SUMMARYThe occurrence of plate tectonics on Earth is rooted in the physics of lithospheric ductile weakening and shear-localization. The pervasiveness of mylonites at lithospheric shear zones is a key piece of evidence that localization correlates with reduction in mineral grain size. Most lithospheric mylonites are polymineralic and the interaction between mineral phases, such as olivine and pyroxene, especially through Zener pinning, impedes normal grain growth while possibly enhancing grain damage, both of which facilitate grain size reduction and weakening, as evident in lab experiments and field observations. The efficacy of pinning, however, relies on the mineral phases being mixed and dispersed at the grain scale, where well-mixed states lead to greater mylonitization. To model grain mixing between different phases at the continuum scale, we previously developed a theory treating grain-scale processes as diffusion between phases, but driven by imposed compressive stresses acting on the boundary between phases. Here we present a new model for shearing rock that combines our theory for diffusive grain mixing, 2-D non-Newtonian flow and two-phase grain damage. The model geometry is designed specifically for comparison to torsional shear-deformation experiments. Deformation is either forced by constant velocity or constant stress boundary conditions. As the layer is deformed, mixing zones between different mineralogical units undergo enhanced grain size reduction and weakening, especially at high strains. For constant velocity boundary experiments, stress drops towards an initial piezometric plateau by a strain of around 4; this is also typical of monophase experiments for which this initial plateau is the final steady state stress. However, polyphase experiments can undergo a second large stress drop at strains of 10–20, and which is associated with enhanced phase mixing and resultant grain size reduction and weakening. Model calculations for polyphase media with grain mixing and damage capture the experimental behaviour when damage to the interface between phases is moderately slower or less efficient than damage to the grain boundaries. Other factors such as distribution and bulk fraction of the secondary phase, as well as grain-mixing diffusivity also influence the timing of the second stress drop. For constant stress boundary conditions, the strain rate increases during weakening and localization. For a monophase medium, there is theoretically one increase in strain rate to a piezometric steady state. But for the polyphase model, the strain rate undergoes a second abrupt increase, the timing for which is again controlled by interface damage and grain mixing. The evolution of heterogeneity through mixing and deformation, and that of grain size distributions also compare well to experimental observations. In total, the comparison of theory to deformation experiments provides a framework for guiding future experiments, scaling microstructural physics to geodynamic applications and demonstrates the importance of grain mixing and damage for the formation of plate tectonic boundaries.