In this paper, we show that Ye–Todd–Mizuno's O(√(n)L)-iteration homogeneous and self-dual linear programming (LP) algorithm possesses quadratic convergence of the duality gap to zero. In the case of infeasibility, this shows that a homogenizing variable quadratically converges to zero (which proves that at least one of the primal and dual LP problems is infeasible) and implies that the iterates of the (original) LP variable quadratically diverge. Thus, we have established a complete asymptotic convergence result for interior-point algorithms without any assumption on the LP problem.
