Light-scattering experiments in dye-doped liquid crystals both to determine crystal parameters and to construct consistent neural network empirical physical formulas for scattering amplitudes

• Published in: Optics communications Vol. 284; no. 8; pp. 2173 - 2181
• Main Authors: Yildiz, Nihat, San, Sait Eren, Polat, Ömer
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 15.04.2011
• Subjects: Amplitudes; Construction; Electric potential; Liquid crystal; Liquid crystals; Neural network; Neural networks; Nonlinear optics; Nonlinearity; Regression; Scattering amplitude; Scattering amplitude; Neural network; Liquid crystal; Nonlinear optics
• ISSN: 0030-4018; 1873-0310
• DOI: 10.1016/j.optcom.2010.12.093

The aim of this paper is two-fold. Firstly, static laser light-scattering amplitude measurements in azo-dye doped nematic liquid crystals (NLCs) were made versus scattering angle, temperature and applied bias voltage. Three NLC parameters were determined: the elastic constant ratios K11/K22 by regression, phase transition temperatures, and Freedericksz voltages from the graphs. They were all doping ratio dependent. Secondly, as a novel approach, by a nonlinear universal function approximator layered feedforward neural network (LFNN) we constructed an explicit form of empirical physical formulas (EPFs) for theoretically unknown nonlinear azo-dye doped NLC scattering amplitude functions. Excellent LFNN test set (i.e. yet-to-be measured experimental data) predictions prove that the constructed LFNN-EPPs estimate unknown amplitude functions consistently. The LFFN-EPFs, too, confirmed the doping-ratio dependency. Also, comparing LFNN and regression amplitude fits, the LFNN fits were significantly better. In conclusion, physical laws embedded in the physical data can be consistently extracted by LFNN. One major potential application in the nonlinear optics domain is that these LFNN-EPFs, by differentiation, integration, minimization, etc., can be used to obtain further NLC scattering amplitude related molecular structural physical quantities. This could in turn help us to develop new nonlinear optical materials. ►First aim: dye-doped liquid crystal parameters were determined by light scattering. ►All parameters were found to be doping-ratio dependent. ►Second aim: we formed neural network empirical scattering amplitude formulas. ►Excellent agreement was seen between experimental and neural scattering amplitudes. ►Neural networks can be very useful in liquid crystal structure investigations.