Mathematical Contributions — KentaKitamura

GitHub: KitaKen1

Erdős Problems

#CategoryTopicContribution highlightFirst postStatus when posted
176LeanProofDiscrepancy of arithmetic progressionsLean 4 formalized polynomial upper bounds: N(k,2)=O(k³), N(k,√k)=O(k⁵)2026-06-21OPEN 2026-07-09
265LeanProofNumber TheoryCritical base-two growth for rational reciprocal-pair seriesIndependent Lean proof claim: both rational sums force log(aₙ)/2ⁿ → 0, ruling out limsup aₙ^(1/2ⁿ) > 12026-09-07OPEN 2026-09-07
287ProofComputationUnit fractions summing to 1 — must some denominator gap be ≥ 3?Counterexample-size lower bounds, culminating in k ≥ 686349210761570896312026-05-28OPEN 2026-07-09
301LiteratureLargest subset of {1,…,N} with no unit-fraction relationLiterature note: Wang's 667/806 upper bound; suggestion for the next finite experiment2026-07-04OPEN 2026-07-09
302LeanProofNumber TheoryMaximum size of unit-fraction-triple-free subsetsLean proof report: f(N) ≤ (0.8461739827964010 + o(1))N, improving the recorded 0.9 upper bound; the full estimation remains open2026-09-25OPEN 2026-09-25
341ComputationEventual periodicity of greedy sum-avoiding sequencesComputation: for {1,4,9,16,25} (Dickson rule) the differences have period 224, not thousands of terms2026-07-08OPEN 2026-07-09
346LeanProofRatio limit (1+√5)/2 for boundedly-complete sequencesLean 4 formalization: affirmative answer under the limit-exists interpretation — listed as 🟢 “Full solution (Lean)” in the AI-contributions wiki2026-06-21OPEN 2026-07-09
352ProofComputationTriangles of area 1 in sets of positive measure (Koizumi's discrete model)Witness table for N=1…40, an upper-bound scheme (exact for N≤7), and a sharp low-side theorem2026-05-31OPEN 2026-07-09
383ComputationLargest prime factor of (p²)(p²+1)⋯(p²+k)Explicit witness: p = 9188057 works for all k ≤ 10 (search over p ≤ 10⁷)2026-06-21OPEN 2026-07-09
389ComputationDivisibility between two consecutive blocks of k integersExplicit witnesses extend the verified existence range from n ≤ 27 to n ≤ 332026-08-01OPEN 2026-07-09
404ComputationLeanp-adic valuation of sums of factorials — the function f(a,p)Two landscape repositories: exact f(a,2) for a ≤ 256 (incl. two-sided f(2,2)=254) and odd-prime columns with Lean-checked cells2026-07-07OPEN 2026-07-09
409ComputationIterating n ↦ φ(n)+1 until a primeNew record-depth examples: three n with F(n) = 67 (beyond exhaustive sweep to 1.5·10⁹)2026-06-21OPEN 2026-07-09
425ProofLeanMultiplicative Sidon sets: F(n) = π(n) + (c+o(1))·n^{3/4}/(log n)^{3/2}?Lean-verified improvement of the lower-bound coefficient to 3.499 (from 2.95), plus a sharpness theorem for the layered ansatz2026-06-06OPEN 2026-07-09
503ProofLargest isosceles sets in ℝᵈNote: f(22) = 276, from Musin's tight two-distance bound + Blokhuis's upper bound2026-05-29OPEN 2026-07-09
516LeanProofComplex AnalysisFejér-gap variant of Pólya's minimum-modulus problemLean counterexample: an explicit entire Fejér-gap series has log m(r)/log M(r) ≤ 1/2 for every sufficiently large r2026-09-21PROVED (LEAN) 2026-09-21
522LeanProofComplex AnalysisRoots of random ±1 polynomials in the closed unit diskLean proofs that R_n/(n/2) → 1 almost surely for ±1 and for 0/1 coefficients, covering both readings of the problem2026-09-26OPEN 2026-09-26
536LeanProofSets avoiding three elements with pairwise equal lcmLean 4 formalized upper bound |A| ≤ ⌈5N/6⌉2026-06-22OPEN 2026-07-09
603LeanColouring unions of countable families with |Aᵢ∩Aⱼ| ≠ 2Lean 4 / Mathlib formalization attempts for both interpretations of the problem2026-05-17SOLVED 2026-07-09
612LeanProofGraph TheoryDiameter bounds for connected graphs with a forbidden fixed cliqueSix formerly open fixed-clique cases settled in Lean: three positive bounds and three explicit counterexample families2026-09-07OPEN 2026-09-07
647ProofIs there n > 24 with max_{m<n}(m+τ(m)) ≤ n+2?New necessary condition: any candidate n > 84 must have (n−3)/3 prime2026-05-29OPEN 2026-07-09
768LiteratureDensity of n where every prime p | n has a divisor d ≡ 1 (mod p)Literature note: Eric Li's preprint resolving the problem, with c = 1/(2√log 2)2026-07-09OPEN 2026-07-09
811LeanProofGraph TheoryBalanced rainbow cliques: Axenovich–Clemen conjectureLean proof claim: every K_q (q≥4) has arbitrarily large balanced counterexamples; the general graph classification remains open2026-09-15OPEN 2026-09-15
835ProofColouring k-subsets of {1,…,2k} so every (k+1)-set shows all k+1 coloursSpectral verification: any remaining colouring must give each k-set the same colour as its complement2026-05-30OPEN 2026-07-09
873LeanProofNumber TheoryLCM windows: the supplementary all-scale conjectureLean refutation: no increasing sequence sustains F(A,X,3) ≫ X^(1/3) log X at every scale; main #873 remains open2026-09-11OPEN 2026-09-11
878LeanProofNumber TheoryUnconventional functions f and F: four solved statementsLean proofs of Main Questions 1-2, the density-one F(n) ~ (1/2)n log log n conjecture, and Erdős's formula (17); four other clauses remain open2026-09-13OPEN 2026-09-13
889LeanProofNumber TheoryCapital-V₁ finite-exception conjectureLean proof that only finitely many n satisfy V₁(n)=1; also classifies every exception through n=80 without claiming that 80 is globally largest2026-09-22OPEN 2026-09-22
920LiteratureChromatic number of Kₖ-free graphsStatus observation: #986 (now proved) + the implication already on the page ⇒ #920 should be positive2026-06-30OPEN 2026-07-09
939ComputationSums of coprime r-powerful numbersA new small r = 5 example: 2⁵·17⁶ + 7¹¹ + 2⁵·3⁶·7⁸ = 3⁷·13⁷2026-05-25OPEN 2026-07-09
944LeanProofGraph TheoryDirac's critical-edge conjecture: the remaining k=4, r=1 caseExplicit 48-vertex Lean witness settles k=4; with Jensen's k≥5 theorem, this completes Dirac's r=1 conjecture for every k≥42026-09-10OPEN 2026-09-10
1011LeanProofGraph TheoryTriangle-forcing thresholds at r=4 and r=5Lean: complete f₄(n) for every n, plus f₅(n)=⌊n²/4⌋−3n+15 for every n≥802026-09-10OPEN 2026-09-19
1039LeanLargest disc inside {z : |f(z)| < 1} for monic polynomialsLean 4 / Mathlib formalization of the proof strategy discussed in the thread2026-05-15OPEN 2026-07-09
1040LeanProofTransfinite diameter and minimal areas of polynomial lemniscatesIndependent Lean proof of μ(F)=0 for the finite-Fekete iInf formulation of part (ii); the bridge to classical logarithmic capacity remains to formalize2026-09-06OPEN 2026-09-06
1047LeanProofComplex AnalysisGoodman's maximum number of non-convex componentsLean determination by degree: 0 for n≤1, 1 for n=2, 2 for n=3, and n for every n≥42026-09-18DISPROVED (LEAN) 2026-09-18
1056ComputationConsecutive intervals with product ≡ 1 (mod p)Explicit finite witnesses for every 2 ≤ k ≤ 14 via OEIS A0604272026-06-21OPEN 2026-07-09
1103LeanProofNumber TheoryTao's finite admissible-set conjecture modulo prime squaresComplete Lean classification: A(N)=|SF∩[N]| exactly for positive N in {1,…,8,17}, proving A(N)>|SF∩[N]| for every N≥182026-09-22OPEN 2026-09-22
1134LeanProofNumber TheoryKlarner seed-0 density variantLean proof that the set generated by 0 under 2x, 3x+2, and 6x+3 has natural density 02026-09-20DISPROVED (LEAN) - $24 2026-09-20
1187LeanMonochromatic arithmetic progressions of primes / prime common differenceLean formalization of the standard mod-4 counterexample for the second part2026-05-12SOLVED 2026-07-09
1207LiteraturePoint sets in ℝᵈ with no isosceles triangle — P_d(n)Literature note: Lee–Pohoata–Zhu preprint settles the conjectural part (Erdős 1980)2026-07-08OPEN 2026-07-09

Formal Conjectures

#CategoryTopicContribution highlightFirst postStatus when posted
WOWII #2LeanProofGraph TheoryNeighbourhood independence and leaves in spanning treesSubmitted an independent Lean 4 proof on 2026-07-27; Formal Conjectures records its maximum-edge triangle-free-spanning-subgraph argument as the third formal proof, and PR #4654 was approved and merged on 2026-08-082026-07-27OPEN2026-07-27
WOWII #31LeanProofLiteratureGraph TheoryInduced-path length versus graph radiusFormalized the published Erdős–Saks–Sós / Fan Chung proof of p(G) ≥ 2·rad(G) − 1 in Lean 4 and submitted the proof link in PR #46582026-07-28SOLVED2026-07-28
WOWII #217LeanProofGraph TheoryLeaf number, graph residue, and Hamiltonian pathsSubmitted a complete Lean 4 proof that the WOWII #217 leaf-number and residue bound forces a Hamiltonian path; PR #4656 was reviewer-approved and merged on 2026-08-042026-07-28OPEN2026-07-28
WOWII #316LeanProofGraph TheoryPendant vertices and well total dominationSubmitted a complete Lean 4 proof of WOWII Graph Conjecture 316 to Formal Conjectures on 2026-07-13; the contribution was accepted by a Formal Conjectures reviewer, and PR #4426 was merged on 2026-08-072026-07-13OPEN2026-07-13
Lonely Runner variant
Tao (2017) asymptotic lower bound
LeanStatementLiteratureImproved asymptotic lower bound for the gap of lonelinessAdded the Lean definitions of the gap of loneliness and the formal statement of Tao's 2017 asymptotic lower bound; PR #2177 was approved and merged on 2026-05-062026-02-06SOLVED2026-02-06
OQP #40Quantum InformationLiteratureStatusRefinement of the Bessis–Moussa–Villani conjectureLiterature note: Cha–Lee counterexamples disprove the proposed upper bound; the violation ratio can be arbitrarily large2026-07-13OPEN2026-07-13
MathOverflow #507128
R = Q(R) · proper invertible ideal I ⊊ R
LeanProofCounterexampleExplicit cuspidal-cubic idealization counterexampleConstructed and formally verified an explicit R = Q(R) with a proper invertible ideal; the MathOverflow asker validated the Lean4Web proof, and PR #4644 was reviewer-approved and merged on 2026-08-042026-07-27OPEN2026-07-27
Bézier–Bernstein Voronovskaja formula
α > 0, α ≠ 1 · f ∈ C²[0,1]
LeanProofCentral-limit-scale asymptotics of Bézier-type Bernstein operatorsDetermined and Lean-formalized the exact √n-limit μ(α)√(x(1−x))f′(x) for α > 0, α ≠ 1 and f ∈ C²[0,1], and submitted the complete Lean 4 proof in PR #46462026-07-27OPEN2026-07-27
Convex additive VC₂ dimension
convex C ⊆ ℝ³
proposed upper bound 1
LeanProofCounterexampleSix-halfspace counterexample to the proposed upper boundConstructed and Lean-verified a convex polyhedron in ℝ³, defined by six halfspaces, that disproves the proposed additive VC₂-dimension upper bound 1 in PR #46572026-07-28OPEN2026-07-28
Monochromatic Quantum Graph
even N ≥ 6 · D ≥ 3
ℤ and {-1,0,1}-valued weights
LeanProofQuantum InformationGraph TheoryNonexistence for all even integer-weight casesSubmitted a solver-free Lean 4 proof of nonexistence for every even N ≥ 6 and D ≥ 3 over ℤ, settling six integer and six trinary-integer declarations in PR #46592026-07-29OPEN2026-07-29
Monochromatic Quantum Graph
even N ≥ 6 · D ≤ N − 2
commutative integral domains
LeanProofQuantum InformationGraph TheorySharp upper bound on the number of colorsSubmitted a solver-free Lean 4 proof of the sharp bound D ≤ N − 2 over every commutative integral domain, settling three Formal Conjectures high-color cases in PR #46612026-07-29OPEN2026-07-29
Monochromatic Quantum Graph
N=6, D=4 over ℂ
LeanProofQuantum InformationGraph TheoryNonexistence of a monochromatic quantum graph equation-system solutionSubmitted a complete Lean 4 proof that the N=6, D=4 monochromatic quantum graph equation system has no solution over ℂ; the proof works over every commutative integral domain2026-07-30OPEN2026-07-30
Weak Tiling 4.3LeanProofCounterexampleWeak tiling measures versus convex combinations of proper tilingsFound an explicit counterexample to the statement formalized as Weak Tiling Problem 4.3 and submitted a kernel-checked Lean 4 proof in PR #47042026-08-04OPEN2026-08-04
Erdős #42
constructive Sidon-set variant
LeanProofAdditive CombinatoricsThreshold function for Sidon sets with disjoint difference setsSubmitted a complete Lean 4 proof of the constructive Erdős #42 variant; reviewer mo271 approved it and PR #4853 was merged on 2026-08-10, marking the variant solved2026-08-10OPEN2026-08-10
Moving Sofa uniqueness
unrestricted-set misformalization
Misformalization DetectionGerver-sofa uniqueness stated over every planar setIdentified that the Moving Sofa uniqueness theorem quantified over every planar set; the counterexample was credited in reviewer-approved PR #4852, which corrected the statement to uniqueness of moving sofas up to rigid motion and was merged on 2026-08-102026-08-10OPEN2026-08-10
Green Problem 14
AKS14 lower-bound status correction
LiteratureStatusCorrectionEstablished lower bounds W(3,t) for 20 ≤ t ≤ 39Identified that 20 AKS14 lower-bound declarations for 20 ≤ t ≤ 39 were established by finite certificates rather than open conjectures, and proposed marking them research solved2026-08-10OPEN2026-08-10
Erdős #973
negative-answer status correction
LiteratureStatusCorrectionExterior power sums and nonexistence of an exponential constantIdentified Luo–Yang–Zhu's negative solution of Erdős #973, documented the exact asymptotic obstruction, and proposed changing the Formal Conjectures entry to research solved with answer(False)2026-08-10OPEN2026-08-10
Square Packing
negative-radius least-element flaw
Misformalization DetectionNonexistence of least feasible radii when radii range over all realsShowed that allowing r ∈ ℝ makes two circle-container least-radius statements false: every feasible radius has a smaller feasible radius; proposed restricting r ≥ 0 or using ℝ≥02026-08-10OPEN2026-08-10
Erdős #80
infeasible c=2 misformalization
Misformalization DetectionBook-size asymptotics quantified beyond the feasible edge-density rangeDetected that both Erdős #80 declarations quantify over every c>0: at c=2 no simple n-vertex graph can have 2n² edges, so f(2,n)=0 and both formal statements are false2026-08-11OPEN2026-08-11
Fernandes Conjecture 1
2-generation of Γ(m ⊕ n)
LeanProofGroup TheoryTwo-generation of equal-sign permutation-pair subgroupsSubmitted a complete Lean 4 proof that Γ(m ⊕ n) is 2-generated outside the four stated exception pairs, using explicit generators, Goursat kernels, and alternating-group containment2026-08-11OPEN2026-08-11
Erdős #608
known counterexample status report
LiteratureStatusGraph TheoryEdges contained in 5-cycles above the Mantel thresholdReported that Erdős #608 was already disproved by the Füredi–Maleki construction documented in 2016, and linked primateria’s complete external Lean 4 disproof2026-08-11SOLVED2026-08-11
Erdős #539
three exponent variants
LeanProofAdditive CombinatoricsGrowth exponent of the cofactor-set thresholdProved that the zero-inclusive Formal Conjectures threshold has logarithmic exponent 1/2, refuting both proposed n^(2/3) growth variants; reviewer-approved and merged into Formal Conjectures on 2026-08-152026-08-11OPEN2026-08-11
MathOverflow #10799
Kahn–Kalai Conjecture 7
LeanProofCounterexampleCritical-probability optimality of monotone Boolean familiesIndependently formalized the Diskin–Kreitner scale-dense dual-tribes counterexample, disproving the fixed-1000 Formal Conjectures variant and the original Conjecture 7 form; reviewer-approved and merged into Formal Conjectures on 2026-08-152026-08-11OPEN2026-08-11
Microscopic Weighting
ten-point metric counterexample
LeanProofCounterexampleMetric GeometryMicroscopic weighting versus finite concentrationConstructed and Lean-verified a ten-point metric space with finite concentration but no microscopic weighting, disproving Roff–Willerton Conjecture 3.32026-08-11OPEN2026-08-11
Green Problem 52
logarithmic variant
LeanProofLiteratureCounterexampleAdditive CombinatoricsAffine subspaces in double sumsets over Boolean cubesFormalized the Hamming-ball counterexample credited to Kaave Hosseini and Ryan Alweiss in Lean 4; reviewer-approved PR #4887 was merged, closing FC's O(log K) variant while the qualitative O_K(1) question remains open2026-08-12OPEN2026-08-12
Erdős #692 Part II
maximum-existence variant
LeanProofNumber TheoryDensity of integers with exactly one divisor in an intervalProved that δ₁(n,m) attains a maximum over m > n+1 for every fixed n, by reducing the search to an explicit finite interval2026-08-12OPEN2026-08-12
Erdős #319
underformalized Big-O target
Misformalization DetectionLeanProofNumber TheoryThe “simplest upper bound” reduced to an arbitrary upper boundDetected that the FC target does not encode “simplest” or optimal: the ambient-set bound c(N)≤N closes it without using the signed reciprocal-sum condition2026-08-12OPEN2026-08-12
Erdős #357
√n lower Big-O variant
LeanProofNumber TheoryDistinct consecutive interval sums in increasing integer sequencesConstructed an admissible sequence of length ⌊√n⌋ and proved its consecutive interval sums distinct, yielding the explicit FC answer √n = O(f(n))2026-08-13OPEN2026-08-13
Erdős #357
three additional lower-growth targets
LeanProofNumber TheoryLower growth and strict-versus-weak monotonicity for distinct interval sumsProved log n = o(f(n)), √n = O(h(n)), and log n = o(h(n)) by combining the strict √n lower bound with a formal proof that the weak and strict extremal functions agree2026-08-15OPEN2026-08-15
Erdős #688
constant-answer upper-bound flaw
Misformalization DetectionLeanProofNumber TheoryAsymptotic decay of the extremal covering exponentDetected that the FC upper-bound target admits g(n)=1: a Lean witness proves 0≤ε_n≤1 without estimating decay, so filling the answer hole would not solve Erdős #6882026-08-13OPEN2026-08-13
Erdős #142
trivial linear upper-bound flaw
Misformalization DetectionLeanProofNumber TheoryUpper estimates for progression-free subsets of {1,…,N}Detected that the FC upper-bound target admits g(N)=N: the ambient cardinality bound r_k(N)≤N closes it without proving a nontrivial estimate or asymptotic formula2026-08-13OPEN2026-08-13
OEIS A237271
square and hexagonal parity
LeanProofLiteratureNumber TheoryParity of the number of parts in symmetric divisor-sum representationsProved both square and hexagonal parity targets with one complementary-divisor involution, formalizing the existing OEIS middle-divisor parity characterization in Lean 4; mo271 approved the PR on its first review2026-08-13OPEN2026-08-13
OEIS A287616
known-solution status
LiteratureStatusCorrectionNumber TheoryUniversal sum of triangular, pentagonal, and heptagonal numbersReported that Cao–Guo–Qiu–Feng–Gao prove the exact A287616 representation theorem, and proposed changing the Formal Conjectures declaration to research solved2026-08-13OPEN2026-08-13
Green Problem 3
known-solution status
LiteratureStatusCorrectionAdditive CombinatoricsProduct-free open subsets of the unit intervalReported Franchi–Gowers–Yip's affirmative solution of Green Problem 3 and proposed changing the Formal Conjectures answer to True with research-solved status2026-08-13OPEN2026-08-13
Green Problem 31
two upper-bound status corrections
LiteratureStatusCorrectionAdditive CombinatoricsImproved upper bound for finite Sidon setsReported Hou–Zhao's bound F(N)≤√N+0.9435N^(1/4)+O(1), which improves 0.98183 and settles both the infinitely-often and eventual Formal Conjectures upper-bound targets2026-08-13OPEN2026-08-13
Open Quantum Problem 35
AME(7,6) and AME(7,10)
LiteratureStatusCorrectionQuantum InformationExistence of seven-party absolutely maximally entangled statesReported Shi–Zhang–Zhao–Li's existence theorem for every AME(7,d) with d≥3, giving answer(True) for the open d=6 and d=10 benchmark declarations2026-08-13OPEN2026-08-13
Open Quantum Problem 35
AME(12,5) known-solution status
LiteratureStatusCorrectionQuantum InformationExistence of an absolutely maximally entangled state on twelve ququintsReported Bevins–Bidav's explicit AME(12,5) construction and proposed setting the Formal Conjectures benchmark to answer(True) with research-solved status2026-08-13OPEN2026-08-13
Independent Domination
even and odd bound status
LiteratureStatusCorrectionGraph TheoryIndependent domination number for bounded-degree graphsReported that Cho–Kim–Kim–Oum's Corollary 1.3 proves both displayed independent-domination bounds and proposed marking the even and odd Formal Conjectures declarations solved2026-08-13OPEN2026-08-13
MathOverflow #31809
counterexample status
LiteratureStatusCorrectionCategory TheoryPre-triangulated categories that are not triangulatedReported Chen–Liu–Lu–Zhang's explicit pre-triangulated non-triangulated category, giving answer(False) to the Formal Conjectures version of MathOverflow #318092026-08-13OPEN2026-08-13
Green Problem 19
internally implied bound status
LiteratureStatusCorrectionAdditive CombinatoricsCorner-density exponent lower and upper boundsDetected that the solved theorem C=4 already in the same file implies both open bounds C≥3.13 and C≤4, and proposed marking the redundant declarations solved2026-08-13OPEN2026-08-13
Erdős #272
main-asymptotic status
LiteratureStatusCorrectionAdditive CombinatoricsMaximum size of families with arithmetic-progression intersectionsDetected that the recorded Szabó estimate N²/2+O(N^(5/3)log³N) already yields the open main asymptotic N²/2, and proposed marking the principal declaration solved2026-08-13OPEN2026-08-13
Green Problem 37
sublinear upper bound m(N,k)=o(N)
LeanProofAdditive CombinatoricsSparse sets containing a k-term progression of every difference up to NConstructed a Fermat-number CRT periodic cover and proved in Lean 4 that m(N,k)=o(N) for every fixed k; the Big-O target follows from the same explicit answer N↦N2026-08-14OPEN2026-08-14
Erdős #361
asymptotic growth Θ(n)
LeanProofAdditive CombinatoricsNumber TheoryLargest subset of [1,⌊cn⌋] avoiding n as a subset sumProved in Lean 4 that the maximum subset-sum-avoiding cardinality has order Θ(n) for every fixed c>0, using an explicit interval-block lower-bound construction2026-08-14OPEN2026-08-14
Poisson n-Lie Conjecture 3.5
scalar-matrix determinant bracket
LeanProofNonassociative AlgebraStatementDeterminant brackets from scalar matrices and commuting derivationsProved the exact general target proposed in PR #4893, first establishing the same Poisson n-Lie identities for arbitrary natural n and m without the registered lower bounds2026-08-14OPEN2026-08-14
OEIS A113019
third fixed point 9⁹
LeanProofCounterexampleNumber TheoryFixed points of the digit-length–digital-root power mapFound and Lean-verified the third fixed point 387420489 = 9⁹, disproving the proposed classification of 1 and 32 as the only fixed points; OEIS approved and published the result2026-08-15OPEN2026-08-15
OEIS A100478
eventual periodicity
LeanProofNumber TheoryPrime-counting Pentanacci recurrence from arbitrary initial dataProved that every orbit of the prime-counting Pentanacci recurrence from arbitrary nonnegative initial data is bounded and therefore eventually periodic2026-08-15OPEN2026-08-15
OEIS A108306
general INVERT–matrix identity
LeanProofNumber TheoryINVERT transforms and powers of a 2×2 matrixProved the full OEIS A108306 conjecture for arbitrary natural parameters by realizing the INVERT convolution as a two-state matrix recurrence2026-08-15OPEN2026-08-15
OEIS A105801
eventual constancy modulo 3ᵏ
LeanProofNumber Theory3-adic stabilization of the Fibonacci–Collatz sequenceProved that the Fibonacci–Collatz sequence is eventually constant modulo 3ᵏ for every positive k, resolving the full registered OEIS conjecture2026-08-15OPEN2026-08-15
OEIS A112970
three dyadic-ray identities
LeanProofNumber TheoryDyadic rays in a generalized Stern sequenceProved all three OPEN A112970 targets from the stronger dyadic-ray identity a(c·2ⁿ−1)=a(c−1), using only the sequence's odd recurrence2026-08-15OPEN2026-08-15
OEIS A113250
odd-indexed terms are squares
LeanProofNumber TheorySquare terms in the m=4 specialization of a fourth-order recurrence familyProved every odd-indexed term of A113250 is a square by specializing the parameterized identity A₂ₙ₊₁=Yₙ² at m=42026-08-15OPEN2026-08-15
OEIS A113252
odd-indexed terms are squares
LeanProofNumber TheorySquare terms in the m=6 specialization of a fourth-order recurrence familyProved every odd-indexed term of A113252 is a square by specializing the parameterized identity A₂ₙ₊₁=Yₙ² at m=62026-08-15OPEN2026-08-15
OEIS A113255
odd-indexed terms are squares
LeanProofNumber TheorySquare terms in the m=9 specialization of a fourth-order recurrence familyProved every odd-indexed term of A113255 is a square by specializing the parameterized identity A₂ₙ₊₁=Yₙ² at m=92026-08-15OPEN2026-08-15
OEIS A103425
prime-free weighted Tribonacci witness
LeanProofNumber TheoryPrime-free third-order linear recurrencesAnswered the exact OPEN target with the relatively prime coefficients (1,1,−1) and the constant prime-free sequence xₙ=42026-08-15OPEN2026-08-15
OEIS A114831
asymptotic ratio √3
LeanProofOEISNumber TheoryAsymptotics of a harmonic-mean recurrenceProved a(n+1)/a(n) → √3 by expressing the ratio recurrence as a vanishingly perturbed contraction with fixed point √32026-08-15OPEN2026-08-15
OEIS A102371
A105033 complement identity
LeanProofOEISNumber TheorySloping binary numbers, bitwise carries, and XOR recurrenceProved a(n)=2ⁿ−1−A105033(n−1) for every n≥1 through a bitwise carry recurrence and a fixed-width XOR complement identity2026-08-15OPEN2026-08-15
OEIS A102722
a(n) ~ (1−γ)n
LeanProofOEISNumber TheoryFractional-part sums and the Dirichlet divisor problemProved a(n) ~ (1−γ)n by formalizing Dirichlet's hyperbola method, controlling floor errors, and deriving the normalized limit2026-08-15OPEN2026-08-15
OEIS A112521
recursive-array main diagonal
LeanProofOEISCombinatoricsNOR bracketings, recursive arrays, and WZ telescopingProved a(n)=T(n,n) for all n≥1 using transformed Fibonacci polynomials, a WZ recurrence, and positivity of the resulting signed binomial sum2026-08-15OPEN2026-08-15
OEIS A211417
four divisibility targets
LeanProofOEISNumber TheoryFactorial ratios, p-adic valuations, and Landau step functionsProved three atomic divisibility conjectures for the factorial ratio a(n), then derived the fourth product divisibility target by controlling pairwise gcds2026-08-16OPEN2026-08-16
Lₚ Rogers–Shephard
Conjecture 5 equality rigidity
LeanProofConvex GeometryEquality rigidity for planar centrally symmetric convex bodiesProved that equality in the planar Lₚ Rogers–Shephard bound forces the convex body to be a parallelogram with a vertex at the origin2026-08-16OPEN2026-08-16
OEIS A105751
2-adic valuation asymptotic
LeanProofOEISNumber TheoryGaussian-integer products and exact 2-adic valuationsProved ν₂(a(n)) ~ n/4 via exact formulas in all four residue classes, obtained from a dyadic block induction for the Gaussian-integer product2026-08-16OPEN2026-08-16
OEIS A100474
first semiprime after a(11)
LeanProofComputationOEISNumber TheoryCertified recurrence evaluation, minimality, and large-prime proofDetermined a(36) as the first semiprime after a(11), excluded every index 12–35, and certified its 131-digit prime cofactor in Lean2026-08-16OPEN2026-08-16
Erdős #979
k=3 prime-cube representations
LeanProofNumber TheoryUnbounded representation counts for sums of three prime cubesIndependently proved that sums of three prime cubes have unbounded representation multiplicity, using the CM theory of the Fermat cubic and Hecke-coefficient asymptotics2026-08-17SOLVED2026-08-17
Convex VCₙ bound
false n=0 case
Misformalization DetectionLeanProofCounterexampleConvex GeometryFinite additive VCₙ dimension of convex sets in ℝⁿ⁺¹Showed in Lean that the universal convex VCₙ-bound declaration is false at n=0: translates of the singleton {0}⊆ℝ realize every pattern, defeating every finite d2026-08-18OPEN2026-08-18
OEIS A129365
four valuation conjectures
LeanProofOEISNumber TheoryGCD-product ratios and an exact p-adic valuation formulaProved all four conjectures from one exact valuation identity, establishing integrality, prime support, block invariance, and the A004125 valuation sum2026-08-18OPEN2026-08-18
OEIS A003625
quadratic irreducibility over GF(p)
LeanProofLiteratureOEISNumber TheoryQuadratic reciprocity and irreducibility over GF(p)Formalized the classical criterion p mod 7∈{3,5,6} iff x²+x+2 is irreducible over GF(p), using completion of the square and quadratic reciprocity2026-08-19OPEN2026-08-19
Erdős #394
reversed lower-bound direction
Misformalization DetectionNumber TheoryGrowth of the Erdős–Hall divisor-product function t₂(n)Detected that the two sides of ≫ are swapped: the declaration labeled as the known lower bound instead states an upper bound strong enough to collapse the open Hall conjecture2026-08-19SOLVED2026-08-19
OEIS A078590
integrality counterexample at n=7
LeanProofCounterexampleOEISNumber TheoryIntegrality of a nonlinear exponential recurrenceDisproved integrality at the first failing step n=7: the numerator is 9 modulo 171, so the resulting rational term has reduced denominator 192026-08-21OPEN2026-08-21
OEIS A070823
squarefree-cofactor counterexample at n=20
LeanProofCounterexampleOEISNumber TheoryDecimal-concatenation recurrence and squarefree cofactorsProved the divisibility-by-3 half for every n>2, but disproved the squarefree-cofactor claim at n=20 by exhibiting the unavoidable square factor 13²2026-08-21OPEN2026-08-21
OEIS A159829 Conjecture 1
odd-exponent prime-value obstruction
LeanProofCounterexampleOEISNumber TheoryPrime values of sums of two like powersDisproved the all-k infinitude claim at k=3 and proved the stronger classification that for every odd k≥3 the only prime representable as nᵏ+mᵏ is 22026-08-21OPEN2026-08-21
OEIS A185895 Conjecture 3
prime-power Gauss congruences
LeanProofOEISCombinatoricsNumber TheoryGauss congruences for signed distinct-block partition countsProved the full prime-power Gauss congruences by rewriting the coefficients as signed multinomial sums and reducing them coefficientwise modulo pᵏ2026-08-21OPEN2026-08-21
OEIS A022030
alternating nonlinear recurrence
LeanProofOEISNumber TheoryLinear recurrence forced by an alternating nonlinear rounding processProved the recurrence for the alternating sequence 4,16,63,249,… by identifying a linear candidate and bounding its alternating determinant defect2026-08-21OPEN2026-08-21
OEIS A049473
ζ(3) tail and Beatty difference classification
LeanProofOEISNumber Theoryζ(3) approximation and complementary Beatty sequencesProved both the ζ(3) tail bracketing and the exact A001954/A001953 classification of the zero and one steps of the nearest-integer sequence2026-08-21OPEN2026-08-21
OEIS A076141
binary sub-pattern uniqueness in n²
LeanProofOEISCombinatoricsNumber TheoryBinary words occurring in the square n²Proved that the binary word of n occurs at most once in the binary word of n², including overlapping occurrences2026-08-21OPEN2026-08-21
Erdős #367
false ∀ε higher-full-parts variant
Misformalization DetectionLeanProofCounterexampleNumber TheoryLimsup growth of products of r-full partsDisproved the universal-ε variant at (r,k,ε)=(3,2,2): divisibility Bᵣ(m)∣m bounds the normalized product by 2, so its limsup is finite2026-08-22OPEN2026-08-22
Melnikov valency-variety
37-vertex counterexample
LeanProofCounterexampleGraph TheoryChromatic number versus the number of distinct vertex degreesConstructed a 37-vertex tripartite graph with w(G)=30 and χ(G)=3; the proposed lower bound is also 3, so the required strict inequality fails by equality2026-08-22OPEN2026-08-22
Green Problem 29
approximate-group subset bound is false
LeanProofLiteratureCounterexampleAdditive CombinatoricsGroup TheoryLarge subsets of approximate groups with controlled product setsAdapted the certified public slab counterexample to FC: a 3-approximate group in Multiplicative ℤ×H forces every S⊆A with S⁸⊆A⁴ to have cardinality at most 12026-08-22OPEN2026-08-22
Erdős #539
two square-root variants are false
LeanProofCounterexampleAdditive CombinatoricsNumber TheoryGrowth of the cofactor-set threshold beyond every constant multiple of √nProved in Lean 4 that h(n)/√n tends to infinity, so h(n) is neither O(√n) nor Θ(√n); the exact growth of h remains open2026-09-05OPEN2026-09-05
Erdős #354(ii)
common base γ=√φ
LeanProofAdditive CombinatoricsNumber TheoryCompleteness of floor-power sequences for a common exponential baseProved that γ=√((1+√5)/2) works for every α>0; the α-sequence alone is complete, so β and the irrational-ratio assumption are unnecessary2026-09-05OPEN2026-09-05
Erdős #879(ii)
optimal-set factor bound
LeanProofNumber TheoryPrime-factor structure of maximum-weight pairwise-coprime subsetsProved the stronger universal bound ω(a)≤2 for all members of every sufficiently large optimal admissible set, so the proposed unbounded-prime-factor claim already fails at k=32026-09-06OPEN2026-09-06
Bugeaud Problem 10.5
lacunary fractional parts
LeanProofNumber TheoryUniversal lacunary sequences over real number fieldsBuilt a single positive lacunary integer sequence whose fractional parts have a cluster point in every ε-interval for every ξ∉K, proving both registered targets2026-09-20OPEN2026-09-20
Convex additive VCₙ
finite bound for n≥1
LeanProofConvex GeometryFinite additive VCₙ dimension of convex sets in ℝⁿ⁺¹Proved VCₙ(C)≤2^(8(n+2)^n)−1 for every n≥1 and convex C⊆ℝⁿ⁺¹; the sharper bounds 2 and 3 remain open2026-09-20OPEN2026-09-20
Erdős #1095
upper-conjecture status
LiteratureStatusCorrectionNumber TheoryUpper bound for the Ecklund–Erdős–Selfridge function g(k)Reported that Ecklund–Erdős–Selfridge (1974) already prove g(k) < exp(k(1+o(1))), the coefficient-1 exponential bound, and proposed marking the declaration solved2026-09-21OPEN2026-09-21
Erdős #160
better upper and lower bounds
LiteratureStatusCorrectionAdditive CombinatoricsColourings in which every four-term progression receives three coloursReported Shi–Dong's improved upper bound and the stronger lower bound implied by Bloom–Sisask, and proposed marking both better-bound declarations solved2026-09-21OPEN2026-09-21
TxGraffiti Conjecture 2
counterexample status
LiteratureStatusCorrectionGraph TheoryZero forcing versus independence in subcubic graphsReported Fischer's connected 24-vertex counterexample of maximum degree 3, which falsifies Z(G) ≤ α(G)+1 and gives answer False2026-09-21OPEN2026-09-21
Erdős #968
increasing and decreasing triples
LiteratureStatusCorrectionAnalytic Number TheoryMonotone triples in the normalized primes pₙ/nReported that Ford–Maynard–Tao's chains of large prime gaps and Maynard's bounded prime clusters give infinitely many increasing and decreasing triples, answering both True2026-09-21OPEN2026-09-21
Cyclic-subgroup solvability
Conjecture 5.5 status
LiteratureStatusCorrectionGroup TheorySolvability from the number of cyclic subgroupsReported that Theorem 7.1 of the revised Das–Dey–Galindo–Sharma paper proves the cyclic-subgroup solvability threshold cyc(G) < 2^(t+2), and proposed marking it solved2026-09-21OPEN2026-09-21
Bugeaud 10.4
uncountable-spectrum counterexample
LiteratureStatusCorrectionNumber TheorySpectrum of the geometric sequence ξαⁿ modulo oneReported Özcan's result that the spectrum of (ξαⁿ) can be uncountable, a stronger counterexample for every α > 1, giving answer False2026-09-21OPEN2026-09-21
Erdős #1099
factorial and LCM variants
LiteratureStatusCorrectionNumber TheoryBounded divisor-ratio sums along n! and lcm(1,…,n)Reported that Theorem 1 of Tenenbaum (1987) bounds h_α(n!) and h_α(lcm(1,…,n)) for every α > 1, and proposed marking both variants solved2026-09-21OPEN2026-09-21
Erdős #331
Ruzsa variant status
LiteratureStatusCorrectionAdditive CombinatoricsCommon differences of sets with square-root asymptotic densityReported that Theorem 4 of Erdős–Freud (1984) proves Ruzsa's asymptotic-density variant, and proposed marking the declaration solved2026-09-21OPEN2026-09-21
Erdős #906
known-solution status
LiteratureStatusCorrectionComplex AnalysisDense zeros of high derivatives of an entire functionReported that Hou's preprint Cofinite Zeros of High Derivatives proves the existence statement of Erdős 906, and proposed marking it solved2026-09-21OPEN2026-09-21
Kourovka 19.25
counterexample status
LiteratureStatusCorrectionGroup TheorySimplicity from group order and totient sums of element ordersReported that the order-6048 counterexample of van Doorn–Judin–Monticone–Morrison applies to the current Kourovka 19.25 declaration, giving answer False2026-09-21OPEN2026-09-21
Kourovka 20.76
counterexample status
LiteratureStatusCorrectionGroup TheoryAbelian subgroups of p-groups with small abelian normal subgroupsReported Muliarchyk's Proposition 6.1, a 2-group of order 2¹⁴⁵ whose abelian normal subgroups have order at most 2⁴² but which has an abelian subgroup of order 2⁸⁵, giving answer False2026-09-21OPEN2026-09-21
Green Problem 44
counterexample status
LiteratureStatusCorrectionAnalytic Number TheorySieving half the residue classes modulo 1000 primesReported that Green Problem 44 is the same question as Erdős 1202, whose negative resolution gives a counterexample and answer False2026-09-21OPEN2026-09-21
Tu–Deng conjecture
known-solution status
LiteratureStatusCorrectionCombinatoricsBinary weights of residue pairs modulo 2ᵏ − 1Reported Liu–Luo–Xie's Theorem 1.1 and Cusick's preprint as proofs of the exact Tu–Deng statement, and proposed marking the declaration solved2026-09-21OPEN2026-09-21
Bugeaud 10.6
two existence variants
LiteratureStatusCorrectionNumber TheoryRapidly increasing integer sequences dense modulo oneReported that the Ajtai–Havas–Komlós existence theorem, as recorded by Gao–Mooroogen–Yip, yields both Bugeaud 10.6 variants after integer rounding2026-09-21OPEN2026-09-21
Erdős #700(b)
square-root exceedance status
LiteratureStatusCorrectionNumber TheoryMinimal gcd of n with its binomial coefficientsReported a public July 24 solution combining a three-prime lemma with Maynard's theorem to give infinitely many composite n with f(n) > √n, answering part (b) True2026-09-21OPEN2026-09-21
Erdős #872
prime-question status
LiteratureStatusCorrectionCombinatoricsSmallest maximal primitive subsets of {2,…,n}Reported the May 20 forum proof, supported by a 2023 paper, that every maximal primitive subset of {2,…,n} has at least π(n) elements, answering the prime question True2026-09-21OPEN2026-09-21
Erdős #269
rationality-variant counterexample
LiteratureStatusCorrectionNumber TheoryReciprocal sums of partial LCMs of P-smooth numbersReported Steve Fan's forum proof that the sum is transcendental for two primes, a counterexample to universal rationality giving answer False2026-09-21OPEN2026-09-21
Erdős #1063
better-upper-bound status
LiteratureStatusCorrectionNumber TheoryLeast n for which all but one of n−i divide C(n,k)Reported a public July 25 proof of nₖ ≤ exp(o(k)), improving Cambie's k·lcm(1,…,k−1) bound, and proposed marking the declaration solved2026-09-21OPEN2026-09-21
Erdős #289
known-solution status
LiteratureStatusCorrectionNumber TheoryUnit fractions from separated intervals of consecutive integersReported the public Cambie–Land–Tang proof and Lean formalization of the separated-interval statement, and proposed marking Erdős 289 solved2026-09-21OPEN2026-09-21
Erdős #105
n−4 variant counterexample
LiteratureStatusCorrectionDiscrete GeometryLines through two points of A that avoid a disjoint set BReported a July 24 forum counterexample with ten points in A and six in B, refuting the n−4 variant and giving answer False2026-09-21OPEN2026-09-21
Erdős #263(ii)
corrected-statement status
LiteratureStatusCorrectionNumber TheoryGrowth of strictly increasing irrationality sequencesReported Liam Price's May 9 forum construction as settling the corrected, strictly increasing version of Erdős 263(ii), and proposed marking it solved2026-09-21OPEN2026-09-21
Erdős #270
linear-variant status
LiteratureStatusCorrectionNumber TheoryIrrationality of the f(n) = n seriesReported Kovač's July 12 forum proof that Σ n!/(2n)! is irrational, settling the f(n)=n case of Erdős 270, and proposed marking the linear variant solved2026-09-21OPEN2026-09-21
Green Problem 47
exact-containment counterexample
LeanProofCounterexampleAnalytic Number TheoryInverse large sieve: sparse-or-quadratic dichotomyConstructed and Lean-verified a locally sparse set that is too large for the proposed bound yet lies in no single rational quadratic image2026-09-21OPEN2026-09-21
Open Quantum Problem 35
AME(11,10) existence
LeanProofQuantum InformationAbsolutely maximally entangled state with eleven ten-dimensional partiesConstructed an AME(11,10) state by coupling certified binary and quinary graph-state branches, proving the exact open benchmark affirmatively2026-09-22OPEN2026-09-22
Green Problem 25
improved upper bound
LeanProofAdditive CombinatoricsPartition colour count versus unions of restricted sumsetsProved that Green Problem 25 already fails at k(N)=⌈N^(23/40)⌉, improving the previous N/log N upper-bound scale2026-09-23OPEN2026-09-23
Open Quantum Problem 35
AME(9,6) existence
LeanProofQuantum InformationAbsolutely maximally entangled state with nine six-dimensional partiesConstructed and Lean-verified an AME(9,6) state, proving that all 126 four-party reduced density matrices equal I/6⁴2026-09-23OPEN2026-09-23
Square Packing
three squares in a circle
LeanProofMetric GeometryPacking three freely placed unit squares in a circleProved in Lean that 5√17/16 is the least circle radius for three unit squares under arbitrary placements2026-09-24OPEN2026-09-24
Weak Tiling 4.1
bounded support density
LeanProofWeak tiling measures and density of their supportsProved that the support of every weak tiling measure of a finite union of intervals has uniformly bounded point counts in unit intervals2026-09-24OPEN2026-09-24
Erdős #302
0.84617 upper bound
LeanProofNumber TheoryUnit-fraction triple-free setsSubmitted a Lean proof of $f(N)\le(0.8461739827964010+o(1))N$, improving the recorded $0.9N$ upper bound; the exact asymptotic remains open2026-09-25OPEN2026-09-25
Green Problem 41
double-exponential bound
LeanProofDynamical SystemsDouble-exponential pyjama-covering boundSubmitted a Lean proof of the double-exponential pyjama-covering bound, improving the known triple-exponential bound; the exact and polynomial cases remain open2026-09-25OPEN2026-09-25
KLM Problem 7.2
three-dimensional convex bodies
LeanProofHarmonic AnalysisConvex GeometrySpectral products with three-dimensional convex bodiesProved in Lean that if A ⊂ ℝ³ is a convex body and A × B is spectral for a bounded measurable B, then B is spectral; dimensions n ≥ 4 remain open2026-09-25OPEN2026-09-25
Erdős #143(i)
lower-density status
LiteratureStatusCorrectionNumber TheoryDensity of well-separated sets of realsReported that Koukoulopoulos–Lamzouri–Lichtman force well-separated sets to have upper logarithmic density 0, hence lower density 0, answering part (i) True2026-09-25OPEN2026-09-25
Erdős #348
deletion-threshold answer
LiteratureStatusCorrectionNumber TheoryComplete sequences robust to deleting m but not n termsReported Geneson's theorem that such sequences exist exactly when m ≤ 1, giving the answer set {(m, n) : m < n, m ≤ 1}2026-09-25OPEN2026-09-25
Erdős #413(ii)
ω-barrier status
LiteratureStatusCorrectionNumber TheoryInfinitely many barriers for ε·ω(n)Reported that Lau's Theorem 1.3, giving ω(n − k) ≤ C log k for all 1 < k < n infinitely often, makes n − 1 a (1/C)·ω barrier, answering part (ii) True2026-09-25OPEN2026-09-25
Erdős #817
negative-answer status
LiteratureStatusCorrectionAdditive CombinatoricsSubset sums free of three-term progressionsReported Costa's theorem liminf g₃(n)/3ⁿ = 0, which rules out 3ⁿ = O(g₃(n)) and gives answer False2026-09-25OPEN2026-09-25
Erdős #1046
lemniscate-diameter status
LiteratureStatusCorrectionComplex AnalysisDiameter of connected polynomial lemniscatesReported that monic lemniscates have logarithmic capacity 1 and that cap(K) ≤ diam(K)/2, so their diameter is at least 2, answering the variant True2026-09-25OPEN2026-09-25
Erdős #1084
planar triangular-lattice case
LiteratureStatusCorrectionDiscrete GeometryPlanar contact numbers of triangular-lattice point setsReported Harborth's formula f₂(N) = ⌊3N − √(12N − 3)⌋, which at N = 3n² + 3n + 1 gives 9n² + 3n, and proposed marking the variant solved2026-09-25OPEN2026-09-25
Erdős #1106(i)
partition-number prime factors
LiteratureStatusCorrectionNumber TheoryDistinct prime factors of products of partition numbersReported Schinzel–Wirsing's bound F(n) ≫ log n and Ono's theorem that every prime divides some p(n), answering part (i) True2026-09-25OPEN2026-09-25
Erdős #1133
robust-interpolation status
LiteratureStatusCorrectionApproximation TheoryRobust polynomial interpolation on [−1, 1]Reported that Yang's Theorem 2.1 with ρ = 0 contains the Erdős 1133 statement, repeated nodes included, answering it True2026-09-25OPEN2026-09-25
Erdős #1167
finite-target case
LiteratureStatusCorrectionSet TheoryNegative stepping-up for partition relations with finite targetsReported that Erdős–Hajnal–Máté–Rado's Combinatorial Set Theory proves the all-finite-targets case of the negative stepping-up lemma, and proposed marking it solved2026-09-25OPEN2026-09-25
Erdős #1207
isosceles-free subset status
LiteratureStatusCorrectionDiscrete GeometryLarge isosceles-free subsets of planar point setsReported that Lee–Pohoata–Zhu's point sets, in which every subset repeats some distance many times, force isosceles-free subsets of size ≲ n^(1−δ), answering True2026-09-25OPEN2026-09-25
Green Problem 1
sum-free subset status
LiteratureStatusCorrectionAdditive CombinatoricsSum-free subsets beyond n/3Reported Bedert's theorem that every n-element set has a sum-free subset of size n/3 + c log log n, answering Green Problem 1 True2026-09-25OPEN2026-09-25
Green Problem 14
W(3,29) and W(3,30) values
LiteratureStatusCorrectionCombinatoricsTwo-colour van der Waerden numbers W(3, r)Reported Chowdhury–Codel–Heule's SAT bounds W(3,29) ≥ 869 and W(3,30) ≥ 904, which refute the conjectured values 868 and 903 (answer False for both)2026-09-25OPEN2026-09-25
Green Problem 27
improved lower bound
LiteratureStatusCorrectionAdditive CombinatoricsSmallest sets in ℤ/pℤ with no unique sumsReported Cao–Yuan's bound m(p) ≫ log p · log log p, which beats Bedert's bound and supplies the answer function for the lower-bound declaration2026-09-25OPEN2026-09-25
Green Problem 35
lower bound at p = 2
LiteratureStatusCorrectionHarmonic AnalysisAutoconvolution norms of densities on [0, 1]Reported that ‖f∗f‖₂² ≥ 0.574575 (Martin–O'Bryant) gives c(2) ≥ 0.758 > √(4/7), supplying a lower-bound answer2026-09-25OPEN2026-09-25
Green Problem 35
upper bound at p = ∞
LiteratureStatusCorrectionHarmonic AnalysisAutoconvolution norms of densities on [0, 1]Reported that rescaling the TTT-Discover step function (autoconvolution ratio 1.50286) to [0, 1] gives c(∞) ≤ 0.75143 < 0.75162026-09-25OPEN2026-09-25
Erdős #123
snug {2,3,5}-smooth sums
LiteratureStatusCorrectionNumber TheorySnug sums of distinct numbers of the form 2ᵏ3ˡ5ᵐReported that the sorry-free Lean theorem erdos123_dcomplete_real, specialised to (2, 3, 5) with 1 < ρ < min(1 + ε, 2), gives the snug statement, answering True2026-09-25OPEN2026-09-25
Erdős #1054
parts (i) and (ii)
LiteratureStatusCorrectionNumber TheorySums of the smallest divisors of mReported Tao's argument with Kovač's bound #{n ≤ X : f(n) ≤ δn} ≪ δ²X, which disproves f(n) = o(n) and its almost-all version (answer False for both)2026-09-25OPEN2026-09-25
Erdős #1101(ii)
subexponential good sequence
LiteratureStatusCorrectionNumber TheoryGood sequences with subexponential growthReported Peter Li's subexponential good sequence from the forum, which settles part (ii); part (i) remains open2026-09-25OPEN2026-09-25
Erdős #1110
coprime non-representables
LiteratureStatusCorrectionNumber TheoryPairwise coprime integers not representable by pᵏqˡ sumsReported that Yu–Chen (2022) handle all pairs except (5,2), (9,2), (5,3) and that a public Lean-formalized proof covers those three, answering True2026-09-25OPEN2026-09-25
Erdős #1212
composite visible-lattice path
LiteratureStatusCorrectionNumber TheoryPaths to infinity in the visible lattice that avoid prime pairsReported Alex Chengyu Li's construction of a composite-restricted path to infinity in the visible lattice, with a Lean formalization, answering True2026-09-25OPEN2026-09-25
TxGraffiti Conjecture 3
counterexample status
LiteratureStatusCorrectionGraph TheoryIndependent domination versus saturation in regular graphsReported a public 48-vertex cubic graph with i(G) = 16 > μ*(G) = 15, after an earlier 50-vertex example, as counterexamples; the statement is false2026-09-25OPEN2026-09-25
Erdős #457
q(n, log n) variant
LiteratureStatusCorrectionNumber TheoryLeast prime not dividing a product of consecutive integersDetected that the solved erdos_457 already gives infinitely many n with q(n, log n) > (2+ε) log n, answering the variant True2026-09-25OPEN2026-09-25
Monochromatic Quantum Graph
N = 6 consequences of D = 3
LiteratureStatusCorrectionQuantum InformationMonochromatic quantum graph equations for N = 6Detected that the solved N = 6, D = 3 complex case implies the three open N = 6 declarations by restricting colours and mapping ℝ into ℂ2026-09-25OPEN2026-09-25
Monochromatic Quantum Graph
N=10, D=8 over ℂ
LeanProofQuantum InformationGraph TheoryThe equality case D = N − 2 for ten-vertex monochromatic quantum graphsProved in Lean that the N=10, D=8 monochromatic quantum graph system has no solution over ℂ or any integral domain, settling the equality case D = N − 2 at N = 102026-09-26OPEN2026-09-26
Erdős #522
almost-sure root count (±1 and 0/1)
LeanProofProbabilityRoots of random polynomials in the closed unit diskProved in Lean that R_n/(n/2) → 1 almost surely for random ±1 and 0/1 polynomials; the ±1 case had earlier informal proofs, and no earlier 0/1 proof is known2026-09-26OPEN2026-09-26
Tensor rank
mrank(3, 3, 5) = 6
LeanProofLinear AlgebraMaximal rank of complex 3 × 3 × 5 tensorsProved in Lean that the maximal rank of complex 3 × 3 × 5 tensors is 6, settling the one undetermined entry of the mrank(3, 3, p) table for p ≤ 92026-09-26OPEN2026-09-26
Cyclic-subgroup solvability
Conjecture 5.5 Lean proof
LeanProofGroup TheorySolvability from the number of cyclic subgroupsFormalized Conjecture 5.5 in Lean without the classification of finite simple groups, using Feit–Thompson and Gorenstein–Walter; the result itself was already proved in the paper's second version2026-09-26OPEN2026-09-26
Green Problem 22
N₀(r) ≤ exp exp(r⁴³)
LeanProofAdditive CombinatoricsMonochromatic x + y and xy in r-colourings of {1, …, N}Proved in Lean that N₀(r) ≤ ⌈exp exp(r⁴³)⌉ by keeping all colours, improving Green–Sawhney's exp exp(r⁵⁰); they had noted that a slightly smaller exponent was reachable2026-09-27OPEN2026-09-27
Kourovka 21.149
statement and Lean proof
LeanProofStatementGroup TheoryOrder automorphisms of Dlab groups and conjugation in larger Dlab groupsFormalized Kourovka 21.149 and proved in Lean that a Dlab-group order automorphism is not induced by conjugation under any injective embedding into any Dlab group, extending Gong–Yang–Zeng's rank-one, order-preserving case2026-09-27OPEN2026-09-27
Littlewood Problem 7
at most seven touching cylinders
LeanProofStatementDiscrete GeometryComputationMutually touching congruent infinite cylindersProved in Lean that at most seven congruent infinite cylinders touch pairwise, excluding eight lines at pairwise distance one and certifying seven exactly; the literature bound was 7 ≤ N ≤ 92026-09-30OPEN2026-09-30
Reverse Littlewood–Offord
odd-n rate → 1/√2
LeanProofStatementProbabilityRandom signed sums of planar unit vectors in the unit discProved in Lean that F₂,₁(n)^(1/n) → 1/√2 for odd n, with a lower bound matching Hollom–Sorkin's 2^(−⌊n/2⌋) construction at the exponential scale2026-10-01OPEN2026-10-01
Green Problem 41
quasi-polynomial bound
LeanProofStatementDynamical SystemsQuasi-polynomial pyjama-covering boundProved in Lean that exp(10¹⁶(log(1/ε)+2)⁴) rotated pyjama sets cover the plane, a quasi-polynomial bound improving the earlier double-exponential one; the polynomial bound remains open2026-10-02OPEN2026-10-02
Ryser's conjecture
4-partite, ν ≤ 3
LeanProofStatementCombinatoricsComputationCovers versus matchings in 4-partite 4-uniform hypergraphsProved Ryser's conjecture for 4-partite hypergraphs with matching number 3 (τ ≤ 9), a case not covered by earlier results, in a computer-assisted Lean proof that also covers the known cases ν = 1, 22026-10-03OPEN2026-10-03
Quantum extremal numbers
Qex(9) = 112; bounds for 10, 11
LeanProofStatementQuantum InformationCombinatoricsMaximally mixed half-size reductions of 9-, 10- and 11-qubit pure statesProved in Lean that Qex(9) = 112, settling the nine-qubit quantum extremal number (previously 112 ≤ Qex(9) ≤ 120), and improved the upper bounds to Qex(10) ≤ 208 and Qex(11) ≤ 4222026-10-04OPEN2026-10-04
Bapat–Lal q-permanent
Conjectures 1, 2 disproved, n = 144
LeanProofCounterexampleStatementLinear AlgebraStrict monotonicity in q of q-permanents of positive definite matricesDisproved in Lean the Bapat–Lal conjecture that the q-permanent of every non-diagonal Hermitian positive definite matrix is strictly increasing on [−1, 1], and da Fonseca's half-line extension, with an order-144 counterexample2026-10-05OPEN2026-10-05
Irrationality exponent
μ(log 2) = 2
LeanProofStatementNumber TheoryRational approximation of the natural logarithm of 2Proved in Lean that the irrationality exponent of log 2 is 2, adapting OpenAI's interpolation-determinant proof that μ(π) = 2; the best published bound had been about 3.572026-10-07OPEN2026-10-07
Kalai's flag conjecture
via symmetric Funk volume
LeanProofStatementDiscrete GeometryConvex GeometryFunk-ball volumes and complete flags of centrally symmetric convex bodiesProved in Lean the Funk-volume inequality conjectured by Faifman, Vernicos and Walsh for every centrally symmetric convex body and every radius, a family interpolating between the symmetric Mahler inequality and Kalai's flag conjecture, and derived the flag conjecture (at least 2ⁿ n! complete flags)2026-10-08OPEN2026-10-08
Frantzikinakis Problem 19
commuting pointwise convergence
LeanProofStatementDynamical SystemsHarmonic AnalysisAlmost-everywhere convergence of multiple ergodic averages for commuting mapsProved in Lean that the averages (1/N)Σ f(Tⁿx)g(Sⁿx) converge almost everywhere for any two commuting invertible measure-preserving transformations, answering Frantzikinakis' Problem 19, and extended the result to L² functions2026-10-08OPEN2026-10-08
Extended Permanence Conjecture
endotactic mass-action systems
LeanProofStatementDynamical SystemsPermanence of endotactic mass-action systems with time-dependent ratesProved in Lean the Extended Permanence Conjecture of Craciun, Nazarov and Pantea for endotactic mass-action systems with natural-number exponents, with one absorbing box for all bounded measurable rate paths2026-10-09OPEN2026-10-09
Auslander–Reiten / Tachikawa
counterexample over ℚ
LeanProofCounterexampleStatementRepresentation TheorySelf-extensions of modules over symmetric algebras in characteristic zeroConstructed in Lean a finite-dimensional symmetric ℚ-algebra with a non-projective module whose positive self-Ext groups all vanish, a formally verified characteristic-zero counterexample to the Auslander–Reiten conjecture and to Tachikawa's second conjecture, adapting OpenAI's characteristic-two construction2026-10-09OPEN2026-10-09
Erdős #952
binary-form moat variants
LeanProofStatementNumber TheoryBounded-step walks through lattice points where a binary form has few prime factorsProposed and proved in Lean two variants of Erdős Problem 952: uniform moats for points where an irreducible binary form of degree at least 2 has at most r prime factors, and a classification showing that long bounded-step walks exist exactly for the forms c(aX+bY)^m2026-10-10SOLVED2026-10-10
Fisher information convexity
counterexample in dimension 41
LeanProofCounterexampleStatementProbabilityTime-convexity of the Fisher information of the Gaussian heat flow in high dimensionDisproved in Lean the convexity of Fisher information along the Gaussian heat flow, Problem (iii) of Zou–Fan–Gao–Wang, with an explicit admissible density on ℝ⁴¹ whose Fisher information is strictly concave for small times2026-10-10OPEN2026-10-10
Hajłasz–Onninen Question 1
centered maximal W¹,¹ endpoint
LeanProofStatementHarmonic AnalysisGradient bounds for the centered Hardy–Littlewood maximal function in W¹,¹Proved in Lean the Hajłasz–Onninen W¹,¹ endpoint bound ‖∇Mf‖₁ ≤ C_d‖∇f‖₁ for the centered ball maximal function in every dimension, extending OpenAI's planar proof to d ≥ 32026-10-08OPEN2026-10-08
Isometric embeddings in ℝ⁴
closed orientable surfaces
LeanProofStatementDifferential GeometrySmooth isometric embeddings of Riemannian surfaces in Euclidean four-spaceProved in Lean that every closed orientable smooth surface with any smooth Riemannian metric admits a smooth isometric embedding into ℝ⁴, extending OpenAI's isometric immersion theorem to embeddings2026-10-09OPEN2026-10-09
Strong Holevo additivity
arbitrary qubit channels
LeanProofStatementQuantum InformationAdditivity of the Holevo information of qubit channelsProved in Lean that χ(Φ⊗Ω) = χ(Φ) + χ(Ω) for every qubit channel Φ and every finite-dimensional channel Ω, without the unital or pure-output hypotheses of earlier theorems, and hence additivity over any number of uses of a qubit channel2026-10-11OPEN2026-10-11
Melbourne–Tkocz Conjecture 6.1
entropic Rogers–Shephard inequality
LeanProofStatementProbabilityConvex GeometryEntropy of the difference of two independent log-concave random vectorsProved in Lean the sharp entropic Rogers–Shephard inequality h(X − X′) − h(X) ≤ n log 2 for independent random vectors with a common log-concave density in every dimension, Conjecture 6.1 of Melbourne–Tkocz, with equality examples and optimality of the constant2026-10-12OPEN2026-10-12
Kalai's 3^d conjecture
dimension five
LeanProofStatementConvex GeometryCombinatoricsNonempty faces of centrally symmetric convex polytopesFormalized Kalai's 3^d conjecture and submitted a Lean proof of the dimension-five bound of 243 nonempty faces; the general conjecture remains open2026-10-11OPEN2026-10-11
Finitely presented residually finite p-groups
every prime
LeanProofStatementGroup TheoryAll-prime strengthening of the finitely presented Burnside problemFormalized the finitely presented Burnside problem and proved in Lean the stronger all-prime variant: an infinite finitely presented residually finite p-group exists for every prime p, extending the earlier p=2 result2026-10-13OPEN2026-10-13
Simultaneous primitive roots
unconditional positive lower density
LeanProofStatementAnalytic Number TheoryPrimes for which every member of a finite set is a primitive rootProved in Lean the unconditional bound N_A(x) ≥ c_A x/log x for positive integer bases satisfying the square condition, and the prime-base c_A x/(log x)^2 corollary without the four analytic hypotheses2026-10-09OPEN2026-10-09
Infinity Laplacian
interior regularity with bounded forcing
LeanProofStatementPartial Differential EquationsInterior gradient regularity for viscosity solutions of the infinity Laplace equationProved in Lean interior C¹,α estimates for |Δ∞u| ≤ M in every dimension n ≥ 2, with forcing term (RM)^(1/3), and formalized the homogeneous theorem and its scaling corollary2026-10-12OPEN2026-10-12
Sphere energy and cap discrepancy
dimensions 2, 8 and 24
LeanProofStatementPotential TheoryDiscrete GeometryLattice coefficients in optimal Riesz energy, logarithmic energy and cap discrepancyProved in Lean the predicted lattice coefficients for Riesz energy (−2 < s < d, s ≠ 0), logarithmic energy and spherical cap L² discrepancy on S², S⁸ and S²⁴; general dimensions and d=4 remain open2026-10-14OPEN2026-10-14
Reversible Finsler spheres
infinitely many closed geodesic images
LeanProofStatementDifferential GeometryDynamical SystemsGeometrically distinct closed geodesics for arbitrary reversible Finsler metricsProved in Lean infinitely many geometrically distinct prime closed geodesics for every smooth strongly convex reversible Finsler metric on Sⁿ, n ≥ 3; the n=2 case is cited from the literature2026-10-09OPEN2026-10-09
Green #14
explicit colouring witnessing W(3,r) ≥ 2r²
LeanProofAdditive CombinatoricsNumber TheoryAn explicit asymmetric van der Waerden colouring avoiding red triples and long blue progressionsConstructed a deterministic colouring witnessing W(3,r) ≥ 2r² for r = 520822699705116213452472320 and proved its correctness in Lean; Comparator, nanoda and Lean kernel verification passed2026-11-18OPEN2026-11-18
Erdős #104 variant
exact count of unit-circumradius triples
LeanProofStatementDiscrete GeometryCombinatoricsTriples on unit circles in planar point sets not contained in one unit circleDetermined in Lean g(2)=g(3)=0 and g(n)=binom(n−1,3)+n−2+1_{3 divides n−1} for n≥4; every n≥4 maximizer has n−1 points on one unit circle2026-11-19OPEN2026-11-19

OEIS

SequenceCategoryTopicContribution highlightSubmittedStatus
A059304ProofDouble binomial-sum identityProved Peter Bala's 2024 conjectured double-binomial identity by coefficient extraction; OEIS promoted it from a conjecture to a theorem and approved the contribution on 2026-08-082026-08-04APPROVED 2026-08-08

Lean-Eval

BenchmarkAreaContribution highlightModel / workflowSubmittedEvaluation
pi_succ_sphere_n_mulEquiv_zmod_twoAlgebraic topology · LeanFirst complete Lean formalization of $\pi_{n+1}(S^n) \cong \mathbb{Z}/2$ for every $n \ge 3$ to be accepted by Lean-Eval; recorded as a newly solved benchmarkGPT-5.6 and Fable 5 (human-in-the-loop)2026-08-06FIRST / PASS newly solved (1/1)
conway_schneeberger_fifteenNumber theory · Quadratic forms · LeanFirst complete Lean formalization of the Conway–Schneeberger fifteen theorem to be accepted by Lean-Eval; recorded as a newly solved benchmarkGPT-5.6 and Fable 5 (human-in-the-loop)2026-08-03FIRST / PASS newly solved (1/1)

Research Notes

Proof notes are separate from first solutions, formalizations, and service. Priority and review status are stated for each note.

ItemCategoryTopicContribution highlightWrittenScope
Barbados 2026 Problem 33ProofGraph TheoryOdd-hole-free triangle Ramsey graphsA proof note gives the 9-vertex construction, proves minimum order and uniqueness at that order, and settles the K₄-free variant using a known colouring theorem2026-09-24Proof note; priority unverified

Service

Mathematical status reporting, conjecture curation, review, and community service, listed separately from research contributions.

DateRoleItemService contributionEvidence
2026-08-07Invited reviewerPR #4693 — Erdős #176 formalizationInvited by Formal Conjectures collaborator mo271 to review statement fidelity; the author revised the documentation, references, and certificate packaging in responsereview
2026-08-01Status reporterErdős #113 — remaining-implication reportReported that OpenAI's result refutes the forward implication remaining after the previously known counterexample to the reverse implication related to Erdős #113report
2026-08-01Status & formalization curatorErdős #146 — result-status curationReported OpenAI's Lean-verified counterexample on the Erdős forum and updated the existing Formal Conjectures issue so #146 could be formalized as solvedforum report
2026-08-01Status & formalization curatorErdős #180 — result-status curationReported OpenAI's formal counterexample on the Erdős forum and updated the existing Formal Conjectures issue so #180 could be formalized as solvedforum report
2026-08-01Formalization curatorErdős #183 — Formal Conjectures status updateUpdated the existing Erdős #183 issue with the exact Ramsey-number divergence theorem and immutable Lean certificate so it could be formalized as solvedissue comment
2026-08-01Status & formalization curatorErdős #575 — result-status curationReported OpenAI's disproof of the corrected compactness conjecture on the Erdős forum and updated the corresponding Formal Conjectures issueforum report
2026-08-01Conjecture curatorEhrhart's volume conjecture — Formal Conjectures proposalOpened issue #4680 proposing OpenAI's Lean-formalized Ehrhart volume result for inclusion as a research-solved Formal Conjectures entryproposal issue
2026-08-01Conjecture curatorQuantum parallel repetition — Formal Conjectures proposalOpened issue #4681 proposing OpenAI's Lean-formalized quantum parallel repetition result for inclusion as a research-solved Formal Conjectures entryproposal issue
2026-08-12Status & scope reviewerErdős #692 — status and formalization-scope reviewClarified that the solved Part I does not answer Part II, and detected that the FC Part II target asks only for existence although the source asks where the maximum occursissue comment
2026-08-12Statement-audit coordinatorFormal Conjectures — possible-misformalization audit trackerOpened a central tracker for possible source mismatches, degenerate boundary cases, answer-semantics issues, and review blockers; added Erdős 1208 and 190 PR blockers, and a maintainer subsequently validated six earlier leadstracking issue
2026-08-13Statement-audit coordinatorFormal Conjectures — possible misformalizations IIOpened a second structured tracker for exact self-answers, reflexive asymptotic answers, degenerate witnesses, and indexing flaws; later added a positive A211417 repair candidatetracking issue
2026-08-16Volunteer batch auditorPR #4946 — Auto-OEIS statement and data auditSampled the 45-conjecture Auto-OEIS import and reported five reproducible or source-backed defects; mo271 said all were useful and fixed, including upstream OEIS typosaudit comment
2026-08-17Invited batch auditorPR #5016 — second Auto-OEIS statement-fidelity auditAt mo271's request, reviewed the 97-file OEIS import and reported 30 concrete statement or definition questions; the author called the review very useful and addressed the commentsaudit comment

Problem-by-problem details

#176Discrepancy of arithmetic progressions

LeanProofOPEN
  • 🟢 Status when posted OPEN — the 2026-07-09 snapshot of the problem page, taken after the posts below, still shows OPEN (statuses only move from open to solved)
  • 📖 AI-contributions wiki entries matching these posts: 🟡 Partial result (Lean) alongside Hunter (2026), under “AI alongside literature” (21 Jun 2026), and ⚪ Candidate partial result (Lean) under “AI standalone” (23 Jun 2026) — wiki · archived snapshot
  • 💬 Comments on record archived thread · live thread · live problem page
Problem. Let $N(k,\ell)$ be the minimal $N$ such that for any $f:\{1,\ldots,N\}\to\{-1,1\}$ there must exist a $k$-term arithmetic progression $P$ such that $ \left\lvert \sum_{n\in P}f(n)\right\rvert\geq \ell.$ Find good upper bounds for $N(k,\ell)$. Is it true that for any $c>0$ there exists some $C>1$ such that $N(k,ck)\leq C^k?$ What about $N(k,2)\leq C^k$ or $N(k,\sqrt{k})\leq C^k?$ — erdosproblems.com/176
  • 2026-06-23 Lean formalization
    Lean 4 formalization of a polynomial upper bound for $N(k,\sqrt{k})$: for every $k\ge 2$, $N(k,\sqrt{k}) \le \lfloor \tfrac{4}{3}(k^2-k+1)(k-1)(k(k-1)+1)\rfloor + 1$, i.e. $N(k,\sqrt{k}) = O(k^5)$ — in particular an exponential-type bound $N(k,\sqrt{k})\le C^k$. Checked with no sorry/admit; axioms are the standard [propext, Classical.choice, Quot.sound].
  • 2026-06-21 Lean formalization
    Lean 4 formalization of $N(k,2) \le \lfloor 2(k^2-1)(k-1)(2k+1)/(k-3)\rfloor + 1$ for $k\ge 5$ (an $O(k^3)$ bound), with the small cases $N(2,2)=3$, $N(3,2)=9$, $N(4,2)=13$ handled by finite Boolean checks — together giving the $C^k$-type upper bound for all $k\ge 2$.

#265Critical base-two growth for rational reciprocal-pair series

LeanProofNumber TheoryOPEN
Problem. Let $1\leq a_1<a_2<\cdots$ be an increasing sequence of integers. How fast can $a_n\to\infty$ grow if $\sum 1/a_n$ and $\sum 1/(a_n-1)$ are both rational? In particular, can one achieve $\limsup_{n\to\infty} a_n^{1/2^n}>1?$ — erdosproblems.com/265
  • 2026-09-07 independent Lean proof of the critical-base-two case
    Published an independent Lean 4 negative-answer proof for the remaining critical base-two question in OPEN Erdős Problem 265. The fully expanded theorem assumes a strictly increasing natural-number sequence with 2≤a₀, summability of both ∑1/aₙ and ∑1/(aₙ-1), and rational values for both sums. It proves first that log(aₙ)/2ⁿ→0, then rules out the robust infinitely-often growth condition: there cannot be a fixed c>1 with c^(2ⁿ)≤aₙ infinitely often. This gives a negative answer to limsup aₙ^(1/2ⁿ)>1 and, in ordinary real-analysis terms, forces the base-two roots to tend to one. The proof converts rational tail sums into integral lower bounds, introduces an envelope satisfying an eventual quadratic recurrence, takes its binary-normalized logarithmic limit, excludes a positive limit using a second residual estimate, and transfers the zero limit back to the original sequence. The pinned modular Lean 4.27.0 project completed a local lake --wfail build with 1,902 jobs, and a source scan found no sorry or admit; the repository's axiom audit reports only propext, Classical.choice, and Quot.sound. The Erdős Problems page still records the overall problem as OPEN, and this record does not claim external peer review or acceptance.

#287Unit fractions summing to 1 — must some denominator gap be ≥ 3?

ProofComputationOPEN (FALSIFIABLE)
Problem. Let $k\geq 2$. Is it true that, for any distinct integers $1<n_1<\cdots <n_k$ such that $1=\frac{1}{n_1}+\cdots+\frac{1}{n_k}$ we must have $\max(n_{i+1}-n_i)\geq 3$? — erdosproblems.com/287
  • 2026-05-29 discussion
    Process note on how the successive bounds were found (interactive human+AI workflow: pointing the AI at unused information in the denominator sequence; the AI helped with algebra, arithmetic, drafting and verifier/search code).
  • 2026-05-28 bound + verified chain extension
    Finite strengthening of catsflowers5544's good-prime-chain certificate: any counterexample must have $k \ge 68634921076157089631$. Two improvements: a sharpened conversion from chain coverage to an $n_1$ lower bound, and a certified extension of the good-prime chain by safe-prime/Pocklington-certified primes up to $p_m = 147573952589666836319$ (giving $n_1 > 39943925344138028689$). Verifier/search scripts published.
  • 2026-05-28 bound
    Harmonic-sum corollary of the same certificate: any counterexample must have $k \ge 704390616921701072$ (from $n_1 > B = 409938931585744826$ and $1 = \sum 1/n_i < \log(1+k/B)$, so $k > (e-1)B$).
  • 2026-05-28 bound
    Combining the certificate with Woett's prime-power observation: any counterexample must have $k \ge 23$ (if $k\le 22$ then $n_1 \mid \mathrm{lcm}(1,\dots,42) = 219060189739591200$, contradicting the certificate). Rules out $k=19,\dots,22$ by a route different from the Lean verification of $k\le 18$.

#301Largest subset of {1,…,N} with no unit-fraction relation

LiteratureOPEN
Problem. Let $f(N)$ be the size of the largest $A\subseteq \{1,\ldots,N\}$ such that there are no solutions to $\frac{1}{a}= \frac{1}{b_1}+\cdots+\frac{1}{b_k}$ with distinct $a,b_1,\ldots,b_k\in A$? Estimate $f(N)$. In particular, is it true that $f(N)=(\tfrac{1}{2}+o(1))N$? — erdosproblems.com/301
  • 2026-07-04 literature note
    Pointed out the preprint of Xinjun Wang, “A 667/806 Upper Bound for Erdős Problem #301 on Unit-Fraction-Free Sets” (May 2026): $f(N) \le (667/806 + o(1))N \approx 0.8275N$, improving the bound $25/28 \approx 0.8929$ recorded on the problem page (the conjectural $\tfrac12$ remains open).
  • 2026-07-04 observation
    Observation that Wang's $667/806$ configuration uses divisors of $720 = 2^4\cdot 3^2\cdot 5$, so a natural next finite experiment is to add the prime $7$.

#302Maximum size of unit-fraction-triple-free subsets

LeanProofNumber TheoryOPEN
  • 🟢 Status when posted OPEN — the supplied 2026-09-25 PDF capture shows Problem 302 as OPEN; discussion post #9181 reports the improved Lean upper bound
  • 🧩 Scope the bound improves the previous $9/10$ upper bound; the asymptotic value of $f(N)$ remains unknown. The forum post and Formal Conjectures PR #6580 document the same result, not two separate proofs
  • 💬 Public record discussion post #9181 · Formal Conjectures PR #6580 · live problem page
Problem. Let $f(N)$ be the size of the largest $A\subseteq\{1,\ldots,N\}$ such that there are no solutions to $\frac{1}{a}=\frac{1}{b}+\frac{1}{c}$ with distinct $a,b,c\in A$. Estimate $f(N)$. In particular, is $f(N)=(\tfrac12+o(1))N$? — erdosproblems.com/302
  • 2026-09-25 Lean upper-bound proof report
    Reported in the Erdős Problems discussion a Lean-verified upper bound $f(N)\le(0.8461739827964010+o(1))N$ for the largest subset of $\{1,\ldots,N\}$ without distinct $a,b,c$ satisfying $1/a=1/b+1/c$. This improves the $9/10$ upper bound attributed on the problem page to Wouter van Doorn and was also submitted as Formal Conjectures PR #6580. The forum comment links the proof repository and Lean4Web artifact and reports successful Lean compilation with only standard axioms. The 2026-09-25 PDF capture shows the problem remained OPEN; the contribution improves one bound but does not determine the asymptotic value of $f(N)$.

#341Eventual periodicity of greedy sum-avoiding sequences

ComputationOPEN
Problem. Let $A=\{a_1<\cdots<a_k\}$ be a finite set of positive integers and extend it to an infinite sequence $\overline{A}=\{a_1<a_2<\cdots \}$ by defining $a_{n+1}$ for $n\geq k$ to be the least integer exceeding $a_n$ which is not of the form $a_i+a_j$ with $i,j\leq n$. Is it true that the sequence of differences $a_{m+1}-a_m$ is eventually periodic? — erdosproblems.com/341
  • 2026-07-08 computation
    For the example $\{1,4,9,16,25\}$ the page says thousands of terms are needed before periodicity appears; under the stated Dickson rule (excluding sums $a_i+a_j$ with $i,j\le n$, $i=j$ allowed) the computation finds a period of length only 224: the periodic part starts at $a_{87}=440$ and $a_{n+224} = a_n + 1176$ for all $n \ge 87$. Reproducible Python check on Colab.

#346Ratio limit (1+√5)/2 for boundedly-complete sequences

LeanProofOPEN
Problem. Let $A=\{1\leq a_1< a_2<\cdots\}$ be a set of integers such that $A\backslash B$ is complete for any finite subset $B$ and $A\backslash B$ is not complete for any infinite subset $B$. (Here 'complete' means all sufficiently large integers can be written as a sum of distinct members of the sequence.) Is it true that if $a_{n+1}/a_n \geq 1+\epsilon$ for some $\epsilon>0$ and all $n$ then $\lim_n \frac{a_{n+1}}{a_n}=\frac{1+\sqrt{5}}{2}?$ — erdosproblems.com/346
  • 2026-06-21 Lean formalization
    Distinguished two readings of the problem (convergence-from-hypotheses vs. limit-exists). For the limit-exists interpretation, gave a Lean 4 formalization of an affirmative answer: if the two deletion hypotheses hold, the sequence has a uniform ratio gap, and $a_{n+1}/a_n \to L$ for some $L>1$, then $L = \varphi$. Axiom printout is the standard set, no sorryAx.

#352Triangles of area 1 in sets of positive measure (Koizumi's discrete model)

ProofComputationOPEN
Problem. Is there some $c>0$ such that every measurable $A\subseteq \mathbb{R}^2$ of measure $\geq c$ contains the vertices of a triangle of area 1? — erdosproblems.com/352
  • 2026-06-03 method + code
    A finite upper-bound scheme for $f(N)$ in Koizumi's discrete $N$-avoiding lattice model: normalize a diameter pair, discard points by a triangle test, and bound the rest via chromatic numbers of compatibility graphs ($\omega \le \chi$), taking the max over branches and diameter vectors. Verified $f(N) \le 1,4,4,7,9,10,12$ for $N=1,\dots,7$ — matching the known lower-bound witnesses, so these small values are now exact.
  • 2026-05-31 witness repository
    Public repository of verified lower-bound witnesses for $N=1,\dots,40$ in Koizumi's model: $N\le 27$ from BorisAlexeev's posted witnesses (converted to uniform JSON and re-verified), $N=28,\dots,40$ new witnesses from own search; all checked by an exact integer verifier, with an interactive HTML viewer.
  • 2026-05-31 theorem
    A sharp theorem for the low-side subproblem: if $F_{\mathrm{low}}(N)$ is the maximum size of $S\subseteq\mathbb{Z}^2$ all of whose triangles satisfy $\mathrm{area}+\mathrm{diam} < N$, then $F_{\mathrm{low}}(N) = (4\pi/\sqrt{27})N + O(\sqrt{N})$ — the disk constant $4\pi/\sqrt{27}$ is sharp. Proof via the Blaschke–Sas theorem and Pick's theorem.

#383Largest prime factor of (p²)(p²+1)⋯(p²+k)

ComputationOPEN
Problem. Is it true that for every $k$ there are infinitely many primes $p$ such that the largest prime divisor of $\prod_{0\leq i\leq k}(p^2+i)$ is $p$? — erdosproblems.com/383
  • 2026-06-21 computation
    Finite computational witness: the prime $p = 9188057$ satisfies $P^+\big(\prod_{0\le i\le k}(p^2+i)\big) = p$ for every $0 \le k \le 10$ (full factorizations given; $p^2+11$ breaks it). Search log over $p \le 10^7$: the number of primes surviving to $k = 1,2,\dots,10$ is $181281, 49789, 13811, 3229, 954, 227, 64, 13, 5, 1$, and none survive to $k=11$.

#389Divisibility between two consecutive blocks of k integers

ComputationOPEN
Problem. Is it true that for every $n\geq 1$ there is a $k$ such that $n(n+1)\cdots(n+k-1)\mid (n+k)\cdots(n+2k-1)?$ — erdosproblems.com/389
  • 2026-08-01 computation
    Found explicit witnesses for $n=28,\ldots,33$: $(28,18253129921815)$, $(29,18253129921814)$, $(30,359503904702161)$, $(31,359503904702160)$, $(32,2394789405254690)$, and $(33,2394789405254689)$. For each pair $(n,k)$, the product of $n,\ldots,n+k-1$ divides the product of $n+k,\ldots,n+2k-1$. Since OEIS A375071 already gives exact minimal values through $n=27$, these witnesses extend verification of the existence assertion through $n=33$. The new values are upper-bound witnesses only; no minimality is claimed.

#404p-adic valuation of sums of factorials — the function f(a,p)

ComputationLeanOPEN
Problem. For which integers $a\geq 1$ and primes $p$ is there a finite upper bound on those $k$ such that there are $a=a_1<\cdots<a_n$ with $p^k \mid (a_1!+\cdots+a_n!)?$ If $f(a,p)$ is the greatest such $k$, how does this function behave? Is there a prime $p$ and an infinite sequence $a_1<a_2<\cdots$ such that if $p^{m_k}$ is the highest power of $p$ dividing $\sum_{i\leq k}a_i!$ then $m_k\to \infty$? — erdosproblems.com/404
  • 2026-07-08 witness repository
    Attempt repository for the odd-prime columns $p = 3,5,7,11,13$ of $f(a,p)$, with a clickable GitHub-Pages table of certificates. Contribution 1: baseline-case lower bounds $f(1,3), f(1,5), f(1,7) \ge 20000$ and $f(1,11), f(1,13) \ge 5000$ (finite witnesses, checked directly mod $p^K$) — in the direction of Terence Tao's comments about $f(1,5)$ in the thread. Contribution 2: exact rows such as $f(3,3)=3$, $f(39,3)=25$, $f(700,3)=363$, $f(110,5)=30$, $f(236,5)=61$, $f(42,7)=7$, $f(113,11)=11$, $f(163,13)=13$ (with small Lean4Web checks), plus scan-survivor lower-bound rows $f(a,p) \ge v_p(a!)$.
  • 2026-07-07 witness repository
    Repository of exact values in the $p=2$ column: $f(a,2)$ for $1 \le a \le 256$. Highlight: $f(2,2) = 254$ — the problem page records Lin's upper bound $f(2,2)\le 254$, and the repository supplies the matching lower-bound witness plus a two-sided Lean/Lean4Web check. For odd $a$ the value freezes at $f(a,2) = v_2(a!)$; for even $a$ cancellations make the landscape wild (largest value found: $f(34,2) = 18444$).

#409Iterating n ↦ φ(n)+1 until a prime

ComputationOPEN
Problem. How many iterations of $n\mapsto \phi(n)+1$ are needed before a prime is reached? Can infinitely many $n$ reach the same prime? What is the density of $n$ which reach any fixed prime? — erdosproblems.com/409
  • 2026-06-21 computation
    Explicit examples with $F(n) = 67$ (iterations of $n \mapsto \phi(n)+1$ to reach a prime): $2548851069$, $3398468092$, $5097702138$, all reaching the prime $9005041$ (full 67-step trajectory listed). Method: exhaustive totient-sieve sweep of $n \le 1.5\cdot 10^9$ (max found there: $F = 66$ at $n = 1371524731$), then a targeted inverse-totient extension from the high-$F$ frontier.

#425Multiplicative Sidon sets: F(n) = π(n) + (c+o(1))·n^{3/4}/(log n)^{3/2}?

ProofLeanOPEN
Problem. Let $F(n)$ be the maximum possible size of a subset $A\subseteq\{1,\ldots,N\}$ such that the products $ab$ are distinct for all $a<b$. Is there a constant $c$ such that $F(n)=\pi(n)+(c+o(1))n^{3/4}(\log n)^{-3/2}?$ If $A\subseteq \{1,\ldots,n\}$ is such that all products $a_1\cdots a_r$ are distinct for $a_1<\cdots <a_r$ then is it true that $\lvert A\rvert \leq \pi(n)+O(n^{\frac{r+1}{2r}})?$ — erdosproblems.com/425
  • 2026-06-06 Lean formalization
    (Edited 2026-06-11 to link the complete Lean 4 formalization.) Lean 4 proof, assuming only PNT as a named hypothesis, that for every fixed $c < 3.499$ and all sufficiently large $n$: $F(n) \ge \pi(n) + c\,n^{3/4}(\log n)^{-3/2}$ — improving the previous lower coefficient $2.95$ from the thread. Construction upgrades the layered projective-plane approach: polarity graphs $ER_q$ on merged prime windows plus an averaging embedding using every layer pair under the hyperbola $pq\le n$; a uniform 20-band variant gives $C_{20} = 3.49901\ldots > 3.499$. Works in the stronger product-Sidon convention, hence implies the strict-pair version.
  • 2026-06-06 proof (natural language)
    Natural-language proof that $\Lambda = 2^{11/4}/3^{3/4} = 2.9511517859\ldots$ is the exact supremum of the standard layered rectangular semiprime ansatz: every $\lambda < \Lambda$ is attainable and no $c > \Lambda$ is, via a Bellman-inequality certificate (Cauchy–Schwarz), maximized at cutoff $a_k = \sqrt{2/3}$. So beating $\Lambda$ requires leaving the rectangular layered ansatz (as the 3.499 construction above does).

#503Largest isosceles sets in ℝᵈ

ProofOPEN
Problem. What is the size of the largest $A\subseteq \mathbb{R}^d$ such that every three points from $A$ determine an isosceles triangle? That is, for any three points $x,y,z$ from $A$, at least two of the distances $\lvert x-y\rvert,\lvert y-z\rvert,\lvert x-z\rvert$ are equal. — erdosproblems.com/503
  • 2026-05-29 observation
    Noted that in dimension $22$ one gets the exact value $f(22) = 276$: the Delsarte–Goethals–Seidel bound for spherical two-distance sets is tight in $\mathbb{R}^{22}$ (Musin), giving a spherical two-distance set of size $275$; adding the centre gives an isosceles set of size $276$, matching Blokhuis's general upper bound $f(22) \le \binom{24}{2} = 276$. (Also recorded as Corollary 6.4 in Chojecki's note.)

#516Fejér-gap variant of Pólya's minimum-modulus problem

LeanProofComplex AnalysisPROVED (LEAN)
  • 🔵 Status when posted PROVED (LEAN) — the supplied 2026-09-21 PDF capture shows that the finite-order main problem had already been proved by Fuchs and separately states the conjectural infinite-order Fejér-gap variant; discussion post #9121 records the new counterexample to that variant
  • 🧩 Scope of the solution the new theorem gives a negative answer to the additional Fejér-gap question by constructing a series whose ratio is eventually at most 1/2. It does not contradict the finite-order theorem or alter the separate Kövari growth result
  • ⏳ Formal Conjectures status PR #6442 is OPEN, unreviewed, and unmerged as of 2026-09-21; all required CI checks pass
  • 💬 Public record discussion post #9121 · Formal Conjectures PR #6442 · live problem page
Problem. Let $f(z)=\sum_{k\geq1}a_kz^{n_k}$ be an entire function of finite order with $n_k/k\to\infty$, and let $M(r)=\max_{|z|=r}|f(z)|$ and $m(r)=\min_{|z|=r}|f(z)|$. Is $\limsup_{r\to\infty}\log m(r)/\log M(r)=1$? The page also asks whether the same conclusion holds for arbitrary entire functions under the Fejér-gap condition $\sum_k1/n_k<\infty$. — erdosproblems.com/516
  • 2026-09-21 variant counterexample + Formal Conjectures proof submission
    Submitted Formal Conjectures PR #6442 with a complete Lean 4 counterexample to the conjectural infinite-order Fejér-gap variant stated in the additional text of Erdős Problem 516. The proof constructs an explicit entire lacunary series with strictly increasing exponents, nonzero coefficients, and summable reciprocal exponents, but shows that its registered minimum/maximum-modulus quotient log m(r)/log M(r) is at most 1/2 for every sufficiently large radius; consequently its limsup is not one. The construction uses widely separated polynomial blocks containing z^N(1−z/R)^m: on the same circle the series is controlled at the positive point and large at the negative point, while block separation proves local uniform convergence and the Fejér-gap condition. This answers only the additional infinite-order variant in the negative. It does not contradict or re-prove Fuchs's already-solved finite-order Fabry-gap theorem, which gives the page its pre-existing PROVED (LEAN) badge, and it does not change the separate Kövari growth variant. At fixed commit 675d078, both complete proof files report successful builds; the final axiom audit lists only propext, Classical.choice, and Quot.sound, with no sorryAx. PR #6442 changes the exact Fejér-gap target from answer(sorry) to answer(False). As of 2026-09-21 it is OPEN, unreviewed, and unmerged, and all required CI checks pass.

#522Roots of random ±1 polynomials in the closed unit disk

LeanProofComplex AnalysisOPEN
  • 🟢 Status when posted OPEN — the supplied 2026-09-26 PDF capture shows Problem 522 as OPEN with no proof claims recorded; discussion post #9188 reports Lean proofs of the almost-sure statement for both $\pm1$ and $0/1$ coefficients
  • 🧩 Scope and priority Yakir proved the convergence in probability. Informal almost-sure arguments for $\pm1$ coefficients were shared in the same thread in April 2026 by P. Chojecki and, independently, by Y. Kwon; the new contribution is kernel-checked proofs of both readings, including $0/1$ coefficients, for which the proof repository reports no earlier proof. The forum post and Formal Conjectures PR #6608 document the same result, not two separate proofs
  • 💬 Public record discussion post #9188 · Formal Conjectures PR #6608 · live problem page
Problem. Let $f(z)=\sum_{0\leq k\leq n}\epsilon_kz^k$ be a random polynomial, where $\epsilon_k\in\{-1,1\}$ independently uniformly at random for $0\leq k\leq n$. Is it true that, if $R_n$ is the number of roots of $f(z)$ in $\{z\in\mathbb{C}:\lvert z\rvert\leq1\}$, then $\frac{R_n}{n/2}\to1$ almost surely? The page notes that Erdős may instead have intended coefficients in $\{0,1\}$. — erdosproblems.com/522
  • 2026-09-26 Lean proof report
    Reported in the Erdős Problems discussion Lean proofs of both almost-sure statements that Formal Conjectures registers for this problem: $R_n/(n/2)\to1$ almost surely for coefficients uniform on $\{-1,1\}$ (erdos_522) and on $\{0,1\}$ (erdos_522.variants.zero_one). Since the thread leaves unclear which coefficient set Erdős intended, the post notes that the proofs answer the question affirmatively under either reading. The same proofs were submitted as Formal Conjectures PR #6608; the comment links the proof repository and Lean4Web artifact and reports Lean compilation with only standard axioms. Yakir had proved the convergence in probability, and informal almost-sure arguments for $\pm1$ coefficients were shared in the thread in April 2026 by P. Chojecki and, independently, by Y. Kwon. The 2026-09-26 PDF capture shows the problem still OPEN with no proof claims recorded.

#536Sets avoiding three elements with pairwise equal lcm

LeanProofOPEN
Problem. Let $f(N)$ be the largest size of $A\subseteq \{1,\ldots,N\}$ with the property that there are no distinct $a,b,c\in A$ such that $[a,b]=[b,c]=[a,c],$ where $[a,b]$ denotes the least common multiple. Estimate $f(N)$ - in particular, is it true that $f(N)=o(N)$? — erdosproblems.com/536
  • 2026-06-22 Lean formalization
    Lean 4 formalization of the upper bound $|A| \le N - \lfloor N/6 \rfloor = \lceil 5N/6 \rceil$ for $A \subseteq \{1,\dots,N\}$ with no three distinct $a,b,c$ satisfying $\mathrm{lcm}(a,b) = \mathrm{lcm}(a,c) = \mathrm{lcm}(b,c)$. Repository builds with lake build; no sorry/axiom/admit; standard axiom printout.

#603Colouring unions of countable families with |Aᵢ∩Aⱼ| ≠ 2

LeanSOLVED
  • ℹ️ Formalization of the resolution The problem was resolved by GPT 5.4 Pro (prompted by Chojecki), as credited in the page's remarks; these posts are Lean formalizations of the resolution material discussed in the thread, responding to the site's formalisation programme.
  • 🟢 Badge history last archived as OPEN on 2026-01-13; marked SOLVED by 2026-07-09 — snapshot
  • 📖 AI-contributions wiki this Lean formalization is listed under “Formalization” — of the Chojecki–GPT-5.4 Pro (2026) solution (17 May 2026) — wiki · archived snapshot
  • 💬 Comments on record archived thread · live thread · live problem page
Problem. Let $(A_i)$ be a family of countably infinite sets such that $\lvert A_i\cap A_j\rvert \neq 2$ for all $i\neq j$. Find the smallest cardinal $C$ such that $\cup A_i$ can always be coloured with at most $C$ colours so that no $A_i$ is monochromatic. — erdosproblems.com/603
  • 2026-05-17 Lean formalization
    Lean 4 / Mathlib formalization-attempt repository based on Chojecki's PDF and the thread discussion, covering two mathematical interpretations: the countable-sequence reading (find a finite colouring so no set is monochromatic) and the arbitrary-size-family reading (conditional on the Erdős–Rado partition relation, encoded as ArrowOmegaTwo κ μ). Both Lean4Web files linked.

#612Diameter bounds for connected graphs with a forbidden fixed clique

LeanProofGraph TheoryOPEN
Problem. Let $G$ be a connected graph with $n$ vertices, minimum degree $d$, and diameter $D$. Show that if $G$ contains no $K_{2r}$ and $(r-1)(3r+2)\mid d$, then $D\leq \frac{2(r-1)(3r+2)}{2r^2-1}\frac{n}{d}+O(1),$ and if $G$ contains no $K_{2r+1}$ and $3r-1\mid d$, then $D\leq \frac{3r-1}{r}\frac{n}{d}+O(1).$ — erdosproblems.com/612
  • 2026-09-08 three counterexamples completing the remaining fixed-clique cases
    Extended the Erdős 612 Lean development with counterexample families for the other three formerly open fixed-clique cases. A 71-layer periodic family of connected K₇-free graphs has minimum degree 800, order 21296p+960, and diameter at least 71p+1; its excess over the proposed coefficient 8/3 is at least p/75-11/5, which is unbounded. Since 8∣800, this single family refutes both the original r=3 / K₇ case and the amended k=6 / K₇ case. A separate 33-layer periodic family of connected K₆-free graphs has minimum degree 2200, order 27923m+8800, diameter at least 33m+3, and excess at least m/11000-37/5; it refutes the amended k=5 / K₆ case. Together with the three positive results posted the previous day, this supplies full finite-graph Lean targets for all six cases that the project's status catalog had identified as open: original K₅ proved, original K₇ refuted, amended K₄ and K₅ proved, and amended K₆ and K₇ refuted. The updated root module gives FC-shaped answer(False) wrappers for the new cases and prints their axioms; the repository reports no sorryAx and only propext, Classical.choice, and Quot.sound. This is a project proof claim rather than a merged Formal Conjectures contribution: issue #828 still contains only the earlier three-positive-case proposal, and the Erdős Problems page remains marked OPEN.
  • 2026-09-07 three fixed-clique solutions + Lean formalization proposal
    Published complete Lean 4 finite-graph proofs for three formerly open fixed-clique cases associated with OPEN Erdős Problem 612. For the original conjecture at r=2, every connected K₅-free graph satisfies 2δ(D+1)+4≤5n; this is stronger than the requested asymptotic coefficient, uses additive constant C=0, and removes the divisibility assumption. For the Czabarka–Singgih–Székely amended conjecture at k=3, every connected K₄-free graph satisfies 3δ(D+1)+6≤7n. Its k=4 / K₅ case follows from the stronger original K₅ bound. These prove the uniform interpretation of O(1) and therefore also the weaker interpretation where the constant may depend on the minimum degree. The modular development proves the graph statements directly from connectedness and clique-freeness; the final FC-shaped wrappers contain no local inequality as an assumption. The project's axiom audit reports only propext, Classical.choice, and Quot.sound, with no sorryAx. A prospective catalog of the original and amended families and their fixed-clique variants was submitted on Formal Conjectures issue #828. At the time of this first post, other fixed-clique cases still remained open; the 2026-09-08 follow-up supplies the remaining three results.

#647Is there n > 24 with max_{m<n}(m+τ(m)) ≤ n+2?

ProofOPEN (VERIFIABLE)
Problem. Let $\tau(n)$ count the number of divisors of $n$. Is there some $n>24$ such that $\max_{m<n}(m+\tau(m))\leq n+2?$ — erdosproblems.com/647
  • 2026-05-29 observation
    Additional necessary condition on top of ScottHughes's reduction: any candidate $n > 84$ must satisfy $(n-3)/3$ prime — equivalently, the two prime-chain families acquire the extra conditions $(8s+5)/3$ prime and $(16s+5)/3$ prime. Proof: the candidate condition at $m = n-3$ forces $\tau(n-3) \le 5$; since $2520 \mid n$, write $n - 3 = 3r$ with $r > 27$; composite $r$ would give $\tau(3r) \ge 6$.

#768Density of n where every prime p | n has a divisor d ≡ 1 (mod p)

LiteratureOPEN
Problem. Let $A\subset\mathbb{N}$ be the set of $n$ such that for every prime $p\mid n$ there exists some $d\mid n$ with $d>1$ such that $d\equiv 1\pmod{p}$. Is it true that there exists some constant $c>0$ such that for all large $N$ $\frac{\lvert A\cap [1,N]\rvert}{N}=\exp(-(c+o(1))\sqrt{\log N}\log\log N).$ — erdosproblems.com/768
  • 2026-07-09 literature note
    Pointed out the recent preprint of Eric Li, “The Sylow Divisor Condition: a Resolution of Erdős Problem 768”, which appears to settle the problem affirmatively: $|A\cap[1,x]|/x = \exp(-(c+o(1))\sqrt{\log x}\,\log\log x)$ with the explicit constant $c = 1/(2\sqrt{\log 2})$.

#811Balanced rainbow cliques: Axenovich–Clemen conjecture

LeanProofGraph TheoryOPEN
  • 🟢 Status when posted OPEN — the supplied 2026-09-15 PDF capture shows the general graph-classification problem as OPEN; discussion post #9045 reports the all-cliques variant proof
  • 🧩 Scope of the solution the Axenovich–Clemen conjecture is answered for all cliques K_q with q≥4 on the No side. The general classification, rainbow C₆ challenge, and quantitative threshold remain open
  • ⏳ Formal Conjectures status issue #956 is OPEN as of 2026-09-15. The nine-entry draft and proof were reported in a comment, not submitted as a PR; no independent review, acceptance, or merge is recorded
  • 💬 Public record discussion post #9045 · Formal Conjectures issue #956 comment · live problem page
Problem. Suppose $n\equiv1\pmod m$. An $m$-colour edge-colouring of $K_n$ is balanced if every vertex sees exactly $\lfloor n/m\rfloor$ edges of each colour. For which finite graphs $G$, with $m=e(G)$, does every balanced colouring of $K_n$ contain a rainbow copy of $G$ for all sufficiently large $n\equiv1\pmod m$? — erdosproblems.com/811
  • 2026-09-15 variant solution claim + Formal Conjectures statement proposal
    Published a Lean 4 proof claim for Axenovich–Clemen Conjecture 1.3, a variant on the OPEN Erdős Problem 811 page, and reported the proof and a nine-entry FC-shaped statement draft in Formal Conjectures issue #956. For every clique order q≥4 and every lower bound n₀, the final theorem constructs an n≥n₀ with n≡1 (mod q.choose 2) and a completely balanced colouring with exactly q.choose 2 colours but no rainbow K_q. This proves the No-side classification for cliques with at least four vertices, including the previously known K₄ case and extending Axenovich–Clemen's negative odd-palette cases to all clique orders. The proof uses explicit small constructions, a uniform probabilistic argument for q≥12, and a proved lexicographic amplification and definition bridge. It does not classify arbitrary graphs, answer the rainbow C₆ challenge, or determine the general colour-degree threshold, so Erdős #811 remains OPEN. The FC-shaped catalogue has nine intentional sorry placeholders; the separate standalone proof reports a successful Lean 4.27.0 check and only propext, Classical.choice, and Quot.sound for the final theorem, with no sorryAx, native_decide, or project-specific mathematical axiom. As of 2026-09-15, the result is a public proof claim in an OPEN FC issue, not a reviewed, accepted, or merged PR.

#835Colouring k-subsets of {1,…,2k} so every (k+1)-set shows all k+1 colours

ProofOPEN (VERIFIABLE)
Problem. Does there exist a $k>2$ such that the $k$-sized subsets of $\{1,\ldots,2k\}$ can be coloured with $k+1$ colours such that for every $A\subset \{1,\ldots,2k\}$ with $\lvert A\rvert=k+1$ all $k+1$ colours appear among the $k$-sized subsets of $A$? — erdosproblems.com/835
  • 2026-05-30 verification
    Independent verification (Hoffman ratio bound + Johnson-scheme eigenvalues, instead of edkirkby's Walsh–Hadamard idea) of the implication: if a proper $(k+1)$-colouring of $J(2k,k)$ exists in the remaining Ma–Tang cases (where $k+1$ is prime, $k$ even), then every colour class is closed under complementation — every $k$-set has the same colour as its complement. Uses equality analysis in Hoffman's bound and the Eberlein eigenvalue formula for the distance-$k$ graph.

#873LCM windows: the supplementary all-scale conjecture

LeanProofNumber TheoryOPEN
  • 🟢 Status when posted OPEN — the supplied 2026-09-11 PDF capture shows the original variable-k problem as OPEN; discussion post #8964 records the negative Lean proof claim for its supplementary all-X conjecture
  • 🧩 Scope of the solution the supplementary assertion that one sequence satisfies the Erdős–Szemerédi-order lower bound at every positive scale is refuted; the original “for every ε there exists k” question remains open
  • ⏳ Formal Conjectures status PR #5544 is OPEN, unreviewed, and unmerged as of 2026-09-11; its build is pending while the other reported checks pass
  • 💬 Public record discussion post #8964 · Formal Conjectures PR #5544 · live problem page
Problem. Let $A=\{a_1<a_2<\cdots\}\subset\mathbb{N}$, and let $F(A,X,k)$ count the indices $i$ such that $\operatorname{lcm}(a_i,\ldots,a_{i+k-1})<X$. Is it true that for every $\varepsilon>0$ there exists some $k$ such that $F(A,X,k)<X^{\varepsilon}$? — erdosproblems.com/873
  • 2026-09-11 supplementary-conjecture refutation + Formal Conjectures submission
    Published a Lean 4 proof and submitted Formal Conjectures PR #5544 giving a negative answer to the remaining supplementary all-scale conjecture associated with OPEN Erdős Problem 873. The proof uses the same F definition as Formal Conjectures and shows, for every positive strictly increasing sequence, that liminf F(A,X,3)/X^(1/3) ≤ 6; an all-scale lower bound F(A,X,3) ≥ c·X^(1/3)·log X with c>0 would force that ratio to diverge, a contradiction. Its arithmetic core combines an lcm/gcd identity for triples, a gap bound, and a weighted telescoping estimate, then bridges the finite-prefix count to the exact Set.encard formulation. The PR also adds FC-shaped formalizations of the two already-known Erdős–Szemerédi bounds and the all-scale statement, but the newly solved claim is only the all-scale strengthening: the original variable-k question in #873 remains OPEN. At fixed commit 44cbf18, the project reports only propext, Classical.choice, and Quot.sound in its axiom audit, with no sorry or admit. As of 2026-09-11, PR #5544 is OPEN, has no submitted review, and is not merged; its build is pending while the other reported checks pass.

#878Unconventional functions f and F: four solved statements

LeanProofNumber TheoryOPEN
  • 🟢 Status when posted OPEN — the supplied 2026-09-13 PDF capture shows the multi-part problem as OPEN; discussion post #9014 records the four Lean proof claims and fixed proof artifacts
  • 🧩 Scope of the solutions Main Questions 1 and 2, the density-one sharp 1/2 variant for F, and original formula (17) are proved. The eventual maxima equality, the two requested asymptotic-formula questions, and the quadruple-log upper bound remain open; therefore #878 remains OPEN overall
  • ⏳ Formal Conjectures status issue #997 is OPEN as of 2026-09-13. The complete draft and four proof artifacts were posted there, but there is no PR, independent review, acceptance, or merge yet
  • 💬 Public record discussion post #9014 · Formal Conjectures issue #997 comment · live problem page
Problem. For the unconventional arithmetic functions $f(n)$ and $F(n)$ defined on the problem page, is it true that, for almost all $n$, $f(n)=o(n\log\log n)$ and $F(n)\gg n\log\log n$? Is $\max_{n\le x}f(n)\sim x\log x/\log\log x$? Is $\max_{n\le x}f(n)=\max_{n\le x}F(n)$ for every, or every sufficiently large, $x$? Find asymptotics for the count of $n<x$ with $f(n)=F(n)$ and for $H(x)=\sum_{n<x}f(n)/n$; is $H(x)\ll x\log\log\log\log x$? — erdosproblems.com/878
  • 2026-09-13 multi-part solution + proposed Formal Conjectures formalization
    Published Lean 4 proofs for four FC-shaped targets within OPEN Erdős Problem 878 and reported them in Formal Conjectures tracking issue #997: (1) the first question, combining f(n)=o(n log log n) almost everywhere with F(n) ≫ n log log n; (2) the all-endpoint maximal-order asymptotic max_{n≤x} f(n) ~ x log x / log log x, strengthening the previously known subsequence result; (3) the official-remarks sharpening F(n) ~ (1/2)n log log n almost everywhere; and (4) Erdős's unproved formula (17), (x h(x)-m(x))/x → ∞, from the 1984 paper. These are four proved formal targets, not four independent arguments: in particular, the sharp 1/2 theorem supplies the decisive lower-bound component of Question 1, while the little-o component follows the established analytic framework. The draft also formalizes all seven questions on the current page, six results or conjectures in its official remarks, and formula (17). It does not claim the still-open eventual equality of the two maxima, the two requested counting/sum asymptotics, or the quadruple-log upper bound, so #878 remains OPEN overall. At fixed commit d4a8eec, the project reports successful local builds and only propext, Classical.choice, and Quot.sound for the four final targets, with no sorry, admit, sorryAx, or project-specific mathematical axiom in those proofs. As of 2026-09-13 this is a public proof claim and proposed FC formalization in OPEN issue #997; it has not been submitted as a PR, independently reviewed, accepted, or merged.

#889Capital-V₁ finite-exception conjecture

LeanProofNumber TheoryOPEN
  • 🟢 Status when posted OPEN — the supplied 2026-09-22 PDF capture shows the main lowercase-v divergence problem as OPEN; discussion post #9147 records the separate capital-V₁ finite-exception theorem
  • 🧩 Scope of the solution the registered capital-V₁ exception set is proved finite, and all exceptions through 80 are classified. The main v₀ problem, the lowercase-v₁ variant, and whether 80 is the largest capital-V₁ exception remain open
  • ⏳ Formal Conjectures status PR #6453 is OPEN, unreviewed, and unmerged as of 2026-09-22; all required CI checks pass
  • ℹ️ Artifact version the forum comment says v4.35.0-rc2, while the fixed repository toolchains and verification record pin Lean/Mathlib v4.27.0; the immutable artifact version is used here
  • 💬 Public record discussion post #9147 · Formal Conjectures PR #6453 · proof repository · live problem page
Problem. For $k\geq0$ and $n\geq1$, let $v(n,k)$ count the prime factors of $n+k$ which divide none of $n,\ldots,n+k-1$, and let $v_0(n)=\max_{k\geq0}v(n,k)$. Does $v_0(n)\to\infty$ as $n\to\infty$? Erdős and Selfridge separately asked whether the capital-$V_1$ equation $V_1(n)=1$ has only finitely many solutions. — erdosproblems.com/889
  • 2026-09-22 variant solution + Formal Conjectures proof submission
    Submitted Formal Conjectures PR #6453 and published a Lean 4 affirmative proof of the capital-V₁ finite-exception conjecture proposed by Erdős and Selfridge in their 1967 paper: the set {n : ℕ | V₁(n)=1} is finite. This is the exact registered FC target Erdos889.erdos_889.variants.V1_eq_1_finite, and the PR changes only that declaration from research open / answer(sorry) to research solved / answer(True). For an exceptional positive n, the proof takes the least positive integer Q not dividing n, proves Q=p^e, and derives the normal form n=B(p^t−1). A cubic-factor argument bounds t; multiplicative-order classes, divisor bounds, and an elementary prime-counting lower bound then force Q, t, and finally n below a uniform constant. The same files kernel-check that the exceptions up to 80 are exactly 1,2,3,4,6,7,8,12,15,16,24,30,48,80, but do not exclude additional exceptions above 80. The result does not solve the main lowercase-v₀(n)→∞ problem, the lowercase-v₁ finite-exception problem, or the stronger conjecture that 80 is the largest capital-V₁ exception. At fixed commit 44d38dd, the FC-importing and standalone sources pin Lean/Mathlib v4.27.0; the recorded direct Lean compiler checks succeeded and report only propext, Classical.choice, and Quot.sound, while a source scan finds no proof hole, custom axiom, native_decide, or unsafe declaration. The forum comment labels its Lean4Web link v4.35.0-rc2, but the immutable repository toolchains and verification record both pin v4.27.0, which is the version recorded here. As of 2026-09-22, PR #6453 is OPEN, unreviewed, and unmerged; all required CI checks pass.

#920Chromatic number of Kₖ-free graphs

LiteratureOPEN
Problem. Let $f_k(n)$ be the maximum possible chromatic number of a graph with $n$ vertices which contains no $K_k$. Is it true that, for $k\geq 4$, $f_k(n) \gg \frac{n^{1-\frac{1}{k-1}}}{(\log n)^{c_k}}$ for some constant $c_k>0$? — erdosproblems.com/920
  • 2026-06-30 status observation
    Observed that the #920 page already notes a positive answer to #986 implies a positive answer to #920, and #986 is now marked proved (following Bradač's off-diagonal Ramsey lower bound) — so #920 should now be positive / ready for a status update.

#939Sums of coprime r-powerful numbers

ComputationOPEN
Problem. Let $r\geq 2$. An $r$-powerful number $n$ is one such that if $p\mid n$ then $p^r\mid n$. If $r\geq 4$ then can the sum of $r-2$ coprime $r$-powerful numbers ever be itself $r$-powerful? Are there at most finitely many such solutions? Are there infinitely many triples of coprime $3$-powerful numbers $a,b,c$ such that $a+b=c$? — erdosproblems.com/939
  • 2026-05-25 computation
    For $r=5$, an explicit example not listed on the page: $2^5\cdot 17^6 + 7^{11} + 2^5\cdot 3^6\cdot 7^8 = 3^7\cdot 13^7$ (numerically $772402208 + 1977326743 + 134481277728 = 137231006679$; equivalently $17\cdot 34^5 + 7\cdot 49^5 + 1029\cdot 42^5 = 1521\cdot 39^5$). All four numbers are $5$-powerful and the three summands have joint gcd $1$ (not pairwise coprime). Appears smaller than the displayed $r=5$ example; reproducible verification published.

#944Dirac's critical-edge conjecture: the remaining k=4, r=1 case

LeanProofGraph TheoryOPEN
  • 🟢 Status when posted OPEN — the supplied 2026-09-10 PDF capture shows the full arbitrary-r problem as OPEN; discussion post #8949 records the proof claim for the last k=4 case of Dirac's r=1 conjecture
  • 🧩 Scope of the solution this completes Dirac's r=1 conjecture for every k≥4 when combined with Jensen's k≥5 theorem, but does not solve Erdős #944 for arbitrary r
  • ⏳ Formal Conjectures status PR #5467 is OPEN, unreviewed, and unmerged as of 2026-09-10; its build and required checks pass
  • 💬 Public record discussion post #8949 · Formal Conjectures PR #5467 · live problem page
Problem. A critical vertex, edge, or set of edges is one whose deletion lowers the chromatic number. Let $k\geq 4$ and $r\geq 1$. Must there exist a graph $G$ with chromatic number $k$ such that every vertex is critical, yet every critical set of edges has size greater than $r$? — erdosproblems.com/944
  • 2026-09-10 subproblem solution + Formal Conjectures proof submission
    Submitted Formal Conjectures PR #5467 and published an independent Lean 4 proof claim for the last open case of Dirac's 1970 conjecture inside OPEN Erdős Problem 944. The proof constructs an explicit 48-vertex, 6-regular right Cayley graph of S₄×C₂, proves that it is 4-vertex-critical, and proves that deleting any single edge does not lower its chromatic number. Consequently every critical edge set has cardinality greater than one, giving an affirmative witness for k=4, r=1. Lean checks an explicit four-colouring, finite recursive conflict certificates excluding three-colourings after each representative edge deletion, group-symmetry transport to every edge, and explicit three-colourings after deleting any vertex. Together with Jensen's published result for every k≥5, this completes Dirac's r=1 conjecture for all k≥4. It does not settle the full Erdős Problem 944 for arbitrary r. At fixed commit 9606d77, the standalone project reports only propext, Classical.choice, and Quot.sound in the final theorem's axiom audit, with no sorry, admit, native_decide, or declared axiom. PR #5467 changes the general Dirac and k=4 FC declarations from research open / answer(sorry) to research solved / answer(True). As of 2026-09-10 the PR is OPEN, has no submitted review, and is not merged; the upstream build and checks pass.

#1011Triangle-forcing thresholds at r=4 and r=5

LeanProofGraph TheoryOPEN
  • 🟢 Status when posted OPEN — the supplied 2026-09-19 PDF capture shows the general all-r problem as OPEN; discussion post #9099 records the added fixed-r=5, n≥80 Lean proof claim
  • 🧩 Scope of the solutions the fixed r=4 threshold is determined for every natural n, and the fixed r=5 threshold is determined for every n≥80. The r=5, n<80 range and the original all-r problem remain open
  • ⏳ Formal Conjectures status issue #1069 is OPEN as of 2026-09-19. Both results are documented in issue comments; there is no PR, independent review, acceptance, or merge yet
  • 💬 Public record discussion post #8945 · discussion post #9099 · FC r=4 comment · FC r=5 follow-up · live problem page
Problem. Let $f_r(n)$ be minimal such that every graph on $n$ vertices with at least $f_r(n)$ edges and chromatic number at least $r$ contains a triangle. Determine $f_r(n)$. — erdosproblems.com/1011
  • 2026-09-10 fixed-parameter solution + Lean proof publication
    Published a prospective Formal Conjectures statement and an independent Lean 4 proof claim completely determining the fixed r=4 threshold in OPEN Erdős Problem 1011 for every natural-number order n: f₄(n)=0 for n≤10, f₄(11)=21, and f₄(n)=⌊(n−3)²/4⌋+6 for n≥12. The Erdős Problems page had recorded the last formula only for n≥150, from Ren–Wang–Wang–Yang; this proof extends it to every n≥12 and handles all smaller orders separately. The formal target asks for an unknown function and proves that each supplied value is the least triangle-forcing edge count for n-vertex graphs of chromatic number at least four. The development contains explicit lower constructions, a graph-theoretic upper-bound reduction, and kernel-checked embedded CNF/LRAT certificates for the small-order range. At fixed commit ab52ebc, the project reports only propext, Classical.choice, and Quot.sound in the final theorem's axiom audit, with no sorryAx, native_decide axiom, or project-specific mathematical axiom. This settles only the fixed r=4 case: the original all-r problem and the all-n fixed r=5 problem remain open. Formal Conjectures issue #1069 contains the proposal and proof record; it is not a merged upstream entry.
  • 2026-09-19 fixed-parameter range solution + Lean proof publication
    Extended the same project with a kernel-checked Lean 4 proof for the fixed r=5 case throughout the infinite range n≥80, proving M₅(n)=⌊n²/4⌋−3n+14 and the triangle-forcing threshold f₅(n)=⌊n²/4⌋−3n+15. The final theorem proves both extremal formulas together with f₅(n)=M₅(n)+1; the development includes the graph-theoretic upper bound and explicit lower witnesses, rather than assuming a certificate hypothesis or near-extremal classification. At fixed commit c1a089d, the published verification record reports a complete local re-elaboration of all 359 embedded modules under Lean 4.34.0. The final axiom audit is exactly propext, Classical.choice, and Quot.sound, with no sorryAx, native-computation axiom, or project-specific mathematical axiom. The result was recorded in Erdős Problems discussion post #9099 and Formal Conjectures issue #1069. Its scope is r=5, n≥80: the remaining r=5, n<80 range and the original all-r problem remain open, and this is an issue comment, not an accepted or merged upstream PR.

#1039Largest disc inside {z : |f(z)| < 1} for monic polynomials

LeanOPEN
  • 🟢 Status when posted OPEN — the 2026-07-09 snapshot of the problem page, taken after the posts below, still shows OPEN (statuses only move from open to solved)
  • 📖 AI-contributions wiki this Lean formalization is listed under “Formalization” — of the GPT-5.5 Pro (2026) partial result (15 May 2026) — wiki · archived snapshot
  • 💬 Comments on record archived thread · live thread · live problem page
Problem. Let $f(z)=\prod_{i=1}^n(z-z_i)\in \mathbb{C}[z]$ with $\lvert z_i\rvert \leq 1$ for all $i$. Let $\rho(f)$ be the radius of the largest disc which is contained in $\{z: \lvert f(z)\rvert< 1\}$. Determine the behaviour of $\rho(f)$. In particular, is it always true that $\rho(f)\gg 1/n$? — erdosproblems.com/1039
  • 2026-05-15 Lean formalization
    Lean 4 / Mathlib formalization attempt of the proof strategy discussed in the thread, based on Liam Price's natural-language proof PDF.

#1040Transfinite diameter and minimal areas of polynomial lemniscates

LeanProofOPEN
Problem. Let $F\subseteq\mathbb{C}$ be closed and infinite, and let $\mu(F)$ be the infimum of the areas of $\{z:|f(z)|<1\}$ over positive-degree monic polynomials whose roots lie in $F$. Is $\mu(F)$ determined by the transfinite diameter of $F$? In particular, is $\mu(F)=0$ whenever the transfinite diameter of $F$ is $1$? — erdosproblems.com/1040
  • 2026-09-06 independent Lean proof publication
    Published an independent Lean 4 proof claim for the second question of OPEN Erdős Problem 1040. For every closed infinite F⊆ℂ, the kernel-checked theorem proves μ(F)=0 whenever the project's transfinite diameter—defined as the infimum of its finite Fekete diameters—is at least one. Here μ(F) is the infimum of the planar areas of the strict unit sublevel sets of positive-degree monic root polynomials with all roots in F. For bounded F, the proof constructs an equilibrium measure, obtains a global Frostman-type potential lower bound, perturbs the measure so that the nonpositive-potential region has arbitrarily small area, and samples finitely many polynomial roots; unbounded sets are handled directly by a two-root polynomial with widely separated roots. The publication includes a modular Lean 4.33.1 build, an FC-style expanded target, and a generated mathlib-only Lean4Web file. The final declarations use only propext, Classical.choice, and Quot.sound, with no sorry, admit, project-specific axiom, native_decide, or unsafe theorem. The formalization boundary is material: the classical theorem identifying this local iInf with the usual limit or limsup of finite Fekete diameters, and hence with logarithmic capacity, is not yet formalized in the project. Discussion post #8824 also documented that three Lean proof claims appeared on 2026-09-06—by Kenta Kitamura, shlummi, and Declan Gessel—and treated the timestamps as public-disclosure records rather than a correctness or priority judgment.

#1047Goodman's maximum number of non-convex components

LeanProofComplex AnalysisDISPROVED (LEAN)
  • 🔵 Status when posted DISPROVED (LEAN) — the supplied 2026-09-18 PDF capture shows that the original Grunsky question had already been disproved; discussion post #9085 records the separate Goodman-variant answer
  • 🧩 Scope of the solution the new result determines the exact maximum number of non-convex components in every degree. It is not the proof responsible for the page's pre-existing DISPROVED badge
  • ✅ Formal Conjectures review reviewer mo271 approved PR #6040 with “LGTM” on 2026-09-18, and all required CI checks pass. The PR remains OPEN and unmerged as of that date
  • 💬 Public record discussion post #9085 · Formal Conjectures PR #6040 · reviewer approval · live problem page
Problem. Let $f\in\mathbb{C}[x]$ be monic with $m$ distinct roots, and let $c>0$ be small enough that $\{z:|f(z)|\le c\}$ has $m$ distinct connected components. Must all these components be convex? Goodman further asks for the maximum possible number of non-convex components as a function of the degree of $f$. — erdosproblems.com/1047
  • 2026-09-18 variant solution + reviewer-approved Formal Conjectures proof submission
    Submitted Formal Conjectures PR #6040 with a complete Lean 4 answer to Goodman's follow-up question on the Erdős Problem 1047 page: among strict polynomial lemniscate sublevel regions {z : |f(z)|<c} for monic degree-n polynomials, the maximum attainable number of non-convex connected components is 0 for n≤1, 1 for n=2, 2 for n=3, and n for every n≥4. The formal target uses IsGreatest, so it proves both the universal upper bound and an attaining example in every degree. Every positive-degree component contains a root, giving the general upper bound; low degrees are classified separately, and explicit degree-four and degree-five examples plus a quadratic-factor extension construct the sharp examples in all higher degrees. This is a solution of the Goodman variant, not the original Grunsky convexity question: the latter was already disproved and the page was marked DISPROVED (LEAN) before this post. At fixed commit 87cc866, the project reports successful checks of both complete proof files and only propext, Classical.choice, and Quot.sound for the final theorem, with no sorryAx. Formal Conjectures reviewer mo271 approved PR #6040 with ‘LGTM’ on 2026-09-18, and all required CI checks pass; as of that date the PR is still OPEN and unmerged. The result is recorded as a complete formal answer to the variant, without claiming first mathematical publication.

#1056Consecutive intervals with product ≡ 1 (mod p)

ComputationOPEN
Problem. Let $k\geq 2$. Does there exist a prime $p$ and consecutive intervals $I_1,\ldots,I_k$ such that $\prod_{n\in I_i}n \equiv 1\pmod{p}$ for all $1\leq i\leq k$? — erdosproblems.com/1056
  • 2026-06-21 witnesses
    Every $2 \le k \le 14$ has an explicit finite witness, extracted from OEIS A060427 and its b-file (following up StijnC's comment): for $p = 10428007$, fifteen listed factorials are all $\equiv 8978998 \pmod p$, so the fourteen adjacent intervals between them each have product $\equiv 1 \pmod p$. Primality and all fourteen interval products verified directly; the $k=14$ witness yields witnesses for every smaller $k$.

#1103Tao's finite admissible-set conjecture modulo prime squares

LeanProofNumber TheoryOPEN
  • 🟢 Status when posted OPEN — the supplied 2026-09-22 PDF capture shows the infinite-sequence problem as OPEN; discussion post #9144 records the separate finite admissible-set theorem
  • 🧩 Scope of the solution the new Lean theorem completely resolves Tao's finite extremal conjecture by classifying every positive N, but it does not answer how fast an infinite sequence with squarefree pairwise sums must grow
  • ℹ️ Review status the proof repository and Lean4Web artifact are public, but the supplied record shows no independent review or acceptance yet
  • 💬 Public record discussion post #9144 · proof repository · Tao's conjecture · live problem page
Problem. Let $A$ be an infinite increasing sequence of integers such that every $n\in A+A$ is squarefree. How fast must $A$ grow? In a related finite problem, let $A(N)$ be the maximum size of a subset of $[N]$ that omits at least one residue class modulo $p^2$ for every prime $p$, and let $\mathcal{SF}$ be the positive squarefree integers. Is $A(N)>|\mathcal{SF}\cap[N]|$ for every $N\geq18$? — erdosproblems.com/1103
  • 2026-09-22 related finite-conjecture solution + Lean proof publication
    Published a Lean 4 proof of the finite admissible-set conjecture suggested by Wouter van Doorn and Terence Tao in work related to Erdős Problem 1103. Let A(N) be the maximum size of a subset of [N] that omits at least one residue class modulo p² for every prime p, and let SF be the positive squarefree integers. The proof establishes the conjectured strict inequality A(N)>|SF∩[N]| for every N≥18 and gives the full positive-integer classification: equality holds exactly for 1≤N≤8 and N=17; strict inequality also holds for 9≤N≤16. Small ranges are handled by exact upper bounds or explicit admissible constructions, a finite chain of prime-square intervals covers 121≤N<176400, and a Bertrand–CRT argument handles the unbounded tail. This is a complete solution of the separate finite extremal conjecture from Section 1.5 of the van Doorn–Tao paper, not a solution of the OPEN infinite-sequence growth problem #1103 itself. At fixed commit 78a0a75, the modular proof and equivalent standalone Lean4Web source pin Lean/Mathlib v4.34.0-rc2, include explicit axiom-audit commands, and a source scan finds no sorry, admit, project-specific axiom, or native_decide. The repository reports the usual propext, Classical.choice, and Quot.sound dependencies. The result was posted publicly on the Erdős Problems thread; this record does not claim independent peer review or acceptance.

#1134Klarner seed-0 density variant

LeanProofNumber TheoryDISPROVED (LEAN) - $24
  • 🔵 Status when posted DISPROVED (LEAN) - $24 — the supplied 2026-09-20 PDF capture shows this badge for the original seed-1 problem and still describes the seed-0 Klarner variant as unknown; discussion post #9107 records the new variant solution
  • 🧩 Scope of the solution the new theorem concerns the seed-0 set under 2x, 3x+2, and 6x+3 and proves natural density zero. The pre-existing disproof and AxiomProver formalization concern the different seed-1 problem under 2x+1, 3x+1, and 6x+1
  • ⏳ Formal Conjectures status PR #6432 is OPEN, unreviewed, and unmerged as of 2026-09-20; all required CI checks pass
  • 💬 Public record discussion post #9107 · Formal Conjectures PR #6432 · AxiomProver original-problem source · live problem page
Problem. Let $A\subseteq\mathbb{N}$ be the smallest set containing $1$ and closed under $x\mapsto2x+1$, $x\mapsto3x+1$, and $x\mapsto6x+1$. Does $A$ have positive lower density? Klarner also asks whether the smallest set containing $0$ and closed under $x\mapsto2x$, $x\mapsto3x+2$, and $x\mapsto6x+3$ has positive density. — erdosproblems.com/1134
  • 2026-09-20 variant solution + Formal Conjectures proof submission
    Submitted Formal Conjectures PR #6432 with a Lean 4 proof answering the Klarner variant stated on the Erdős Problem 1134 page. For the smallest set A⊆ℕ that contains 0 and is closed under x↦2x, x↦3x+2, and x↦6x+3, the proof establishes the stronger conclusion that A has natural density zero; therefore both its lower and upper natural densities are zero, and the answer to the positive-density question is No. The argument embeds the generated set diagonally in ℤ₂×ℤ₃, studies the Haar measure of its closure, derives a contradiction from the stationary decomposition and binary-cylinder recurrences if that measure is positive, and returns to integer counting through Chinese-remainder residue classes. This is a solution of the separate seed-0 Klarner variant, not the original seed-1 problem, which was already disproved and carried the page's DISPROVED (LEAN) - $24 badge. The earlier AxiomProver comment and source concern that original seed-1 problem, not this variant. At fixed commit 57042c7, the project reports successful lake build checks of both complete proof files. The final axiom audit lists only propext, Classical.choice, and Quot.sound, with no sorryAx. PR #6432 changes only the Klarner declaration to research solved with answer(False). As of 2026-09-20, the PR is OPEN, has no submitted review, and is unmerged; all required CI checks pass.

#1187Monochromatic arithmetic progressions of primes / prime common difference

LeanSOLVED
  • ℹ️ Formalization of the resolution The problem page (last edited 2026-04-08, a month before the post) already recorded the resolution (the first part follows from Green–Tao; the second is answered by the standard mod-4 colouring). This post is a Lean formalization of that standard counterexample — the page's “results could be formalisable” list names KentaKitamura. No archive capture of the page predates 2026-07-09.
  • 🔵 Status as of 2026-07-09 SOLVED — snapshot
  • 📖 AI-contributions wiki this Lean formalization is listed under “Formalization” — second part, standard counterexample (12 May 2026) — wiki · archived snapshot
  • 💬 Comments on record archived thread · live thread · live problem page
Problem. Let $k\geq 3$. Is it true that, in any finite colouring of the integers, there are monochromatic arithmetic progressions of primes of length $k$? Are there monochromatic arithmetic progressions of length $k$ whose common difference is a prime? — erdosproblems.com/1187
  • 2026-05-12 Lean formalization
    Formalized in Lean the standard mod-4 counterexample for the second part of the problem, with a Lean 4 Web typecheck.

#1207Point sets in ℝᵈ with no isosceles triangle — P_d(n)

LiteratureOPEN
Problem. Let $P_d(n)$ be such that in any set of $n$ points in $\mathbb{R}^d$ there exist at least $P_d(n)$ many points which do not contain an isosceles triangle. Estimate $P_d(n)$ - in particular, is it true that $P_2(n)<n^{1-c}$ for some constant $c>0$? — erdosproblems.com/1207
  • 2026-07-08 literature note
    Pointed out the preprint of Lee–Pohoata–Zhu, “The Minkowski grid has robustly many repeated distances”, which appears to settle the conjectural part of the problem — the authors state their first result “confirms a conjecture of Erdős from 1980”.

WOWII Graph Conjecture #2Neighbourhood independence and leaves in spanning trees

LeanProofGraph TheorySOLVED
Problem. For every finite connected simple graph $G$, the maximum number $L_s(G)$ of leaves in a spanning tree satisfies $L_s(G) \ge 2\bigl(\overline{\alpha(N(v))}-1\bigr)$, where the barred term is the average, over vertices $v$, of the independence number of the subgraph induced by the neighbours of $v$. — Formal Conjectures: WOWII Graph Conjecture 2 (merged)
  • 2026-07-27 formal conjectures submission
    Submitted Formal Conjectures PR #4654 with a complete Lean 4 proof of WOWII Graph Conjecture 2. The merged Formal Conjectures comment explicitly distinguishes three proofs: the first uses the methods of arXiv:2605.22763; the second combines a spanning-tree leaf bound from connected domination with ordered-pair double counting for adjacent neighbourhoods; and this independently developed contribution is the third. Its distinct argument chooses an edge-maximal triangle-free spanning subgraph, bounds each neighbourhood independence number by the corresponding degree, and uses the handshake lemma and Cauchy–Schwarz to find an edge with a large endpoint-degree sum. The resulting double star is extended to a spanning tree, whose leaf-count identity gives the required lower bound. The project contains no sorry, admit, native_decide, custom axiom, or unsafe declaration, and the axiom audit reports only propext, Classical.choice, and Quot.sound. A Formal Conjectures collaborator approved the contribution, and PR #4654 was merged on 2026-08-08.
  • 2026-08-08 formal proof merge
    Formal Conjectures collaborator mo271 approved the independent proof contribution with “Thanks, LGTM”, and PR #4654 was merged. The project now records this alternative argument as the third formal proof of WOWII Graph Conjecture 2 and describes its method separately from the two preceding proofs.

WOWII Graph Conjecture #31Induced-path length versus graph radius

LeanProofLiteratureGraph TheorySOLVED
Problem. Every finite connected simple graph $G$ contains an induced path on at least $2\,\operatorname{rad}(G)-1$ vertices; equivalently, $p(G) \ge 2\,\operatorname{rad}(G)-1$. — Formal Conjectures: WOWII Graph Conjecture 31
  • 2026-07-28 formal conjectures submission
    Submitted Formal Conjectures PR #4658 with an immutable link to a complete Lean 4 formalization of the induced-path bound $p(G) \ge 2\,\operatorname{rad}(G)-1$. The artifact proves the exact assumptions and conclusion of the Formal Conjectures theorem and formalizes the argument in Theorem 2.2 of Erdős, Saks, and Sós (1986), where the proof is credited to Fan Chung. Both the local and Lean4Web versions contain no sorry or admit. This is a complete formalization of a published theorem, not a claim of a new mathematical proof.

WOWII Graph Conjecture #217Leaf number, graph residue, and Hamiltonian paths

LeanProofGraph TheorySOLVED
Problem. Let $G$ be a finite connected simple graph. If $L_s(G) \le 4\,\mathbf{1}_{\operatorname{residue}(G)=2}+2$, where $L_s(G)$ is the maximum number of leaves among spanning trees of $G$, then $G$ has a Hamiltonian path. — Formal Conjectures: WOWII Graph Conjecture 217 (merged)

WOWII Graph Conjecture #316Pendant vertices and well total domination

LeanProofGraph TheorySOLVED
Problem. Let $G$ be a finite simple connected graph and let $P$ be its set of pendant vertices. If $|P|$ is at least the average degree of the complement of $G$, then $G$ is well totally dominated. — Formal Conjectures: WOWII Graph Conjecture 316 (merged)
  • 2026-07-13 formal conjectures submission
    Submitted Formal Conjectures PR #4426 with a complete Lean 4 proof of WOWII Graph Conjecture 316, matching the theorem then recorded as research open. The project version was checked with Lean 4.27.0 and Mathlib v4.27.0 against Formal Conjectures commit b2e608f; the self-contained version also checks in Lean4Web. Both files compile without sorry, admit, or user-declared axioms, and the axiom printout is the standard [propext, Classical.choice, Quot.sound].
  • 2026-08-07 formal proof merge
    After review, a Formal Conjectures reviewer accepted the complete Lean 4 proof contribution and PR #4426 was merged. The upstream entry now marks WOWII Graph Conjecture 316 as research solved and records the proof through the repository's formal_proof using lean4 mechanism.

Lonely Runner Conjecture — Tao (2017) asymptotic variantImproved asymptotic lower bound for the gap of loneliness

LeanStatementLiteratureSOLVED
Problem. There exists an absolute constant $c>0$ such that, for all sufficiently large $n$, $\delta_n \ge 1/(2n)+c\log n/(n^2(\log\log n)^2)$. — Formal Conjectures: Lonely Runner, Tao (2017) variant (merged)
  • 2026-02-06 formal conjectures statement submission
    Submitted Formal Conjectures PR #2177 adding the Tao (2017) variant of the Lonely Runner Conjecture. The contribution defines the unit circle $\mathbb{R}/\mathbb{Z}$, distance to the nearest integer, the maximal loneliness of an integer-velocity tuple, and the infimum $\delta_n$ over admissible tuples. It then states Tao's improved asymptotic lower bound $\delta_n \ge 1/(2n)+c\log n/(n^2(\log\log n)^2)$ for some $c>0$ and all sufficiently large $n$. This is a Lean formalization of a known theorem statement, not a formal proof of Tao's theorem.
  • 2026-05-06 statement merge
    Formal Conjectures collaborator Paul-Lez approved the revised statement with “LGTM”, and PR #2177 was merged into the upstream repository.

Open Quantum Problem #40Refinement of the Bessis–Moussa–Villani conjecture

Quantum InformationLiteratureStatusOPEN
Problem. For positive semidefinite matrices $A,B$, let $p_{n,m}(A,B)$ be the normalized $t^n s^m$ coefficient of $\operatorname{tr}(tA+sB)^{n+m}$. Is it always true that $\operatorname{tr}(A^nB^m) \ge p_{n,m}(A,B) \ge \operatorname{tr}(\exp(n\log A+m\log B))$? — Open Quantum Problem #40
  • 2026-07-13 literature note
    Pointed out the preprint of Hyunho Cha and Jungwoo Lee, “One-parameter counterexamples to the refined Bessis-Moussa-Villani conjecture” (March 2026), which gives a class of positive-semidefinite matrix counterexamples to the proposed upper bound $\operatorname{tr}(A^nB^m) \ge p_{n,m}(A,B)$. Moreover, the ratio $p_{n,m}(A,B)/\operatorname{tr}(A^nB^m)$ can be arbitrarily large. This falsifies the upper half of the two-sided refinement posed in Formal Conjectures issue #3457; it does not by itself settle the proposed exponential lower bound.

MathOverflow #507128 — proper invertible ideal in a ring equal to its total quotient ringExplicit cuspidal-cubic idealization counterexample

LeanProofCounterexampleSOLVED
Problem. There exist a commutative ring $R$ that is its own total quotient ring and a proper ideal $I\subsetneq R$ whose underlying $R$-module is invertible. — Formal Conjectures: MathOverflow 507128 (merged)
  • 2026-07-27 formal conjectures submission
    Submitted Formal Conjectures PR #4644 with an explicit Lean 4 construction solving the exact target extracted from MathOverflow 507128. Let $D=\mathbb{C}[X,Y]/(Y^2-X^3)$ be the cuspidal-cubic coordinate ring and $P=(X-1,Y-1)$ its invertible ideal. The proof assembles a detector module from evaluation fibres away from the point corresponding to $P$, forms the idealization $R=D\ltimes M$, and takes the ideal induced by the multiplication map $R\otimes_D P\to R$. It proves that every nonunit of $R$ is a zero divisor, hence $R$ is its own total quotient ring, and proves that the induced ideal is both proper and invertible. The concrete algebraic input is constructed in Lean rather than assumed opaquely. The fixed proof contains no sorry or problem-specific axiom, and its axiom audit reports only propext, Classical.choice, and Quot.sound. The MathOverflow question itself states that a proper invertible ideal in such an $R$ is equivalent to an infinite-order Picard class comparable to $[R]$. The asker subsequently published an answer identifying this construction as an independent solution.
  • 2026-08-04 formal proof merge
    The MathOverflow asker inspected the 735-line Lean4Web file and said it was fair to mark the problem formally solved. Formal Conjectures collaborator mo271 then approved the changes with “Thanks, LGTM!” and merged PR #4644.

Bézier–Bernstein Voronovskaja formula — α > 0, α ≠ 1; C² functions on [0,1]Central-limit-scale asymptotics of Bézier-type Bernstein operators

LeanProofOPEN
Problem. Let $\alpha>0$ with $\alpha\ne1$, let $f$ be $C^2$ on $[0,1]$, and let $x\in[0,1]$. Then $\sqrt n\,(B_{n,\alpha}f(x)-f(x))$ converges to $\mu(\alpha)\sqrt{x(1-x)}\,f'(x)$, where $\mu(\alpha)=\int_0^\infty\bigl((1-\Phi(t))^\alpha-(1-\Phi(t)^\alpha)\bigr)\,dt$ and $\Phi$ is the standard normal distribution function. — Formal Conjectures: Bézier–Bernstein Voronovskaja problem
  • 2026-07-27 formal conjectures submission
    Submitted Formal Conjectures PR #4646 with an explicit answer and a complete Lean 4 proof of the Bézier–Bernstein Voronovskaja problem. The proof constructs the powered-binomial probability law, proves its Gaussian-tail limit and uniform moment bounds, identifies the scaled first centered moment with $\mu(\alpha)\sqrt{x(1-x)}$, shows that the scaled second centered moment vanishes, and combines these facts with a uniform quadratic Taylor remainder. It also handles the endpoint cases $x=0$ and $x=1$. The original problem retains $\alpha\ne1$, although the formal argument works for every $\alpha>0$. The fixed proof contains no sorry or admit, and its axiom audit reports only propext, Classical.choice, and Quot.sound.

Convex additive VC₂ dimension — convex subsets of ℝ³, proposed upper bound 1Six-halfspace counterexample to the proposed upper bound

LeanProofCounterexampleOPEN
Problem. Must every convex set $C \subseteq \mathbb{R}^3$ have additive $\mathrm{VC}_2$ dimension at most $1$? — Formal Conjectures: convex additive-VC₂ bound in ℝ³
  • 2026-07-28 formal conjectures submission
    Submitted Formal Conjectures PR #4657 with an explicit counterexample to the claim that every convex subset of $\mathbb{R}^3$ has additive $\mathrm{VC}_2$ dimension at most $1$. The counterexample is a convex polyhedron cut out by six halfspaces with integer coefficients; a four-point $2\times2$ additive grid and sixteen explicit translations realize all sixteen membership patterns. The Lean 4 proof checks the certificate by exact arithmetic and has no sorry, admit, native_decide, custom axiom, or unsafe declaration; its axiom audit reports only propext, Classical.choice, and Quot.sound. A different 13-halfspace counterexample had previously appeared in closed draft fork PR #88, but its pinned Lean CI did not pass; the contribution here is the smaller six-halfspace construction and a completed kernel-checked formalization. PR #4657 also preserves the possible upper bound $2$ as a separate open statement.

Monochromatic Quantum Graph — even N ≥ 6, D ≥ 3 over ℤ and trinary integer weightsNonexistence for all even integer-weight cases

LeanProofQuantum InformationGraph TheoryOPEN
Problem. For every even $N \ge 6$ and every $D \ge 3$, the monochromatic quantum graph equation system has no solution with weights in $\mathbb{Z}$; consequently it has no solution when every weight is restricted to $\{-1,0,1\}$. — Formal Conjectures: Monochromatic Quantum Graph integer cases
  • 2026-07-29 formal conjectures submission
    Submitted Formal Conjectures PR #4659 with a general Lean 4 theorem proving that for every even $N \ge 6$ and every $D \ge 3$, no integer-valued weighting satisfies the monochromatic quantum graph equation system. This settles all six integer declarations in the Formal Conjectures file, and the six declarations with weights restricted to $\{-1,0,1\}$ follow immediately. The proof uses no sorry, declared axiom, or native_decide; the wrapper axiom audit reports only propext, Classical.choice, and Quot.sound. The $N=6,D=3$ integer case and its trinary corollary overlap with PR #4511; the contribution here is the uniform all-even theorem.

Monochromatic Quantum Graph — even N ≥ 6, D ≤ N − 2 over commutative integral domainsSharp upper bound on the number of colors

LeanProofQuantum InformationGraph TheoryOPEN
Problem. Over every commutative integral domain, any solution of the monochromatic quantum graph equation system on an even number $N \ge 6$ of vertices has at most $N-2$ colors. — Formal Conjectures: Monochromatic Quantum Graph high-color cases
  • 2026-07-29 formal conjectures submission
    Submitted Formal Conjectures PR #4661 with a single solver-free Lean 4 theorem proving that every monochromatic quantum graph equation-system solution on an even number $N \ge 6$ of vertices over any commutative integral domain satisfies $D \le N-2$. This simultaneously proves nonexistence for $N=6,D=5$ over both $\mathbb{C}$ and $\mathbb{R}$ and for $N=10,D=9$ over $\mathbb{C}$, settling three declarations recorded as open. The proof uses no SAT solver, finite-classification certificate, native_decide, or sorry; its axiom audit reports only propext, Classical.choice, and Quot.sound.

Monochromatic Quantum Graph — N=6, D=4 over ℂNonexistence of a monochromatic quantum graph equation-system solution

LeanProofQuantum InformationGraph TheoryOPEN
Problem. For $N=6$ and $D=4$, does the monochromatic quantum graph equation system have no solution over $\mathbb{C}$? — Formal Conjectures: Monochromatic Quantum Graph (N=6, D=4)
  • 2026-07-30 formal conjectures submission
    Submitted Formal Conjectures PR #4664 with a complete Lean 4 proof of eqSystem6_no_solution_d4, establishing that no weights $W : \mathrm{WeightsN}(6,4,\mathbb{C})$ satisfy the monochromatic quantum graph equation system. The formal result is stronger than the recorded complex case: it works over every commutative integral domain. The proof contains no sorry or admit, and its axiom audit reports only propext, Classical.choice, and Quot.sound. The PR proposes changing the Formal Conjectures entry from answer(sorry) to answer(True) and recording the external Lean proof.

Weak Tiling Problem 4.3Weak tiling measures versus convex combinations of proper tilings

LeanProofCounterexampleOPEN
Problem. Let $\Omega \subset \mathbb{R}$ be a finite union of intervals and let $\nu$ be a weak tiling measure for $\Omega$. Must $\nu$ be expressible as a convex combination of proper tiling measures? — Formal Conjectures: Weak Tiling Problem 4.3
  • 2026-08-04 formal conjectures submission
    Submitted Formal Conjectures PR #4704 after an AI-assisted investigation found a counterexample to the statement recorded as Weak Tiling Problem 4.3. For $\Omega=(0,1)\cup(2,3)\cup(12,13)\cup(30,31)$, every convex combination of proper tiling measures assigns equal mass to $\{7\}$ and $\{15\}$, while the constructed weak tiling measure has masses $1/2$ and $0$, respectively. The linked Lean 4 development gives a complete kernel-checked proof with no sorryAx; #print axioms reports only propext, Classical.choice, and Quot.sound. The PR proposes changing the Formal Conjectures entry from answer(sorry) to answer(False) and marking it solved.

Erdős #42 — constructive Sidon-set variantThreshold function for Sidon sets with disjoint difference sets

LeanProofAdditive CombinatoricsSOLVED
Problem. There exists a function $f:\mathbb{N}\to\mathbb{N}$ such that, whenever $M\ge1$ and $N\ge f(M)$, every maximal Sidon set $A\subseteq\{1,\ldots,N\}$ admits a Sidon set $B\subseteq\{1,\ldots,N\}$ of size $M$ whose difference set intersects $A-A$ only at $0$. — Formal Conjectures: Erdős #42 constructive variant (merged)
  • 2026-08-10 formal conjectures submission
    Submitted Formal Conjectures PR #4853 with a complete Lean 4 proof of the constructive variant attached to Erdős Problem 42. The development imports Shashi456's formalization of the solved main Erdős #42 theorem, proves that its Sidon and maximal-Sidon predicates agree with the Formal Conjectures definitions, extracts a threshold for each $M$ from the Filter.atTop eventual statement, and uses classical choice to obtain the requested function $f$. This establishes existence of a threshold function; it does not compute an explicit numerical bound. The fixed proof contains no sorryAx, and #print axioms reports only propext, Classical.choice, and Quot.sound. During review, the proof metadata was updated to formal_proof using formal_conjectures and the dependency on Shashi456's main-theorem formalization was credited. Reviewer mo271 then approved the final revision, and PR #4853 was merged on 2026-08-10, changing the upstream declaration to research solved with answer(True).

Moving Sofa — unrestricted-set uniqueness misformalizationGerver-sofa uniqueness stated over every planar set

Misformalization DetectionOPEN
Problem. For every set $s\subseteq\mathbb{R}^2$, is $\operatorname{sofaConstant}=\operatorname{volume}(s)$ equivalent to $s=\operatorname{gerversSofa}$? — Formal Conjectures: Wikipedia/MovingSofa before correction
  • 2026-08-10 formal conjectures misformalization report
    Opened Formal Conjectures issue #4851 to show that the recorded uniqueness theorem was too broad: removing one point from Gerver's sofa gives a different planar set with the same area. This disproves the formal statement but not the intended moving-sofa uniqueness claim. William Blair then observed that merely adding a moving-sofa hypothesis would still leave translated copies as counterexamples, and authored PR #4852 to state uniqueness up to a rigid motion. The PR explicitly credits KitaKen1's point-removal counterexample, was approved by Formal Conjectures reviewer mo271, and was merged on 2026-08-10; issue #4851 was closed by the correction. The corrected conjecture itself remains research open.

Green Problem 14 — AKS14 lower-bound status correctionEstablished lower bounds W(3,t) for 20 ≤ t ≤ 39

LiteratureStatusCorrectionOPEN
Problem. AKS14 establishes the twenty lower bounds for $W(3,t)$ recorded in Formal Conjectures for $20\le t\le39$, with finite good-partition certificates in Appendix A. For $20\le t\le30$ it separately conjectures equality with those bounds; for $31\le t\le39$ it states only lower bounds. — Formal Conjectures: Green Problem 14 lower bounds
  • 2026-08-10 formal conjectures misformalization report
    Opened Formal Conjectures issue #4854 to document a mathematical-status error in Green Problem 14. The repository labels the twenty declarations W_3_20_lower through W_3_39_lower as research open and describes them as conjectured, whereas AKS14 presents all twenty inequalities as established lower bounds and supplies good-partition certificates in Appendix A. The report distinguishes these proved inequalities from the stronger exact-value conjectures: AKS14 conjectures equality only for $20\le t\le30$, while for $31\le t\le39$ it records further lower bounds without an equality conjecture. The proposed correction is therefore to mark the twenty lower-bound declarations research solved.

Erdős #973 — negative-answer status correctionExterior power sums and nonexistence of an exponential constant

LiteratureStatusCorrectionOPEN
Problem. Does there exist $C>1$ such that, for every $n≥2$, there are $z_1,…,z_n∈ℂ$ with $z_1=1$ and $|z_i|≥1$ for which $\max_{2\le k\le n+1}|\sum_i z_i^k|<C^{-n}$? — Formal Conjectures: Erdős Problem 973
  • 2026-08-10 formal conjectures status-correction report
    Opened Formal Conjectures issue #4855 to report that Yanping Luo, Ruiyi Yang, and Keheng Zhu's paper Exterior power sums gives a negative answer to Erdős Problem 973. For every fixed $\lambda>0$ and all sufficiently large $n$, they prove that any $z_1,…,z_n∈ℂ$ with $|z_j|≥1$ satisfies $\max_{2\le k\le n+1}|\sum_j z_j^k|>e^{-\lambda n}$. Consequently, the $n$th root of the optimal maximum tends to 1, so no constant $C>1$ can satisfy the Formal Conjectures statement. The report proposes answer(False), research solved, and adding the paper to the references. The Lean development linked by the authors checks parts of the argument, not the full theorem.

Square Packing — negative-radius least-element flawNonexistence of least feasible radii when radii range over all reals

Misformalization DetectionOPEN
Problem. Do the sets of real radii admitting, respectively, a packing of three unit squares in $\operatorname{Circle}(r)$ and a packing of fifteen unit circles in $\operatorname{Circle}(r)$ have least elements? — Formal Conjectures: Wikipedia/SquarePacking
  • 2026-08-10 formal conjectures misformalization report
    Opened Formal Conjectures issue #4859 to identify a shared flaw in least_three_square_packing_in_circle and least_fifteen_circle_packing_in_circle. Because Circle r is defined using only $r^2$, if $r>0$ then $\operatorname{Circle}(-r)=\operatorname{Circle}(r)$, so $-r$ is a smaller feasible radius. If $r\le0$, then $r^2<(r-1)^2$, hence $\operatorname{Circle}(r)\subseteq\operatorname{Circle}(r-1)$, and $r-1$ is smaller and feasible. Thus every feasible radius has a smaller one and neither formal set has a least element. This disproves only the misformalized statements, not the intended packing problems. The proposed repair is to require $0\le r$ or use $\mathbb{R}_{\ge0}$; the three square-container declarations are unaffected.

Erdős #80 — infeasible-c quantification misformalizationBook-size asymptotics quantified beyond the feasible edge-density range

Misformalization DetectionOPEN
Problem. For every real $c>0$, must the forced book size $f_c(n)$ eventually exceed $n^\varepsilon$ for some $\varepsilon>0$, or at least satisfy $f_c(n)\gg\log n$? — Formal Conjectures: Erdős Problem 80
  • 2026-08-11 formal conjectures misformalization report
    Opened Formal Conjectures issue #4867 to identify a shared quantifier flaw in erdos_80 and erdos_80.variants.log. An admissible graph at density parameter $c$ must have at least $cn^2$ edges, but a simple graph on $n$ vertices has at most $n(n-1)/2$ edges. At $c=2$ no such graph exists for $n\ge1$, so the repository's empty-set convention gives $f(2,n)=0$. Consequently neither eventual lower bound by $n^\varepsilon$ nor domination of $\log n$ can hold for every $c>0$. This disproves only the over-quantified formal statements, not the intended Erdős problem. The report proposes restricting both declarations to the feasible range, for example $0<c<1/2$.

Fernandes Conjecture 1 — 2-generation of Γ(m ⊕ n)Two-generation of equal-sign permutation-pair subgroups

LeanProofGroup TheoryOPEN
Problem. For integers $m\ge n\ge2$ outside $(2,2),(3,3),(4,3),(4,4)$, the equal-sign subgroup $\Gamma_{m\oplus n}\le S_m\times S_n$ is generated by two elements. — Formal Conjectures: Fernandes Conjecture 1
  • 2026-08-11 formal conjectures submission
    Submitted Formal Conjectures PR #4868 with a complete Lean 4 proof of Fernandes Conjecture 1. For each degree the proof constructs a transposition and an even permutation that generate the full symmetric group. For unequal degrees, paired generators give surjective coordinate projections; explicit powers and commutators place a three-cycle in the relevant Goursat kernel, so normality forces that kernel to contain the alternating group. The parity condition then yields the entire equal-sign subgroup. Equal degrees $m=n\ge5$ use a related pair whose square supplies the required three-cycle, and the low-degree cases are handled explicitly. The final audit checks the proved declaration and the Formal Conjectures declaration against the same proposition. The fixed proof contains no sorry or admit, uses neither native_decide nor bv_decide, and #print axioms reports only propext, Classical.choice, and Quot.sound. The PR proposes changing the upstream declaration from research open to research solved and attaching the fixed proof.

Erdős #608 — known counterexample status reportEdges contained in 5-cycles above the Mantel threshold

LiteratureStatusGraph TheorySOLVED
  • 🔵 Status when posted SOLVED — Grzesik–Hu–Volec (2016), p. 2 states immediately before Construction 2 that the Füredi–Maleki construction disproves Conjecture 1.1 for k=2, the C₅ case recorded as Erdős #608.
  • 🔗 Source problem Erdős Problem 608
Problem. If $G$ has $n$ vertices and more than $n^2/4$ edges, must at least $2n^2/9$ edges of $G$ be contained in a $C_5$? — Erdős Problem 608
  • 2026-08-11 formal conjectures literature/status report
    Posted a literature and status report in Formal Conjectures issue #824. The report identifies the exact passage in Grzesik–Hu–Volec (2016): the paragraph immediately before Construction 2 says that the Füredi–Maleki construction disproves Conjecture 1.1 for k=2, which is the C₅ case recorded as Erdős #608. It also links a complete external Lean 4/mathlib disproof published on 2026-07-29 by Emerson Hsieh (primateria), using a rational specialization of the Füredi–Maleki template and proving an explicit uniform gap below 2/9. The Lean proof is credited to its external author; this contribution is the evidence-backed literature and status report to Formal Conjectures, not a claim to either the original counterexample or its formalization.

Erdős #539 — three exponent variantsGrowth exponent of the cofactor-set threshold

LeanProofAdditive CombinatoricsSOLVED
  • 🟢 Status when posted OPEN — Formal Conjectures source when submitted marks sq_cube_root, sq_cube_root_isBigO, and limit as research open. The main Erdős #539 declaration and the Θ(√n) variants are separate and remain open.
  • 🔵 Current status SOLVED — Formal Conjectures PR #4869 approval was merged on 2026-08-15; Formal Conjectures collaborator mo271 approved the contribution with “Thanks, nice!”. The merged source marks all three exponent-related variants research solved: the logarithmic exponent is 1/2, while the two proposed n^(2/3) statements have answer False. The main Erdős #539 declaration and the Θ(√n) variants remain open.
  • 🔗 Source problem Formal Conjectures: Erdős Problem 539 variants (merged)
Problem. For the Formal Conjectures threshold $h(n)$ over $n$-element subsets of $\mathbb N$, determine $\lim \log h(n)/\log n$, and decide whether $n^{2/3}=O(h(n))$ and $h(n)=\Theta(n^{2/3})$. — Formal Conjectures: Erdős Problem 539 variants (merged)
  • 2026-08-11 formal conjectures submission
    Submitted Formal Conjectures PR #4869 with complete Lean 4 proofs resolving three OPEN exponent-related variants of Erdős #539. Theorem A.1 of Schmitt et al. proves the exponent 1/2 for a threshold defined using finite sets of positive integers, whereas Formal Conjectures allows zero. The submitted bridge proves erdosH n ≤ cofactorThreshold (n+1) ≤ erdosH (n+1) and transfers the logarithmic limit by a squeeze argument. It then uses the limit to prove that n^(2/3) = O(cofactorThreshold n) is false and consequently that cofactorThreshold n = Θ(n^(2/3)) is false. The three exported declarations are checked against the corresponding Formal Conjectures propositions; the fixed proof contains no sorry or admit, and its axiom audit reports only propext, Classical.choice, and Quot.sound. This does not claim to resolve the main Erdős #539 problem or the Θ(√n) variants, which remain open.
  • 2026-08-15 formal proof merge
    Formal Conjectures collaborator mo271 approved the complete Lean proof contribution with “Thanks, nice!”, and PR #4869 was merged. The upstream file now marks sq_cube_root, sq_cube_root_isBigO, and limit as research solved and links the fixed external proof; the main problem and Θ(√n) variants are still open.

MathOverflow #10799 — Kahn–Kalai Conjecture 7Critical-probability optimality of monotone Boolean families

LeanProofCounterexampleSOLVED
  • 🟢 Status when posted OPEN — Formal Conjectures source when submitted marks mathoverflow_10799.variants.kahn_kalai_conjecture_7 as research open with an unknown Boolean answer. The unconditioned and weak variants are separate.
  • 🔵 Current status SOLVED — Formal Conjectures PR #4872 approval was merged on 2026-08-15; Formal Conjectures collaborator mo271 approved the status change and proof link with “Thanks, LGTM!”. The upstream entry now records the Diskin–Kreitner counterexample as research solved with answer(False), credits Sahar Diskin and Uri Kreitner for the counterexample, and credits Kenta Kitamura for adapting it to the exact Formal Conjectures statement and formalizing it in Lean.
  • 🔗 Source problem Formal Conjectures: MathOverflow 10799, Kahn–Kalai variant (merged)
Problem. For every increasing family $F\subseteq2^{[n]}$, if $\mu_t(F)=1/2$ and $t/s>1000\log n$, must some $p\in[s,t]$ satisfy the Formal Conjectures optimality inequality? — Formal Conjectures: MathOverflow 10799, Kahn–Kalai variant (merged)
  • 2026-08-11 formal conjectures submission
    Submitted Formal Conjectures PR #4872 with a complete Lean 4 disproof of the OPEN Kahn–Kalai Conjecture 7 variant registered under MathOverflow #10799. The mathematical counterexample is due to Sahar Diskin and Uri Kreitner; this contribution is an independent formalization, not the discovery of the counterexample. It constructs a scale-dense dual-tribes family whose members meet every block, proves the exact Bernoulli-measure and boundary identities, and chooses closely spaced block scales so that continuity gives μ_t(F)=1/2 and t/s > 1000 log n. At every p in [s,t], the normalized boundary is strictly too large for the Formal Conjectures IsOptimal inequality, so no required p exists. The repository proves both the literal negation of the fixed-1000 Formal Conjectures proposition and the original Kahn–Kalai Conjecture 7 quantifier pattern. The fixed proof contains no sorry, admit, custom axiom, or native_decide; #print axioms reports only propext, Classical.choice, and Quot.sound.
  • 2026-08-15 formal proof merge
    Formal Conjectures collaborator mo271 approved the status change and external Lean proof with “Thanks, LGTM!”, and PR #4872 was merged. The merged source marks the variant research solved with answer(False), while preserving the distinction between Diskin and Kreitner's mathematical counterexample and Kenta Kitamura's adaptation and Lean formalization.

Roff–Willerton Conjecture 3.3 — ten-point counterexampleMicroscopic weighting versus finite concentration

LeanProofCounterexampleMetric GeometryOPEN
Problem. Does every nonempty finite metric space admit a microscopic weighting if and only if its distance matrix has finite concentration? — Formal Conjectures: microscopic weighting conjecture
  • 2026-08-11 formal conjectures submission
    Submitted Formal Conjectures PR #4875 with a complete Lean 4 counterexample to the OPEN microscopic-weighting conjecture of Roff and Willerton. The proof constructs an explicit integral 10×10 distance matrix and checks symmetry, positive off-diagonal entries, and every triangle inequality. An explicit rational gauging proves finite concentration. For a kernel vector v with coordinate sum zero and A v = 0, the second-order expansion of Z(t) = exp(-tA) forces every possible microscopic limit g′ to satisfy vᵀ(A ∘ A)g′ = 0. A rational row-space certificate shows this quantity is unchanged across all gaugings, while the explicit gauging gives the nonzero value -103175560609824, yielding a contradiction. The analytic part also proves that the exponential similarity matrix is invertible for all sufficiently small positive t, so the inverse weighting used in the limit is the intended one. The fixed proof establishes the exact Formal Conjectures answer(False) proposition, contains no sorry, admit, or custom axiom, and its axiom audit reports only propext, Classical.choice, and Quot.sound.

Green Problem 52 — logarithmic variantAffine subspaces in double sumsets over Boolean cubes

LeanProofLiteratureCounterexampleAdditive CombinatoricsSOLVED
Problem. If $A\subseteq\mathbb F_2^n$ has an additive complement of size $K$, must $2A$ contain a coset of codimension $O(\log K)$ with universal constants? — Formal Conjectures: Green Problem 52 logarithmic variant (merged)
  • 2026-08-12 formal conjectures literature formalization submission
    Submitted Formal Conjectures PR #4887 to close the OPEN logarithmic strengthening of Green's Problem 52 by formalizing its known literature counterexample in Lean 4; reviewer mo271 approved the contribution and the PR was merged on 2026-08-12. Green's 2025 update credits Kaave Hosseini and Ryan Alweiss with independently observing that the logarithmic bound is too optimistic and records a Hamming-ball counterexample; this contribution is its formalization, not a claim to discovering the mathematical counterexample. For N = 2^(2t), the proof takes a Hamming ball of radius (N-2^t)/2 and uses Walsh characters, their complements, and Parseval's identity to construct an additive complement S with |S| ≤ 2N. Every vector in A+A has Hamming weight at most N-2^t, while a Lean-proved pivot-coordinate lemma forces any nonempty affine subspace of dimension d to contain a vector of weight at least d. Thus every affine subspace in A+A has codimension at least 2^t = √N, which eventually dominates C log |S| + D for all proposed universal real constants C,D. The fixed proof establishes the exact Formal Conjectures answer(False) proposition, contains no sorry, admit, custom axiom, native_decide, or unsafe theorem, and its axiom audit reports only propext, Classical.choice, and Quot.sound. It does not settle the first, qualitative O_K(1) question in Problem 52.

Erdős #692 Part II — maximum-existence formal variantDensity of integers with exactly one divisor in an interval

LeanProofNumber TheoryOPEN
Problem. For fixed $n$, if $\delta_1(n,m)$ is the natural density of integers with exactly one divisor in $(n,m)$, does $\delta_1(n,m)$ attain a maximum as $m$ ranges over integers greater than $n+1$? — Formal Conjectures: Erdős #692 maximum-existence variant
  • 2026-08-12 formal conjectures submission
    Submitted Formal Conjectures PR #4888 with a complete Lean 4 proof of the literal OPEN declaration associated with Part II of Erdős Problem 692. For fixed n, the proof first evaluates the base endpoint δ₁(n,n+2) = 1/(n+1). For all sufficiently large m, it places two disjoint finite blocks of primes inside (n,m) whose reciprocal sums are large. An integer having exactly one divisor in the interval cannot be divisible by primes from both blocks, so the density is bounded by the sum of two prime-avoidance densities and is strictly below the base value. Consequently every global maximizer lies in an explicit finite interval; a maximizing endpoint is then chosen from that finite set. This proves that a maximum is attained. The source problem asks where the maximum occurs, so this is a solved maximum-existence formal variant rather than a solution locating or characterizing a maximizing endpoint. It does not claim a closed formula or uniqueness. The fixed proof contains no sorry, admit, or custom axiom, and its axiom audit reports only propext, Classical.choice, and Quot.sound.

Erdős #319 — missing optimality in the Big-O variantThe “simplest upper bound” reduced to an arbitrary upper bound

Misformalization DetectionLeanProofNumber TheoryOPEN
Problem. The docstring asks for the simplest $g(N)$ such that $c(N)=O(g(N))$, but the Lean declaration checks only that one supplied function is an upper bound; where is simplicity or optimality enforced? — Formal Conjectures: underformalized Erdős #319 Big-O variant
  • 2026-08-12 formal conjectures misformalization report with lean witness
    Detected an underformalization in the OPEN isBigO variant of Erdős Problem 319. Its docstring asks for the simplest asymptotic upper bound, whereas its Lean proposition accepts any supplied function g for which c = O(g). PR #4890 demonstrated the gap with a kernel-checked witness g(N)=N: the IsGreatest hypothesis yields a maximizing set A ⊆ {1,…,N} with c(N)=|A|, so the ambient cardinality bound alone gives c(N)≤N. The signed reciprocal-sum condition is never used. Maintainer mo271 flagged the PR as a potential misformalization, and Issue #4892 was opened with the repository's misformalization label, proposing a matching lower bound or another explicit optimality condition. This is a valid elementary upper bound and a Lean witness to the statement flaw, not a solution of Erdős Problem 319. The fixed proof contains no sorry, admit, or custom axiom; its audit reports only propext, Classical.choice, and Quot.sound.

Erdős #357 — √n lower Big-O formal variantDistinct consecutive interval sums in increasing integer sequences

LeanProofNumber TheoryOPEN
Problem. If $f(n)$ is the maximum length of a strictly increasing sequence in $[1,n]$ whose consecutive interval sums are all distinct, prove the explicit lower-growth answer $\sqrt n=O(f(n))$. — Formal Conjectures: Erdős #357 lower Big-O variant
  • 2026-08-13 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #4918 with a direct, kernel-checked Lean 4 proof for the OPEN lower Big-O answer hole attached to Erdős Problem 357. For r=⌊√n⌋, the proof uses the final r integers n-r+1,…,n. Intervals of different lengths have disjoint ranges of possible sums because r²≤n; intervals of the same length have equal sums only when they have the same starting point. Hence all consecutive interval sums are distinct and ⌊√n⌋≤f(n). The inequalities √n<⌊√n⌋+1≤2⌊√n⌋≤2f(n) then establish √n=O(f(n)) with constant 2. The proof is independent of the stronger Weisenberg lower bound already recorded in Formal Conjectures. Accordingly, this is a direct formal proof of a mathematically nontrivial registered variant, not a claim to a new lower bound or a solution of the central open question f(n)=o(n). It also leaves the upper Big-O, Big-Theta, little-o, infinite-density, reciprocal-sum, and weakly monotone declarations unchanged. The fixed proof contains no sorry, admit, custom axiom, native_decide, or unsafe theorem; its axiom audit reports only propext, Classical.choice, and Quot.sound.

Erdős #357 — three additional lower-growth formal variantsLower growth and strict-versus-weak monotonicity for distinct interval sums

LeanProofNumber TheoryOPEN
Problem. For the strict extremal function $f(n)$ and weakly monotone extremal function $h(n)$, prove the three registered targets $\log n=o(f(n))$, $\sqrt n=O(h(n))$, and $\log n=o(h(n))$. — Formal Conjectures: Erdős #357 additional lower-growth variants
  • 2026-08-15 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #4967 with kernel-checked Lean 4 proofs for three OPEN answer holes attached to Erdős Problem 357. The strict result combines the previously formalized direct bound √n=O(f(n)) with Mathlib's asymptotic relation log n=o(√n) to obtain log n=o(f(n)). For the weakly monotone formulation, the proof observes that admissibility already forces injectivity: singleton intervals with equal values would have equal sums. A monotone injective sequence in a linear order is strictly monotone, so the weak and strict extremal functions satisfy h(n)=f(n). Transferring the strict estimates then gives √n=O(h(n)) and log n=o(h(n)). This answers exactly the three named Formal Conjectures variants; it does not claim to solve Erdős Problem 357's central f(n)=o(n) question or its weakly monotone analogue. The fixed proof has no sorry, admit, custom axiom, native_decide, or unsafe theorem; its axiom audit reports only propext, Classical.choice, and Quot.sound.

Erdős #688 — trivial constant-answer upper-bound misformalizationAsymptotic decay of the extremal covering exponent

Misformalization DetectionLeanProofNumber TheoryOPEN
Problem. The source asks for an upper estimate on $\epsilon_n$, in particular whether $\epsilon_n=o(1)$, but the Lean declaration accepts any $g$ satisfying $\epsilon=O(g)$. Why does the constant answer $g(n)=1$ count? — Formal Conjectures: underformalized Erdős #688 upper-bound variant
  • 2026-08-13 formal conjectures misformalization report with lean witness
    Opened Formal Conjectures issue #4919, with the repository's misformalization label, to show that the OPEN upper-bound declaration for Erdős Problem 688 admits the trivial constant answer g(n)=1. For n≥2, applying a feasible covering to m=1 yields a prime with n^ε<p≤n; strict monotonicity of real powers gives ε<1. If the feasible exponent set is nonempty, exponent zero is feasible because every witness prime exceeds 1; if it is empty, Lean's totalized supremum is sSup ∅=0. Thus 0≤epsilonFunction(n)≤1 eventually, and the linked Lean4Web witness proves epsilonFunction=O(1). This bound neither estimates the decay or sharp order of epsilonFunction nor proves the separate conjecture epsilonFunction=o(1). The report therefore treats the proof as a concrete witness to a statement-design flaw, not as a solution of Erdős Problem 688, and proposes either fixing a meaningful target, explicitly requiring decay, or adding a warning that trivial answer instantiations must not trigger an open-to-solved status change.

Erdős #142 — trivial linear-answer upper-bound misformalizationUpper estimates for progression-free subsets of {1,…,N}

Misformalization DetectionLeanProofNumber TheoryOPEN
Problem. The source asks for useful upper estimates or an asymptotic formula for $r_k(N)$, but the Lean declaration accepts any $g$ satisfying $r_k=O(g)$. Why does the ambient-set bound $g(N)=N$ count? — Formal Conjectures: underformalized Erdős #142 upper-bound variant
  • 2026-08-13 formal conjectures misformalization report with lean witness
    Opened Formal Conjectures issue #4922, with the repository's misformalization label, to show that the OPEN one-sided upper-bound declaration for Erdős Problem 142 admits the trivial identity answer g(N)=N. Every admissible progression-free witness S is a subset of Finset.Icc 1 N, so |S|≤|Finset.Icc 1 N|=N; taking the supremal cardinality gives r(k,N)≤N for every k,N, and the linked kernel-checked Lean witness proves r_k=O(N) with Big-O constant 1. This uses no information about arithmetic-progression avoidance and supplies no nontrivial upper bound, order estimate, or asymptotic formula. The report therefore presents the proof as a concrete witness to a statement-design flaw, not as a solution of Erdős Problem 142. It proposes replacing the answer hole with fixed meaningful targets, adding an explicit warning against trivial answer instantiations, or encoding a genuinely sublinear requirement with the necessary condition on k. The issue also records that the standalone proof contains no sorry, admit, native_decide, or custom axiom, and its audit reports only propext, Classical.choice, and Quot.sound.

OEIS A237271 — square and hexagonal parity conjecturesParity of the number of parts in symmetric divisor-sum representations

LeanProofLiteratureNumber TheoryOPEN
Problem. For every positive $n$, prove that both $a(n^2)$ and $a(n(2n-1))$ are odd, where $a(N)$ counts the parts in the symmetric representation of $\sigma(N)$. — Formal Conjectures: OEIS A237271 parity conjectures
  • 2026-08-13 formal conjectures literature-backed lean proof submission
    Submitted Formal Conjectures PR #4924 with kernel-checked Lean 4 proofs of the two OPEN parity declarations for OEIS A237271: a(n²) is odd and a(n(2n-1)) is odd for every positive n. The Formal Conjectures definition writes a(N) as one plus the number of adjacent divisor pairs (d,e) for which e is odd and 2d≤e. On adjacent pairs this condition is equivalent to the strict gap 2d<e; divisor complementation sends (d,e) to (N/e,N/d), preserves the gap, reverses the sorted divisor list, and is an involution. A fixed pair would satisfy de=N. For a square or a hexagonal number, the divisor n would then lie strictly between the paired divisors, contradicting adjacency. Thus the counted pairs occur in two-element orbits, their number is even, and the initial one makes a(N) odd. OEIS A237271 has recorded since 2021 that the parity of the sequence is the characteristic function of numbers with middle divisors; the contribution is a direct, self-contained Lean bridge to the current Formal Conjectures definitions, not a claim of a new mathematical discovery. It does not address the separate Carmichael-number declaration. The fixed proofs contain no sorry, admit, custom axiom, native_decide, or unsafe theorem; both axiom audits report only propext, Classical.choice, and Quot.sound. Formal Conjectures collaborator mo271 approved the submission on its first review, congratulated the resolution of the conjectures, enabled auto-merge, and placed the PR directly into the merge queue.
  • 2026-08-13 first-review acceptance and automatic merge queue
    Formal Conjectures collaborator mo271 approved PR #4924 on the first review, writing “congrats on resolving the conjectures”. No revision was requested. The reviewer then enabled auto-merge and placed the PR directly into the merge queue.

OEIS A287616 — known-solution status correctionUniversal sum of triangular, pentagonal, and heptagonal numbers

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Every nonnegative integer is representable as $x(x+1)/2+y(3y+1)/2+z(5z+1)/2$ for nonnegative integers $x,y,z$. — Formal Conjectures: OEIS A287616
  • 2026-08-13 formal conjectures literature-status report
    Reported in issue #4927 that Cao, Guo, Qiu, Feng, and Gao's 2026 paper Every Nonnegative Integer Is a Sum of a Triangular, a Pentagonal, and a Heptagonal Number settles the exact statement encoded by OeisA287616.conjecture. The paper's abstract explicitly says that it settles OEIS A287616, so the requested upstream action is to replace the open status with research solved and cite the result. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Green Problem 3 — known-solution status correctionProduct-free open subsets of the unit interval

LiteratureStatusCorrectionAdditive CombinatoricsOPEN
Problem. Every open $A\subseteq[0,1]$ of Lebesgue measure greater than $1/3$ contains $x,y,z\in A$ satisfying $xy=z$. — Formal Conjectures: Green Problem 3
  • 2026-08-13 formal conjectures literature-status report
    Reported in issue #4927 that Franchi, Gowers, and Yip's 2026 paper Product-free subsets of (0,1) answers Green Problem 3 affirmatively: an open product-free subset of the unit interval has measure at most $1/3$. This is equivalent to the positive answer(True) requested by the Formal Conjectures declaration. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Green Problem 31 — two upper-bound status correctionsImproved upper bound for finite Sidon sets

LiteratureStatusCorrectionAdditive CombinatoricsOPEN
Problem. Improve the coefficient $0.98183$ in $F(N)\le\sqrt N+0.98183N^{1/4}+O(1)$, both infinitely often and eventually. — Formal Conjectures: Green Problem 31 upper bounds
  • 2026-08-13 formal conjectures literature-status report
    Reported in issue #4927 that Hou and Zhao's 2026 paper Vector-valued smoothing for finite Sidon sets proves $F(N)\le\sqrt N+0.9435N^{1/4}+O(1)$. Since $0.9435<0.98183$ and the theorem holds eventually, it supplies a witness for the eventual declaration and hence also for the weaker infinitely-often declaration. The requested correction is to mark both upper-bound targets solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Open Quantum Problem 35 — AME(7,6) and AME(7,10) status correctionExistence of seven-party absolutely maximally entangled states

LiteratureStatusCorrectionQuantum InformationOPEN
Problem. Do absolutely maximally entangled states AME(7,6) and AME(7,10) exist? — Formal Conjectures: Open Quantum Problem 35 AME(7,d) cases
  • 2026-08-13 formal conjectures literature-status report
    Reported in issue #4927 that Shi, Zhang, Zhao, and Li's 2026 paper Complete Existence Classification of Seven-Partite Absolutely Maximally Entangled States proves that AME(7,d) exists exactly when $d\ge3$. The theorem directly gives positive answers to the Formal Conjectures benchmark cases AME(7,6) and AME(7,10). The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Open Quantum Problem 35 — AME(12,5) status correctionExistence of an absolutely maximally entangled state on twelve ququints

LiteratureStatusCorrectionQuantum InformationOPEN
Problem. Does an absolutely maximally entangled state AME(12,5) exist? — Formal Conjectures: Open Quantum Problem 35 AME(12,5)
  • 2026-08-13 formal conjectures literature-status report
    Reported in issue #4927 that Bevins and Bidav's 2026 paper Symmetry-guided constructions of absolutely maximally entangled states in five open cases constructs a Hermitian self-dual $[12,6,7]_{25}$ maximum-distance-separable code. Its stabilizer construction proves the existence of AME(12,5), giving the positive answer requested by the Formal Conjectures benchmark. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Independent Domination — even and odd bound status correctionIndependent domination number for bounded-degree graphs

LiteratureStatusCorrectionGraph TheoryOPEN
Problem. For isolate-free graphs, prove the displayed even-degree bound $(D+2)^2i(G)\le(D^2+4)n$ and odd-degree bound $(D+1)(D+3)i(G)\le(D^2+3)n$. — Formal Conjectures: Independent Domination even and odd cases
  • 2026-08-13 formal conjectures literature-status report
    Reported in issue #4927 that Cho, Kim, Kim, and Oum's paper Independent domination of graphs with bounded maximum degree, published in JCTB in 2023, supplies the even- and odd-maximum-degree bounds displayed in the two open Formal Conjectures declarations. The requested correction is to cite Corollary 1.3 and mark both declarations solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

MathOverflow #31809 — counterexample status correctionPre-triangulated categories that are not triangulated

LiteratureStatusCorrectionCategory TheoryOPEN
Problem. Must every pre-triangulated category in the stated setting be triangulated? — Formal Conjectures: MathOverflow 31809
  • 2026-08-13 formal conjectures literature-status report
    Reported in issue #4927 that Chen, Liu, Lu, and Zhang's 2026 paper A pre-triangulated category which is not triangulated constructs an explicit counterexample using finitely generated projective modules over the type-$A_5$ preprojective algebra over $\mathbb F_2$. This falsifies the universal proposition encoded in the Formal Conjectures declaration, so the proposed answer is False. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Green Problem 19 — internally implied bound status correctionCorner-density exponent lower and upper bounds

LiteratureStatusCorrectionAdditive CombinatoricsOPEN
Problem. Prove the recorded bounds $C\ge3.13$ and $C\le4$ for the corner-density exponent. — Formal Conjectures: Green Problem 19 bounds
  • 2026-08-13 formal conjectures literature-status report
    Reported in issue #4927 that Green19.green_19 is already tagged research solved and states the exact identity $C=4$, while the immediately following weaker declarations $C\ge3.13$ and $C\le4$ remain tagged open. Both inequalities are direct consequences of the recorded exact value, so the report proposes correcting their status without making a new mathematical claim. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #272 — main-asymptotic status correctionMaximum size of families with arithmetic-progression intersections

LiteratureStatusCorrectionAdditive CombinatoricsOPEN
Problem. Determine the main asymptotic of the largest family of subsets of $[N]$ whose pairwise intersections are nonempty arithmetic progressions. — Formal Conjectures: Erdős Problem 272
  • 2026-08-13 formal conjectures literature-status report
    Reported in issue #4927 that the same Formal Conjectures file already tags Szabó's bound $t(N)=N^2/2+O(N^{5/3}\log^3N)$ as solved. Since the error term is $o(N^2)$, this directly implies $t(N)\sim N^2/2$, supplying the answer function for the still-open main asymptotic declaration. The stronger open question asking for an $O(N)$ error term is separate and is not claimed to be solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Green Problem 37 — sublinear upper boundSparse sets containing a k-term progression of every difference up to N

LeanProofAdditive CombinatoricsOPEN
Problem. For each fixed $k$, let $m(N,k)$ be the minimum size of a set of natural numbers that contains a $k$-term arithmetic progression with common difference $d$ for every $1\le d\le N$. Prove $m(N,k)=o(N)$, and hence $m(N,k)=O(N)$. — Formal Conjectures: Green Problem 37 qualitative upper bounds
  • 2026-08-14 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #4943 with a complete Lean 4 proof of the two qualitative upper-bound declarations for Green Problem 37, using the explicit answer function N ↦ N. For a fixed progression length k, the construction chooses consecutive Fermat numbers as pairwise-coprime moduli. The Chinese remainder theorem assigns, for every difference d≤N, a start point whose jth progression term is divisible by the jth modulus. Taking the union of these sparse periodic residue classes inside a finite interval therefore supplies all required progressions. A cardinality bound makes the linear coefficient arbitrarily small by moving to sufficiently large Fermat moduli, proving $m(N,k)=o(N)$; the registered $O(N)$ theorem follows immediately. The fixed proof contains no sorryAx; its axiom audit reports only propext, Classical.choice, and Quot.sound. This closes only green_37_littleO and green_37_bigO. It does not determine the exact minimum, eventual formula, or Theta class, all of which remain open.

Erdős #361 — linear asymptotic variantLargest subset of [1,⌊cn⌋] avoiding n as a subset sum

LeanProofAdditive CombinatoricsNumber TheoryOPEN
Problem. For fixed $c>0$, determine the order of growth of the maximum size of a set $A\subseteq\{1,\ldots,\lfloor cn floor\}$ for which no subset of $A$ sums to $n$. — Formal Conjectures: Erdős #361 asymptotic variant
  • 2026-08-14 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #4944 with a complete Lean 4 proof of the asymptotic variant of Erdős Problem 361, filling the answer with n ↦ n. The upper bound is the ambient-set estimate $|A|\le\lfloor cn floor$. For the lower bound, after choosing an integer k large enough in terms of c, the proof takes a block of consecutive integers immediately above $n/(k+1)$ and of length comparable to $n/(4k(k+1))$. A subset with at most k elements has sum below $n$, while one with at least k+1 elements has sum above $n$; hence the entire block avoids $n$ as a subset sum and gives a positive linear lower bound. Together these estimates prove that the extremal cardinality is $\Theta(n)$ for every fixed $c>0$. The fixed proof contains no sorryAx; its axiom audit reports only propext, Classical.choice, and Quot.sound. This solves only the registered asymptotic variant; it does not determine the exact extremal function or address its possible irregular dependence on $n$.

Poisson n-Lie Conjecture 3.5 — scalar-matrix bracketDeterminant brackets from scalar matrices and commuting derivations

LeanProofNonassociative AlgebraStatementOPEN
Problem. Given a field $F$, a commutative $F$-algebra $A$, $n+m$ pairwise commuting derivations, and a scalar matrix $M\in M_{n+m,m}(F)$, prove that the determinant bracket formed from the derivative columns and the columns of $M$ satisfies the Poisson $n$-Lie identities encoded in Formal Conjectures PR #4893. — Formal Conjectures PR #4893: proposed Poisson n-Lie target
  • 2026-08-14 independent lean proof publication
    Published a complete Lean 4 proof of the exact general target proposed in Formal Conjectures PR #4893 for the scalar-matrix determinant bracket associated with pairwise commuting derivations. The proof first establishes a stronger theorem for arbitrary natural numbers $n$ and $m$, so the registered hypotheses $n\ge2$ and $m\ge1$ are not needed. Alternation and the Leibniz rule follow from determinant identities. For the Filippov identity, the proof expresses the relevant Hamiltonian operator through adjugate cofactors, proves a Piola-type divergence cancellation, and cancels the mixed second-derivative terms using commutativity of the derivations and determinant antisymmetry. The fixed proof contains no sorry, admit, custom axiom, native_decide, or unsafe theorem; both final declarations report only propext, Classical.choice, and Quot.sound. The exact Lean target is proved, but the source correspondence still needs review: the proposed IsPoissonNLie predicate does not state $F$-multilinearity, while the paper's Conjecture 3.5 cites additional assumptions that are absent from the Lean target. Accordingly, this record does not yet claim that every semantic detail of the paper conjecture has been captured.

OEIS A113019 — third fixed point 387420489 = 9⁹Fixed points of the digit-length–digital-root power map

LeanProofCounterexampleNumber TheorySOLVED
  • 🟢 Status when posted OPEN — Formal Conjectures source when submitted marks OeisA113019.conjecture as research open with an unknown Boolean answer; the OEIS entry itself asks whether 1 and 32 have any other fixed-point companions.
  • 🔵 Current status SOLVED — OEIS A113019 revision #24 was approved and published on 2026-08-15; OEIS editor Alois P. Heinz approved the revision, and the public A113019 entry now states that 9^9 = 387420489 is also a fixed point, credited to Kenta Kitamura. This publishes the counterexample on OEIS; Formal Conjectures PR #4953 remains open separately.
  • 🔗 Source problem Formal Conjectures: OEIS A113019 fixed-point conjecture
Problem. For $a(n)=(\text{number of decimal digits of }n)^{\operatorname{digitalRoot}(n)}$, are $1$ and $32$ the only fixed points? — Formal Conjectures: OEIS A113019 fixed-point conjecture
  • 2026-08-15 formal conjectures counterexample submission
    Submitted Formal Conjectures PR #4953 with a concrete counterexample to the OPEN fixed-point question in OEIS A113019. For n = 387420489 = 9⁹, the decimal representation has nine digits and digit sum 45, hence digital root 9; therefore the sequence definition gives a(n) = 9⁹ = n. This is a third fixed point beyond 1 and 32, so the proposed classification is false. The Lean file proves both the explicit fixed-point identity and the exact Formal Conjectures target with answer(False). The proof uses simplification of the registered definition followed by arithmetic contradiction, contains no sorry or admit, and its axiom audit reports only propext, Classical.choice, and Quot.sound.
  • 2026-08-15 oeis update submission
    Submitted OEIS draft edit #18 adding the explicit fixed point 387420489 = 9⁹, the nine-digit and digital-root calculation, and attribution to Kenta Kitamura. Draft edit #19 then changed the draft status from editing to proposed. This records submission to OEIS, not editorial approval or publication.
  • 2026-08-15 oeis publication
    After editorial compression of the submitted explanation, OEIS revision #24 was approved by Alois P. Heinz. The public entry now includes the comment “9^9 = 387420489 is also a fixed point,” with attribution to Kenta Kitamura, thereby publishing the counterexample and resolving the displayed fixed-point question. Formal Conjectures PR #4953 remains open as the separate Lean-status update.

OEIS A100478 — eventual periodicity from arbitrary nonnegative initial valuesPrime-counting Pentanacci recurrence from arbitrary initial data

LeanProofNumber TheoryOPEN
Problem. Given any five positive initial values, prove that the recurrence $a_{n+5}=\pi(a_n+a_{n+1}+a_{n+2}+a_{n+3}+a_{n+4})$ is eventually periodic. — Formal Conjectures: OEIS A100478 eventual-periodicity target
  • 2026-08-15 formal conjectures proof submission
    Submitted Formal Conjectures PR #4954 with a complete Lean 4 proof of the OPEN eventual-periodicity target associated with OEIS A100478. For initial maximum B, the proof sets t = B + 123 and K = 6006t. An elementary residue-class estimate for the prime-counting function gives π(30030t) ≤ 6006t, hence π(5K) ≤ K. Strong induction then shows that every term of the recurrence is at most K: whenever five consecutive terms lie below K, their sum is at most 5K and so is the next term. The five-term state therefore ranges over a finite set. Two states must coincide, and determinism of the recurrence propagates that equality to yield a positive eventual period. The core theorem is stronger than the registered target: it proves eventual periodicity for every nonnegative initial five-tuple, so the FC theorem does not use its positivity hypothesis. It does not claim that every orbit is eventually constant or classify the possible periods or constant values. The fixed proof contains no sorry, admit, custom axiom, native_decide, or unsafe theorem; its final axiom audit reports only propext, Classical.choice, and Quot.sound.
  • 2026-08-15 oeis update submission
    Submitted OEIS draft edit #22 recording both the original all-ones orbit and the general theorem. Since a(54)=⋯=a(58)=66 and π(330)=66, the published sequence is constant at 66 from index 54 onward. More generally, the linked Lean proof establishes eventual periodicity for every five nonnegative initial values, ruling out the nonperiodic “third pattern” while leaving open whether every such orbit is eventually constant. The draft also adds the fixed-commit proof link and attribution to Kenta Kitamura. Draft edit #23 changed the state from editing to proposed; this records submission, not editorial approval or publication.

OEIS A108306 — general INVERT-transform and matrix-power identityINVERT transforms and powers of a 2×2 matrix

LeanProofNumber TheoryOPEN
Problem. For arbitrary $a,b,n\in\mathbb N$, prove that the $n$-th term of the INVERT transform of $(1,a,ab,ab^2,ab^3,\ldots)$ equals the upper-left entry of $\begin{pmatrix}1&a\\1&b\end{pmatrix}^{n}$. — Formal Conjectures: OEIS A108306 INVERT–matrix conjecture
  • 2026-08-15 formal conjectures proof submission
    Submitted Formal Conjectures PR #4957 with a complete Lean 4 proof of the OPEN general identity attached to OEIS A108306. Writing D n for the INVERT transform, the proof introduces an auxiliary tail state T n and derives the simultaneous recurrences D(n+1)=D(n)+a·T(n) and T(n+1)=D(n)+b·T(n). These are exactly multiplication by the matrix [[1,a],[1,b]]. A simultaneous induction from the initial state (1,0) then identifies D n with the upper-left entry of the nth matrix power for every natural a, b, and n, proving the exact Formal Conjectures target. The proof is kernel checked and contains no sorry, admit, custom axiom, native_decide, or unsafe theorem; its axiom audit reports only propext, Classical.choice, and Quot.sound. The separate textbook theorem for the specialized sequence a=5, b=2 is outside this proof's claimed scope.

OEIS A105801 — Fibonacci–Collatz sequence stabilizes modulo every 3ᵏ3-adic stabilization of the Fibonacci–Collatz sequence

LeanProofNumber TheoryOPEN
Problem. For the Fibonacci–Collatz sequence $a(1)=1$, $a(2)=2$, with the next term equal to half the sum when it is even and three times the sum plus one when it is odd, prove that for every $k>0$ the sequence is eventually constant modulo $3^k$. — Formal Conjectures: OEIS A105801 modular-stabilization conjecture
  • 2026-08-15 formal conjectures proof submission
    Submitted Formal Conjectures PR #4959 with a complete Lean 4 proof of the OPEN conjecture attached to OEIS A105801. The proof transforms the sequence by y(n)=5a(n)+1. On consecutive transformed values, an even branch replaces the pair by its averaging step, whereas an odd branch triples their sum. Repeated averaging strictly decreases the gap between unequal same-parity entries until mixed parity is reached. From an odd–even pair, two consecutive triple branches produce three times another positive unequal pair. Since odd scaling commutes with both branches, this pumping argument iterates to reach a consecutive state divisible by 3^k for every positive k; that divisibility is then preserved by every later step. Thus 3^k eventually divides 5a(n)+1. Coprimality of 5 and 3^k forces all sufficiently late terms to have the same residue modulo 3^k, which is the exact Formal Conjectures target. The proof is kernel checked and contains no sorry, admit, custom axiom, native_decide, or unsafe theorem; its axiom audit reports only propext, Classical.choice, and Quot.sound. The same pumping theorem also applies mathematically to the related A181717 initial condition, but only the registered A105801 target is claimed as formalized here.

OEIS A112970 — three identities on dyadic rays of a generalized Stern sequenceDyadic rays in a generalized Stern sequence

LeanProofNumber TheoryOPEN
Problem. For the generalized Stern sequence satisfying $a(2r+1)=a(r)$, prove for every $n\in\mathbb N$ that $a(2^n)=a(2^{n+1}+1)$, $a(2^n-1)=a(3\cdot2^n-1)$, and $a(2^n-1)=1$. — Formal Conjectures: three OEIS A112970 conjectures
  • 2026-08-15 formal conjectures proof submission
    Submitted Formal Conjectures PR #4960 with complete Lean 4 proofs of all three OPEN targets attached to OEIS A112970. The proof first establishes the odd recurrence a(2r+1)=a(r) for the registered generalized Stern sequence, then iterates it to prove the stronger dyadic-ray lemma a(c·2^n−1)=a(c−1) for every c≥1 and natural n. The specializations c=1 and c=3 yield a(2^n−1)=1 and a(2^n−1)=a(3·2^n−1); a direct odd-recurrence step proves a(2^n)=a(2^(n+1)+1). The even recurrence is not needed. These are exactly the three Formal Conjectures declarations, but the separate OEIS identification a(2^n)=A033638(n) is not claimed as formalized. The proof is kernel checked and contains no sorry, admit, custom axiom, native_decide, or unsafe theorem; all three axiom audits report only propext, Classical.choice, and Quot.sound.

OEIS A113250 — every odd-indexed term in the m=4 recurrence is a squareSquare terms in the m=4 specialization of a fourth-order recurrence family

LeanProofNumber TheoryOPEN
Problem. For the fourth-order integer recurrence corresponding to $m=4$ in the OEIS A113249 family, prove that every odd-indexed term $a(2n+1)$ is a perfect square. — Formal Conjectures: OEIS A113250 square-term conjecture
  • 2026-08-15 formal conjectures proof submission
    Submitted Formal Conjectures PR #4963 with a complete Lean 4 proof of the OPEN square-term conjecture for OEIS A113250, the m=4 specialization of the A113249 recurrence family. Setting p=m²=16, the shared proof defines a second-order sequence Y and proves the exact identity A(2n+1)=Y(n)². The odd subsequence of the fourth-order recurrence and the square sequence Y(n)² satisfy the same third-order linear recurrence, and their first three values agree; induction therefore proves that every odd-indexed term is an integer square. Specializing the parameterized theorem at p=16 proves the exact OeisA113250.conjecture target. A113250, A113252, and A113255 are recorded separately because they are distinct OEIS and Formal Conjectures entries, even though one family theorem proves all three. The proof is kernel checked and contains no sorry, admit, custom axiom, native_decide, or unsafe theorem; its axiom audit reports only propext, Classical.choice, and Quot.sound.

OEIS A113252 — every odd-indexed term in the m=6 recurrence is a squareSquare terms in the m=6 specialization of a fourth-order recurrence family

LeanProofNumber TheoryOPEN
Problem. For the fourth-order integer recurrence corresponding to $m=6$ in the OEIS A113249 family, prove that every odd-indexed term $a(2n+1)$ is a perfect square. — Formal Conjectures: OEIS A113252 square-term conjecture
  • 2026-08-15 formal conjectures proof submission
    Submitted Formal Conjectures PR #4963 with a complete Lean 4 proof of the OPEN square-term conjecture for OEIS A113252, the m=6 specialization of the A113249 recurrence family. Setting p=m²=36, the shared proof defines a second-order sequence Y and proves the exact identity A(2n+1)=Y(n)². The odd subsequence of the fourth-order recurrence and the square sequence Y(n)² satisfy the same third-order linear recurrence, and their first three values agree; induction therefore proves that every odd-indexed term is an integer square. Specializing the parameterized theorem at p=36 proves the exact OeisA113252.conjecture target. A113250, A113252, and A113255 are recorded separately because they are distinct OEIS and Formal Conjectures entries, even though one family theorem proves all three. The proof is kernel checked and contains no sorry, admit, custom axiom, native_decide, or unsafe theorem; its axiom audit reports only propext, Classical.choice, and Quot.sound.

OEIS A113255 — every odd-indexed term in the m=9 recurrence is a squareSquare terms in the m=9 specialization of a fourth-order recurrence family

LeanProofNumber TheoryOPEN
Problem. For the fourth-order integer recurrence corresponding to $m=9$ in the OEIS A113249 family, prove that every odd-indexed term $a(2n+1)$ is a perfect square. — Formal Conjectures: OEIS A113255 square-term conjecture
  • 2026-08-15 formal conjectures proof submission
    Submitted Formal Conjectures PR #4963 with a complete Lean 4 proof of the OPEN square-term conjecture for OEIS A113255, the m=9 specialization of the A113249 recurrence family. Setting p=m²=81, the shared proof defines a second-order sequence Y and proves the exact identity A(2n+1)=Y(n)². The odd subsequence of the fourth-order recurrence and the square sequence Y(n)² satisfy the same third-order linear recurrence, and their first three values agree; induction therefore proves that every odd-indexed term is an integer square. Specializing the parameterized theorem at p=81 proves the exact OeisA113255.conjecture target. A113250, A113252, and A113255 are recorded separately because they are distinct OEIS and Formal Conjectures entries, even though one family theorem proves all three. The proof is kernel checked and contains no sorry, admit, custom axiom, native_decide, or unsafe theorem; its axiom audit reports only propext, Classical.choice, and Quot.sound.

OEIS A103425 — existence of a prime-free weighted Tribonacci sequencePrime-free third-order linear recurrences

LeanProofNumber TheoryOPEN
Problem. Do there exist relatively prime integer coefficients $(a,b,c)$ and an integer sequence satisfying $x_{n+3}=ax_{n+2}+bx_{n+1}+cx_n$ such that no $|x_n|$ is prime? — Formal Conjectures: OEIS A103425 prime-free recurrence question
  • 2026-08-15 formal conjectures proof submission
    Submitted Formal Conjectures PR #4964 with an explicit Lean 4 witness answering the OPEN OEIS A103425 existence question affirmatively. Take coefficients (a,b,c)=(1,1,−1) and the constant sequence x(n)=4. The coefficients are relatively prime, the recurrence holds because 4=4+4−4, and every absolute term is 4, hence nonprime. This proves the exact registered statement with answer(True). The source and Formal Conjectures target do not require the sequence to be nonconstant, increasing, unbounded, or genuinely of minimal order three; this record therefore does not claim a proof of any such stronger nondegenerate variant. It also does not claim historical priority: Ivan Lunev published a stronger positive, nonconstant prime-free Tribonacci-like construction in 2017. The fixed Lean proof contains no sorry, admit, custom axiom, native_decide, or unsafe theorem, and its axiom audit reports only propext and Quot.sound.

OEIS A114831 — convergence of the consecutive-term ratio to √3Asymptotics of a harmonic-mean recurrence

LeanProofOEISNumber TheoryOPEN
Problem. Let $a(1)=1$, $a(2)=2$, and $a(n+1)=a(n)+\lfloor 2a(n)a(n-1)/(a(n)+a(n-1))\rfloor$. Prove that $a(n+1)/a(n)\to\sqrt3$. — Formal Conjectures: OEIS A114831 asymptotic conjecture
  • 2026-08-15 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #4969 with a kernel-checked Lean 4 proof of the OPEN asymptotic target for OEIS A114831. Writing r(n)=a(n+1)/a(n) and F(x)=1+2/(x+1), the floored recurrence gives r(n+1)=F(r(n))−e(n), where the nonnegative floor error tends to zero because the positive increasing sequence satisfies a(n)≥n. All relevant ratios lie in [1,3]; on this range, F has the positive fixed point √3 and contracts distance to it by a factor at most 1/2. A perturbed-contraction argument therefore proves r(n)→√3, exactly matching OeisA114831.conjecture3. This resolves the registered ratio-limit theorem but does not claim a closed formula or a sharper asymptotic expansion for a(n). The fixed proof has no sorry, admit, custom axiom, native_decide, or unsafe theorem; its axiom audit reports only propext, Classical.choice, and Quot.sound.

OEIS A102371 — binary-complement formula using A105033Sloping binary numbers, bitwise carries, and XOR recurrence

LeanProofOEISNumber TheoryOPEN
Problem. For every $n\ge1$, prove that the OEIS A102371 term satisfies $a(n)=2^n-1-\operatorname{A105033}(n-1)$. — Formal Conjectures: OEIS A102371 identity
  • 2026-08-15 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #4970 with a kernel-checked Lean 4 proof answering the OPEN OEIS A102371 question affirmatively for every n≥1. Let R(n) be the finite bit-selected sum occurring in A105033 and define the sloping and carry parts by S(n)=n−R(n) and C(n)=R(n)+1. The proof establishes S(n)+C(n)=n+1, C(n+1)=2·(S(n) AND C(n))+1, and S(n+1)=S(n) XOR C(n); the last identity uses the binary full-adder relation (x XOR y)+2·(x AND y)=x+y. It then proves a fixed-width complement/XOR lemma and applies it inductively to the recurrence defining A102371. This yields a(n)=2^n−1−A105033(n−1), exactly the formula asked in the OEIS comment and registered in OeisA102371.conjecture. No additional generalization about either sequence is claimed. The fixed proof has no sorry, admit, custom axiom, native_decide, or unsafe theorem; its axiom audit reports only propext, Classical.choice, and Quot.sound.

OEIS A102722 — asymptotic with the Euler–Mascheroni constantFractional-part sums and the Dirichlet divisor problem

LeanProofOEISNumber TheoryOPEN
Problem. Let $a(n)=\left\lfloor\sum_{k=1}^{n}\{n/k\}\right\rfloor$. Prove that $a(n)\sim(1-\gamma)n$, where $\gamma$ is the Euler–Mascheroni constant. — Formal Conjectures: OEIS A102722 asymptotic conjecture
  • 2026-08-15 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #4971 with a kernel-checked Lean 4 proof of the OPEN OEIS A102722 asymptotic conjecture, with no additional hypothesis. Write S(n)=∑_{k≤n}{n/k}, D(n)=∑_{k≤n}⌊n/k⌋, and r=⌊√n⌋. The proof first establishes the exact identity S(n)=nHₙ−D(n) and formalizes Dirichlet's hyperbola identity D(n)=2∑_{k≤r}⌊n/k⌋−r² by counting lattice points with product at most n. Bounding each floor error by 1 controls the truncated divisor sum within r of nHᵣ. Mathlib's harmonic-number limit together with r/n→0, r²/n→1, and 2 log r−log n→0 then gives D(n)/n−log n→2γ−1 and S(n)/n→1−γ. Finally, taking the outer floor changes the normalized value by at most 1/n, so a(n)/n→1−γ; since 1−γ≠0, this is converted to the exact asymptotic equivalence a(n)~(1−γ)n. The fixed proof has no sorry, admit, custom axiom, native_decide, or unsafe theorem; its axiom audit reports only propext, Classical.choice, and Quot.sound.

OEIS A112521 — equality with the main diagonal of a recursive arrayNOR bracketings, recursive arrays, and WZ telescoping

LeanProofOEISCombinatoricsOPEN
Problem. Let $T(1,1)=1$, let $T(i,j)=0$ outside the positive quadrant, and define $T$ by the stated five-term recurrence. Prove that $a(n)=T(n,n)$ for every $n\ge1$. — Formal Conjectures: OEIS A112521 diagonal conjecture
  • 2026-08-15 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #4972 with a complete, kernel-checked Lean 4 proof of the OPEN OEIS A112521 main-diagonal conjecture. Define S(d)=∑_{0≤j≤d}(−1)^j C(2j,j) C(2d−j,d−j). The proof constructs homogeneous Fibonacci polynomials and applies the linear change v=x+y, u=x−y; the transformed coefficients satisfy exactly the five-term recurrence defining T, which identifies T(d+1,d+1) with S(d). A fully formalized Wilf–Zeilberger telescoping certificate then proves, for n≥3, the recurrence 5n(n+1)S(n+1)=2n(2n+1)S(n)+4(16n²−1)S(n−1). Together with four explicit initial cases, its positive coefficients establish S(d)≥0. Rewriting the signed sum inside a(d+1) as S(d) and using nonnegativity to remove Int.toNat completes a(n)=T(n,n) for every n≥1. All WZ identities and boundary terms are proved in Lean; no unproved bridge assumption is used. The fixed proof has no sorry, admit, custom axiom, native_decide, or unsafe theorem, and its axiom audit reports only propext, Classical.choice, and Quot.sound.

OEIS A211417 — divisibility of an integral factorial ratioFactorial ratios, p-adic valuations, and Landau step functions

LeanProofOEISNumber TheoryOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures source when submitted marks all four displayed divisibility declarations as research open, each with a theorem-body sorry. The separate general_divisibility declaration remains open and is vacuous as written because it permits D=0. Issue #4923 now records an explicit positive witness proposed for the intended repair; its paper proof is reported complete but has not yet been kernel-checked in Lean. The independent supercongruence target also remains open.
  • 🔗 Source problem Formal Conjectures: four OEIS A211417 divisibility conjectures
Problem. For $a(n)=(30n)!n!/((15n)!(10n)!(6n)!)$, prove $(2n+1)\mid7a(n)$, $(3n+1)\mid a(n)$, and $(5n+1)\mid a(n)$; deduce $(2n+1)(3n+1)(5n+1)\mid42a(n)$. — Formal Conjectures: four OEIS A211417 divisibility conjectures
  • 2026-08-16 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #5010 with kernel-checked Lean 4 proofs of four OPEN OEIS A211417 targets. For the three mathematically substantive atomic assertions, the proof expresses the p-adic valuation of the factorial ratio through the nonnegative Landau term F(n,m)=⌊30n/m⌋+⌊n/m⌋−⌊15n/m⌋−⌊10n/m⌋−⌊6n/m⌋ and sums its prime-power contributions using Legendre's formula. General residue lemmas cover all but finitely many exceptional prime powers; the exceptional 2-, 3-, 5-, and 7-primary parts are supplied by binomial-coefficient valuation identities and explicit higher-power Landau contributions. This proves (2n+1)∣7a(n), (3n+1)∣a(n), and (5n+1)∣a(n) for every natural n. The fourth FC target is then derived rather than counted as an independent discovery: gcd(2n+1,3n+1)=1, while the remaining pairwise overlaps divide 3 and 2, so the product divides 42a(n). The fixed proofs contain no sorry, admit, custom axiom, native_decide, or unsafe theorem for these four targets; their axiom audits report only propext, Classical.choice, and Quot.sound. The general divisibility declaration and the separate supercongruence are not claimed solved. A 2026-08-16 update to issue #4923 proposes the positive general witness D(r)=lcm(1,…,r)^|{1≤i≤r : gcd(i,30)=1}|; its proof is paper-complete but is explicitly not yet Lean-checked.

Lₚ Rogers–Shephard Conjecture 5 — equality-case rigidityEquality rigidity for planar centrally symmetric convex bodies

LeanProofConvex GeometryOPEN
Problem. Let $K\subset\mathbb R^2$ be a compact convex body with a centre of symmetry, containing the origin. For $p>1$, equality in the sharp $L_p$ Rogers--Shephard volume bound implies that $K$ is a parallelogram with a vertex at the origin. — Formal Conjectures: arXiv 2607.03582 Conjecture 5
  • 2026-08-16 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #5011 with a complete Lean 4 proof of the OPEN Conjecture 5 from Fradelizi–Manui–Meyer–Ndiaye (2026). The argument develops an endpoint–fiber rigidity theorem. A lower bound for each fiber of the Lₚ sum, together with the assumed equality of total volumes, layer-cake integration, and continuity, forces equality for every interior fiber. Choosing a support point orthogonal to the centre of symmetry then determines the exact fiber intervals along the relevant ray. Their endpoints trace four boundary segments; supporting functionals trap the centered body inside the quadrilateral generated by the symmetry centre and support point. After translation this identifies the original body as a parallelogram having the origin as a vertex. The fixed proof contains no sorry, admit, custom axiom, native_decide, or unsafe theorem, and its axiom audit reports only propext, Classical.choice, and Quot.sound. The linked repository also contains a separate proof of Conjecture 4, but PR #5011 targets Conjecture 5, so this entry counts only the Conjecture 5 contribution.

OEIS A105751 — asymptotic 2-adic valuationGaussian-integer products and exact 2-adic valuations

LeanProofOEISNumber TheoryOPEN
Problem. For $a(n)=\operatorname{Im}\prod_{k=0}^n(1+ki)$, prove $\nu_2(a(n))\sim n/4$ as $n\to\infty$. — Formal Conjectures: OEIS A105751 2-adic conjecture
  • 2026-08-16 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #5013 with a kernel-checked Lean 4 proof of the OPEN 2-adic asymptotic for OEIS A105751. Writing the defining complex product as the image of a Gaussian-integer product gives an exact recurrence for its real and imaginary coordinates. The proof normalizes four-factor blocks, pairs adjacent blocks into even- and odd-start polynomial superblocks, and uses their parity and cross-difference divisibility in a dyadic induction. This yields exact formulas for every residue class: ν₂(a(4m+1))=m, ν₂(a(4m+2))=m, ν₂(a(4m))=m+2+ν₂(⌈m/2⌉), and ν₂(a(4m+3))=m+3+ν₂(⌈m/2⌉) away from the two zero terms. Consequently the error from n/4 is logarithmic; Lean proves the normalized error bound (12+4 log₂ n)/n, shows that it tends to zero, and applies the squeeze theorem to the exact Formal Conjectures target. The fixed theorem has no sorry, admit, custom axiom, native_decide, or unsafe dependency, and its axiom audit reports only propext, Classical.choice, and Quot.sound. The separate conjecture for primes p≡1 (mod 4) is outside this proof and remains open.

OEIS A100474 — the next semiprime after a(11)Certified recurrence evaluation, minimality, and large-prime proof

LeanProofComputationOEISNumber TheoryOPEN
Problem. For the recursively defined sequence $a(n)$ in OEIS A100474, determine the first semiprime term occurring after $a(11)$. — Formal Conjectures: OEIS A100474 next-semiprime question
  • 2026-08-16 formal conjectures exact-answer submission
    Submitted Formal Conjectures PR #5014 with the exact answer to the OPEN A100474 next-semiprime question: the first qualifying term after a(11) is a(36), a 133-digit integer factoring as 251 × p, where p is a 131-digit prime. The Lean development certifies the first 665 primes, evaluates the defining recurrence sequentially through index 36, and excludes every index from 12 through 35 using explicit three-factor decompositions. It proves the large cofactor prime with a recursive Lucas certificate and kernel-checked square-and-multiply modular traces, then establishes that 36 is the least qualifying index and rewrites the exact sInf target. The proof uses neither native_decide nor compiler-trusted evaluation; the final theorem has no sorryAx, Lean.trustCompiler, or Lean.ofReduceBool, and its axiom audit reports only propext, Classical.choice, and Quot.sound. This is a complete certified answer to the semiprime question, not a solution of the separate prime-occurrence conjecture, which remains open.

Erdős #979 — unbounded representations by three prime cubesUnbounded representation counts for sums of three prime cubes

LeanProofNumber TheorySOLVED
  • 🔵 Status when posted SOLVED — Formal Conjectures source when submitted already marks the k=3 variant research solved, crediting an unpublished result of Erdős, while the all-k statement remains research open. PR #5017 adds the new independent kernel-checked proof as its formal_proof; this entry is therefore not counted as a newly solved open problem.
  • 🔗 Source problem Formal Conjectures: Erdős #979 k=3 variant
Problem. If $f_3(n)$ counts representations of $n$ as a sum of three prime cubes, prove $\limsup_{n\to\infty} f_3(n)=\infty$. — Formal Conjectures: Erdős #979 k=3 variant
  • 2026-08-17 independent lean proof registration
    Submitted Formal Conjectures PR #5017 to register a complete Lean 4 proof of the exact k=3 variant of Erdős Problem 979: the limsup of the number of representations of an integer as a sum of three prime cubes is infinite. Formal Conjectures already classified this variant as solved on the strength of an unpublished assertion by Erdős, but no text of his proof is publicly cited; the submitted development is an independent proof, not a reconstruction of the unpublished argument. It follows a modern complex-multiplication route. Point counts for the Fermat cubic are controlled through cubic Jacobi sums; primary Eisenstein integers encode the relevant Hecke angles; lattice-point and thin-annulus estimates produce Dirichlet-series models; and Wiener–Ikehara together with the prime number theorem in the progression 1 modulo 3 yields an unbounded supply of suitable prime triples. An injection from sorted triples to multisets then proves the literal Formal Conjectures target. The modular development spans 52 project modules, and the standalone Lean4Web file is about 25,700 lines. The final axiom audit reports only propext, Classical.choice, and Quot.sound, with no sorryAx. The proof does not settle the all-k form of Problem 979 or give a quantitative growth rate for f₃.

VCDimConvex — degenerate n=0 misformalizationFinite additive VCₙ dimension of convex sets in ℝⁿ⁺¹

Misformalization DetectionLeanProofCounterexampleConvex GeometryREPAIRED; PROOF SUBMITTED
  • 🟢 Status when posted OPEN — Formal Conjectures source when reported marks the all-natural-n declaration research open, but issue #5022 gives a complete Lean disproof of its literal n=0 instance. The suggested repair is to require 0 < n, or 2 ≤ n in line with the neighboring conjecture. The declaration was later repaired by adding 1 ≤ n, and PR #6436 supplies a proof of that repaired positive-n target.
  • 🔵 Current status REPAIRED; PROOF SUBMITTED — Formal Conjectures PR #6436 records that the repaired positive-n target received a proof submission on 2026-09-20; PR #6436 proves the finite bound for every n≥1; it is open and unmerged as of that date.
  • 🔗 Source problem Formal Conjectures: VCDimConvex universal bound
Problem. The declaration quantifies over every $n\in\mathbb N$: for each $n$, is there a finite $d$ bounding the additive $\mathrm{VC}_n$ dimension of every convex set in $\mathbb R^{n+1}$? What happens at $n=0$? — Formal Conjectures: VCDimConvex universal bound
  • 2026-08-18 formal conjectures misformalization report with lean disproof
    Opened Formal Conjectures issue #5022, labeled misformalization, to show that exists_hasAddVCNDimAtMost_n_of_convex_rn_add_one is false because it quantifies over all natural numbers including zero. For an arbitrary proposed bound d, the Lean proof chooses the convex singleton C={0} in Fin 1 → ℝ. Since Fin 0 → Fin (d+1) has only one function, the degenerate zero-parameter setup allows translations of that singleton to realize every required membership pattern; hence HasAddVCNDimAtMost C 0 d fails for every d. The counterexample compiles against Formal Conjectures commit 9943b0a on Lean 4.27.0, and its axiom audit reports only propext, Classical.choice, and Quot.sound. The issue proposes adding 0<n, or the stronger 2≤n used by the adjacent upper-bound conjecture. This is a precise disproof of the then-registered statement, not by itself a solution of the intended positive-dimensional problem. The declaration was subsequently repaired by adding 1≤n; PR #6436 separately supplies a kernel-checked proof of that intended target. The n=0 disproof is independent of PR #4657's n=2, d=1 counterexample.

OEIS A129365 — integrality, prime support, and p-adic valuationsGCD-product ratios and an exact p-adic valuation formula

LeanProofOEISNumber TheoryOPEN
Problem. For the ratio $a(n)$ of the double GCD product to $\prod_{k=1}^n\lfloor n/k\rfloor!^k$, prove its integrality; characterize its prime divisors by $p\le\lfloor n/3\rfloor$; prove that $\operatorname{ord}_p a(np+k)$ is constant for $0\le k<p$; and express that valuation as $\sum_{i\ge0}b(\lfloor n/p^i\rfloor)$ for OEIS A004125's remainder sum $b$. — Formal Conjectures PR #5016: OEIS A129365 targets
  • 2026-08-18 formal conjectures lean proof publication
    Published kernel-checked Lean 4 proofs of all four research open OEIS A129365 targets introduced in Formal Conjectures PR #5016. The proof establishes the master identity vₚ(a(N)) = Σ_{r≥1} b(⌊N/pʳ⌋), where b(m)=Σ_{k=1}^m(m mod k) is OEIS A004125. At each prime-power level, the numerator's double GCD product contributes m², because exactly m=⌊N/pʳ⌋ indices are divisible by pʳ. Legendre's factorial formula turns the denominator contribution into Σ_{k=1}^m k⌊m/k⌋; subtracting and using division with remainder gives m²-Σ k⌊m/k⌋=Σ(m mod k)=b(m). Nonnegativity of this sum proves that the denominator divides the numerator. Since b(m)>0 exactly when m≥3, a prime divides a(n) exactly when p≤⌊n/3⌋. The floor quotients are unchanged throughout each block np,…,np+p-1, proving valuation invariance, and substituting N=np with an index shift gives the exact infinite-sum formula, whose support is finite. Both the Formal Conjectures and standalone Mathlib builds use Lean 4.27.0; the four axiom audits report only propext, Classical.choice, and Quot.sound, with no sorryAx, custom axiom, native_decide, or unsafe theorem. This entry counts as one solved problem family and four proved target statements.

OEIS A003625 — irreducibility of x²+x+2 over finite fieldsQuadratic reciprocity and irreducibility over GF(p)

LeanProofLiteratureOEISNumber TheoryOPEN
Problem. For every prime $p$, prove that $p\equiv3,5,$ or $6\pmod 7$ if and only if $x^2+x+2$ is irreducible over $\operatorname{GF}(p)$. — Formal Conjectures PR #5016: OEIS A003625 target
  • 2026-08-19 formal conjectures lean proof publication
    Published a kernel-checked Lean 4 proof of the exact research open OEIS A003625 target introduced in Formal Conjectures PR #5016. Completing the square gives 4(x²+x+2)=(2x+1)²+7, so for odd primes away from 7, reducibility is equivalent to −7 being a quadratic residue. Quadratic reciprocity and the first supplement cancel their signs to identify the Legendre symbols (−7/p)=(p/7). The nonresidues modulo 7 are precisely 3, 5, and 6, while the exceptional primes 2 and 7 are checked directly and make both sides false. The Formal Conjectures and standalone Mathlib versions both build with Lean 4.27.0; the final axiom audit reports only propext, Classical.choice, and Quot.sound, with no sorryAx, custom axiom, native_decide, or unsafe theorem. Because the equivalence was already stated on OEIS by Federico Provvedi in 2018 and is a classical quadratic-reciprocity consequence, this record carries the Literature tag and is excluded from counts of newly solved open mathematical problems.

Erdős #394 — lower-bound asymptotic notation reversedGrowth of the Erdős–Hall divisor-product function t₂(n)

Misformalization DetectionNumber TheorySOLVED
Problem. The docstring records the known lower bound $\sum_{n\le x}t_2(n)\gg x^2/\log x$. Does the Lean declaration encode that direction, while leaving the Erdős–Hall upper-bound conjecture open? — Formal Conjectures: reversed Erdős #394 lower_bound
  • 2026-08-19 formal conjectures misformalization report
    Opened Formal Conjectures issue #5062, labeled misformalization, after detecting that erdos_394.variants.lower_bound reverses the two sides of the project's ≫ notation. The docstring correctly says Σ_{n≤x}t₂(n) ≫ x²/log x, a trivial known lower bound coming from t₂(p)=p−1 on primes. But since f≫g is defined as g=O(f), the code instead says Σt₂(n)=O(x²/log x), an upper bound. That accidental upper bound is far stronger: because log 2<1, it would imply the file's research open Erdős–Hall statement Σt₂(n)=o(x²/(log x)^c) for every c<log 2, allowing the open conjecture to be closed without new mathematics. The proposed fix simply swaps the arguments of ≫; the corrected lower bound stays research solved, while hall_conjecture meaningfully remains open. No other declaration is affected. This is a statement-direction correction, not a proof of Erdős Problem 394 or the Erdős–Hall conjecture.

OEIS A078590 — first nonintegral term at n=7Integrality of a nonlinear exponential recurrence

LeanProofCounterexampleOEISNumber TheoryOPEN
Problem. Starting with $a(1)=a(2)=1$ and $a(n)=(2^{a(n-1)}+1)/a(n-2)$, are all terms integers? — Formal Conjectures: open OEIS A078590 integrality target
  • 2026-08-21 formal conjectures counterexample submission
    Published a kernel-checked Lean 4 counterexample to the open integrality question for OEIS A078590 and submitted it in Formal Conjectures PR #5078. The first six recurrence terms are integers, with a(5)=171. Rather than expanding the enormous value 2^a(6), the proof computes a(6)≡3 (mod 18) and uses 2^18≡1 (mod 171) to obtain 2^a(6)+1≡9 (mod 171). Thus 171 does not divide the seventh numerator, so the exact Formal Conjectures universal-divisibility target is false at n=7; moreover the gcd is 9, making the seventh rational term's reduced denominator 171/9=19. The PR replaces the false positive proposition by the yes/no statement with answer(False) and marks it research solved. The Formal Conjectures and standalone Mathlib versions use Lean 4.27.0. The axiom audit reports only propext, Classical.choice, and Quot.sound, with no sorryAx, custom axiom, native_decide, or unsafe theorem. This is a complete negative answer to the OEIS question, not merely a formalization of a previously known result.

OEIS A070823 — squarefree-cofactor failure at n=20Decimal-concatenation recurrence and squarefree cofactors

LeanProofCounterexampleOEISNumber TheoryOPEN
Problem. For the recurrence formed by the absolute difference of the two decimal concatenations of consecutive terms, is every $a(n)$ with $n>2$ divisible by $3$ and expressible as $2^j3^k s$ with $s$ squarefree? — Formal Conjectures: open OEIS A070823 squarefree target
  • 2026-08-21 formal conjectures counterexample submission
    Published a kernel-checked Lean 4 counterexample to the open squarefree-cofactor question for OEIS A070823 and submitted it in Formal Conjectures PR #5079. The proof also settles the statement's first conjunct positively: because 10≡1 (mod 3), concatenating x,y or y,x gives the same residue x+y, so their absolute difference is divisible by 3 for every n>2. The second conjunct fails at n=20. Kernel reduction gives the compact certificate a(20) mod 13³ = 1859 = 13²·11, hence 13²∣a(20). Since 13 is coprime to every factor 2^j3^k, any decomposition a(20)=2^j3^k s forces 13²∣s, contradicting the requirement that s be squarefree. The PR keeps the existing single target, replaces its false universal positive form by answer(False), and marks it research solved. The Formal Conjectures and standalone Mathlib versions use Lean 4.27.0, contain no sorry, admit, custom axiom, native_decide, or unsafe theorem, and include explicit axiom audits. This is a complete negative answer to the OEIS question, not merely a formalization of a previously known result.

OEIS A159829 Conjecture 1 — only the prime 2 for odd exponentsPrime values of sums of two like powers

LeanProofCounterexampleOEISNumber TheoryOPEN
Problem. For every integer $k\ge3$, are there infinitely many primes of the form $n^k+m^k$ with $n,m\ge1$? — Formal Conjectures: open OEIS A159829 Conjecture 1
  • 2026-08-21 formal conjectures counterexample submission
    Published a kernel-checked Lean 4 negative answer to OEIS A159829 Conjecture 1 and submitted it in Formal Conjectures PR #5081. The proof establishes the stronger theorem that for every odd k≥3, the set of prime values n^k+m^k with positive n,m is exactly the singleton {2}. Oddness gives the factorization n+m ∣ n^k+m^k. If the sum is prime, the nontrivial divisor n+m must equal the entire sum; but as soon as either input exceeds 1, the inequality n+m<n^k+m^k contradicts that equality. Hence n=m=1 and the value is 2. Specializing to k=3 makes the prime-value set finite and refutes the original universal infinitude assertion. The PR replaces Conjecture 1 by the explicit negative answer answer(False) and marks it research solved, while leaving the separate n^k+m^k+1 Conjecture 2 unchanged and open. The Formal Conjectures and standalone Mathlib versions use Lean 4.27.0; the axiom audits report only propext, Classical.choice, and Quot.sound, with no sorryAx, custom axiom, native_decide, or unsafe theorem. This is a complete negative answer to Conjecture 1, not a claim about the unresolved even-exponent cases or Conjecture 2.

OEIS A185895 Conjecture 3 — prime-power Gauss congruencesGauss congruences for signed distinct-block partition counts

LeanProofOEISCombinatoricsNumber TheoryOPEN
Problem. For the coefficients $a(N)=N![x^N]\prod_{j\ge1}(1-x^j/j!)$, prove for every prime $p$ and positive $n,k$ that $a(np^k)\equiv a(np^{k-1})\pmod{p^k}$. — Formal Conjectures: open OEIS A185895 Conjecture 3
  • 2026-08-21 formal conjectures lean proof submission
    Published a kernel-checked Lean 4 proof of OEIS A185895 Conjecture 3 and submitted it in Formal Conjectures PR #5083. Expanding the defining product and eliminating the rational coefficient followed by floor rewrites a(N) as a signed sum over sets of distinct positive block sizes whose sum is N; each summand is the associated multinomial coefficient. The proof then works coefficientwise in a multivariate polynomial ring over ZMod(p^k). A prime-power freshman's-dream argument shows that, at total degree p·m, a multinomial coefficient vanishes modulo p^k unless every multiplicity is divisible by p; in the surviving case it reduces to the coefficient with all multiplicities divided by p. The surviving finite sets are reindexed by the bijection s↔p·s, which preserves cardinality and therefore the sign. Taking m=n·p^(k−1) yields a(np^k)≡a(np^(k−1)) (mod p^k) for every prime p and positive n,k. The proof uses the general Gauss/Dold mechanism as mathematical background but establishes the exact OEIS target directly. The Formal Conjectures and standalone Mathlib versions use Lean 4.27.0; the final axiom audit reports only propext, Classical.choice, and Quot.sound, with no sorryAx or custom axiom. Conjectures 1 and 2 concern separate sign-change phenomena and are explicitly left open.

OEIS A022030 — recurrence for the alternating ceiling/floor sequenceLinear recurrence forced by an alternating nonlinear rounding process

LeanProofOEISNumber TheoryOPEN
Problem. For the alternating sequence beginning $4,16,63,249,984,\ldots$, prove that for every $n\ge4$, $a(n)=4a(n-1)-a(n-3)+a(n-4)$. — Formal Conjectures: open OEIS A022030 recurrence target
  • 2026-08-21 formal conjectures lean proof submission
    Published a kernel-checked Lean 4 proof of the exact open Formal Conjectures target for OEIS A022030 and submitted it in PR #5085. The proof carefully targets the current alternating definition, whose terms begin 4,16,63,249,984,…: even steps use ceiling minus one and odd steps use floor plus one. It introduces the linear candidate L with the claimed recurrence and proves positivity plus the growth bound 3Lₙ≤Lₙ₊₁. A signed determinant defect Eₙ=(-1)ⁿ(Lₙ₊₁²−LₙLₙ₊₂) satisfies a positive-coefficient order-ten recurrence; checking its initial values and propagating the bounds 0<Eₙ≤Lₙ show that the nonlinear ceiling/floor step is exactly Lₙ₊₂ at every index. Therefore the original sequence equals L, giving a(n)=4a(n−1)−a(n−3)+a(n−4) for all n≥4. This is the alternating Formal Conjectures target, not the older non-alternating sequence also discussed in OEIS comments. The proof repository uses Lean 4.27.0, has no sorry, admit, custom axiom, native_decide, or unsafe declaration, and its axiom audit reports only propext, Classical.choice, and Quot.sound.

OEIS A049473 — zeta-tail inequalities and Beatty difference positionsζ(3) approximation and complementary Beatty sequences

LeanProofOEISNumber TheoryOPEN
Problem. For the nearest-integer sequence $a(n)=\operatorname{round}(n/\sqrt2)$ and $s(m)=\zeta(3)-\sum_{k=1}^m k^{-3}$, prove both $s(a(n))<1/n^2<s(a(n)-1)$ and the A001954/A001953 classification of the zero/one positions of $a(n)-a(n-1)$. — Formal Conjectures: open OEIS A049473 target
  • 2026-08-21 formal conjectures lean proof submission
    Published a kernel-checked Lean 4 proof of the full open Formal Conjectures target for OEIS A049473 and submitted it in PR #5087. For the nearest-integer sequence a(n)=⌊n/√2+1/2⌋, the proof first establishes two telescoping bounds for the ζ(3) tail s(m)=ζ(3)−Σₖ₌₁ᵐ1/k³: an upper bound below 1/(2m(m+1)) and a lower bound above 1/(2m²−2m+1). The nearest-integer inequalities yield the matching quadratic bounds 2m²−2m+1≤n²≤2m(m+1), which combine to give s(a(n))<1/n²<s(a(n)−1). The proof then derives the difference classification from exact floor inequalities and irrationality of √2: one-steps occur at the A001953 Beatty positions and zero-steps at the complementary A001954 positions. Thus both conjuncts of the original theorem are proved, not just the more elementary Beatty component. The repository uses Lean 4.27.0; its proof files contain no sorry, admit, custom axiom, native_decide, or unsafe declaration, and the axiom audits report only propext, Classical.choice, and Quot.sound.

OEIS A076141 — unique binary sub-pattern occurrences in n²Binary words occurring in the square n²

LeanProofOEISCombinatoricsNumber TheoryOPEN
Problem. Let $a(n)$ count the (possibly overlapping) occurrences of the most-significant-bit-first binary word of $n$ in the binary word of $n^2$. Prove that $a(n)\le1$ for every $n\in\mathbb{N}$. — Formal Conjectures: open OEIS A076141 target
  • 2026-08-21 formal conjectures lean proof submission
    Published a kernel-checked Lean 4 proof of the exact open Formal Conjectures target for OEIS A076141 and submitted it in PR #5088. The imported definition counts every occurrence, including overlaps, of the most-significant-bit-first binary word of n in the binary word of n². Assuming two distinct starts, the proof converts the two list-prefix matches into arithmetic block equations for n²; their overlap gives a period for the binary word of n. Careful lower and upper block decompositions, together with the needed power-of-two and coprimality bounds, then split into the possible signs of an auxiliary integer and force an impossible negative or oversized remainder. Hence two distinct occurrences cannot exist, while the n=0 case is discharged directly from the exact list definition. The proof uses the literal Formal Conjectures definitions rather than a surrogate theorem. The repository uses Lean 4.27.0; its proof files contain no sorry, admit, custom axiom, native_decide, or unsafe declaration, and the axiom audit reports only propext, Classical.choice, and Quot.sound.

Erdős #367 — false universal-ε higher-full-parts variantLimsup growth of products of r-full parts

Misformalization DetectionLeanProofCounterexampleNumber TheoryOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures source immediately before submission marks the literal universal-ε higher-full-parts question research open; the Erdős Problems page uses the same universal wording. PR #5090 proves that this literal statement is false and changes only this variant to research solved. The main declaration erdos_367.parts.i, Erdős Problem 367 overall, and the mathematically meaningful corrected variant with ∃ ε>0 all remain open. PR #5090 remains open as of 2026-08-22.
  • 🔗 Source problem Formal Conjectures: open literal higher-full-parts variant
Problem. For every $r\ge3$, $k\ge2$, and $\varepsilon>0$, must $\limsup_{n\to\infty}\prod_{n\le m<n+k}B_r(m)/n^{1+\varepsilon}=\infty$, where $B_r(m)$ is the $r$-full part of $m$? — Formal Conjectures: open literal higher-full-parts variant
  • 2026-08-22 formal conjectures literal-target disproof submission
    Published a kernel-checked Lean 4 disproof of the literal open Formal Conjectures higher-full-parts variant attached to Erdős Problem 367 and submitted it in PR #5090. The target quantifies over every positive ε; specializing to (r,k,ε)=(3,2,2) already gives a counterexample. From the definition of the r-full part, the proof establishes B r m ∣ m, hence B₃(n)B₃(n+1)≤n(n+1). After division by n^(1+2)=n³, the resulting EReal-valued sequence is eventually bounded by the finite value 2, so its limsup cannot be ⊤. This proves the exact imported target with answer(False). The scope boundary is essential: the result detects the flaw in the literal ∀ ε>0 formulation; it does not solve erdos_367.parts.i, Erdős Problem 367 as a whole, or the corrected question asking whether a suitable ε>0 exists. The proof uses Lean 4.27.0, contains no sorry, admit, custom axiom, native_decide, or unsafe declaration, and its axiom audit reports only propext, Classical.choice, and Quot.sound.

Melnikov's valency-variety problem — 37-vertex counterexampleChromatic number versus the number of distinct vertex degrees

LeanProofCounterexampleGraph TheoryOPEN
  • 🟢 Status when posted OPEN — archived Open Problem Garden problem page presents Melnikov's strict chromatic lower bound as an unanswered problem. Formal Conjectures did not yet contain this target; issue #5092 proposes adding it directly as research solved with answer(False) and links the completed Lean proof. The issue is open and marked up for grabs as of 2026-08-22, so this records a solution of the external problem and an addition proposal, not an upstream Formal Conjectures merge. Counterexample reports were also submitted to Open Problem Garden itself and to its UnsolvedMath mirror on 2026-08-22; both reports remain in their moderation queues, with no public report URL yet.
  • 🔗 Source problem Open Problem Garden: Melnikov's valency-variety problem (archived)
Problem. For every finite simple graph $G$ with at least two vertices, is $\chi(G)>\left\lceil\lfloor w(G)/2\rfloor/(|V(G)|-w(G))\right\rceil$, where $w(G)$ is the number of distinct vertex degrees? — Open Problem Garden: Melnikov's valency-variety problem (archived)
  • 2026-08-22 counterexample publication, formal conjectures addition proposal, and external status reports
    Published a kernel-checked Lean 4 counterexample to Melnikov's valency-variety problem and opened Formal Conjectures issue #5092 proposing that the problem be added as an already solved answer(False) entry. The explicit graph has 37 vertices partitioned into independent sets of sizes 6, 10, and 21. Its degree set is exactly {0,…,29}, so w(G)=30; the displayed tripartition gives a 3-colouring, while an explicit triangle forces chromatic number at least 3, hence χ(G)=3. Melnikov's proposed bound evaluates to ceil(floor(30/2)/(37−30))=ceil(15/7)=3, so the universal strict inequality fails by equality. The graph is disconnected and contains one isolated vertex, which is sufficient for the problem exactly as stated because no connectedness hypothesis is imposed; the proof does not claim a connected counterexample or determine the optimal replacement bound. The prospective Formal Conjectures-style statement is explicitly unofficial because no target was registered upstream when the issue was opened. On 2026-08-22, the counterexample and proof were also reported directly through the Open Problem Garden problem page and through the corresponding UnsolvedMath entry OPG-46575. Both submissions are still awaiting moderation, so they are recorded as submitted rather than published or accepted, and there is not yet a public permalink for either report. The proof uses Lean 4.27.0, contains no sorry, admit, custom axiom, native_decide, or unsafe declaration, and its axiom audit reports only propext, Classical.choice, and Quot.sound.

Green Problem 29 — counterexample to the polynomial-density subset boundLarge subsets of approximate groups with controlled product sets

LeanProofLiteratureCounterexampleAdditive CombinatoricsGroup TheoryOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures source when submitted marks Green29.green_29 research open with an unknown Boolean answer. A public validator result had already accepted, kernel-verified, and review-approved a counterexample to this exact pinned target on 2026-08-06. PR #5094 adapts and packages that public proof for the Formal Conjectures import and proposes answer(False). The already-solved weaker declaration Green29.green_29.variant, which drops S ⊆ A and allows the density constant to depend on K, is separate and unchanged. PR #5094 remains open as of 2026-08-22.
  • 🔗 Source problem Formal Conjectures: open Green Problem 29 target
Problem. Do there exist absolute constants $C,c>0$ such that every $K$-approximate group $A$ contains $S\subseteq A$ with $|S|\ge C K^{-c}|A|$ and $S^8\subseteq A^4$? — Formal Conjectures: open Green Problem 29 target
  • 2026-08-22 formal conjectures proof adaptation and solved-status submission
    Prepared an FC-facing Lean 4 repository and submitted Formal Conjectures PR #5094 to change the exact OPEN target Green29.green_29 from answer(sorry) to answer(False). The mathematical proof is explicitly adapted from public validator submission 82ab85ee-5dfc-4775-b3e1-8abc16e213b9, credited there to hotkey 5GeGrYFpMrNSh3Nwcx987zWz4cME9A9NbCkEbjBvv4uLUScV; it was accepted, kernel-verified, and review-approved on 2026-08-06. Thus this record credits the contribution as a proof adaptation and upstream status correction, not as an independent discovery of the counterexample. The construction takes an arbitrarily large finite group H and the slab A=({−1}×H)∪({1}×H)∪{(0,1)} inside Multiplicative ℤ×H. Three translates cover A², so A is a 3-approximate group and |A| grows with |H|. If S⊆A and S⁸⊆A⁴, comparing integer coordinates forces every element of S into the sole zero-layer point, hence |S|≤1. This contradicts every proposed positive lower density C·3^(−c) once H is large enough. The fixed proof establishes the exact imported proposition, contains no sorry, admit, custom axiom, native_decide, or unsafe declaration, and its axiom audit reports only propext, Classical.choice, and Quot.sound. The known weaker green_29.variant is not used as a substitute and remains unchanged.

Erdős #539 — negative answers to both square-root variantsGrowth of the cofactor-set threshold beyond every constant multiple of √n

LeanProofCounterexampleAdditive CombinatoricsNumber TheoryOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures source when submitted marks both Erdos539.erdos_539.variants.isBigO_sq and Erdos539.erdos_539.variants.sq research open as positive propositions. PR #5276 supplies answer(False) for both. The main problem asking for the exact asymptotic order of h(n) remains open, while the three distinct exponent-related variants handled in merged PR #4869 were already solved. PR #5276 remains open as of 2026-09-05.
  • 🔗 Source problem Formal Conjectures: open Erdős #539 square-root variants
Problem. For the cofactor threshold $h(n)$, is $h(n)=O(\sqrt n)$, and consequently is $h(n)=\Theta(\sqrt n)$? — Formal Conjectures: open Erdős #539 square-root variants
  • 2026-09-05 formal conjectures negative-answer submission
    Submitted Formal Conjectures PR #5276 with kernel-checked negative answers to both OPEN square-root variants of Erdős Problem 539. The proof establishes the stronger asymptotic statement h(n)/√n → ∞ for the original zero-inclusive Formal Conjectures definition. Prime-factor exponent vectors turn the cofactor set into positive coordinate differences. Assuming its cardinality q satisfies q²≤K|A|, a proved weak-PFR theorem supplies a large subset of bounded affine rank. After an injective coordinate restriction, a fixed-dimensional geometric lower bound for positive differences gives q≥c_d|B|^a_d with a_d>1/2. Combining this with q≤√(K|A|) bounds |A| in terms of K, which forces the threshold divided by √n to diverge. The exact target wrappers therefore prove answer(False) for both isBigO_sq and sq, and a metaprogram compares their complete types with the imported FC declarations to exclude extra hypotheses. This is a substantive solution of the two registered variants, not a solution of the main request to determine the exact order of h(n). The repository also states that publication and independent novelty review are separate from its local formal verification; this record therefore does not assert an external novelty determination. Both builds passed on 2026-09-05, contain no sorry, admit, or custom axiom, and the final theorems report only propext, Classical.choice, and Quot.sound.

Erdős #354(ii) — an explicit common base γ=√φ in (1,2)Completeness of floor-power sequences for a common exponential base

LeanProofAdditive CombinatoricsNumber TheoryOPEN
Problem. Does there exist a common $\gamma\in(1,2)$ such that, for every $\alpha,\beta>0$ with $\alpha/\beta$ irrational, the interleaving of $\lfloor\alpha\gamma^n\rfloor$ and $\lfloor\beta\gamma^n\rfloor$ is an additive complete sequence? — Formal Conjectures: open Erdős #354(ii) target
  • 2026-09-05 formal conjectures affirmative-proof submission
    Submitted Formal Conjectures PR #5286 with a kernel-checked affirmative solution of the OPEN Part II of Erdős Problem 354. The explicit witness is γ=√φ=√((1+√5)/2)∈(1,2). In fact, for every α>0, the single sequence ⌊αγⁿ⌋ is additively complete, so the second sequence involving β and the hypothesis that α/β is irrational are not needed for this witness. Splitting into even and odd indices turns each subsequence into a golden-ratio floor sequence. On the even indices, the identity φ²=φ+1 yields recurrences with a carry in {0,1}; carry-one events occur arbitrarily far out, since eventual absence of carries would force bounded nonnegative fractional parts to obey an exact Fibonacci recurrence and ultimately make φ rational. Disjoint carry-one triples allow subset sums to represent arbitrarily long intervals of consecutive integers. The positive odd-indexed tail grows by at most a factor of two, so adding its terms makes translated representable intervals overlap and extend without bound. This proves every sufficiently large natural number is a sum of terms with distinct indices. The exact answer(True) wrapper is linked to the imported FC target. This solves only the existential replacement-base question in Part II; the original base-2 question in Part I remains open. Both the FC and standalone versions build successfully, contain no sorry, admit, custom axiom, native_decide, or unsafe theorem, and the final axiom audit reports only propext, Classical.choice, and Quot.sound.

Erdős #879(ii) — bounded prime-factor count in optimal admissible setsPrime-factor structure of maximum-weight pairwise-coprime subsets

LeanProofNumber TheoryOPEN
Problem. For each $n$, let an admissible set be a subset of $\{1,\ldots,n\}$ whose distinct members are pairwise coprime, and call it optimal when its sum is maximal. Is it true that, for every $k\ge2$ and all sufficiently large $n$, some optimal set contains an integer having at least $k$ distinct prime factors? — Formal Conjectures: Erdős #879 formalization in PR #5302
  • 2026-09-06 formal conjectures negative-answer submission
    Submitted Formal Conjectures PR #5302 with a new formalization of both parts of Erdős Problem 879 and a kernel-checked negative solution of Part II. The proof establishes a strictly stronger statement: for every sufficiently large n, every element of every maximum-sum pairwise-coprime subset of {1,…,n} has at most two distinct prime factors. Thus the claimed eventual existence of an element with arbitrarily many distinct prime factors is false already for k=3. The argument observes that each prime can occur in at most one selected integer and injects large primes blocked inside nonsingleton selected integers into a controlled set of small primes. A first exchange rules out an element with two sufficiently small prime factors. In the remaining three-factor configuration, one small and two large factors can be replaced by a prime-power product and a large square, increasing the total weight; the prime number theorem supplies unblocked primes in the required fixed-ratio intervals. The repository includes an exact wrapper for the Formal Conjectures Part II statement and a Lean4Web version. The final theorem reports only propext, Classical.choice, and Quot.sound, with no sorryAx or project-specific axiom. This result does not solve Part I, and the record makes no claim of an exhaustive external novelty review.

Bugeaud Problem 10.5Universal lacunary sequences over real number fields

LeanProofNumber TheoryOPEN
Problem. For every real number field $K$ and every $\varepsilon>0$, construct a positive lacunary sequence $(t_n)$ in $K$ such that, simultaneously for every $\xi\notin K$, each subinterval of $[0,1]$ of length $\varepsilon$ contains a cluster point of $(\{\xi t_n\})$. In particular, prove $\limsup_n\{\xi t_n\}\ge 1-\varepsilon$. — Formal Conjectures: Bugeaud Problem 10.5
  • 2026-09-20 formal conjectures proof submission
    Submitted Formal Conjectures PR #6434 with kernel-checked Lean 4 proofs of both OPEN targets in Bugeaud's Problem 10.5. The proof establishes the strictly stronger moreover statement: for every real number field K and every ε>0, it constructs one positive lacunary sequence in K that works simultaneously for every real ξ∉K, placing a cluster point of the fractional parts {ξtₙ} inside every requested subinterval of [0,1] of length ε. The registered limsup bound limsup {ξtₙ} ≥ 1−ε follows from this statement. Choosing an integer Q>1/ε, setting M=Q! and J=10M, the construction uses the explicit integer sequence t(Jk+r)=(J+r)2ᵏ. The proof combines late escape of irrational doubling orbits, Dirichlet approximation, divisibility by Q!, nearest-integer rounding, and compactness to obtain the cluster point. It actually works for every intermediate field of ℝ/ℚ; finite dimensionality is used only because it appears in the registered target. At fixed commit fa5533b, the exact-target Formal Conjectures project and standalone mathlib project both report successful Lean 4.33.1 builds. The final axiom audit lists only propext, Classical.choice, and Quot.sound, with no sorryAx, custom axiom, native_decide, or unsafe theorem. PR #6434 changes both declarations to research solved; as of 2026-09-20 it is OPEN, unreviewed, and unmerged. All reported required checks except the still-running upstream project build have completed successfully.

Convex additive VCₙ — uniform finite bound for n≥1Finite additive VCₙ dimension of convex sets in ℝⁿ⁺¹

LeanProofConvex GeometryOPEN
Problem. For every integer $n\ge1$, does there exist a finite $d$ such that every convex set $C\subseteq\mathbb R^{n+1}$ has additive $\mathrm{VC}_n$ dimension at most $d$? — Formal Conjectures: convex additive VCₙ bounds
  • 2026-09-20 formal conjectures proof submission
    Submitted Formal Conjectures PR #6436 with a complete Lean 4 proof of the repaired uniform finite-bound target for convex additive VC dimension. For every n≥1 and every convex C⊆ℝⁿ⁺¹, the proof supplies the explicit witness d=2^(8*(n+2)^n)−1 to HasAddVCNDimAtMost C n d. The argument first obtains small representative subsets for convex membership patterns through geometric sparsification, then applies a direct polynomial sign-pattern bound to the determinant polynomials. The exact Formal Conjectures wrapper and axiom audit are pinned at commit 9d7685c; #print axioms reports only propext, Classical.choice, and Quot.sound, with no sorryAx. This completes the intended positive-dimensional problem after issue #5022 identified the false n=0 version and the declaration was repaired by adding 1≤n. It is compatible with PR #4657's counterexample to the proposed n=2 bound 1: the present result proves existence of a finite bound, not the sharper open n=2 bound 2 or uniform bound 3. PR #6436 changes only the finite-existence declaration to research solved; as of 2026-09-20 it is OPEN, unreviewed, and unmerged. The upstream scan and other completed required checks pass; the project build is still running.

Erdős #1095 — upper-conjecture status correctionUpper bound for the Ecklund–Erdős–Selfridge function g(k)

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Let $g(k)$ be the least $n > k+1$ such that every prime factor of $\binom{n}{k}$ exceeds $k$. Then $g(k)\le\exp((1+o(1))k)$. — Formal Conjectures: Erdős Problem 1095
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Ecklund, Erdős, and Selfridge's 1974 paper A New Function Associated with the Prime Factors of $\binom{n}{k}$ (Math. Comp. 28, 647–649) obtains the upper bound $g(k) < \exp(k(1+o(1)))$. This is the coefficient-1 exponential bound encoded by erdos_1095.variants.upper_conjecture, so the requested correction is to mark the declaration solved and cite the paper. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #160 — better upper- and lower-bound status correctionsColourings in which every four-term progression receives three colours

LiteratureStatusCorrectionAdditive CombinatoricsOPEN
Problem. Let $h(n)$ be the least number of colours of $\{1,\ldots,n\}$ for which every four-term arithmetic progression receives at least three distinct colours. Improve the recorded bounds $\exp(c(\log n)^{1/12})\ll h(n)\ll n^{2/3}$. — Formal Conjectures: Erdős Problem 160 better bounds
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Shi and Dong's 2026 preprint An Improved Upper Bound for Colorings Without Symmetrically Colored $k$-Term Arithmetic Progressions gives an upper bound for $h(n)$ stronger than $n^{2/3}$, while Bloom and Sisask's An improvement to the Kelley-Meka bounds on three-term arithmetic progressions implies a lower bound stronger than $\exp(c(\log n)^{1/12})$, the bound the file derives from Hunter's observation and the Kelley–Meka bounds. The requested correction is to mark better_upper and better_lower solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

TxGraffiti Conjecture 2 — counterexample status correctionZero forcing versus independence in subcubic graphs

LiteratureStatusCorrectionGraph TheoryOPEN
Problem. Every connected graph $G\ne K_4$ with maximum degree at most $3$ satisfies $Z(G)\le\alpha(G)+1$, where $Z(G)$ is the zero forcing number and $\alpha(G)$ the independence number. — Formal Conjectures: TxGraffiti Conjecture 2
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Mikko Fischer's 2026 preprint A counterexample to the zero forcing versus independence conjecture for cubic and subcubic graphs gives a connected 24-vertex graph of maximum degree 3 with $Z(G) > \alpha(G)+1$. This falsifies the universal statement encoded by tx_graffiti_conjecture_2, so the proposed answer is False. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #968 — increasing- and decreasing-triple status correctionsMonotone triples in the normalized primes pₙ/n

LiteratureStatusCorrectionAnalytic Number TheoryOPEN
Problem. Let $u_n=p_n/n$, where $p_n$ is the $n$th prime. Are there infinitely many $n$ with $u_n < u_{n+1} < u_{n+2}$, and infinitely many with $u_n > u_{n+1} > u_{n+2}$? — Formal Conjectures: Erdős Problem 968 triple variants
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that consecutive large prime gaps from Ford, Maynard, and Tao's Chains of large gaps between primes, together with the bounded clusters of primes in Maynard's Small gaps between primes, imply that there are infinitely many increasing triples $u_n < u_{n+1} < u_{n+2}$ and infinitely many decreasing triples $u_n > u_{n+1} > u_{n+2}$. The proposed answer is True for both declarations. The report explicitly leaves open the main question erdos_968, whether $u_n < u_{n+1}$ holds on a set of positive lower density. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Cyclic-subgroup solvability threshold — Conjecture 5.5 status correctionSolvability from the number of cyclic subgroups

LiteratureStatusCorrectionGroup TheoryOPEN
Problem. If a finite group $G$ has fewer than $2^{t+2}$ cyclic subgroups, where $t$ is the number of distinct prime divisors of $\lvert G\rvert$, then $G$ is solvable. — Formal Conjectures: arXiv 2604.08040 Conjecture 5.5
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that version 2 of arXiv:2604.08040, now titled Group Structure from Subgroup and Cyclic Subgroup Counts by Das, Dey, Galindo, and Sharma, proves the cyclic-subgroup solvability threshold of Conjecture 5.5 as Theorem 7.1. The requested correction is to mark solvable_of_cyc_lt solved with a positive answer. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Bugeaud 10.4 — countable-spectrum status correctionSpectrum of the geometric sequence ξαⁿ modulo one

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. For every real $\xi\ne0$ and $\alpha > 1$, the spectrum of $(\xi\alpha^n)_{n\ge1}$, that is, the set of irrational $\theta\in(0,1)$ for which $(\xi\alpha^n-n\theta)_{n\ge1}$ is not uniformly distributed modulo one, is at most countable. — Formal Conjectures: Bugeaud Problem 10.4
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Hikmet Burak Özcan's 2026 preprint The spectrum of $(\xi\alpha^n)$ can be uncountable gives a stronger counterexample for every $\alpha > 1$. This refutes the universal countability statement spectrum_xi_alpha_pow_countable, posed by Mendès France and recorded as Problem 10.4 in Bugeaud's Distribution modulo one and Diophantine approximation, so the proposed answer is False. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1099 — factorial and LCM variant status correctionsBounded divisor-ratio sums along n! and lcm(1,…,n)

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. For $\alpha > 1$ let $h_\alpha(n)=\sum_i(d_{i+1}/d_i-1)^\alpha$, where $1=d_1 < \cdots < d_{\tau(n)}=n$ are the divisors of $n$. Are $h_\alpha(n!)$ and $h_\alpha(\mathrm{lcm}(1,\ldots,n))$ bounded for every $\alpha > 1$? — Formal Conjectures: Erdős Problem 1099 variants
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Theorem 1 of Gérald Tenenbaum's 1987 paper Sur un problème extrémal en arithmétique (Annales de l'Institut Fourier) proves, for every $\alpha > 1$, that $h_\alpha(n!)$ and $h_\alpha(\mathrm{lcm}(1,\ldots,n))$ are bounded. These are the two explicit sequences that Erdős suggested as candidates, and the requested correction is to mark erdos_1099.variants.factorial and erdos_1099.variants.lcm solved with positive answers. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #331 — Ruzsa variant status correctionCommon differences of sets with square-root asymptotic density

LiteratureStatusCorrectionAdditive CombinatoricsOPEN
Problem. Let $A,B\subseteq\mathbb N$ with $\lvert A\cap[1,N]\rvert\sim c_A N^{1/2}$ and $\lvert B\cap[1,N]\rvert\sim c_B N^{1/2}$ for constants $c_A,c_B > 0$. Are there infinitely many solutions of $a_1-a_2=b_1-b_2\ne0$ with $a_1,a_2\in A$ and $b_1,b_2\in B$? — Formal Conjectures: Erdős Problem 331 Ruzsa variant
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Theorem 4 of Erdős and Freud's 1984 paper On Disjoint Sets of Differences (J. Number Theory 18, 99–109) proves the variant suggested by Ruzsa, in which both counting functions are asymptotic to positive multiples of $N^{1/2}$. The requested correction is to mark erdos_331.variants.ruzsa solved and cite the theorem. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #906 — known-solution status correctionDense zeros of high derivatives of an entire function

LiteratureStatusCorrectionComplex AnalysisOPEN
Problem. Is there a nonzero transcendental entire function $f$ such that, for every strictly increasing sequence $n_0 < n_1 < \cdots$, the zeros of the derivatives $f^{(n_k)}$ together form a dense subset of $\mathbb C$? — Formal Conjectures: Erdős Problem 906
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Eric Hou's 2026 preprint Cofinite Zeros of High Derivatives proves the existence statement required by erdos_906. The requested correction is to mark the declaration solved with a positive answer. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Kourovka 19.25 — counterexample status correctionSimplicity from group order and totient sums of element orders

LiteratureStatusCorrectionGroup TheoryOPEN
Problem. Let $G$ and $H$ be finite groups of the same order with $\sum_{g\in G}\phi(\operatorname{ord}g)=\sum_{h\in H}\phi(\operatorname{ord}h)$, where $\phi$ is Euler's totient function. If $G$ is simple, must $H$ be simple? — Formal Conjectures: Kourovka Problem 19.25
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that the order-6048 counterexample in van Doorn, Judin, Monticone, and Morrison's 2026 preprint On Some Problems from the Kourovka Notebook is a counterexample to the current kourovka_19_25 declaration, whose same-order hypothesis was added by Formal Conjectures PR #4607. The proposed answer is therefore False. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Kourovka 20.76 — counterexample status correctionAbelian subgroups of p-groups with small abelian normal subgroups

LiteratureStatusCorrectionGroup TheoryOPEN
Problem. Let $G$ be a finite $p$-group in which every abelian normal subgroup has order at most $p^k$. Must every abelian subgroup of $G$ have order at most $p^{2k}$? — Formal Conjectures: Kourovka Problem 20.76
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Proposition 6.1 of Kyrylo Muliarchyk's note Abelian subgroups of finite 2-groups: counterexamples to Problems 20.75 and 20.76 gives a finite 2-group of order $2^{145}$ whose abelian normal subgroups have order at most $2^{42}$ but which contains an elementary abelian subgroup of order $2^{85} > 2^{84}$. Taking $p=2$ and $k=42$ falsifies the Formal Conjectures declaration, so the proposed answer is False. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Green Problem 44 — counterexample status correctionSieving half the residue classes modulo 1000 primes

LiteratureStatusCorrectionAnalytic Number TheoryOPEN
Problem. Sieve $[N]$ by removing $\lfloor p_i/2\rfloor$ residue classes modulo each of the primes $2\le p_1 < \cdots < p_{1000} < N^{9/10}$. Must the remaining set have size at most $N/10$? — Formal Conjectures: Green Problem 44
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Green Problem 44 is the same problem as Erdős Problem 1202. The erdosproblems.com page for Problem 1202 identifies it with Green's Problem 44 and records it as resolved in the negative by Price and GPT-5.4 Pro, with a construction that sieves out half of the residue classes modulo many primes while $\gg N$ integers survive. This supplies a counterexample to the Formal Conjectures declaration, so the proposed answer is False. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Tu–Deng conjecture — known-solution status correctionBinary weights of residue pairs modulo 2ᵏ − 1

LiteratureStatusCorrectionCombinatoricsOPEN
Problem. For $k\ge2$ and every nonzero residue $t$ modulo $2^k-1$, at most $2^{k-1}$ pairs of residues $(a,b)$ satisfy $a+b\equiv t\pmod{2^k-1}$ and $w(a)+w(b)\le k-1$, where $w$ is the binary weight. — Formal Conjectures: Tu–Deng conjecture
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Theorem 1.1 of Liu, Luo, and Xie's 2026 preprint A Complete Proof for Tu-Deng Conjecture and Thomas W. Cusick's 2026 preprint on the same conjecture both prove the exact combinatorial statement encoded by tu_deng_conjecture. The requested correction is to mark the declaration solved and cite the proofs. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Bugeaud 10.6 — sublacunary and intermediate-growth variant status correctionsRapidly increasing integer sequences dense modulo one

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Find a strictly increasing sequence $(m_n)$ of positive integers such that $(\{\xi m_n\})$ is dense modulo one for every irrational $\xi$, and which is either genuinely sublacunary ($m_{n+1}/m_n\ge1+c/\log n$ eventually, for some $c > 0$) or of intermediate growth ($m_n\ge\exp(n^\alpha)$ eventually, for some $0 < \alpha < 1$). — Formal Conjectures: Bugeaud Problem 10.6
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that the Ajtai–Havas–Komlós existence theorem, as recorded in Example 2.4(d) of Gao, Mooroogen, and Yip's On an Erdős similarity problem in the large, yields sequences satisfying both Formal Conjectures variants of Bugeaud's Problem 10.6 after rounding to integers. The requested correction is to mark problem_10_6_variant_1 and problem_10_6_variant_2 solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #700(b) — square-root exceedance status correctionMinimal gcd of n with its binomial coefficients

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Let $f(n)=\min_{1 < k\le n/2}\gcd(n,\binom{n}{k})$. Are there infinitely many composite $n$ with $f(n) > n^{1/2}$? — Formal Conjectures: Erdős Problem 700(b)
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that a July 24, 2026 write-up on Erdős Problem a Day combines a three-prime lemma with Maynard's theorem to prove that infinitely many composite $n$ satisfy $f(n) > n^{1/2}$. The write-up's own title says that one of the three sub-questions “may be unconditionally settled”. The proposed answer to erdos_700.parts.ii is True. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #872 — prime-question status correctionSmallest maximal primitive subsets of {2,…,n}

LiteratureStatusCorrectionCombinatoricsOPEN
Problem. For every $n\ge2$, does every maximal primitive subset of $\{2,\ldots,n\}$ have at least $\pi(n)$ elements? — Formal Conjectures: Erdős Problem 872 prime question
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that a May 20 post in the Erdős Problems forum thread for Problem 872 proves, using the largest prime powers up to $n$, that every maximal primitive subset of $\{2,\ldots,n\}$ has at least $\pi(n)$ elements; Sayed Elsakhawy's 2023 paper Independence and domination in divisor graph and mod-difference graphs (Journal of the Egyptian Mathematical Society) also supports the statement. The proposed answer to erdos_872.variants.prime_question is True. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #269 — rationality-variant status correctionReciprocal sums of partial LCMs of P-smooth numbers

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Let $P$ be a finite set of at least two primes, and let $a_1 < a_2 < \cdots$ be the positive integers whose prime factors all lie in $P$. Is $\sum_{n\ge1}1/[a_1,\ldots,a_n]$ rational for every such $P$? — Formal Conjectures: Erdős Problem 269 rationality variant
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Steve Fan's June 26 post in the Erdős Problems forum thread for Problem 269 proves that the sum is transcendental when $\lvert P\rvert=2$. Any such $P$ is a counterexample to the universal rationality statement encoded by erdos_269.variants.rational, so the proposed answer is False. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1063 — better-upper-bound status correctionLeast n for which all but one of n−i divide C(n,k)

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Let $n_k$ be the least $n\ge2k$ such that all but one of the integers $n-i$ with $0\le i < k$ divide $\binom{n}{k}$. Improve Cambie's upper bound $n_k\le k\cdot\mathrm{lcm}(1,\ldots,k-1)$. — Formal Conjectures: Erdős Problem 1063 better upper bound
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that a public July 25 proof in the GitHub repository pw/erdos1063-upper-bound gives $n_k\le\exp(o(k))$. Since $\mathrm{lcm}(1,\ldots,k-1)=e^{(1+o(1))k}$, this bound is $o(k\cdot\mathrm{lcm}(1,\ldots,k-1))$ and supplies the improvement requested by erdos_1063.better_upper. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #289 — separated-interval status correctionUnit fractions from separated intervals of consecutive integers

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Is it true that, for all sufficiently large $k$, there are intervals $I_1,\ldots,I_k\subset\mathbb N$, pairwise neither overlapping nor adjacent and each of length at least $2$, with $1=\sum_{i=1}^k\sum_{n\in I_i}1/n$? — Formal Conjectures: Erdős Problem 289
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that the public proof by Cambie, Land, and Tang, together with its Lean formalization in the GitHub repository beetree/math_erdos_289, establishes the separated-interval statement currently encoded by erdos_289. The requested correction is to mark the declaration solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #105 — n−4 variant counterexample status correctionLines through two points of A that avoid a disjoint set B

LiteratureStatusCorrectionDiscrete GeometryOPEN
Problem. Let $A,B\subset\mathbb R^2$ be disjoint sets of sizes $n$ and $n-4$, with $A$ not contained in a line. Is there a line containing at least two points of $A$ and no point of $B$? — Formal Conjectures: Erdős Problem 105 n−4 variant
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that a July 24 post in the Erdős Problems forum thread for Problem 105 gives an explicit counterexample to the $n-4$ strengthening, with $10$ points in $A$ and $6$ in $B$ such that every line through two points of $A$ meets $B$. The configuration is stated in the real projective plane, and a projective transformation carries it into $\mathbb R^2$. The proposed answer to erdos_105.variants.sub_four is False. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #263(ii) — corrected-statement status correctionGrowth of strictly increasing irrationality sequences

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Call a strictly increasing sequence $a_n$ of positive integers an irrationality sequence if $\sum 1/b_n$ is irrational for every sequence of positive integers $b_n$ with $a_n/b_n\to1$. Must every irrationality sequence satisfy $a_n^{1/n}\to\infty$? — Formal Conjectures: Erdős Problem 263(ii)
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Liam Price's May 9 post in the Erdős Problems forum thread for Problem 263, which presents a GPT-5.5 Pro disproof of the second question together with a Lean auto-formalization, settles part (ii) in the corrected form adopted on erdosproblems.com on 2026-04-02, where irrationality sequences must be strictly increasing. The requested correction is to mark erdos_263.parts.ii solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #270 — linear-variant status correctionIrrationality of the f(n) = n series

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Is $\sum_{n\ge1}\frac{1}{(n+1)(n+2)\cdots(2n)}=\sum_{n\ge1}\frac{n!}{(2n)!}$ irrational? — Formal Conjectures: Erdős Problem 270 linear variant
  • 2026-09-21 formal conjectures literature-status report
    Reported in issue #6448 that Vjekoslav Kovač's July 12 post in the Erdős Problems forum thread for Problem 270, an observation made with Tonći Crmarić, proves that $\sum_{n\ge1}n!/(2n)!$ is irrational by a Fourier-style argument for Cantor series. This settles the case $f(n)=n$, which the Formal Conjectures file still calls unknown, so the requested correction is to mark erdos_270.variants.linear solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Green Problem 47 — exact-containment inverse-sieve counterexampleInverse large sieve: sparse-or-quadratic dichotomy

LeanProofCounterexampleAnalytic Number TheoryOPEN
Problem. If $A\subseteq\mathbb N$ occupies at most $(p+1)/2$ residue classes modulo every sufficiently large prime $p$, must either $|A\cap[0,X]|=O(\sqrt X/\log^{100}X)$, or must all of $A$ lie in the image of the integers under one rational quadratic? — Formal Conjectures: Green Problem 47
  • 2026-09-21 formal conjectures counterexample submission
    Submitted Formal Conjectures PR #6449 with a complete Lean 4 negative answer to Green Problem 47. The explicit set is B={2}∪{4^k:k∈ℕ}∪{p²:p prime, p≡1 (mod 8)}. It occupies at most (q+1)/2 residue classes modulo every odd prime q, but its prime-square layer makes its counting function too large to be O(√X/(log X)^100). The powers 4^k force any rational quadratic containing all of B on integer inputs to have zero discriminant; the additional element 2 then contradicts the resulting square-like form. Thus neither branch of the exact registered dichotomy holds. This is a genuine counterexample to Green Problem 47 as printed and faithfully formalized, not a defect in its Lean encoding. Its scope is nevertheless narrower than the broader Green–Harper inverse large sieve program: deleting the single element 2 leaves a subset of the squares, so it does not refute Green–Harper (2014), Conjecture 1.7, whose structural alternative permits finitely many exceptions. At fixed commit d6912b6, both complete proof versions report successful builds; the final axiom audit lists only propext, Classical.choice, and Quot.sound, with no sorryAx. PR #6449 changes the target from answer(sorry) to answer(False); as of 2026-09-21 it is OPEN, unreviewed, and unmerged. All completed required checks pass, while the upstream project build is still running.

Open Quantum Problem 35 — existence of AME(11,10)Absolutely maximally entangled state with eleven ten-dimensional parties

LeanProofQuantum InformationOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures source before PR #6455 marks OpenQuantumProblem35.ame_11_10_open research open with an unknown Boolean answer. PR #6455 changes only this case to research solved with answer(True) and links the fixed external Lean proof. The other unresolved parameter pairs in Open Quantum Problem 35 are outside this submission.
  • 🔗 Source problem Formal Conjectures: open AME(11,10) target
Problem. Does an absolutely maximally entangled state $\mathrm{AME}(11,10)$ exist? — Formal Conjectures: open AME(11,10) target
  • 2026-09-22 formal conjectures affirmative-proof submission
    Submitted Formal Conjectures PR #6455 with a complete Lean 4 affirmative proof of the OPEN AME(11,10) benchmark. The construction identifies each local ten-dimensional system with ZMod 2 × ZMod 5, couples two branches from a twelve-qubit graph state and the published quinary graph construction of an AME(12,5) state, and then discards the extra branch coordinate. Lean checks 792 binary five-versus-seven cut certificates and 924 quinary six-versus-six cut certificates, together with coverage of every required subsystem. Character orthogonality makes the cross terms vanish and gives the coupled raw reduction 2,000,000·I; exact scaling produces I/10^5 on every five-party subsystem and normalization follows from the trace. Coordinate equivalences then transport the state to the exact Formal Conjectures definition ExistsAME 11 10. This settles the single pair (n,d)=(11,10) and does not claim a complete classification of all pairs in Open Quantum Problem 35. At fixed commit 6b7d009, the repository reports a fresh 9,001-job build of the modular FC project and a separate compilation of the 1.1 MB standalone source under Lean/Mathlib v4.33.1. The final axiom audit lists only propext, Classical.choice, and Quot.sound; a source scan finds no sorry, admit, custom axiom, native_decide, or unsafe theorem. PR #6455 changes only the target's answer and metadata. As of 2026-09-22 it is OPEN, unreviewed, and unmerged; all completed required checks pass, while the upstream project build is still running.

Green Problem 25 — improved upper bound for restricted sumsetsPartition colour count versus unions of restricted sumsets

LeanProofAdditive CombinatoricsOPEN
Problem. Can the known $N/\log N$ upper bound be lowered for the number of classes in a partition of $[N]$ whose within-class restricted sumsets cover fewer than $N/10$ values? — Formal Conjectures: open Green 25 upper-bound target
  • 2026-09-23 formal conjectures improved-upper-bound submission
    Submitted Formal Conjectures PR #6508 with a complete Lean 4 solution of the OPEN upper-bound target in Green Problem 25. For every N > 2^(10240000−1), the proof constructs a partition of {1,…,N} into exactly ⌈N^(23/40)⌉ nonempty classes for which the union of all within-class restricted sumsets has size below N/10. Since N^(23/40)=o(N/log N), this strictly improves the previously recorded upper-bound scale. The argument uses random linear colourings of binary vectors, controls carries through a fractional union bound with exponent 1/16, and reduces the growth estimate to an exact 256-word, eight-bit transfer-matrix certificate checked by Lean's kernel. Restriction to [N] and refinement then give exactly the required number of nonempty classes without creating new monochromatic pairs. The fixed proof repository reports successful builds of both the FC-compatible and standalone versions; its axiom audit lists only propext, Classical.choice, and Quot.sound, and its sources contain no sorry, admit, custom axiom, native_decide, or unsafe theorem. This solves only Green25.green_25.upper: the exact admissible set of colour counts, the improved lower bound, and optimality of the exponent 23/40 remain open. As of 2026-09-23, PR #6508 is OPEN, unreviewed, and unmerged; all completed required checks pass, while the upstream project build is still running.

Open Quantum Problem 35 — existence of AME(9,6)Absolutely maximally entangled state with nine six-dimensional parties

LeanProofQuantum InformationOPEN
Problem. Does an absolutely maximally entangled state $\mathrm{AME}(9,6)$ exist? — Formal Conjectures: open AME(9,6) target
  • 2026-09-23 formal conjectures affirmative-proof submission
    Submitted Formal Conjectures PR #6512 with a complete Lean 4 affirmative proof of the OPEN AME(9,6) benchmark. The explicit normalized nine-party state uses binary and ternary labels for each six-dimensional local basis, with phases defined by four finite matrices. Character-sum cancellation eliminates every off-diagonal entry in each four-party reduced density matrix, while the diagonal entries equal 1/6⁴. Lean checks a finite certificate covering all C(9,4)=126 choices of four parties and proves the exact imported ExistsAME 9 6 proposition. It also proves answer(True) ↔ ExistsAME 9 6 without using the unfinished FC declaration as a premise. The fixed repository records a passing lake --wfail build AME96 AME96.Audit and an axiom audit containing only propext, Classical.choice, and Quot.sound; the separate Lean4Web edition was also reported to build. The proof settles this single (n,d)=(9,6) existence case; the general OQP 35 classification remains open. As of 2026-09-23, PR #6512 is OPEN, unreviewed, and unmerged; the completed required checks pass, while the upstream project build is running.

Square Packing — exact minimum circle radius for three unit squaresPacking three freely placed unit squares in a circle

LeanProofMetric GeometryOPEN
Problem. What is the least radius of a circle containing three non-overlapping unit squares, with arbitrary rotations and reflections? — Formal Conjectures: open three-square minimum target
  • 2026-09-24 formal conjectures exact-minimum proof submission
    Submitted Formal Conjectures PR #6524 with a Lean 4 proof that the least radius of a circle packing three non-overlapping unit squares is exactly 5√17/16. The proof imports the pinned FC Packing, UnitSquare, and Circle definitions and establishes their exact IsLeast statement for arbitrary rotations and reflections. A matching placement gives the upper bound. For the lower bound, proven symmetries reduce 512 separation branches to 64 representatives; center elimination and rational Bernstein certificates rule out every smaller radius. The fixed repository records successful checks of the FC-compatible modules and the standalone Lean source. Its integration audit confirms the FC feasible set, and the final axiom audit lists only propext, Classical.choice, and Quot.sound. The numerical candidate and prior rigorous bounds predate this contribution; The-Anh Vu-Le also has a separate Lean 4 proof, acknowledged in the PR. This contribution supplies a separately checked exact proof for the FC definitions. PR #6524 leaves the long proof external, so its FC theorem body remains by sorry. As of 2026-09-24 the PR is OPEN and unmerged, with no reviewer decision; its required CI checks, including the project build, pass.

Weak Tiling Problem 4.1 — bounded density of a weak tiling measure's supportWeak tiling measures and density of their supports

LeanProofOPEN
Problem. If $\Omega\subset\mathbb R$ is a finite union of intervals and $\nu$ is a weak tiling measure for $\Omega$, must the support of $\nu$ have uniformly bounded numbers of points in every open unit interval? — Formal Conjectures: open Weak Tiling Problem 4.1 target
  • 2026-09-24 formal conjectures affirmative-proof submission
    Submitted Formal Conjectures PR #6532 with a complete Lean 4 affirmative proof of Weak Tiling Problem 4.1 from Kolountzakis, Lev, and Matolcsi. For every finite union of intervals Ω and every weak tiling measure ν for Ω, the proof establishes a uniform bound on the number of points of ν.support in any open unit interval. The external theorem proves the exact FC statement answer(True) ↔ … HasBoundedDensity ν.support using FC's own definitions. Its argument combines the pure-point structure of the measure with lattice coordinates and Fourier recurrence analysis, ultimately placing the support in a finite set together with finitely many arithmetic progressions. The fixed repository reports successful FC-compatible and standalone Lean builds; #print axioms lists only propext, Classical.choice, and Quot.sound, with no sorryAx. This contribution concerns Problem 4.1; the separately registered Problem 4.3 counterexample is a different result. As of 2026-09-24 PR #6532 is OPEN, unreviewed, and unmerged; completed required checks pass while the upstream project build is still running. The proof has not yet received an independent human review.

Erdős #302 — Unit-fraction triple-free sets

LeanProofNumber TheoryOPEN
Problem. Let $f(N)$ be the largest size of $A\subseteq\{1,\ldots,N\}$ containing no three distinct $a,b,c$ with $1/a=1/b+1/c$. Estimate $f(N)$. — Formal Conjectures: open Erdős 302 estimation target
  • 2026-09-25 formal conjectures improved-upper-bound submission
    Submitted Formal Conjectures PR #6580 adding a research solved variant for the upper bound $f(N)\le(0.8461739827964010+o(1))N$. Here $f(N)$ is the largest subset of $\{1,\ldots,N\}$ with no three distinct $a,b,c$ satisfying $1/a=1/b+1/c$. The previous bound recorded in Formal Conjectures is $0.9N$ asymptotically. The linked fixed-commit Lean theorem uses the unchanged FC definition IsMaxNoTripleCard and proves the exact new declaration. The proof repository reports kernel checks of its finite certificates and an axiom audit listing only propext, Classical.choice, and Quot.sound, with no sorryAx; these builds were not rerun for this site update. This is a partial improvement, not a determination of the limit or the optimal constant. As of 2026-09-25, PR #6580 is open and unmerged, with no independent human review reported.
  • 2026-09-25 erdős problems forum report
    Reported the same Lean upper bound in Erdős Problem 302 discussion post #9181, linking the Formal Conjectures submission, public proof repository, and Lean4Web artifact. The PDF capture taken on 2026-09-25 shows the overall problem still OPEN. This is a public report of the same result, not a second independent proof.

Green Problem 41 — Double-exponential pyjama-covering bound

LeanProofDynamical SystemsOPEN
Problem. For sufficiently small $\varepsilon>0$, can the plane be covered by at most $\exp\exp(\varepsilon^{-C})$ rotated copies of the pyjama set for some constant $C$? — Formal Conjectures: open Green 41 double-exponential variant
  • 2026-09-25 formal conjectures double-exponential-bound proof submission
    Submitted Formal Conjectures PR #6581 with an external Lean 4 proof of the exact double-exponential variant answer(True) ↔ ∃ C, ε₀ > 0, ∀ 0 < ε ≤ ε₀, minCopies ε ≤ exp(exp(ε^(-C))). Here minCopies ε is the minimum number of rotations of the set {z : ∃ k ∈ ℤ, |Re z − k| ≤ ε} needed to cover the plane. Kravitz and Leng's earlier bound is triple exponential, so this removes one exponential from the bound recorded in Formal Conjectures. The proof uses the official FC definitions and produces explicit constants C = 102 and ε₀ = 1/(10 + log(10000·101)). Its fixed repository reports successful FC-compatible and standalone Lean builds, with #print axioms listing only propext, Classical.choice, and Quot.sound and no sorryAx; these builds were not rerun for this site update. The exact minimum and polynomial-bound variant are not proved. As of 2026-09-25, PR #6581 is open and unmerged, with no independent human review recorded.
  • 2026-09-28 formal conjectures review response
    On 2026-09-27 Formal Conjectures maintainer mo271 commented on PR #6581 (“Thanks, nice!”) and asked for a sentence in the docstring and a reference to the proof repository. On 2026-09-28 the PR added a docstring sentence crediting the Lean proof and cited the repository as [Ki26]. The PR remained open and unmerged.
  • 2026-10-02 formal conjectures pr extension
    The same PR was extended with a solved quasi-polynomial variant, green_41.variants.quasi_polynomial_bound, proved in a new repository and recorded separately on this site. The quasi-polynomial bound implies the double-exponential one; the double-exponential statement and its proof link are unchanged.

Kolountzakis–Lev–Matolcsi Problem 7.2 — three-dimensional caseSpectral products with three-dimensional convex bodies

LeanProofHarmonic AnalysisConvex GeometryOPEN
Problem. If $A\subset\mathbb R^3$ is a convex body and $B\subset\mathbb R^m$ is a bounded measurable set such that $A\times B$ is spectral, must $B$ be spectral? — Formal Conjectures: open KLM Problem 7.2 three-dimensional target
  • 2026-09-25 formal conjectures affirmative-proof submission
    Submitted Formal Conjectures PR #6590 with an external Lean 4 proof of the three-dimensional case of Problem 7.2 in Kolountzakis, Lev, and Matolcsi's Spectral sets and weak tiling: if $A\subset\mathbb R^3$ is a convex body, $B\subset\mathbb R^m$ is bounded and measurable, and $A\times B$ is spectral, then $B$ is spectral. The one- and two-dimensional cases are due to Greenfeld and Lev. The external theorem proves the exact FC statement answer(True) ↔ ∀ m, 0 < m → spectralProductImpliesRightSpectral 3 m with the definitions imported from Formal Conjectures. The argument shows that $A$ weakly tiles its complement, is a centrally symmetric zonotope whose belts have four or six facets, and hence is one of five normal forms: a box, a hexagonal prism, or the graphical zonotope of $C_4$, $K_4-e$, or $K_4$. The box and prism follow from Greenfeld–Lev's product theorems; for the other three solids the dual Voronoi cell supplies the windows required by Greenfeld–Lev's window theorem, whose existence they had left as a conjecture for non-cylindrical three-dimensional parallelohedra. The fixed repository reports FC-compatible (Lean v4.33.1) and standalone Lean4Web (v4.35.0-rc3) checks without errors or warnings, with #print axioms listing only propext, Classical.choice, and Quot.sound and no sorryAx; these checks were not rerun for this site update. Problem 7.1 and the statement for convex bodies in dimension $n\ge4$ remain open. As of 2026-09-25, PR #6590 is open and unmerged, its completed upstream checks pass, and no independent human review is recorded.

Erdős #143(i) — lower-density status correctionDensity of well-separated sets of reals

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Let $A\subseteq(1,\infty)$ be countably infinite with $\lvert kx-y\rvert\ge1$ for all distinct $x,y\in A$ and all integers $k\ge1$. Must $\liminf_{x\to\infty}\lvert A\cap[1,x]\rvert/x=0$? — Formal Conjectures: Erdős Problem 143(i)
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Koukoulopoulos, Lamzouri, and Lichtman's Erdős's integer dilation approximation problem and GCD graphs shows that a well-separated set has upper logarithmic density $0$, that is, $\sum_{a\in A,\,a\le x}1/a=o(\log x)$; by partial summation its lower density is $0$. The proposed answer to erdos_143.parts.i is True, while erdos_143.parts.ii remains open. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #348 — deletion-threshold status correctionComplete sequences robust to deleting m but not n terms

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. For which $0\le m < n$ is there a complete nondecreasing sequence of integers that remains complete after removing any $m$ terms but is not complete after removing any $n$ terms? — Formal Conjectures: Erdős Problem 348
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Jesse Geneson's 2026 preprint Deletion thresholds and exponential examples for complete sequences proves that the pairs $0\le m < n$ admitting such a sequence are exactly those with $m\le1$, answering Erdős Problem 348. The proposed answer set is $\{(m,n) : m < n,\ m\le1\}$; the issue flags that this should be double-checked against the Formal Conjectures encoding of deletion by Function.updateFinset … 0. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #413(ii) — ε·ω barrier status correctionInfinitely many barriers for ε·ω(n)

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Call $n$ a barrier for $f$ if $m+f(m)\le n$ for all $m < n$. Is there some $\varepsilon > 0$ such that infinitely many $n$ are barriers for $\varepsilon\,\omega$, where $\omega(m)$ is the number of distinct prime factors of $m$? — Formal Conjectures: Erdős Problem 413(ii)
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Theorem 1.3 of Cheuk Fung Lau's 2026 preprint On the Number of Prime Factors of Consecutive Integers gives infinitely many $n$ with $\omega(n-k)\le C\log k$ for all $1 < k < n$. Since $\log k\le k-1$, each such $n-1$ is a barrier for $\omega/C$, so the proposed answer to erdos_413.parts.ii is True. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #817 — negative-answer status correctionSubset sums free of three-term progressions

LiteratureStatusCorrectionAdditive CombinatoricsOPEN
Problem. Let $g_3(n)$ be the least $N$ such that some $n$-element set $A\subseteq\{1,\ldots,N\}$ has subset sums containing no non-trivial three-term arithmetic progression. Is $g_3(n)\gg3^n$? — Formal Conjectures: Erdős Problem 817
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Simone Costa's 2026 preprint A negative answer to the Erdős-Sárkőzy question proves $\liminf_{n\to\infty}g_3(n)/3^n=0$. Hence $3^n=O(g_3(n))$ fails, and the proposed answer to erdos_817 is False. The issue also notes that the author reports a Lean proof starting from this declaration. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1046 — lemniscate-diameter status correctionDiameter of connected polynomial lemniscates

LiteratureStatusCorrectionComplex AnalysisOPEN
Problem. If $f$ is a monic polynomial and $\{z : \lvert f(z)\rvert\le1\}$ is connected, must its diameter be at least $2$? — Formal Conjectures: Erdős Problem 1046 diameter variant
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that for monic $f$ the set $K=\{z : \lvert f(z)\rvert\le1\}$ has logarithmic capacity $1$, while every compact set satisfies the classical inequality $\operatorname{cap}(K)\le\operatorname{diam}(K)/2$. Hence $\operatorname{diam}(K)\ge2$ even without the connectedness hypothesis, and the proposed answer to erdos_1046.variants.diameter is True. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1084 — planar triangular-lattice status correctionPlanar contact numbers of triangular-lattice point sets

LiteratureStatusCorrectionDiscrete GeometryOPEN
Problem. Let $f_2(N)$ be the maximum number of pairs at distance exactly $1$ among $N$ points in the plane whose pairwise distances are all at least $1$. Is $f_2(3n^2+3n+1)=9n^2+3n$? — Formal Conjectures: Erdős Problem 1084 triangular case
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Harborth (Elem. Math. 29 (1974), 14–15) proved $f_2(N)=\lfloor3N-\sqrt{12N-3}\rfloor$, as recorded on erdosproblems.com. For $N=3n^2+3n+1$ one has $12N-3=(6n+3)^2$, so the formula gives exactly $9n^2+3n$, the value in erdos_1084.variants.triangular_optimal_d2. The requested correction is to mark the variant solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1106(i) — known-solution status correctionDistinct prime factors of products of partition numbers

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Let $p(n)$ be the partition function and $F(n)$ the number of distinct prime factors of $\prod_{i=1}^n p(i)$. Does $F(n)\to\infty$? — Formal Conjectures: Erdős Problem 1106(i)
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Schinzel and Wirsing (1987) proved $F(n)\gg\log n$, and that Ono (2000) showed every prime divides some partition number $p(n)$, as recorded on erdosproblems.com. Either result gives $F(n)\to\infty$, so the proposed answer to erdos_1106.parts.i is True; erdos_1106.parts.ii remains open. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1133 — robust-interpolation status correctionRobust polynomial interpolation on [−1, 1]

LiteratureStatusCorrectionApproximation TheoryOPEN
Problem. For every $C > 0$, is there $\varepsilon > 0$ such that, for all large $n$ and all nodes $x_1,\ldots,x_n\in[-1,1]$, some values $y_1,\ldots,y_n\in[-1,1]$ force every polynomial of degree less than $(1+\varepsilon)n$ that takes the value $y_i$ at at least $(1-\varepsilon)n$ of the nodes to exceed $C$ in absolute value somewhere on $[-1,1]$? — Formal Conjectures: Erdős Problem 1133
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Theorem 2.1 of Jia-Qi Yang's 2026 preprint Erdős's Robust Polynomial Interpolation Problem: the Optimal Exponential Scale, specialised to $\rho=0$, contains the statement encoded by erdos_1133, including configurations with repeated nodes. The proposed answer is True. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1167 — finite-target status correctionNegative stepping-up for partition relations with finite targets

LiteratureStatusCorrectionSet TheoryOPEN
Problem. Let $r\ge2$, let $\lambda$ be an infinite cardinal, and let the targets $\kappa_\alpha$ ($\alpha < \gamma$) be finite. Does $2^\lambda\to(\kappa_\alpha+1)^{r+1}_{\alpha<\gamma}$ imply $\lambda\to(\kappa_\alpha)^{r}_{\alpha<\gamma}$? — Formal Conjectures: Erdős Problem 1167 finite targets
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that the case of Erdős Problem 1167 in which all targets $\kappa_\alpha$ are finite is proved in Erdős, Hajnal, Máté, and Rado's Combinatorial Set Theory (1984) via the negative stepping-up lemma, as recorded on erdosproblems.com. The issue notes that the Formal Conjectures variant's omission of the hypothesis $\kappa_\alpha > r$ is harmless, and requests marking erdos_1167.variants.finite_targets solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1207 — isosceles-free subset status correctionLarge isosceles-free subsets of planar point sets

LiteratureStatusCorrectionDiscrete GeometryOPEN
Problem. Let $P_2(n)$ be the largest $m$ such that every set of $n$ points in the plane contains $m$ points spanning no isosceles triangle. Is $P_2(n) < n^{1-c}$ for some $c > 0$? — Formal Conjectures: Erdős Problem 1207
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Lee, Pohoata, and Zhu's 2026 preprint The Minkowski grid has robustly many repeated distances constructs $n$-point planar sets in which every subset $A$ repeats some distance $\gtrsim\lvert A\rvert^2/n^{1-\delta}$ times. An isosceles-free set repeats each distance at most $\lvert A\rvert$ times, so every isosceles-free subset has size $\lesssim n^{1-\delta}$, and the proposed answer to erdos_1207 is True. This follows the earlier forum literature note on the same preprint. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Green Problem 1 — known-solution status correctionSum-free subsets beyond n/3

LiteratureStatusCorrectionAdditive CombinatoricsOPEN
Problem. Does every set $A$ of $n$ positive integers contain a sum-free subset of size at least $n/3+\Omega(n)$, where $\Omega(n)\to\infty$? — Formal Conjectures: Green Problem 1
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Theorem 1.2 of Benjamin Bedert's 2025 preprint Large sum-free subsets of sets of integers via $L^1$-estimates for trigonometric series gives a sum-free subset of size at least $n/3+c\log\log n$ in every set of $n$ integers, and that Green's list marks Problem 1 as solved. Taking $\Omega(n)=c\log\log n$, the proposed answer to green_1 is True. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Green Problem 14 — W(3,29) and W(3,30) status correctionTwo-colour van der Waerden numbers W(3, r)

LiteratureStatusCorrectionCombinatoricsOPEN
Problem. Let $W(3,r)$ be the least $N$ such that every red–blue colouring of $\{1,\ldots,N\}$ has a red three-term or a blue $r$-term arithmetic progression. Is $W(3,29)=868$, and is $W(3,30)=903$? — Formal Conjectures: Green Problem 14 van der Waerden values
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that §6.3 of Chowdhury, Codel, and Heule's 2023 paper A Linear Weight Transfer Rule for Local Search (NFM 2023) solves the SAT instances Green-29-868 and Green-30-903, giving $W(3,29)\ge869$ and $W(3,30)\ge904$. These refute the conjectured values $W(3,29)=868$ and $W(3,30)=903$, so the proposed answer is False for both W_3_29_eq and W_3_30_eq. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Green Problem 27 — lower-bound status correctionSmallest sets in ℤ/pℤ with no unique sums

LiteratureStatusCorrectionAdditive CombinatoricsOPEN
Problem. Let $m(p)$ be the size of the smallest $A\subseteq\mathbb Z/p\mathbb Z$ with $\lvert A\rvert\ge2$ in which no element of $A+A$ has a unique representation. Improve Bedert's lower bound $m(p)\gg\log p\cdot\sqrt{\log\log\log p}/\log\log\log\log p$ along primes. — Formal Conjectures: Green Problem 27 lower bound
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Theorem 1.1 of Jiao-Long Cao and Ye Yuan's 2026 preprint A Second-Logarithm Lower Bound for Sets with No Unique Sums proves $m(p)\gg\log p\,\log\log p$. This improves Bedert's bound recorded as lowerBest, so the answer function $\log p\cdot\log\log p$ works for green_27.lower; green_27.upper remains open. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Green Problem 35 — lower-bound status correctionAutoconvolution norms of densities on [0, 1]

LiteratureStatusCorrectionHarmonic AnalysisOPEN
Problem. Let $c(p)$ be the infimum of $\lVert f\ast f\rVert_p$ over nonnegative $f$ supported on $[0,1]$ with $\int f=1$. Improve the known lower bound $\sqrt{4/7}$ at $p=2$ or $0.64$ at $p=\infty$. — Formal Conjectures: Green Problem 35 lower bound
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that the inequality $\lVert f\ast f\rVert_2^2\ge0.574575$ for unit-interval densities, from Corollary 2.8 of Martin and O'Bryant's The Symmetric Subset Problem in Continuous Ramsey Theory (see also White's An almost-tight $L^2$ autoconvolution inequality and Rechnitzer's The first 128 digits of an autoconvolution inequality), gives $c(2)\ge0.758 > \sqrt{4/7}\approx0.756$. A valid answer is therefore $0.758$ at $p=2$ and $0$ elsewhere. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Green Problem 35 — upper-bound status correctionAutoconvolution norms of densities on [0, 1]

LiteratureStatusCorrectionHarmonic AnalysisOPEN
Problem. Let $c(p)$ be the infimum of $\lVert f\ast f\rVert_p$ over nonnegative $f$ supported on $[0,1]$ with $\int f=1$. Improve the best known upper bound $c(\infty)\le0.7516$. — Formal Conjectures: Green Problem 35 upper bound
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that the step function found by TTT-Discover (Learning to Discover at Test Time), whose autoconvolution ratio is $1.50286$, gives $c(\infty)\le0.75143 < 0.7516$ after rescaling to $[0,1]$; the issue adds that the GGTW step function itself already gives $0.75158$. A valid answer is $0.7515$ at $p=\infty$ and $\infty$ elsewhere, and the issue notes that OptimizationConstants/1a.lean already records $C_{1a}\le1.5029$ as solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #123 — snug {2,3,5}-smooth variant status correctionSnug sums of distinct numbers of the form 2ᵏ3ˡ5ᵐ

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. For every $\varepsilon > 0$, can all large integers $n$ be written as sums of distinct integers $b_1 < \cdots < b_t$ of the form $2^k3^l5^m$ with $b_t < (1+\varepsilon)b_1$? — Formal Conjectures: Erdős Problem 123 snug variant
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that the sorry-free Lean theorem erdos123_dcomplete_real in the public repository antoshashakov/Principia-Math-Solutions, specialised to $(a,b,c)=(2,3,5)$ with $1 < \rho < \min(1+\varepsilon,2)$, gives the snug statement: every large $n$ is a sum of distinct $\{2,3,5\}$-smooth numbers lying in some interval $[x,\rho x)$. The proposed answer to erdos_123.variants.powers_2_3_5_snug is True. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1054 — parts (i) and (ii) status correctionSums of the smallest divisors of m

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Let $f(n)$ be the least $m$ such that $n$ is the sum of the $k$ smallest divisors of $m$ for some $k\ge1$. Is $f(n)=o(n)$, or at least $f(n)=o(n)$ for almost all $n$? — Formal Conjectures: Erdős Problem 1054
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Tao's argument, together with Kovač's bound $\#\{n\le X : f(n)\le\delta n\}\ll\delta^2X$, disproves both parts. The erdosproblems.com page for Problem 1054 notes that Tao disproved the claim $f(n)=o(n)$ by showing that $\{n : f(n)\le\delta n\}$ has upper density $\ll\delta^2$. The same bound rules out $f(n)=o(n)$ on a set of density $1$, so the proposed answer is False for both erdos_1054.parts.i and erdos_1054.parts.ii. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1101(ii) — known-solution status correctionGood sequences with subexponential growth

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Is there a good sequence $u_1 < u_2 < \cdots$ (pairwise coprime, with convergent reciprocal sum and the required gap condition) that grows subexponentially, that is, with $\log u_n=o(n)$? — Formal Conjectures: Erdős Problem 1101(ii)
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Peter Li's subexponential good sequence, posted in the Erdős Problems forum thread for Problem 1101 on 27 April 2026 with help from GPT-5.5 Pro and reported in a reply to pass a standard check, settles erdos_1101.parts.ii. The requested correction is to mark it solved; erdos_1101.parts.i remains open. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1110 — known-solution status correctionPairwise coprime integers not representable by pᵏqˡ sums

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Let $p > q\ge2$ be coprime with $\{p,q\}\ne\{2,3\}$, and call $n$ representable if it is a sum of integers $p^kq^l$ none of which divides another. Are there infinitely many pairwise coprime non-representable integers? — Formal Conjectures: Erdős Problem 1110
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Yu and Chen's 2022 paper On a conjecture of Erdős and Lewin (J. Number Theory) handles all pairs $(p,q)$ except $(5,2)$, $(9,2)$, and $(5,3)$, and that a public proof in the GitHub repository Apiros3/erdos1110, with a Lean formalization, treats these three exceptional pairs. Together they give infinitely many pairwise coprime non-representable integers for every admissible pair, so the proposed answer to erdos_1110 is True. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #1212 — known-solution status correctionPaths to infinity in the visible lattice that avoid prime pairs

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. In the graph on coprime pairs $(x,y)\in\mathbb N^2$ that joins pairs differing by $\pm1$ in one coordinate, is there a path to infinity along which every vertex has $\min(x,y) > 1$ and at least one composite coordinate? — Formal Conjectures: Erdős Problem 1212
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that Alex Chengyu Li's 2026 working paper Infinite paths in the composite-restricted visible lattice, together with its Lean formalization in the GitHub repository crabsatellite/erdos-1212-visible-lattice-path, constructs a path to infinity of the required kind. The proposed answer to erdos_1212 is True. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

TxGraffiti Conjecture 3 — counterexample status correctionIndependent domination versus saturation in regular graphs

LiteratureStatusCorrectionGraph TheoryOPEN
Problem. Every $r$-regular graph $G$ with $r\ge1$ satisfies $i(G)\le\mu^*(G)$, where $i(G)$ is the independent domination number and $\mu^*(G)$ the minimum size of a maximal matching. — Formal Conjectures: TxGraffiti Conjecture 3
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that a public 48-vertex connected cubic graph, released in the GitHub repository ipitchford/txgraffiti-order48-successor, has $i(G)=16 > \mu^*(G)=15$, and that an earlier 50-vertex cubic counterexample is recorded by Evidence Press. Both releases present themselves as certificate-backed candidate counterexamples; either graph falsifies tx_graffiti_conjecture_3, and the issue proposes recording the statement as false. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Erdős #457 — q(n, log n) variant status correctionLeast prime not dividing a product of consecutive integers

LiteratureStatusCorrectionNumber TheoryOPEN
Problem. Let $q(n,k)$ be the least prime not dividing $\prod_{1\le i\le k}(n+i)$. Is there $\varepsilon > 0$ such that $q(n,\log n)\ge(2+\varepsilon)\log n$ for infinitely many $n$? — Formal Conjectures: Erdős Problem 457 q(n, k) variant
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that erdos_457 is already tagged research solved with answer(True): for some $\varepsilon > 0$ there are infinitely many $n$ such that every prime $p\le(2+\varepsilon)\log n$ divides $\prod_{1\le i\le\log n}(n+i)$. For each such $n$ the least prime not dividing the product exceeds $(2+\varepsilon)\log n$, so the same $\varepsilon$ answers erdos_457.variants.qnk positively without any new mathematical claim. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Monochromatic Quantum Graph — three N = 6 status correctionsMonochromatic quantum graph equations for N = 6

LiteratureStatusCorrectionQuantum InformationOPEN
Problem. For $N=6$, is there no solution of the monochromatic quantum graph equation system over $\mathbb C$ for any $D\ge3$, and none over $\mathbb R$ for $D=3$ or for any $D\ge3$? — Formal Conjectures: Monochromatic Quantum Graph N = 6
  • 2026-09-25 formal conjectures literature-status report
    Reported in issue #6591 that eqSystem6_no_solution_d3, solved by Formal Conjectures PR #4610, implies eqSystem6_no_solution_ge3, eqSystem6_no_solution_d3_real, and eqSystem6_no_solution_ge3_real by restricting colours and mapping real solutions into $\mathbb C$. The requested correction is to mark all three declarations solved. The omnibus issue explicitly describes its list as AI-assisted and possibly mistaken; this entry therefore records a literature-status report pending Formal Conjectures maintainer verification, not an independent proof claim.

Monochromatic Quantum Graph — N=10, D=8 over ℂThe equality case D = N − 2 for ten-vertex monochromatic quantum graphs

LeanProofQuantum InformationGraph TheoryOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures source before PR #6607 marks MonochromaticQuantumGraph.eqSystem10_no_solution_d8 research open with answer(sorry). PR #6607 changes only this target to research solved with answer(True) and links the external proof; the cases $N=10$, $3\le D\le7$ and the general conjecture eqSystem_no_solution_ge6_ge3 remain open.
  • 🔗 Source problem Formal Conjectures: open N=10, D=8 target
Problem. For $N=10$ and $D=8$, does the monochromatic quantum graph equation system have no solution over $\mathbb{C}$? — Formal Conjectures: open N=10, D=8 target
  • 2026-09-26 formal conjectures affirmative-proof submission
    Submitted Formal Conjectures PR #6607 with an external Lean 4 proof that the monochromatic quantum graph equation system has no solution for $N=10$ vertices and $D=8$ colours over $\mathbb{C}$, and more generally over every integral domain. The external theorem proves the exact FC statement answer(True) ↔ ¬ ∃ W : WeightsN 10 8 ℂ, EqSystemN 10 8 W with the definitions imported from Formal Conjectures. The sharp bound $D\le N-2$ of the earlier PR #4661 already excludes $D=9$ for $N=10$; this proof settles the equality case $D=N-2$. Building on that sharp-bound development, the argument selects for each vertex and colour a neighbour whose weight matrix is supported on a single column, splits the vertices into mutual pairs and unpaired vertices, and counts the colours that are bad for some pair to find at least three clean colours; the three edge-disjoint perfect matchings they induce yield a non-constant colouring with a nonzero matching sum, a contradiction. The fixed repository reports kernel-checked FC-compatible (Lean v4.33.1) and standalone Lean4Web (v4.35.0-rc3) builds, with #print axioms listing only propext, Classical.choice, and Quot.sound and no sorryAx; these builds were not rerun for this site update. The cases $N=10$ with $3\le D\le7$, the $D=3$ cases for $N=8,12,14,16$, and the general Krenn–Gu conjecture remain open. As of 2026-09-26, PR #6607 is open and unmerged, its upstream build checks have not yet run, and no independent human review is recorded.

Erdős #522 — almost-sure convergence for ±1 and 0/1 coefficientsRoots of random polynomials in the closed unit disk

LeanProofProbabilityOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures source before PR #6608 marks both Erdos522.erdos_522 and Erdos522.erdos_522.variants.zero_one research open with answer(sorry). PR #6608 changes both targets to research solved with answer(True) and links the external proofs; the already solved number_real_roots and yakir_solution are unchanged. erdosproblems.com still lists Problem 522 as open and records Yakir's convergence in probability.
  • 🔗 Source problem Formal Conjectures: open Erdős 522 targets
Problem. Let $f(z)=\sum_{0\le k\le n}\epsilon_k z^k$ with independent coefficients $\epsilon_k$ uniform on $\{-1,1\}$ (or on $\{0,1\}$), and let $R_n$ be the number of roots of $f$ in $\{z\in\mathbb C : \lvert z\rvert\le1\}$. Is it true that $R_n/(n/2)\to1$ almost surely? — Formal Conjectures: open Erdős 522 targets
  • 2026-09-26 formal conjectures affirmative-proof submission
    Submitted Formal Conjectures PR #6608 with an external Lean 4 proof of both open Erdős 522 declarations: for one infinite sequence of independent coefficients uniform on $\{-1,1\}$, and separately on $\{0,1\}$, the number $R_n$ of roots of $f_n(z)=\sum_{k\le n}\epsilon_kz^k$ in the closed unit disk satisfies $R_n/(n/2)\to1$ almost surely. The external theorems prove the exact FC statements with the KacCoefficients definitions imported from Formal Conjectures. Yakir had proved the convergence in probability. Informal almost-sure arguments for $\pm1$ coefficients had already been given in April 2026 by P. Chojecki and, independently, by Y. Kwon and J. Zou, both shared in the Erdős Problems thread, and the repository reports no earlier proof for $0/1$ coefficients; this contribution is a kernel-checked proof of both. The argument reduces almost-sure convergence to summable deviation probabilities, brackets $R_n$ by Jensen's formula at three radii, and controls the resulting logarithmic averages by a small-ball estimate from separation, martingale concentration on the cube, and a Lindeberg replacement with Abel summation, without Berry–Esseen or log-Sobolev inputs. The fixed repository reports kernel-checked FC-compatible (Lean v4.33.1) and standalone Lean4Web (v4.35.0-rc3) builds, with #print axioms listing only propext, Classical.choice, and Quot.sound for both theorems; these builds were not rerun for this site update. Rates of convergence and other coefficient distributions are not claimed. As of 2026-09-26, erdosproblems.com still lists Problem 522 as open, PR #6608 is open and unmerged, its completed upstream checks pass while the project build is still running, and no independent human review is recorded.
  • 2026-09-26 erdős problems forum report
    Reported the same Lean proofs in Erdős Problem 522 discussion post #9188, linking the Formal Conjectures submission, public proof repository, and Lean4Web artifact, and noting that they answer the question under either coefficient reading. The PDF capture taken on 2026-09-26 shows the problem still OPEN. This is a public report of the same result, not a second independent proof.

Maximal rank of 3 × 3 × 5 tensors — mrank(3, 3, 5) = 6Maximal rank of complex 3 × 3 × 5 tensors

LeanProofLinear AlgebraOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures source before PR #6616 marks Arxiv.«0805.3777».isMaxRank_three_three_five research open with answer(sorry), describing it as the one undetermined entry of the table of $\operatorname{mrank}(3,3,p)$ for $p\le9$. PR #6616 changes only this target to research solved with answer(6) and links the external proof; isMaxRank_cube and Friedland's generic-rank conjecture remain open.
  • 🔗 Source problem Formal Conjectures: open mrank(3, 3, 5) target
Problem. Determine the maximal rank $\operatorname{mrank}(3,3,5)$ of a complex $3\times3\times5$ tensor, previously known only to be $6$ or $7$. — Formal Conjectures: open mrank(3, 3, 5) target
  • 2026-09-26 formal conjectures exact-value proof submission
    Submitted Formal Conjectures PR #6616 with an external Lean 4 proof that $\operatorname{mrank}(3,3,5)=6$: every complex $3\times3\times5$ tensor has rank at most $6$, and some tensor has rank exactly $6$. Previously only $\operatorname{mrank}(3,3,5)\in\{6,7\}$ was known, the one undetermined entry of the table of $\operatorname{mrank}(3,3,p)$ for $p\le9$ in Bruzda, Friedland, and Życzkowski's survey. The external theorem proves the exact FC statement IsMaxRank [3, 3, 5] answer(6) with the IsMaxRank definition imported from Formal Conjectures, and also derives the file's recorded bounds $6\le\operatorname{mrank}(3,3,5)\le7$. The proof works with the span of the tensor's five slices in $M_3(\mathbb C)$: the trace-zero upper-triangular matrices give the lower bound, while the upper bound combines a reflexivity lemma for spans whose determinantal cubic is squarefree, a Bertini-type lemma for pencils of plane cubics, a case analysis of the degenerate configurations, and Gerstenhaber's bound on spaces of nilpotent matrices. The fixed repository reports kernel-checked FC-compatible (Lean v4.33.1) and standalone Lean4Web (v4.35.0-rc3) builds, with #print axioms listing only propext, Classical.choice, and Quot.sound; these builds were not rerun for this site update. The result concerns complex tensors only, and the repository reports no earlier proof in the literature while noting that unpublished work cannot be excluded. As of 2026-09-26, PR #6616 is open and unmerged, its completed upstream checks pass while the project build is still running, and no independent human review is recorded.

Das–Dey–Sharma Conjecture 5.5 — Lean proof without the classificationSolvability from the number of cyclic subgroups

LeanProofGroup TheoryOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures source before PR #6621 marks Arxiv.«2604.08040».solvable_of_cyc_lt research open with answer(sorry), although version 2 of the source paper proves it as Theorem 7.1, as reported earlier in issue #6448. PR #6621 changes it to research solved with answer(True), cites that theorem, and links the external Lean proof.
  • 🔗 Source problem Formal Conjectures: open Conjecture 5.5 target
Problem. If a finite group $G$ has fewer than $2^{t+2}$ cyclic subgroups, where $t$ is the number of distinct prime divisors of $\lvert G\rvert$, then $G$ is solvable. — Formal Conjectures: open Conjecture 5.5 target
  • 2026-09-26 formal conjectures lean proof submission
    Submitted Formal Conjectures PR #6621, which marks Das, Dey, and Sharma's Conjecture 5.5 solved and links an external Lean 4 proof: every finite group with fewer than $2^{\omega(\lvert G\rvert)+2}$ cyclic subgroups is solvable, and the bound is sharp because $A_5$ has exactly $2^{3+2}=32$ cyclic subgroups. The statement itself was already proved as Theorem 7.1 of the second version of arXiv:2604.08040, with C. Galindo, through lower bounds for every finite nonabelian simple group obtained from the classification of finite simple groups, as reported earlier in issue #6448; Das and Sharma had also settled the range below 50 cyclic subgroups. The contribution here is a kernel-checked proof by a different route that avoids the classification: a minimal counterexample has no solvable normal subgroup, root counts and families of cyclic subgroups with a fixed 2-part bound its Sylow 2-subgroup, and the Feit–Thompson and Gorenstein–Walter theorems handle the remaining cases. The complete development, about 1.23 million lines including Feit–Thompson and Gorenstein–Walter closures vendored from the Qiuzhen-CFSG project, proves the exact FC statement with the definitions imported from Formal Conjectures; the repository reports a from-scratch build of all 10,749 jobs with only propext, Classical.choice, and Quot.sound in every axiom report, which was not rerun for this site update. A standalone Lean4Web file of about 7,000 lines proves the statement from the Feit–Thompson theorem taken as an explicit hypothesis, stated as the Lean Eval problem feit_thompson. As of 2026-09-26, PR #6621 is open and unmerged, its completed upstream checks pass while the project build is still running, and no independent human review is recorded.

Green Problem 22 — improved monochromatic sum-product boundMonochromatic x + y and xy in r-colourings of {1, …, N}

LeanProofAdditive CombinatoricsOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures source before PR #6632 marks Green22.green_22 research open with answer(sorry), asking for a bound that is $o(\exp\exp(r^{50}))$. PR #6632 fills the answer with $\lceil\exp\exp(r^{43})\rceil$, marks it research solved, and links the external proof; the true growth rate of $N_0(r)$ remains open.
  • 🔗 Source problem Formal Conjectures: open Green 22 target
Problem. Let $N_0(r)$ be the least $N$ such that every $r$-colouring of $\{1,\ldots,N\}$ contains integers $x,y\ge3$ with $x+y$ and $xy$ in $\{1,\ldots,N\}$ of the same colour. Find a bound for $N_0(r)$ that improves the Green–Sawhney bound $\exp\exp(r^{50})$. — Formal Conjectures: open Green 22 target
  • 2026-09-27 formal conjectures improved-upper-bound submission
    Submitted Formal Conjectures PR #6632 with an external Lean 4 proof that $N_0(r)\le\lceil\exp\exp(r^{43})\rceil$ for all sufficiently large $r$, where $N_0(r)$ is the least $N$ such that every $r$-colouring of $\{1,\ldots,N\}$ has $x,y\ge3$ with $x+y$ and $xy$ of the same colour. Since $\lceil\exp\exp(r^{43})\rceil=o(\exp\exp(r^{50}))$, this improves the bound of Green and Sawhney, and the external theorem proves the exact FC statement with the definitions imported from Formal Conjectures. Green and Sawhney had remarked that minor numerical tweaks would give a slightly smaller exponent than $50$, so a qualitative improvement was anticipated. The contributions here are the explicit exponent $43$, obtained by keeping all colours and initial indices together (positivity for the colour sum, energy increments over a subpartition of unity, and random signs) instead of pigeonholing one colour class, and a complete kernel-checked proof that includes Green and Sawhney's inverse theorem (their Proposition 5.1) by an elementary route without the Siegel–Walfisz theorem, the prime number theorem, or exponential sums over primes. The fixed repository reports kernel-checked FC-compatible (Lean v4.33.1) and standalone Lean4Web (v4.35.0-rc3) builds, with #print axioms listing only propext, Classical.choice, and Quot.sound; these builds were not rerun for this site update. The true order of growth of $N_0(r)$, between exponential and doubly exponential, remains open. As of 2026-09-27, PR #6632 is open and unmerged, its completed upstream checks pass while the project build is still running, and no independent human review is recorded.

Kourovka 21.149 — order automorphisms of Dlab groups not induced by conjugationOrder automorphisms of Dlab groups and conjugation in larger Dlab groups

LeanProofStatementGroup TheoryOPEN
  • 🟢 Status when posted OPEN — Kourovka Notebook, 21st edition (arXiv:1401.0300v46) lists the revised question as Problem 21.149, which Formal Conjectures did not yet contain. Gong, Yang, and Zeng had answered it affirmatively for rank-one slope groups and order-preserving embeddings. The original, weaker question, an order automorphism that is not inner, had been answered by Aristotle (Harmonic) with a Lean proof, which led to the revision.
  • 🔗 Source problem Kourovka Notebook Problem 21.149
Problem. Does some Dlab group have an order automorphism that cannot be realized as conjugation by an element of a Dlab group, possibly a larger one into which it embeds? — Kourovka Notebook Problem 21.149
  • 2026-09-27 formal conjectures statement and proof submission
    Submitted Formal Conjectures PR #6643, which adds Kourovka Notebook Problem 21.149 to Formal Conjectures and marks it solved with an external Lean 4 proof. Following Dlab and Gong–Yang–Zeng, the statement defines the six Dlab groups as subgroups of the order automorphisms of $\mathbb R$, together with Dlab's order, and asks for an order automorphism $\alpha$ of some Dlab group $G$ such that no injective homomorphism $e$ of $G$ into any Dlab group $A$ and no $u\in A$ satisfy $e(\alpha(f))=u^{-1}e(f)u$ for all $f\in G$. The proof takes $G=D_{\langle2\rangle}(I)$ and $\alpha(f)=h^{-1}fh$, where $h$ is built from bumps on $[0,1]$ whose breakpoints accumulate at both ends, so that $h$ lies in no Dlab group; simplicity of local commutator subgroups (Higman), first action components, and a staircase along the fixed points of $h$ then rule out every conjugating element. The same development proves the statement for every nontrivial slope group. Gong, Yang, and Zeng had answered the revised question for rank-one slope groups and order-preserving embeddings using classifications of the orders of Dlab groups; this proof allows any slope group and any injective homomorphism and uses no order classification. The original, weaker form of the question, an order automorphism that is not inner, was answered earlier by Aristotle (Harmonic) with a Lean proof; the PR records it as kourovka_21_149.variants.not_inner, linking that proof and a derivation from the main statement. The fixed repository reports warning-free kernel-checked builds, FC-compatible at Lean v4.33.1 and standalone for Lean4Web at v4.35.0-rc3, with #print axioms listing only propext, Classical.choice, and Quot.sound; these builds were not rerun for this site update. As of 2026-09-27, PR #6643 is open and unmerged, its completed upstream checks pass while the project build is still running, and no independent human review is recorded.

Littlewood's mutually touching cylinders — at most sevenMutually touching congruent infinite cylinders

LeanProofStatementDiscrete GeometryComputationOPEN
Problem. What is the maximum number of congruent infinite circular cylinders in $\mathbb R^3$ with pairwise disjoint interiors such that every two of them touch? The maximum-seven formulation asks whether this number is $7$, equivalently whether $7$ is the largest number of lines in $\mathbb R^3$ at pairwise distance $1$. — Formal Conjectures: proposed Littlewood cylinder statement (PR head)
  • 2026-09-30 formal conjectures statement and proof submission
    Submitted Formal Conjectures PR #6741, which adds Littlewood's mutually touching cylinders problem to Formal Conjectures in its maximum-seven form and marks it solved with an external Lean 4 proof. After scaling the radius to $1/2$, the statement says that $7$ is the greatest number of affine lines in $\mathbb R^3$ whose pairwise distances are all $1$, with the geometry written directly in the theorem and no general-position hypothesis. Bozóki, Lee, and Rónyai had constructed seven such cylinders, and the published upper bounds had fallen from $24$ to $18$ (Koizumi), $10$ (Dillon, Koizumi, and Luo), and $9$ (Höfer, 2026), leaving $7\le N\le9$. The Lean development proves both halves. For seven lines, checked rational bounds show that a contraction maps a cube around an approximate solution into itself, so the fixed-point theorem yields an exact real configuration. For eight lines, the geometry forces nondegeneracy and a system of contact and orientation signs obeying Grassmann–Plücker, tetrahedron, closed-polygon, and five-direction constraints; a parity argument excludes a five-label pattern, an inductive graph cover with 5,115 checked extension certificates reduces the remaining cases to 131 contact matrices, and LRAT refutations reconstructed as Lean proof terms exclude each of them. The fixed repository reports a warning-free modular build at Lean 4.33.1 and a standalone 10.3 MB file checked locally with Lean 4.34.1 and on public Lean4Web, with #print axioms listing only propext, Classical.choice, and Quot.sound; these checks were not rerun for this site update. The formal theorem concerns the cylinders' axes, and the correspondence between touching cylinders and lines at distance one is not itself formalized. As of 2026-09-30, PR #6741 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Reverse Littlewood–Offord problem at radius one — the odd-n exponential rateRandom signed sums of planar unit vectors in the unit disc

LeanProofStatementProbabilityOPEN
Problem. For odd $n$, let $F_{2,1}(n)$ be the infimum over unit vectors $v_1,\ldots,v_n\in\mathbb R^2$ of $\mathbb P(\lVert\varepsilon_1v_1+\cdots+\varepsilon_nv_n\rVert_2\le1)$ for independent uniform signs $\varepsilon_i$. Does $\lim F_{2,1}(n)^{1/n}$ exist over odd $n$, and if so, what is it? — Formal Conjectures: proposed reverse Littlewood–Offord statement (PR head)
  • 2026-10-01 formal conjectures statement and proof submission
    Submitted Formal Conjectures PR #6772, which adds the open problem stated after Question 7.1 of Hollom, Portier, and Souza's Double-jump phase transition for the reverse Littlewood–Offord problem and marks it solved with answer(1 / Real.sqrt 2) and an external Lean 4 proof. For odd $n$, $F_{2,1}(n)$ is the least probability, over unit vectors $v_1,\ldots,v_n$ in the plane, that $\lVert\varepsilon_1v_1+\cdots+\varepsilon_nv_n\rVert_2\le1$ for independent uniform signs. Hollom, Portier, and Souza showed that $F_{2,1}(n)$ is at least exponentially small, roughly $0.525^n$, and asked whether $F_{2,1}(n)^{1/n}$ converges; Hollom and Sorkin's construction gives $F_{2,1}(n)\le2^{-\lfloor n/2\rfloor}$. The new lower bound is $F_{2,1}(n)\ge2^{-(n+1)}(1-\rho)^{-((n-3)/2-K_\rho)}$ for every $0<\rho<1/2$, with $K_\rho=\lfloor1152/(1-2\rho)^2\rfloor$, so $F_{2,1}(n)^{1/n}\to1/\sqrt2$ and the Hollom–Sorkin construction is optimal at the exponential scale. The proof normalizes the vectors so that their alternating sum is real, bounds alternating sums of vectors in a half-plane, pairs adjacent vectors around a pivot, and shows by a second-moment argument that random flips of light pairs keep the signed sum in the unit disc with probability at least $1/2$. Both bounds are formalized, and the external theorem proves the exact FC statement. The fixed repository reports a warning-free build at Lean 4.33.1 against Formal Conjectures 137aec5 and a standalone Lean4Web file, with #print axioms listing only propext, Classical.choice, and Quot.sound; these checks were not rerun for this site update. As of 2026-10-01, PR #6772 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Green Problem 41 — Quasi-polynomial pyjama-covering bound

LeanProofStatementDynamical SystemsOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures source before PR #6581 has no quasi-polynomial variant: the exact-value problem Green41.green_41 and the polynomial-bound variant are research open, and the published bound recorded there is Kravitz and Leng's triple-exponential one. PR #6581, which already marked the double-exponential variant solved, adds green_41.variants.quasi_polynomial_bound as research solved with answer(True) and a link to the external proof.
  • 🔗 Source problem Formal Conjectures: proposed quasi-polynomial variant (PR head)
Problem. Is a quasi-polynomial number of rotations, $\exp((\log\varepsilon^{-1})^C)$ for some constant $C$, enough to cover the plane by rotated copies of the pyjama set $\{z : \lvert\operatorname{Re}z-k\rvert\le\varepsilon\text{ for some }k\in\mathbb Z\}$ for all sufficiently small $\varepsilon>0$? — Formal Conjectures: proposed quasi-polynomial variant (PR head)
  • 2026-10-02 formal conjectures statement and proof submission
    Extended Formal Conjectures PR #6581 with a new solved variant green_41.variants.quasi_polynomial_bound and an external Lean 4 proof that $\exp((\log\varepsilon^{-1})^C)$ rotations of the pyjama set suffice, with $C=5$ and $\varepsilon_0=e^{-10^{18}}$. The proof gives the explicit bound $\mathrm{minCopies}\,\varepsilon\le\exp(10^{16}(\log(1/\varepsilon)+2)^4)$ for $0<\varepsilon<1/10$, and on the way the single-exponential bound $\exp(\varepsilon^{-201})$. It builds on the author's double-exponential proof of 2026-09-25, replacing the cyclic orbit of a rotation by base-$p$ digit rotations $(1+p^ki)/(1-p^ki)$, whose common denominator has norm $p^{O(N^2)}$ instead of $p^{p^N}$; using them in Step A and in the Fejér-kernel discrete core makes every parameter $O(\log^2(1/\varepsilon))$. The proofs construct an explicit covering set, and the external theorem proves the exact FC statement with the definitions imported from Formal Conjectures. The fixed repository reports kernel-checked FC-compatible (Lean v4.33.1) and standalone Lean4Web (v4.35.0-rc3) builds, with #print axioms listing only propext, Classical.choice, and Quot.sound; these builds were not rerun for this site update. The exact value of $\mathrm{minCopies}\,\varepsilon$ and Green's polynomial question remain open, and the repository explains why its method cannot reach a polynomial bound. As of 2026-10-02, PR #6581 is open and unmerged with passing upstream checks; a maintainer commented positively on the earlier double-exponential version, and the quasi-polynomial addition has not yet been reviewed.

Ryser's conjecture — 4-partite hypergraphs with matching number at most 3Covers versus matchings in 4-partite 4-uniform hypergraphs

LeanProofStatementCombinatoricsComputationOPEN
Problem. Ryser's conjecture states that every finite $r$-partite $r$-uniform hypergraph $H$ satisfies $\tau(H)\le(r-1)\nu(H)$, where $\nu$ is the matching number and $\tau$ the cover number. For $r=4$: does every such hypergraph with $\nu(H)\le3$ satisfy $\tau(H)\le3\nu(H)$? — Formal Conjectures: proposed Ryser's conjecture statement (PR head)
  • 2026-10-03 formal conjectures statement and proof submission
    Submitted Formal Conjectures PR #6812, which adds Ryser's conjecture $\tau(H)\le(r-1)\nu(H)$ for finite $r$-partite $r$-uniform hypergraphs to Formal Conjectures, with one variant for each case of the current status, and marks the cases $r=4$, $\nu(H)\le3$ solved with external Lean 4 proofs. The cases $\nu=1$ (Gyárfás; Tuza) and $\nu=2$ (White's 2026 preprint, confirming a claim of Tuza) were known. The case $\nu=3$, $\tau(H)\le9$, is not among the known cases of the 2021 survey of DeBiasio, Kamel, McCourt, and Sheats, and the PR states that to the author's knowledge it was open. The proof projects onto pairs of vertex classes and applies Kőnig's theorem to scattered families of edges. A computer-assisted proposition, that $\nu(H)\le2$ together with seven $\{0,1\}$-scattered edges forces $\tau(H)\le5$, is reduced to 232 templates and refuted by certificates: case-split trees for a bitmask checker whose soundness is proved in Lean, and LRAT refutations replayed through Mathlib. All certificates are checked by the Lean kernel with decide +kernel and no native_decide; the only unsafe code is in three LRAT elaborator commands whose output proof terms are kernel checked. The complete development has 2,706 modules (about 180 MB, mostly certificate data, about 7.5 CPU hours to build), and a re-encoded single-file version for Lean4Web checks in about eight minutes; the repository reports only propext, Classical.choice, and Quot.sound in its axiom reports, which were not rerun for this site update. The cases $r=4$ with $\nu\ge4$, $r=5$ with $\nu\ge2$, and every $r\ge6$ remain open. As of 2026-10-03, PR #6812 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Quantum extremal numbers — Qex(9) = 112, with new upper bounds for 10 and 11 qubitsMaximally mixed half-size reductions of 9-, 10- and 11-qubit pure states

LeanProofStatementQuantum InformationCombinatoricsOPEN
Problem. For a pure state $\psi$ of $n$ qubits, let $m_k(\psi)$ be the number of $k$-sets $A$ of qubits whose reduced state is maximally mixed, $\rho_A=I/2^k$, and let $\mathrm{Q}_{ex}(n)=\max_\psi m_{\lfloor n/2\rfloor}(\psi)$. Determine $\mathrm{Q}_{ex}(9)$, known to lie between $112$ and $120$, and $\mathrm{Q}_{ex}(10)$ and $\mathrm{Q}_{ex}(11)$. — Formal Conjectures: proposed quantum extremal number statements (PR head)
  • 2026-10-04 formal conjectures statement and proof submission
    Submitted Formal Conjectures PR #6836, which adds the quantum extremal numbers $\mathrm{Q}_{ex}(n,k)$ of Zhang, Ning, Shi, and Zhang (Extremal maximal entanglement, Phys. Rev. A 111, 052410 (2025)) to Formal Conjectures: the general problem, the known values $\mathrm{Q}_{ex}(4)=4$, $\mathrm{Q}_{ex}(7)=32$ and $\mathrm{Q}_{ex}(8)=56$, and the cases $n=9,\ldots,12$ of the paper's Table I. $\mathrm{Q}_{ex}(n)=\mathrm{Q}_{ex}(n,\lfloor n/2\rfloor)$ is the largest number of maximally mixed $\lfloor n/2\rfloor$-qubit reductions of an $n$-qubit pure state; it equals $\binom{n}{\lfloor n/2\rfloor}$ exactly when an absolutely maximally entangled state exists. The PR marks $\mathrm{Q}_{ex}(9)=112$ as solved with an external Lean 4 proof, and the docstrings of the still-open $\mathrm{Q}_{ex}(10)$ and $\mathrm{Q}_{ex}(11)$ cite the new upper bounds $208$ and $422$, also proved in Lean, which improve the bounds $240$ and $461$ of Table I. The lower bounds $112$, $200$ and $396$ come from graph states in Danielsen's database, as cited in the paper; the upper bounds $112$, $208$ and $422$ are new to the author's knowledge, and the August 2026 follow-up of Zhang, Han, and Zhang still gives $120$ for nine qubits. The proof works only with purities of reduced states: positivity inequalities for two copies of the state, including Rains' shadow inequalities, are proved from explicit integer sums of squares, and extremal set theory finishes each case, with the Kneser graph $KG(9,4)$ for nine qubits and Mathlib's Kruskal–Katona theorem for eleven. The theorems cover every complex pure state, with no stabilizer or graph-state assumption. The matching state is the graph state of H?SvCz}, and the only computations, the cut criterion for its 126 four-sets and two colex-segment sizes, are checked by the Lean kernel with decide +kernel and no native_decide. The external project imports Formal Conjectures at df3f12d, copies the definitions verbatim (an audit script checks the copy character for character), and proves qex_nine : Qex 9 4 = answer(112) with the exact FC statement. The fixed repository reports a warning-free build at Lean 4.33.1 and a standalone Lean4Web file that checks in about 25 seconds of CPU time, with #print axioms listing only propext, Classical.choice, and Quot.sound; these checks were not rerun for this site update. $\mathrm{Q}_{ex}(10)$, $\mathrm{Q}_{ex}(11)$, $\mathrm{Q}_{ex}(12)$ and every $n\ge13$ remain open. As of 2026-10-04, PR #6836 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Bapat–Lal q-permanent monotonicity and da Fonseca's half-line extension — a counterexample of order 144Strict monotonicity in q of q-permanents of positive definite matrices

LeanProofCounterexampleStatementLinear AlgebraOPEN
  • 🟢 Status when posted OPEN — Mitchell (2020), Operators and Matrices 14, 915–919 calls Conjecture 1 difficult even for $n=4$ and, in a remark on p. 917, states da Fonseca's half-line extension (Conjecture 2) for non-diagonal positive definite matrices. Conjecture 1 had been proved for $n\le3$ and for tridiagonal matrices, and da Fonseca's 2018 note (arXiv:1804.02231) states both conjectures. The OpenProblemsInNLA catalogue (MI-18) marks Conjecture 1 partially resolved, after adding in September 2026 a computer-assisted proof of the order-four case that it describes as AI-produced and not independently reviewed. Formal Conjectures did not yet contain either problem.
  • 🔗 Source problem Formal Conjectures: proposed q-permanent statements (PR head)
Problem. For an $n\times n$ complex matrix $A$ and real $q$, the $q$-permanent is $P_q(A)=\sum_{\sigma\in S_n}q^{\ell(\sigma)}\prod_i a_{i\sigma(i)}$, where $\ell(\sigma)$ is the number of inversions of $\sigma$; it equals $\det A$ at $q=-1$ and $\operatorname{per}A$ at $q=1$. Conjecture 1 (Bapat–Lal): is $P_q(A)$ strictly increasing in $q\in[-1,1]$ for every non-diagonal Hermitian positive definite matrix $A$? Conjecture 2 (da Fonseca): does every such $A$ admit some $\varepsilon<-1$ such that $P_q(A)$ is strictly increasing on $(\varepsilon,\infty)$? — Formal Conjectures: proposed q-permanent statements (PR head)
  • 2026-10-05 formal conjectures statement and disproof submission
    Submitted Formal Conjectures PR #6857, which adds the Bapat–Lal $q$-permanent monotonicity conjecture, Conjecture 1 in da Fonseca's The $\mu$-permanent revisited (arXiv:1804.02231), going back to Bapat (1992) and Bapat and Lal (1994), and records its negative answer with an external Lean 4 proof. The $q$-permanent $P_q(A)=\sum_{\sigma}q^{\ell(\sigma)}\prod_i a_{i\sigma(i)}$ equals $\det A$, $\prod_i a_{ii}$ and $\operatorname{per}A$ at $q=-1,0,1$, and the conjecture was motivated by the Hadamard inequality and Marcus's permanental analogue. It was known for $n\le3$ and for tridiagonal matrices, and Mitchell (2020) showed that it is equivalent to its extension to singular positive semidefinite matrices without a zero row; in September 2026 the OpenProblemsInNLA catalogue added a computer-assisted, AI-produced proof of the order-four case. The counterexample starts from a rank-two Gram matrix $H=VV^*$ of order 144 whose rows have an integer and a Gaussian-integer coordinate. Exact identities express the permanent of $H$ and the derivative at $q=1$ through weighted coefficient norms $P$ and $S$ of two polynomials: $2\frac{d}{dq}P_q(H)$ at $q=1$ equals $\binom{144}{2}P-S$, and $S>10300P$, so the derivative is negative. The certificate computations, with $P$ and $S$ of 768 and 772 digits, are checked by the Lean kernel with decide +kernel and no native_decide. By continuity, $A=H+\varepsilon I$ is Hermitian positive definite and non-diagonal for some $\varepsilon>0$, and its $q$-permanent strictly decreases between two points of $[-1,1]$. The formal counterexample is existential in $\varepsilon$ and the two points, and it is complex; the proof does not determine the least order or give a real symmetric counterexample. Targeted searches of arXiv, of works citing Bapat–Lal and Mitchell, and of the web found no earlier counterexample. The fixed repository reports that the shared definitions match the FC file character for character and that the compiled FC target and the proof have definitionally equal types, with a build at Lean 4.33.1 against Formal Conjectures 89294ea, a standalone Lean4Web file at Lean 4.35.0-rc3, and #print axioms listing only propext, Classical.choice, and Quot.sound; these checks were not rerun for this site update. As of 2026-10-05, PR #6857 is open and unmerged, its upstream checks pass, and no independent human review is recorded.
  • 2026-10-05 open-problem catalogue report
    Reported the same Lean disproof in OpenProblemsInNLA issue #329 on entry MI-18, Bapat's q-permanent monotonicity conjecture. MI-18 states the conjecture for non-diagonal Hermitian positive semidefinite matrices with positive diagonal and marks it partially resolved, with the computer-assisted order-four proof. A positive definite counterexample of order 144 lies in that class, so it refutes the arbitrary-order assertion; the issue proposes the status Lean verified (negative resolution) while keeping the order-four and other special-case results. As of 2026-10-05 the issue is open with no maintainer response. This is a report of the same result, not a second proof.
  • 2026-10-05 formal conjectures pr extension
    Extended PR #6857 with da Fonseca's half-line extension, Conjecture 2 of arXiv:1804.02231: does every non-diagonal Hermitian positive definite matrix have some $\varepsilon<-1$ such that its $q$-permanent is strictly increasing on $(\varepsilon,\infty)$? The 2018 note omits the non-diagonal hypothesis, without which diagonal matrices, whose $q$-permanent is constant, would make the statement trivially false; the PR uses the non-diagonal formulation stated in Mitchell's remark on p. 917. Every such half-line contains $[-1,1]$, so the order-144 counterexample also refutes Conjecture 2, and a short Lean proof, whose statement matches the FC file character for character, derives this negative answer from the Conjecture 1 theorem. The FC file also records, without formalizing them, that the note's Conjectures 3 and 4 are already known to be false: Conjecture 3 by Colbrook's counterexample, formalized in Lean by Stepaniants, and Conjecture 4 at $q=1$ by Shchesnovich's 2016 disproof of the permanent-on-top conjecture, with a simpler counterexample by Tran Hoang Anh (2021). The repository's recorded build log predates the Conjecture 2 module, and no build was rerun for this site update; the upstream checks on the updated PR pass, and no independent human review is recorded.

Irrationality exponent of log 2 — μ(log 2) = 2Rational approximation of the natural logarithm of 2

LeanProofStatementNumber TheoryOPEN
Problem. The irrationality exponent $\mu(x)$ of a real number $x$ is the supremum of the exponents $\mu$ for which $0<|x-p/q|<q^{-\mu}$ has infinitely many rational solutions $p/q$; every irrational $x$ has $\mu(x)\ge2$. What is $\mu(\log 2)$? — Formal Conjectures: proposed log 2 statement (PR head)
  • 2026-10-08 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #6899 on 2026-10-07 with 11 pages of proof notes, and on 2026-10-08 submitted PR #6923, which adds the irrationality exponent of $\log 2$ from Bugeaud and Kim (arXiv:2510.02059v2, Section 1) and marks $\mu(\log 2)=2$ solved with an external Lean 4 proof. The FC definition takes the supremum of the positive $\mu$ for which infinitely many reduced fractions $p/q$ with $q>1$ satisfy $0<|x-p/q|<q^{-\mu}$. This supremum is $0$ for an empty or unbounded set, so the value $2$ requires the hard bound $\mu(\log 2)\le2$. The proof adapts the interpolation-determinant method of OpenAI's The irrationality exponent of π is 2 to the interpolation centres $(2^j, jp_i/q_i)$. Assuming arbitrarily good approximations, a geometric interpolation theorem, proved through jet surjectivity on a blow-up and Serre vanishing, gives a nonzero full-row minor; its arithmetic lower bound, from clearing denominators, contradicts an analytic upper bound from the row translation $e^{j\log 2}=2^j$. Of the roughly 105,000 lines of the split Lean proof, about 97,000 are OpenAI's $\mu(\pi)$ library at a pinned revision, credited in the repository's notices, and about 8,800 are specific to $\log 2$. The FC definition is copied verbatim, and the target agrees up to the answer wrapper. The fixed repository reports an audit at Lean 4.34.1 and a single-file Lean4Web edition of about 95,000 lines that passed at Lean 4.35.0-rc4, with #print axioms listing only propext, Classical.choice, and Quot.sound. The proof notes were posted at 13:40 UTC on 2026-10-07. At 14:54 UTC the same day, Liu, Jiang, and Zhang submitted arXiv:2610.10192, a paper proof by a similar method that $\mu(\log r)=2$ for every positive rational $r\ne1$, and on 2026-10-08 R. M. Casper proposed the same statement for all positive rational logarithms with a Lean formalization (FC issue #6924 and PR #6925). As of 2026-10-09, PR #6923 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Kalai's full flag conjecture via the symmetric Funk-volume inequality — at least 2ⁿ n! complete flagsFunk-ball volumes and complete flags of centrally symmetric convex bodies

LeanProofStatementDiscrete GeometryConvex GeometryOPEN
Problem. Does every centrally symmetric convex polytope $P\subset\mathbb R^n$ with nonempty interior, $n\ge1$, have at least $2^n\,n!$ complete flags, that is, chains $F_0\subset F_1\subset\cdots\subset F_{n-1}$ of faces of $P$ with $\dim F_i=i$? — Formal Conjectures: proposed Kalai full flag statement (PR head)
  • 2026-10-08 formal conjectures statement and proof submission
    Submitted Formal Conjectures PR #6926, which adds Kalai's full flag conjecture as discussed by Faifman, Vernicos, and Walsh (arXiv:2306.09268, Section 1) and marks it solved with an external Lean 4 proof: every centrally symmetric convex polytope in $\mathbb R^n$ with $0$ in its interior, $n\ge1$, has finitely many complete flags and at least $2^n\,n!$ of them, the number for the cube. A complete flag is a strictly increasing chain of nonempty proper convex extreme subsets with affine dimensions $0,\ldots,n-1$. The proof goes through a stronger theorem, also proved in Lean: for every origin-symmetric convex body $K\subset\mathbb R^n$ and every $0<\tau<1$, the Holmes–Thompson volume $\int_{\tau K}|(K-x)^\circ|\,dx$ of the Funk ball $\tau K$ is at least $(4\operatorname{artanh}\tau)^n/n!$, the value for the cube. This is the inequality in Faifman, Vernicos, and Walsh's symmetric Funk-volume conjecture (their Conjecture 1.1, without the equality cases), a one-parameter family that interpolates between the symmetric Mahler inequality, its limit as $\tau\to0$, and Kalai's flag conjecture, which the proof obtains at large radius by comparison with the volume growth of polytopal Funk balls, governed by the number of flags as in their Theorem 1.2. OpenAI's symmetric Mahler proof covers only the end $\tau\to0$, and Winter's implication (below) only connects the two ends, so the intermediate radii are new in this proof. The argument is a $\tau$-deformation of the conformal-lens argument in OpenAI's proof of the symmetric Mahler conjecture, using OpenAI's Lean modules for the flux, Stokes and holomorphic-mass steps. The project has about 17,000 lines of new Lean code in 163 modules, together with 228 vendored OpenAI modules (about 21,000 lines) credited in the repository; its definitions and the flag theorem match the FC file character for character, and the Funk-volume theorem has an FC-style statement in the repository. The fixed repository reports kernel replays and an exact-type check of both targets, with #print axioms listing only propext, Classical.choice, and Quot.sound, and a single-file Lean4Web edition that passes at Lean 4.35.0-rc4. A manuscript was posted at 17:03 UTC on 2026-10-07 and the Lean proof on 2026-10-08. About four hours before the manuscript, at 12:51 UTC on 2026-10-07, Winter (arXiv:2610.09990) proved that the Mahler conjecture implies the flag conjecture, so together with OpenAI's claimed symmetric Mahler proof, released on 2026-10-06, his result also yields the flag bound, but not the Funk-volume inequality. As of 2026-10-09, PR #6926 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Frantzikinakis Problem 19 — pointwise convergence of bilinear averages for two commuting transformationsAlmost-everywhere convergence of multiple ergodic averages for commuting maps

LeanProofStatementDynamical SystemsHarmonic AnalysisOPEN
Problem. Let $T$ and $S$ be commuting invertible measure-preserving transformations of a probability space $(X,\mu)$. Do the averages $\frac1N\sum_{n=1}^{N}f(T^nx)\,g(S^nx)$ converge for $\mu$-almost every $x$, for all bounded measurable functions $f$ and $g$? — Formal Conjectures: proposed commuting pointwise convergence statements (PR head)
  • 2026-10-08 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #6937 and submitted PR #6960, which adds Problem 19 of Frantzikinakis' survey Some open problems on multiple ergodic averages (arXiv:1103.3808v3, §5.2.5): for two commuting invertible measure-preserving transformations $T$ and $S$ of a probability space, do the averages $\frac1N\sum_{n<N}f(T^nx)g(S^nx)$ converge almost everywhere for bounded $f$ and $g$? The PR answers it affirmatively with an external Lean 4 proof, and adds a variant for square-integrable $f$ and $g$. The statements quantify over every probability space and every pair of complex-valued functions, without ergodicity, and the repository also proves convergence, to the same limit, for the survey's sums over $n=1,\ldots,N$. The problem was open in general; Frantzikinakis' progress list records it as solved only for distal actions. The proof establishes a planar maximal estimate with growth $C\sqrt m$ for two polynomial-Gaussian kernels, using matrix inequalities, Gaussian heat identities and maximal estimates adapted from OpenAI's preprint Annular variation of the triangular Hilbert transform at the symmetric point (October 5, 2026), which the repository credits. It transfers the estimate to commuting measurable flows to obtain almost-everywhere convergence of continuous-time averages, reaches the interval kernel by an $L^1$ density argument, and recovers the discrete averages through a suspension and an exact phase identity; the $L^2$ case follows by truncation and Birkhoff's theorem. A manuscript was posted on 2026-10-07 (UTC) and the Lean proof on 2026-10-08. The external targets match the FC statements character for character, with True in place of answer(True). The fixed repository reports a 37,884-line proof checked at Lean 4.33.1 with Formal Conjectures' Mathlib revision, 2,095 audited declarations, and a single-file Lean4Web edition at Lean 4.35.0-rc4, with #print axioms listing only propext, Classical.choice, and Quot.sound for both targets. Targeted searches found no earlier proof of the general case. As of 2026-10-09, PR #6960 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Extended Permanence Conjecture — endotactic κ-variable mass-action systems are permanentPermanence of endotactic mass-action systems with time-dependent rates

LeanProofStatementDynamical SystemsOPEN
Problem. A reaction network is endotactic if, for every direction $r$ and every reaction $y\to y'$ with $r\cdot(y'-y)<0$, some reaction $\tilde y\to\tilde y'$ has $r\cdot(\tilde y'-\tilde y)>0$ and $r\cdot\tilde y<r\cdot y$. Craciun, Nazarov and Pantea conjectured that every endotactic $\kappa$-variable mass-action system, with time-dependent rates bounded away from $0$ and $\infty$, is permanent: on each positive stoichiometric class, all trajectories eventually enter a common compact subset of the positive orthant. Is this true? — Formal Conjectures: proposed endotactic permanence statements (PR head)
  • 2026-10-09 formal conjectures statement and proof submission
    Submitted Formal Conjectures PR #7010, which adds the Extended Permanence Conjecture of Craciun, Nazarov, and Pantea (arXiv:1010.3050v2, Section 4) for mass-action systems with natural-number exponents and marks it solved with an external Lean 4 proof. For every endotactic network and every rate path $k(t)\in[\ell,u]^E$ with $0<\ell\le u$, measurable and differentiable off finitely many times on bounded intervals, each positive stoichiometric class has unique global positive solutions and an $\varepsilon>0$ such that every trajectory eventually stays in $[\varepsilon,\varepsilon^{-1}]^d$. A second target strengthens this to one box for all bounded measurable rate paths at once, the differential-inclusion form proposed by Gopalkrishnan, Miller, and Shiu. The power-law version mentioned in the source is not covered. The proof adapts the affine-approximation and trapping method of OpenAI's preprints on weakly reversible mass-action systems with fixed rates (September 25 and October 5, 2026). Endotacticity gives a positive projected flux in every direction; finitely many affine barriers with arbitrarily small tolerance turn these directional estimates into compact plateau traps that are forward invariant for all admissible rates; and a square-root iteration of scales gives late absorption. The new ingredients are the endotactic flux step and the existence, uniqueness and absorption theory for measurable rates. The project has about 13,900 lines of Lean in 121 files and does not reuse OpenAI's Lean code; its definitions and targets match the FC file character for character after renaming the namespace. The fixed repository reports a full build at Lean 4.33.1 with Formal Conjectures' Mathlib revision, an axiom audit of 754 theorems with #print axioms listing only propext, Classical.choice, and Quot.sound for both targets, a check that the compiled FC types agree with the proof's, and a single-file Lean4Web edition at Lean 4.35.0-rc4. Targeted searches found no earlier proof of the endotactic case. As of 2026-10-09, PR #7010 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Auslander–Reiten conjecture and Tachikawa's second conjecture — a counterexample over ℚSelf-extensions of modules over symmetric algebras in characteristic zero

LeanProofCounterexampleStatementRepresentation TheoryOPEN
  • 🟢 Status when posted OPEN — OpenAI, A counterexample to Tachikawa's second conjecture (September 23, 2026) disproves Tachikawa's second conjecture, and a companion OpenAI preprint the Auslander–Reiten conjecture, over fields of characteristic two. On 2026-10-07 Q. Tang (唐琪越) posted an AI-generated note on Hexagon (2610.00143) extending these constructions, without a formalization, to every field containing an element of infinite multiplicative order, which includes $\mathbb Q$. Enomoto and Marczinzik (arXiv:2609.19172) proved that the Tachikawa, Nakayama, generalized Nakayama and Auslander–Reiten conjectures are globally equivalent over any fixed commutative artinian ring. Formal Conjectures did not yet contain Tachikawa's conjectures; its Auslander–Reiten entry is marked open, and open PR #6898 marks it refuted using OpenAI's example.
  • 🔗 Source problem Formal Conjectures: proposed rational Tachikawa statement (PR head)
Problem. The Auslander–Reiten conjecture asserts that a finitely generated module $M$ over an Artin algebra $\Lambda$ with $\operatorname{Ext}^n_\Lambda(M,M\oplus\Lambda)=0$ for all $n\ge1$ is projective. Tachikawa's second conjecture is its self-injective case, where $\operatorname{Ext}^n_\Lambda(M,\Lambda)=0$ holds automatically: $\operatorname{Ext}^n_\Lambda(M,M)=0$ for all $n\ge1$ should force $M$ to be projective. Do these hold for finite-dimensional symmetric algebras over $\mathbb Q$? — Formal Conjectures: proposed rational Tachikawa statement (PR head)
  • 2026-10-10 formal conjectures statement and disproof submission
    Opened Formal Conjectures issue #7020 on 2026-10-09 with a 10-page manuscript, and on 2026-10-10 submitted PR #7067, which adds Tachikawa's second conjecture (Tachikawa, Quasi-Frobenius Rings and Generalizations, LNM 351, Section 8) for finite-dimensional symmetric algebras over $\mathbb Q$ and records a negative answer with an external Lean 4 counterexample. The FC statement uses Mathlib's derived-category groups $\operatorname{Ext}^n_\Gamma(M,M)$ and calls $\Gamma$ symmetric when $\Gamma\cong\operatorname{Hom}_{\mathbb Q}(\Gamma,\mathbb Q)$ as bimodules. OpenAI's preprint A counterexample to Tachikawa's second conjecture (September 23, 2026) had disproved the conjecture over fields of characteristic two; this proof carries that construction over to characteristic zero, keeping its starting algebra and architecture and replacing the parity argument in its trace obstruction by a twisted-trace argument. Starting from a 10-dimensional algebra, the construction forms a trivial extension, a tensor square and a triangular algebra $\Lambda$, and takes $\Gamma=\Lambda\ltimes D\Lambda$ with $M=\Gamma\otimes_\Lambda Z$; the vanishing of all positive self-Ext groups is transferred along the trivial extension, and non-projectivity comes from a nonzero Ext group of a tensor-square module. The same pair is a counterexample over $\mathbb Q$ to the Auslander–Reiten conjecture of 1975: a symmetric algebra is self-injective, so $\operatorname{Ext}^n_\Gamma(M,\Gamma)=0$ as well, and the repository proves, as a separate Lean theorem, the negation of the Artin-algebra Auslander–Reiten statement in Formal Conjectures. OpenAI's explicit Auslander–Reiten counterexample, also dated September 23, 2026, lives in characteristic two. Two days before issue #7020, on 2026-10-07, Q. Tang (唐琪越) posted an AI-generated note on Hexagon (2610.00143) that extends OpenAI's constructions to every field containing an element of infinite multiplicative order, including $\mathbb Q$, and claims counterexamples there to both Tachikawa conjectures, the Auslander–Reiten conjecture and the Nakayama-type conjectures, without a formalization. The rational counterexample here was found independently, and the author became aware of that note only after this announcement; what this work adds is a counterexample over $\mathbb Q$ verified in Lean. By Enomoto and Marczinzik's global equivalences, the Nakayama, generalized Nakayama and Tachikawa conjectures also fail for some algebras over $\mathbb Q$. The Lean development has about 14,700 new lines together with about 19,200 lines of OpenAI's Lean code, credited in the repository, and the FC definition and target match the proof character for character. The repository reports that both final theorems depend only on propext, Classical.choice, and Quot.sound, and provides a single-file Lean4Web edition at Lean 4.35.0-rc4. The PR registers the Tachikawa statement and leaves the update of the Auslander–Reiten entry to a separate PR. As of 2026-10-10, PR #7067 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Erdős Problem 952 — moats for binary forms and a classification of walkable formsBounded-step walks through lattice points where a binary form has few prime factors

LeanProofStatementNumber TheorySOLVED
Problem. Erdős Problem 952 asks whether one can walk to infinity through Gaussian primes with steps of bounded length. For a homogeneous form $F\in\mathbb Z[X,Y]$, consider the lattice points $(x,y)$ with $F(x,y)\ne0$ and at most $r$ distinct prime factors of $|F(x,y)|$, joined when their distance is at most $D$. For which $F$ are the connected components uniformly bounded for every $D$ and $r$? — Formal Conjectures: Erdős Problem 952
  • 2026-10-10 formal conjectures variant proposals with proofs
    Proposed in Formal Conjectures issue #7091 three variants of Erdős Problem 952, the Gaussian-moat problem, all as solved and with Lean 4 proofs. OpenAI's preprint Bounded-Step Walks on Gaussian Primes (September 26, 2026) answered the problem negatively by proving, with a Lean formalization, that for every step bound $D$ the graph joining Gaussian primes at distance at most $D$ has uniformly bounded components. Variant 1 is that theorem, with an FC-style statement and a new Lean proof derived from variant 3 with $F=X^2+Y^2$. Variant 2 generalizes it to binary forms: for every irreducible homogeneous $F\in\mathbb Z[X,Y]$ of degree at least $2$ and all $D$ and $r$, the lattice points $(x,y)$ with $F(x,y)\ne0$ and at most $r$ distinct prime factors of $|F(x,y)|$, joined at distance at most $D$, form components of uniformly bounded size. Variant 3 classifies all nonzero homogeneous forms of positive degree: arbitrarily long bounded-step self-avoiding walks through such points exist for some $D$ and $r$ exactly when $F=c(aX+bY)^m$, in which case the walk $(a+bn,\,b-an)$ keeps $F$ constant. The proof adapts OpenAI's entropy-sieve argument: primes that split completely, supplied by Chebotarev's theorem, give $\deg F$ forbidden residue classes modulo each prime, and a Hamming-graph isoperimetric inequality replaces the Boolean cube; disjoint blocks of primes then handle the bound $r$ on distinct prime factors, and reducible forms are treated by divisibility and by periodic forbidden strips for two independent linear factors. The Lean statements forbid arbitrarily long injective walks, which on $\mathbb Z^2$ with bounded steps is equivalent to uniformly bounded components, and the FC-style statements match the proofs character for character. The Lean development does not reuse OpenAI's Lean code and vendors the TauCeti, PFR and LeanPool libraries; the repository reports strict builds of all three targets with #print axioms listing only propext, Classical.choice, and Quot.sound, an earlier full audit with independent kernel replays, and a single-file Lean4Web edition at Lean 4.35.0-rc4. For quadratic forms, variant 2 is equivalent to a theorem on quadratic number fields formalized in Lean by GitHub user Renriviera (quadratic-moats-lean), public on 2026-10-07 for irreducible elements and on 2026-10-09 for elements with at most $b$ distinct prime-ideal factors; no earlier statement of the cases of degree at least $3$ or of the classification in variant 3 was found. As of 2026-10-11, the issue is open with no comments.

Convexity of Fisher information along the heat flow — a counterexample in dimension 41Time-convexity of the Fisher information of the Gaussian heat flow in high dimension

LeanProofCounterexampleStatementProbabilityOPEN
Problem. For a probability density $f$ on $\mathbb R^d$ let $P_tf=f*\gamma_t$, with $\gamma_t$ the centred Gaussian of covariance $t\,\mathrm{Id}$, and $I_f(t)=\int|\nabla\log P_tf|^2\,P_tf$. With $H=-\nabla^2\log f$, Problem (iii) asks whether $\mathcal D[f]=\int(|\nabla H|^2+2\operatorname{tr}H^3)f$ can be negative; since $I_f''(t)=\mathcal D[P_tf]$, this asks whether $I_f$ is convex in $t$ for every admissible density. Is it convex in every dimension? — Formal Conjectures: proposed Fisher information statements (PR head)
  • 2026-10-11 formal conjectures statement and disproof submission
    Opened Formal Conjectures issue #7070 on 2026-10-10 with an 8-page manuscript, and on 2026-10-11 submitted PR #7113, which adds Problem (iii) of Zou, Fan, Gao, and Wang (A Hexagonal Counterexample to Log-Convexity of Fisher Information Along the Heat Flow, arXiv:2605.18081v3, Section 7.3) and marks it solved negatively with an external Lean 4 proof. The FC file defines the heat flow with the covariance-$t$ Gaussian kernel, the Fisher information $\int|\nabla\log\rho|^2\rho$, and the admissible class of the paper (positive Schwartz probability densities with $\mathcal I,\mathcal Q,\mathcal D<\infty$ and $\mathcal Q>0$), and asks whether $t\mapsto I_f(t)$ is convex on $(0,\infty)$ for every admissible $f$; a variant, proposed in the issue, asks the same for all smooth positive densities with finite Fisher information. The counterexample lives on $\mathbb R^{41}$: a radial bump $w(x)=\tilde u(|x|^2/2)$ built from $u(s)=\tfrac43\log(1+e^{15-2s-s^2})$ is periodized over the lattice $20\mathbb Z^{41}$, exponentiated, and multiplied by a very wide Gaussian. For such radial profiles $|\nabla H|^2+2\operatorname{tr}H^3$ reduces to a cubic expression in the derivatives of $u$, and kernel-checked interval arithmetic on 293 cells shows that its weighted radial integral is at most $-5\cdot10^{18}$, which gives $\mathcal D[f]<0$. Differentiating under the integral and integrating by parts, the proof establishes $I_f'=-\mathcal Q$ and $\mathcal Q'=-\mathcal D(P_tf)$, and continuity at $t=0$ gives $I_f''(t)<0$ for all small $t>0$, so $I_f$ is not convex; the variant follows. The definitions and both targets match the FC file character for character, with bridge lemmas to the project's own definitions. The project has about 11,000 lines of Lean and uses seven heat-kernel modules from the Tau Ceti library; the fixed repository reports a build at Lean 4.35.0-rc3 with Mathlib 1b5006ca and #print axioms listing only propext, Classical.choice, and Quot.sound for both targets, and a single-file edition of about 13,000 lines. Convexity was known for $d\le4$, and related conjectures had recently been disproved: the Gaussian completely monotone conjecture by Gu and Sellke (arXiv:2605.11656) and log-convexity of $I_f$ by Zou, Fan, Gao, and Wang. No earlier counterexample to convexity itself was found. As of 2026-10-11, PR #7113 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Hajłasz–Onninen Question 1 — the W¹,¹ endpoint for the centered ball maximal function in every dimensionGradient bounds for the centered Hardy–Littlewood maximal function in W¹,¹

LeanProofStatementHarmonic AnalysisOPEN
Problem. For $f\in W^{1,1}(\mathbb R^d)$ let $Mf(x)=\sup_{r>0}|B(x,r)|^{-1}\int_{B(x,r)}|f|$ be the centered ball maximal function. Hajłasz and Onninen asked whether $f\mapsto|\nabla Mf|$ is bounded from $W^{1,1}(\mathbb R^d)$ to $L^1(\mathbb R^d)$, that is, whether $\|\nabla Mf\|_1\le C_d(\|f\|_1+\|\nabla f\|_1)$; by scaling this is equivalent to $\|\nabla Mf\|_1\le C_d\|\nabla f\|_1$. Does this hold in every dimension? — Formal Conjectures: proposed Hajłasz–Onninen statements (PR head)
  • 2026-10-11 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #6938 on 2026-10-08 with a 13-page manuscript for $d\ge2$, and on 2026-10-11 submitted PR #7129, which adds Question 1 of Hajłasz and Onninen (On boundedness of maximal functions in Sobolev spaces, Ann. Acad. Sci. Fenn. Math. 29 (2004), p. 169) and marks it solved with external Lean 4 proofs: for every $d\ge1$ there is $C_d$ such that every signed $f\in W^{1,1}(\mathbb R^d)$ has an almost everywhere finite centered ball maximal function $Mf$ with an integrable weak gradient and $\|\nabla Mf\|_1\le C_d(\|f\|_1+\|\nabla f\|_1)$; a variant for $d\ge2$, proposed in the issue, removes the $\|f\|_1$ term. This is the $p=1$ endpoint of Kinnunen's theorem. The centered case was known for $d=1$ by Kurka (2015), and OpenAI's preprint An Endpoint Gradient Bound for the Centered Disk Maximal Operator (dated 2026-09-26, released 2026-10-06 with a Lean formalization) proved the homogeneous bound for $d=2$. The proof generalizes the architecture of OpenAI's planar argument to every dimension, replacing its complex-number factor by a real matrix correction: a signed finite-band estimate for suprema over dyadic ranges of radii, assembled from ball moment identities, shell lemmas, dyadic tent refinements, cone palettes and a Markov label law, is passed to $W^{1,1}$ by smooth approximation, a BV bound and removal of the singular part. The case $d=1$ is obtained by adapting the same argument, and no OpenAI Lean code is reused. The FC definitions and both targets match the proof's statements, which use definitionally equal wrappers. The project has about 66,000 lines of Lean, with classical maximal-function lemmas from the Tau Ceti library; the fixed repository reports a full build of the single-file edition (about 64,500 lines) at Lean 4.35.0-rc4 with Mathlib 260040a4 and #print axioms listing only propext, Classical.choice, and Quot.sound for both targets. No earlier proof for $d\ge3$ was found. As of 2026-10-11, PR #7129 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Smooth isometric embeddings in ℝ⁴ — every closed orientable Riemannian surfaceSmooth isometric embeddings of Riemannian surfaces in Euclidean four-space

LeanProofStatementDifferential GeometryOPEN
Problem. Let $M$ be a closed orientable smooth surface, compact and without boundary, and let $g$ be any smooth Riemannian metric on $M$, with no curvature assumption. Is there a smooth embedding $F\colon M\hookrightarrow\mathbb R^4$ that is isometric, that is, $\langle dF_p(v),dF_p(w)\rangle=g_p(v,w)$ for all $p\in M$ and $v,w\in T_pM$? — Formal Conjectures: proposed smooth isometric embedding statement (PR head)
  • 2026-10-12 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #7048 on 2026-10-09 with a 7-page manuscript, and on 2026-10-12 submitted PR #7211, which adds the question whether every closed orientable smooth surface with any smooth Riemannian metric admits a smooth isometric embedding into $\mathbb R^4$ and marks it solved affirmatively with an external Lean 4 proof. In the FC file the surface is a compact Hausdorff space with a smooth atlas modelled on $\mathbb R^2$, possibly empty or disconnected. Mathlib has no orientability for manifolds at the pinned revision, so the file defines it for surfaces as an orientation of every tangent plane that is locally constant when read through each chart. The statement asks for a smooth injective map into $\mathbb R^4$ preserving the metric, which on a compact surface is an embedding. Gromov's survey Geometric, algebraic and analytic descendants of Nash isometric embedding theorems records, as a conjecture proposed by Chern around 1950, the existence of smooth isometric immersions of surfaces into $\mathbb R^4$, and OpenAI's preprint Smooth isometric immersions of closed surfaces into Euclidean four-space (dated 2026-09-23), with a Lean formalization, proved it for every closed surface, orientable or not, allowing self-intersections. This proof extends OpenAI's construction to embeddings in the orientable case. It starts from an embedding instead of an immersion: by the classification of surfaces, a closed connected orientable surface embeds in $\mathbb R^3$ as the boundary of a handlebody, inverse stereographic projection places it in $S^3\subset\mathbb R^4$, and the spherical preparation and scaling keep it an embedding whose induced metric lies below $g$. The metric deficit is then added in finitely many primitive steps $a^2\,dx\otimes dx$ supported on coordinate disks, each realized exactly by OpenAI's oscillatory construction. The new argument shows that every exact correction can be taken injective: the positive phase Hessian makes the oscillation plane, the orthogonal complement of $F_y$ and $F_{yy}$, complementary to the tangent plane, an oblique tubular retraction along this complement cancels the leading derivative oscillation, and the composition is $C^1$-close to the previous embedding, so the corrected map is itself an embedding at small scales. Disconnected surfaces are handled by translating the component images apart. The orientability definition and the target of the PR file coincide character for character with the proved statement, apart from answer(True) written as True. The Lean release has 2,494 modules besides Mathlib: 114 written for this proof and 2,380 pinned third-party modules, credited in the repository, namely OpenAI's Lean formalization of the analytic construction, used unchanged (1,467 modules), parts of qinz1yang/differential-geometry (789), and the classification of surfaces from Lean Pool, which originates in mccorvie/classification-of-surfaces (124). The repository's public GitHub Actions run on the fixed commit records a full build at Lean 4.34.1 with Mathlib d13f23b7, acceptance by Lean Comparator of the proof against the challenge extracted from the FC file, and a fresh kernel replay of the whole environment, with #print axioms listing only propext, Classical.choice, and Quot.sound for the FC target. No earlier proof of the embedding statement was found. As of 2026-10-12, PR #7211 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Strong Holevo additivity — every qubit channel with an arbitrary partner channelAdditivity of the Holevo information of qubit channels

LeanProofStatementQuantum InformationOPEN
  • 🟢 Status when posted OPEN — Tang–Zhu–Bai–Wang, arXiv:2609.28592 proves Holevo additivity with every finite-dimensional partner for qubit channels admitting a pure output, which include all qubit channels of Kraus rank at most two. King (2002) had proved it for unital qubit channels, and OpenAI's preprint dated 2026-09-24 for generalized amplitude damping channels. No proof for arbitrary qubit channels was found. Formal Conjectures did not yet contain the problem.
  • 🔗 Source problem Formal Conjectures: proposed qubit Holevo additivity statements (PR head)
Problem. For a quantum channel $\Lambda$, the Holevo information $\chi(\Lambda)$ is the supremum of $S(\Lambda(\sum_ip_i\rho_i))-\sum_ip_iS(\Lambda(\rho_i))$ over finite ensembles of input states $\rho_i$ with probabilities $p_i$, where $S$ is the von Neumann entropy. Is $\chi(\Phi\otimes\Omega)=\chi(\Phi)+\chi(\Omega)$ for every qubit channel $\Phi\colon M_2(\mathbb C)\to M_2(\mathbb C)$ and every channel $\Omega\colon M_n(\mathbb C)\to M_m(\mathbb C)$, with no unitality, rank or symmetry condition on $\Phi$ and with entangled inputs allowed? — Formal Conjectures: proposed qubit Holevo additivity statements (PR head)
  • 2026-10-13 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #7125 on 2026-10-11 with a 12-page manuscript, and on 2026-10-13 submitted PR #7212, which adds strong Holevo additivity for qubit channels and marks it solved with an external Lean 4 proof: $\chi(\Phi\otimes\Omega)=\chi(\Phi)+\chi(\Omega)$ for every qubit channel $\Phi$ and every channel $\Omega$ between systems of positive finite dimension. A second target, $\chi(\Phi^{\otimes n})=n\,\chi(\Phi)$ for every qubit channel and every $n\ge0$, follows by taking the partner to be a power of $\Phi$. The FC file uses Mathlib alone: a channel is a finite Kraus family with $\sum_iK_i^\dagger K_i=1$, the von Neumann entropy is taken with the natural logarithm through the continuous functional calculus, and $\chi$ is a supremum over finite ensembles of density matrices, so entangled inputs to $\Phi\otimes\Omega$ are included. Additivity of the Holevo information fails for some channels of high dimension (Hastings, 2009). With a qubit factor and an arbitrary partner it had been proved when $\Phi$ is unital (King, 2002), when $\Phi$ has a pure output (Tang, Zhu, Bai, and Wang, arXiv:2609.28592), and for generalized amplitude damping channels (OpenAI's preprint Classical capacity and entropy inequalities for generalized amplitude damping, dated 2026-09-24, whose Lean formalization covers repeated uses; Fang, arXiv:2609.14621, determines the capacity of these channels). This proof imposes no unitality, pure-output, Kraus-rank or symmetry condition. It reduces the upper bound to pure joint inputs, shows that every qubit channel, after a rotation of its output found from a zero of a binary quartic, has a normal form in which every Kraus operator but one maps into a single output basis vector, and obtains the pure-input bound from a four-block matrix entropy inequality. That inequality is proved through a dual value function whose minimizer exists, is unique and depends differentiably on the data: the Hessian of the value is a Schur complement of Kubo–Mori operators, nonpositive by a comparison of a curve of positive matrices with weighted geometric means, and concavity with homogeneity reduces the claim to rank-one terms and a scalar case. The theorem is proved first for Physlib's bundled CPTP maps and then transferred to Kraus families. The definitions and both targets of the PR file coincide character for character with the proved statements. The Lean project has 80 modules (53 of them for the four-block inequality) and about 11,800 lines, including six modules of lemmas adapted from OpenAI's Lean library, and vendors 51 files of Physlib and 14 further files of OpenAI's library, all credited in the repository. The repository's public GitHub Actions run records a full build at Lean 4.34.1 with Mathlib d13f23b7, acceptance by Lean Comparator of the proof against the challenge extracted from the FC file, and a fresh kernel replay of the whole environment, with #print axioms listing only propext, Classical.choice, and Quot.sound for both targets, and a single-file edition of about 39,400 lines is provided for Lean4Web. The results are equalities of Holevo quantities; the coding theorem relating $\chi$ to the classical capacity is not formalized. No earlier proof for arbitrary qubit channels was found. As of 2026-10-13, PR #7212 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Melbourne–Tkocz Conjecture 6.1 — the sharp entropic Rogers–Shephard inequalityEntropy of the difference of two independent log-concave random vectors

LeanProofStatementProbabilityConvex GeometryOPEN
  • 🟢 Status when posted OPEN — Melbourne–Tkocz, arXiv:2005.10930v1, Conjecture 6.1 states the conjecture, attributing it to Madiman and Kontoyiannis, and proves its Rényi analogue for orders $\alpha\ge2$; for Shannon entropy it records the weaker bounds $n\log4$ of Madiman–Kontoyiannis and $n\log e$ of Bobkov–Madiman. In July 2026 Fu and Li (arXiv:2510.09206v3) still describe the extremality of exponential distributions for such differences as suspected, proved only for orders at least 2. No proof of the Shannon case was found. Formal Conjectures did not yet contain the problem.
  • 🔗 Source problem Formal Conjectures: proposed entropic Rogers–Shephard statement (PR head)
Problem. Let $X$ and $X'$ be independent random vectors with the same log-concave probability density on $\mathbb R^n$, $n\ge1$, and let $h$ be the differential Shannon entropy with natural logarithms. The conjecture, an entropic analogue of the Rogers–Shephard inequality for difference bodies, asserts $h(X-X')-h(X)\le n\log2$, with equality for products of one-sided exponential distributions. Does it hold in every dimension? — Formal Conjectures: proposed entropic Rogers–Shephard statement (PR head)
  • 2026-10-13 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #7196 on 2026-10-12 with an 8-page manuscript, and on 2026-10-13 submitted PR #7283, which adds Conjecture 6.1 of Melbourne and Tkocz (Reversals of Rényi Entropy Inequalities under Log-Concavity, arXiv:2005.10930v1), attributed there to Madiman and Kontoyiannis, and marks it solved with an external Lean 4 proof: for independent random vectors $X,X'$ with the same log-concave probability density on $\mathbb R^n$, $n\ge1$, one has $h(X-X')-h(X)\le n\log2$. The FC file defines probability densities and multiplicative log-concavity, with zeros and endpoint weights allowed, and takes $X,X'$ on an arbitrary probability space. Its theorem concludes, without assuming them, that $X-X'$ has the density $g(s)=\int f(x)f(x-s)\,dx$, that $f\log f$ and $g\log g$ are integrable, so both entropies are finite, and that the entropy difference is at most $n\log2$. The inequality is an entropic analogue of the Rogers–Shephard bound for difference bodies. Madiman and Kontoyiannis had proved it with $n\log4$ and Bobkov and Madiman with $n\log e$, and Melbourne and Tkocz proved the sharp Rényi analogue for orders $\alpha\ge2$. The proof first bounds the entropy increase for uniform distributions on $N$-dimensional convex bodies by $F(N)=N(H_{2N}-H_N)$, with $H_k$ the harmonic numbers, using Brunn–Minkowski for the overlap volumes, and computes the difference densities of uniform vectors on a simplex, for which the increase equals $F(m)$. A density proportional to the $m$-th power of a concave function on a convex body is then a marginal of the uniform distribution on a convex body in dimension $n+m$ whose fibers are simplices; a mixture of simplex-difference densities and concavity of the entropy give the bound $F(n+m)-F(m)$. Approximating a log-concave density by such concave powers on compact sets, with common exponential bounds to pass to the limit in the entropies, and letting $m\to\infty$ yields $n\log2$. The project also proves that products of one-sided exponential distributions and all their invertible affine images attain equality and that no smaller constant works in any dimension; these are among four supporting statements checked together with the FC target. The definitions and the target of the PR file coincide character for character with the proved statement. The Lean project has 58 modules and a file of intermediate results, about 7,600 lines; three modules contain material adapted from the Tau Ceti library, and three unmodified Lean Pool files supply the Brunn–Minkowski and Prékopa–Leindler inequalities, all credited in the repository. The repository's public GitHub Actions run on the fixed commit records a full build at Lean 4.33.1 with Mathlib 0df444a3, the revision pinned by Formal Conjectures, acceptance by Lean Comparator of the proof against the five challenge statements, a 451-target axiom audit listing only propext, Classical.choice, and Quot.sound, and a fresh kernel replay. OpenAI's preprint A sharp entropy bound and the simplex inequality for isotropic constants (dated 2026-10-05) proves a different sharp entropy inequality for log-concave densities with the same extremizers, by another method. No earlier proof of the Shannon case was found. As of 2026-10-13, PR #7283 is open and unmerged, its upstream checks pass, and no independent human review is recorded.

Kalai's 3^d conjecture — the dimension-five face-count boundNonempty faces of centrally symmetric convex polytopes

LeanProofStatementConvex GeometryCombinatoricsOPEN
Problem. Kalai's Conjecture A asks whether every centrally symmetric convex polytope of dimension $d$ has at least $3^d$ nonempty faces, including the polytope itself and excluding the empty face. In dimension five this is the bound $s(P)=1+\sum_{i=0}^{4}f_i(P)\ge3^5=243$, where $f_i(P)$ counts the $i$-dimensional faces. — Formal Conjectures: proposed dimension-five statement (PR head)
  • 2026-10-15 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #7141 on 2026-10-11 with a manuscript presenting a proposed proof of Kalai's Conjecture A in dimension five, and submitted PR #7341 on 2026-10-15. The PR adds four statements in namespace Kalai: the general conjecture, the established dimension-at-most-four case of Sanyal, Werner and Ziegler, the dimension-five theorem with an external Lean proof, and the remaining dimension-at-least-six conjecture. Only the dimension-five statement has this project's external proof. The theorem Kalai.three_pow_five proves $s(P)\ge243$ for every full-dimensional centrally symmetric convex polytope in $\mathbb R^5$. The five-cube attains the bound. Its hypotheses use finite convex hulls, nonempty interior and symmetry about the origin, and Set.ncard counts nonempty exposed faces, which coincide with nonempty faces for polytopes. There are no additional rigidity, certificate, vertex-count or facet-count assumptions. The proof uses Euler's relation and antipodal symmetry to reduce the count to $s(P)=3+2(f_1+f_3)$ with $f_1+f_3$ even. Rigidity inequalities, incidence bounds and polarity reduce a putative counterexample to fifteen possible vertex/facet-count pairs; local facet geometry, cube sections, polar fibres and exact finite arithmetic exclude these cases. The repository supplies a 240-page manuscript, 976 readable proof modules and an optimized Lean archive. Public CI run 38654268610 verified proof commit 38bc3bb using Lean 4.33.1 and Mathlib 0df444a3, matching the FC pins checked on 2026-10-15. The run replayed all 42,660 archived declarations with the ordinary kernel, accepted the solution against the independently compiled Comparator challenge, passed a separate leanchecker --fresh FCSolution replay and passed the exact arithmetic supplement. The archived axiom output lists only propext, Classical.choice and Quot.sound for both the public theorem and its underlying result. The proof and the PR use the same dimension-five target. As of 2026-10-15, PR #7341 is open and unmerged with no submitted human review; external proof verification does not itself establish upstream acceptance. The general conjecture and the unrestricted cases in dimensions at least six remain open.

Infinite finitely presented residually finite p-groups for every primeAll-prime strengthening of the finitely presented Burnside problem

LeanProofStatementGroup TheoryOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures issue #7284 proposes the all-prime existence problem and links Kitamura's solution note. OpenAI's September 2026 periodic-group result and October 2026 residually finite 2-group result precede this submission; the latter already implies the residually finite periodic variant. The new contribution extends the residually finite p-group existence result to every prime.
  • 🔗 Source problem Formal Conjectures: proposed all-prime statement (fixed PR head)
Problem. For every prime $p$, does there exist an infinite finitely presented residually finite abstract group $G_p$ such that $\forall g\in G_p\;\exists a\in\mathbb N_0,\ g^{p^a}=1$? Finite presentation means finitely many generators and relators; residual finiteness means every nonidentity element survives in a finite quotient. The group may depend on $p$, and $a$ may depend on the element. — Formal Conjectures: proposed all-prime statement (fixed PR head)
  • 2026-10-16 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #7284 on 2026-10-13 and submitted PR #7412 on 2026-10-16, adding the finitely presented Burnside problem and two residually finite variants. The main statement reuses OpenAI's theorem that an infinite finitely presented periodic group exists. Kitamura's all-prime theorem proves that for every prime $p$ there is an infinite finitely presented residually finite abstract $p$-group; specializing to $p=2$ also gives the intermediate residually finite periodic variant. OpenAI's separate preprint of 2026-10-05 had already supplied the residually finite 2-group case and hence the intermediate existence statement. The new result covers every prime; the main statement is credited to OpenAI and the intermediate variant is recorded as a corollary. The FC targets use Mathlib's standard finite-presentation, residual-finiteness and p-group predicates. The group may depend on the prime, and the p-power killing an element may depend on that element; a common finite exponent is not asserted. The construction develops a characteristic-p coefficient algebra with a matrix-nil augmentation ideal, proves separation through finite word-length quotients and finite-rank injectivity for Steinberg groups, and takes the kernel of reduction to the finite Steinberg group over $\mathbb F_p$. For every rank at least seven, the same kernel is proved infinite, finitely presented, residually finite and a p-group. The proof reuses 63 unchanged OpenAI word-machine and periodic-group modules, with adapted material from Tau Ceti and Lean Pool credited in the repository. Public CI run 38865696145 verified commit c1ce611 on Lean 4.34.1: all 204 project modules built, Comparator accepted all three FC targets, 494 core and five connection axiom audits passed, and a separate leanchecker --fresh FCSolution replay succeeded. The archived target axiom output lists only propext, Classical.choice and Quot.sound. The PR reports a separate successful local compilation of the FC statement file on FC's Lean 4.33.1; this is distinct from the external proof verification on Lean 4.34.1. As of 2026-10-16, PR #7412 is open and unmerged with no submitted human review.

Simultaneous primitive roots — unconditional lower bounds for integer basesPrimes for which every member of a finite set is a primitive root

LeanProofStatementAnalytic Number TheoryOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures issue #7011 proposes the positive lower-density statement and links Kitamura's solution note. OpenAI's 2026-10-04 simultaneous theorem assumes four analytic inputs and gives a lower bound of order $x/(\log x)^2$ for prime bases. The submitted variants remove those external hypotheses, and the stronger one allows integer bases satisfying the square condition with a bound of order $x/\log x$.
  • 🔗 Source problem Formal Conjectures: proposed positive lower-density statement (fixed PR head)
Problem. Let $A\subset\mathbb Z_{\ge2}$ be finite and nonempty, and suppose every odd-cardinality subset of $A$ has nonsquare product. Let $N_A(x)$ count the primes $x<p\le2x$ for which every $a\in A$ is a primitive root modulo $p$. Are there constants $c_A>0$ and $x_A\ge2$ such that $N_A(x)\ge c_Ax/\log x$ for every real $x\ge x_A$, without additional analytic hypotheses? — Formal Conjectures: proposed positive lower-density statement (fixed PR head)
  • 2026-10-17 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #7011 on 2026-10-09 and submitted PR #7425 on 2026-10-17, adding two statements with external Lean proofs. The stronger theorem, positive_lower_density, applies to every nonempty finite set of integers at least two for which every odd-cardinality subset product is nonsquare. It proves that, for all sufficiently large real $x$, at least $c_Ax/\log x$ primes in $(x,2x]$ make every member of the set a primitive root, for some $c_A>0$. The prime-base theorem, simultaneous_primitive_roots_unconditional, follows from this result and gives $c_Ax/(\log x)^2$ primes in the open interval $(x,2x)$, without the four analytic hypotheses of OpenAI's conditional simultaneous theorem of 2026-10-04. Distinct prime bases automatically satisfy the square condition. The submitted types use Mathlib's IsSquare, orderOf and ZMod directly, with no problem-specific definitions. The argument arranges quadratic nonresidue classes using the square condition and quadratic reciprocity, counts candidate primes with a large prime factor of $p-1$, bounds order obstructions through Kummer-field estimates and a product of $a^j-1$, and fixes a cutoff that leaves a positive multiple of $x/\log x$. The project credits OpenAI's libraries and the existing ArtinPrimitiveRoots formalization. Public CI run 38910471634 verified commit 0987a8b on Lean 4.33.1 with Mathlib 0df444a3; all nine jobs succeeded. The independent Mathlib-only challenge matches both submitted types, Comparator accepted the two proofs and their dependent definitions, the axiom audit reports only propext, Classical.choice and Quot.sound, and fresh Lean kernel replay passed. Acceptance and rejection controls accepted the valid proof and rejected a wrong type and an extra axiom. The separate four-input conditional reference and the square-condition obstruction supplement are outside the two-target Comparator scope. This result supplies a positive lower bound, not Artin's predicted exact limiting density; rational and negative bases are outside its scope. Priority for a first unconditional proof is not established. As of 2026-10-17, PR #7425 is open and unmerged with no submitted human review.

Infinity Laplacian — interior C¹,α estimates with bounded forcingInterior gradient regularity for viscosity solutions of the infinity Laplace equation

LeanProofStatementPartial Differential EquationsOPEN
Problem. For each dimension $n\ge2$, are there constants $0<\alpha_n\le1/3$ and $C_n\ge0$, depending only on $n$, such that every bounded continuous function with $|\Delta_\infty u|\le M$ in the viscosity sense on $B_R(c)\subset\mathbb R^n$, $R>0$, $M\ge0$, belongs to $C^{1,\alpha_n}(B_{R/2}(c))$ and satisfies $\|Du\|_{L^\infty(B_{R/2}(c))}+R^{\alpha_n}[Du]_{C^{0,\alpha_n}(B_{R/2}(c))}\le C_n(\operatorname{osc}_{B_R(c)}u/R+(RM)^{1/3})$? Here $\Delta_\infty u=D^2u[Du,Du]$ is the unnormalized operator, and the endpoint $\alpha_n=1/3$ is not asserted. — Formal Conjectures: proposed infinity-Laplacian statements (fixed PR head)
  • 2026-10-18 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #7198 on 2026-10-12 with a proposed bounded-forcing regularity theorem, and submitted PR #7446 on 2026-10-18 with three formal statements and immutable external Lean proof links. The main statement records OpenAI's 2026-10-04 uniform interior regularity theorem for bounded infinity-harmonic functions on the unit ball in dimensions at least three. A second statement records its translation and scaling corollary on arbitrary balls. Kitamura's bounded-forcing variant proves interior $C^{1,\alpha_n}$ regularity in every dimension $n\ge2$ for bounded continuous functions satisfying $|\Delta_\infty u|\le M$ in the viscosity sense, with the estimate $\|Du\|_\infty+R^{\alpha_n}[Du]_{\alpha_n}\le C_n(\operatorname{osc}_{B_R(c)}u/R+(RM)^{1/3})$ on the half-radius ball. The exponent $0<\alpha_n\le1/3$ and constant $C_n$ precede all solution, center, radius and forcing data, so depend only on dimension. The formalization specifies Euclidean distances and norms and the unnormalized infinity Laplacian. Upper and lower C² viscosity tests include zero-gradient tests, and no boundary values are imposed. The proof establishes a homogeneous tangent estimate through cone comparison, blow-ups, compactness and Liouville rigidity, then lifts the function by an extra variable and constructs homogeneous replacements. Vertical quadratic perturbations bound the replacement error; affine approximations on nested balls produce the derivative and its Hölder bound, and rescaling gives the cubic-root forcing term. The two homogeneous FC conclusions are obtained in Lean by specializing the bounded-forcing theorem at zero forcing. Public CI run 39112978961 verified commit 90d031a on Lean 4.35.0-rc3 with Mathlib f6090c70; all four jobs succeeded. Comparator accepted all three targets, the source audit covered 1,292 declarations, the FC audit covered eight declarations, and a separate fresh kernel replay passed. The archived axiom output lists only propext, Classical.choice and Quot.sound for all three theorems. The FC statement file was compiled separately on FC's Lean 4.33.1, and its mathematical statements were checked against the verified proofs on the proof toolchain. The result does not assert the endpoint exponent $1/3$; the normalized-operator appendix is outside the certified targets. As of 2026-10-18, PR #7446 is open and unmerged with no submitted human review.

Next-order sphere energy and spherical cap discrepancy in dimensions 2, 8 and 24Lattice coefficients in optimal Riesz energy, logarithmic energy and cap discrepancy

LeanProofStatementPotential TheoryDiscrete GeometryOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures issue #7313 proposes the next-order Riesz and logarithmic energy statements on $\mathbb S^8$ and links Kitamura's manuscript, which also treats dimensions two and 24. The PR formalizes the broader conjectures of Brauchart–Hardin–Saff (2012) and the cap discrepancy predictions of Brauchart (2011) and Bilyk–Brauchart (2025), while retaining the general statements and dimension-four variants as open.
  • 🔗 Source problem Formal Conjectures: proposed sphere energy statements (fixed PR head)
Problem. For optimal configurations of $N$ distinct points on the unit sphere $\mathbb S^d$, do the next-order Riesz and logarithmic energy terms and the leading spherical cap $L^2$ discrepancy have the predicted lattice coefficients? For $d=2,8,24$ and $-2<s<d$, $s\ne0$, the Riesz prediction is $\mathcal E_s(d,N)=V_{s,d}N^2+A_d^{-s/d}Z_{L_d}(s)N^{1+s/d}+o(N^{1+s/d})$, where $A_d$ is the sphere area and $L_d$ is the covolume-one triangular, $E_8$ or Leech lattice. The corresponding logarithmic coefficient is $Z'_{L_d}(0)+d^{-1}\log A_d$, and the optimal cap discrepancy is asymptotic to $\sqrt{-\kappa_d A_d^{1/d}Z_{L_d}(-1)}N^{-1/2-1/(2d)}$, with $\kappa_d=\Gamma((d+1)/2)/(d\sqrt\pi\,\Gamma(d/2))$. — Formal Conjectures: proposed sphere energy statements (fixed PR head)
  • 2026-10-19 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #7313 on 2026-10-14 for energy on $\mathbb S^8$, and submitted PR #7455 on 2026-10-19 with eighteen statements across Riesz energy, logarithmic energy and spherical cap discrepancy. Each family separates general coefficient existence from the predicted lattice formula, followed by dimension-specific variants. Nine variants on $\mathbb S^2$, $\mathbb S^8$ and $\mathbb S^{24}$ carry external Lean proofs; nine general or dimension-four statements remain open. The Riesz proofs establish the triangular, E8 and Leech Epstein coefficients for $-2<s<d$, $s\ne0$, including the removable E8 value at $s=2$. This covers part of the parent conjectures' larger range $-2<s<d+2$, $s\ne0,d$. The logarithmic proofs determine the linear coefficient after subtracting $W_dN^2-(N/d)\log N$, and the discrepancy proofs determine the leading constant at order $N^{-1/2-1/(2d)}$. All distances are chordal and energy sums are over ordered distinct pairs. Riesz energy is minimized for positive exponents and maximized for negative exponents; the logarithmic kernel is $-\log r$. Cap discrepancy uses uniform directions and Lebesgue height measure on $[-1,1]$. The proof obtains sharp Gaussian bounds from lattice minorants, localizes to sphere charts, subtracts the continuous background and takes Mellin integrals to obtain the Epstein coefficients. Matching lattice-patch configurations give limits along all integer sizes; a Frullani integral yields the logarithmic term, and Stolarsky's identity transfers the $s=-1$ energy result to discrepancy. Public CI run 39173822333 verified commit 56b3016 on Lean 4.34.1 with Mathlib d13f23b7. Six dependency-ordered build jobs completed from a cold project build, and all three dimension-specific Comparator jobs passed. The eighteen-statement Challenge is the pre-submission draft; the PR adds its standard license header with the mathematical body unchanged. Comparator selected the nine solved variants, compared their types and dependent definitions, audited the permitted axioms, and replayed the selected proofs and dependencies in Lean's builtin kernel. The actual output lists only propext, Classical.choice and Quot.sound for all nine targets. The optional separate replay of every imported declaration was skipped. The Challenge build used a recorded import-only FC utility adapter; compilation and linting in upstream FC's Lean 4.33.1 environment remain unverified. As of 2026-10-19, PR #7455 is open and unmerged with no submitted human review.

Reversible Finsler spheres — infinitely many prime closed geodesics in dimensions n ≥ 3Geometrically distinct closed geodesics for arbitrary reversible Finsler metrics

LeanProofStatementDifferential GeometryDynamical SystemsOPEN
  • 🟢 Status when posted OPEN — Formal Conjectures issue #7041 proposes the reversible Finsler sphere assertion for dimensions at least three and links Kitamura's solution note. OpenAI's 2026-09-24 preprint supplies the Riemannian sphere result; De Philippis, Marini, Mazzucchelli and Suhr (2022) supply the reversible Finsler two-sphere case. The new external Lean proof covers the reversible Finsler assertion for n ≥ 3.
  • 🔗 Source problem Formal Conjectures: proposed sphere geodesic statements (fixed PR head)
Problem. For every integer $n\ge3$, does every smooth, strongly convex, reversible Finsler metric on the standard sphere $S^n$ have infinitely many geometrically distinct prime closed geodesics? Smoothness is required off the zero section, and geometric distinctness means distinct images in the sphere: iterates, changes of starting point and reversals contribute no new image. No curvature, bumpiness or nondegeneracy hypothesis is imposed. — Formal Conjectures: proposed sphere geodesic statements (fixed PR head)
  • 2026-10-22 formal conjectures statement and proof submission
    Opened Formal Conjectures issue #7041 on 2026-10-09 and submitted PR #7526 on 2026-10-22 with four statements: OpenAI's Riemannian sphere assertion for $n\ge2$, the all-dimensional reversible Finsler variant, the literature theorem on the two-sphere, and the higher-dimensional reversible Finsler theorem. Kitamura's external Lean proof establishes the complete $n\ge3$ target, ReversibleFinslerSpheres.infinitely_many_prime_closed_geodesics. Together with De Philippis–Marini–Mazzucchelli–Suhr (2022), this gives the mathematical result for all $n\ge2$; the contribution supplies neither a Lean proof of the two-sphere case nor an all-dimensional Lean proof. The definitions use the standard smooth sphere and its actual tangent spaces. Finsler smoothness holds off the zero section, and strong convexity is positivity of the fiber Hessian of $F^2/2$. The geodesics are smooth regular periodic curves stationary for the actual energy under every jointly smooth periodic variation. Their least positive period is normalized to one; infinitude counts distinct images, so iteration, phase changes and reversal do not add images. There are no curvature or nondegeneracy restrictions. The higher-dimensional argument assumes finitely many prime images, obtains a common spectral rate with bounded error from loop-space products and local index estimates, and derives uniform filling and resonance bounds. Reversal equivariant cohomology and half-period doubling bound a defect count; at a maximal-defect interval, a new class must acquire an evaluation-cup partner that vanishes at its first critical crossing, a contradiction. The project credits OpenAI's Riemannian sphere result and reference framework. Public CI run 39703676590 verified commit be3aaf6 using Lean 4.35.0-rc3 with Mathlib f6090c70. It built the project from source without restoring a project cache, using pinned dependency caches. The unmodified pinned Comparator CLI and real Linux landrun passed target and dependent-definition comparison, built-in Lean kernel replay and the post-comparison axiom audit. The exact target uses only propext, Classical.choice and Quot.sound. Its mathematical definitions and theorem are extracted from the standalone FC file. The earlier four-statement draft elaborated in FC's Lean 4.33.1 environment with a header warning; a warning-free full module build of the revised submission remains unverified. The earlier PDF is historical supporting material and has not been audited line by line against the current Lean snapshot. As of 2026-10-22, PR #7526 is open and unmerged with no submitted human review.

Green #14 — a deterministic explicit colouring for W(3,r) ≥ 2r²An explicit asymmetric van der Waerden colouring avoiding red triples and long blue progressions

LeanProofAdditive CombinatoricsNumber TheoryOPEN
Problem. Let $W(3,r)$ be the least $N$ such that every red–blue colouring of $\{1,\ldots,N\}$ contains a red three-term arithmetic progression or a blue $r$-term arithmetic progression. Give an explicit integer $r\ge3$ and a colouring of $\{1,\ldots,2r^2-1\}$ avoiding both types of monochromatic progression, thereby witnessing $W(3,r)\ge2r^2$. — Formal Conjectures: Green 14 explicit-colouring problem (fixed pre-submission commit)

Erdős #104 related question — exact unit-circumradius triple counts and all maximizersTriples on unit circles in planar point sets not contained in one unit circle

LeanProofStatementDiscrete GeometryCombinatoricsOPEN
Problem. Let $g(n)$ be the maximum number of unordered triples with circumradius one among $n\ge2$ distinct planar points, subject to no unit circle containing all the points. Determine $g(n)$. The exact answer is $g(2)=g(3)=0$ and, for $n\ge4$, $g(n)=\binom{n-1}{3}+n-2+\mathbf1_{3\mid(n-1)}$. Moreover, every admissible maximizing configuration with $n\ge4$ has exactly $n-1$ points on some unit circle. — Formal Conjectures: proposed unit-circumradius triple variant (fixed PR head)
  • 2026-11-19 formal conjectures statement and proof submission
    Submitted Formal Conjectures PR #8081 on 2026-11-19 (JST), adding Erdős's 1975 question about unit-circumradius triples as a solved value variant of Problem 104, with the exact formula in its answer definition and a citation to Kitamura's Lean proof. For every $n\ge2$, g(n) counts the maximum number of unordered three-point subsets lying on a unit circle, over configurations not all contained in one unit circle. The proof gives $g(2)=g(3)=0$ and $g(n)=\binom{n-1}{3}+n-2+\mathbf1_{3\mid(n-1)}$ for $n\ge4$. The external project also proves that every admissible maximizer with at least four points has exactly n−1 points on some unit circle, strengthening Erdős's suggested form for an attaining configuration. The main #104 question counts distinct circles: four points on one unit circle contribute one circle there but four triples here. This record concerns the separate triple-counting question. For n−1 points on a unit circle and one point off it, qualifying pairs of circle points form a graph whose components are isolated vertices, edges and triangles, giving the sharp additional-triple bound. Orbits of a 120-degree rotation on the circle, together with its center, attain the formula. Point-deletion counting yields the general upper bound by induction; separate small-case and equality arguments show that every maximizer has the asserted circle. The formalization counts three-element subsets using bundled Euclidean unit spheres. Checked bridges identify this count and its set of attainable values with the proof's distance-based definitions. Public CI run 44324826773 verified commit dd3f6c1 using Lean 4.35.0 and Mathlib 56a00585. Comparator checked both the exact-value and maximizer targets and their reachable definitions in separate Lean environments under Lake's bubblewrap sandbox. Both nanoda and Lean's kernel accepted the proof export, and a fresh Lean replay of the solution module passed. The axiom audits for both theorems list only propext, Classical.choice and Quot.sound. The public CI artifact's SHA-256 matches GitHub's recorded digest, and its input hashes identify the checked statement and proof sources. As of 2026-11-19, PR #8081 is open and unmerged with no submitted human review.

OEIS Details

A059304Double binomial-sum identity

ProofAPPROVED
Theorem. For every $n\ge 0$, $a(n)=\sum_{0\le j,k\le n}\binom{n}{j}\binom{n}{k}\binom{j+k}{n}$, where $a(n)=2^n\binom{2n}{n}$. — A059304
  • 2026-08-04 OEIS proof submission
    Submitted a proof of Peter Bala's conjectured identity for OEIS A059304. Using coefficient extraction, $\binom{j+k}{n}=[x^n](1+x)^{j+k}$, so the double sum is $[x^n](\sum_{j=0}^n\binom{n}{j}(1+x)^j)^2=[x^n](2+x)^{2n}=2^n\binom{2n}{n}=a(n)$. The published entry now labels the identity a theorem and includes this proof under Kenta Kitamura's name. OEIS editor Michel Marcus revised and reproposed the submission on 2026-08-04; Sean A. Irvine approved it on 2026-08-08.

Lean-Eval Details

pi_succ_sphere_n_mulEquiv_zmod_two$\pi_{n+1}(S^n) \cong \mathbb{Z}/2$ for $n \ge 3$

Lean-EvalFIRST / PASS
Benchmark. For every $n \ge 3$ and basepoint $x \in S^n$, the multiplicative group $\pi_{n+1}(S^n,x)$ is equivalent to $\mathbb{Z}/2$.
  • 2026-08-06 Lean-Eval submission
    Produced the first complete Lean formalization accepted by Lean-Eval for this benchmark, proving it uniformly for all $n \ge 3$. Lean-Eval's official result records it under Newly-solved problems, establishing it as the first accepted solution for this benchmark within Lean-Eval. The proof follows Pontryagin's framed-cobordism route, without using a Hopf map or degree theory. The evaluated artifact comprises 448 submission modules and builds 9101 jobs; it uses only propext, Classical.choice, and Quot.sound, with no sorry, admit, native_decide, or bv_decide. The evaluation attempted 1 benchmark and succeeded on 1.

conway_schneeberger_fifteenConway–Schneeberger fifteen theorem

Lean-EvalFIRST / PASS
Benchmark. A positive-definite quadratic form defined by an integral symmetric matrix is universal if and only if it represents every integer from $1$ through $15$.
  • 2026-08-03 Lean-Eval submission
    Produced the first complete Lean formalization accepted by Lean-Eval for this benchmark. Lean-Eval's official result records conway_schneeberger_fifteen under Newly-solved problems, establishing it as the first accepted solution for the benchmark within Lean-Eval. The development formalizes the Conway–Schneeberger fifteen theorem via Bhargava's escalator-tree argument, including the supporting quadratic-form and number-theory machinery not previously available in Mathlib. The accepted artifact is a complete, sorry-free Lean 4 proof; it uses only propext, Classical.choice, and Quot.sound, with no native_decide or declared axioms. The evaluation attempted 1 benchmark and succeeded on 1.

Research Note Details

Barbados 2026 Problem 33 — Odd-hole-free triangle Ramsey graphs

ProofGraph Theory
  • 📄 Author proof note (2026-09-24) full argument
  • 🔗 Problem statement Barbados 2026 Problem 33
  • ℹ️ Scope Not counted as a new first solution; priority unverified; no Lean formalization or independent review.
Result. Among K₆-free, odd-hole-free graphs G with G → (3,3), the minimum order is 9 and the only 9-vertex example up to isomorphism is K₂ + complement(C₇). The K₄-free variant has no example.

Wrote a self-contained argument for the explicit 29-edge graph K₂ + complement(C₇), its minimum order and uniqueness among 9-vertex graphs satisfying all three conditions. The K₄-free case follows from the published 4-colour theorem for odd-hole-free graphs. Earlier Ramsey constructions already imply a positive answer to the existence question, and priority of the more specific statements remains unverified. This is not counted as a new first solution to an open problem; there is no Lean formalization or independent review.

Service Details

PR #4693 — Erdős #176 formalization

Invited reviewer

At the invitation of Formal Conjectures collaborator mo271, reviewed arex1337's PR #4693, “feat(Erdős 176): formalize discrepancy progression numbers and exact values”. The review was explicitly scoped to the definitions and formalization of parts (i)–(iii), not the [Ly26] certificates or the claimed exact values. It confirmed the arithmetic-progression encoding looked faithful to the source and raised four non-blocking issues: the undocumented $2 \le k$ hypothesis, the missing explanation that threshold monotonicity covers $N \le 156$, the rationale for choosing $k=13,15$, and the inaccessible 5.17-GiB certificate archive. The author credited the careful review and revised the documentation, references, and archive packaging in response.

Erdős #113 — remaining-implication report

Status reporter

Posted a deliberately scoped update on Erdős #113: the problem had already been marked disproved via Janzer's counterexample to the reverse implication, while OpenAI's Chapter 10 result refutes the remaining forward implication. The entry does not present this as the first disproof of the whole problem.

Erdős #146 — result-status curation

Status & formalization curator

Posted a source-backed update that OpenAI had resolved Erdős #146 by counterexample. The report identifies TwoDegenerateGraphs.not_erdos_146, links the official announcement and paper, and pins the Lean certificate to immutable commit a13547c. Also updated Formal Conjectures issue #382 with the theorem and recommended treating it as a solved-result formalization target. Williamjblair later added #146 as research solved in PR #4715.

Erdős #180 — result-status curation

Status & formalization curator

Posted a source-backed update that OpenAI had announced Erdős #180 as resolved. The report identifies CompactnessConjecture.not_erdos_180, which proves the direct negation of the formal conjecture statement, and links the immutable Lean certificate. Also updated Formal Conjectures issue #407 and recommended treating it as a solved-result formalization target. Williamjblair later added #180 as research solved in PR #4715.

Erdős #183 — Formal Conjectures status update

Formalization curator

Reported in Formal Conjectures issue #409 that OpenAI had resolved Erdős #183. The comment identifies ErdosProblems.MulticolourTriangleRamsey.erdos_183, states that it proves R(3;k)^(1/k) → ∞, and recommends a solved-result formalization target. Williamjblair later added the result as research solved in merged PR #4715.

Erdős #575 — result-status curation

Status & formalization curator

Posted a source-backed update that Chapter 10 of OpenAI's paper disproves the corrected Erdős–Simonovits compactness conjecture and explicitly cites Erdős #575. The report also links the accompanying Lean formalization at immutable commit a13547c. Also updated Formal Conjectures issue #804 with the disproof and immutable Lean source.

Ehrhart's volume conjecture — Formal Conjectures proposal

Conjecture curator

Opened Formal Conjectures issue #4680 to propose the Ehrhart volume result from OpenAI's ten-proofs release for inclusion as @[category research solved]. The proposal links the official article and paper, the immutable full Lean proof, and its compact Comparator Challenges statement template.

Quantum parallel repetition — Formal Conjectures proposal

Conjecture curator

Opened Formal Conjectures issue #4681 to propose the finite two-player quantum parallel repetition result from OpenAI's ten-proofs release for inclusion as @[category research solved]. The proposal links the official article and paper, the immutable full Lean proof, and its compact Comparator Challenges statement template.

Erdős #692 — status and formalization-scope review

Status & scope reviewer

Posted a detailed status and statement-scope review on Formal Conjectures issue #4839. The Erdős Problems page contains two questions: Cambie's counterexamples settle the unimodality question in Part I, but neither the existence of many local maxima nor the recorded n=3 Lean counterexample locates a global maximizer for general n, as requested by Part II. The review therefore explains why the page-level DISPROVED (LEAN) status reflects Part I and why Erdős Problem 692 as a whole should remain open. It also identifies a scope mismatch in Formal Conjectures: its Part II declaration asks only for an endpoint at which the maximum is attained, not where that endpoint is or how to characterize it. PR #4888 supplies a kernel-checked proof of this literal existence target, so the review proposes recording it as a separate solved variant while retaining the source Part II as research open.

Formal Conjectures — possible-misformalization audit tracker

Statement-audit coordinator

Opened Formal Conjectures issue #4896 to turn a broad statement audit into a checkable triage queue. Its 18 checklist lines cover source/statement mismatches, vacuous or degenerate boundary cases, answer-slot semantics, and blockers in open pull requests; it also links the existing dedicated trackers for Erdős 80, 319, and 769. The issue explicitly labels the entries as leads rather than asserting that every item is a confirmed bug, and records that the list was organized with AI assistance from longer working notes. Formal Conjectures maintainer williamjblair then checked Erdős 694, 195, and 887 and reported that all three findings were correct. He also confirmed the Erdős 940, Green 21, and Erdős 939 boundary-case leads, while narrowing exactly what the Erdős 939 witness establishes. The follow-up mathematical-review-guide PR #4899 incorporates examples from the audit and explicitly credits its distinction between unverified leads and confirmed defects. On 2026-08-14, the tracker was expanded with two open-PR blockers: PR #4258 proposes an all-dimension exponent for Erdős 1208 whose d=2 case conflicts with the fixed power saving in arXiv:2607.05374, and PR #3588 encodes only three-term rainbow progressions for Erdős 190 although the cited problem and solution concern rainbow k-term progressions.

Formal Conjectures — possible misformalizations II

Statement-audit coordinator

Opened Formal Conjectures issue #4923, with the repository's misformalization label, as a follow-up to issue #4896. Its independently checkable entries identify four exact self-answer patterns in Erdős 33, 329, 348, and 409(iii); three reflexive Big-O or Big-Theta answers in Erdős 422, 539, and 789; the zero-witness defect in OEIS A211417; and boundary or indexing problems in OEIS A111291 and A109074. On 2026-08-16, the A211417 entry was strengthened from a defect report to a proposed mathematical repair: the intended positive statement appears to hold with D(r)=lcm(1,…,r)^|{1≤i≤r : gcd(i,30)=1}|. The issue records the general proof as paper-complete but not yet Lean-checked, and distinguishes it from the four fixed divisibility statements submitted in PR #5010. The checklist remains a review queue rather than a claim that every underlying problem has been solved.

PR #4946 — Auto-OEIS statement and data audit

Volunteer batch auditor

Performed a voluntary quality-control pass on Formal Conjectures PR #4946, which imports 45 conjectures from the auto_oeis branch. The audit did not claim exhaustive review and presented all findings as leads for independent checking. It supplied a concrete modular counterexample candidate to A357565 at (m,p,r)=(2,3,2); identified an apparent duplicated value in A383327, with direct enumeration suggesting the final 67 should be 69; found that A379240's helper named A376418 appears not to encode OEIS A376418 and maps 56 and 60 to the same triple although the target function distinguishes them; linked a public Lean proof of unboundedness apparently conflicting with A374265's affirmative boundedness statement; and observed that A363414's existential is already witnessed by n=0, suggesting a missing positivity condition. Formal Conjectures collaborator mo271 thanked KitaKen1 for the useful review, stated that all five had been fixed, specifically noted that the audit also caught typos upstream at OEIS, and then invited a review of the follow-up PR #5016.

PR #5016 — second Auto-OEIS statement-fidelity audit

Invited batch auditor

After Formal Conjectures collaborator mo271 explicitly invited a follow-up review in PR #4946, performed a statement-fidelity audit of all 97 OEIS files proposed in PR #5016. The report lists 30 concrete questions with a corresponding repair suggestion for each, plus three further points explicitly separated as possible interpretation or file-splitting differences. Recurring defects include omitted conjectures, generalizations, minimality claims, and occurrence directions; empty sInf searches that silently become zero; natural-number division or conversion before integrality and nonnegativity are established; and uncharacterized axioms used in place of sequence definitions. It also gives sharper test cases, including a likely false n=0 boundary in A017666 and p=11 counterexamples suggesting reversed Legendre-symbol arguments in A179537. Other entries flag supplied finiteness assumptions, missing positivity or existence conditions, and source constants weakened to arbitrary ones. The PR author called the review very useful, applied the suggestions broadly, and explained the few places where known instances or interpretive choices were intentionally outside the open targets; KitaKen1 then narrowed or withdrew the corresponding objections. After A129365 was repaired to keep the quotient rational until integrality is proved, KitaKen1 also published proofs of all four of its open targets, recorded separately as a mathematical contribution. A second downstream project supplied a Lean proof of A003625's classical irreducibility criterion, recorded separately as a literature-backed formalization.

Related repositories