Robustness, Fidelity and Prediction-Looseness of Models

Assessment of the credibility of a mathematical or numerical model of an engineering system must combine three components: (1) The fidelity of the model to test data. (2) The robustness, of model fidelity, to lack of understanding of the underlying processes. (3) The prediction looseness of the model. ‘Prediction looseness’ is the range of predictions of models which are equivalent in terms of fidelity. The main result of this paper is that high fidelity, high robustness, and small prediction looseness are mutually incompatible. A model with high fidelity to data and high robustness to imperfect understanding of the process, will have low predictive focus. Our analysis is based on info-gap models of uncertainty.