Nunes Grapiglia, GeovaniDavar, DânâDânâDavar2025-05-142025-05-142025-05-142023https://hdl.handle.net/2078.2/32387In this work we present a derivative-free trust-region method, based on finite-difference gradient approximations, for smooth convexly constrained optimization problems. We establish a worst-case complexity bound for the number of function evaluations that the method needs to find an approximate stationary point. Notably, the obtained bound depends only linearly on the problem dimension. Illustrative numerical results are also presented.Derivative-free optimizationTrust-region methodFinite-differencesWorst-case complexitytext::thesis::master thesisthesis:39215