Multi-fidelity Regression of Sparse Plasma Transport Data Available in Disparate Physical Regimes

Physical data, such as material properties, are typically generated by several methods, including experiments and computations, in limited parameter regimes accessible to those methods. When datasets generated using such disparate methods are combined into one dataset, the resulting combined dataset is typically sparse with dense "islands" in a potentially high-dimensional parameter space, and predictions must be interpolated among such islands. Using plasma transport data as our example, we use a non-linear multi-fidelity Gaussian-process regression framework that exploits physical data from multiple sources at multiple fidelities. We find that including data from multiple fidelities can reduce single-fidelity-only interpolation error by an order of magnitude resulting in predictions that are accurate across disparate regiems. We discuss how the multi-fidelity Gaussian process regression framework reveals the effectiveness of the low-fidelity data. When the low-fidelity model is insufficient, the framework reduces to single-fidelity Gaussian process regression. This framework suggests where data should be collected to generate more accurate predictions and can be readily applied to look-up tables for transport coefficients or equations of state that are often used in hydrodynamic simulations.