Database generation for validation of TGLF and retraining of neural network accelerated TGLF-NN

The trapped gyro-Landau fluid (TGLF) code is a quasi-linear model of gyrokinetic turbulence that relies on so-called saturation rules SAT0, SAT1, and SAT2 to capture the nonlinear physics necessary to compute particle, heat and momentum transport in fusion plasmas. Any free parameters in these saturation rules are calibrated against high-fidelity simulations with the CGYRO code. A large database of five million TGLF simulations has been generated with experimental parameters from 2500 DIII-D discharges in order to increase the accuracy of TGLF-NN, which is a neural network accelerated representation of TGLF. This database includes perturbations in four plasma parameters, namely the electron and ion temperature gradients, the density gradient and the E×B shearing rate. The response of TGLF fluxes to these perturbations also serves as a sensitivity study to aid validation of TGLF against this experimental database. For example, the TGLF heat flux is found to be predominantly driven by ion temperature gradient modes, trapped electron modes and kinetic ballooning modes, which is verified by linear CGYRO simulations on selected cases. Lastly, the sensitivity study allows estimation of critical gradients for turbulence onset and heat flux stiffness as a function of saturation rule to complement ongoing validation and model development efforts.