Hanna's A382125 residue pattern

OEIS A382125 states: "Conjectures: a(3n) == A382124(n) (mod 3) for n >= 0; a(3n+1) == 0 (mod 3) and a(3*n+2) == 0 (mod 3) for n >= 0."

Write $A(x)=\exp(\sum_{n\ge1}\sigma(n)\sigma(2n)x^n/(3n))$ and $B(x)=\exp(\sum_{n\ge1}\sigma(n)\sigma(2n)x^n/n)$, with $\sigma(n)=\sum_{d\mid n}d$. The public rational-series definitions are exactly the preregistered expressions.

Statement echo

| Source clause | Lean counterpart | Fidelity | | --- | --- | --- | | Every index $n\ge0$ | claim quantifies over every natural index | Exact | | A382124 generating function | HannaSigmaProductExpModThree.A | Exact rational exponential substitution | | A382125 generating function | HannaSigmaProductExpModThree.B | Exact rational exponential substitution | | $b(3n)\equiv a(n)\pmod3$ | First conjunct: coeff (3*n) B = u, coeff n A = v, u ≡ v [ZMOD 3] | Exact | | $b(3n+1)\equiv0\pmod3$ | Second conjunct: coeff (3*n+1) B = u and u ≡ 0 [ZMOD 3] | Exact | | $b(3n+2)\equiv0\pmod3$ | Second conjunct: coeff (3*n+2) B = v and v ≡ 0 [ZMOD 3] | Exact | | Integer coefficients | Both conjuncts supply integer witnesses and rational coefficient equalities | Witnesses make congruence well typed; no extra hypothesis |

The named conjecture is preregistered in #18864. The exact rational formal series carry integer witnesses for the source's congruence statements. Its settling declaration is D5/S3/Arith/Robin/HannaSigmaProductExpModThreeCube.result.

The literature screen in #18863 and #18864 covers the OEIS entries and their links, the OEIS Open benchmark and four OEIS-proof papers, arXiv:2610.15849, arXiv:2410.07237, arXiv:2608.11941 and arXiv:2607.24832. No settlement was found in that searched scope. Completeness is ASSUMED-UNVERIFIED.

Put $c(n)=\sigma(n)\sigma(2n)/3$. The arithmetic argument proves that $c$ is integer valued and satisfies the Dold congruences. For the prime three, writing $m=3^a k$ with $3\nmid k$ gives $c(3^a k)=\sigma(3^a)^2c(k)$, and $\sigma(3^{a+1})-\sigma(3^a)=3^{a+1}$. This supplies the required power of three. The other primes use the Dold congruences of the divisor sum and their product and index-scaling closure.

The shared DoldExponentialIntegrality.integral_exp_of_dold theorem proves the sufficient Dold integrality criterion. HannaSigmaProductExpModThree.AIntegral applies it to $c$. The identity integerCube_map identifies the cube of the integral lift of $A$ with $B$. The theorem cube_residues applies characteristic-three Frobenius to that lift.

A nonintegral coefficient of $A$, or an index violating a source congruence, contradicts the formal settlement under the displayed definitions. The finite experiment below is supporting evidence; the universal result is the Lean theorem.

The settling theorem is D5/S3/Arith/Robin/HannaSigmaProductExpModThreeCube.result. The axiom closure of every public declaration is contained in {propext, Classical.choice, Quot.sound}. The full statement appears in its Blueprint theorem node.

The supporting source is adapted from AINTLIB at commit 1e63feb21e2dd9e54908d14c9c07058eff3ff1d4, with attribution and Apache-2.0 license in Library/RecurrenceComposition/aintlib2026dieudonnedwork.md.

The escape audit is unfinished: #19891. The published owner declarations D5.S3.Arith.Robin.PrimeFiberPropagation.sigma_prime_pow_succ, D5.S1.Recurrence.LinearRows.LinearExponentDyadicSupport.Agree, D5.S1.Recurrence.LinearRows.LinearExponentDyadicSupport.pow_change and D5.S1.Recurrence.Residue.ParametricExponentialSquareCongruence.linear_ode_unique are included in that unfinished audit. No delivered Reg registration is classified as declared_validated.

Experiment

Pinned experiment directory. Run python3 check.py 300 in that directory; exit 0, RESULT PASS. The script SHA-256 is 319ee58b884641c875833eb3e830906af72d593ab73e1d6b0da6ff155cd352f3.

Computed scope: $c(n)$ is integral for $1\le n\le900$; the Dold congruences hold for primes $p<40$ and $pm\le900$; $A$ is integral through degree 300 and $B$ through degree 900; their tested coefficients match the b-files; $B=A^3$ through degree 300; $a(n)\equiv b(3n)\pmod3$ for $0\le n\le300$; and $b(3n+1)\equiv b(3n+2)\equiv0\pmod3$ for $0\le n<300$.

Declaration classification

Names in this table are relative to D5.S3.Arith.Robin.HannaSigmaProductExpModThreeCube. The shared live escape witness is D5.S1.Recurrence.Algebraic.DoldExponentialIntegrality.IsRIntegralPS.dieudonneDwork_mpr; content classification expands the same-delivery support.

| Declaration | proof_shape | escape_witness | Consumers | | --- | --- | --- | --- | | claim | bind-only | none | result | | result | content | shared witness | external named settlement |

What the settlement shows

• Proved in Lean — mechanism and integrality. DoldExponentialIntegrality.integral_exp_of_dold proves integrality for every integer ghost sequence satisfying Dold. HannaSigmaProductExpModThree.AIntegral establishes the integrality of $A$; the arithmetic factorization at three closes the denominator-prime case.
• Proved in Lean — exact scope. HannaSigmaProductExpModThree.result and HannaSigmaProductExpModThreeCube.result prove the two complete preregistered claims at every natural index. They use no coefficient cutoff.
• Proved in Lean — reusable structure. The general sufficient Dold criterion has an arbitrary integer ghost sequence. CubicFifteenSubstitutionParity.prime_power_subst supplies Frobenius for arbitrary series over a prime residue field. No further named conjecture is claimed settled.
• Computed — owner integration. The source inventory publishes LinearExponentDyadicSupport.pow_change and ParametricExponentialSquareCongruence.linear_ode_unique, with complete Scribe nodes. CubicFifteenSubstitutionParity.prime_power_subst is already public on the protected base. These are reused through their owners.
• Computed — shared support. The delivered source inventory has one shared DoldExponentialIntegrality module, consumed by HannaSigmaProductExpModThree.AIntegral; no second copy is delivered. Open: integration of the independent fibpow lane is outside this delivery.
• Computed — publication structure. Each of the two named problems has its own dossier and one OpenProblemResolutionClaim on its settling result. Open: four-slot registration evidence is incomplete under #19891; required CI and the review triplet govern admission.
• Computed — coefficient checks. The experiment entry gives the command, script identity and exact finite ranges. Those ranges carry no uniform numerical conclusion.
• Open. The residues of $A$ itself modulo three are not determined here.

Inventory evidence: git grep -n -E "^(private )?(noncomputable )?(def|theorem) (pow_change|linear_ode_unique|prime_power_subst|integral_exp_of_dold)|OpenProblemResolutionClaim" -- D5/S1/Recurrence D5/S3/Arith/Robin Blueprint/D5/S1/Recurrence Blueprint/D5/S3/Arith/Robin (exit 0). The module and Scribe inventories establish the publication and sharing statements; they are not mathematical proofs.

Generalization proved in prose

Let $q$ be prime and let $f:\mathbb N\to\mathbb Z$ satisfy the Dold congruences. Suppose $q\mid f(n)$ for every $n$ and $q^{v_q(m)+2}\mid f(qm)-f(m)$ for every positive $m$. Put $g(n)=f(n)/q$. At primes $p\ne q$, cancellation of the factor $q$ preserves the required $p$-power divisibility. At $p=q$, the extra power of $q$ gives $q^{v_q(m)+1}\mid g(qm)-g(m)$. Thus $g$ satisfies Dold, and the sufficient criterion makes $\exp(\sum_{n\ge1}g(n)x^n/n)$ integral. Reduction modulo $q$ and Frobenius identify its $q$-th power with substitution of $x^q$. This is a prose consequence of the general criterion, not a new Lean declaration or another named settlement.

The source's assertion that $B$ is the self-convolution cube of $A$ is represented by integerCube_map after rational mapping. The two stated residue conjectures follow. No other source result is asserted to depend on them.

Literature completeness, priority, the source's identification with the formal generating functions and the fidelity of the prose-to-Lean correspondence are outside the kernel. The source and literature readings are attributed to the preregistration's orchestrator and literature seat. The experiment is finite. The escape-registration evidence remains incomplete under #19891.