[Todos CMAT] LPE - Ponencia de profesor visitante.

[Franco Robledo]

Estimados Profesores y Estudiantes:
      Mediante el presente mail el Laboratorio de Probabilidad y 
Estadística (Facultad de Ingeniería-Facultad de Ciencias),
los invita muy cordialmente a la siguiente ponencia a realizarse el 
Miercoles 4 de junio de 14:00 a 16:00 horas
en el Salón de Seminarios I de Facultad de Ciencias (subsuelo).

Profesor Invitado:
Mauricio de Souza, de la U.Federal de Minas Gerais. (profesor visitante 
del Dpto. de Inv. Operativa).

Título de la Ponencia: Multicommodity Flow Problems with Separable 
Piecewise Convex Costs.

Resumen en inglés:
We consider here a multicommodity flow network optimization problem with
non-convex but piecewise convex arc cost
functions. We derive complete optimality conditions for local minima 
based on
negative-cost cycles associated with each commodity. These conditions do 
not
extend to the convex non-smooth case.

We present cycle-based algorithmic approaches to find local minima of a non
convex and non smooth model for capacity expansion of a network supporting
multicommodity flows. By exploiting complete optimality conditions for 
local
minima, we give the convergence analysis of the negative-cost cycle 
canceling
method. The cycle canceling method is embedded in a tabu search strategy to
explore the solution space beyond the first local optimum. Reaching a local
optimum, the idea is to accept a cost-increasing solution by pushing flow
around a positive-cost cycle, and then, to make use of the cycle cancelling
method incorporating tabu search memory structures to find high quality 
local
optima. Computational experiments on instances of the literature show 
that the
tabu search algorithm can significantly improve feasible solutions 
obtained by
the local optimization procedure, and outperforms the capacity and flow
assignment heuristic in terms of solution quality.

Desde ya muchas gracias.
Salúdalo muy atentamente,
         Dr. Ing. Franco Robledo Amoza