Electron Paramagnetic Resonance, Optimization and Automatic Differentiation
This chapter describes an optimization problem applied to electron paramagnetic resonance spectroscopy. Levenberg-Marquardt fails to converge using a divided differences Jacobian approximation in single precision, while it succeeds using Odyssée-generated forward mode Jacobian values. In double precision, the optimizer returns a smaller “minimum” objective function with AD compared to DD in 46% more CPU time.
Unable to display preview. Download preview PDF.