slug: oeis-a369083-reflected-quadratic-quarter-parity bibkey: hanna2024a369083 doi: null url: https://oeis.org/A369083 triage: theorem motivation_gids:
- D5/S1/Recurrence/Parity/ReflectedQuadraticQuarterParity
Binomial parity of the A369083 coefficients
Problem
OEIS A369083, Paul D. Hanna, Jan 12 2024, gives the following NAME and
COMMENT, quoted verbatim from Library/Recurrence/hanna2024a369083.md:
Expansion of g.f. A(x) satisfying A(x) = 1 + x*(5*A(x)^2 - A(-x)^2)/4.
Conjecture: a(n) == binomial(4*n+3,n) (mod 2) for n >= 0 (cf. A263133).
Motivation
This is a first-tier OEIS conjecture from the 2024 entry. The target is the assertion for every natural index, including zero. KPI: open problems resolved.
Gap
The supplied search found no proof: the search seat read the OEIS entry and revision history on 2026-09-09 and searched the identifier on arXiv, MathOverflow, and GitHub. This Stage-B seat had no network access and did not repeat those external checks. This search scope does not establish exhaustive absence of a published proof.
Route
Construct A over the integers by coefficient stabilisation with exact division.
The odd-index coefficient of a square is even, proved using the Mathlib
Frobenius identity after reduction modulo two. Thus generating_equation
states A(0) = 1 and 4*(A-1) = X*(5*A^2-rescale(-1,A)^2), the NAME with its
denominator cleared. generating_unique covers every integer series B with
constant coefficient one satisfying that equation.
reflection_linear combines the equation with its x -> -x image and cancels
four in integer power series, giving A(-x) = 5*A - 4 - 6*x*A^2.
Substitution yields quartic_equation:
1 - 4*x + (10*x-1)*A - (5*x+12*x^2)*A^2 + 15*x^2*A^3 - 9*x^3*A^4 = 0.
Over ZMod 2 this becomes F*(1+x*F)^3 = 1, so H := 1+x*F satisfies
H^4-H^3 = X.
The reduced series G of the frozen AbsoluteReciprocalCubeParity satisfies
the same equation, using its generating_equation with absolute value
invisible modulo two. Its mod_two_identity identifies G with
1 + sum_{n>=1} binomial(4*n-1,n)*X^n. Factoring
H^4-H^3-G^4+G^3 through a unit gives this module’s mod_two_identity,
1 + X*reduction(A) = reduction(integer G). Coefficient extraction gives
a(n) == binomial(4*n+3,n+1) (mod 2). The identity choose_shift,
binomial(4*n+3,n+1) = 3*binomial(4*n+3,n), then gives
hanna_conjecture: a n % 2 = (Nat.choose (4*n+3) n : Int) % 2 for every n.
Target generality is I because the proof imports the frozen
AbsoluteReciprocalCubeParity, itself generality G.
Falsifier
A natural index n for which the coefficient a(n) has parity different from
binomial(4*n+3,n) would contradict the assertion. The orchestrator’s exact
coefficient check is supporting evidence only, not a proof for all indices.
Evidence
- Lean module:
D5/S1/Recurrence/Parity/ReflectedQuadraticQuarterParity.lean. - Main theorem:
hanna_conjecture. - Companions:
generating_equation,generating_unique,reflection_linear,quartic_equation,choose_shift,mod_two_identity. generating_equationstates the NAME multiplied by 4, together withconstantCoeff A = 1, over integer power series. After embedding into rational power series, this is equivalent to the NAME’s divided equation.- Axioms: std3 (
propext,Classical.choice,Quot.sound) for all seven public theorems, as reported by the implementation seat. - Frozen prerequisite:
D5/S1/Recurrence/Parity/AbsoluteReciprocalCubeParity, statement_idsha256:a7484b050055466f1a3091c41ffc3856281a6de69fd76b14ba71577ef1c07c75.
Triage
theorem. The formal proof resolves the coefficient assertion at every
natural index for the uniquely constructed integer series.
ASSUMED-UNVERIFIED
The quotations and attribution were supplied by the orchestrator and copied from the Library note. The OEIS entry and revision history were read by the search seat on 2026-09-09, not by this seat. The reported literature scope was identifier search on arXiv, MathOverflow, and GitHub; this seat had no network. The searches did not establish exhaustive literature coverage or first-publication priority. Source-to-Lean identification and the orchestrator’s exact numerical check are not kernel-checked facts.