Answers are grounded in this paper and cite their sources.
Theorem. Consider the Reissner–Nordström–de Sitter black hole with horizons at r₋ = 19/20, r₊ = 1 and r_c = 2 (cosmological constant Λ = 1200/4301, M = 6903/8602, Q² = 3002/4301). For the massless scalar field, every nonzero quasinormal mode of the exterior, for every angular momentum ℓ, satisfies Im ω < −(3/5)κ₋, where κ₋ is the surface gravity of the Cauchy horizon. The frequency ω = 0 occurs only for ℓ = 0, with a constant mode, and is a simple zero. By the theorem of Hintz and Vasy (arXiv:1512.08004), solutions with smooth initial data extend across the Cauchy horizon in H¹_loc and are continuous up to it, so the Christodoulou formulation of strong cosmic censorship fails for this linear model at this parameter point. Nothing is claimed for rough data, other parameters, or the nonlinear Einstein–Maxwell–scalar problem. Method. For ℓ ≤ 9 and bounded frequencies, interval Taylor models of the radial Wronskian are evaluated on 415 discs, with a Rouché disc at ω = 0. All other frequencies are excluded by a Hermite-type monotone quantity integrated along complex contours in the tortoise coordinate. A slit-domain condition controls the analytic continuation of the outgoing solutions. For ℓ ≥ 10 the contours and weights scale with ℓ(ℓ+1), so finitely many interval inequalities cover all angular momenta at once. Trust assumptions. The theorems of Hintz and Vasy as cited; standard ODE and complex-analysis facts; and the correctness of the Arb ball-arithmetic library used through python-flint. Part of the computation was rechecked independently with mpmath interval arithmetic. The remaining parts were checked only by two independent implementations built on Arb. The article lists which parts these are. Files. The article (PDF) and a supplementary archive with: the certificate programs and records; the frozen specifications and complete working proofs; a one-step check (check.py, which with --replay reruns every certificate); the reviews and independent checks; the trust list. AI tools. AI systems (Claude Opus 5.5, Claude Fable 5.2 and GPT-6 Astra) did the computational and drafting work under the author's direction; the article's section on AI use describes it. Blind reviews crossed developers: what one developer's model drafted, the other developer's model examined. Models of both families re-proved the theorem starting from its statement; the cross-family run was also given the numerical records. Two adversarial reviews and a recomputation with a second interval library were carried out. No human expert has yet checked the proofs line by line.
No takes yet. Share an insight, caveat, or question.
Neko2333OWO (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: