How the underlying mantle responds to the subduction of an oceanic plate remains controversial. Numerical models demonstrate the presence of 3D flow in the sub-slab mantle with a poloidal component induced by slab entrainment and a toroidal component driven by slab rollback. These flows produce plate–motion–parallel and trench-parallel fast directions of seismic anisotropy, respectively, allowing them to be distinguished via the anisotropy pattern. However, seismology studies have so far documented the dominance of either, but not both, type of flow in most subduction zones, evoking debates concerning the flow mechanisms. We employed source-side anisotropy technique to map the sub–slab mantle of the Cocos subduction zone. The observed anisotropy pattern can be best explained by a plate–motion–parallel anisotropy layer, roughly 130 km thick, underlain by a trench-oblique anisotropy regime. We interpret this phenomenon as an entrained poloidal flow transitioning to a rollback–induced toroidal flow modified to suit the regional tectonics. This finding reconciles the current debate over which flow prevails behind a subduction zone. The upper, entrained layer in our model extends below the lithosphere–asthenosphere boundary previously defined in the vicinity of the middle America trench, suggesting strong coupling along the lithosphere–asthenosphere boundary in this region.

潜在的地幔如何对洋洋板块的俯冲做出反应仍然是有争议的。数值模型证明了3D流动存在于亚板状地幔中，板状夹带引起了极向分量，而板坯回滚驱动了超环面分量。这些流动分别产生了地震各向异性的板块运动平行和沟槽平行快方向，从而可以通过各向异性模式来区分它们。但是，到目前为止，地震学研究已经证明，在大多数俯冲带中，一种或两种流动类型均占主导地位，引发了关于流动机理的争论。我们采用了源侧各向异性技术绘制了科科斯俯冲带的下平板地幔图。大约130 km厚的板块运动平行各向异性层可以最好地解释观测到的各向异性模式，沟槽倾斜各向异性机制为基础。我们将这种现象解释为夹带的多极体流转变为回滚诱导的环形流，并对其进行了修改以适应区域构造。这一发现与当前关于在俯冲带后普遍存在水流的争论相吻合。我们模型中的上层夹带层在先前在美国中部海沟附近定义的岩石圈-软流圈边界下方延伸，表明该地区沿岩石圈-软流圈边界强烈耦合。