Read around and understand Runge-Kutta 4 (RK4) integration method and its computational implementation.
Consider attempt to integrate a self-regulating stepsize control into RK4 functions. eg. Variable stepsize depending on difference between k2 and k3 (2nd and 3rd gradients calculated).
Adaptive step-size achieved through error analysis during RK4 loop
Research and work through Electromagnetic equation of motion involving RK4 method.
Understand RK4 in reference to a parabolic path.
Understand errors in all of above (accuracy of method).
Write and run through C++ program showing electron trajectory in simple EM field.
C++ RK4 integrator up and running using Lorentz differential equation to track particle in EM field. Seems to be accurate to 0.07% in 14 million steps.
Create simple EM field Geant4 example in vacumn (or very low density gas) and run electrons through it to compare trajectories with previous C++ code.
Port current C++ RK4 version to CUDA.
Above involves deciding how to change RK4 algorithm to parallel, what to put in CUDA kernal.