Seminar 2022.2
November 3, 2022
A Cubic Regularization of Newton Method with Finite-Difference Hessian Approximations
Max Leandro Nobre Gonçalves
In this talk, we present a version of the Cubic Regularization of the Newton method for unconstrained nonconvex optimization, in which the Hessian matrices are approximated by forward gradient differences. The regularization parameter of the cubic models and the accuracy of the Hessian approximations are jointly adjusted using a nonmonotone line-search criterion. Complexity analysis of the proposed algorithm is discussed and preliminary numerical experiments are presented to confirm our theoretical findings.
November 10, 2022
Fenchel Conjugate via Busemann Function on Hadamard Manifolds
Glaydston de Carvalho Bento
In this paper we introduce a Fenchel-type conjugate, given as the supremum of convex functions, via the Busemann functions. Due the absence of non-constant affine functions in Hadamard manifolds and taking into account that the Busemann functions are smooth convex functions with constant gradient, our study ensures that our proposal of Fenchel conjugate seems to be the most adequate to cover the absence of approximations by non-constant affine functions. Besides, it is possible to evidence the influence of the sectional curvature in obtaining the main results. In particular, we have illustrated that the difference between a proper, lsc, convex function and its biconjugate is a constant that depends on the sectional curvature of the manifold, showing that in general a Fenchel-Moreau type theorem is directly influenced by the sectional curvature.
We also present some applications formulated in terms of the Fenchel's conjugate.
November 17, 2022
An inexact scalarization proximal point algorithm for constrained multiobjective minimization with quasiconvex objective functions
Erik Alex Papa Quiroz
In this talk, an inexact proximal point algorithm with proximal distances is introduced to solve multiobjective minimization problems with locally Lipschitz quasiconvex objective functions constrained to a closed convex set with nonempty interior. We prove the convergence of the sequence generated by the algorithm to a Pareto-Clarke critical point and convergence to a Pareto solution when the objective functions are convex or when we consider the exact proximal algorithm for the quasiconvex problem when the proximal parameter converges to zero. We give some conditions to obtain the finite termination of the algorithm and prove linear or superlinear rate of convergence for a large class of proximal distances. Thus, this work extends the convergence of the proximal point method for quasiconvex multiobjective minimization problems and improves recent convergence results of previous works for the convex case
December 1, 2022
Copositivity with respect to general cones and existence results for complementarity problems
Orizon Pereira Ferreira
We test copositivity of operators with respect to general cones by using a gradient projection algorithm for solving constrained convex problems on the sphere in finite dimensional vector spaces. This approach can also be used to analyse solvability of nonlinear cone-complementarity problems. The convergence analysis of the gradient projection algorithm on the sphere is also presented. To our best knowledge this is the first algorithm which can be used to test copositivity of operators with respect to general cones. Numerical results, including testing such copositivity of operators, are presented
December 8, 2022
Convergence of Quasi-Newton Methods for Solving Constrained Generalized Equations
Paulo César da Silva Júnior
In this paper, we focus on quasi-Newton methods to solve constrained generalized equations.
As is well-known, this problem was firstly studied by Robinson and Josephy in the 70's.
Since then, it has been extensively studied by many other researchers, specially Dontchev and Rockafellar.
Here, we propose a Broyden-type quasi-Newton approach to deal with constrained generalized equations. Projections onto the feasible set are inexact. The local convergence of general quasi-Newton approach is established under a bounded deterioration of the update matrix and Lipschitz continuity hypotheses. In particular, we prove that a general scheme converges linearly to the solution under suitable assumptions.
January 12, 2023
Nonlinear conjugate gradient methods for Vector Optimization with approximated steepest descent direction and without Wolfe conditions
Luis Román Lucambio Pérez
We will present three variants of our algorithm. The first is for convex problems, the second uses the Lipschitz constant of the objective, and finally, a variant that needs a good approximation of the Lipschitz constant.
January 19, 2023
Proximal Gradient Methods for Composite Multiobjective Optimization
Jefferson D. Gonçalves de Melo
In this talk, we will review some proximal gradient methods for solving composite multiobjective/vectorial optimization problems whose objective function can be expressed as the sum of two convex functions, a differentiable and a prox-friendly non-smooth one. We will also discuss two line-searches procedures that allow us to consider problems for which the Lipschitz constant of the Jacobian of the differentiable component of the objective function is hard to compute or does not exist.
February 2, 2023
Métodos Quase-Newton com Busca Linear de Wolfe para Otimização Multiobjetivo.
Danilo Rodrigues de Souza
Propomos três métodos tipo BFGS com busca linear de Wolfe para otimização multiobjetivo irrestrita. Os algoritmos são bem definidos mesmo para problemas gerais não convexos. O primeiro mimetiza o método BFGS clássico para otimização escalar, para o qual a convergência global e R-linear para um ponto Pareto ótimo são estabelecidas para problemas fortemente convexos. Na análise de convergência local, a taxa é Q-superlinear. Os outros dois algoritmos são versões globalmente convergentes do método BFGS para problemas não convexos. Finalmente, caracterizamos explicitamente de maneira não assintótica a convergência local superlinear do método BFGS para otimização multiobjetivo.
February 9, 2023
Globally convergent inexact quasi-Newton methods for solving nonlinear systems
Paulo César da Silva Júnior
Large scale nonlinear systems of equations can be solved by means of inexact quasi-Newton methods. A global convergence theory is introduced that guarantees that, under reasonable assumptions, the algorithmic sequence converges to a solution of the problem. Under additional standard assumptions, superlinear convergence is preserved.
Schedule:
November 3 - Max 10 - Glaydston 17 - Erick 24 - Conpeex December 1 - Orizon 08 - Paulo César 15 - Leandro January 12 - Luis Roman 19 - Jefferson 26 - Paulo César Fevereiro 2 - Danilo 9 - 16-