Recursive central rounding for mixed integer programs

We introduce a new rounding heuristic for mixed integer programs. Starting from a fractional solution, the new approach is based on recursively fixing a subset of the discrete variables while using the analytic center to re-center the remaining ones. The proposed rounding approach can be used independently or integrated with other heuristics. We demonstrate both setups by first using the proposed approach to round the optimal solution of the linear programming relaxation. We then integrate the proposed rounding heuristic with the feasibility pump by replacing the original simple rounding function of the feasibility pump. We conduct computational testing on mixed integer problems from MIPLIB and CORAL and on mixed integer quadratic problems from MIQPLIB. The proposed algorithm is computationally efficient and provides good quality feasible solutions.