Aragón Artacho–Borwein–Tam Douglas–Rachford strong-convergence question
Let be an infinite-dimensional Hilbert space, let be nonempty closed convex sets with , and let . Define the Douglas–Rachford iterates by , where and . Must weak convergence of imply strong convergence, that is, does necessarily imply ?
References
- A Convex Feasibility Problem for Which Douglas-Rachford Converges Only Weakly — arXiv — Jan Quan, Panagiotis Patrinos
Source: arXiv
Progress summary
A new unrefereed paper claims a counterexample showing that Douglas–Rachford iterations can approach a solution weakly without approaching it in distance, so the answer is no if the construction is correct.
The question asks whether weak convergence of Douglas–Rachford iterations for two intersecting closed convex sets in an infinite-dimensional Hilbert space must also be convergence in distance. Earlier literature treated this as open for the two-set method.
Known results
- Weak convergence holds for the standard method under the stated convexity and intersection assumptions.
- Hundal exhibited failure of convergence in distance for alternating projections.
- Damek, Davis, and Yin (2015) proved arbitrarily slow Douglas–Rachford convergence in infinite dimensions, without giving this counterexample.
- Convergence in distance is known under additional regularity assumptions.
October 2026 claimed counterexample
On October 20, 2026, Jan Quan and Panagiotis Patrinos reported an explicit lifted construction in A Convex Feasibility Problem for Which Douglas–Rachford Converges Only Weakly. If valid, it gives a negative answer to the exact question. The preprint is unrefereed, so the resolution remains unverified.
Current status (as of October 2026): The general implication is claimed to be false by an explicit preprint construction, but that claimed counterexample has not been independently verified.
Solutions 0
No solutions have been posted yet.