Development of accurate predictive models for multiphase multicontaminant transport in groundwater requires contemporaneous advances on a multitude of fronts. Fundamental physical understanding of upscaled macroscopic multiphase flow and of the validity of the local equilibrium assumption has progressed but is far from complete. Both of these are critical in determining the form and computational complexity of numerical models. Data describing partitioning behavior of multicontaminant systems, presently rather sparse, will need to be enhanced, with realistic assessments of the number of components necessary to represent a given compositional system adequately. Because a fully implicit nonequilibrium m-phase n-component model appears to be impractical, modelers have had to make compromises, and this can be expected to continue. With the local equilibrium assumption, generalized compositional formulations have been successful in the petroleum industry, and groundwater modeling is trending in some of the same directions. For nonequilibrium, analogous or alternative formulations will require new ideas, such as approximate couplings between kinetics and transport, to achieve effective compromises. Innovations in numerical methods, which will substantially reduce the computing effort for a given accuracy, especially on vector and parallel computers, will ease the need for compromises but cannot be expected to eliminate it.
It will be important to develop perspective on the balancing of various uncertainties and errors. Is the multiphase Darcy equation accurate enough to justify the expense of nonequilibrium? Nonequilibrium may greatly increase the computational burden of a simulation, necessitating a coarser grid. Is the error of assuming equilibrium, with a fine grid, greater than the discretization error of a coarser grid for nonequilibrium? Is it sensible to perform rigorous nonequilibrium calculations if the results are used in transport equations solved by numerically dispersive methods such as upstream finite differences? How effectively will improved numerical discretizations reduce numerical errors, making the effort of improving other modeling aspects appropriate? To make intelligent compromises, at least at an intuitive engineering level, modelers will need the answers, possibly different answers for different problems, to these types of questions. This will involve understanding of fundamental and upscaled physical processes, knowledge of properties of alternative mathematical formulations of multicomponent systems, and assessment of constraints imposed on accuracy by limited computational resources, even with efficient, accurate numerical techniques.