Capturing Scientists ’ Insight for DDDAS.pdf

Capturing Scientists’ Insight for DDDAS Paul Reynolds, David Brogan, Joseph Carnahan, Yannick Loitie?re, and Michael Spiegel Computer Science Department University of Virginia Abstract. One of the intended consequences of utilizing simulations in dynamic, data-driven application systems is that the simulations will ad- just to new data as it arrives. These adjustments will be difficult because of the unpredictable nature of the world and because simulations are so carefully tuned to model specific operating conditions. Accommodating new data may require adapting or replacing numerical methods, sim- ulation parameters, or the analytical scientific models from which the simulation is derived. In this research, we emphasize the important role a scientist’s insight can play in facilitating the runtime adaptation of a simulation to accurately utilize new data. We present the tools that serve to capture and apply a scientist’s insight about opportunities for, and limitations of, simulation adaptation. Additionaly, we report on the two ongoing collaborations that serve to guide and evaluate our research. 1 Introduction In dynamic, data-driven application systems (DDDAS), we have observed that scientists are regularly confronted with the challenge of creating simulations ca- pable of adapting to unanticipated runtime conditions. Runtime conditions may trigger adjustment or replacement of the analytical scientific models from which the system is derived, the numerical methods that implement those models, or the computational infrastructure that executes the numerical methods [1]. How, for example, should a weather simulation respond to newly acquired data that invalidates its predictions? Can some simulation parameters be adjusted auto- matically by a Kalman filter or must the Kalman filter itself be reparameterized? Perhaps an entirely different underlying model is required to appropriately sim- ulate the new portion of state space exposed by the new data. Because so many a