Many scientific phenomena are studied via mathematical (i.e., computer) models and field experiments, simultaneously.
- Real experiments are expensive, and for this and other reasons (ethics, lack of materials/infrastructure, etc.), limited configurations can be entertained.
Computer simulations are lots cheaper, but usually not so cheap as to allow infinite exploration of the configuration space(s).
- Plus the simulations usually idealize reality (i.e., they are biased)
- and can involve more "knobs", or tuning parameters, than can be controlled (or even known) in the field.
So the goal here is to build an apparatus that can harmonize the two data types
- for the purpose of learning about/predicting the "real" process,
- or possibly optimizing some aspect of it.