A Discontinuous Future for Numerical Modeling in Soil and Rock

The case is made that continuum methods (such as finite elements) for rock and soil may be completely replaced by particle models (such as DEM) in 10-20 years. Reproducing complex behavior with continuum methods requires complex constitutive models, containing sometimes dozens of parameters and/or internal variables, such as yield-surface descriptors. The DEM model needs very few material properties because the behavioral complexity arises as an emergent property of the system. There is a lot of current interest about emergent systems in which collections of simple things behave---collectively--in complicated, and completely different, ways. For example, in rock and soil, the following behaviors appear "automatically" from a particle model of the material: continuously nonlinear stress/strain response, dilation related to mean stress, transition from brittle to ductile behavior, hysteresis and memory (nested yield surfaces), nonlinear strength envelopes, acoustic emissions, and many more. It is very complicated to reproduce all these things with a continuum constitutive law, and such laws tend to contain many ad hoe parameters and equations. However, all the phenomena noted above appear as emergent properties of a macroscopic system composed of particles---either unbonded for soil, or bonded for rock. Further, a DEM model naturally exhibits localization (fractures in a brittle solid, or shear bands in a granular material). It is difficult to capture such localization in a continuum model that uses a mesh. It is already accepted that the DEM is well suited to simulating systems such as ball mills, blocky rock and assemblies of discrete fragments. Further, the method has been used to study the micromechanics of materials such as soil at the element level. However, we suggest that in a few years the DEM should be applied in the traditional domain of continuum methods, representing complete rock & soil systems such as dams, tunnels and foundations. Clearly, this can only be done when computers become powerful enough, but it is argued that this will occur in 10-20 years. The talk will explore the reasons why all the complex behaviors of soil and