UCLA

Aspects of backward stimulated Raman scattering (SRS), in which a light wave incident on a plasma decays into a backscattered light wave at a lower frequency and a forward propagating electron plasma wave (EPW), are investigated for situations of relevance to laser driven inertial fusion energy (IFE). Results from one- and two-dimensional particle-in-cell (PIC) simulations---both of nonlinear electron plasma waves impulsively driven by a ponderomotive driver of short duration, and of self-consistent SRS from a continuous driver---are shown which explore effects of small magnetic fields ($\omega_c / \omega_p \ll 1$) oriented perpendicular to the direction of laser propagation on SRS across a range of laser intensities of relevance to IFE. While it is confirmed that a magnetic field increases the intensity threshold for kinetic inflation to occur and decreases the average reflectivity near this threshold, it is found that saturated levels of reflectivity can be similar or enhanced due to a magnetic field. Even when the average reflectivity is similar, however, the simulations show that the qualitative dynamics of SRS remain distinctly different with and without a magnetic field. Analysis indicates that the magnetic field alters the recurrence pattern of SRS by enhancing the damping of EPWs, limiting modifications to the electron distribution function, and reducing the magnitude of the nonlinear frequency shift. In addition, at these relatively high intensities, backscattered light from SRS can be of sufficient intensity to undergo backward SRS itself. Such rescatter prevents backscattered light from reaching the incident boundary, causing a reduction in measured reflectivity. The magnetic field is shown to prevent rescatter from occurring, leading to a relative enhancement in reflectivity.

Fully relativistic PIC simulations such as those used above require robust and feature-rich codes that can fully leverage modern graphics processing unit (GPU)-based supercomputers. Two major algorithmic improvements to the PIC code \textsc{Osiris} are thus undertaken.First, an implementation of dynamic load balancing in \textsc{Osiris} is described, which divides the simulation space into many small, self-contained regions or ``tiles.'' The implementation shows low overhead and improved scalability with OpenMP thread number on simulations with both uniform load and severe load imbalance. Compared to other load-balancing techniques, the presented algorithm gives order-of-magnitude improvement in parallel scalability for simulations with severe load imbalance.

Second, an implementation of GPU acceleration for \textsc{Osiris}, built on the aforementioned tile data structures, is described. An overview of the algorithm, which features a CUDA extension to the underlying Fortran architecture, is given. Detailed performance benchmarks for thermal plasmas are presented which demonstrate excellent weak scaling and high levels of absolute performance. The robustness of the code to model a variety of physical systems is demonstrated via simulations of Weibel filamentation and laser-wakefield acceleration. Finally, energy consumption benchmarks are provided which indicate that the GPU algorithm is up to $\sim14$ times faster and $\sim$7 times more energy-efficient than the optimized CPU algorithm on a node-to-node basis. The described developments address the PIC simulation community's computational demands both by contributing a robust and performant GPU-accelerated PIC code and by providing insight on efficient use of GPU hardware.