Jean Charles GILBERT
INRIA (research center of Paris)
2, rue Simone Iff
75589 Paris Cedex 12
Tel : (33) 1 80 49 41 20
Optimization, complementarity problems, computational differentiation, and applications.
Current and previous activities
Augmented Lagrangian approaches in quadratic optimization
Interior-point methods in nonlinear optimization
Sequential quadratic programming (a Newton method in constrained optimization)
Algebraic methods in derivative-free optimization
Decomposition-coordination algorithms for large-scale problems, including in stochastic optimization
Stochastic optimisation : methods for problems decomposed on a scenario tree
Positive semi-definie optimization : SDP relaxation of polynomial problems
Polynomial optimisation: moment-sos method
Convex optimization: convex quadratic optimization, L1 and gauge recovery
Numerical methods for complementarity problems
Newton-min algorithm for linear and nonlinear problems
2013-2016: Global optimization of the power flow in an electricity transportation network, with Cédric Josz (RTE, Inria, Paris VI), Jean Maeght (RTE), and Patrick Panciatici (RTE).
Ibtihel Ben Gharbia,
Marc Le Bret,
J. Le Foll,
Georges Le Vey,
Lecture notes (some in French)
Libopt: an environment for testing solvers on
heterogeneous collections of problems (with X. Jonsson).
M1CG1: implements a CG algorithm for solving in sequence large
scale linear systems with similar positive definite matrices; the solver generates an l-BFGS preconditioning matrix that can be
used to accelarate finding the solution of the next linear system.
large-scale unconstrained optimization by a limited memory BFGS method
(with C. Lemaréchal).
Qpalm, Oqla (with Émilie
Joannopoulos): two solvers of convex quadratic optimization problems, with the following features:
written in Matlab (Qpalm) and C++ (Oqla),
augmented Lagrangian method, with minimization of the augmented Lagrangian by an active set method,
can deal with an infeasible problem, in which case a solution to the closest feasible problem is computed,
can deal with an unbounded problem, in which case an unboundedness direction is computed,
the speed of convergence prescribed by the user is ensured (provided rounding error does not prevail),
adaptation to an SQP solver of nonlinear optimizatin problems.
SQPlab, SQPlight et SQPpro: a suite of optimization
solvers based on the SQP Newtonian approach.
Sdolab (version 0.4): Matlab code that solves a semidefinite optimization problem in complex numbers (dense
Small pieces of software
- pgr: polyhedral gauge recovery problem solver
(minimization of a polyhedral gauge on an affine subspace, with uniqueness detection).
François Akoa (2000),
Ibtihel Ben Gharbia,
Laurent Chauvier (1995),
G. Do (1995),
Xavier Jonsson (1997),
Souaid Mezouar (2006),
Houssem Miled (2005),
Philippe Ségalat (1998),
Jan Stuchlý (2006),