2 optimization seminars: 22/10/07 and 5/12/07
Forum 'Annonces' - Sujet créé le 2007-10-02 par Leo Liberti
Two optimization seminars are forthcoming at LIX, Ecole Polytechnique.
Please feel free to forward this notice to all interested parties.
Yours,
Leo Liberti
-------------------
Lundi 22/10/2007, Salle de Seminaires du LIX, 10:30:
Split Cuts from Elementary Disjunctions
Andrea Lodi
DEIS, University of Bologna, Italy
andrea.lodi@unibo.it
Split cuts are a famous class of valid inequalities for mixed integer linear
programs. For this class of inequalities, the separation problem can be
formulated as a (non linear) mixed integer program, which easily reduces to
a linear program if the disjunction associated with the split cuts is xed.
Another important class of valid inequalities are Mixed Integer Gomory(MIG)
cuts. Violated MIG cuts can be read from the optimal simplex tableau with
limited computational eort. We inves- tigate the correspondence between MIG
cuts from the simplex tableau and split cuts. In particular, we present a
linear programming model for separating violated inequalities (LPSEP), whose
feasible solutions correspond to violated split cuts. We show that MIG cuts
from the simplex tableau correspond to basic solutions of LPSEP, which can
be viewed as a protable tool for strengthening MIG cuts. A computa- tional
analysis trying to highlight both algebraic and geometric aspects of cuts is
reported and discussed. Joint work with Matteo Fischetti and Andrea
Tramontani.
--------------------
Mercredi 5/12/2007, Salle de Seminaires du LIX, 10:30:
Certainty equivalent predictions for dynamic
systems through multivalued non linear variational
inequalities
Giacomo Patrizi
DSPSA, Universita` di Roma - La Sapienza, Roma, Italy
An empirical process is a data set, through which it is desired to formulate
a model of the operation of the underlying phenomenon, so that suitable
controls to achieve determinate objectives may be defined and applied. The
aim of this talk is to formulate an algorithm based on generalized
variational inequalities with set-valued maps to determine at the same time
the optimal model of the process and the optimal control to be applied,
which must be certainty equivalent to the ex post facto optimal control of
the process. This formulation will be contrasted to the well known robust
optimization approach. The problem so defined can be formulated as a
generalized variational inequality problem and it will be shown how this
formulation generalizes classical variational problems. Uncertainty effects,
often defined as additive components in the traditional approach, may be an
essential part of the nonlinear empirical process specied and thus cannot be
excluded .
Please feel free to forward this notice to all interested parties.
Yours,
Leo Liberti
-------------------
Lundi 22/10/2007, Salle de Seminaires du LIX, 10:30:
Split Cuts from Elementary Disjunctions
Andrea Lodi
DEIS, University of Bologna, Italy
andrea.lodi@unibo.it
Split cuts are a famous class of valid inequalities for mixed integer linear
programs. For this class of inequalities, the separation problem can be
formulated as a (non linear) mixed integer program, which easily reduces to
a linear program if the disjunction associated with the split cuts is xed.
Another important class of valid inequalities are Mixed Integer Gomory(MIG)
cuts. Violated MIG cuts can be read from the optimal simplex tableau with
limited computational eort. We inves- tigate the correspondence between MIG
cuts from the simplex tableau and split cuts. In particular, we present a
linear programming model for separating violated inequalities (LPSEP), whose
feasible solutions correspond to violated split cuts. We show that MIG cuts
from the simplex tableau correspond to basic solutions of LPSEP, which can
be viewed as a protable tool for strengthening MIG cuts. A computa- tional
analysis trying to highlight both algebraic and geometric aspects of cuts is
reported and discussed. Joint work with Matteo Fischetti and Andrea
Tramontani.
--------------------
Mercredi 5/12/2007, Salle de Seminaires du LIX, 10:30:
Certainty equivalent predictions for dynamic
systems through multivalued non linear variational
inequalities
Giacomo Patrizi
DSPSA, Universita` di Roma - La Sapienza, Roma, Italy
An empirical process is a data set, through which it is desired to formulate
a model of the operation of the underlying phenomenon, so that suitable
controls to achieve determinate objectives may be defined and applied. The
aim of this talk is to formulate an algorithm based on generalized
variational inequalities with set-valued maps to determine at the same time
the optimal model of the process and the optimal control to be applied,
which must be certainty equivalent to the ex post facto optimal control of
the process. This formulation will be contrasted to the well known robust
optimization approach. The problem so defined can be formulated as a
generalized variational inequality problem and it will be shown how this
formulation generalizes classical variational problems. Uncertainty effects,
often defined as additive components in the traditional approach, may be an
essential part of the nonlinear empirical process specied and thus cannot be
excluded .