For the Pell numbers one has Pn+1−Pn−1=2Pn and Pn+12−Pn−12=2P2n. We look for all the solutions of (Pn+1(k))x−(Pn−1(k))x=2Pm(k) in positive integers k≥2, n, m, x, where P(k) is the k-generalized Pell sequence. We prove that for k=2 the two identities above give all the solutions, and that for k≥3 the only solutions are (n,m,x)=(1,1,1),(1,2,2),(2,2,1). In contrast with the analogous equation for k-generalized Fibonacci numbers, there are no parametric families of solutions. The proof uses an auxiliary equation with Fibonacci numbers which does not depend on k, Matveev's lower bound for linear forms in logarithms, the reduction of Dujella and Pethő, and, for large k, the fact that the dominant root of P(k) is very close to φ2.
Introduction
The Fibonacci numbers satisfy Fn+12−Fn−12=F2n. Patel and Chaves [12] asked when Fn+1x−Fn−1x is again a Fibonacci number and solved the equation Fn+1x−Fn−1x=Fm completely. Apart from trivial cases, the only solutions come from x=2, m=2n. Gómez, Luca and Gómez [9] treated Fn+1x−Fn−1x=Fmy. The same question for the k-generalized Fibonacci numbers was studied by Bensella, Patel and Behloul [2], and solved completely by Gómez, Gómez and Luca [8]; in that case there are two infinite parametric families of solutions, with n∈{4,5} and k growing exponentially with the parameter. These families come from the fact that the k-generalized Fibonacci numbers Fr(k) are powers of 2 for 2≤r≤k+1.
In this paper, we are interested in the Pell analogue because of the two identities
where Pn is the n-th Pell number (the second one follows from the first and from Pn+1+Pn−1=Qn, PnQn=P2n, with Qn the Pell–Lucas numbers). So, for Pell numbers, the natural question is when Pn+1x−Pn−1x is twice a Pell number. Let k≥2 and let P(k)=(Pn(k))n be the k-generalized Pell sequence. We study the equation
(Pn+1(k))x−(Pn−1(k))x=2Pm(k)(2)
in positive integers k≥2, n, m and x. In a companion paper [7] we treat, in the same spirit, (Pn(k))x+(Pn+1(k))x=Pm(k), which generalizes the Pell identity Pn2+Pn+12=P2n+1 to arbitrary indices.
In a first version of this work we tried to follow the method of [8] step by step, and we obtained two parametric families similar to the families for k-Fibonacci numbers. A numerical check showed that they are not solutions. For example, the first member (k,n,m,x)=(7,4,11,3) gives
343−53=39179while2P11(7)=21854.
The reason is that the property used in [8] has no analogue here. By a result of Kılıç [10], the first k+1 non-zero terms of P(k) are not powers of 2 but their are equal to the Fibonacci numbers with odd index, Pr(k)=F2r−1 for 1≤r≤k+1. Hence, when n≤k and m≤k+1, (2) becomes an equation which does not depend on k, namely
F2n+1−F2n−3=2F2m−1
and this equation has no solution (see Theorem 1.2). This is the main reason why the final answer is so short. In fact, in the range where the k-Fibonacci numbers produce whole families of solutions, the k-Pell numbers produce none.
Theorem 1.1. Let k≥2, n≥1, m≥1 and x≥1 be integers satisfying (2).
(i) If k=2, then (m,x)=(n,1) or (m,x)=(2n,2); conversely, these are solutions for every n≥1.
(ii) If k≥3, then (n,m,x)∈{(1,1,1),(1,2,2),(2,2,1)}, and these three triples are solutions for every k≥3.
Theorem 1.2. The equation
F2n+1−F2n−3=2F2m−1(3)
has no solution in integers n≥2, m≥1 and x≥2.
We use the following strategy for the proof. The elementary cases (x=1, n=1, and k=2 with x=2) are settled in § 3, and Theorem 1.2 is proved in § 4, with three linear forms in logarithms in Q(5). For an arbitrary k we use the same three forms, now in Q(α), where α is the dominant root of P(k). This bounds x in terms of k (Section 5). This gives first k<1.05×1045, and six rounds of reduction bring this down to k≤257 (§ 6). The remaining values 2≤k≤257 are treated one by one in Section 7.
Preliminaries
The generalized Pell sequence
Let k≥2 be an integer. The k-generalized Pell sequence P(k):=(Pn(k))n≥2−k is defined by
Pn(k)=2Pn−1(k)+Pn−2(k)+⋯+Pn−k(k)for all n≥2,
with the initial conditions P−(k−2)(k)=P−(k−3)(k)=⋯=P−1(k)=P0(k)=0 and P1(k)=1. For k=2 we obtain the Pell sequence 0,1,2,5,12,29,70,…. The terms of this sequence are called k-generalized Pell numbers. When k is fixed we simply write Pn:=Pn(k).
Kılıç [10] showed that the first k+1 non-zero terms of P(k) are Fibonacci numbers with odd index, namely
Pn(k)=F2n−1for all 1≤n≤k+1.(4)
The characteristic polynomial of P(k),
Ψk(x):=xk−2xk−1−xk−2−⋯−x−1,
is irreducible over Q and has just one root outside the unit circle [4]. Let α:=α(k) denote that root. It was proved in [4] that
φ2(1−φ−k)<α<φ2,(5)
where φ=(1+5)/2 is the golden section. We write α1:=α,α2,…,αk for the roots of Ψk; thus ∣αi∣<1 for 2≤i≤k. A direct computation gives
For x0=1+2 we have x02−3x0+1=1−2, so (6) gives (x0−1)Ψk(x0)=1−(2−1)x0k−1=1−x0k−2≤0. Since Ψk has exactly one positive root, we get α≥1+2, with equality for k=2.
For large k the dominant root α is very close to φ2, and Pn(k) remains very close to F2n−1 (in relative terms) also for n>k+1, as long as n is small compared with φ2k. We put
c:=φ2+11=5φ1=105−5=0.2763…
For instance F2n−1=cφ2n+φ−2n+1/5, since cφ=1/5.
Lemma 2.4. Let k≥30 and write θ:=φ2−α, δ:=θ/φ2 and g=c(1+η). Then
Notations and terminology from algebraic number theory
Let η be an algebraic number of degree d with minimal primitive polynomial over the integers
a0xd+a1xd−1+⋯+ad=a0i=1∏d(x−η(i)),
where the leading coefficient a0 is positive and the η(i)'s are the conjugates of η. Then the logarithmic height of η is given by
h(η):=d1(loga0+i=1∑dlog(max{∣η(i)∣,1})).
In particular, if η=p/q is a rational number with gcd(p,q)=1 and q>0, then h(η)=logmax{∣p∣,q}. The following properties of h(⋅) will be used without reference:
Since Ψk is monic, irreducible, and has exactly one root of modulus greater than 1, we have h(α)=(logα)/k. For g we use the bound of Bravo, Herrera and Luca [5 Lemmas 3 and 6]:
h(g)<H(k):=4logφ+log(k+1)(k≥2).(10)
In Q(5) we shall use h(2)=log2, h(5)=21log5, h(φ)=21logφ, h(2/5)=21log5 (minimal polynomial 5y2−4), and h(2c)=21log(5(1+1/5))=0.98953…, since 2c=1−1/5 has minimal polynomial 5y2−10y+4.
Linear forms in logarithms and the reduction lemma
Theorem 2.5 (Matveev). Let γ1,…,γt be positive real algebraic numbers in a real algebraic number field K of degree D, b1,…,bt be nonzero integers, and assume that
Λ:=γ1b1⋯γtbt−1
is nonzero. Then
log∣Λ∣>−1.4×30t+3×t4.5×D2(1+logD)(1+logB)A1⋯At,
where B≥max{∣b1∣,…,∣bt∣} and Ai≥max{Dh(γi),∣logγi∣,0.16} for all i=1,…,t.
We write C(D,t):=1.4×30t+3×t4.5×D2(1+logD).
For a real number r we denote by ∥r∥ the distance from r to the nearest integer. The following is the reduction lemma of Dujella and Pethő [6 Lemma 5(a)]; we state it with a real exponent w, the proof being the same.
Lemma 2.6 (Dujella and Pethő). Let M be a positive integer, let τ be an irrational number and let p/q be a convergent of the continued fraction of τ with q>6M. Let A, B, μ be real numbers with A>0 and B>1, and put ε:=∥μq∥−M∥τq∥. If ε>0, then there is no solution of
0<∣uτ−v+μ∣<AB−w
in integers u, v and a real number w with 1≤u≤M and w≥log(Aq/ε)/logB.
In practice we take the first convergent with q>6M, and the next ones if ε≤0; we never needed more than one extra convergent. We shall also use, without further mention, that if ∣Λ∣<1/2 and Λ=eΓ−1 with Γ real, then ∣Γ∣<2∣Λ∣. When ∣Λ∣≥1/2, the upper bound for ∣Λ∣ that we use already gives a bound for the exponent w which is smaller than the bounds obtained with Lemma 2.6.
Elementary cases
Let (k,n,m,x) be a solution of (2) in positive integers with k≥2.
Lemma 3.1 (The case x=1). If x=1, then m=n when k=2, and (n,m)∈{(1,1),(2,2)} when k≥3.
Proof. For k=2, Pn+1−Pn−1=2Pn by (1), so Pm=Pn and m=n. Let k≥3. For n=1, P2−P0=2=2P1, and for n=2, P3−P1=4=2P2. Let n≥3. Then
since k≥3 and Pn−2≥P1=1, and Pn+1−Pn−1<2Pn+1. Since (Pr)r≥1 is strictly increasing, 2Pm cannot lie strictly between 2Pn and 2Pn+1.
Lemma 3.2 (The case n=1). If n=1, then (m,x)∈{(1,1),(2,2)}.
Proof. Since P2=2 and P0=0, (2) becomes Pm=2x−1. For x=1 we get m=1; for x≥2, Lemma 2.3 gives m=2 and x=2.
Lemma 3.3 (The case k=2, x=2). If k=2 and x=2, then m=2n.
Proof. By (1), Pn+12−Pn−12=2P2n, so Pm=P2n and m=2n.
The triples listed in Theorem 1.1 are the solutions. This follows from (1) for k=2, and from P2−P0=2P1, P22−P02=2P2 and P3−P1=2P2 for k≥3. Therefore it remains to show that (2) has no solution with
n≥2,x≥2,(k,x)=(2,2).(11)
Lemma 3.4 (Size of m). If n≥2 and x≥2, then (n−1)x+1≤m≤nx+1. In particular m>x, and the integer
j:=(n+1)x−m
satisfies x−1≤j≤2x−1.
Proof. By Lemma 2.1(a), 2αm−2≤2Pm<Pn+1x≤αnx, so m−2<nx and m≤nx+1. By (8), Pn−1x≤5−xPn+1x, hence 2αm−1≥2Pm≥(1−5−x)Pn+1x≥2524α(n−1)x. Thus
m−1≥(n−1)x−logαlog(25/12)>(n−1)x−1
because log(25/12)<log(1+2). Hence m≥(n−1)x+1.
Proof of Theorem 1.2
We begin with the case in which all the indices are at most k+1, which reduces to (3). Everything takes place in Q(5), the heights are small, and the argument shows on a simple example the steps that we shall repeat later for an arbitrary k.
In this section (n,m,x) is a solution of (3) with n≥2 and x≥2. We use Fr=(φr−(−φ)−r)/5 and φr−2≤Fr≤φr−1 for r≥1.
Since Fr+4=3Fr+1+2Fr≥5Fr, we have F2n−3≤F2n+1/5. Exactly as in Lemma 3.4 (with φ2 in place of α) we get (n−1)x+1≤m≤nx+1. We put
e:=2m−1−(2n+1)x,
so that −3x+1≤e≤−x+1.
Write a:=φ−4n−2 and b:=φ−4n+6, so that F2n+1=5φ2n+1(1+a) and F2n−3=5φ2n−3(1+b), and 2F2m−1=52(φ2m−1+φ−2m+1). We consider
Φ1:=52φ2m−1F2n+1−x−1,
Φ2:=2⋅5x−1φe−1,
Φ3:=2⋅5x−1φe(1−φ−4x)−1−1.
Dividing (3) by F2n+1x, we get Φ1=−(F2n−3/F2n+1)x−52φ−2m+1F2n+1−x, hence
Then (1+a)x−1≤1.026xa, (1+b)x≤e0.05, and, using m≥(n−1)x+1, r≤0.014xa. This gives
∣Φ2∣<2.7max{xφ−4n−2,φ−4x}.(14)
In the same way, the numerator of Φ3 is [(1+a)x−1]−φ−4x[(1+b)x−1]−r. Note that φ−4xb≤a for x≥2, and the denominator is 1−φ−4x≥1−φ−8; hence, under (12a),
∣Φ3∣<2.66xφ−4n−2.(15)
The norms from Q(5) to Q satisfy ∣N(φ)∣=1, ∣N(5)∣=5, ∣N(2/5)∣=4/5. If Φ1=0, then ∣N(2φ2m−1/5)∣=4/5 would be equal to F2n+12∈Z, which is impossible. If Φ2=0, then 1=∣N(25x−1φe)∣=4⋅5x−1, which is impossible. If Φ3=0, then
that is, F2x2=4⋅5x−2. This is false for x=2 (F4=3); for x≥3 it would give 5∣F2x, hence 5∣x, and then 2v5(x)=2v5(F2x)=x−2, which is impossible since x−2>2logx/log5 for x≥5. Hence Φ1Φ2Φ3=0.
to Φ2, with t=3, (γ1,γ2,γ3)=(2,5,φ), (b1,b2,b3)=(1,x−1,e), B=3x, and (A1,A2,A3)=(2log2,log5,logφ). We get log∣Φ2∣>−K2(1+log3x), with
K2<1.0413×1012
to Φ3, with t=4, the same data and γ4=1−φ−4x, b4=−1, A4=log(φ4x−1)<4xlogφ where the conjugate of γ4 is 1−φ4x. More,
log∣Φ3∣>−4xlogφ⋅K3(1+log3x)
K3:=C(2,4)<1.140×1014
to Φ1, with t=3, (γ1,γ2,γ3)=(2/5,φ,F2n+1), (b1,b2,b3)=(1,2m−1,−x), B=2m, (A1,A2,A3)=(log5,logφ,2logF2n+1) with 2logF2n+1≤4nlogφ where
log∣Φ1∣>−4nlogφ⋅K1(1+log2m)
K1:=C(2,3)log5logφ<7.511×1011.
Suppose first that (13) holds and that φ−4x≥xφ−4n−2. Then (14) and the bound for Φ2 give 4xlogφ<log2.7+K2(1+log3x), hence x<X0:=1.7634×1013; then (15) and the bound for Φ3 give n<6.553×1028. Otherwise, either (13) fails, and then 4n<6+log(20x)/logφ, or xφ−4n−2>φ−4x, and then (4n+2)logφ<log(2.7x)+K2(1+log3x). In both cases n is bounded by an explicit function of x; inserting this and m≤nx+1 in (12) and the bound for Φ1, we get x<3.0554×1027. Therefore, in all cases,
x<3.06×1027(16)
For the reduction phase, we put Γ2:=log2+(x−1)log5+elogφ, so that Φ2=eΓ2−1. With τ:=log5/logφ and μ:=log2/logφ, (14) gives (if ∣Φ2∣<1/2)
Lemma 2.6 with u=x−1≤M gives w<W for an explicit W. If w=4xlogφ, then x<W/(4logφ), and for each such x and each e∈[−3x+1,−x+1] the number Φ3 is explicit and non-zero, so that (15) bounds n. If w is the other term, or if (13) fails, n is bounded directly. Once n≤N, we apply Lemma 2.6 to Γ1/logF2n+1, where
Γ1=log(2/5)+(2m−1)logφ−xlogF2n+1
with u=2m−1≤2(NM+1), A=2.8/logF2n+1, B=5, separately for each 2≤n≤N; this bounds x. Table 1 shows the three passes we made.
pass
M
W
x if w=4xlogφ
n from Φ3
N
x≤
1
3.06×1027
69.18
35
3.8
68
46
2
47
11.00
5
2.3
7
7
3
8
7.42
3
1.8
4
7
Table 1. Reduction for equation (3). Here M is the bound for x used in the pass, W comes from Φ2, N is the resulting bound for n and the last column is the bound for x given by Φ1.
As usual, almost all the work is done by the first pass, which brings the bound for x from 3×1027 down to 46. Thus n≤4 and x≤7. A direct search with 2≤n≤60 and 2≤x≤60 shows that (3) has no solution. This proves Theorem 1.2.
Linear forms in Q(α) and a bound for x
The forms below are those of Section 4, with g and α in the place of 1/5 and φ2. We then obtained that we work is now in a field of degree k and that the height of g grows like logk, so that the bound for x will depend on k.
In this section k≥2 and (k,n,m,x) is a solution of (2) satisfying (11). We use the notation of Lemma 2.1, and j=(n+1)x−m is as in Lemma 3.4. We consider
(b) If (15a) holds, then0<∣Λ2∣<2.6max{α−2x,xα−n}and∣Λ3∣<1.5xα−n.
Proof. (a) Since 2gαm=2Pm−2em, dividing (2) by Pn+1x gives
Λ1=−(Pn−1/Pn+1)x−2emPn+1−x.
By (8) and Pn+1≥5 we get ∣Λ1∣<2⋅5−x. If Λ1=0, then 2gαm=Pn+1x. Applying an automorphism of the Galois closure of Q(α) sending α to α2 and using Lemma 2.2, we get
Pn+1x=2∣gk(α2)∣∣α2∣m<2,
a contradiction.
(b) Put U:=gαn+1 and V:=gαn−1, and write Pn+1=U(1+ε1), Pn−1=V(1+ε2). By Lemma 2.1(c),(d), ∣ε1∣<1.812α−n−1 and ∣ε2∣<1.812α−n+1, so that, by (15a), x∣ε2∣≤0.05; moreover (15a) implies αn−1≥72.8, hence n≥6 and V>20. Dividing (2) by Ux and using V/U=α−2,
Λ2=[(1+ε1)x−1]−α−2x(1+ε2)x−2emU−x.
Now
∣(1+ε1)x−1∣≤ex∣ε1∣−1≤1.026x∣ε1∣<0.78xα−n
as α≥1+2, (1+∣ε2∣)x<1.052, and U−x=α−2xV−x<0.0025α−2x. This gives the bound for Λ2. In the same way the numerator of Λ3 is
[(1+ε1)x−1]−α−2x[(1+ε2)x−1]−2emU−x,
of absolute value less than 1.026x(∣ε1∣+α−2x∣ε2∣)+U−x<0.91xα−n, and its denominator is 1−α−2x≥1−α−4>0.97; hence ∣Λ3∣<1.5xα−n. Finally, if Λ2=0, then gx−1αj=2; conjugating as in (a) we get 2=∣gk(α2)∣x−1∣α2∣j<1, since j≥x−1≥1, a contradiction. □
Lemma 5.2. (a) Fork=2, x≥2andx−1≤j≤2x−1, we haveΛ3=0if and only if(x,j)=(2,2).
(b) For2≤k≤257, 2≤x≤64, x−1≤j≤2x−1and(k,x,j)=(2,2,2), we have∣Λ3∣>1.22×10−3.
Proof. (a) For k=2 we have α=1+2 and g=1/(22). For (x,j)=(2,2),
2g−1α−2=42α−2=1−α−4,
because α2−α−2=(α−α−1)(α+α−1)=2⋅22. This correspond to the identity
Pn+12−Pn−12=2P2n
seen in Q(2). Conversely, if Λ3=0, taking norms from Q(2) to Q gives 23x−1=∣N(1−α−2x)∣=Q2x−2, where Qr=αr+(−α)−r are the Pell–Lucas numbers. For x even, Q2x−2=8Px2, so Px is a power of 2. Thus x=2 by Lemma 2.3, and then j=2. For x odd, Q2x−2=Qx2, and Qx is twice an odd number larger than 1 for x≥3, which is impossible.
(b) This is a direct computation (Appendix A). The minimum is attained for k=4, x=62, j=80, where ∣Λ3∣>1.22×10−3. □
Part (a) explains the only zero we have got. Since (k,x)=(2,2) is excluded by (11), it causes no problem. Part (b) covers all the pairs (x,j) for which Λ3 is used in Section 7, since there x≤64.
Now we can apply Matveev for Λ1 and Λ2. Let K=Q(α), D=k. For Λ1 we take t=3, (γ1,γ2,γ3)=(2g,α,Pn+1), (b1,b2,b3)=(1,m,−x) and B=m (since m>x). By (10) we may take A1=k(log2+H(k)), A2=logα and A3=klogPn+1≤knlogα. For Λ2 we take (γ1,γ2,γ3)=(2,g,α), (b1,b2,b3)=(1,1−x,−j), B=2x, A1=k(log2+H(k)), A2=kH(k) and A3=logα. Using logα<2logφ, we get
Then every solution of (2) satisfying (11) has x<X(k).
Proof. If (15a) holds and the minimum in (17) is 2xlogα, then x satisfies the first inequality. Otherwise, either (15a) fails or the minimum is nlogα−logx; in both cases n<Ns(x). Then m≤nx+1 by Lemma 3.4, and (16) gives the second inequality. □
Large values of k
The idea of this section is to exploit the error made by replacing α by φ2. This is of size about kφ−2k, which is negligible as soon as k is large, and the forms we obtain no longer depend on k.
In this section k≥30. Lemma 2.4 allows us to replace g by c=1/(5φ) and α by φ2 in Λ1,Λ2,Λ3. The new forms live in Q(5), and their coefficients do not depend on k. Since c1−x=(5φ)x−1, the forms are
L1:=2cφ2mF2n+1−x−1,
L2:=25x−1φx−1−2j−1,
L3:=25x−1φx−1−2j(1−φ−4x)−1−1.
Note that L2 and L3 are the forms Φ2 and Φ3 of § 4, with e=x−1−2j. In particular they are non-zero, and Matveev gives the same lower bounds as for Φ2,Φ3. Note that here ∣x−1−2j∣≤3x. The form L1 is non-zero because ∣N(2cφ2m)∣=4/5 is not an integer. Put
Ek(x):=1.011x(0.45k+2.02)φ−2k.
Lemma 6.1. Let k≥30, and let (k,n,m,x) be a solution of (2) which satisfies (11) and 2.022Ek(x)≤0.01. Put λ∗:=−log(2.022Ek(x)).
(a) If (15a) holds, then ∣L2∣<e−λ2 and ∣L3∣<e−λ3, where
λ2:=min{2xlogα−log5.26,nlogα−log(5.26x),λ∗},
λ3:=min{nlogα−log(3.04x),λ∗}.
:::`
(b) If n≤k and ∣ζm∣≤0.01, then ∣L1∣<e−λ1, where
and Lemma 5.1(b) gives the claim. For L3 the argument is the same, with the additional factor (1−φ−4x)/(1−α−2x), which differs from 1 by less than 4xδφ−8 and is absorbed in Ek(x).
(b) If n≤k, then Pn+1=F2n+1 by (4), and by (9), Λ1+1=(L1+1)(1+ζm). Hence
For L1 we apply Theorem 2.5 in Q(5) with (γ1,γ2,γ3)=(2c,φ,F2n+1), (b1,b2,b3)=(1,2m,−x), B=2m, (A1,A2,A3)=(log(5+5),logφ,2logF2n+1) and we get
log∣L1∣>−4nlogφ⋅K1′(1+log2m)
with K1′:=C(2,3)log(5+5)logφ<9.236×1011.
Lemma 6.2. For every k≥258 we have 2.022Ek(X(k))<10−55.
Proof. Let k≥258 and λ:=logk. We have C(k,3)k2=ck4(1+λ) with c:=1.4⋅306⋅34.5, and 2clogφ<1.379⋅1011. Moreover H(k)≤λ+1.93, log2+H(k)≤λ+2.627 and (1+λ)(λ+2.627)(λ+1.93)≤k1.1 (at k=258 the left side is <401.2, the right side is >449, and k1.1 divided by the left side increases with k). Since 2logφ<H(k), we get
E1(k)<E2(k)≤M:=1.379⋅1011k5.1,M≥2.7⋅1023.
Put N(x):=1+(log(94.64x)+M(1+log2x))/log(1+2), and let x≥2 satisfy one of the two inequalities of Proposition 5.3. Since 2log(1+2)>log5, Ns(x)≤N(x) and E2(k)(1+log2x)≤N(x), in both cases
xlog5<log2.6+MN(x)(1+log(N(x)x+1)).
Now let x≥M3 and u:=logx, so that u≥160 and logM≤u/3. Then N(x)≤1.135M(u+4.7) and, since log(u+4.7)≤0.035u, 1+log(N(x)x+1)≤1.377u. Hence the right-hand side of the preceding inequality is at most 1.62M2u2≤1.62u2e2u/3<eulog5=xlog5, so the inequality fails. Therefore
Since 15.3/k+0.45/(0.45k+2.02)<2logφ, the function G is decreasing for k≥258, and G(309)<10−55. This proves the claim for k≥309. For 258≤k≤308 we computed X(k) directly. The largest value of log10(2.022Ek(X(k))) is −55.40, attained at k=258.
Proposition 6.3.(2) has no solution satisfying (11) with k≥Kabs:=1.0427×1045.
Proof. Let k≥Kabs and x<X:=X(k); at k=Kabs we have X<2.99×10398. By Lemma 6.2, Lemma 6.1 applies. Comparing with Matveev’s bound for L2, the third term λ∗ cannot be the minimum in λ2, because
Hence, if (15a) holds, either 2xlogα−log5.26<K2(1+log3x), which gives x<X0′=1.7634×1013 as in Section 4, or
n<Nb:=(K2(1+log3X)+log(5.26X))/logα>9.949×1014
if (15a) fails, n<1+log(36.4X)/logα<Nb.
Case x<X0′. Then Matveev’s bound for L3 gives, again because λ∗ cannot be the minimum, n<Na:=1.32×1029. Hence n≤k and m≤nx+1<NaX0′+1<k+1, so that all the terms in (2) are odd-indexed Fibonacci numbers by (4), and (2) becomes (3), which has no solution by Theorem 1.2.
Case n<Nb. Then n≤k and m≤NbX+1, and ∣ζm∣ is extremely small. Matveev’s bound for L1 and Lemma 6.1(b) give x<XG=1.048×1030. Then m≤NbXG+1≤Kabs≤k. This is the condition which determines Kabs. We conclude with Theorem 1.2 as before.
All the inequalities used are of the form k>F(logk), where F is a polynomial with positive coefficients. Since they hold at k=Kabs, they hold for all k≥Kabs. We also checked them numerically for k=Kabs⋅10t, t=0,5,10,…,180.
We now lower Kabs by reduction. Let K be an upper bound for k, with K≥258, and consider 258≤k≤K; then x<X(K) and, by Lemma 6.2, Lemma 6.1 applies. One round of reduction consists of the following steps, with logα≥2logφ+log(1−φ−30) and x≤M:
(R1) Lemma 2.6 applied to L2, with τ=log5/logφ, μ=log2/logφ, u=x−1≤M, A=2/logφ, B=e, gives λ2<W. Hence either x≤Xa:=(W+log5.26)/(2logα), or n<(W+log(5.26M))/logα, or
k<Kc:=2logφW+log(2.022⋅1.011M(0.45K+2.02))
or (15a) fails.
(R2) For each 2≤x≤Xa and x−1≤j≤2x−1 the number L3 is explicit and non-zero; with ℓ:=−logminj∣L3∣, Lemma 6.1(a) gives either n<(ℓ+log(3.04x))/logα, or
2klogφ<ℓ+log(2.022⋅1.011x(0.45K+2.02)).
(R3) Steps (R1)–(R2) give n≤N unless k is smaller than an explicit bound. For each 2≤n≤N, Lemma 2.6 applied to L1, with τ=2logφ/logF2n+1, μ=log(2c)/logF2n+1, u=m≤NM+1, A=2/logF2n+1, B=e, gives x≤XG unless k is smaller than an explicit bound.
So, if k is larger than a bound on K1, let us say the maximum of the bounds met in (R1)–(R3) and of N, we get n≤N and x≤max{Xa,XG}. We repeat (R1)–(R3) with this new bound for x until it stabilizes, and we get final bounds n≤N′, x≤X′, valid for k>K2. Finally, if k≥N′X′+1, then n≤k and m≤N′X′+1≤k+1, so that (2) becomes (3), which has no solution. Thus a round gives the new bound k≤max{K1,K2,N′X′+1}.
Table 2. Rounds of reduction for large k. Here K is the bound for k at the start of the round. For the first pass of each round we give ⌊log10X(K)⌋, the value W of (R1), Xa, the bound N for n, and the bound XG of (R3). The last pass of every round gives W=6.40, N′=10, X′=6, hence N′X′+1=61.
round
K
log10X(K)
W
Xa
N
XG
new K
1
1.0427×1045
398
923.69
480
1914
580
2021
2
2022
57
138.32
72
283
92
297
3
298
50
122.13
64
249
80
261
4
262
50
120.60
63
246
79
258
5
259
50
120.56
63
246
79
257
6
258
50
120.55
63
246
79
257
Table 2.
Proposition 6.4.Equation(2)has no solution satisfying(11)withk≥258.
Proof. By Proposition 6.3 we may assume k<Kabs. Table 2 shows the rounds. Each round starts from the bound given by the previous one plus one, and the last round gives again k≤257.
The cases 2≤k≤257
Now, it remains is a finite, but very long, computation. We have 256 values of k, each with its own α, its own g and a bound X(k) up to 1050. The steps are those of § 4, in Q(α) instead of Q(5).
Let 2≤k≤257. We start with x<X(k)≤1.08×1050 (Proposition 5.3) and make three passes of the following three steps, with x≤M.
(S1) With Γ2=log2+(1−x)logg−jlogα, Lemma 5.1(b) gives, under (15a),
0<∣xτ−j+μ∣<logα5.2e−w,w=min{2xlogα,nlogα−logx},
with τ=−logg/logα and μ=log(2g)/logα. Lemma 2.6 with u=x≤M gives w<W. Hence either x<W/(2logα), or n<(W+logM)/logα, or (15a) fails and n<1+log(36.4M)/logα.
(S2) For each 2≤x<W/(2logα) and x−1≤j≤2x−1 we computed Λ3 (it is nonzero, except for (k,x,j)=(2,2,2), see Lemma 5.2) and n<log(1.5x/∣Λ3∣)/logα from Lemma 5.1(b). Together with (S1) this gives n≤N.
(S3) For each 2≤n≤N, Lemma 2.6 applied to Γ1/logPn+1, where Γ1=log(2g)+mlogα−xlogPn+1, with τ=logα/logPn+1, μ=log(2g)/logPn+1, u=m≤NM+1, A=4/logPn+1 and B=5, gives a new bound for x.
Proposition 7.1. Let 2≤k≤257. Every solution of (2) with n≥2 and x≥2 satisfies n≤12 and x≤8. For k≥3 there is no such solution, and for k=2 all of them have x=2 and m=2n.
Proof. The bounds follow from Table 3. For each k, a search in exact integer arithmetic over 2≤n≤Nk, 2≤x≤Xk, where Nk≤12 and Xk≤8 are the final bounds of the third pass for this k, comparing (Pn+1x−Pn−1x)/2 with the list of the Pm, m≤NkXk+3, gives the result. □
pass
M
W in (S1)
x in (S2)
N
x from (S3)
1
1.08×1050
124.45
64
248
82
2
82
14.99
7
19
11
3
11
9.63
5
12
8
Table 3. Maximum over 2≤k≤257 of the quantities met in the three passes.
Proof of Theorem 1.1
Let (k,n,m,x) be a solution of (2). If x=1 or n=1, or if k=2 and x=2, we conclude with Lemmas 3.1, 3.2 and 3.3. Otherwise (11) holds, that is, n≥2, x≥2 and (k,x)=(2,2). By Proposition 6.4 we have k≤257. By Proposition 7.1, every solution with 2≤k≤257, n≥2 and x≥2 has k=2 and x=2, which is excluded by (11). Hence no solution satisfies (11), and this proves Theorem 1.1. □
Appendix A. Computational details
All computations were done in Python 3. Real numbers were handled with the decimal module; the precision was chosen in each step as twice the number of digits of the bound M plus 60 to 120 digits, so that the products τq and μq in Lemma 2.6, where q is about 6M⋅1020, are correct. The dominant root α was computed by Newton’s method on the equation x2−3x+1+x1−k=0 from (6), which is well conditioned for large k, starting from φ2, and g from (7). The k-Pell numbers were computed exactly. In all the applications of Lemma 2.6 the first convergent with q>6M or the next one gave ε>0.
As a consistency check, independent of the proof, we also verified by a direct search that, for 3≤k≤100, 1≤n≤150, 1≤x≤60, and for 2≤k≤60, 1≤n≤80, 1≤x≤30, the only solutions of (2) are those of Theorem 1.1. The minimum in Lemma 5.2(b) was computed with 150 significant digits.
For k≥30 the minimum of ∣L3∣ over 2≤x≤480 and x−1≤j≤2x−1 is greater than 8.43×10−4, attained at (x,j)=(457,610); so (R2) never meets a small value.
Declarations
Data availability. The Python scripts used in this paper are available from the second author on request.
Use of AI tools. Claude (Anthropic) was used to find the error in an earlier version of this work, to write and run the Python code described in Appendix A. The authors take full responsibility for the content.
[2]H. Bensella, B. K. Patel, D. Behloul, On the exponential Diophantine equation (Fₘ₊₁⁽ᵏ⁾)ˣ − (Fₘ₋₁⁽ᵏ⁾)ˣ = Fₙ⁽ᵏ⁾, preprint (2022), arXiv:2205.13168.arxiv.org/abs/2205.13168
[3]J. J. Bravo, J. L. Herrera, Repdigits in generalized Pell sequences, Arch. Math. (Brno) 56 (2020), 249–262.DOI
[4]J. J. Bravo, J. L. Herrera, F. Luca, On a generalization of the Pell sequence, Math. Bohem. 146 (2021), no. 2, 199–213.DOI
[5]J. J. Bravo, J. L. Herrera, F. Luca, Common values of generalized Fibonacci and Pell sequences, J. Number Theory 226 (2021), 51–71.DOI
[6]A. Dujella, A. Pethő, A generalization of a theorem of Baker and Davenport, Quart. J. Math. Oxford Ser. (2) 49 (1998), no. 195, 291–306.DOI
[7]F. Faye, B. Faye, On the exponential Diophantine equation (Pₙ⁽ᵏ⁾)ˣ + (Pₙ⁽ᵏ⁾)ˣ = Pₘ⁽ᵏ⁾, preprint (2026).
[8]C. A. Gómez, J. C. Gómez, F. Luca, The complete solution of the Diophantine equation (Fₙ₊₁⁽ᵏ⁾)ˣ − (Fₙ₋₁⁽ᵏ⁾)ˣ = Fₘ⁽ᵏ⁾, Mediterr. J. Math. 21 (2024), Paper No. 13. https://doi.org/10.1007/s00009-023-02529-5
[9]J. C. Gómez, F. Luca, C. A. Gómez, On the exponential Diophantine equation Fₙ₊₁ˣ − Fₙ₋₁ˣ = Fₘ, Taiwanese J. Math. 26 (2022), no. 4. https://doi.org/10.11650/tjm/220303
[10]E. Kılıç, On the usual Fibonacci and generalized order-k Pell numbers, Ars Combin. 88 (2008), 33–45.
[11]E. M. Matveev, An explicit lower bound for a homogeneous rational linear form in logarithms of algebraic numbers II, Izv. Ross. Akad. Nauk Ser. Mat. 64 (2000), no. 6, 125–180; English translation in Izv. Math. 64 (2000), no. 6, 1217–1269.
[12]B. K. Patel, A. P. Chaves, On the exponential Diophantine equation Fₙ₊₁ˣ − Fₙ₋₁ˣ = Fₘ, Mediterr. J. Math. 18 (2021), Paper No. 187. https://doi.org/10.1007/s00009-021-01831-4