Olivine, the major mineralogical species in the upper mantle of the Earth, is highly anisotropic [Kumazawa and Anderson, 1969]. If a significant fraction of the olivine grains in a volume within the mantle have their crystallographic axed aligned parallel to one another - as can happen through the effect of strain [Zhang and Karato, 1995] - that volume can take on a net seismic anisotropy. Two neighboring volumes, with different amounts of alignment or different alignment orientations will possess different seismic properties. There is ample evidence from the study of the birefringence (splitting) of the two polarizations of teleseismic shear phases (e.g. SKS) that most of the upper mantle is in fact anisotropic [e.g. Vinnik et al. 1992; Silver, 1996]. A worldwide summary by Silver [1996] indicates that a typical SKS wave is split by 0.5-1.5s. This translates into 50-150 km of anisotropic mantle beneath a typical point on the Earths surface, presuming 5% shear anisotropy, on the high end of mantle rocks exposed in ophiolites and kimberlite nodules [Mainprice and Silver, 1993]. The time of creation of this anisotropic mantle fabric is controversial. Vinnik et al. [1992] argued for an origin related to the Earths current stress field, via a "dynamic" fabric imposed within the upper mantle below the continents. In this view, indicators of seismic anisotropy should be fairly coherent over large areas, and generally independent of the features exposed on the surface.
Silver [1996], citing the general agreement between mantle fabric directions and surface geology, argues that the anisotropy is "frozen" into the continental lithosphere. This model provides a mechanism for the production of small (100-200 km) anisotropic domains - regions within the mantle that have coherent fabric, and display similar anisotropic properties. It envisages large regions with uniform mantle fabric being formed in the process of continental lithosphere creation, getting torn up by later tectonic events, reoriented and reassembled into stable cratons.
Levin et al. [1996] have determined that a typical velocity heterogeneity of about 2.5% (i.e. teleseismic P wave fluctuations of about 0.3s) would be expected from randomly oriented anisotropic domains, if these domains have a degree of anisotropy sufficient to produce the 1-s global median splitting times reported by Silver [1996]. Since shear-wave splitting demonstrates that the mantle is anisotropic, and since anisotropy can be shown to cause velocity heterogeneity as large as that deduced from mantle tomography, we might ask whether it can account for most - or all - of the heterogeneity in older terranes.
The northeastern US presents an excellent testing ground for the above
hypothesis.
The region is in a stable passive margin tectonic regime, with most
recent activity 
100 Ma (the passage of the Monteregian hotspot).
On the other hand, the tectonic history
of the region spans two Wilson cycles.
Remnants of two collisional orogens (Grenvillian and Appalachian)
comprise most of the area (Figure 1).
Numerous accreted terranes have been identified within the younger
Appalachians [Zen, 1983; Taylor, 1989].
The more ancient Grenvillian Orogen also has internal variations
[Moore, 1986; Rivers et al., 1989].
If the "anisotropic domain" paradigm is universally applicable,
it should be clearly manifested here.
Seismic properties of the continental lithosphere of the eastern North America have been extensively studied over the last two decades. Seismic velocities in the lithosphere and upper mantle have been studied using teleseismic traveltime delays of short period P waves [Taylor and Toksoz, 1979; Levin et al, 1995], and also using surface wave traveltime delays [Van der Lee and Nolet, 1997; Li et al, 1999]. Seismic anisotropy in the upper mantle has been identified in numerous studies of shear-wave splitting [Silver and Chan, 1988; Vinnik et al, 1992, Levin et al., 1996, 1999; Barruol et al, 1997; Fouch et al, 1999]. Interpretation of the observed splitting parameters varies. Levin et al. [1996] and Barruol et al. [1997] cited a rapid lateral variation of parameters that correlated with major tectonic features to argue in favor of the "frozen fabric" model. Fouch et al. [1999] argue that most lateral variability in splitting parameters may be explained by a complex flow pattern in the mantle, and favor the "dynamic fabric" model.
In this paper we examine properties of teleseismic shear waves observed by a network of stations in the Northeastern US (Figure 2). We measure shear-wave splitting of core-refracted phases like SKS, and determine relative traveltime delays of shear waves. Our main finding is that, given a sufficient volume of data, the pattern of anisotropy is more laterally homogeneous than was previously thought. Moreover, there is no obvious correlation between the splitting parameters (indicative of anisotropy) and the relative delays (indicative of velocity distribution). We also find significant back-azimuthal variation in shear-wave splitting parameters at all sites. This variation is similar among stations, and is fit well by a simple model that contains two horizontal layers with different anisotropic parameters.
In the following sections we describe our observations and a two-layer model that fits them, and discuss potential mechanisms that could lead to the formation of such structure in this stable continental region.