Adaptive Resolvent Analysis: Application to High Enthalpy Boundary Layers

A method is presented to adaptively and efficiently sample the regions in spectral and physical space where the resolvent operator produces the highest gain. At each iteration, Gaussian Process Regression exploits previously acquired gains and their sensitivities to produce a prior and related uncertainties. The next point to sample is chosen by maximizing an acquisition function that balances increasing the value of the prior while decreasing the uncertainty in the prediction, as in Bayesian Optimization. Compared to a fixed grid resolution, which can be expensive for complicated geometries and base flows, this method utilizes fewer query points clustered in the most energetic regions in the spectral space. The method is then applied to a high-enthalpy boundary-layer flow along with masking of the resolvent to isolate the input-output relationships between variables and phenomena, identifying key length and time scales for each relationship.