Cyclic Douglas–Rachford best approximation conjecture
Let be the underlying Hilbert space, and let be closed and convex with . Write for the metric projection onto . A pair is a best approximation pair when . The two-set cyclic Douglas–Rachford scheme is the iteration generated by the corresponding cyclic Douglas–Rachford operator.
Cyclic Douglas–Rachford best approximation conjecture. If a best approximation pair relative to exists, then the two-set cyclic Douglas–Rachford scheme converges weakly to a point such that is a best approximation pair relative to .
The conjecture concerns convergence in the inconsistent case, where the two closed convex sets are disjoint. The supplied source explicitly notes that non-convex settings can make the conjecture false; for the stated closed convex setting, the claim is presented as a conjecture and is therefore recorded here as refuted only according to the supplied status evidence.
References
Primary source
Jonathan M. Borwein and Matthew K. Tam, “A Cyclic Douglas-Rachford Iteration Scheme”, arXiv:1303.1859 (2013).
Progress summary
A 2013 paper appears to claim the conjecture is proved, but the supplied sources do not establish that the proof has been independently verified.
The conjecture concerns weak convergence of the cyclic Douglas–Rachford method for disjoint closed convex sets when a best-approximation pair exists. The supplied sources disagree about whether the original paper presents this as a conjecture or as a theorem.
Known results
- The original paper reports numerical evidence for the asserted behavior in inconsistent two-set problems.
- A related non-convex example, with and , fails, but does not address the closed-convex conjecture.
- For affine subspaces, the 2014 journal account reports norm convergence.
March–October 2013 claimed resolution
The arXiv version presents the closed-convex assertion as proved, and an October 2013 follow-up says the method yields best-approximation pairs when they exist. Neither supplied source gives enough proof verification to treat this as conclusively settled.
Current status (as of August 2026): A 2013 source claims the closed-convex conjecture is proved and a follow-up supports the conclusion, but independent verification is not documented, so the exact conjecture remains open as a verified theorem.
Solutions 0
No solutions have been posted yet.