A distributed Jacobi algorithm for large scale constrained convex optimization

Dang Doan
Systems Control and Optimization Laboratory
University of Freiburg, Germany

This video was captured during his presentation at Group For Research in Analysis, Université de Montréal, Montréal, Canada on May 29, 2017.
Link: https://www.gerad.ca/en/events/1461

For further information, please freely access to his research group at https://www.syscop.de/people/dang-doan

Abstract:
We consider sparse convex optimization problems, under general affine constraints. There are couplings between variables in both the cost function and constraints of the optimization problem.

We propose a distributed Jacobi algorithm to solve this problem in a cooperative manner. Using a local update in each iteration which is equivalent to a convex combination of new local solutions and old iterates, the Jacobi algorithm guarantees to achieve feasible solutions at every iteration.

We provide the a posteriori certification for centralized optimality of distributed solutions, and a priori conditions that guarantee convergence to optimality in several problem settings.

The proposed approach is useful for distributed model predictive control applications where feasibility is an important requirement. It fosters distributing the computations, especially in settings with a large number of subsystems, a sparse coupling structure, and local communication is available.