A new RF Ray Tracing code using auto differentiation

To design a fusion pilot plant or simulate an ITER pulse, a high-fidelity, whole-device simulation capability including RF heating is needed. Ray tracing is a method to determine to deposition location of RF power or the originating location of the emission in a plasma medium. Adapting legacy codes to modern HPC systems is challenging due to GPU incompatibility and a lack of abstraction. Errors transcribing derivatives of complex dispersion relations pose a challenge when attempting to extend existing or develop new codes. Auto differentiation technology eliminates this error and has emerged as a preferred method for abstracting backend resources from front end equations in machine learning applications. We present a new ray tracing code built from a pure C++ auto differentiation framework. This approach results in a modern extensible code that can adapt to current and future HPC systems and embed within other codes. New physics is added by simply providing expressions for the dispersion relation and equilibria. The framework then performs symbolic computations to convert these into optimized expressions for the differential equations that can be solved to trace a ray. This new ray tracer shall be validated against expected analytic expressions for simple dispersions functions.