What was already known
Nikiforov proved the λ₁-only inequality; Bollobás–Nikiforov conjectured adding λ₂. Prior proofs covered triangle-free, regular, weakly perfect and sparse-triangle graph classes, while Coutinho–Spier–Zhang had weaker constants for general graphs. Giacomelli’s March 2026 arXiv:2603.26379 addressed special cases, not this general weighted theorem.
AI involvement
GPT-6 Astra is credited with mathematical ideation in a collaboration with Coutinho, Liu, Spier, Tang and Zhang. Grok 4.6 agents later formalized the proof under Zhang’s statement-level direction. The available exposition does not document a complete human explanation of the weighted mechanism.
GPT 6 Astra is expressly credited with ideation of the central weighted proof in a human collaboration; Grok later formalized it. More autonomous attribution is unsupported.
Importance
The weighted theorem implies the longstanding general Bollobás–Nikiforov conjecture and strengthens earlier special-class results.
Limitations
The proof does not establish the proposed inequality for up to ω(G) eigenvalues, and the authors do not separate every central lemma by human or model origin.
Technical verification
The repository’s dated record reports five theorem statements matched to Challenge, Lean kernel acceptance, standard axioms and a clean build.
Formal checking
clean lake build, axiom report and statement matching to Challenge
- Checker
- Lean 4 kernel, NanoDa and Comparator
- Scope
- Bollobás–Nikiforov and weighted, matrix, Gram and conic headline theorems
- Basis
- Reported by the linked checker
Checking a formal statement does not establish that it matches the original problem.
Human understanding
The preliminary note was model-generated and a more human-readable author exposition was still in progress; the available record does not document how fully humans can explain the central weighted proof.
Correction and publication history
The weighted theorem implies the longstanding general Bollobás–Nikiforov conjecture and strengthens earlier special-class results.
Read this published revision