betaFeedback

WorksMathematics

On Primitive Divisor Multiplicities of Lucas Sequences

Abstract

Let (Un)(U_{n}) be a nondegenerate Lucas sequence with coprime integer parameters. Assuming the rational abcabc conjecture, we prove that almost every index nn has a primitive prime divisor of UnU_{n} of multiplicity one. More quantitatively, the logarithmic contribution of repeated primitive prime factors tends to zero relative to the primitive part along any family with ϕ(n)/nϕ(n) / n bounded below, and also in Cesàro mean over all indices. We derive explicit logarithmic lower bounds for Lucas non-Wieferich primes with ranks in prescribed residue classes. Both signs of the discriminant are allowed. Independently, we describe initial multiplicities at good odd primes by a unique Hensel root in the recurrence-parameter disk and by the exact contraction rate of a normalized Dickson-polynomial iteration. Applications include rank and period lifting for Fibonacci numbers and the Wall-Sun-Sun condition.

Introduction

Let u,v∈Zu,v \in\mathbb{Z} be coprime integers, v≠0v \ne0, and let α,β\alpha,\beta be the roots of X2−uX+vX^2-uX+v. We assume throughout that α/β\alpha/\beta is not a root of unity. In particular, Δ=u2−4v≠0\Delta=u^2-4v \ne0 and u≠0u \ne0. Define the Lucas sequences of the first and second kinds by

Un=αn−βnα−β,Vn=αn+βn(n≥0).U_n=\frac{\alpha^n-\beta^n}{\alpha-\beta},\qquad V_n=\alpha^n+\beta^n \qquad(n\ge0).

Equivalently, both satisfy

xn+2=uxn+1−vxn,(1)x_{n+2}=ux_{n+1}-vx_n, \tag*{(1)}

with initial values (x0,x1)=(0,1)(x_0,x_1)=(0,1) and (2,u)(2,u) respectively. Put A=max⁡(∣α∣,∣β∣)>1A=\max(|\alpha|,|\beta|)>1. For a prime p∤vp\nmid v, let kU(p)k_U(p) denote its rank of apparition, the least positive nn with p∣Unp\mid U_n, and write rU(p)=νp(UkU(p))r_U(p)=\nu_p(U_{k_U(p)}), where νp\nu_p is the pp-adic valuation, with νp(0)=+∞\nu_p(0)=+\infty. Such a rank always exists (see Proposition 2.3). We call pp a primitive divisor at index nn if kU(p)=nk_U(p)=n, and simple if additionally rU(p)=1r_U(p)=1. The usual Lucas primitive-divisor convention excludes primes dividing Δ\Delta; our first-occurrence convention includes them. This affects only finitely many indices for any fixed sequence.

The classical primitive-divisor theorem ensures that, for n>30n>30, UnU_n has a primitive divisor in the usual sense [4]; for real roots, Carmichael’s earlier theorem gives a stronger threshold [6]. For a comprehensive account of Lucas sequences, see Ballot and Williams [3]. These existence results do not control the multiplicity of a primitive divisor. For the Fibonacci sequence (u,v)=(1,−1)(u,v)=(1,-1), the assertion that every primitive divisor is simple is equivalent to the nonexistence of Wall–Sun–Sun primes. In general it is false: even within Q(5)\mathbb{Q}(\sqrt{5}), one can construct Lucas sequences in which a prescribed odd prime p≠5p\ne5 has initial multiplicity at least two (Proposition 6.11).

We investigate two aspects of initial multiplicity: its global distribution over the indices, conditional on abcabc, and its local description as a function of the recurrence parameter. For the first, write

Pn=∏kU(p)=nprU(p),Sn=∏kU(p)=nrU(p)=1p,Hn=∏kU(p)=nrU(p)≥2prU(p).(2)\begin{aligned} P_n&=\prod_{k_U(p)=n}p^{r_U(p)},\qquad S_n=\prod_{\substack{k_U(p)=n\\r_U(p)=1}}p,\qquad H_n=\prod_{\substack{k_U(p)=n\\r_U(p)\ge2}}p^{r_U(p)}. \tag*{(2)} \end{aligned}

Thus Pn=SnHnP_n=S_nH_n. Empty products are 1. Throughout, ϕ\phi denotes Euler’s totient function, μ\mu the Möbius function, and ω(n)\omega(n) the number of distinct prime factors of nn. By the rational abcabc conjecture we mean that, for each ϵ>0\epsilon>0, coprime positive integers a+b=ca+b=c satisfy c≪ϵrad⁡(abc)1+ϵc\ll_\epsilon\operatorname{rad}(abc)^{1+\epsilon}, where rad⁡(m)=∏p∣mp\operatorname{rad}(m)=\prod_{p\mid m}p for nonzero mm.

Theorem 1.1 (Conditional abundance of simple primitive divisors). Assume the rational abcabc conjecture. For every fixed nondegenerate Lucas sequence with coprime integer parameters the following hold.

  1. For every η,δ>0\eta,\delta>0, uniformly over indices n>1n>1 with ϕ(n)≥ηn\phi(n)\ge\eta n,

Hn≪u,v,η,δPnδ,Sn≫u,v,η,δPn1−δ.(3)H_n\ll_{u,v,\eta,\delta}P_n^\delta,\qquad S_n\gg_{u,v,\eta,\delta}P_n^{1-\delta}. \tag*{(3)}

In particular, log⁡Sn/log⁡Pn→1\log S_n/\log P_n\to1 along any unbounded family of such indices.

  1. Put Rn=log⁡Hn/log⁡PnR_n=\log H_n/\log P_n if Pn>1P_n>1, and Rn=0R_n=0 otherwise. Then

lim⁡X→∞1X∑n≤XRn=0.\lim_{X\to\infty}\frac{1}{X}\sum_{n\le X}R_n=0.
  1. The indices nn with Sn>1S_n>1 have natural density one. Moreover, for every fixed prime pp, all sufficiently large indices pap^a have Spa>1S_{p^a}>1.

These conclusions also hold if primitive divisors are required to be coprime to the discriminant.

This result refines the conditional method of extracting non-Wieferich primes from cyclotomic factors. Ribenboim and Walsh [17] obtained bounds on powerful parts of binary recurrent sequences of positive discriminant; see also Yabuta [25] for negative discriminant and Ribenboim [16] for conditional infinitude of Lucas non-Wieferich primes. Peng [11] treats a generalization of Fibonacci–Wieferich primes. Most directly related to our argument is the work of Anitha, Fathima and Vijayalakshmi [2], who, using the number-field abcabc conjecture, obtain logarithmic lower bounds for Lucas non-Wieferich primes in the residue classes ±1\pm1 modulo a fixed integer. Their cyclotomic extraction uses a positive-proportion family of indices with ϕ(n)/n\phi(n)/n bounded below. We use the rational Lucas identity to work with rational abcabc and obtain a density-one statement for ranks themselves, without a restriction on the sign of the discriminant. The powerful-part bound and cyclotomic estimates are established tools; the focus of Theorem 1.1 is their consequence for the distribution of simple primitive divisors.

Corollary 1.2 (Conditional rank distribution). Assume rational abcabc. For h≥1h \ge1 and bb (mod hh), let Nh,b(x)N_{h,b}(x) count primes p≤xp \le x, p∤2vΔp \nmid2v\Delta, with rU(p)=1r_U(p) = 1 and kU(p)≡bk_U(p) \equiv b (mod hh). Then

lim inf⁡x→∞Nh,b(x)log⁡x≥1hlog⁡A.(4)\liminf_{x \to\infty} \frac{N_{h,b}(x)}{\log x} \ge\frac{1}{h\log A}. \tag*{(4)}

In particular, for h≥2h \ge2 the same lower bound holds for the count of Lucas non-Wieferich primes p≤xp \le x with p≡±1p \equiv\pm1 (mod hh).

Here a prime p∤2vΔp \nmid2v\Delta is Lucas non-Wieferich if νp(Up−(Δ/p))=1\nu_p(U_{p-(\Delta/p)}) = 1, equivalently rU(p)=1r_U(p) = 1 by the rank and lifting formulas, where (⋅/p)(\cdot/p) is the Legendre symbol. The conclusion for arbitrary bb restricts the rank rather than just the prime’s residue class.

Our second result concerns variation of uu at fixed vv. Write Un(X,v),Vn(X,v)∈Z[X]U_n(X,v), V_n(X,v) \in\mathbb{Z}[X] for the Lucas polynomials obtained by replacing uu by XX in (1.1). They are Dickson polynomials of the second and first kinds, respectively (with index n−1n-1 in the second-kind case); see [8]. At an odd prime p∤vΔp \nmid v\Delta, set

λp=(vp)v−(p−1)/2∈Zp×,Tp(X)=λpVp(X,v).(5)\lambda_p = \left(\frac{v}{p}\right)v^{-(p-1)/2} \in\mathbb{Z}_p^\times,\qquad T_p(X) = \lambda_p V_p(X,v). \tag*{(5)}

The normalization ensures λp2vp=v\lambda_p^2v^p = v and λp≡1\lambda_p \equiv1 (mod pp), so that the iteration preserves the norm parameter and the residue class of uu.

Theorem 1.3 (Local multiplicity and normalized iteration). Let p∤2vΔp \nmid2v\Delta be prime, k=kU(p)k = k_U(p) and r=rU(p)r = r_U(p). There is a unique u^∈u+pZp\hat{u} \in u + p\mathbb{Z}_p with Uk(u^,v)=0U_k(\hat{u},v) = 0. For x,y∈u+pZpx,y \in u + p\mathbb{Z}_p,

νp(Uk(x,v))=νp(x−u^),(6)\nu_p(U_k(x,v)) = \nu_p(x-\hat{u}), \tag*{(6)}
νp(Tp(x)−Tp(y))=1+νp(x−y).(7)\nu_p(T_p(x)-T_p(y)) = 1+\nu_p(x-y). \tag*{(7)}

In particular, u^\hat{u} is the unique fixed point of TpT_p in this disk. Setting uj=Tpj(u)u_j = T_p^j(u), for every j≥0j \ge0 one has

uj=λp(pj−1)/(p−1)Vpj(u,v),(8)u_j = \lambda_p^{(p^j-1)/(p-1)}V_{p^j}(u,v), \tag*{(8)}
νp(uj−u^)=νp(uj+1−uj)=r+j.(9)\nu_p(u_j-\hat{u}) = \nu_p(u_{j+1}-u_j) = r+j. \tag*{(9)}

Consequently νp(Tp(u)−u)=r\nu_p(T_p(u)-u)=r, and the exceptional parameter modulo pap^a is uniquely determined for every a≥1a \ge1.

For v=−1v=-1, the normalization is trivial, and the iteration consists of the usual trace polynomials. In particular, for Fibonacci numbers Tpj(1)=LpjT_p^j(1)=L_{pj} and νp(Lpj−1)=rU(p)\nu_p(L_{pj}-1)=r_U(p). Bragman and Rowland [5 Theorem 10] express the Fibonacci Wall exponent using the principal-unit part of the golden ratio. The local description here is consistent with theirs: its focus is the recurrence parameter, the norm-preserving normalization for arbitrary vv, and the exact valuation of each iteration step. McIntosh and Roettger [9 Theorem 1(v)] already give a congruence characterization of Lucas–Wieferich primes in terms of Vp(u,v)V_p(u,v) for p∤2uvΔp \nmid2uv\Delta. Our valuation identity refines this characterization to all exponents and also allows p∣up \mid u. The underlying Dickson composition and power-map dynamics are classical [8, 7]. For related congruences and derivative identities for Lucas polynomials, see Yamagishi [26]; Müller [10] studies the distribution of ranks when a recurrence parameter varies over a finite field.

The proofs are organized as follows. § 2 records the lifting and rank results needed below, including the discriminant-prime case. § 3 establishes the primitive-factor estimates, including the prime 2. § 4 proves Theorem 1.1 and Corollary 1.2; § 5 proves Theorem 1.3. Finally, § 6 gives Fibonacci applications, drawing on the author’s earlier notes [12, 13, 14, 15].

Lifting in rings of algebraic integers

Throughout, νp\nu_p is the usual pp-adic valuation on integers, with νp(0)=+∞\nu_p(0)=+\infty. We also use νp\nu_p for the valuation on Qp\mathbb{Q}_p normalized by νp(p)=1\nu_p(p)=1. For a number field KK we use congruence modulo pj\mathfrak{p}^j in OK\mathcal{O}_K to mean congruence modulo the ideal pjOK\mathfrak{p}^j\mathcal{O}_K. In particular, an integer in pjOK\mathfrak{p}^j\mathcal{O}_K belongs to pjZp^j\mathbb{Z}, since OK∩Q=Z\mathcal{O}_K \cap\mathbb{Q}=\mathbb{Z}.

Theorem 2.1 (Algebraic lifting). Let pp be an odd rational prime, and let α,β∈OK\alpha,\beta\in\mathcal{O}_K be units modulo pOKp\mathcal{O}_K. Put

An={∑i=0n−1αn−1−iβi,n≥1,0,n=0.A_n = \begin{cases} \sum_{i=0}^{n-1}\alpha^{n-1-i}\beta^i, & n \ge1,\\ 0, & n=0. \end{cases}

Suppose 0≠Ak∈pOK0 \ne A_k \in p\mathcal{O}_K. For every positive integer mm and prime ideal p∣p\mathfrak{p} \mid p,

vp(Akm)=vp(Ak)+epνp(m),(10)v_{\mathfrak{p}}(A_{km})=v_{\mathfrak{p}}(A_k)+e_{\mathfrak{p}}\nu_p(m), \tag*{(10)}

where ep=vp(p)e_{\mathfrak{p}}=v_{\mathfrak{p}}(p) and vpv_{\mathfrak{p}} is the normalized ideal valuation. In particular, if AnA_n is integral over Z\mathbb{Z} and rational for every nn, then

νp(Akm)=νp(Ak)+νp(m).(11)\nu_p(A_{km})=\nu_p(A_k)+\nu_p(m). \tag*{(11)}

Proof. The polynomial identity Akt=AkSt(c,d)A_{kt}=A_kS_t(c,d) holds with c=αkc=\alpha^k, d=βkd=\beta^k and St(c,d)=∑i=0t−1ct−1−idiS_t(c,d)=\sum_{i=0}^{t-1}c^{t-1-i}d^i. Since c−d=(α−β)Ak∈pOKc-d=(\alpha-\beta)A_k \in p\mathcal{O}_K, we have c≡d(modp)c \equiv d \pmod{\mathfrak{p}}. If p∤tp \nmid t, then St(c,d)≡tdt−1(modp)S_t(c,d) \equiv td^{t-1} \pmod{\mathfrak{p}} is a unit modulo pOKp\mathcal{O}_K. For t=pt=p, write c=d+pzc=d+pz, where z∈OKz\in\mathcal{O}_K. The binomial theorem gives

Sp(c,d)=∑j=1p(pj)dp−j(pz)j−1≡pdp−1(modp2).S_p(c,d)=\sum_{j=1}^{p}\binom{p}{j}d^{p-j}(pz)^{j-1}\equiv pd^{p-1}\pmod{p^2}.

Here the intermediate terms are in p2OKp^2\mathcal{O}_K, and so is the last term because p≥3p\geq3. Hence Sp(c,d)=pwS_p(c,d)=pw with ww a unit modulo pOKp\mathcal{O}_K. Replacing c,dc,d by their ppth powers preserves their congruence and unit properties. Factor m=patm=p^a t with p∤tp\nmid t and apply the two calculations successively. The factors prime to every p∣p\mathfrak{p}\mid p have valuation zero, and each factor pp has valuation epe_{\mathfrak{p}}. This proves (2.1); for rational integers it implies (2.2). □

Remark 2.2. No unramifiedness hypothesis is required. For p=2p=2 the same step is valid if c≡d(mod4)c\equiv d\pmod{4}, since S2(c,d)=c+d≡2d(mod4)S_2(c,d)=c+d\equiv2d\pmod{4}. It need not hold when only c≡d(mod2)c\equiv d\pmod{2}; for instance, F3=2F_3=2 but F6=8F_6=8.

Ranks of apparition

Proposition 2.3. The sequence (Un)(U_n) satisfies gcd⁡(Um,Un)=∣Ugcd⁡(m,n)∣\gcd(U_m,U_n)=\lvert U_{\gcd(m,n)}\rvert for positive m,nm,n. For every M>1M>1 coprime to vv, the least positive index kU(M)k_U(M) with M∣UkU(M)M\mid U_{k_U(M)} exists, and

M∣Un⟺kU(M)∣n.M\mid U_n\quad\Longleftrightarrow\quad k_U(M)\mid n.

For an odd prime p∤vp\nmid v, put ε=(Δ/p)\varepsilon=(\Delta/p). Then

kU(p)∣p−ε,Up≡ε(modp).(12)k_U(p)\mid p-\varepsilon,\qquad U_p\equiv\varepsilon\pmod p. \tag*{(12)}

If p∣Δp\mid\Delta, then kU(p)=pk_U(p)=p. A prime dividing vv divides no UnU_n with n≥1n\geq1.

Proof. The identity Um+n=UmUn+1−vUm−1UnU_{m+n}=U_mU_{n+1}-vU_{m-1}U_n follows from the recurrence (1.1). Every prime dividing vv is coprime to each positive-index UnU_n, since Un≡un−1U_n\equiv u^{n-1} modulo that prime. The recurrence then shows gcd⁡(Um,Um−1)=1\gcd(U_m,U_{m-1})=1 by descending induction. The addition identity implies gcd⁡(Um,Um+n)=gcd⁡(Um,Un)\gcd(U_m,U_{m+n})=\gcd(U_m,U_n), so the Euclidean algorithm proves strong divisibility. The recurrence state (Un,Un+1)(U_n,U_{n+1}) evolves by an invertible matrix modulo MM if gcd⁡(M,v)=1\gcd(M,v)=1. It returns to (0,1)(0,1), giving existence; strong divisibility and minimality yield the equivalence.

Modulo pp, if Δ\Delta is a nonzero square, the two nonzero roots lie in Fp\mathbb{F}_p and have (p−1)(p-1)st powers equal to 11. If Δ\Delta is a nonsquare, the roots lie in Fp2\mathbb{F}_{p^2} and Frobenius interchanges them, so their (p+1)(p+1)st powers are both vv. In either case the Binet expression gives Up−ε=0U_{p-\varepsilon}=0 and Up=εU_p=\varepsilon modulo pp. In the remaining case the polynomial has a double root c=u/2≠0c=u/2\neq0 in Fp\mathbb{F}_p; the polynomial sum defining UnU_n reduces to ncn−1nc^{n-1}. This gives kU(p)=pk_U(p)=p and Up≡0(modp)U_p\equiv0\pmod p. □

Corollary 2.4. Let pp be an odd prime not dividing vv, k=kU(p)k=k_U(p) and r=νp(Uk)r=\nu_p(U_k). Then for every positive nn and aa,

νp(Un)={r+νp(n/k),k∣n,0,k∤n,(13)\nu_p(U_n)= \begin{cases} r+\nu_p(n/k), & k\mid n,\\ 0, & k\nmid n, \end{cases} \tag*{(13)}
kU(pa)=pmax⁡(0,a−r)k.(14)k_U(p^a)=p^{\max(0,a-r)}k. \tag*{(14)}

Proof. Apply Theorem 2.1 to the roots of X2−uX+vX^2-uX+v (both are units modulo pp because their product is vv) and use Proposition 2.3. The smallest multiplier with valuation at least a−ra-r gives (14).

Primes dividing the discriminant

Proposition 2.5. For odd primes pp one has p∣Upp \mid U_p if and only if p∣Δp \mid\Delta. If p∣Δp \mid\Delta and p>3p>3, then

Up≡p(modp2).(15)U_p \equiv p \pmod{p^2}. \tag*{(15)}

Moreover 9∣U39 \mid U_3 if and only if Δ≡6(mod9)\Delta\equiv6 \pmod{9}.

Proof. For p∤vp \nmid v the first assertion follows from Proposition 2.3; if p∣vp \mid v, then p∤UpΔp \nmid U_p\Delta by coprimality. The symmetric expansion is

2p−1Up=∑j=0(p−1)/2(p2j+1)up−2j−1Δj.2^{p-1}U_p=\sum_{j=0}^{(p-1)/2}\binom{p}{2j+1}u^{p-2j-1}\Delta^j.

If p∣Δp \mid\Delta and p>3p>3, every term except pup−1pu^{p-1} is divisible by p2p^2, including the last term since (p−1)/2≥2(p-1)/2 \ge2. Since p∤up \nmid u, Fermat’s little theorem gives (15). Finally 4U3=3u2+Δ4U_3=3u^2+\Delta. If 9∣U39 \mid U_3 then 3∣Δ3 \mid\Delta, and u2≡1(mod3)u^2 \equiv1 \pmod{3}. Conversely Δ≡6(mod9)\Delta\equiv6 \pmod{9} implies 3∤u3 \nmid u; the displayed identity proves the equivalence.

Remark 2.6. This corrects the one-sided binomial expansion in [15 Lemma 4.1]: p∣Δp \mid\Delta does not in general imply Δ∈pOK\sqrt{\Delta} \in p\mathcal{O}_K. The symmetric expansion avoids that issue. The valuation formulas agree with the general results of Sanna [21]. In particular, νp(Un)=νp(n)\nu_p(U_n)=\nu_p(n) for p>3p>3 dividing Δ\Delta; for p=3∣Δp=3 \mid\Delta and 3∣n3 \mid n it is ν3(U3)+ν3(n)−1\nu_3(U_3)+\nu_3(n)-1.

Cyclotomic factors and their primitive parts

For n>1n>1 let

Cn=∣Φn(α,β)∣=∏d∣n∣Ud∣μ(n/d).(16)C_n=\left|\Phi_n(\alpha,\beta)\right|=\prod_{d\mid n}\left|U_d\right|^{\mu(n/d)}. \tag*{(16)}

where Φn(X,Y)=Yφ(n)Φn(X/Y)\Phi_n(X,Y)=Y^{\varphi(n)}\Phi_n(X/Y) is the homogeneous cyclotomic polynomial. The identity follows by Möbius inversion from the cyclotomic factorization of αn−βn\alpha^n-\beta^n. The polynomial is symmetric for n>1n>1, so its value is a rational algebraic integer and CnC_n is a positive integer. Moreover ∣Un∣=∏d∣nd>1Cd\left|U_n\right|=\prod_{\substack{d\mid n\\d>1}}C_d. These factors, also known as Lucas atoms or Möbius duals, have been studied in [20, 1, 19]. We record their valuations to control the nonprimitive contribution, with a short proof from Theorem 2.1.

Proposition 3.1 ([1, 19]). Let n>1n > 1, let p∤vp \nmid v be prime, and write k=kU(p)k = k_U(p), r=rU(p)r = r_U(p). If pp is odd, then

νp(Cn)={r,n=k,1,n=pak, a≥1,0,otherwise.(17)\nu_p(C_n) = \begin{cases} r, & n = k, \\ 1, & n = p^a k,\ a \ge1, \\ 0, & \text{otherwise}. \end{cases} \tag*{(17)}

If p=2p = 2, then vv is odd, k=2k = 2 for even uu and k=3k = 3 for odd uu, and

ν2(Cn)={r,n=k,ν2(Vk),n=2k,1,n=2ak, a≥2,0,otherwise.(18)\nu_2(C_n) = \begin{cases} r, & n = k, \\ \nu_2(V_k), & n = 2k, \\ 1, & n = 2^a k,\ a \ge2, \\ 0, & \text{otherwise}. \end{cases} \tag*{(18)}

A prime dividing vv divides no CnC_n.

Proof. For odd pp, insert (2.4) into the Möbius transform in (3.1). If k∤nk \nmid n, all terms vanish. If n=kmn = km, the transform is

∑t∣mμ(m/t)(r+νp(t)).\sum_{t\mid m} \mu(m/t)\left(r+\nu_p(t)\right).

The constant contributes only at m=1m = 1. The transform of νp(t)=∑a≥11pa∣t\nu_p(t) = \sum_{a\ge1} 1_{p^a\mid t} is 11 precisely at m=pam = p^a for a≥1a \ge1. This also covers primes dividing Δ\Delta, with their own initial value rr.

For p=2p = 2, the ranks follow from the recurrence (1.1) modulo 2. If 2∣Uj2 \mid U_j, the roots αj\alpha^j, βj\beta^j are congruent modulo 2 in their ring of integers and are units modulo 2, so the quotient Ujt/UjU_{jt}/U_j is odd for odd tt by the geometric-sum identity. Since Vj=2Uj+1−uUjV_j = 2U_{j+1} - uU_j, VkV_k is even. The identities

U2j=UjVj,V2j=Vj2−2vjU_{2j} = U_jV_j,\qquad V_{2j} = V_j^2 - 2v^j

show that the first doubling increases the valuation by ν2(Vk)\nu_2(V_k), and each subsequent doubling increases it by one, since V2j≡2(mod4)V_{2j} \equiv2 \pmod4 whenever VjV_j is even and vv is odd. Thus ν2(Ukt)=r\nu_2(U_{kt}) = r for odd tt, and r+ν2(Vk)+ν2(t)−1r + \nu_2(V_k) + \nu_2(t) - 1 for even $t. Its Möbius transform gives (3.3). Finally, p∣vp \mid v implies p∤Udp \nmid U_d for every positive dd by Proposition 2.3, hence p∤Cnp \nmid C_n. ∎

Corollary 3.2. For n>1n > 1, write Cn=DnPnC_n = D_nP_n. The odd part of DnD_n is either 11 or the largest prime factor QQ of nn, occurring to exponent one. In particular,

1≤Dn≪u,vn.(19)1 \le D_n \ll_{u,v} n. \tag*{(19)}

Proof. By Proposition 3.1, a nonprimitive odd divisor pp requires n=pakU(p)n = p^a k_U(p) for a≥1a \ge1. If p∣Δp \mid\Delta, then kU(p)=pk_U(p) = p; otherwise kU(p)∣p−(Δp)k_U(p) \mid p - \left(\frac{\Delta}{p}\right), whose prime factors are all smaller than pp (also for p+1p + 1, which is even). Thus p=Qp = Q and its exponent is one. The contribution of 2 is bounded by 2max⁡(1,ν2(VkU(2)))2^{\max(1,\nu_2(V_{k_U(2)}))} if vv is odd, and is absent otherwise. This proves the bound. ∎

Lemma 3.3 (Growth of the primitive part). For every fixed nondegenerate Lucas sequence with coprime integer parameters,

log⁡Pn∼ϕ(n)log⁡A(n→∞).(20)\log P_n \sim\phi(n)\log A \qquad(n \to\infty). \tag*{(20)}

If Δ>0\Delta> 0, the sharper estimate log⁡Pn=ϕ(n)log⁡A+Ou,v(log⁡n)\log P_n = \phi(n)\log A + O_{u,v}(\log n) holds.

Proof. Choose ∣α∣=A≥∣β∣|\alpha| = A \ge|\beta|. Since αβ=v\alpha\beta= v is a nonzero integer, A≥1A \ge1. Equality would force both absolute values to be one; if the roots are real their quotient is then a root of unity, and if complex, v=1v = 1 and u∈{−1,0,1}u \in\{-1,0,1\}, again giving a root of unity. Hence A>1A > 1. By (16),

log⁡Cn=ϕ(n)log⁡A+∑d∣nμ(n/d)log⁡∣1−(β/α)d∣.(21)\log C_n = \phi(n)\log A + \sum_{d\mid n}\mu(n/d)\log\left|1-(\beta/\alpha)^d\right|. \tag*{(21)}

For real roots, ∣β/α∣<1|\beta/\alpha| < 1 by nondegeneracy. The absolute sum of the error terms over all d≥1d \ge1 converges, so log⁡Cn=ϕ(n)log⁡A+Ou,v(1)\log C_n = \phi(n)\log A + O_{u,v}(1).

For complex conjugate roots the ratio z=β/αz = \beta/\alpha has modulus one but is not a root of unity. The standard bound for a linear form in logarithms, in the form of [23 Lemma 3], gives that the logarithm log⁡∣1−zd∣\log|1-z^d| is at least −cu,v(log⁡(2d))2-c_{u,v}(\log(2d))^2 for some positive constant; it is at most log⁡2\log2. Therefore the error in (21) is Ou,v(2ω(n)(log⁡(2n))2)O_{u,v}(2^{\omega(n)}(\log(2n))^2). The elementary estimates 2ω(n)=no(1)2^{\omega(n)} = n^{o(1)} and ϕ(n)=n1−o(1)\phi(n) = n^{1-o(1)} show this is o(ϕ(n))o(\phi(n)). For completeness, for each ϵ>0\epsilon> 0 the inequalities 2≤pϵ2 \le p^\epsilon and 1−1/p≥p−ϵ1 - 1/p \ge p^{-\epsilon} hold for all sufficiently large primes pp; the remaining primes contribute a fixed constant to the two products. Combining with (19) proves both assertions.

Conditional density of simple primitive divisors

We use the rational Lucas identity to bound the entire powerful part, and then remove the nonprimitive factors. The estimate below is a version of the powerful-part bounds of Ribenboim and Walsh [17]; the proof also works for negative discriminant, as in the setting of Yabuta [25].

Lemma 4.1 (Conditional powerful-part bound). Assume rational abc. For n≥1n \ge1, define the powerful part

Tn=∏νp(Un)≥2pνp(Un).T_n = \prod_{\nu_p(U_n)\ge2} p^{\nu_p(U_n)}.

For each δ>0\delta> 0, it satisfies Tn≪u,v,δAδnT_n \ll_{u,v,\delta} A^{\delta n}. In particular, log⁡Hn=o(n)\log H_n = o(n).

Proof. The identity

Vn2−ΔUn2=4vn(22)V_n^2-\Delta U_n^2=4v^n \tag*{(22)}

has three nonzero terms: nondegeneracy excludes Un=0U_n = 0 and Vn=0V_n = 0 for n≥1n \ge1. If p∣vp \mid v, then Un≡un−1U_n \equiv u^{n-1} and Vn≡un(modp)V_n \equiv u^n \pmod p, so neither is divisible by pp. It follows that the common divisor gg of the three terms in (22) divides 44 (if vv is even then uu is odd and the common divisor is 11). Dividing by gg gives pairwise coprime integers, and moving the negative terms to the other side gives an equation of the form a+b=ca+b=c with a,b,c>0a,b,c>0.

The maximum of the absolute values of the original terms is ≍u,vA2n\asymp_{u,v} A^{2n}. For real roots, this follows from ∣Vn∣≍u,vAn|V_n|\asymp_{u,v} A^n; for complex roots, use ∣4vn∣=4A2n|4v^n|=4A^{2n}. Also ∣Un∣,∣Vn∣≪u,vAn|U_n|,|V_n|\ll_{u,v} A^n. The radical of 4vnΔ4v^n\Delta is fixed, so the conjecture implies, for every ϵ>0\epsilon>0,

A2n≪u,v,ϵ(Anrad⁡(∣Un∣))1+ϵ.A^{2n}\ll_{u,v,\epsilon}\left(A^n\operatorname{rad}(|U_n|)\right)^{1+\epsilon}.

The radical is thus ≫u,v,ϵAn(1−ϵ)/(1+ϵ)\gg_{u,v,\epsilon} A^{n(1-\epsilon)/(1+\epsilon)}. On the other hand,

rad⁡(∣Un∣)=∣Un∣Tnrad⁡(Tn)≤∣Un∣Tn≪u,vAnTn.\operatorname{rad}(|U_n|)=\frac{|U_n|}{T_n}\operatorname{rad}(T_n)\le\frac{|U_n|}{\sqrt{T_n}}\ll_{u,v}\frac{A^n}{\sqrt{T_n}}.

Combining gives Tn≪u,v,ϵA4ϵn/(1+ϵ)T_n\ll_{u,v,\epsilon} A^{4\epsilon n/(1+\epsilon)}. Taking ϵ\epsilon sufficiently small proves the claim. Since Hn∣TnH_n\mid T_n, the final assertion follows.

Proof of Theorem 1.1. By Lemma 3.3, for nn sufficiently large we have log⁡Pn≥12ϕ(n)log⁡A\log P_n\ge\frac{1}{2}\phi(n)\log A. In particular, if ϕ(n)≥ηn\phi(n)\ge\eta n, apply Lemma 4.1 with exponent ηδ/2\eta\delta/2 to obtain Hn≪u,v,η,δPnδH_n\ll_{u,v,\eta,\delta} P_n^\delta. The bound for Sn=Pn/HnS_n=P_n/H_n follows. The finitely many small indices can be absorbed in the constants. Taking arbitrary δ>0\delta>0 proves the logarithmic limiting assertion.

For the mean-value assertion, observe that

1X∑n≤Xnϕ(n)≤∑d≥1μ(d)2dϕ(d)=∏p prime(1+1p(p−1))=:C<∞.(23)\frac{1}{X}\sum_{n\le X}\frac{n}{\phi(n)}\le\sum_{d\ge1}\frac{\mu(d)^2}{d\phi(d)}=\prod_{p\ \mathrm{prime}}\left(1+\frac{1}{p(p-1)}\right)=:C<\infty. \tag*{(23)}

Indeed, the identity n/ϕ(n)=∑d∣nμ(d)2/ϕ(d)n/\phi(n)=\sum_{d\mid n}\mu(d)^2/\phi(d) follows by multiplicativity, and the product converges by comparison with ∑m≥2m−2\sum_{m\ge2}m^{-2}. By Lemma 4.1, for each ϵ>0\epsilon>0, log⁡Hn≤ϵn\log H_n\le\epsilon n for all sufficiently large nn. Hence for such nn,

Rn≤2ϵlog⁡Anϕ(n).R_n\le\frac{2\epsilon}{\log A}\frac{n}{\phi(n)}.

The finite initial portion contributes o(1)o(1) to the mean. By (4.2) the limsup of the mean is at most 2Cϵ/log⁡A2C\epsilon/\log A, for every ϵ>0\epsilon>0, and is therefore zero.

By Lemma 3.3, Pn>1P_n>1 for all sufficiently large nn. When Sn=1S_n=1 and Pn>1P_n>1 we have Rn=1R_n=1. The mean-value conclusion therefore shows that the set of such indices has density zero. For n=pan=p^a we have ϕ(n)/n=1−1/p\phi(n)/n=1-1/p, so the first assertion with any δ<1\delta<1 and Lemma 3.3 show that Sn>1S_n>1 for all sufficiently large aa. Finally, removing the primes dividing Δ\Delta alters only the finitely many indices at which such a prime first appears; the estimates and limits remain valid after a change of constants.

Proof of Corollary 1.2. By Theorem 1.1, the number of indices n≤Yn\le Y in the class b(modh)b\pmod h with Sn>1S_n>1 is Y/h+o(Y)Y/h+o(Y). Excluding the finitely many ranks of primes dividing 2vΔ2v\Delta, choose a simple primitive divisor pnp_n of each such UnU_n. The primes are distinct because the ranks are distinct. Since pn≤∣Un∣≤cu,vAnp_n \le\lvert U_n\rvert\le c_{u,v}A^n, taking Y=⌊log⁡(x/cu,v)/log⁡A⌋Y=\left\lfloor\log(x/c_{u,v})/\log A\right\rfloor proves (1.4). Finally, when b=0b=0 the rank bound in Proposition 2.3 gives h∣pn−(Δ/pn)h\mid p_n-(\Delta/p_n), so pn≡±1(modh)p_n\equiv\pm1\pmod h. The equivalence of simplicity with the stated Lucas non-Wieferich condition follows from (2.4) because pnp_n is coprime to the index pn−(Δ/pn)p_n-(\Delta/p_n).

Remark 4.2. These statements concern the distribution of ranks and not a prime-density asymptotic. They neither assert that every sufficiently large index has a simple primitive divisor nor exclude repeated primitive divisors. The argument uses the unspecified constants of the rational abcabc conjecture and gives no explicit rate for its exceptional set.

Local multiplicity in recurrence-parameter space

We prove Theorem 1.3. In this section vv is held fixed. The Binet expressions for the Lucas polynomials give the derivative identities (the second is also [26 Lemma 2.1]

(X2−4v)∂Un∂X=nVn−XUn,∂Vn∂X=nUn.(24)(X^2-4v)\frac{\partial U_n}{\partial X}=nV_n-XU_n,\qquad\frac{\partial V_n}{\partial X}=nU_n. \tag*{(24)}

Indeed, if a,ba,b are the roots of Y2−XY+vY^2-XY+v over Q(X)\mathbb{Q}(X), then a′=a/(a−b)a'=a/(a-b) and b′=−b/(a−b)b'=-b/(a-b). Differentiating the Binet expressions gives (5.1). We also have Vp(X,v)≡Xp(modp)V_p(X,v)\equiv X^p\pmod p as polynomials by the Frobenius identity for the trace; see also [26 Lemma 3.1].

Proof of Theorem 1.3. The rank bound yields k∣p−(Δ/p)k\mid p-(\Delta/p) and p∤kp\nmid k. By (4.1), p∤Vk(u,v)p\nmid V_k(u,v) because p∣Uk(u,v)p\mid U_k(u,v) and p∤2vp\nmid2v. The first identity in (5.1) gives Uk′(u,v)≢0(modp)U'_k(u,v)\not\equiv0\pmod p. Hensel’s lemma supplies a unique u~\tilde{u} in u+pZpu+p\mathbb{Z}_p. On that disk the polynomial divided difference of Uk(X,v)U_k(X,v) at any two points is congruent to Uk′(u,v)U'_k(u,v) modulo pp, hence a unit. Taking one point to be u~\tilde{u} proves (1.6).

Euler’s criterion implies λp≡1(modp)\lambda_p\equiv1\pmod p; the normalization also gives λpp2=v\lambda_p^{p^2}=v. Consequently Tp(x)≡x(modp)T_p(x)\equiv x\pmod p for all x∈Zpx\in\mathbb{Z}_p. For xx in the disk, the derivative Tp′(x)=λppUp(x,v)T'_p(x)=\lambda_p^pU_p(x,v) has valuation one, by (2.3) applied at the residue of uu. To obtain the exact contraction, write y=x+hy=x+h, s=νp(h)≥1s=\nu_p(h)\ge1, and expand the polynomial Tp(x+h)T_p(x+h). Its coefficient of hh has valuation one. The coefficients of hih^i for 2≤i<p2\le i<p are in pZpp\mathbb{Z}_p, since Tp(X)≡Xp(modp)T_p(X)\equiv X^p\pmod p; the coefficient of hph^p is a unit. All these higher terms have valuation strictly greater than 1+s1+s (here p≥3p\ge3). The linear term therefore gives (1.7). If x=yx=y, both sides have the usual infinite-valuation interpretation.

Let cj=λp(pj−1)/(p−1)c_j=\lambda_p^{(p^j-1)/(p-1)}. The roots at parameter uj=Tpj(u)u_j=T_p^j(u) are cjαpjc_j\alpha^{p^j} and cjβpjc_j\beta^{p^j}. This follows by induction, as their product is cj2vpj=vc_j^2v^{p^j}=v and the next iteration scales their ppth powers by λp\lambda_p. It proves (1.8). The Binet expression gives

Uk(uj,v)=cjk−1Ukpj(u,v)Upj(u,v).U_k(u_j,v)=c_j^{k-1}\frac{U_{kp^j}(u,v)}{U_{p^j}(u,v)}.

Here cjc_j is a pp-adic unit and the denominator is a unit, since k>1k>1 and p∤kp\nmid k. By (2.4), the valuation of the numerator is r+jr+j. Applying (1.6) yields νp(uj−u~)=r+j\nu_p(u_j-\tilde{u})=r+j. Thus uj→u~u_j\to\tilde{u}, and continuity gives Tp(u^)=u^T_p(\hat{u})=\hat{u}. The two successive distances to u^\hat{u} have unequal valuations, so νp(uj+1−uj)=r+j\nu_p(u_{j+1}-u_j)=r+j. Finally, (7) rules out two distinct fixed points. This proves the theorem.

Corollary 5.1. Under the hypotheses of Theorem 1.3, for each a≥1a \ge1 there is exactly one residue class modulo pap^a in u+pZpu+p\mathbb{Z}_p on which pa∣Uk(x,v)p^a \mid U_k(x,v). Of its pp lifts modulo pa+1p^{a+1}, exactly p−1p-1 have valuation aa and one has higher valuation. In particular,

νp(Uk(u,v))=νp(λpVp(u,v)−u),u^≡Tp(u)(modpr+1).\nu_p(U_k(u,v))=\nu_p(\lambda_pV_p(u,v)-u), \qquad\hat{u}\equiv T_p(u)\pmod{p^{r+1}}.

Proof. The class is x≡u^(modpa)x\equiv\hat{u}\pmod{p^a} by (6); the remaining statements are immediate from (9).

To compare the valuation identity with [9 Theorem 1(v)], put w=(v/p)v(p−1)/2=λp−1w=(v/p)v^{(p-1)/2}=\lambda_p^{-1}. Since w≡1(modp)w\equiv1\pmod p, we have w2+1≡2w(modp2)w^2+1\equiv2w\pmod{p^2}. Thus λp−1≡(vp−1+1)/2(modp2)\lambda_p^{-1}\equiv(v^{p-1}+1)/2\pmod{p^2}, and λpVp(u,v)≡u(modp2)\lambda_pV_p(u,v)\equiv u\pmod{p^2} is precisely their trace congruence. The exceptional class at precision p2p^2 has relative frequency 1/p1/p as the parameter varies in the disk. This exact statement about a variable parameter does not imply equidistribution over primes for a fixed integer uu.

For example, take (u,v)=(3,2)(u,v)=(3,2) and p=7p=7. Here Un=2n−1U_n=2^n-1, Vn=2n+1V_n=2^n+1, k=3k=3, r=1r=1, and λ7=1/8\lambda_7=1/8. Hence T7(3)=129/8T_7(3)=129/8 and ν7(T7(3)−3)=ν7(105/8)=1\nu_7(T_7(3)-3)=\nu_7(105/8)=1, as predicted. The first iterate satisfies T7(3)≡10(mod49)T_7(3)\equiv10\pmod{49}, the unique lift of 33 modulo 77 solving U3(X,2)=X2−2≡0(mod49)U_3(X,2)=X^2-2\equiv0\pmod{49}.

The Dickson normalization in (5) allows any fixed nonzero norm parameter, rather than just v=−1v=-1. The iteration itself is a rescaled power map, with classical Dickson composition; our application records the rank exponent as its exact initial distance and step size. For v=−1v=-1, λp=1\lambda_p=1, so no rescaling is needed.

Fibonacci applications

Ranks and periods

Specialize to (u,v)=(1,−1)(u,v)=(1,-1), writing Fn=UnF_n=U_n, Ln=VnL_n=V_n, φ=(1+5)/2\varphi=(1+\sqrt{5})/2, and ψ=(1−5)/2=−φ−1\psi=(1-\sqrt{5})/2=-\varphi^{-1}. Write k(m)=kU(m)k(m)=k_U(m) for the rank of apparition modulo mm and ℓ(m)\ell(m) for the Pisano period modulo mm. Put rp=νp(Fk(p))r_p=\nu_p(F_{k(p)}) for each prime pp, and ϵp=(5/p)\epsilon_p=(5/p) for odd p≠5p\ne5. The quantities CnC_n, DnD_n, PnP_n, SnS_n, HnH_n in this section are their Fibonacci specializations. A prime p≠2,5p\ne2,5 is Wall–Sun–Sun when νp(Fp−ϵp)≥2\nu_p(F_{p-\epsilon_p})\ge2. Since r2=r5=1r_2=r_5=1 and rp=νp(Fp−ϵp)r_p=\nu_p(F_{p-\epsilon_p}) otherwise, the assertion that every first-occurring prime is simple is equivalent to the nonexistence of Wall–Sun–Sun primes. The problem goes back to Wall [24]; see also Sun and Sun [22]. No such prime is known; the customary heuristic instead predicts infinitely many [9, 5].

In R=Z[φ]R=\mathbb{Z}[\varphi] the identity

φn=Fnφ+Fn−1(n≥1)(25)\varphi^n=F_n\varphi+F_{n-1}\qquad(n\ge1) \tag*{(25)}

follows by induction. The ring RR is the ring of integers of Q(5)\mathbb{Q}(\sqrt{5}). The period ℓ(M)\ell(M) is exactly the order of φ\varphi in (R/MR)×(R/MR)^{\times}, by (25) and the independence of 1,φ1,\varphi as a Z\mathbb{Z}-basis. This interpretation uses ideal congruence and does not depend on whether a prime dividing MM splits, ramifies or remains inert in Q(5)\mathbb{Q}(\sqrt{5}).

Proposition 6.1. For M>1M > 1, ℓ(M)/k(M)\ell(M)/k(M) divides 44. If pp is an odd prime, then

ℓ(pa)=pmax⁡(0,a−rp)ℓ(p)(a≥1).(26)\ell(p^a)=p^{\max(0,a-r_p)}\ell(p) \qquad(a \ge1). \tag*{(26)}

Proof. Let k=k(M)k=k(M). Equation (25) implies φk≡Fk−1(modM)\varphi^k \equiv F_{k-1} \pmod{M}. Taking norms gives Fk−12≡(−1)k(modM)F_{k-1}^2 \equiv(-1)^k \pmod{M}. Thus Fk−1F_{k-1} is a unit of order dividing 44. The powers of φ\varphi that are rational scalars modulo MM occur exactly at multiples of kk, by strong divisibility. Hence ℓ(M)/k(M)\ell(M)/k(M) equals the order of Fk−1F_{k-1} modulo MM and divides 44.

The kernel of reduction (R/paR)×→(R/pR)×(R/p^aR)^{\times} \to(R/pR)^{\times} has order a power of pp (it is 1+pR/paR1+pR/p^aR). Therefore ℓ(pa)/ℓ(p)\ell(p^a)/\ell(p) is a power of pp. On the other hand k(pa)/k(p)k(p^a)/k(p) is a power of pp by (14), and the ratios of period to rank are powers of 22. Since pp is odd, these latter ratios are equal at pap^a and pp. This gives (26). □

Corollary 6.2. For an odd prime p≠5p \ne5, the following are equivalent: pp is Wall–Sun–Sun; rp≥2r_p \ge2; k(p2)=k(p)k(p^2)=k(p); ℓ(p2)=ℓ(p)\ell(p^2)=\ell(p).

Proof. Use k(p)∣p−εpk(p)\mid p-\varepsilon_p from (12), so p∤(p−εp)/k(p)p\nmid(p-\varepsilon_p)/k(p), and apply (13), (14) and (26). □

Theorem 6.3 (Odd-rank congruences). Let p≠2,5p\ne2,5 be prime with odd rank n=k(p)n=k(p). Then ℓ(p)=4n\ell(p)=4n and

p≡{1(mod4n),εp=1,2n−1(mod4n),εp=−1.(27)p \equiv \begin{cases} 1 \pmod{4n}, & \varepsilon_p=1,\\ 2n-1 \pmod{4n}, & \varepsilon_p=-1. \end{cases} \tag*{(27)}

In particular, p≡1(mod4)p\equiv1\pmod{4}.

Proof. As in Proposition 6.1, Fn−12≡−1(modp)F_{n-1}^2\equiv-1\pmod{p}; for odd pp its order is 44. Thus ℓ(p)=4n\ell(p)=4n. In the split case φp−1≡1(modp)\varphi^{p-1}\equiv1\pmod{p} in R/pR≅Fp×FpR/pR\cong\mathbb{F}_p\times\mathbb{F}_p, giving 4n∣p−14n\mid p-1. In the inert case Frobenius sends φ\varphi to ψ=−φ−1\psi=-\varphi^{-1}, so φp+1≡−1(modp)\varphi^{p+1}\equiv-1\pmod{p}. Since φ2n≡−1(modp)\varphi^{2n}\equiv-1\pmod{p}, we have 4n∣p+1−2n4n\mid p+1-2n. This proves the stated congruences. □

Corollary 6.4. If q>5q>5 is prime and pp is a prime divisor of FqF_q, then k(p)=qk(p)=q, ℓ(p)=4q\ell(p)=4q, and (27) holds with n=qn=q. If e=νp(Fq)e=\nu_p(F_q), then ℓ(pa)=4pqmax⁡(0,a−e)\ell(p^a)=4p_q^{\max(0,a-e)} for every a≥1a\ge1. In particular, p2∣Fqp^2\mid F_q implies that pp is a Wall–Sun–Sun prime.

Proof. The rank divides qq and exceeds 11. Also p≠2,5p\ne2,5 since their ranks are 3,53,5. Apply Theorem 6.3 and Corollary 6.2. □

Remark 6.5. This extends the prime-index classification of [13 Proposition 2] to all odd ranks. It does not imply that a hypothetical Wall–Sun–Sun prime has prime, or even odd, rank. In particular, squarefreeness of every prime-index Fibonacci number is only a necessary consequence of nonexistence of Wall–Sun–Sun primes.

Cyclotomic factors and splitting

For Fibonacci numbers, k(2)=3k(2)=3, r2=1r_2=1, and ν2(L3)=2\nu_2(L_3)=2. Thus (18) has the exceptional value 2 at n=6n=6 and is at most one elsewhere. The odd part of DnD_n is either 1 or the largest prime factor of nn, as in Corollary 3.2. We recall the squarefreeness characterization of Ross, Shen and Cai.

Corollary 6.6 ([19 Corollary 5.1]). The following assertions are equivalent:

  1. No Wall–Sun–Sun prime exists.

  2. PnP_n is squarefree for every n>1n>1.

  3. The odd part of CnC_n is squarefree for every n>1n>1.

Proof. The first two are equivalent by definition and r2=r5=1r_2=r_5=1. For an odd prime pp, (17) shows that p2∣Cnp^2\mid C_n if and only if k(p)=nk(p)=n and rp≥2r_p\ge2. This proves the third equivalence. By (18), the third condition is also equivalent to squarefreeness of CnC_n at every n≠6n\ne6. This is exactly the characterization of Ross, Shen and Cai [19 Corollary 5.1]. □

Splitting and parity of primitive multiplicities

The valuation formula above is a specialization of [1 Theorems 14–15] and [19 Theorem 2.2]. Ross, Shen and Cai [18 Corollaries 2.5–2.6] establish the parity of primitive multiplicities for general regular Lucas sequences and explicitly for Fibonacci numbers. In particular, their Corollary 2.6(b), (e) includes the inert prime-power existence result below. We provide a direct proof using the character modulo 5 and combine it with the odd-rank congruences to obtain the modulus 4n4n. Their Corollary 2.4(b) also includes the prime-power quotient congruence modulo pap^a. We next restrict to n>1n>1 with gcd⁡(n,30)=1\gcd(n,30)=1, so every prime dividing CnC_n is odd and different from 5. This restriction is convenient rather than essential. For odd dd one has

Fd≡d3d−1(mod5),(Fd5)=(d5).(28)F_d\equiv d3^{d-1}\pmod5,\qquad\left(\frac{F_d}{5}\right)=\left(\frac{d}{5}\right). \tag*{(28)}

Indeed, the polynomial T2−T−1T^2-T-1 has a double root 3 modulo 5. Its divided difference FdF_d reduces to d3d−1d3^{d-1}, and 3d−13^{d-1} is a square when dd is odd. (The first congruence holds also for even dd.)

Theorem 6.7 (Character of the primitive part). Let n>1n>1 and gcd⁡(n,30)=1\gcd(n,30)=1. Define

ηn={(p5),n=pa for a prime p, a≥1,1,n is not a prime power.\eta_n= \begin{cases} \left(\frac{p}{5}\right), & n=p^a\text{ for a prime }p,\ a\ge1,\\ 1, & n\text{ is not a prime power.} \end{cases}

Then

(Cn5)=ηn,(29)\left(\frac{C_n}{5}\right)=\eta_n, \tag*{(29)}
(Pn5)=ηn(Dn5),(30)\left(\frac{P_n}{5}\right)=\eta_n\left(\frac{D_n}{5}\right), \tag*{(30)}
Pn≡{1(mod4n),(Pn5)=1,2n−1(mod4n),(Pn5)=−1.(31)P_n\equiv \begin{cases} 1 \pmod{4n}, & \left(\frac{P_n}{5}\right)=1,\\ 2n-1 \pmod{4n}, & \left(\frac{P_n}{5}\right)=-1. \end{cases} \tag*{(31)}

In particular, the parity of the total multiplicity of inert primitive divisors of FnF_n is determined by nn and DnD_n.

Proof. By (3.1) and (6.4),

(Cn5)=∏d∣n(d5)μ(n/d)=(∏d∣ndμ(n/d)5).\left(\frac{C_n}{5}\right)=\prod_{d\mid n}\left(\frac{d}{5}\right)^{\mu(n/d)} =\left(\frac{\prod_{d\mid n}d^{\mu(n/d)}}{5}\right).

The integer inside the symbol in the final expression equals pp for n=pan=p^a and 11 otherwise, by taking the valuation at each prime and using the same Möbius transform as in the proof of Proposition 3.1. This proves (6.5), and the identity Cn=DnPnC_n=D_nP_n proves (6.6).

If k(p)=nk(p)=n, Theorem 6.3 implies that p≡1p\equiv1 or 2n−12n-1 modulo 4n4n according as pp splits or is inert. The latter residue is an involution since (2n−1)2≡1(mod4n)(2n-1)^2\equiv1\pmod{4n}. Consequently PnP_n is congruent to the latter residue precisely when the sum of the multiplicities of inert factors is odd. By quadratic reciprocity, (5/p)=(p/5)(5/p)=(p/5), so that parity is measured by (Pn/5)(P_n/5). This proves (6.7).

Corollary 6.8 (Prime-power indices). Let p>5p>5 be prime and a≥1a\geq1. Put Bp,a=Fpa/Fpa−1B_{p,a}=F_{p^a}/F_{p^{a-1}}. Then

Bp,a=Cpa=Ppa≡{1(mod4pa),(p/5)=1,2pa−1(mod4pa),(p/5)=−1.(32)B_{p,a}=C_{p^a}=P_{p^a}\equiv \begin{cases} 1 \pmod{4p^a}, & (p/5)=1,\\ 2p^a-1 \pmod{4p^a}, & (p/5)=-1. \end{cases} \tag*{(32)}

If p≡±2(mod5)p\equiv\pm2\pmod{5}, then FpaF_{p^a} has an inert primitive divisor qq of odd multiplicity. Such a qq satisfies q≡2pa−1(mod4pa)q\equiv2p^a-1\pmod{4p^a} and q≡13q\equiv13 or 17(mod20)17\pmod{20}.

Proof. For n=pan=p^a, (3.1) reduces to the quotient in the statement. If Dn≠1D_n\ne1, Corollary 3.2 would give n=pbk(p)n=p^{bk(p)} for b≥1b\geq1. But 1<k(p)∣p−ϵp1<k(p)\mid p-\epsilon_p and p∤k(p)p\nmid k(p), a contradiction. Thus Dn=1D_n=1 and (6.8) follows from Theorem 6.7. When (p/5)=−1(p/5)=-1, the total inert multiplicity is odd and at least one such prime has odd multiplicity. The congruences follow from Theorem 6.3 and quadratic reciprocity.

Remark 6.9. For a fixed inert prime p>5p>5, Corollary 6.8 gives infinitely many distinct inert primes, one of odd primitive multiplicity at each rank pap^a. Each such multiplicity is either one or at least three; the character argument cannot distinguish the two possibilities. For example, with p=7p=7, a=2a=2, one has

B7,2=598364773=97⋅6168709.B_{7,2}=598364773=97\cdot6168709.

The prime 97 is inert, has rank 49, and occurs exactly once. This is an illustration, not a general proof of simple primitive divisibility.

The local Wall exponent

For Fibonacci numbers, Tp(X)=Vp(X,−1)T_p(X)=V_p(X,-1) by (5), so Theorem 1.3 gives an exact description in terms of Lucas numbers.

Corollary 6.10. For every odd prime p≠5p \ne5,

νp(Lp−1)=rp=νp(Fp−ϵp).(33)\nu_p(L_p-1)=r_p=\nu_p(F_{p-\epsilon_p}). \tag*{(33)}

Moreover, Lpj→u^L_{p^j}\to\hat{u} in Zp\mathbb{Z}_p as j→∞j\to\infty, and νp(Lpj−u^)=rp+j\nu_p(L_{p^j}-\hat{u})=r_p+j for every j≥0j\ge0. In particular, pp is Wall–Sun–Sun if and only if Lp≡1(modp2)L_p\equiv1\pmod{p^2}.

Proof. Set (u,v)=(1,−1)(u,v)=(1,-1) in Theorem 1.3 and apply (13).

The congruence characterization at exponent two is classical [9 Theorem 1(v)]. The fixed point is also explicit. In the unramified local extension containing φ\varphi, let ζ\zeta be its Teichmüller representative. Then the representative of ψ=−φ−1\psi=-\varphi^{-1} is −ζ−1-\zeta^{-1} and

u^=ζ−ζ−1.(34)\hat{u}=\zeta-\zeta^{-1}. \tag*{(34)}

The sum belongs to Zp\mathbb{Z}_p (in the inert case Frobenius interchanges the two summands); their quotient has order k(p)k(p), and their difference is a unit as p≠5p\ne5. Thus Uk(p)(ζ−ζ−1,−1)=0U_{k(p)}(\zeta-\zeta^{-1},-1)=0, and Hensel uniqueness proves (34). In fact

1−u^=(φ−ζ)(1+(φζ)−1),1-\hat{u}=(\varphi-\zeta)(1+(\varphi\zeta)^{-1}),

where the second factor is a unit, as it reduces to 5/φ\sqrt{5}/\varphi modulo pp. This recovers the local Wall exponent of Bragman and Rowland [5 Theorem 10].

Proposition 6.11. Let k≥1k\ge1. If p>3p>3 is a prime and p∣Fkp\mid F_k, then

FkpFk≡p(modp2).\frac{F_{kp}}{F_k}\equiv p\pmod{p^2}.

Independently, for every odd prime p≠5p\ne5, the Lucas sequence Bm=Fpm/FpB_m=F_{pm}/F_p has discriminant coprime to pp, norm parameter v=−1v=-1, and first occurrence of pp at m=k(p)m=k(p) with multiplicity rp+1r_p+1.

Proof. The first quotient is the ppth term of the sequence Am=Fkm/FkA_m=F_{km}/F_k, whose parameters are u=Lku=L_k, v=(−1)kv=(-1)^k, and Δ=5Fk2\Delta=5F_k^2. Thus Proposition 2.5 applies. For the second assertion use the same construction with index pp. Now p∤Fpp\nmid F_p by Proposition 2.3, and p∤k(p)p\nmid k(p). Hence p∣Bmp\mid B_m if and only if k(p)∣mk(p)\mid m. Its initial valuation is νp(Fpk(p))=rp+1\nu_p(F_{pk(p)})=r_p+1 by (13).

This example shows that a general multiplicity-one theorem for Lucas sequences is false, even within the field Q(5)\mathbb{Q}(\sqrt{5}) and with norm parameter −1-1. Any proof for Fibonacci numbers must use additional information about their fixed recurrence parameter.

Further questions

The conditional density result allows repeated primitive divisors. For Fibonacci numbers it does not imply that Hn=1H_n=1 at every index, which would exclude all Wall–Sun–Sun primes. An unconditional result establishing Spa>1S_{p^a}>1 for infinitely many aa for a single fixed prime pp would already prove the infinitude of non-Wall–Sun–Sun primes with prescribed prime-power ranks. A related question, suggested by Corollary 6.8, is whether for some inert prime p>5p>5 infinitely many FpaF_{p^a} have an inert simple primitive divisor. The quadratic-character constraint alone only guarantees an inert divisor of odd multiplicity, not of multiplicity one.

Locally, the multiplicity-one conjecture for Fibonacci numbers is the assertion νp(1−u^)=1\nu_p(1-\widehat{u})=1 for every p≠2,5p\ne2,5. The exact contraction of Theorem 1.3 controls all further lifts but not that initial distance. The subsequence construction of Proposition 6.11 illustrates the obstruction to a parameter-independent multiplicity-one assertion. Neither the conditional global estimate nor the local reformulation resolves this open problem.

Declaration of generative AI use

The author used ChatGPT (OpenAI, GPT-6) to assist in the consolidation and presentation of previously developed mathematical results, the exploration of further mathematical consequences, literature searches, proof refinement, and manuscript preparation. The research builds on the author’s earlier independent work on Fibonacci and Lucas sequences. The author takes full responsibility for the originality, accuracy, and integrity of the contents.

MSC2020: 11B39, 11A07, 11A41, 11D75, 11S05

References

  1. [1]G. Alecci, P. Miska, N. Murru, and G. Romeo. On alternative definition of Lucas atoms and their p-adic valuations. Monatsh. Math., 207:175–196, 2025.arxiv.org/abs/2308.10216
  2. [2]K. Anitha, I. Mumtaj Fathima, and A. R. Vijayalakshmi. Lucas non-Wieferich primes in arithmetic progressions and the abc conjecture. Open Math., 21:20220563, 2023.arxiv.org/abs/2101.04901
  3. [3]C. J.-C. Ballot and H. C. Williams. The Lucas Sequences: Theory and Applications, volume 8 of CMS/CAIMS Books in Mathematics. Springer, Cham, 2023.
  4. [4]Yu Bilu, G. Hanrot, and P. M. Voutier. Existence of primitive divisors of Lucas and Lehmer numbers. J. Reine Angew. Math., 539:75–122, 2001.DOI
  5. [5]N. Bragman and E. Rowland. Limiting density of the Fibonacci sequence modulo powers of a prime. Res. Number Theory, 11:88, 2025.arxiv.org/abs/2202.00704
  6. [6]R. D. Carmichael. On the numerical factors of the arithmetic forms αⁿ ± βⁿ. Ann. of Math. (2), 15:30–70, 1913.
  7. [7]B. Diarra and D. Sylla. p-adic dynamical systems of Chebyshev polynomials. p-Adic Numbers Ultrametric Anal. Appl., 6(1):21–32, 2014.DOI
  8. [8]R. Lidl, G. L. Mullen, and G. Turnwald. Dickson Polynomials, volume 65 of Pitman Monographs and Surveys in Pure and Applied Mathematics. Longman, 1993.
  9. [9]R. J. McIntosh and E. L. Roettger. A search for Fibonacci-Wieferich and Wolstenholme primes. Math. Comp., 76:2087–2094, 2007.DOI
  10. [10]S. Müller. On the rank of appearance and the number of zeros of the Lucas sequences over F_q. In D. Jungnickel and H. Niederreiter, editors, Finite Fields and Applications, pages 390–408. Springer, 2001.DOI
  11. [11]W. Peng. ABC implies there are infinitely many non-Fibonacci-Wieferich primes. J. Number Theory, 212:354–375, 2020.
  12. [12]S. Riasat. Z/[φ] and the Fibonacci sequence modulo n. Mathematical Reflections, pages 1–7, 2011. Issue 1. Author’s online version.
  13. [13]S. Riasat. On Fibonacci numbers with prime index. Note, 2012. Author’s online version.
  14. [14]S. Riasat. A generalisation of ‘lifting the exponent’. Note, 2013. Author’s online version.
  15. [15]S. Riasat. Generalising ‘LTE’ and application to Fibonacci-type sequences. Note, 2014. Author’s online version.
  16. [16]P. Ribenboim. On square factors of terms of binary recurring sequences and the ABC conjecture. Publ. Math. Debrecen, 59:459–469, 2001.DOI
  17. [17]P. Ribenboim and G. Walsh. The ABC conjecture and the powerful part of terms in binary recurring sequences. J. Number Theory, 74:134–147, 1999.DOI
  18. [18]T. Ross, Z. Shen, and T. Cai. Cyclotomic congruences and Lucas sequences. Preprint, 2026. arXiv:2512.03468v2 (8 January 2026).arxiv.org/abs/2512.03468
  19. [19]T. Ross, Z. Shen, and T. Cai. The p-adic valuations of Möbius duals of Lucas sequences. Fibonacci Quart., pages 1–10, 2026.DOI
  20. [20]B. E. Sagan and J. Tirrell. Lucas atoms. Adv. Math., 374:107387, 2020.
  21. [21]C. Sanna. The p-adic valuation of Lucas sequences. Fibonacci Quart., 54:118–124, 2016.
  22. [22]Z. H. Sun and Z. W. Sun. Fibonacci numbers and Fermat’s last theorem. Acta Arith., 60:371–388, 1992.DOI
  23. [23]P. M. Voutier. Primitive divisors of Lucas and Lehmer sequences, II. J. Théorie des Nombres Bordeaux, 8:251–274, 1996. Online version.DOI
  24. [24]D. D. Wall. Fibonacci series modulo m. Amer. Math. Monthly, 67:525–532, 1960.
  25. [25]M. Yabuta. The ABC-conjecture and the powerful numbers in Lucas sequences. Fibonacci Quart., 45:362–365, 2007.DOI
  26. [26]M. Yamagishi. A short proof of congruences for Lucas sequences. Fibonacci Quart., 57:260–264, 2019.DOI

Paper details

vlnc:W_9zks642m
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT

View statement
Looking for related papers…
Contents