Numerical models of subduction zones have advanced to the point where we can  consider how to move beyond the notion of simple conceptual model towards something that we could call a simulation. This requires a process of data assimilation to tune generic models so that they can address questions regardin a specific location and time but, of course, it also demands that the forwards models have sufficient predictive power that they can actually address the observations directly.

In this talk I will give an overview of recent time-dependent modelling developments in 3D by members of the Underworld geodynamics group and how they lead to a view of subduction zone dynamics that differs markedly from the classical, static "textbook” cross sections. 

Higher resolution models are possible in 2D and these have considerable predictive power in regions away from slab edges or other significant along-strike variations in strength or buoyancy. I will discuss how these models can be used to better understand the state of stress and seismicity in slabs and how this relates to another oversimplification in the textbook model.