The Iterates of the Frank-Wolfe Algorithm May Not Converge

Abstract

The Frank-Wolfe algorithm is a popular method for minimizing a smooth convex function f over a compact convex set C. While many convergence results have been derived in terms of function values, hardly nothing is known about the convergence behavior of the sequence of iterates (xt)t2N. Under the usual assumptions, we design several counterexamples to the convergence of (xt)t2N, where f is d-time continuously differentiable, d > 2, and f(xt) ---> minC f. Our counterexamples cover the cases of open-loop, closed-loop, and line-search step-size strategies. We do not assume misspecification of the linear minimization oracle and our results thus hold regardless of the points it returns, demonstrating the fundamental pathologies in the convergence behavior of (xt)t2N.

Replaced by

Jérôme Bolte, Cyrille Combettes, and Edouard Pauwels, “The Iterates of the Frank–Wolfe Algorithm May Not Converge”, Mathematics of Operations Research, 2024, forthcoming.

Reference

Jérôme Bolte, Cyrille Combettes, and Edouard Pauwels, “The Iterates of the Frank-Wolfe Algorithm May Not Converge”, TSE Working Paper, n. 22-1311, February 2022.