
 GC scarp exposure Unit 4 nonfoliated cataclasite, Waikukupa Thrust Unit 3 fault gouge, and Waikukupa Thrust Unit 4 cataclasite all showed a positive correlation between hold time and Δμ (Figure 5c). This behavior is consistent with mechanistic interpretations of healing in rocks composed of framework silicates, where the real area of surface contact increases with time because of contact junction growth and/or strengthening [Dieterich, 1978; Marone, 1998].
 Along thrust segments at Gaunt Creek and Waikukupa River, the east dipping Alpine Fault plane is overlain by 10–50 m of cataclasite comprising clasts of quartzofeldspathic and metabasic mylonites in a finely comminuted chloritized matrix. Near the striated basal contact with footwall fluvioglacial gravels, strain localization in the hanging wall has resulted in further grain size reduction and fault gouge formation. In places, mechanically aligned phyllosilicates form a planar fabric aligned subparallel to the fault plane (e.g., Figure 3e).
 Quantitative X-ray diffraction analyses indicate that strain localization and grain size reduction was accompanied by changes in the nature and abundance of phyllosilicate phases at the Gaunt Creek localities (Table 1). At the GC scarp exposure, fault gouges comprising Units 2 and 3 contain a high percentage of clay minerals (57–60%) relative to Unit 4 cataclasite (20%). Unit 4 cataclasite contains higher temperature clay minerals chlorite (clinochlore-1MIIb) and muscovite (2M1). Unit 3, however, contains the lower temperature clay minerals smectite (montmorillonite) and kaolinite as well as chlorite (clinochlore-1MIIb) [e.g.,Moore and Reynolds, 1997]. Based on mineralogy, microstructures and grain size, the Unit 4 foliated cataclasite contains components of both the cataclasite and fault gouge (Table 1) (Figure 3b). Basal Unit 2 gouges contain predominantly low-temperature phyllosilicates, with the exception of lizardite (1M) at the GC scarp exposure (Table 1). The origin of lizardite (1M) is unknown; it is possible this mineral was derived from footwall gravels or a protolith present downdip or along strike.
 Hand specimen and thin section identification of relict clasts in the fault gouges and cataclasites, along with X-ray diffraction analysis, indicate that comminution of hanging wall mylonites and protocataclasites formed the granulated cataclasites and gouges analyzed in this study. By an analysis of garnet fragments in cataclasites,Cooper and Norris  reached the same conclusion, but they could not discount a contribution from footwall granitoids at depth. Most mylonite hanging wall rocks are quartzofeldspathic, composed of the primary minerals: quartz, plagioclase, muscovite, and biotite with minor amounts of garnet, ilmenite, and calcite. Metabasic mylonites contain amphibole, as reflected in the mineralogy of the Waikukupa Thrust samples [Toy et al., 2010].
 In addition to primary minerals, GC exposure fault gouges and foliated cataclasite fault gouges contain kaolinite and the weak dioctahedral clay mineral smectite (Table 1) (μ = 0.15–0.32) [Saffer and Marone, 2003]. Warr and Cox found that smectite forms from low-temperature alteration in Alpine Fault mylonite-derived cataclasites and gouges. Smectite can form from low-temperature alteration (<120°C) of the primary minerals illite/muscovite and chlorite. In our samples, smectite occurrence coincides with a depletion in chlorite relative to adjacent hanging wall cataclasites. Oxidized alteration of chlorite forms smectite, which further alters into kaolinite and amorphous iron oxide (goethite) [e.g.,Craw, 1984; Chamberlain et al., 1999]; these minerals are all present in Units 2, 3 and 4 on the GC scarp exposure. Warr and Cox  suggested that the growth of smectite leads to mechanical weakening in Alpine Fault gouges, but its association with a secondary cement strengthens the GC scarp exposure fault gouge (μ = 0.44) relative to the GC terrace exposure fault gouge (μ = 0.31).
 GC terrace exposure Unit 3, GC scarp exposure Unit 4 cataclasite, Waikukupa Thrust Unit 3 fault gouge and Waikukupa Thrust Unit 4 cataclasite all contain the primary mineral phases quartz, albite ± orthoclase, calcite, illite-muscovite (2M1), and chlorite (clinochlore-1MIIB). These minerals are likely to remain stable to subgreenschist conditions (<320°C) near the base of the brittle seismogenic zone [Warr and Cox, 2001]. Using a geothermal gradient of 40°C, this equates to a depth of 8 km on the Alpine Fault [Toy et al., 2010]. At effective normal stresses >60 MPa, however, clay-rich gouge behavior becomes increasingly nonlinear, and pressure solution creep may compete with friction to accommodate grain sliding in clay-rich gouges containing a soluble mineral phase such as quartz or calcite [e.g.,Rutter, 1983; Bos and Spiers, 2002; Niemeijer and Spiers, 2007; Gratier et al., 2011]. Whether deformation is accommodated in the gouges via frictional sliding and cataclasis or pressure solution creep depends on many factors, foremost temperature and magnitude of differential stress resolved on the Alpine Fault [Rutter, 1976; Cox and Etheridge, 1989; Gratier et al., 2009].
 Our experiments were conducted at room temperature on saturated fault gouges loaded at effective normal stresses ∼30 MPa and loading rates between 1 and 300 μm/s. At these conditions, clay-rich Unit 3 fault gouges are frictionally weak, exhibit velocity-strengthening frictional behavior, and do not undergo frictional healing. The weak (μ = 0.31), velocity-strengthening (a-b = 0.0075–0.016), nonhealing (Δμ = –0.003 after t = 100 s) properties of the GC terrace exposure Unit 3 gouge (illite-muscovite 32%; chlorite 9%) best represents the frictional behavior of clay-rich fault gouges at depth.
Figure 4aillustrates that the GC terrace exposure Unit 3 fault gouge did not display peak friction. Experimental observations indicate that aligned layers of chlorite (clinochlore-1MIIb) and illite-muscovite (2M1) facilitated strain accommodation during the early stages of shearing, suppressing peak strength [Saffer and Marone, 2003]. At higher strains during the slide-hold-slid tests, little or no contact strengthening occurred in these aligned clay grains, and we observed no frictional healing with hold time [e.g.,Bos and Spiers, 2000].
 An additional factor influencing peak strength and frictional healing is dilation. Dilation is a physical response to shear-induced strain localization observed in many granular materials. By increasing the thickness of the shearing granular layer, dilation increases effective normal stress, decreases pore fluid pressures, and suppresses frictional instabilities. The lack of frictional healing in the GC terrace exposure fault gouge suggests that aligned clay layers do not exhibit classical healing behavior governed by a combination of frictional strengthening during the hold and dilantancy strengthening upon reshear. These units have very low permeabilities, and time-dependent compaction may result in high pore fluid pressures and undrained loading [Lockner and Byerlee, 1994; Garagash and Rudnicki, 2003; Faulkner et al., 2011; Schmitt et al., 2011].
 Unit 3 fault gouges also exhibit velocity-strengthening behavior, a frictional property that promotes stable (aseismic) sliding, inhibits earthquake rupture nucleation, and may contribute to earthquake rupture arrest at shallow depths (≲3 km) [Marone, 1998]. However, paleoseismological, geomorphological and geodetic evidence suggests that the Alpine Fault fails coseismically in large earthquakes, forming surface ruptures with an average slip per event of ∼8 m dextrally and ∼1 m vertically [Sutherland et al., 2007].
 Along the Alpine Fault, the upper stability transition, which defines the updip limit of the seismogenic zone, is either too shallow to arrest seismic rupture propagating from below, or the upper, stable region is not sufficiently velocity strengthening to arrest rupture [Marone, 1998]. Alternatively, theoretical considerations, numerical modeling and high-velocity rock friction experiments show that rapid undrained loading in low-permeability fault gouge results in thermal pressurization of pore fluids at slip rates greater than 0.1 m/s and displacements on the order of a few microns [e.g.,Sibson, 1973; Lachenbruch, 1980; Rice, 2006].
 Sufficient shear heating can induce thermal pressurization, a weakening mechanism that operates before seismic slip speeds are obtained [Schmitt et al., 2011]. Faulkner et al. presented experimental data on saturated clay-rich gouges sheared in a high-velocity rotary apparatus at low normal stress (1.63 MPa) and high slip velocity (1.3 m/s). They attributed the immediate frictional weakening to rapid thermal pressurization of the pore fluid, which resulted in a negligible critical slip weakening distance and fracture energy. A rapid loss of strength at the tip of an earthquake rupture may make propagation through clay rich gouges energetically favorable [e.g.,Tinti et al., 2005; Bizzarri and Cocco, 2006a, 2006b]. As slip speeds grow to within a few orders of magnitude of elastic wave speeds, however, additional factors contribute to the physics of earthquake propagation through velocity-strengthening material, including elastically radiated energy, fracture energy, and poromechanical effects [Rice, 1993; Lapusta et al., 2000; Perfettini and Ampuero, 2008].
 The inclusion of reworked clasts of fault gouge and hanging wall cataclasites in Unit 3 fault gouges indicates that large magnitude surface ruptures on the Alpine Fault have occurred repeatedly within the same fault core. Local pore fluid overpressures are indicated by the irregular contacts between Unit 3 fault gouge and Unit 4 cataclasite at the GC scarp exposure (Figure 2c). Questions remain, however, about whether these irregular contacts formed interseismically by creep, compaction, and pore fluid pressurization in the impermeable, clay-rich gouge or coseismically as gouge injection structures formed by shear-induced thermal pressurization [e.g.,Cowan, 1999; Meneghini et al., 2010; Lin, 2011].
 Oblique thrust fault segments of the Alpine Fault form surface outcrops at Gaunt Creek and Waikukupa River. Samples of cataclastic fault rocks were collected and investigated using a double-direct shear configuration under controlled pore pressure. At effective normal stresses between 30 and 31 MPa, fault core gouges have fault normal permeabilities up to 3 orders of magnitude lower than hanging wall cataclasites (Table 2). Fault core gouges are velocity strengthening and have lower friction coefficients (μ = 0.31–0.44) than hanging wall cataclasites. Cataclasites are frictionally strong (μ = 0.50–0.57), exhibit normal to high rates of frictional healing, and display both velocity-strengthening and velocity-weakening behaviors. These results support general models of mature fault cores as weak with respect to the surrounding crust [e.g.,Rice and Cocco, 2007; Carpenter et al., 2011].
 Current geodetic estimates indicate that the Alpine Fault is locked to a depth of 13–18 km, and total elastic strain accumulation is released in large magnitude (Mw7.5–8) earthquakes that propagate along-strike distances between 300 and 600 km [Beavan et al., 2010; Sutherland et al., 2007]. Petrological observations of clasts of reworked fault gouge in fault core rocks sampled in this study reveal that earthquake ruptures preferentially propagated through velocity-strengthening fault gouges exposed in surface outcrops at Gaunt Creek and Waikukupa River. Understanding the mechanisms facilitating earthquake rupture propagation through clay-rich gouges remains an important goal in rock mechanics and earthquake physics.
 The first author received support for this research from the Claude McCarthy Foundation and the University of Canterbury Roper Scholarship. We would like to thank to Sam Haines for doing preliminary X-ray diffraction analyses and Mark Raven for doing more detailed, quantitative analyses. Greg De Pascale is credited with discovering the Alpine Fault terrace exposure at Gaunt Creek. The laboratory work was supported by NSF grants OCE-0648331, EAR-0746192, and EAR-0950517 to C. Marone and Marsden grant UOO0919 to V. Toy. Reviews by Dan Faulkner, an anonymous reviewer, and Editor Thorsten Becker greatly improved the quality of this manuscript.