Kinetic modelling of CO2-water-rock interactions

A system perturbated by CO2 injection reacts by dissolving primary minerals and form new secondary phases. The importance of such mineral reactions for safe long-term storage is highly system dependent, with some reservoirs rich in reactive phases that contain divalent metal cations, which promote mineral carbonte growth, and other reservoirs being almost pure quartz (SiO2) sands. Because of the complexity in the reaction paths of natural systems, numerical simulations have proved to be valuable to predict the reactivity on time scales from laboratory experiments to tens or hundreds of thousands of years. Such numerical simulations require quality thermodynamic and kinetic data, as well as proper mathematical expressions. We here provide a guide to model kinetically constrained CO2-water-rock interactions aimed for both experienced geochemists and researchers with more limited background in modelling geochemical reactions. Because rates of mineral reactions are coupled through common aqueous species, and rates vary by more than ten orders of magnitude, the system of ordinary differential equations (ODE's) is commonly stiff. This leads to challenges in solving such equations and limits the possible system size and spatial and geometrical complexity that can be solved. We have here focused on how different simplifications can be made to reduce the CPU time required to solve mineral reactions and to allow reaction path modelling even for the largest scales. For a 10,000 years batch reaction simulation, we showed how the CPU run time decreased from hours when all minerals were defined by ODE's (fully kinetic), to a few seconds if secondary phases were allowed to grow according to the local equilibrium assumption (semi-kinetic). We then showed how the system can be further simplified by allowing far-from-equilibrium dissolving minerals to be expressed by first-order decay analytical expressions. We finally suggest a step-wise procedure where significant mineral reactions were first identified by running batch simulations, and the system were then simplified according to the system size, geometric complexity, and the time of interest. (c) 2013 Elsevier Ltd. All rights reserved.