Paper claims proof of the 38-year-old k-server conjecture

Overview

Updated Sep 15

A paper posted to arXiv on September 14 claims to settle the k-server conjecture, a problem in theoretical computer science open since 1988. Authors Christian Coester, Elias Koutsoupias, and Marek Zbysiński say they have proven that the Work Function Algorithm achieves the optimal competitive ratio of k on every metric space.

The k-server problem concerns moving k servers to handle requests arriving without advance notice. The best previous bound, a competitive ratio of 2k-1, came from Koutsoupias and Papadimitriou in 1995. If this proof holds, it resolves a central question in competitive analysis.

Why it matters

This proof settles a 38-year-old open problem and shows the Work Function Algorithm is optimal for online server movement on every metric space.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

Years the conjecture remained open

Posed in 1988, with a claimed proof posted in September 2026.

2k-1

Best previous competitive ratio

Koutsoupias and Papadimitriou proved this bound for the Work Function Algorithm in 1995.

Claimed competitive ratio

The conjectured optimum, which the new paper claims to prove.

Voices

Curated perspectives — historical figures and your fellow readers.

People Involved

Christian Coester

Computer scientist, co-author of the 2026 proof

Co-author of the claimed proof

Elias Koutsoupias

Computer scientist, co-author of the 2026 proof

Co-author of the claimed proof

Marek Zbysiński

Computer scientist, co-author of the 2026 proof

Co-author of the claimed proof

Daniel Sleator

Computer scientist, co-proposer of the conjecture

Posed the conjecture in 1988

Organizations Involved

arXiv

Preprint server

Hosted the proof paper on September 14, 2026

Preprint server where the claimed proof was posted.

Timeline

1988 September 2026

5 events Latest: September 14th, 2026 · 2 months ago

Tap a bar to jump to that date

September 2026
Claimed proof posted
Latest Mathematical

Coester, Koutsoupias, and Zbysiński post a proof of the k-server conjecture to arXiv.
September 14th, 2026
July 2021
Unifying potential fails on circle
Mathematical

Coester and Koutsoupias' potential provably fails for three servers on a circle.
July 2021
1995
2k-1 bound established
Mathematical

Koutsoupias and Papadimitriou prove WFA reaches competitive ratio 2k-1.
1995
1991
Tree metrics resolved
Mathematical

Chrobak and Larmore prove the conjecture for tree metrics.
1991
1988
Conjecture posed
Mathematical

Manasse, McGeoch, and Sleator pose the k-server conjecture.
1988

Scenarios

k-Server Proof Accepted by Peer Review

Likely Resolves by End of 2027

Discussed by: Theoretical computer science community and arXiv watchers

The paper passes review at a top conference like STOC or FOCS, or in a journal such as the Journal of the ACM. The proof becomes accepted knowledge in competitive analysis. Acceptance for a result of this magnitude typically takes 6 to 18 months.

Gap Found in k-Server Proof

Possible Resolves by Q2 2027

Discussed by: Referees and competitive analysis researchers

An error surfaces during peer review or community scrutiny, forcing a retraction or a major revision of the claim. This is a real risk. The 2021 paper by Coester and Koutsoupias identified a similar potential that failed on the circle, showing how subtle the problem is.

Formal Verification Confirms the Proof

Possible Resolves by End of 2028

Discussed by: Formal methods groups and ProofAtlas

Researchers encode the proof in an interactive theorem prover like Lean, Coq, or Isabelle, machine-checking every step. This would provide an independent check of correctness. Formalization efforts for major proofs typically take several years.

Historical Context

3 moments from history that rhyme with this story — and how they unfolded.

1904–2006

Poincaré Conjecture (1904–2006)

Grigori Perelman posted his proof of the 100-year-old conjecture to arXiv in 2002 and 2003. The mathematics community spent years scrutinizing it, with doubts persisting into 2006.

Then

Perelman's proof was verified and accepted by 2006; Perelman declined the Fields Medal and the Clay Millennium Prize.

Now

Settled the first of the seven Millennium Prize Problems and set a modern standard for how arXiv proofs get verified.

Why this matters now

Shows how claims resolving famous open problems take years of peer scrutiny, which is what the k-server proof now faces.

1637–1995

Fermat's Last Theorem (1637–1995)

Andrew Wiles announced a proof in June 1993; a gap appeared in his manuscript months later. He repaired it with Richard Taylor in 1994, publishing the corrected proof in 1995.

Then

The corrected proof passed review within about a year of the initial gap.

Now

Resolved a 358-year-old problem and demonstrated that major proofs can have fixable gaps.

Why this matters now

Even proofs of famous conjectures can have flaws identified during review, but those flaws are often repairable.

1852–1976

Four Color Theorem (1852–1976)

Kenneth Appel and Wolfgang Haken proved the 120-year-old conjecture in 1976 by checking nearly 2,000 cases with computers, which drew initial skepticism.

Then

The computational method was debated for years; alternative proofs appeared in 1997 and 2005.

Now

Now accepted, and it established standards for computer-assisted proof.

Why this matters now

New or unusual proof techniques, like those used for the k-server claim, can be controversial before being accepted.