isotropy, VTI). With this assumption, SH leads SV here, corresponding to "'=±90 " in our notation (Fig. 1c). A further limitation is using only one azimuth of rays in D!: this cannot distinguish VTI from the case of an arbitrarily tilted axis of rotational symmetry in which wave speed does not vary (tilted transverse isotropy, TTI) when the axis dips towards the receivers or stations. An improvement on this situation can be made by utilising crossing ray paths in D! 10 , but this relies on having the correct source-receiver geometry, which is not possible beneath North America using only deep earthquakes.We address this issue beneath the Caribbean by incorporating measurements from shallow earthquakes in our dataset, and thus reduce the symmetry of the anisotropy which must be assumed.We measure anisotropy in D! using differential splitting in S and ScS phases using an approach described by refs. 10,11 . Both phases travel through the same region of the upper mantle (UM), but only ScS samples D! (Fig. 1a). As the majority of the lower mantle (LM) is relatively isotropic 12 , by removing the splitting introduced in the UM we can measure that which occurs only in D! (see Supplementary Information).Earthquakes in South and Central America, Hawaii, the East Pacific Rise (EPR) and the Mid-Atlantic Ridge (MAR), detected at North American stations, provide a dense coverage of crossing rays which traverse D! beneath southern North America and the Caribbean (Fig. 1b). Three distinct regions are covered (Fig. 2), each sampled along two distinct azimuths. The Caribbean (region ÔSÕ) has been previously well studied 1,4,8 , but the northeast (ÔEÕ) and southwest (ÔWÕ) United States have not. Hence nowhere are our measurements compatible with VTI, because we do not find "«=±90 " within error in both directions for any region.A likely mechanism for the production of anisotropy in D! is the lattice-preferred orientation (LPO) of anisotropic mineral phases present above the CMB such as (Mg,Fe)O, and MgSiO 3 -perovskite (pv) and -postperovskite (ppv). These may give rise to styles of anisotropy more complicated than TTI with lower symmetries, which are compatible with our two-azimuth measurements. We investigate the possibility of LPO in ppv leading to the observed anisotropy rather than other phases because of its likely Our results can differentiate between these candidate mechanisms if we assume that most of the measured anisotropy in D! is a result of deformation-induced LPO in ppv, and we have an accurate estimate of the mantle flow where we measure anisotropy.At present, such models of mantle deformation are in their infancy, but we can nonetheless make inferences from broad-scale trends in subduction and global V S models. We calculate the orientations of the shear planes and slip directions which are compatible with our measurements for the three slip systems in ppv. Aggregate elastic (001) system. These planes and directions are plotted in Fig. 3. We also produce the shear planes predicted for cases of pv and MgO ( Suppl...

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

The D region, which lies in the lowermost few hundred kilometres of the mantle, is a central cog in the Earth's heat engine, influencing convection in the underlying core and overlying mantle. In recent years dense seismic networks have revealed a wealth of information about the seismic properties of this region, which are distinct from those of the mantle above. Here we review observations of seismic anisotropy in this region. In the past it has been assumed that the region exhibits a simple form of transverse isotropy with a vertical symmetry axis (VTI anisotropy). We summarise new methodologies for characterising a more general style of anisotropy using observations from a range of azimuths. The observations can be then used to constrain the mineralogy of the region and its style of deformation by a lattice preferred orientation (LPO) of the constituent minerals. Of specific interest is the recent discovery of the stability of the post-perovskite phase in this region, which might explain many enigmatic properties of D . Mantle flow models based on density models derived from global tomographic seismic velocity models can be used to test plausible mineralogies, such as post-perovskite, and their deformation mechanisms. Here we show how linked predictions from mineral physics, geodynamical modelling and seismic observations can be used to better constrain the dynamics, mineralogy and physical properties of the lowermost mantle.

[1] In order to test the hypothesis that seismic anisotropy in the lowermost mantle is caused by the development of a post-perovskite lattice preferred orientation, and that anisotropy can thus be used as a probe of the dynamics of the mantle's lower boundary layer, an integrated model of texture generation in D″ is developed. This is used to predict the elastic anisotropy of the lowermost mantle as probed by global anisotropic tomographic inversions. The model combines the current 3D mantle flow field with simulations of the deformation of post-perovskite polycrystalline aggregates. Different descriptions of single crystal plasticity can lead to model results which are anti-correlated to each other. In models where post-perovskite deformation is accommodated by dislocations moving on (010) or (100), patterns of anisotropy are approximately correlated with the results of tomographic inversions. On the other hand, in models where dislocations move on (001) patterns of anisotropy are nearly anti-correlated with tomographic inversions. If all the seismic anisotropy in D″ extracted from global anisotropic inversions is due to the presence of a lattice preferred orientation in post-perovskite in the lowermost mantle, and if the results of the tomographic inversions are not strongly biased by the sampling geometries, these results suggest that, in contrast to ideas based on the 1D anisotropic signal, deformation of post-perovskite in the lowermost mantle may be accommodated by dislocations moving on (010) or (100). Alternatively, a significant portion of the anisotropic signal may be caused by mechanisms other than the alignment of post-perovskite crystals.

No abstract