摘要:
A framework has been constructed for predictive modelling of the stimulated Raman scattering (SRS) instability at ignition scale in laser direct-drive inertial confinement fusion (ICF). An extended ray-tracing methodology from literature was found to underpredict SRS, and to be computationally inefficient. This was corrected by modification of the energy-exchange process between laser and Raman light, and introduction of thresholds and bounds physically informed by collisional absorption and Rosenbluth gain. Predictions were further improved by Gaussian process (GP) regression surrogates, which at the first instance fully resolved the SRS solution in the steady-state strong damping limit. Subsequently, additional physics were captured using the plasma wave solver LPSE, with scope for future capture of more complex solvers within the hierarchical machine learning framework. An additional GP surrogate was used to replace the costly resonant frequency search part of the algorithm. The framework was implemented in the 1D ICF code Freyja, allowing for investigation of the effects of SRS on fusion performance for a shock ignition case. It also allowed for efficient training of the GP surrogates by keeping the predicted error below thresholds across a full simulation. SRS was found to have a detrimental effect on fusion performance, but this was mitigated to some extent when modelling the secondary effects of the hot electron populations which are created. These hot electrons were found to strengthen the ignition shock, but an uncertainty quantification study still showed a much reduced probability of ignition compared to the case without SRS. As well as predictive modelling, a separate probabilistic study was carried out to show the approach of calibrating modelling coefficients for laserplasma interactions and energy transport against experimental data. The results of this study were used to both set certain modelling parameters for later simulations, and to highlight th