# The Iterates of the Frank-Wolfe Algorithm May Not Converge

arXiv: https://arxiv.org/abs/2202.08711 · HTML: https://mlsanity.com/paper/2202.08711/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 17 Feb 2022.

## Built on by (most cited first, 5 of 5)

- [Muon Optimizes Under Spectral Norm Constraints](https://mlsanity.com/paper/2506.15054/) (2025)
- [Implicit Bias of AdamW: ℓ_∞ Norm Constrained Optimization](https://mlsanity.com/paper/2404.04454/) (2024)
- [Implicit regularization in AI meets generalized hardness of approximation in optimization -- Sharp results for diagonal linear networks](https://mlsanity.com/paper/2307.07410/) (2023)
- [LITE: Efficiently Estimating Gaussian Probability of Maximality](https://mlsanity.com/paper/2501.13535/) (2025)
- [Broximal Gradient Descent: A Projection-Free Sister of Projected Gradient Descent](https://mlsanity.com/paper/2605.08850/) (2026)

## Abstract (arXiv)

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)_t∈mathbbN. Under the usual assumptions, we design several counterexamples to the convergence of (xt)_t∈mathbbN, where f is d-time continuously differentiable, d≥2, and f(xt)→minCf. 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)_t∈mathbbN.