AI Science Tracker
EN

Meta FAIR

3 discoveries from 2026

Discoveries

Meta FAIR

A Lean proof constructs a maximally entangled state for nine ten-level parties

Source check: Source checked. Not independent scientific confirmation.

Finding: A formal proof generated with GPT-5.6 Sol constructs a pure state for nine ten-level quantum systems whose reduction to any four is maximally mixed. The exact result concerns existence of an AME(9,10) state.

Technology: GPT-5.6 Sol; h1 Lean proof-search harness

Family GPT-5: profile · models page

Publication: Publication status unspecified in annotation · Test context: prospective or independent check. Catalog categories, not an evidence grade.

AI contribution
The h1 harness used GPT-5.6 Sol to generate a proof script with feedback from the Lean 4 checker.
Test and result
The encoded construction supplies a normalized state and verifies its reductions. The Formal Conjectures maintainer linked the original work, and a later independent paper analyzes the construction.
Limitation / still unknown
The early maintainer replay exceeded its time limit for this proof. Source review did not include replay. The mathematical construction does not demonstrate hardware preparation.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://github.com/facebookresearch/atlas-lean/blob/223d8bc0879b8ddd5b5c935ffe2a3a35f6e0b86d/Atlas/FC/OpenQuantumProblem35_ame_9_10_open_solution.lean#L1416 · https://github.com/facebookresearch/atlas-lean/pull/7 · https://arxiv.org/html/2610.35962v1 · https://github.com/google-deepmind/formal-conjectures/pull/6008 · https://arxiv.org/html/2611.17216v1

The legacy label “Not reported” does not distinguish information absent from reviewed material from information not assessed. It is not a scientific finding of absence.

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: other. Paper: not linked in this catalog; this does not mean no paper exists.

Review date: 21 Nov 2026. Review scope: Pinned Atlas Lean theorem and August 28 merge; September Formal Conjectures proof acceptance; h1 Table 1 and Appendix D; follow-up Appendix B.1.

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 3/5 · Validation 4/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Meta FAIR

A Lean proof establishes the word count behind OEIS A108081

Source check: Source checked. Not independent scientific confirmation.

Finding: GPT-5.6 Sol produced a formal proof counting words generated by a reverse-and-shift concatenation rule. For length n, the count is A108081 at index n−1, a known Fibonacci transform whose connection to the word rule is the result proved.

Technology: GPT-5.6 Sol; h1 Lean proof-search harness

Family GPT-5: profile · models page

Publication: Publication status unspecified in annotation · Test context: prospective or independent check. Catalog categories, not an evidence grade.

AI contribution
The h1 harness used GPT-5.6 Sol to generate a proof script with feedback from the Lean 4 checker.
Test and result
The proof relates the words of each positive length to the sequence using the matching coordinate enumeration. A maintainer reviewer independently replayed the linked version and reported only standard Lean axioms.
Limitation / still unknown
This is a theorem about the specified combinatorial word-generation rule. The existing integer sequence predates the proof.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://github.com/facebookresearch/atlas-lean/blob/223d8bc0879b8ddd5b5c935ffe2a3a35f6e0b86d/Atlas/FC/OeisA108081_count_words_in_x_is_a_shifted.lean#L3797 · https://github.com/facebookresearch/atlas-lean/pull/7 · https://arxiv.org/html/2610.35962v1 · https://github.com/google-deepmind/formal-conjectures/pull/6008 · https://oeis.org/A108081

The legacy label “Not reported” does not distinguish information absent from reviewed material from information not assessed. It is not a scientific finding of absence.

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: other. Paper: not linked in this catalog; this does not mean no paper exists.

Review date: 21 Nov 2026. Review scope: Pinned Atlas Lean theorem and August 28 merge; September Formal Conjectures PR and its independent axiom check; h1 Table 1 and Appendix D.

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 4/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Meta FAIR

A Lean proof establishes the two-sum Ruzsa–Turjányi limit

Source check: Source checked. Not independent scientific confirmation.

Finding: GPT-5.6 Sol proved that for a zero-density asymptotic additive basis, the ratio of the number of two-term sums up to 2N to the number of basis elements up to N tends to infinity. The statement is the two-sum variant of the older Ruzsa–Turjányi result for three-term sums.

Technology: GPT-5.6 Sol; h1 Lean proof-search harness

Family GPT-5: profile · models page

Publication: Publication status unspecified in annotation · Test context: prospective or independent check. Catalog categories, not an evidence grade.

AI contribution
The h1 harness used GPT-5.6 Sol to generate a proof script with feedback from the Lean 4 checker.
Test and result
The formal proof encodes zero density and asymptotic additive-basis coverage. It applies the Plünnecke–Ruzsa inequality to control truncated sumsets.
Limitation / still unknown
This result concerns the density-zero two-sum variant. The originally posed bounded-limsup question in Erdős 337 is different and already has a counterexample.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://github.com/facebookresearch/atlas-lean/blob/223d8bc0879b8ddd5b5c935ffe2a3a35f6e0b86d/Atlas/FC/Erdos337_ruzsa_turjanyi_solution.lean#L184 · https://github.com/facebookresearch/atlas-lean/pull/7 · https://arxiv.org/html/2610.35962v1 · https://github.com/google-deepmind/formal-conjectures/pull/6008 · https://github.com/google-deepmind/formal-conjectures/blob/main/FormalConjectures/ErdosProblems/337.lean · https://github.com/mo271/formal-conjectures/commit/9e4cd5617edacb313fb1da32adb4751f1fa8f603

The legacy label “Not reported” does not distinguish information absent from reviewed material from information not assessed. It is not a scientific finding of absence.

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: other. Paper: not linked in this catalog; this does not mean no paper exists.

Review date: 21 Nov 2026. Review scope: Pinned Atlas Lean theorem and August 28 merge; h1 Table 1 and Appendix D; September separate formal proof and Formal Conjectures statement.

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 4/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open the interactive profile for the timeline, or browse all organizations and model families.