Rationality conjecture for generalized Fibonacci coefficient congruence counts

About 5 years old · traced to

Fix k≥2k\geq 2, m≥1m\geq 1, and 0≤a<m0\leq a<m. Set

In(k)(x)=∏i=1n(1+xFi+k−1(k)).I_n^{(k)}(x)=\prod_{i=1}^n\left(1+x^{F_{i+k-1}^{(k)}}\right).

Let hm,a(k)(n)h_{m,a}^{(k)}(n) be the number of coefficients of In(k)(x)I_n^{(k)}(x) congruent to aa modulo mm, and define

Hm,a(k)(x)=∑n≥0hm,a(k)(n)xn.H_{m,a}^{(k)}(x)=\sum_{n\geq 0}h_{m,a}^{(k)}(n)x^n.

Congruence-count rationality conjecture. The generating function Hm,a(k)(x)H_{m,a}^{(k)}(x) is rational.

This extends the corresponding rationality theorem in the ordinary Fibonacci case to generalized Fibonacci numbers. The source gives no proof or resolution.

References

Primary source

Richard P. Stanley, “Theorems and Conjectures on Some Rational Generating Functions”, arXiv:2101.02131 (2021).

Progress summary

Refreshed
Claimed solved

A reader-written argument claims a complete proof in every order and modulus, but no independent verification has been found, so the generalized conjecture is not settled.

Stanley’s 2021 work proves the ordinary Fibonacci case and records the generalized statement, for order k≥2k\geq 2, as a conjecture. The claim concerns rationality of the generating function counting coefficients in each residue class modulo mm.

Known results

  • Ordinary Fibonacci case: rationality proved in Stanley (2021); the generalized case was left conjectural.

Posted attempt

A reader-written argument claims a complete proof for every k≥2k\geq 2, m≥1m\geq 1, and residue aa, using canonical binary representatives, a finite automaton, and a matrix generating-function argument; it has not been independently verified.

Current status (as of August 2026): The ordinary Fibonacci case is settled, while the generalized conjecture has an unverified complete-proof claim and no independent confirmation.

Sources

Solutions 1

ProofThis solution needs a summarySee full solutionHide full solution

The generating function is rational for every order k≥2k\ge2, every modulus m≥1m\ge1, and every residue class, including zero.

Set

wi=Fi+k−1(k),In(X)=∏i=1n(1+Xwi)=∑s=0dncn(s)Xs,dn=∑i=1nwi.w_i=F^{(k)}_{i+k-1}, \qquad I_n(X)=\prod_{i=1}^n(1+X^{w_i}) =\sum_{s=0}^{d_n}c_n(s)X^s, \qquad d_n=\sum_{i=1}^nw_i.

The recurrence wi+k=wi+⋯+wi+k−1w_{i+k}=w_i+\cdots+w_{i+k-1} gives the weight-preserving rewrite

1k0⟶0k1.1^k0\longrightarrow0^k1.

Every represented exponent therefore has a binary representative avoiding 1k01^k0. By the free equal-weight-pair decomposition in Lemma 5.2, every nonsingleton pair generator has one row containing 1k01^k0. Hence two avoiding representatives cannot differ: each represented exponent has exactly one canonical word.

Fix its canonical word bb. The same free decomposition shows that cn(s)c_n(s) equals the number of factorizations of bb into singleton letters and blocks

0k(∗0k−1)j1,j≥0,0^k(*0^{k-1})^j1, \qquad j\ge0,

where each star is an arbitrary binary letter. This equality holds over the integers.

For m≥2m\ge2, count these tilings modulo mm with a state vector

(v0,v1,…,vk)∈(Z/mZ)k+1,(v_0,v_1,\ldots,v_k) \in(\mathbb Z/m\mathbb Z)^{k+1},

initially (1,0,…,0)(1,0,\ldots,0). Here v0v_0 counts completed tilings and the other coordinates track the kk unfinished-block stages. The letter transitions are

letterv0′v1′vj+1′ (1≤j<k)0v0v0+vkvj1v0+vkvk0.\begin{array}{c|ccc} \text{letter}&v_0'&v_1'&v_{j+1}'\ (1\le j<k)\\ \hline 0&v_0&v_0+v_k&v_j\\ 1&v_0+v_k&v_k&0. \end{array}

Also retain the trailing run of ones, capped at kk, and forbid a zero after a run of length kk. This finite automaton accepts precisely the canonical words, and its final coordinate v0v_0 is the corresponding coefficient modulo mm. It has at most

(k+1)mk+1(k+1)m^{k+1}

states.

Let MM be its finite integer adjacency matrix, ee the initial-state vector, and 1a\mathbf1_a the vector selecting final states with v0=av_0=a. The number Ca(n)C_a(n) of represented exponents with coefficient congruent to aa satisfies

Ca(n)=eTMn1a,C_a(n)=e^{\mathsf T}M^n\mathbf1_a,

and therefore

∑n≥0Ca(n)zn=eT(I−zM)−11a∈Q(z).\sum_{n\ge0}C_a(n)z^n =e^{\mathsf T}(I-zM)^{-1}\mathbf1_a \in\mathbb Q(z).

For a≠0a\ne0, every coefficient in the specified residue class corresponds to a represented exponent, so

hm,a(k)(n)=Ca(n).h_{m,a}^{(k)}(n)=C_a(n).

For a=0a=0, internal zero coefficients must also be counted. They are incorporated exactly by

hm,0(k)(n)=dn+1−∑a=1m−1Ca(n).h_{m,0}^{(k)}(n) =d_n+1-\sum_{a=1}^{m-1}C_a(n).

Since wnw_n satisfies the order-kk generalized Fibonacci recurrence, its partial sum dnd_n has a rational generating function, with characteristic polynomial dividing

(T−1)(Tk−Tk−1−⋯−T−1).(T-1)(T^k-T^{k-1}-\cdots-T-1).

Thus the zero-residue generating function is rational as well. For m=1m=1, every coefficient belongs to the zero class, so h1,0(k)(n)=dn+1h_{1,0}^{(k)}(n)=d_n+1 directly.

Therefore all conjectured residue-count generating functions are rational for every k,m,ak,m,a.