Simulação e modelagem computacional em EPR de onda contínua com ênfase a radicais nitróxidos em regime de movimento rápido

• Ano de defesa: 2019
• Autor(a) principal: Gabriel de Albuquerque Silva
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ICX - DEPARTAMENTO DE FÍSICA; Programa de Pós-Graduação em Física; UFMG
• Palavras-chave em Português: cw EPR; Simulação em cw EPR; Regime fast-motion; Ajuste de mínimos quadrados não-linear; Ressonância paramagnética eletrônica; Métodos de simulação
• Link de acesso: http://hdl.handle.net/1843/31436

As the title suggests, in this study we performed simulations and computational modeling in cw (continuous-wave) EPR using some numerical methods. The entire procedure was written in C language and the graphical representations were automatically generated in the GnuPlot program. Hereafter, we intend to "close" the code and create an executable file that can be applied specifically to nitroxide radicals in fast-motion regime. The first part of this procedure is to simulate the spectrum using an appropriate model. This involves obtaining the energies and eigenstates of a spin Hamiltonian by numerical diagonalization, recursive bisection of the magnetic field interval, interpolation by cubic splines of energy levels, search for the resonant fields, and other details for the construction of the final spectrum as lineshapes and linewidths model. Computational modeling is the second part of the procedure. Basically, it is the adjustment of a simulated spectrum to an experimental spectrum. This realization is done by nonlinear least squares fitting. The minimization techniques used were the Levenberg-Marquardt, Nelder-Mead simplex and simulated annealing methods with the Monte Carlo algorithm. Through the covariance matrix it was possible to estimate uncertainties of the parameters of the spin Hamiltonian, linewidths and correlation time for rotational diffusion. However, the nonlinear least squares fitting takes on an experimental spectrum with Gaussian noise, but in practice these deviations can be neither totally random nor fully Gaussian. This may lead to underestimation of parameter uncertainties, thus requiring further evaluation of experimental data, such as observing the residuals of the fit, their distribution, and even other statistical tests as needed. The target function can be chosen according to the distribution of experimental deviations.