Self-Referential Inverse Series Parity
Abstract
The positive-index coefficients of OEIS A393170 are odd exactly at powers of two.
Paul D. Hanna’s OEIS A393170 entry defines an integer series A by requiring coefficient n of the reciprocal of A-nX to vanish at every positive n, fixes a(0)=1, and states the parity conjecture.
PowerSeries(R) is the formal power-series ring over R, X is its indeterminate, coeff(n,F) is the degree-n coefficient, and constantCoeff(F) is the constant coefficient. The expression invOfUnit(F,1) is Mathlib’s inverse for a series whose constant coefficient is one. The operation map(intCast,ZMod(2)) reduces integer coefficients modulo two. All indices and exponents are natural numbers.
Theorem 1.1 (Existence of a normalized source series).
Proof. Machine-checked in Lean as D5/S3/Arith/SelfReferentialInverseSeriesParity.exists_solution (✓ std3). ∎
Source. Repository-derived.
Commentary.
Set a(0)=1. At each positive n, form the prefix containing the already chosen coefficients below n and choose a(n) to be coefficient n of the inverse of that prefix minus nX. The inverse recurrence depends only on coefficients through degree n, and adding a(n)X^n changes its degree-n coefficient by minus a(n). Strong recursion therefore builds an integer series satisfying all the required equations.
Theorem 1.2 (Parity separation).
Proof. Machine-checked in Lean as D5/S3/Arith/SelfReferentialInverseSeriesParity.generating_equation (✓ std3). ∎
Source. Repository-derived.
Commentary.
Put F=map(intCast,A), B=1/F, and C=1/(F+X). The even-index source conditions make the positive even coefficients of B vanish, while the odd-index conditions make the odd coefficients of C vanish. Thus B=1+XU(X^2) and C=V(X^2). From C(1+XB)=B, comparison of odd coefficients gives V=U and hence B=1+XC. Multiplication by F*(F+X), followed by characteristic-two cancellation, gives F^2+F=X.
Theorem 1.3 (The normalized quadratic root is unique).
Proof. Machine-checked in Lean as D5/S3/Arith/SelfReferentialInverseSeriesParity.generating_unique (✓ std3). ∎
Source. Repository-derived.
Commentary.
If F and the constant-one Artin series S satisfy the same quadratic equation, then (1+F+S)*(F-S)=0. The first factor has unit constant coefficient, so cancellation gives F=S.
Theorem 1.4 (Identification with the Artin series).
Proof. Machine-checked in Lean as D5/S3/Arith/SelfReferentialInverseSeriesParity.mod_two_identity (✓ std3). ∎
Source. Repository-derived.
Commentary.
The parity-separation equation and uniqueness identify the reduction of every normalized source solution with the Artin series.
Theorem 1.5 (Support of the constant-one root).
Proof. Machine-checked in Lean as D5/S3/Arith/SelfReferentialInverseSeriesParity.unit_constant_support (✓ std3). ∎
Source. Repository-derived.
Commentary.
The uniqueness theorem identifies any constant-one root with the Artin series. Its coefficient at zero and at each power of two is one, and every other coefficient is zero.
Theorem 1.6 (The A393170 parity conjecture).
Proof. Machine-checked in Lean as D5/S3/Arith/SelfReferentialInverseSeriesParity.a393170_conjecture (✓ std3). ∎
Citation. Paul D. Hanna (2026). OEIS A393170, self-referential inverse series. URL: https://oeis.org/A393170.
Commentary.
At a positive index, reduction modulo two sends the integer coefficient to one exactly when it is odd. The reduction identity and the support of the constant-one root therefore leave precisely the powers of two.
References
- Truth anchor:
D5/S3/Arith/SelfReferentialInverseSeriesParity.a393170_conjecture - Truth anchor:
D5/S3/Arith/SelfReferentialInverseSeriesParity.exists_solution - Truth anchor:
D5/S3/Arith/SelfReferentialInverseSeriesParity.generating_equation - Truth anchor:
D5/S3/Arith/SelfReferentialInverseSeriesParity.generating_unique - Truth anchor:
D5/S3/Arith/SelfReferentialInverseSeriesParity.mod_two_identity - Truth anchor:
D5/S3/Arith/SelfReferentialInverseSeriesParity.unit_constant_support - Dependency: D5/S3/Arith/ArtinSchreierTracePowersOfTwo