Abstract
Let \((u_{n})_{n \ge 0}\) be a non-degenerate binary recurrence sequence with positive discriminant and p be a fixed prime number. In this paper, we are interested in finding a finiteness result for the solutions of the Diophantine equation \(u_{n_{1}} + u_{n_{2}} + \cdots + u_{n_{t}} = p^{z}\) with \(n_1> n_2> \cdots > n_t\ge 0\). Moreover, we explicitly find all the powers of three which are sums of three balancing numbers using lower bounds for linear forms in logarithms. Further, we use a variant of the Baker–Davenport reduction method in Diophantine approximation due to Dujella and Pethő.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
A large number of interesting Diophantine equations arise when one studies the intersection of two sequences of positive integers. For instance, one can ask when the terms of a fixed binary recurrence sequence are perfect powers, factorials or combinatorial numbers etc. As an example, one can consider the solvability of the Diophantine equation
in integers n, x, z with \(z\ge 2\), where \((u_{n})_{n \ge 0}\) is a linear recurrence sequence. Pethő [15, Corollary] and Shorey-Stewart [21, Theorem 2] independently proved that, Eq. (1.1) has only finitely many solutions under certain natural assumptions. The problem of finding all perfect powers in the Fibonacci sequence has a very rich history [9, 16] and this problem was open for a quite long time. In 2006, Bugeaud, Mignotte, and Siksek [8, Theorem 1] by using both the classical and the modular approach, proved that
are the only solutions of Eq. (1.1) where \(u_{n}\) is the Fibonacci sequence. In the same paper [8, Theorem 2], they also studied Eq. (1.1) for the Lucas number sequence. In 2008, Pethő [18, Theorem 2] (see also [10]) solved Eq. (1.1) when \(u_{n}\) is the Pell sequence and proved that 0, 1 and 169 are the only perfect powers.
Now one can extend Eq. (1.1) by taking two terms of a binary recurrence sequence and ask the same question, i.e., solvability of the Diophantine equation
in integers n, m, x, z with \(z\ge 2\). In this direction, several authors studied the problem of finding (n, m, z) such that
where \((u_{n})_{n \ge 0}\) is a fixed recurrence sequence. In particular, Bravo and Luca considered the case when \(u_{n}\) is the Fibonacci sequence [7, Theorem 2] and the Lucas number sequence [6, Theorem 2]. Further, variants of Eq. (1.3) were studied independently by Bravo et al. [5, Theorem 1] and Marques [13, Theorem 1], where \(u_{n}\) is replaced by the generalized Fibonacci sequence. Bravo et al. [4, Theorem 1] investigated the case when the power of two can be expressed as sum of three Fibonacci numbers. Recently, Pink and Ziegler [19, Theorem 1] have generalized the results of Bravo and Luca [6, 7], and considered a more general Diophantine equation
in non-negative integer unknowns \(n, m, z_{1}, \ldots , z_{s}\), where \((u_{n})_{n \ge 0}\) is a binary non-degenerate recurrence sequence, \(p_{1}, \ldots , p_{s}\) are distinct primes and w is a non-zero integer with \(p_{i} \not \mid w\) for \(1\le i \le s\). They proved that under certain assumptions, Eq. (1.4) has finitely many solutions using lower bounds for linear forms of p-adic logarithms. In [2, Theorem 2.1], Bertók et al. solved completely equations of the form \(u_n = 2^a + 3^b + 5^c\), where \(u_n\) is one of the Fibonacci, Lucas, Pell and associated Pell sequences, respectively.
The purpose of this paper is twofold. On one hand, we give a general finiteness result for the solutions of the equation
in non-negative integer unknowns \(n_{1},\ldots , n_{t}, z\), where \((u_{n})_{n \ge 0}\) is a binary non-degenerate recurrence sequence with \(n_{1}> n_{2}> \cdots > n_{t} \ge 0\) and p is a given prime. On the other hand, we completely solve Eq. (1.5) where \(u_{n}\) is a balancing number sequence and \((t, p) = (3, 3)\). To prove our main results, we use lower bounds for linear forms in logarithms of algebraic numbers and a version of the Baker–Davenport reduction method.
2 Notation and main results
The sequence \((u_{n})_{n \ge 0} = u_{n}(P, Q, u_{0}, u_{1})\) is called a binary linear recurrence sequence if the relation
holds, where \(P, Q\in \mathbb {Z}\) with \(PQ\ne 0\) and \(u_{0}, u_{1}\) are fixed rational integers with \(|u_{0}| + |u_{1}| > 0\). The polynomial \(f(x) = x^2 -Px -Q\) is called the companion polynomial of the sequence \(u_n\). Let \(\Delta = P^2 + 4Q\) be the discriminant of f. We call \(\Delta \) the discriminant of the sequence \(u_n\). The roots of the companion polynomial are denoted by \(\alpha \) and \(\beta \). Then for \(n\ge 0\)
where \(a = u_{1} - u_{0}\beta ,\, b = u_{1} - u_{0}\alpha \). The sequence \((u_{n})_{n\ge 0}\) is called non-degenerate, if \(ab\alpha \beta \ne 0\) and \(\alpha /\beta \) is not a root of unity.
Throughout the paper, we assume that \(u_{n}\) is non-degenerate, \(\sqrt{\Delta } = (\alpha -\beta ) >0\) (that is \(\alpha \) and \(\beta \) are real). We label the roots in such a way that \(|\alpha | > |\beta |\). Up to changing the signs to \((\alpha , \beta )\) simultaneously (that is, replacing \((\alpha , \beta )\) by \((-\alpha , -\beta )\), which has as effect replacing (P, Q) by \((-P, Q)\) and the sequence \((u_n)_{n\ge 0}\) by the sequence\(((-1)^nu_n)_{n\ge 0}\), we may assume that \(\alpha \) is positive. From now on, we will assume these conditions.
With these notations, we have the following theorem.
Theorem 1
Suppose \((u_{n})_{n \ge 0}\) is a non-degenerate binary recurrence sequence of integers satisfying recurrence (2.1). Let us assume that \(\alpha > |\beta |\) and \(\Delta = r^2s \) where \(r, s \in \mathbb {Z}_{ >0}, s\ne 1, s\) is square-free. Then there exists an effectively computable constant C depending on \((u_{n})_{n\ge 0}, p, t\) such that all solutions \((n_{1},\ldots , n_{t},z)\) to Eq. (1.5) satisfy
Remark 2.1
Suppose \(u_n = 2^n - 1\) and \(t = p = 2\). In this case \(\alpha = 2,\; \beta = a = b =1\) and \(\Delta = s = 1\). Then (1.5) becomes
which has infinitely many solutions given by \(n_2 = 1\) and \(n_1 = z\). This shows that the assumption \(s \ne 1\) is necessary in Theorem 1.
Balancing numbersn are solutions of the Diophantine equation
for some natural number m [1]. Let us denote the nth balancing number by \(B_{n}\). Also, the balancing numbers satisfy the recurrence relation \(B_{n+1}=6B_{n}-B_{n-1}\) with initial conditions \(B_0 = 0, B_1 = 1\) for \(n\ge 1\). Therefore, the sequence in Eq. (2.1) for \((P, Q, u_{0}, u_{1}) = (6, -1, 0, 1)\) is a balancing sequence. For more details regarding balancing numbers one can refer [1, 11, 20]. We prove the following theorem as an example for the explicit computations of constants.
Theorem 2
There is no solution of the Diophantine equation
in integers \(n_{1}, n_{2}, n_{3}, z\) with \(n_{1}> n_{2} > n_{3} \ge 0\).
3 Auxiliary results
In the following lemma we find an upper bound for \(|u_{n}|\) which is useful in the proof of Theorem 1 and also we find a relation between \(n_1\) and z when Eq. (1.5) holds.
Lemma 3.1
There exist constants \(d_0\) and \(d_1\) depending only on \((u_n)_{n\ge 0}\) such that the following holds.
-
(1)
\(|u_{n}|< d_{0}\alpha ^{n}\).
-
(2)
If Eq. (1.5) holds with \(n_1> n_2> \cdots > n_t\), then \(z\le d_1 n_1\).
Proof
From Eq. (2.2), we have
Since \(\alpha > |\beta |\), the above inequality becomes
where \(d_{0} := (|a| + |b|)/\sqrt{\Delta }\). This proves (1).
As p is given, we can always choose a suitable integer \(d_1\) such that \(d_{0}\alpha ^{n_i} < p^{n_id_1} \, (1\le i\le t)\). Thus, the above inequality becomes
Therefore, we get \(z \le d_{1}n_{1}\). \(\square \)
The following lemma is due to Pethő and de Weger [17, Lemma 2.2].
Lemma 3.2
[17] Let \(u, v \ge 0, h\ge 1\) and \(x\in \mathbb {R}\) be the largest solution of \(x = u + v (\log x)^{h}\). Then
Let \(\eta \) be an algebraic number of degree d with minimal polynomial
where \(a_0\) is the leading coefficient of the minimal polynomial of \(\eta \) over \(\mathbb {Z}\) and the \(\eta ^{(i)}\)’s are conjugates of \(\eta \) in \(\mathbb {C}\). Define the absolute logarithmic height of an algebraic number \(\eta \) by
In particular, if \(\eta = p/q\) is a rational number with \(\gcd (p, q) = 1\) and \(q >0\), then \(h(\eta ) = \log \max \{|p|, q\}\). Some important properties of logarithmic height which we use in our further investigation are as follows (see e.g. [22]).
-
(1)
\(h(\eta \pm \gamma ) \le h(\eta ) + h(\gamma ) + \log 2\),
-
(2)
\(h(\eta \gamma ^{\pm 1}) \le h(\eta ) + h(\gamma )\),
-
(3)
\(h(\eta ^{t}) \le |t|h(\eta ),\;\) for \(t \in \mathbb {Z}\).
Generally it is a very hard problem to find lower bounds for the height of elements in a number field of given degree. For the quadratic fields, we have the following result due to Pink and Ziegler [19, Lemma 3].
Lemma 3.3
[19] For an algebraic number \(\alpha \) of degree two we have \(h(\alpha ) \ge 0.24\) or \(\alpha \) is a root of unity.
To prove our theorem, we use lower bounds for linear forms in logarithms to bound the index \(n_1\) appearing in Eq. (1.5). Generically, we need the following general lower bound for linear forms in logarithms due to Matveev [14, Theorem 2.2] (see also [8, Theorem 9.4]).
Lemma 3.4
[8] Let \(\gamma _1,\ldots ,\gamma _t\) be positive real algebraic numbers and let \(b_{1},\ldots , b_{t}\) be non-zero rational integers. Let \(\mathbb {Q}(\gamma _1,\ldots ,\gamma _t)\) be the quadratic number field over \(\mathbb {Q}\) and let \(A_{j}\) be real numbers satisfying
Assume that \(B\ge \max \{|b_1|, \ldots , |b_{t}|\}\) and \(\Lambda :=\gamma _{1}^{b_1}\cdots \gamma _{t}^{b_t} - 1\). If \(\Lambda \ne 0\), then
The upper bound on \(n_1\) of the Diophantine equation (2.3) is too large for practical purpose, thus the next step is to reduce it. We present a lemma from [3, Lemma 4] and it is a variant of the famous Dujella and Pethő [12, Lemma 5a] reduction lemma. Now for a real number x, let \(||x|| := \min \{|x - n| : n \in \mathbb {Z}\}\) denote the distance from x to the nearest integer.
Lemma 3.5
[3] Suppose that M is a positive integer, and A, B are positive reals with \(B > 1\). Let p / q be the convergent of the continued fraction expansion of the irrational number \(\gamma \) such that \(q > 6M\), and let \(\epsilon := ||\mu q|| - M||\gamma q||\), where \(\mu \) is a real number. If \(\epsilon > 0\), then there is no solution of the inequality
in positive integers u, m and n with
To prove Theorem 1, we apply linear forms in logarithms several times. Every time, we find an upper bound for \((n_1 - n_i)\) in terms of \(n_1\) for \(2\le i \le t\). In order to apply the linear forms in logarithms to bound \((n_1 - n_i)\) for a fixed i, we require upper bounds of \((n_1 - n_j)\) for all \(1\le j < i\). Finally, using these upper bounds for \((n_1 - n_i),\;2\le i \le t\), we get an upper bound for \(n_1\). In order to apply Lemma 3.4 we must ensure that \(\Lambda _i\) does not vanish. In this regard, we have the following lemma.
Lemma 3.6
Suppose \(\Lambda _i := p^{z}\alpha ^{-n_1}|a|^{-1}r\sqrt{s}(1 + \alpha ^{n_2 - n_1} + \cdots + \alpha ^{n_i - n_1})^{-1} - 1\) for all \(1\le i \le t\). If \(\Lambda _i = 0\), then \(n_1 < \ell _i\), where
Proof
Now \(\Lambda _i = 0\) imply
Since s is a square-free integer other than one, by taking the conjugate of (3.3), we get
Since \(\alpha > 0\)
If \(|\beta | \le 1\), then \(|a|\alpha ^{n_1}< |b|i< i (|b|+|a|)\) which implies that \(n_1 < \frac{\log (i(|b|+|a|)/|a|)}{\log \alpha }\). Now for \(|\beta | > 1\),
which implies that \(n_1 < \frac{\log (i(|b|+|a|)/|a|)}{\log (\alpha /|\beta |)}.\)\(\square \)
The following lemma gives a relationship between height of an algebraic number and its logarithm. We can suitably choose a constant \(d_2\) such that
We also denote \(K = \mathbb {Q}(\alpha , \beta ) = \mathbb {Q}(\sqrt{\Delta }) = \mathbb {Q}(\sqrt{s})\) for the quadratic number field corresponding to binary recurrence sequence \((u_n)_{\ge 0}\).
Lemma 3.7
Let \(\gamma _3(i) := |a|^{-1}r\sqrt{s}(1 + \alpha ^{n_2 - n_1} + \cdots + \alpha ^{n_i - n_1})^{-1}\) are the algebraic numbers in the quadratic number field \({\mathbb {Q}}(\sqrt{s})\) for \(1 \le i \le t\). Then there exists a constant \(d_2\) depending on \(\alpha \) and \(\beta \) such that
Proof
First observe that \( r\sqrt{s} \ge 1\), as \(r\sqrt{s}=\sqrt{\Delta } = \alpha - \beta = \sqrt{P^2 + 4 Q} > 0\) and P, Q are rational integers. Since \(\gamma _3(i) = |a|^{-1}r\sqrt{s}(1 + \alpha ^{n_2 - n_1} + \cdots + \alpha ^{n_i - n_1})^{-1}\) and \(\alpha > 0\), we have
and this implies
Since \(\log i < i\log 2\) for \(i\ge 1\), we have
Furthermore, a and b are conjugate algebraic numbers which are roots of the quadratic polynomial
Thus,
Now we estimate height of \(\gamma _3(i)\). We have
Since s is square-free we have
By Lemma 3.3, we have \(h(\alpha ) \ge 0.24\). Finally the lemma follows by comparing (3.6), (3.7), (3.8) and (3.9).\(\square \)
Suppose \(\ell := \max \{\ell _1, \cdots , \ell _{t}\}\), where \(\ell _i \;(1\le i\le t)\) are defined in Lemma 3.6. If \(n_1 \le \ell \), then the conclusion of Theorem 1 follows trivially (see Lemma 3.1(2)). Henceforth, we assume \(n_1 > \ell \).
Now we are ready to prove Theorem 1. The proof is motivated by the ideas of Bravo and Luca [6, 7].
4 Proof of Theorem 1
4.1 Bounding \((n_1 - n_i)\) in terms of \(n_1\) for \(2\le i \le t\)
Here, we claim that for \(2\le i \le t\), we have
where \(C_i\)’s are effectively computable constants depending on \(P, Q, u_0, u_1, t, p\). We use induction on i to find an upper bound of \(n_1 - n_i\) for \(2\le i \le t\). First, we calculate the upper bound of \(n_1 - n_2\). We can rewrite Eq. (1.5) as
From Lemma 3.1, \(|u_{n}|\le d_{0}\alpha ^{n}\). Also we have assumed that \(n_1> n_2> \cdots > n_{t}\). Thus,
If \(|\beta | < 1\), then the above inequality becomes
where \(c_1\) is a suitable constant. Now dividing both sides of the inequality (4.3) by \(|a|\alpha ^{n_1}/r\sqrt{s}\),
where \(c_2\) is a suitable constant. For the case \(|\beta | > 1\), we divide both sides of the inequality (4.2) by \(|a|\alpha ^{n_1}/r\sqrt{s}\),
Thus, for any \(\beta \) it follows that
where \(c_3 := \max \left\{ c_2, \left( \frac{|b|}{|a|} + \frac{(t-1)d_{0}r\sqrt{s}}{|a|}\right) \right\} \) and the left inequality in (4.5) is obtained using triangle inequality.
In order to apply Lemma 3.4, we take
Thus our first linear form is \(\Lambda _1 := \gamma _{1}^{b_{1}}\gamma _{2}^{b_{2}}\gamma _{3}^{b_{3}} - 1\) . Further, to apply Lemma 3.4 we must ensure that \(\Lambda _1 \ne 0\). Suppose on contrary \(\Lambda _1 = 0\). Then from Lemma 3.6, we have \(n_1< \ell _1\) and hence \(n_1\le \ell \) (see the discussion towards the end of the Sect. 3). Thus, this is a contradiction as we have assumed \(n_1 > \ell \). Hence, \(\Lambda _1 \ne 0\). Here we are taking the quadratic number field \({\mathbb {Q}}(\sqrt{s})\) over \({\mathbb {Q}}\) and \(t=3\). Finally, recall \(z\le d_1n_1\) and so we deduce
Hence we can take \(B := d_1n_1\). Also \( h(\gamma _1)= \log p, h(\gamma _2)\le \log \alpha \). Thus, we can take \(A_1 := 2\log p, A_2 = 2\log \alpha \). Further, using Lemma 3.7, we have
Hence, \( A_3 = 2\left( \log d_2+ \log (r\sqrt{s}) + \log 4 \right) \). Using Lemma 3.4, we have
So the above inequality can be rewritten as,
where \(C_1\) is a suitable constant. Taking logarithms in both sides of inequality (4.5) and comparing the resulting inequality with inequality (4.6),
where \(C_2'\) is a suitable constant. This implies that
Therefore, for \(i = 2\), the statement is true. We assume that inequality (4.1) is true for \(i = k\) with \(2\le k \le t-1\). We want to show that inequality (4.1) is true for \(i = k+1\). To formulate the next linear form we rewrite Eq. (1.5) as follows
Taking absolute value and using inequality (3.1), we have
If \(|\beta | < 1\), then (4.8) becomes
where \(c_4\) is a suitable constant. Dividing both sides of the inequality (4.9) by \(|a|(\alpha ^{n_1} +\cdots +\alpha ^{n_{k}})/r\sqrt{s}\),
where \(c_5 = c_4 r\sqrt{s}/|a|\). For the case \(|\beta | > 1\), we divide both sides of the inequality (4.8) by \(|a|(\alpha ^{n_1} +\cdots +\alpha ^{n_{k}})/r\sqrt{s}\),
Thus, for any \(\beta \) it follows that
where \(c_6 := \max \left\{ c_5, \left( \frac{k|b|}{|a|} + \frac{d_{0}r\sqrt{s}}{|a|}\right) \right\} \). To apply Lemma 3.4 we take,
Our next linear form is \(\Lambda _2 := \gamma _{1}^{b_{1}}\gamma _{2}^{b_{2}}\gamma _{3}^{b_{3}} - 1\). Suppose \(\Lambda _2 = 0\). Then from Lemma 3.6, \(n_1<\ell _k\). That is \(n_1\le \ell \) and this is a contradiction as we have assumed \(n_1 > \ell \). Thus, \(\Lambda _2 \ne 0\). Like in the previous case, here we take \(B= d_1n_1\). We already estimate \(A_1, A_2\) in previous case. Using Lemma 3.7 we can take \(A_3^k = c_{7} + 2(|n_2 - n_1| + \cdots +|n_{k} - n_1|) \log \alpha \) where \(c_7 = 2\left( \log d_2 + \log (r\sqrt{s}) \right) + (k+1)\log 4\).
From Lemma 3.4, we have
Now from the inequalities (4.11) and (4.12), we get
where \(c_8\) is a suitable constant. Now taking logarithm on both sides of (4.13)
By induction hypothesis, we have \((n_1 - n_i) < C_i (\log n_1)^{i-1}\) for \(2\le i \le k\). Thus from (4.14), we obtain
4.2 Bounding \(n_1\)
To bound \(n_1\), we write the Eq. (1.5) as
which implies that
If \(|\beta | < 1\), then (4.16) becomes
Dividing both sides of the inequality (4.17) by \(|a|(\alpha ^{n_1} +\cdots +\alpha ^{n_{t}})/r\sqrt{s}\),
where \(c_{9} = t|b|/ |a|\). For the case \(|\beta | > 1\), we divide both sides of the inequality (4.17) by \(|a|(\alpha ^{n_1} +\cdots +\alpha ^{n_{t}})/r\sqrt{s}\),
Thus, for any \(\beta \) we have
where \(c_{10}:= \max \left\{ c_9, \left( \frac{t|b|}{|a|} \right) \right\} \). To apply Lemma 3.4, we take
Thus, the final linear form is \(\Lambda _t := \gamma _{1}^{b_{1}}\gamma _{2}^{b_{2}}\gamma _{3}^{b_{3}} - 1\). Suppose \(\Lambda _t = 0\). Then from Lemma 3.6, \(n_1< \ell _t\). That is \(n_1\le \ell \) and this is a contradiction as we have assumed \(n_1 > \ell \). Thus, \(\Lambda _t \ne 0\). Here we take \(B= d_1n_1, A_1 = 2\log p, A_2 = 2\log \alpha \). From the conclusions of Lemma 3.7, we can take \(A_3^t= c_{11} + 2[(n_1 - n_2) + \cdots + (n_1 - n_t)]\log \alpha \) where \(c_{11} = 2\left( \log d_2+ \log (r\sqrt{s})\right) + (t+1)\log 4\). From Lemma 3.4, we have
Now from the inequalities (4.19) and (4.20), we get
where \(c_{12}\) is a suitable constant. Now taking logarithm on both sides of (4.21)
Since \((n_1 - n_i) < C_i (\log n_1)^{i-1}\) for \(2\le i \le t\), we deduce from (4.22)
where C is a suitable constant. Theorem 1 follows by applying Lemma 3.2 and Lemma 3.1(2) to the inequality (4.23).
5 Proof of Theorem 2
We give the computational details of the resolution of Diophantine Eq. (2.3). Let \((B_{n})_{n \ge 0}\) be the balancing sequence given by \(B_{0} = 0, B_1 = 1\) and \(B_{n+1} = 6B_{n} - B_{n-1}\) for all \(n\ge 1\). Now one can easily see from (2.2), \(a =1, b=1, \alpha =3+2\sqrt{2},\; \beta =3-2\sqrt{2}\) and the nth term of balancing sequence is
5.1 The case \(n_3 =0\)
In this case \(n_2 > 0\), thus we have
which is a reduced form of Eq. (2.3). Hence without loss of generality, we assume \(n_3 > 0\).
5.2 Bounding \((n_1 - n_2)\) and \((n_1 - n_3)\) in terms of \(n_1\)
If \( 1\le n_1 \le 100\), then a brute force search with Mathematica in the range \(1\le n_{3}< n_2 < n_1 \le 100\) gives no solution of Eq. (2.3). Hence from this point onward, we assume that \(n_1 > 100\). Using Lemma 3.1, for balancing sequence we have \(|B_n| \le \alpha ^n\). Now from (2.3) we have
and this implies
We record this estimate for future referencing. We rewrite (2.3) as
Now take the absolute values on the both hand side of the above relation obtaining
Dividing both sides of the above inequality by \(\frac{\alpha ^{n_1}}{4\sqrt{2}}\), we obtain the first linear form
In order to apply Lemma 3.4, we take
Thus, \(B = 2n_1, h(\gamma _1)= \log 3= 1.0986\cdots , h(\gamma _2) = (\log \alpha )/2 = 0.8813\cdots , h(\gamma _3) \le 0.8664\cdots \). We can choose \(A_1 = 2.4, A_2 = 1.9, A_3 = 1.8\). By Lemma 3.4, we obtain the lower bound for the linear form \(\Lambda _1 := \gamma _{1}^{b_{1}}\gamma _{2}^{b_{2}}\gamma _{3}^{b_{3}} - 1\) is
where \(C_1 = 1.4 \times 30^{5}\times 2^{4.5}\times 4\times (1+ \log 2) \times 2.4 \times 1.9\times 1.8 = 7.9\times 10^{12}\). Further, since \((1+\log 2n_1) < 2\log n_1\) for \(n_1 > 100\), we get from (5.5) and (5.6) that
We now construct a second linear form in logarithms by rewriting Eq. (2.3) as follows:
Taking absolute values in the above relation and using the fact that \(|\beta | <1\)
Dividing both sides of the above inequality by \(\frac{\alpha ^{n_1}}{4\sqrt{2}}(1 + \alpha ^{n_2 - n_1})\), we obtain
Here our second linear form is \(\Lambda _2 := \gamma _{1}^{b_{1}}\gamma _{2}^{b_{2}}\gamma _{3}^{b_{3}} - 1\) where
In this case \(A_1\) and \(A_2\) are same as previous case but \(A_3 = \log 8\sqrt{2} + (n_1 - n_2)\log \alpha \). Thus by applying Lemma 3.4 we get
From inequalities (5.8) and (5.9),
5.3 Bounding \(n_1\)
To bound \(n_1\), we consider the analogue equation for Eq. (4.18) which is
with \(c_{9} = t|b|/|a| = 3\). Thus, final linear form is \(\Lambda _3 := \gamma _{1}^{b_{1}}\gamma _{2}^{b_{2}}\gamma _{3}^{b_{3}} - 1\), where
Here by taking \(A_3 = \log 4\sqrt{2} + (n_1 - n_2)\log \alpha + (n_1 - n_3)\log \alpha + \log 4\) and applying Lemma 3.4, we get
From inequalities (5.11) and (5.12),
Further, from Lemma 3.2, we have
We summarize the above discussion in the following proposition.
Proposition 3
Let us assume that \(n_1> n_2 > n_3\) and \(n_1 > 100\). If \((n_1, n_2, n_3, z)\) is a positive integral solution of Eq. (2.3), then
5.4 Reducing the size of \(n_1\)
From Proposition 3, we can see that the bound we have obtained for \(n_{1}\) is very large. Now our job is to reduce this upper bound to a certain minimal range. We return to inequality (5.5). Put
Then (5.5) implies
Note that \(\Lambda _1 > 0\), otherwise \(3^z \le \alpha ^{n_1}/4\sqrt{2}\). But we always have
Using the fact that \(1+x< e^x \) holds for all positive real numbers x, we get
Dividing Eq. (5.16) by \(\log \alpha \), we have
We are now ready to use Lemma 3.5 with parameters
Let \([a_0, a_1, a_2,\ldots ] = [0, 1, 1, 1, 1, 1, 8, 4, 17,\ldots ]\) be a continued fraction expansion of \(\gamma \) and let \(p_k/q_k\) be its kth convergent. We take \(M:= 3\times 10^{45}\), then using Mathematica, we can see that
To apply Lemma 3.5, consider \(\epsilon := \Vert \mu q_{99}\Vert - M\Vert \gamma q_{99}\Vert \) which is positive. If (2.3) has a solution \((n_1, n_2, n_3, z)\), then \((n_1 - n_2) \in [0,70]\). Next we look into inequality (5.8) to estimate the upper bound for \((n_1 - n_3)\). Now putting
where we take \(\phi (x)= 4\sqrt{2}(1+\alpha ^{-x})^{-1}\), it implies
Using the Binet formula of the balancing sequence, one can show that \(\Lambda _2 >0\) since
Altogether we get
Replacing \(\Lambda _2\) in inequality (5.19) by its formula in Eq. (5.18) and arguing as in inequality (5.17), we get
Again we use Lemma 3.5 here with the following parameters
Proceeding like before with \(M := 3\times 10^{45}\) and applying Lemma 3.5 to the inequality (5.20) for all possible choices of \(n_1-n_2 \in [0,70]\) we find that if Eq. (2.3) has a solution \((n_1, n_2, n_3, z)\), then \((n_1 - n_3) \in [0,72]\). Finally, in order to obtain a better upper bound on \(n_1\), we can put
with \(\psi (x_1,x_2):= 4\sqrt{2}(1+ \alpha ^{-x_1}+\alpha ^{-x_2})^{-1}\), which implies
One can observe \(\Lambda _3 \ne 0\). So, for rest of the cases we consider \(\Lambda _3 > 0\) and \(\Lambda _3< 0\) separately. If \(\Lambda _3 > 0\), then
Suppose \(\Lambda _3< 0\). Since \(\frac{3}{\alpha ^{n_1}}< \frac{1}{2}\) for \(n_1> 2\), we get that \(|e^{\Lambda _3}-1|< 1/2\), therefore \(e^{|\Lambda _3|}< 2\). Thus,
Thus for both cases we have
Putting the value of \(\Lambda _3\) from Eq. (5.22) in inequality (5.24) and arguing as previously we obtain
Now, we repeat the same procedure as earlier with \(M:= 3\times 10^{45}\) for inequality (5.25). For all possible choices of \(n_1-n_2 \in [0,70]\) and \(n_1-n_3 \in [0,72]\), we apply Lemma 3.5 to inequality (5.25). If Eq. (2.3) has a solution \((n_1, n_2, n_3, z)\), then \(n_1 \in [0,75]\). This leads to a contradiction to our assumption that \(n_1 > 100\), which completes the proof of Theorem 2.
References
Behera, A., Panda, G.K.: On the square roots of triangular numbers. Fibonacci Q. 37, 98–105 (1999)
Bertók, C., Hajdu, L., Pink, I., Rábai, Z.: Linear combinations of prime powers in binary recurrence sequences. Int. J. Number Theory 13(2), 261–271 (2017)
Bravo, J.J., Luca, F.: On a conjecture about repdigits in \(k\)-generalized Fibonacci sequences. Publ. Math. Debrecen 82, 623–639 (2013)
Bravo, E.F., Bravo, J.J.: Power of two as sums of three Fibonacci numbers. Lith. Math. J. 55, 301–311 (2015)
Bravo, J.J., Gómez, C.A., Luca, F.: Powers of two as sums of two \(k\)-Fibonacci numbers. Miskolc Math. Notes 17, 85–100 (2016)
Bravo, J.J., Luca, F.: Power of two as sums of two Lucas numbers. J. Integer Seq. 17, A.14.8.3 (2014)
Bravo, J.J., Luca, F.: Power of two as sums of two Fibonacci numbers. Quaest. Math. 39, 391–400 (2016)
Bugeaud, Y., Mignotte, M., Siksek, S.: Classical and modular approaches to exponential Diophantine equations I. Fibonacci and Lucas perfect powers. Ann Math. 163, 969–1018 (2006)
Cohn, J.H.E.: On square Fibonacci numbers. J. Lond. Math. Soc. 39, 537–540 (1964)
Cohn, J.H.E.: Perfect Pell powers. Glasg. Math. J. 38, 19–20 (1996)
Dey, P.K., Rout, S.S.: Diophantine equations concerning balancing and Lucas balancing numbers. Arch. Math. (Basel) 108(1), 29–43 (2017)
Dujella, A., Pethő, A.: A generalization of a theorem of Baker and Davenport. Q. J. Math. Oxf. Ser. 49, 291–306 (1998)
Marques, D.: Powers of two as sum of two generalized Fibonacci numbers, available at arXiv:1409.2704
Matveev, E.M.: An explicit lower bound for a homogeneous rational linear form in the logarithms of algebraic numbers II. Izv. Math. 64, 1217–1269 (2000)
Pethő, A.: Perfect powers in second order linear recurrences. J. Number Theory 15, 5–13 (1982)
Pethő, A.: Full cubes in the Fibonacci sequence. Publ. Math. Debr. 30, 117–127 (1983)
Pethő, A., de Weger, B.M.M.: Product of prime powers in binary recurrence sequences. I: The hyperbolic case, with an application to the generalized Ramanujan–Nagell equation. Math. Comp. 47, 713–727 (1986)
Pethő, A.: The Pell sequence contains only trivial perfect powers, Coll. Math. Soc. J. Bolyai, 60 sets, Graphs and Numbers, Budapest, pp. 561–568 (1991)
Pink, I., Ziegler, V.: Effective resolution of Diophantine equations of the form \(u_{n} + u_{m} = wp_{1}^{z_{1}} \cdots p_{s}^{z_{s}}\). Monatsh Math. 185, 103–131 (2018)
Rout, S.S.: Some Generalizations and Properties of Balancing Numbers, Ph.D. Thesis, NIT Rourkela (2015)
Shorey, T.N., Stewart, C.L.: On the Diophantine equation \(ax^{2t}+bx^{t}y+cy^{2}=d\) and pure powers in recurrence sequences. Math. Scand. 52, 24–36 (1983)
Smart, N.P.: The Algorithmic Resolution of Diophantine Equations, London Mathematical Society, Student Texts, Vol 41. Cambridge University Press, Cambridge (1998)
Acknowledgements
We thank the referee for suggestions which improved the quality of this paper. The first author would like to thank Harish-Chandra Research Institute, Allahabad and Institute of Mathematics & Applications, Bhubaneswar for their warm hospitality during the academic visits.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by J. Schoißengeier.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Mazumdar, E., Rout, S.S. Prime powers in sums of terms of binary recurrence sequences. Monatsh Math 189, 695–714 (2019). https://doi.org/10.1007/s00605-019-01282-w
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00605-019-01282-w