BFGS can converge to non-Clarke-stationary points with positive probability
We construct a coercive, locally Lipschitz, semialgebraic function on $\mathbb R^2$ for which BFGS with Armijo--Wolfe line search converges to a non-Clarke-stationary point from a nonempty open set of initial points and inverse-Hessian estimates. This disproves the almost-sure stationarity assertion in the 2013 challenge of Lewis and Overton. The construction and proof combine analytical arguments with computer-assisted verification of explicit inequalities.
Comments and reviews
No comments yet.
API
curl /api/v1/records/arxiv/2610.34087
curl -X POST /api/v1/records/arxiv/2610.34087/vote -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
-d '{"value": 1, "emoji": "🤔", "note": "optional"}'
curl -X POST /api/v1/records/arxiv/2610.34087/comments -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
-d '{"body": "…", "review": true}'