VibeMathedMath problems solved with AI

Lorist-Schwenninger Remark 2 positivity question

orist and Schwenninger prove Crouzeix's conjecture (arXiv:2608.03841, Lemma 1) by combining a lower bound (their inequality (4)) with an upper bound (inequality (5)). In Remark 2 they observe that (5) alone gives κ1+1E1Tx,x\kappa \le 1 + \sqrt{1 - \Re\langle E_1 Tx,x\rangle}, so positivity of E1Tx,x\Re\langle E_1 Tx,x\rangle would prove the lemma outright. They write: "it is unclear whether E1Tx,x0\Re\langle E_1 Tx,x\rangle \ge 0 in general."

Result
Disproved(see note)
Status
Resolved
AI contribution
AI-discovered
Method
Construction
Field
Numerical range / Crouzeix's conjecture
Posed by
Emiel Lorist, Felix L. Schwenninger
Year posed
2026
Years open
0y
Solved
2026-08-14
Model
Claude Opus 5.0
Vendor
Anthropic
Collaborators
Verification
Site-confirmed
Publication
Announced
Significance
4 / 100
Disclosed cost
Wikipedia
No dedicated article

What was actually shown

Nothing in the paper's proof is affected. At the witness, inequality (4) holds with slack +27.0, inequality (5) holds with equality (f is inner), and the theorem itself holds with slack 2 - kappa = +0.80. Only the shortcut Remark 2 floats is refuted.

What the AI did

The counterexample was found and verified computationally by Claude Opus 5.0 in a single session. The search identified the mechanism (the centres of the domain and the numerical range must be separated to make the quantity negative), found a 2x2 witness, reduced it to a closed-form algebraic expression, and verified the sign in exact arithmetic. The sharp consequence — that inequality (5) alone recovers exactly the Crouzeix-Palencia constant 1+sqrt(2) — was derived in the same session. The human operator directed the investigation to Remark 2 and approved the final write-up; all mathematical content was produced by the model.

Verification

Reproduced by this site on 17 August 2026. Remark 2 was confirmed verbatim in the Lorist-Schwenninger LaTeX source, including the exact sentence "it is unclear whether Re<E_1 Tx,x> >= 0 in general". The counterexample repository pins the claim in a statement file and was run here: the exact sympy certificate gives m = -4/95 - 1188*sqrt(90709)/8617355 = -0.0836..., with the sign certified through its minimal polynomial rather than floating point, and the independent mpmath implementation (no shared code) agrees; both also confirm the paper's own bounds still hold at the witness, so the counterexample refutes the remark's hope without touching Lemma 1. The mechanism note - the domain centre and numerical-range centre must separate - matches what the certificate shows. No human peer review; the tier records this site's own reproduction.

Sources

Related entries

Submitted by ZestyWombat854 on

Changelog3 changes
  • Rasmus Lindahlchanged verificationNote from Three independent implementations agree: exact sympy algebra (sign certified via minimal p… to Reproduced by this site on 17 August 2026. Remark 2 was confirmed verbatim in the Lorist-S…, also verification, significanceNote, significance
  • Rasmus Lindahlapproved this entry
  • ZestyWombat854submitted this entry

Discussion