Machine learning and light modeling in soft matter photonicsZaplotnik, Jaka (Avtor) Ravnik, Miha (Mentor) light modellingnumerical simulationneural networksoft matterliquid crystalsphotonicsopticslaserThis thesis presents numerical modelling of light in soft-matter photonic structures, specifically in cholesteric and nematic liquid crystals, together with the application of machine learning to material characterisation and photonic design. The numerical work combines finite-difference time-domain (FDTD) and frequency-domain (FDFD) optical simulations, in some cases extended to include light-matter interactions, and neural network-based supervised learning. In the first research direction, we demonstrate that supervised machine learning with neural networks, trained on numerically generated data, can be used to determine the elastic constants of a real nematic liquid crystal from specific experimental measurements. We further extend this approach, so far without experimental validation, to reconstruct the nematic director field from polarised microscopy images and to optimise liquid crystal director configurations that reshape a Gaussian beam of light into a desired intensity profile. In the second research direction, we characterise photonic eigenmodes in three types of cholesteric liquid crystal (CLC) resonators -- simple planar CLC cells, planar CLC cells with an embedded isotropic layer, and cylindrical CLC resonators -- and show how their eigenfrequencies and quality factors depend on material and geometric parameters. In simple planar CLC cells, we also apply lasing theory to compare lasing thresholds in different configurations. We further introduce optically stable CLC resonators with patterned surfaces and curved cholesteric layers, enabling well-defined transverse modes, as well as resonators with surface topological defects that directly generate vortex beams. We also model whispering-gallery mode spectra in a liquid crystal droplet upon cooling across the isotropic--nematic phase transition temperature, during which a nematic domain grows inward from the surface, shifting the resonance wavelengths. In the third direction, we simulate pulsed liquid crystal microlasers coupled to polymer waveguides, in a cholesteric microcavity and in a nematic droplet, and show that combining a pump pulse with a delayed red-shifted pulse that depletes the excited dye via stimulated emission depletion (STED) allows the lasing intensity to be controlled on a nanosecond timescale, demonstrating an efficient mechanism for all-optical switching in soft matter. Together, the results of this thesis demonstrate that machine learning is a practical tool for characterising liquid crystal materials and optimising their photonic response. Furthermore, this work advances the understanding of light confinement and lasing in CLC structures and numerically confirms that resonant stimulated-emission depletion is a promising all-optical switching mechanism in soft matter. Collectively, these contributions support the development of novel soft matter photonic applications.20262026-06-12 08:15:18Doktorsko delo/naloga183384VisID: 160452COBISS_ID: 281209603sl