Neutron reflectometry is used to probe thin films and buried interfaces, but the interpretation of reflectometry data is often limited by non-uniqueness, finite measurement range, noise, and weak contrast between the sample of interest and the surrounding layers. These difficulties are especially relevant for soft-matter and other low-SLD systems, where different sample configurations can produce similar reflectivity curves. Magnetic reference layers offer one way to improve such experiments by introducing spin-dependent contrast in polarized neutron reflectometry. The choice of reference material, reference-layer thickness, and capping layer is therefore an experimental design problem.

This thesis investigates magnetic reference layer design for polarized neutron reflectometry using both signal-based and Bayesian design criteria. A specular reflectometry solver is implemented using a layered scattering-length-density model, Parratt recursion, and roughness corrections. The solver is embedded in a computational pipeline in which candidate designs are generated, simulated, scored, and optimized. Signal-based figures of merit, especially the total sensitivity figure, are compared with Bayesian objectives based on mutual information and posterior uncertainty reduction under a finite prior over possible samples of interest.

The results show that magnetic reference layer optimization depends on the chosen notion of informativeness. Signal sensitivity, posterior contraction, and polarization-specific information gain are related but not equivalent design criteria. The thesis therefore provides a simulation and optimization pipeline for MRL design and interprets MRL performance through the associated Bayesian inverse problem.