Optimising Neutron Reflectometry Experiments through Sensitivity-Guided Data Analysis and Substrate Engineering

When fitting model parameters to experimental NR data, it is often challenging to decouple material related parameters, such as real and imaginary parts of the SLD, and structural parameters like layer thicknesses and interface roughness. In optical (photonic) analysis, many methods have been developed to solve such correlation problems. One approach is referred to as multiple sample analysis (MSA), where two or more similar samples, but with some parameters varied, are measured. In the subsequent analysis, two or more corresponding models are fitted simultaneously to the measurements. In NR there is an analogous standard technique of contrast variation, where the problem to decouple parameters is even more challenging since only intensities are measured with the loss of phase information – often termed the "phase problem". Furthermore, an additional possibility to find unique solutions of the SLD from reflectivity data is to use switchable magnetic reference layers (MRL). In the layered thin film structure, a MRL is deposited, whose characteristics can be controlled and, therefore, known beforehand. By applying an external magnetic field this layer is magnetised in a specific direction and probed with neutrons of different spin states. The MRL thus provides additional measurement data and a possibility to decouple the model parameters.

Since NR experiments are both extremely expensive to run, as well as difficult to access, it is important to make the best possible use of the experimental time. Reducing measurement time while maintaining high precision is key to expanding the applicability of neutron scattering techniques. To improve the effectiveness in extracting useful information from neutron reflectivity experiments we have designed substrate assemblies comprising a Si slab, a switchable MRL, and an inert top layer, specifically for modelling and characterisation of thin coatings with unknown properties with lipid bilayers or polymers being prime examples. An optimised substrate stack yields significantly different SLD profiles for polarised neutrons upon opposite magnetisations, effectively increasing the available data for obtaining the SLD profile for the unknown coating. The substrate assemblies are designed using the Holistic Optimization for Gaining Better Evidence from Neutrons HOGBEN software employing a sensitivity analysis based on Fisher information FI and correlation matrices, enabling systematic evaluation of the information gain for different configurations. The importance of this research lies in the potential to address the problem with limited beamtime access at neutron facilities by reducing the total measurement time required per sample without compromising obtained data quality.

The present study aims at optimising the design of solid substrates for polarised neutron reflectometry (PNR) experiments at the solid/liquid interface for the structural investigation of soft matter/biology samples. The substrate assembly in this work consisted of a Si single crystal with its native oxide, a ferromagnetic Fe reference layer, and a SiO2 capping layer. By exploiting the magnetic contrast provided by the Fe layer and performing measurements in different ambients (H2O, D2O, and SMW), we obtained several reflectivity datasets from a single sample. The measurements, carried out with the POLREF instrument at the ISIS Neutron and Muon Source, STFC Rutherford Appleton Laboratory, UK provided detailed information on the SLD profile of a head-tail-head bilayer lipid structure. Building on these results, we initiated sensitivity studies using parameter correlation and FI to find the optimal substrate assembly designs that minimise measurement time while preserving data quality. Our results demonstrate that sensitivity is significantly improved by jointly optimising the thicknesses of the Fe and the capping layer. In particular, we find that well-chosen configurations of the MRL and capping layers can yield equivalent experimental information with up to a fivefold reduction in measurement time.