Shear-wave splitting of S waves from earthquakes in the Arakan slab is consistent with strong asthenospheric anisotropy that developed as a consequence of the India-Asia collision. Global positioning system (GPS) site velocities in the India-Asia-Sundaland triple junction region show that deformation along the Arakan subduction zone is partitioned into dextral strike-slip motion, as India moves northwards with respect to Asia, and contraction across the Arakan trench and Chittagong-Tripura fold belt. Indian Ocean lithosphere comprising the Arakan slab is dismembered into three segments as a result of its collision with Asian lithosphere at the East Himalayan syntaxis. Offsets of intermediate-depth earthquake hypocenters at two locations delineate slab segments that form a left-stepping en echelon structure. Arakan slab focal mechanisms are consistent with slab sinking and along-strike compression and bending, and, south of 25°N, dextral strike-slip along the slab. Two regions of N-S contraction within the slab appear to be localized at the slab segment offsets. Teleseismically recorded S waves from earthquakes within the three slab segments, and surroundings, are split systematically: once corrected for receiver-side splitting, fast shear trends are predominantly trench-parallel beneath the east-dipping slab segments; are more nearly trench-normal on the Sundaland (east) side of the Arakan lithosphere; parallel the southern ∼E-W gap between Arakan slab segments; and turn sharply around the extreme northern and southern edges of subducted Arakan lithosphere. Source-side shear-wave splitting beneath India is consistent with observed ∼E-W–trending fast shear polarizations of SK(K)S splitting in northeastern India. The general pattern of both surface site velocities from GPS and shear-wave splitting studies is consistent with material flow around the eastern Himalayan syntaxis and into the mantle wedge above the Arakan slab, and around the northern terminus of the Arakan slab. The upper mantle may also flow through the gap between the central and southern Arakan slab segments.

Despite its ongoing collision with Asia, begun in the Paleogene (e.g., Beck et al., 1995; Patzelt et al., 1996; Najman et al., 2008), India still moves northward some 3.9 cm/yr with respect to stable Eurasia (Socquet et al., 2006). Along its eastern boundary, the Indian plate subducts beneath western Sundaland (here, Indochina, Malaysia, and western Indonesia; Figs. 1 and 2) with varying, but high, obliquity along the Andaman (Vigny et al., 2005) and Arakan trenches (Cummins, 2007). Relative motion between India and overriding Indochina at the surface is dominantly dextral strike slip with some transpression along the Arakan portion of the subduction zone in Myanmar (Chamot-Rooke and Le Pichon, 1999; Simons et al., 1999, 2007; Michel et al., 2001; Nielsen et al., 2004; Socquet et al., 2006). Although the partitioning of GPS-determined deformation between permanent strain and elastic seismic cycle strain is unclear in Myanmar (but see Jade et al., 2007), geologic evidence for transpression and eastward subduction along the Arakan portion of the India-Sundaland includes well-exposed late Mesozoic ophiolite sequences (Acharyya, 2007), Neogene–Holocene deformation in the Chittagong-Tripura fold belt (Le Dain et al., 1984; Alam et al., 2003), active deformation of marine sediments in the Bay of Bengal segment of the subduction zone (Nielsen et al., 2004), and Quaternary–Holocene arc volcanoes developed on overriding Sunda lithosphere in central Myanmar (Maury et al., 2004). In addition, seismicity defines a well-developed east-dipping Wadati-Benioff zone down to ∼150 km (Ni et al., 1989; Guzman-Speziale and Ni, 1996; Dasgupta et al., 2003; Rao and Kalpna, 2005; Stork et al., 2008; Figs. 2 and 3), and N-S–striking thrust focal mechanisms of shallow earthquakes occur in the Chittagong-Tripura fold belt and Central Burma Basin (Fig. 4). At least one large-magnitude subduction earthquake is known to have occurred on the southern Arakan plate interface, in 1762 (Cummins, 2007). Travel-time inversions of arrivals at global seismic networks and regional stations in India, Tibet, and Yunnan indicate that the Arakan slab penetrates to a depth of 300–400 km, and may then flatten beneath Indochina (Li et al., 2008).

Arakan Slab Segmentation

Joint hypocentral relocations of Arakan slab earthquakes by Stork et al. (2008) show that the intermediate depth events are offset by nearly 100 km at two locations beneath Myanmar (Fig. 3). Whether these slab offsets represent contortions or tears in the slab at depths of 75–150 km is not clear, but that the Arakan slab is segmented, forming a set of three left-stepping en echelon portions, is unequivocal. Centroid moment-tensor focal mechanisms (Ekstrom et al., 2006) of Arakan slab events deeper than 40 km (Figs. 5 and 6) show that the slab is simultaneously sinking into the mantle (dip-slip events with ∼N-S strike), bending due to along-strike compression (mixed thrust and strike-slip mechanisms with W-plunging neutral axes; Le Dain et al., 1984; Ni et al., 1989; see Russo et al., 1993, for an explanation of the basic mechanism of this type of slab lateral bending), and south of 25°N, a few mechanisms with dextral strike slip on generally N-striking nodal planes appropriate to shear due to India's northward motion. Note that a group of these earthquakes that occurred within the region of the northern Arakan slab offset (∼25°N, 95.4°E) have nearly pure thrusting mechanisms with WNW-ESE–striking nodal planes (Fig. 6), and Ni et al. (1989) published very similar focal mechanisms for two events that occurred in the vicinity of the southern slab offset (∼22°N, 95.5°E).

The Arakan slab segmentation (Stork et al., 2008), the observed along-strike lateral compression (Le Dain et al., 1984; Ni et al., 1989), and prevalence of slab-bending earthquakes deeper than 70 km would be expected, given India's relatively rapid northward motion (Le Dain et al., 1984), if such motions drive the northern terminus of the Arakan slab into the East Himalayan syntaxis. Figure 7 shows a conceptual model of the slab segmentation and development of the left-stepping en echelon offsets based on the precise relocations of Stork et al. (2007) and the internal slab deformation indicated by available focal mechanisms. Slab contortion or tearing at the two offset locations allows clockwise rotation of the slab segments about steep axes, thus increasing the bending of the slab and shortening its effective length. The combination of N-S compression along the slab at the segment offsets and the general dextral shear couple produced by India's northward motion result in the classic en echelon secondary shear-band deformation of a competent layer in a weaker medium (e.g., Ramsay and Lisle, 2000, see their figures 34.12, 34.15, and accompanying discussion), albeit at subducted slab scale.

The deformation of the Arakan slab, apparently a result of the ongoing India-Asia collision, is a clear indication that the interaction of the two plates affects not only a wide region at the surface, involving crustal motion around the East Himalayan syntaxis and southeastward extrusion of Indochina (e.g., Molnar and Tapponnier, 1975, 1977; Tapponnier et al., 1982; Le Dain et al., 1984; Shen et al., 2005; Zhang et al., 2005; Tanaka et al., 2008), but also extends to considerable depth. Gradients of mantle isotopic tracers ranging from high concentrations beneath China and Tibet and diminishing systematically around the syntaxis indicate that apparent crustal flow was matched by similar upper-mantle extrusion (Flower et al., 1998, 2001; Deng et al., 2004). The form of upper-mantle deformation at depths down to the top of the transition zone (410 km) can be discerned from seismic anisotropy detailed using shear-wave splitting analyses (e.g., Silver, 1996; Savage, 1999). Recent work on splitting of SK(K)S and similar core-traversing waves shows that the continent-continent collision entails surface and upper-mantle material flow around the East Himalayan syntaxis from southeastern Tibet and Yunnan to an approximately E-W zone at ∼26°N beneath northern Indochina (Lev et al., 2006; Sol et al., 2007; Wang et al., 2008). Shear-wave splitting analyses from northeastern India are consistent with generally E-W upper-mantle fabrics south of Tibet and west of the syntaxis (Singh et al., 2006, 2007). These results were derived from data collected at temporary seismic deployments in China and India, and such deployments are not currently possible in Myanmar. However, the form of the upper-mantle deformation field south of the syntaxis in Myanmar and northeastern India can be determined using shear waves from earthquakes in the Arakan slab itself (e.g., Russo and Silver, 1994; Russo, 2009; Russo and Mocanu, 2009; Russo et al., 2010). In the following, I present results from such a study and show that the deformation of the Arakan slab strongly affects upper-mantle anisotropy locally, and that the larger-scale upper-mantle flow in the region is consistent with India's northward motion and flow around the northern and southern ends of the Arakan slab, and perhaps also through the southern slab gap.

Source-side shear-wave splitting measurements are made using teleseismic S waves (30° < Δ < 84°) recorded at global seismic network stations (i.e., Incorporated Research Institutions of Seismology [IRIS] Global Seismic Network [GSN], GEOSCOPE network, GEOFON network) whose substation upper-mantle shear-wave splitting parameters are known. These events must be large enough to be well recorded teleseismically (Mw >5.5) and deeper than 40 km to ensure that the reflected phases pS and sS do not arrive simultaneously with the long-period (10 s) direct S wave. Since 1990, when the broadband global network began to come online, 15 suitable Arakan slab events have occurred (Table 1). One of these events, the earthquake on Julian day 364 during 1997, occurred at a depth shallower than 40 km, but examination of S waves from the event shows little or no evidence of contamination by pS or sS, so it was also used in the study.

Shear-wave splitting that occurs along the downgoing travel path from the Arakan slab can be isolated because these S waves turn in the largely isotropic lower mantle (Meade et al., 1995). Shear-wave splitting at the core-mantle boundary (CMB) region appears to be small (e.g., Hall et al., 2004), and could potentially affect only the few S waves used in the study that travel to the extreme of the distance range (80°–84°). Contributions to splitting from the upper mantle deeper than the olivine stability field is infrequently observed and is attributed to preferred alignment of wadsleyite (Fischer and Wiens, 1996; Fouch and Fischer, 1996; Wookey and Kendall, 2004). Shear-wave splitting due to upper-mantle substation anisotropy can be determined using SK(K)S and PKS phases, and thus the splitting on the receiver side of the S-wave travel path (Fig. 8) can be corrected (Russo and Silver, 1994; Russo, 2009; Russo and Mocanu, 2009).

Observed receiver station splitting parameters used to correct for splitting beneath the receiver stations are detailed in Table 2. Station splitting corrections were applied when clear splitting of Arakan event S waves was observed at the receiver stations; however, when no splitting was observed (i.e., a linear S wave, or null splitting), receiver corrections were not applied. Note, only one receiver station used in this study is considered to be isotropic (KRIS, Schmid et al., 2004), and thus never required correction for receiver-side splitting. Details of the method for correcting known station splitting can be found in Russo and Silver (1994), Russo (2009), and Russo and Mocanu (2009).

The 15 suitable Arakan slab earthquakes (Table 1) yielded 103 receiver-corrected source-side splitting measurements and 152 null splitting observations (Table 3). Splitting nulls derive from observed linear waveforms—either before or after receiver splitting correction—and occur when either the initial S-wave polarization fortuitously parallels one of two possible anisotropic symmetry directions or the medium traversed is isotropic. An example of the splitting measurement before and after receiver correction is shown in Figure 9. Positive splitting results and nulls for each event are shown in Figure 10, and all splitting parameters are detailed in Table 3. The Arakan slab event that occurred on Julian day 209, 2008 (Table 1), yielded no suitable waveforms for splitting and will not be discussed further. Splitting delay times are generally high, with a mean of 3.0 s, similar to that found for source-side splitting in the Cascades region (2.9 s; Russo, 2009) and the Carpathian Arc (2.77 s; Russo and Mocanu, 2009). Results in Figure 10 are plotted at surface projections of the point at 200 km depth along their downgoing path from the source event to the receiver station. This procedure allows discernment of variable upper-mantle anisotropy in the source region (Russo, 2009; Russo and Mocanu, 2009), and is predicated on results of numerical studies of S-wave–effective Fresnel zones (Zhao et al., 2000) and the effects of slowly varying anisotropy on observed splitting (Saltzer et al., 2000).

As is clear from Figure 10, shear-wave splitting in the vicinity of the Arakan slab is highly variable, and appears to depend strongly on the upper-mantle volume sampled by the downgoing S waves. The source earthquakes differ in both location and depth, although the majority of the earthquakes occurred in the central segment of the Arakan slab at depths of ∼110–120 km. Typically, one of the two possible fast shear trends of observed splitting nulls, whether null before or after receiver splitting correction, is consistent with observed splitting fast axes for similar source-receiver paths. In several instances, I measured splitting of S waves recorded at stations at short distances from the source events (6°–8°). Although these waves could have acquired their splitting anywhere along their paths and are potentially subject to phase modifications due to shallow incidence (e.g., Crampin and Booth, 1985), results, once receiver corrected, are consistent with splitting measurements made at teleseismic station distances along similar azimuths (Figs. 10A, 10D, and 10E).

Shear-Wave Splitting and Upper-Mantle Anisotropy

Teleseismic shear-wave splitting is commonly associated with development of linear preferred orientation (LPO) in olivine-dominated upper-mantle aggregates. The LPO aligns olivine crystallographic a-axes (seismically fast) in the finite deformation shear plane parallel to the direction of tectonic extension (Hess, 1964; Carter et al., 1972; Gueguen and Nicolas, 1980; Christensen, 1984; Nicolas and Christensen, 1987; Ribe, 1989a, 1989b; Ribe and Yu, 1991; Zhang et al., 2000; Kaminski and Ribe, 2001; Jung et al. 2006). LPO in natural upper-mantle samples typically follows this basic type-A fabric (Mainprice and Silver, 1993; Ben Ismail and Mainprice, 1998), although petrographically distinct type-C and type-E fabrics that yield shear-wave splitting with fast polarizations in the material flow direction, similar to the A-type fabric, also exist (Jung et al., 2006).

The presence of water under high stress conditions may also complicate anisotropic fabrics, producing a distinctive B-type fabric (Jung and Karato, 2001; Karato, 2003; Jung et al., 2006), and the presence of melt apparently results in similar fabrics (Holtzman et al., 2003). Elevated water content and partial melt thus may modify LPO fabrics, yielding olivine b-axis concentrations in the shear plane and/or material extension direction, or girdles of crystallographic axes, instead of the usual a-axis clustering. Non-coaxial finite strain also typically yields more complicated anisotropic fabrics (Tommasi et al., 1999; Kaminski and Ribe, 2001; Blackman and Kendall, 2002).

The asthenosphere beneath the Arakan slab was likely thoroughly devolatilized during formation of the Indian lithosphere by ridge processes, and the asthenospheric channel beneath the subducted Indian lithosphere is therefore unlikely to include significant water. Although slab dewatering fluids may hydrate the suprasubduction mantle wedge and modify LPO (Jung and Karato, 2001; Bostock et al., 2002; Karato, 2003; Jung et al., 2006; Abt and Fischer, 2008), the mantle beneath the slab is unlikely to have been affected. Also, slow-velocity anomalies visible in the tomographic study of Li et al. (2008) are modest (1% slow) and inconsistent with presence of partial melt beneath the Arakan slab. Thus, a large majority of the shear-wave splitting results at this subduction zone can be linked to A-type anisotropy. Some ray paths (see below) do sample the deeper mantle wedge region, where the increased likelihood of hydration and partial melt fraction make the B-type fabric a potential alternative (e.g., Abt and Fischer., 2008). However, Kneller et al. (2008) show that the conditions for development of B-type anisotropic fabrics (high stress, high water content, and low temperature) are restricted to the shallow mantle wedge region, and so for all downgoing S waves used in this study that do not sample the shallow suprasubduction mantle wedge (see below), we adopt the a-axis olivine LPO model for interpretations.

Homogeneous Anisotropy Models

The sampling of the Arakan–Triple Junction region upper mantle achieved, given the locations of the source events and receiver stations, is heterogeneous, but still dense enough in several quadrants for quite a few of the events (see Fig. 10) to rule out interpretations of the observed splitting variations that invoke homogeneous plunging anisotropies. For example, neglecting the strong along-strike component of plate boundary zone motion discernible in both the earthquake focal mechanisms (Figs. 4–6; see also Le Dain et al., 1984; Ni et al., 1989; Stork et al., 2008) and the GPS results for the area (Socquet et al., 2006; Simons et al., 2007), a potential explanation for the variation in source-side S splitting results visible in Figure 11 is a single, regionally homogeneous anisotropic symmetry that plunges eastward in the Arakan slab dip direction, as might be produced by 2-D entrained upper-mantle flow at the subduction zone. Upper-mantle fabrics with this orientation should yield largely N-S–trending fast shear polarizations both west and east of the Arakan slab (Crampin and Booth, 1985; Chevrot and van der Hilst, 2003), which pattern is considerably simpler than that observed, both for individual events and all the events in aggregate.

Shear-Wave Splitting and Arakan Slab Segmentation

Figure 11, a compilation of all the source-side splitting results of this study, shows that the various source events, which differ in focal mechanism and often in depth, yield shear-wave splitting measurements that are largely consistent with respect to the upper-mantle volumes sampled by S waves traveling similar source-receiver travel paths. For example, splitting fast polarizations observed to the east of the northern Arakan slab segment trend NNW-SSE with delay times near 3 s, although the S waves derive from no less than six source events, and were recorded at receiver stations with distinct station anisotropy (see Tables 2 and 3). Similar consistency of results is observed for splitting measurements from groups of events sampling other portions of the Arakan slab–Triple Junction region. Potential fast and slow axes of observed splitting nulls also often display similar consistency (Fig. 12), pointing to the presence of similar anisotropic fabrics within these disparate regions around the Arakan slab, although the anisotropic fabrics appear to vary from region to region around the slab and vicinity.

The distinct sampling of the source-side shear-wave splitting can be seen in the theoretical ray paths of S waves from the source events to the receiver stations used to make the measurements (Fig. 13). Shear-wave velocities within a Cartesian model volume were assigned based on published variations from a velocity model for the area (Li et al., 2008), which allowed construction of an Arakan slab anomaly 1% higher in velocity than surroundings. Rays were traced along event-station azimuths from the event hypocenters down to 900 km depth. Given the periods of the S waves used (10 s, minimum) and the source-receiver distances, which imply sensitivity to structure off the theoretical ray path (Zhao et al., 2000), the ray tracing is meant to show simply that despite such concerns the rays do largely sample distinct regions azimuthally around the individual sources. Thus, averaging the splitting results, for example, would be unwarranted, and, given the scale of geologic heterogeneity visible at the surface, and implied at upper-mantle depths, the fact of variations in observed splitting is unsurprising.

Anisotropic fabrics related to India's northward motion and along-strike shear within the wide India-Sundaland plate boundary zone can thus be inferred to exist along ray paths that sample the upper-mantle volume west of the slab between ∼20°N and south of Shillong Plateau, particularly west of ∼94°E longitude (see Fig. 13). Note however that a subset of splitting fast trends spanning the full N-S extent of the western study region appears to be parallel to the Arakan subduction zone strike (e.g., main trace of the high topography of the Indo-Burman ranges, Fig. 1). These observations derive from ray paths that sample a central region—here termed the internal zone—between the more E-W fast splits of the Shillong Plateau group (see below), and a group of splitting observations with NW-SE fast trends that sample the region proximal to the bottom surfaces of the northern two Arakan slab segments. Overall, the internal zone measurements follow a pattern such that the general trend of fast shear south of 22°N is slightly NNW, but north of that latitude the mean trend of this splitting subset is NNE, just as the arcuate subduction zone changes strike from NNW in the south to NNE in the north (Fig. 11).

The basic pattern of splitting west of the Arakan slab is disrupted or modulated near, and just south of, the Shillong Plateau, from ∼24° to 25.5°N, where fast splitting axes trend ENE and E-W, similar to the strike of the Plateau itself (E-W) and perhaps indicating interaction of, or transition between, two upper-mantle anisotropic fabrics: N-S–striking fabrics due to India-Sundaland shear, and E-W flattening fabrics developed orthogonal to the N-S compression of the India-Asia collision. Also, projecting southwards along strike of the Arakan slab near the southern limit of deeper slab seismicity, south of 20°N, shear-wave splitting fast axes trend E-W, possibly indicating that upper-mantle flows around the southern slab edge. If so, then subducted Indian lithosphere is not continuous between the southern seismicity of the Arakan slab and the northernmost intermediate depth seismicity of the Andaman subduction segment, an issue not resolved by available tomography (Li et al., 2008).

For ray paths sampling to the east of the Arakan slab, the results are divisible into three groups: north of 24.3°N, splitting fast axes trend NW-SE to NNW-SSE; in a central region (22°–24.3°N), fast axes are E-W; and south of 21°N, predominantly NW-SE fast axes are observed. The northern group of splitting fast axes appears to be part of a limb of upper-mantle flow around the northern edge of the northern Arakan slab segment (Fig. 11). The central group may be related to upper-mantle flow through the gap between the northern and southern Arakan slab segments (Fig. 11), and may also be part of the larger-scale flow field beneath the Shan Plateau (see below). The southern group of measurements east of the Arakan slab appears to be consistent with upper-mantle flow around the southern terminus of the Arakan slab.

Shear-Wave Splitting and Upper-Mantle Flow

The source-side splitting measurements shown in Figure 11 are consistent with observations of SK(K)S/PKS splitting made at nearby seismic stations (Fig. 14) (Vinnik et al., 1992; Lev et al., 2006; Singh et al., 2006, 2007; Huang et al., 2007). In NE India, station fast shear-wave polarizations generally trend E-W, except near the western Shillong Plateau, where fast axes trend ENE-WSW (Singh et al., 2006, 2007). To the east, splitting fast directions trend NNW-SSE, north of ∼26°N, before abruptly changing to E-W trends south of that latitude (data shown from Lev et al., 2006; and Huang et al., 2007; but see also Flesch et al., 2005; Sol et al., 2007; Wang et al., 2008). At station CHTO, at Chiang Mai, Thailand, in the southeast of the study region, the fast shear trend is also nearly E-W (Vinnik et al., 1992). In almost all cases, adjacent source-side and station splitting measurements are similar in fast polarization direction (Fig. 14). Thus, the differing travel paths (upgoing and traversing the entire anisotropic upper mantle for SK(K)S/PKS, and downgoing from the individual event hypocenters for S) and methods used to measure splitting in these studies (those in this study corrected for receiver station splitting, which is unnecessary for the others) yield the same results.

East of 96°E, E-W fast source-side and station splits generally parallel important surface structures of the western Shan Plateau (Fig. 14), i.e., the long, smoothly linear curving valleys visible in Figure 14. These valleys are geomorphic expressions of a series of near-parallel, sinistral strike-slip faults, the Mengxing, Mae Chan, Nam Ma (and other) faults (Shen et al., 2005; Simons et al., 2007). The faults are seismically active, with fairly frequent left-lateral earthquake focal mechanisms (Fig. 4), and appear to form the westernmost part of the Indochina-wide surface expression of upper-mantle flow around the East Himalayan syntaxis (e.g., Sol et al., 2007; Wang et al., 2008). Note that such flow would then comprise a complete reversal of direction, from eastward flow beneath Tibet north of the syntaxis to westward flow into the suprasubduction mantle wedge above the Arakan slab in Myanmar. It does not seem at all coincidental that these Shan Plateau structures and the generally E-W shear-wave splitting both develop at ∼26°N latitude. In fact, this latitude appears to mark an important structural boundary not only at the surface, but also at depth: the offset between the northern and central Arakan slab segments, the Shillong Plateau, and the northern limit of Chittagong-Tripura fold belt structures all lie approximately at this latitude (Fig. 14), and are almost certainly expressions of the combined upper-mantle flow and Arakan slab deformation due to the collisional tectonics of the syntaxis. The slab appears to act as a strong strut, now deforming, which indents Asia at the syntaxis efficiently, and which modulates both the larger-scale upper-mantle flow field on the Asia side of the collision zone (Shen et al., 2005; Lev et al., 2006; Sol et al., 2007; Wang et al., 2008), and also the smaller-scale fabrics of the Arakan slab segments, where flow both below and above the slab segments appears to be affected.

Upper-mantle fabric development should be strongest where coaxial finite deformation is strongest (e.g., Ramsay and Lisle, 2000). Assuming observed splitting delay times are a proxy for anisotropic fabric strength (e.g., Gueguen and Nicolas, 1980; Nicolas and Christensen, 1987), contouring of the delay times could potentially reveal regions where upper-mantle flow is strongest or most coherent. Figure 15 shows the results of such contouring: delay times—and upper-mantle flow fabric strength?—appear to be greatest near the East Himalayan syntaxis, in the SE portion of the study region, and also near the offset between the central and southern Arakan slab segments. Secondary clusters of high delay times occur around the northern edge of the northern Arakan slab segment, near the eastern Shillong Plateau, and as an isolated pocket beneath the central Chittagong-Tripura fold belt. Strong fabric development due to upper-mantle flow around the tightly curved northern edge of the Arakan slab and also around the syntaxis itself would not be surprising. Flow through any kind of narrow channel, such as perhaps exists between the central and southern Arakan slab segments, is also commensurate with observations from fluid dynamics (e.g., Schlichting and Gersten, 2000). The causes of localized high delay times in the southeast of the study area and beneath the fold belt are unclear.

Scale of Upper-Mantle Flow

The distinct anisotropic volumes delineated by the shear-wave splitting observations outlined above are shown schematically in Figure 16. Several conclusions can be drawn: (1) The source events occurred at depths of 50–100 km, and sampled anisotropic fabrics from those depths down to the top of the transition zone. The Arakan slab segments extend from near the surface (Ni et al., 1989; Rao and Kalpna, 2005; Stork et al., 2008) to the top of the transition zone (Li et al., 2008). And finally, the geology of the syntaxis region, NE India, Yunnan, and the Shan Plateau can be related to the splitting results, both source-side observations of this study and station splitting published by others. These observations all imply that the upper-mantle flow field extends from the surface perhaps to the top of the transition zone. (2) Laterally, there seem to be two scales of anisotropic fabric and flow: a smaller-scale flow field, defined by source-side splitting observations around the Arakan slab segments, varies on the order of 50–75 km laterally. A larger-scale structure of anisotropic upper-mantle fabrics is also clearly developed and appears to be determined by the regional form of the India-Asia collision, i.e., the large-scale material flow around the east Himalayan syntaxis. (3) The transitions between these two scales of the upper-mantle flow field, and between the domains of the smaller-scale Arakan slab fabrics, appear to be rather sharp. If so, then strain partitioning of upper-mantle deformation would appear to occur: the transition from the generally E-W flow field of the Himalayas, NE India, and the Shillong Plateau (Singh et al., 2006, 2007) to the slab-parallel fabrics of the Arakan internal zone, appears to be abrupt (Figs. 11, 16; see Fig. 13 for sampling). An abrupt transition between the flow fields suggests the presence of domainal deformations (in the structural geology sense) bounded by much narrower zones of stronger shear, and perhaps even ductile faults in the upper mantle. One implication of the existence of volumes of generally homogeneous coaxial finite strain bounded by high-strain shear zones is that large-scale deformation may not proceed similarly to the deformation of viscous continua.

An Olivine B-Axis Anisotropic Fabric?

An interesting question is whether the mantle wedge above the Arakan slab could actually be characterized by predominantly b-axis anisotropic fabrics, with concomitant orientation of flow in the suprasubduction wedge? If so, the mantle wedge flow field would then actually be N-S, consistent with generally N-S shear between northward-propagating India and Indochina extruding southeastwards, but, as indicated by many splitting observations, this flow field would then extend to at least 102°E (Lev et al., 2006; Sol et al., 2007; Wang et al., 2008), far to the east of the region normally considered to be upper-mantle wedge. If the flow field were actually this wide (some 600 km) and anisotropic b-axis fabrics predominated throughout, some as yet unidentified mechanism for hydrating or partially melting the upper mantle far beyond the normal width of the Arakan upper-mantle wedge would appear to be required (e.g., Kneller et al., 2008). Note that if the shallow Arakan slab (the closest known source of hydrated material) is the source of the water, the hydration process would have to transport material effectively across the dominant N-S mantle fabrics organized by upper-mantle flow. Alternatively, dewatering of the Indian lithosphere lying at the top of the transition zone (Li et al., 2008) could potentially hydrate the upper mantle from below without entailing cross-flow material transport (see also Van der Lee et al., 2008). If we assume b-axis fabrics have formed due to presence of partial melt, then apparent E-W fast splitting in the central part of the Arakan upper-mantle wedge would actually imply N-S flow fabrics, cutting across the E-W grain of surface structures of the Shan Plateau. Note that such flow fabrics would be inconsistent with 2-D mantle wedge corner flow, and would presumably also imply an abnormally wide mantle wedge region.

Finally, simple asthenospheric flow (e.g., Vinnik et al., 1989a, 1989b) could also be invoked to explain generally E-W splitting fast axes beneath eastern Myanmar and Indochina south of 26°N, as observed in this study and those of Lev et al. (2006), Huang et al. (2007), Sol et al. (2007), and Wang et al. (2008). If basal shear beneath a generally east-moving Sundaland (relative to stable Eurasia; Simons et al., 2007) organizes upper-mantle deformation fabrics, then one would expect the asthenospheric channel to show generally E-W fast splitting trends. Such basal shear would extend beneath almost all of Indochina, consistent with GPS results (Simons et al., 2007), and could explain the great eastward extent of the splitting observations—far beyond the normal width of supra-subduction upper-mantle wedge—without requiring either b-axis anisotropic fabrics or ad hoc mechanisms for hydrating or partially melting the upper-mantle wedge and beyond.

Shear-wave splitting of S waves from earthquakes in the Arakan slab is consistent with strong upper-mantle anisotropy in the India-Asia-Sundaland triple junction region. The Arakan slab is dismembered into three segments, defined by high-precision relocations of intermediate earthquake, and approximately E-W offsets of these hypocenters at two locations define a left-stepping en echelon structure. S waves from earthquakes within the three segments and surroundings are split systematically, and, corrected for receiver-side splitting, fast shear trends are predominantly trench-parallel beneath the east-dipping slab segments; are more nearly trench-normal on the Sundaland (east) side of the Arakan lithosphere; parallel the southern ∼E-W gap between Arakan slab segments; and turn sharply around the extreme northern and southern edges of subducted Arakan lithosphere. Source-side shear-wave splitting beneath India is consistent with ∼E-W–trending fast shear polarizations of SK(K)S splitting in northeastern India. The general pattern of both surface site velocities from GPS and upper-mantle flow is consistent with material flow around the eastern Himalayan syntaxis into the mantle wedge above the Arakan slab, and around the northern edge of the Arakan slab. The upper mantle may also flow through the gap between the central and southern Arakan slab segments, and around the apparent southern edge of the Arakan slab.

The shear-wave splitting database published online by the group at the Laboratoire de Tectonophysique, Université de Montpellier II, was indispensable in finding suitable stations for receiver-side correction (see I am grateful to Martin Flower, Mian Liu, Victor Mocanu, Suzan van der Lee, and Liz Widom for many stimulating discussions of Sundaland tectonics and geochemistry. Associate Editor Craig Jones and two anonymous reviewers made very useful suggestions for improving the manuscript. All data used in this study were downloaded from the IRIS Data Management Center.