Fibonacci degree conjecture for Golombic and Levine polynomials

From papers

Let w=(b1m1bkmk)w=(b_1^{m_1}\dots b_k^{m_k}) be a word in reduced form, and let γn(b1m1bkmk)\gamma_n(b_1^{m_1}\dots b_k^{m_k}) and n(b1m1bkmk)\ell_n(b_1^{m_1}\dots b_k^{m_k}) be the nnth terms of its Golombic and Levine sequences. For the Fibonacci numbers (F0,F1,F2,)=(0,1,1,2,3,5,8,13,21,)(F_0,F_1,F_2,\dots)=(0,1,1,2,3,5,8,13,21,\dots), fix j,k,nNj,k,n\in\mathbb N with jkj\leq k.

Fibonacci degree conjecture. Both polynomial expressions γn(b1m1bkmk)\gamma_n(b_1^{m_1}\dots b_k^{m_k}) and n(b1m1bkmk)\ell_n(b_1^{m_1}\dots b_k^{m_k}) have degree FnF_n in mjm_j and degree Fn1F_{n-1} in bjb_j.

This predicts Fibonacci-number degree growth for the polynomial expressions arising from the Golombic and Levine constructions. The source presents it as an observed pattern following the corresponding polynomiality results; its resolution is not given.

Progress summary

Open

The conjecture remains unproved: its defining paper records the Fibonacci degree pattern but gives no proof or counterexample.

The conjecture predicts that the polynomial expressions from the Golombic and Levine constructions have Fibonacci-number degrees in each exponent and base variable. It appears as Conjecture 5.2 in the paper introducing these sequences.

Known results

  • Theorems 3.11 and 3.15 establish that, for fixed kk, the Golombic and Levine quantities are integer-valued polynomials in the parameters m1,,mk,b1,,bkm_1,\dots,m_k,b_1,\dots,b_k.

2026 paper records the conjecture

The paper describes the Fibonacci-degree pattern as an observed phenomenon and formulates it as Conjecture 5.2. No proof, counterexample, verification, withdrawal, or competing resolution was found in the retrieved sources.

Current status (as of August 2026): Polynomiality is established, but the asserted Fibonacci degrees remain an open conjecture with no publicly recorded proof or counterexample.

Sources
Sources & referencesView supporting material

Primary source

Johan Claes and Roland Miyamoto, “Golombic and Levine sequences”, arXiv:2602.10992 (2026).

Solutions 1

Proof

We prove both degree formulas simultaneously. Let

w=b1m1bkmk,w=b_1^{m_1}\cdots b_k^{m_k},

and initially specialize every mh,bhm_h,b_h to a positive integer. For a positive word u=(u1,,uL)u=(u_1,\ldots,u_L), define

Γ(u)=1u12u2LuL,L(u)=rev(Γ(u)).\Gamma(u)=1^{u_1}2^{u_2}\cdots L^{u_L}, \qquad \mathcal L(u)=\operatorname{rev}(\Gamma(u)).

Choose either O=Γ\mathcal O=\Gamma or O=L\mathcal O=\mathcal L, and write

u(t)=Ot(w),Lt=u(t).u^{(t)}=\mathcal O^t(w),\qquad L_t=|u^{(t)}|.

Fix t1t\ge1, set u=u(t)u=u^{(t)}, and put

S=i=1Ltui,T=i=1Ltiui,M=max1iLtui.S=\sum_{i=1}^{L_t}u_i,\qquad T=\sum_{i=1}^{L_t}i u_i,\qquad M=\max_{1\le i\le L_t}u_i.

Both deployment and reversed deployment have the same length and sum of entries, so

S=Lt+1,T=Lt+2.S=L_{t+1},\qquad T=L_{t+2}.

Since all entries of the preceding row are positive, its deployment contains every label 1,,Lt11,\ldots,L_{t-1}. Reversal preserves these labels, and therefore

M=Lt1.M=L_{t-1}.

The upper bound TLtST\le L_tS is immediate. For the lower bound,

S2=iui2+2i<juiujMiui+2Mj(j1)uj=M(2TS)2MT.\begin{aligned} S^2 &=\sum_i u_i^2+2\sum_{i<j}u_iu_j\\ &\le M\sum_i u_i+2M\sum_j(j-1)u_j\\ &=M(2T-S)\le2MT. \end{aligned}

Consequently, both operators satisfy the universal inequalities

Lt+122Lt1Lt+2LtLt+1(t1).(1)\boxed{\quad \frac{L_{t+1}^2}{2L_{t-1}} \le L_{t+2} \le L_tL_{t+1} \qquad(t\ge1). \quad} \tag{1}

Write sh=m1++mhs_h=m_1+\cdots+m_h, with s0=0s_0=0. Directly,

L0=h=1kmh,L1=h=1kmhbh,L_0=\sum_{h=1}^km_h,\qquad L_1=\sum_{h=1}^km_hb_h,

and

L2=h=1kbhi=sh1+1shi=12h=1kbhmh(2sh1+mh+1).(2)L_2 = \sum_{h=1}^kb_h \sum_{i=s_{h-1}+1}^{s_h}i = \frac12\sum_{h=1}^k b_hm_h(2s_{h-1}+m_h+1). \tag{2}

Fix all parameters except one xx at arbitrary positive integer values. If x=mjx=m_j, the initial degrees are

(d0,d1,d2)=(1,1,2),(d_0,d_1,d_2)=(1,1,2),

because the quadratic coefficient in L2L_2 is bj/2>0b_j/2>0. If x=bjx=b_j, the initial degrees are

(d0,d1,d2)=(0,1,1),(d_0,d_1,d_2)=(0,1,1),

because its coefficient in L2L_2 is

i=sj1+1sji>0.\sum_{i=s_{j-1}+1}^{s_j}i>0.

Suppose dt+1=dt+dt1d_{t+1}=d_t+d_{t-1}. Applying (1) as x+x\to+\infty gives

2dt+1dt1dt+2dt+dt+1.2d_{t+1}-d_{t-1} \le d_{t+2} \le d_t+d_{t+1}.

The two bounds coincide because

2dt+1dt1=dt+dt+1.2d_{t+1}-d_{t-1}=d_t+d_{t+1}.

Hence dt+2=dt+dt+1d_{t+2}=d_t+d_{t+1}, and induction yields

degxLt={Ft+1,x=mj,Ft,x=bj.(3)\deg_xL_t= \begin{cases} F_{t+1},&x=m_j,\\ F_t,&x=b_j. \end{cases} \tag{3}

These conclusions hold for every positive-integer specialization of all remaining parameters. The source's Theorems 3.11 and 3.15 establish that LtL_t is a polynomial in the full parameter list. Any coefficient of a higher power of xx would therefore be a polynomial in the remaining parameters that vanishes on the entire positive-integer lattice, hence vanishes identically. Conversely, each positive specialization already attains the degrees in (3). Thus these are the degrees of the full multivariate polynomials.

Finally,

γn(w)=Γn1(w),n(w)=Ln1(w).\gamma_n(w)=|\Gamma^{n-1}(w)|,\qquad \ell_n(w)=|\mathcal L^{n-1}(w)|.

Substituting t=n1t=n-1 into (3) proves

degmjγn(w)=degmjn(w)=Fn,\deg_{m_j}\gamma_n(w) = \deg_{m_j}\ell_n(w) = F_n,

and

degbjγn(w)=degbjn(w)=Fn1,\deg_{b_j}\gamma_n(w) = \deg_{b_j}\ell_n(w) = F_{n-1},

for every 1jk1\le j\le k and every n1n\ge1.

0 endorsements
Shivam Patel ·