The Theta conjecture of Interdonato and Iraci (with the valley Delta conjecture as a special case)
Let be the Bergeron-Garsia nabla operator and the Theta operators of D'Adderio, Iraci and Vanden Wyngaerd, acting on symmetric functions over . Interdonato and Iraci defined a statistic on parking functions (labelled Dyck paths) with decorated rises and decorated contractible valleys, and conjectured that for and Is it true? The case is the rise version of the Delta conjecture of Haglund, Remmel and Wilson, which is a theorem (D'Adderio and Mellit; Blasiak, Haiman, Morse, Pun and Seelinger). The case is the valley version, which was still open. A refinement by touch number replaces by and restricts the sum to paths with .
- Result
- Proved(see note)
- Status
- Candidate (review pending)
- AI contribution
- AI co-developed
- Method
- Argument
- Field
- Algebraic combinatorics; Macdonald polynomials, Delta conjecture
- Posed by
- Giovanni Interdonato and Alessandro Iraci, The Theta conjecture, arXiv:2609.17744 (September 2026)
- Year posed
- 2026
- Years open
- 0y
- Solved
- 2026-11-09
- Model
- ChatGPT 6 Astra (paper); OpenAI Codex (Lean library)
- Vendor
- OpenAI
- Collaborators
- Haojun Bai, Feng Gu, Zhuowei Lin, Dun Qiu, Minhao Zhang
- Verification
- Lean-checked, statement unaudited
- Publication
- Preprint
- Significance
- 30 / 100
- Disclosed cost
- —
- Wikipedia
- No dedicated article
What was actually shown
Theorem 1.4 of arXiv:2611.12451 (Bai, Gu, Lin, Qiu and Zhang, v1 9 November 2026) proves the touching refinement: for and , is the sum of over with . Summing over gives the Theta conjecture (Corollary 1.5(1)). Setting gives the valley Delta conjecture (Corollary 1.5(2)) and, through Iraci and Vanden Wyngaerd, the generalized valley Delta conjecture of Qiu and Wilson (Corollary 1.5(3)). The Lean development checks Theorem 1.4 and the Theta conjecture with the Theta operators on the left side; the Delta-operator forms in Corollary 1.5(2) and (3) are not formalized, and the Lean proof follows a different route from the paper's.
What the AI did
From the paper's section "Use of AI tools": "The theoretical derivations and the proofs in this paper were developed with the help of ChatGPT 6 Astra. The authors studied and checked all of this material, and wrote the paper themselves. We also checked the main identities by computer for small . The authors take full responsibility for the content of the paper." The repository README adds that the Lean proof is the LeanSF library, "developed by Dun Qiu with OpenAI Codex".
Verification
Lean-checked; the formal statement has no independent anchor yet. Read by this site on 19 November 2026: the TeX source of arXiv:2611.12451v1 (the conjectures, Theorem 1.4, Corollary 1.5, Definition 2.4, the AI section) and the conjecture and dinv definition in Interdonato and Iraci's arXiv:2609.17744, which agree. In the repository qiudun123/theta-conjecture at commit 4858863 (its only commit, 9 November 2026) the audit file comparator/Challenge.lean was read in full: it imports only Mathlib, defines the modified Macdonald polynomials by Haiman's characterization, quantifies over all operators with the nabla and Pi eigenvalues, and states theta_touch_identity and theta_total_identity in variables for every ; its four inversion rules, area, touch and dinv corrections match Definition 2.4. A text search of the 856 Lean files found sorry only in that audit file, by design, and no axiom, native_decide, unsafe, opaque or implemented_by. GitHub Actions on that commit report success for the build with axiom audits (9 November) and for the Comparator run (16 November). Nothing was rebuilt here, the shipped numerical script was not run, and the 61-page written proof was not refereed.
Sources
Submitted by MellowGander562 on