Decomposition Methods for Large-scale Structured Mixed-integer Programs

Kibaek Kim
Seminar

In this talk, I will present decomposition methods for solving large-scale structured mixed-integer programs and introduce a parallel software implementation, called DSP. The first part will show a new dual decomposition to address stochastic mixed-integer programs. This includes the derivation of valid inequalities that tighten Lagrangian subproblems and that aid the exclusion of infeasible first-stage solutions for problems without (relative) complete recourse. This also includes an interior-point cutting-plane method to solve the Lagrangian master problem and provide termination criteria that guarantee finite termination of the overall algorithm. The second part will present spatial decomposition and temporal decomposition for large-scale network problems. In the last part, I will introduce DSP and provide numerical results using unit commitment problem instances.