DOI 10.

1007/s10986-020-09505-6
Lithuanian Mathematical Journal

Even perfect numbers in generalized Pell sequences

Jhon J. Bravo1 and Jose L. Herrera
Departamento de Matemáticas, Universidad del Cauca, Calle 5 No 4–70, Popayán, Colombia
(e-mail: jbravo@unicauca.edu.co; joseherrera@unicauca.edu.co)

Received October 19, 2019; revised March 10, 2020

Abstract. In this paper, by using linear forms in logarithms and the Baker–Davenport reduction procedure we prove
that there are no even perfect numbers appearing in generalized Pell sequences. We also deduce some interesting results
involving generalized Pell numbers, which we believe are of independent interest. This paper continues a previous work
that searched for perfect numbers in the classical Pell sequence.
MSC: 11B39, 11J86
Keywords: generalized Pell number, perfect number, linear form in logarithms, reduction method

1 Introduction
A perfect number is defined as any positive integer where the sum of its proper divisors equals the number.
The first integers satisfying this property are

6, 28, 496, 8128, 33550336, 8589869056, 137438691328, . . . (sequence A000396).

Euclid, over two thousand years ago, showed that 2p−1 (2p − 1) is an even perfect number whenever 2p − 1 is
a prime. The prime numbers of the form 2p − 1 are known as Mersenne primes and require p itself to be prime.
A much less obvious result, due to Euler, showed that all even perfect numbers are of the form 2p−1 (2p − 1)
for some p such that 2p − 1 is a Mersenne prime. This is known as the Euclid–Euler theorem.
Theorem [Euclid–Euler]. An even positive integer is a perfect number if and only if it has the form 2p−1 ×
(2p − 1) with 2p − 1 a prime.
This theorem describes the relationship between perfect numbers and Mersenne primes. It is conjectured
that there are infinitely many Mersenne primes. This is equivalent, by the Euclid–Euler theorem, to the conjec-
ture that there are infinitely many even perfect numbers. It is also unknown whether there exists even a single
odd perfect number.
There are a lot of integer sequences used in number theory. For instance, the Fibonacci sequence (Fn )∞ n=0
is one of the most famous and curious numerical sequences in mathematics and has been widely studied in the
literature. Also, there is the Pell sequence, which is as important as the Fibonacci sequence. The Pell sequence
1
The author was supported in part by Projects VRI ID 4689 (Universidad del Cauca) and Colciencias 110371250560.

0363-1672/20/6004-0001 
c 2020 Springer Science+Business Media, LLC

1
2 J.J. Bravo and J.L. Herrera

(Pn )∞n=0 is defined by the recurrence Pn = 2Pn−1 + Pn−2 for n  2 with P0 = 0 and P1 = 1 as initial
conditions. For the beauty and rich applications of these numbers and their relatives, we refer the reader to
Koshy’s book [8].
Let φ(n) and σ(n) be the Euler function and the sum of divisors of the positive integer n, respectively.
Note that a number n is perfect if σ(n) = 2n.
There are many papers in the literature dealing with Diophantine equations involving arithmetic functions
in members of a binary recurrent sequence. For example, in [10], it was shown that 1, 2, and 3 are the only
Fibonacci numbers whose Euler function is also a Fibonacci number, whereas in [11], it was proved that if the
Euler function of the nth Pell number Pn is a power of 2, then n  8. In [9], it was shown that the Fibonacci
sequence contains no any perfect number, whereas in [4], it was proved that there are no such perfect numbers
in the Pell sequence.
In 2014, Facó and Marques [6] looked for perfect numbers in the k-generalized Fibonacci sequence and
proved that there are no even perfect numbers belonging to k-generalized Fibonacci sequences when k ≡ 3
(mod 4). The k-generalized Fibonacci sequence starts with k − 1 consecutive 0s followed by 1, and each term
afterward is the sum of the k preceding terms.
Inspired by these results, in this paper, we study a problem related to that of [4, 6, 9] but with the
k-generalized Pell sequence. The k-generalized Pell sequence or, for simplicity, the k-Pell sequence P (k) =
(k)
(Pn )∞ n=−(k−2) is given by the following linear recurrence of higher order:

(k) (k) (k)

Pn(k) = 2Pn−1 + Pn−2 + · · · + Pn−k for n  2,

(k) (k) (k) (k)

with initial conditions P−(k−2) = P−(k−3) = · · · = P0 = 0 and P1 = 1. Note that the usual Pell sequence
is obtained for k = 2, that is, P (2) = (Pn )∞
n=0 .
More precisely, we determine all the solutions of the Diophantine equation
 
Pn(k) = 2p−1 2p − 1 , (1.1)

where 2p − 1 is prime, and n, k are positive integers with k  2.

Our main result is the following:
Theorem 1. There are no even perfect k-Pell numbers.
It is important to mention that there are many important divisibility properties in the theory of binary
recurrent sequences, and several of them were strongly used in [4, 9] to study the perfect numbers in the
Fibonacci and Pell sequences. Unfortunately, divisibility properties similar to those used in [4,9] are not known
for linear recurrence sequences of higher order, and therefore it is necessary to attack the problem differently.
Our proof uses linear forms in logarithms and the Baker–Davenport reduction procedure. It also uses
properties of generalized Pell sequences and some estimates involving generalized Pell numbers.

2 Preliminary results
2.1 The k-Pell sequence
The first interesting fact about the k-Pell sequence, showed by Kiliç [7], is that the first k + 1 nonzero terms
in P (k) are Fibonacci numbers with odd index, namely,

Pn(k) = F2n−1 for all 1  n  k + 1, (2.1)

(k)
whereas the next term is Pk+2 = F2k+3 −1. Bravo, Herrera, and Luca [3] also investigated the k-Pell sequence
and generalized some well-known properties of (Pn )∞
n=0 to the sequence P
(k) . They proved, for instance, the
 Even perfect numbers in generalized Pell sequences 3

following nice formula involving Fibonacci and generalized Pell numbers:


n−k−1
(k)
Pn(k) = F2n−1 − F2j Pn−k−j for all k  2 and n  k + 2.
j=1

Note that this formula immediately shows that the nth k-Pell number does not exceed the Fibonacci number
with index 2n − 1. In fact,
Pn(k) < F2n−1 for all k  2 and n  k + 2.

It is also known that the characteristic polynomial of P (k) ,

Φk (x) = xk − 2xk−1 − xk−2 − · · · − x − 1,

has just one real root γ := γ(k) outside the unit circle. This means that the single root γ is in fact a Pisot
number (see, e.g., [3]). To simplify notation, we will omit the dependence on k of γ whenever no confusion
may arise. We denote by γ := γ1 , γ2 , . . . , γk the roots of the characteristic polynomial Φk (x). Since γ is a Pisot
number with minimal polynomial Φk (x), it follows that this polynomial is irreducible over Q[x]. Moreover, it
is also known that γ is exponentially close to ϕ2 , where ϕ denotes the golden section. In fact, it was proved
in [3, Lemma 3.2] that γ is located between ϕ2 (1 − ϕ−k ) and ϕ2 .
Let us define, for an integer k  2, the function
z−1 z−1
gk (z) = = . (2.2)
(k + 1)z 2 − 3kz + k − 1 k(z − 3z + 1) + z 2 − 1
2

If we consider the function gk (x) defined in (2.2) as a function of a real variable, then it is not difficult to see
that gk (x) has a vertical asymptote in
√
3k + 5k 2 + 4
c2 (k) := (2.3)
2(k + 1)

and is positive and continuous in (c2 (k), +∞). The following elementary lemma will be needed for some
estimates in the next subsection.
Lemma 1. Keep the above notation and let k  2 be an integer. Then
 
0.276 < gk (γ) < 0.5 and gk (γi ) < 1 for 2  i  k.

Proof. The proof of the first part can be found in [3, Lemma 3.2]. For the second part, that is, for 2  i  k,
we consider the function
 
hk (x) = (x − 1)Φk (x) = xk−1 x2 − 3x + 1 + 1. (2.4)

Evaluating expression (2.4) at γi and rearranging some terms of the resulting expression, we get the relation
γi2 − 3γi + 1 = −1/γik−1 , and so
  k
k γi2 − 3γi + 1 + γi2 − 1 = γi2 − 1 − .
γik−1
Hence
 
  2    k  2  k  2 
k γi − 3γi + 1 + γi − 1 = 
2   
 γ k−1 − γi − 1   |γi |k−1 − γi − 1 > k − 2,
i

Lith. Math. J., Online First, 2020

4 J.J. Bravo and J.L. Herrera

where we used the fact that |γi | < 1 because 2  i  k. Consequently,

  |γi − 1| 2
gk (γi ) = <  1 for all k  4.
|k(γi2 2
− 3γi + 1) + γi − 1| k−2

Finally, the cases k = 2 and 3 can be checked computationally. 



The following “Binet-like” formula for P (k) appears in [3]:


k
Pn(k) = gk (γi )γin . (2.5)
i=1

It was also proved in [3] that the contribution of the roots that are inside the unit circle to formula (2.5) is very
small, namely, we have the approximation
 (k) 
Pn − gk (γ)γ n  < 1 for all n  2 − k. (2.6)
2
From (2.6) we can write
  1
Pn(k) = gk (γ)γ n + ek (n), where ek (n) < . (2.7)
2
We will use estimate (2.7) later. Furthermore, in [3], the inequality
(k)
γ n−2  Pk  γ n−1 for all n  1 (2.8)

is shown, extending a result known for the usual Pell numbers.

We finish this subsection by giving an important estimate of the dominant term in the Binet-like formula
for P (k) . This estimate, based on the fact that the dominant root of the k-Pell sequence is exponentially close
to ϕ2 , will be the key point in addressing the large values of k, and we believe that it is of independent interest.
Lemma 2. Let γ be the dominant root of the characteristic polynomial Φk (x) of the k-Pell sequence. If k  30
and n > 1 are integers satisfying n < ϕk/2 , then

ϕ2n 4
gk (γ)γ n = (1 + ζ), where |ζ| < .
ϕ+2 ϕk/2

Proof. Let λ > 0 be such that γ + λ = ϕ2 . Since γ is located between ϕ2 (1 − ϕ−k ) and ϕ2 , we get that
λ < ϕ2 − ϕ2 (1 − ϕ−k ) = 1/ϕk−2 , that is, λ ∈ (0, 1/ϕk−2 ). Besides,
 
 2 n 2n λ n
= ϕ2n en log(1−λ/ϕ )  ϕ2n e−2λn/ϕ ,
2 2
γ = ϕ −λ =ϕ
n
1− 2
ϕ

where we used the inequality log(1 − x)  −2x for x ∈ (0, 1/2). We also have that e−x  1 − x for all x ∈ R,
and thus γ n  ϕ2n (1 − 2λn/ϕ2 ). Moreover,

2λn 2n 2ϕk/2 2
2
< k
< k
= k/2 .
ϕ ϕ ϕ ϕ
Hence
 
2n 2
n
γ >ϕ 1− .
ϕk/2
 Even perfect numbers in generalized Pell sequences 5

It then follows that the following inequality holds:

2ϕ2n
ϕ2n − < γ n < ϕ2n ,
ϕk/2
that is,
 n  2n
γ − ϕ2n  < 2ϕ . (2.9)
ϕk/2

Using now the mean-value theorem, we get that there exists θ ∈ (γ, ϕ2 ) such that
   
gk (γ) = gk ϕ2 + γ − ϕ2 gk (θ). (2.10)

It is a simple matter to show that

sk (x)
gk (x) = − ,
tk (x)
where
sk (x) = (k + 1)x2 − 2(k + 1)x + 2k + 1
and
 2
tk (x) = (k + 1)x2 − 3kx + k − 1 .

Since the discriminant of sk (x) is negative and x = 1 is the only critical point of sk (x), we deduce that
sk (x) > 0 for all x ∈ R and is increasing in (1, ∞). Additionally, it is not difficult to see that tk (x) is
increasing in (c2 (k), ∞), where c2 (k) is given by (2.3). From the above and using the fact that c2 (k) <
ϕ2 − 1/k < γ < θ < ϕ2 (see [3, Lemma 3.2(c)]), we obtain, after some elementary algebra, that
   2  2
2 1 2ϕ 1 1 2ϕ 1
tk (θ)  tk ϕ − = ϕ− − + 2 +2  ϕ− − + 2 > 12
k k k k k k

for all k  30. As a consequence,

   sk (θ) sk (ϕ2 ) (ϕ + 2)k + ϕ + 1

g (θ) =  = <k
k
tk (θ) 12 12

for all k  30. Hence by (2.10) we get

        k
gk (γ) − gk ϕ2  = γ − ϕ2 gk (θ) = λgk (θ) < . (2.11)
ϕk−2
Writing
 
γ n = ϕ2n + δ and gk (γ) = gk ϕ2 + η,

inequalities (2.9) and (2.11) yield

2ϕ2n k
|δ| < and |η| < .
ϕk/2 ϕk−2

Lith. Math. J., Online First, 2020

6 J.J. Bravo and J.L. Herrera

Besides, since gk (ϕ2 ) = 1/(ϕ + 2), we have

     ϕ2n
gk (γ)γ n = gk ϕ2 + η ϕ2n + δ = (1 + ζ),
ϕ+2
where
δ (ϕ + 2)ηδ
ζ= 2n
+ (ϕ + 2)η + .
ϕ ϕ2n

Finally, we note that

2 (ϕ + 2)k 2(ϕ + 2)k 4

|ζ| < + + 3k/2−2 < k/2 for all k  30.
ϕk/2 ϕk−2 ϕ ϕ

This finishes the proof of the lemma. 



2.2 Linear forms in logarithms

Let η be an algebraic number of degree d with minimal primitive polynomial over the integers


d
 
d
a0 x + a1 x d−1
+ · · · + ad = a0 x − η (i) ,
i=1

where the leading coefficient a0 is positive, and the η (i) are the conjugates of η . Then

1 
d
   
h(η) = log a0 + log max η (i) , 1
d
i=1

is called the logarithmic height of η . In particular, if η = p/q is a rational number with gcd(p, q) = 1 and
q > 0, then h(η) = log max{|p|, q}.
The following are basic properties of the logarithmic height. For α, β algebraic numbers and s ∈ Z, we
have
• h(α ± β)  h(α) + h(β) + log 2;
• h(αβ ±1 )  h(α) + h(β);
• h(αs ) = |s|h(α).
Then we can state the following theorem [12].
Theorem 2 [Matveev’s theorem]. Let K be a number field of degree D over Q, let γ1 , . . . , γt be positive real
numbers of K, and let b1 , . . . , bt be rational integers. Put

Λ := γ1b1 · · · γtbt − 1 and B  max |b1 |, . . . , |bt | .

Let Ai  max{Dh(γi ), | log γi |, 0.16} be real numbers for i = 1, . . . , t. Then, assuming that Λ = 0, we have
 
|Λ| > exp −1.4 · 30t+3 · t4.5 · D 2 (1 + log D)(1 + log B)A1 · · · At .

We now give an estimate for the logarithmic height of the algebraic number gk (γ), which will be used later.
Lemma 3. For k  2, we have that h(gk (γ)) < 4k log ϕ + k log(k + 1).
 Even perfect numbers in generalized Pell sequences 7

Proof. Put

k
 
fk (x) = x − gk (γi ) ∈ Q[x].
i=1

Then the leading coefficient a0 of the minimal polynomial of gk (γ) of degree d over the integers divides
k 2
i=1 ((k + 1)γi − 3kγi + k − 1). But
   
k
  k
  
 2 k
 (k + 1)γi − 3kγi + k − 1  = (k + 1)  c1 (k) − γi c2 (k) − γi ,
   
i=1 i=1

where
 √ √ 
  3k − 5k 2 + 4 3k + 5k 2 + 4
c1 (k), c2 (k) := , .
2(k + 1) 2(k + 1)
Since
 
Φk (y) < max y k , 1 + y + · · · + y k−2 + 2y k−1 < ϕ2k for all 0 < y < ϕ2
and 0 < c1 (k) < c2 (k) < ϕ2 , which are easily seen, it follows that a0 < ϕ4k (k + 1)k . By using this and
Lemma 1 we obtain

  1 
d
 
h gk (γ) = log a0 + log max gk (γi ), 1 < 4k log ϕ + k log(k + 1). 

d
i=1

2.3 Reduction tool

During the calculations, we get too large upper bounds on our variables, and thus we need to reduce them.
To do this, we use the following lemma, which is a slight variation of a result due to Dujella and Pethő [5]
and itself is a generalization of a result of Baker and Davenport [1]. We will use the version given by Bravo,
Gómez, and Luca [2].

Lemma 4. Let A, B , τ , μ be positive real numbers, and let M be a positive integer. Suppose that p/q is
a convergent of the continued fraction expansion of the irrational τ such that q > 6M . Put ε := μq −
M τ q , where · denotes the distance from the nearest integer. If ε > 0, then there is no positive integer
solution (u, v, w) to the inequality
0 < |uτ − v + μ| < AB −w
subject to the restrictions
log(Aq/ε)
uM and w .
log B

3 Proof of Theorem 1

3.1 An initial relation

Assume throughout that (n, k, p) is an integer solution of Eq. (1.1). Suppose further that k  3 since the case
k = 2 was already solved by Bravo and Luca [4]. Note that if 1  n  k + 1, then Eq. (1.1) has no solution
by (2.1) and the main result in [9]. Thus we can assume that n  k + 2, and so n  5.

Lith. Math. J., Online First, 2020

8 J.J. Bravo and J.L. Herrera

On the other hand, since 2 < γ < ϕ2 for all k  3, by (1.1) and (2.8) we have
 
22(p−1) < 2p−1 2p − 1 = Pn(k)  γ n−1 < ϕ2(n−1)
and
 
2n−2 < γ n−2  Pn(k) = 2p−1 2p − 1 < 22p−1 .

Thus
 
log ϕ log ϕ 4n
p< n+ 1− < and n < 2p + 1,
log 2 log 2 5
which gives
5p
< n  2p. (3.1)
4

3.2 An inequality for n and p in terms of k

We now use (1.1) once again and (2.7) to obtain
 
gk (γ)γ n − 22p−1  < 1 + 2p−1 < 2p .
2
Dividing across by 22p−1 , we get that
 
gk (γ)2−(2p−1) γ n − 1 < 2 . (3.2)
2p
To use the result of Matveev’s theorem 2, we take t := 3 and

γ1 := gk (γ), γ2 := 2, γ3 := γ, b1 := 1, b2 := −(2p − 1), b3 := n.

We also take K = Q(γ), which has the degree D := k. Note that the left-hand side of (3.2) is nonzero, since
otherwise gk (γ)γ n = 22p−1 , and so by (1.1) and (2.7) we get 2p−1 < 1/2, which is a contradiction.
On the other hand, by Lemma 3 we have h(γ1 ) < 4k log ϕ + k log(k + 1) < 4k log k for all k  3. Thus we
can take A1 := 4k2 log k. Since h(γ2 ) = log 2 and h(γ3 ) = (log γ)/k < (2 log ϕ)/k, we take A2 := k log 2
and A3 := 2 log ϕ. Further, by (3.1) we obtain that 2p − 1 < (8n/5) − 1 < 2n − 1, and so we can take
B := 2n − 1. Then, Matveev’s theorem, together with a straightforward calculation, gives
   
gk (γ)2−(2p−1) γ n − 1 > exp −1.6 · 1012 k5 log2 k log n , (3.3)

where we used that 1 + log k < 2 log k for k  3 and 1 + log(2n − 1) < 2 log n for n  5. Comparing (3.2)
and (3.3) and using (3.1), we arrive at

n  2p < 4.7 · 1012 k5 log2 k log n,

leading to
n
< 4.7 · 1012 k5 log2 k. (3.4)
log n

To get an upper bound for n depending on k, we next use the fact that x/ log x < A implies x < 2A log A
whenever A  3. Indeed, taking x := n and A := 4.7 · 1012 k5 log2 k and using the inequality 29.2 +
5 log k + 2 log log k < 32 log k for k  3, from (3.4) we get n < 3.1 · 1014 k5 log3 k. Hence by (3.1) we get
p < 4.8 · 1014 k5 log3 k. We summarize this in the following lemma.
 Even perfect numbers in generalized Pell sequences 9

Lemma 5. If (n, k, p) is a solution of Eq. (1.1) with k  3 and n  k + 2, then

n < 3.1 · 1014 k5 log3 k and p < 4.8 · 1014 k5 log3 k.

3.3 The case of large k

Suppose that k > 400. In this case, we have the inequalities

n < 3.1 · 1014 k5 log3 k < ϕk/2 .

At this point, we require the estimate from Lemma 2 to prove that Eq. (1.1) has no solutions for large values
of k. Indeed, since n < ϕk/2 , Lemma 2, combined with (1.1) and (2.7), gives
 2n   
 ϕ   (k)  ϕ2n ζ 
 2p−1   2p−1
ϕ + 2 − 2  =  Pn − 2 − ek (n) −
ϕ + 2
 
2(p−1) ϕ2n ϕ + 2 4
<ϕ + + k/2 ,
ϕ + 2 2ϕ2n ϕ

where in the last inequality we used the fact 2 < ϕ2 . Dividing this inequality by ϕ2n /(ϕ + 2), we get
 
1 − (ϕ + 2)ϕ−2n 22p−1  < (ϕ + 2)ϕ2(p−n−1) + ϕ + 2 + 4 . (3.5)
2ϕ2n ϕk/2
Note that 2(p − n − 1) < −(2 + 2n/5) by the left-hand side of inequality (3.1). By using this and (3.5) we
obtain
 
1 − (ϕ + 2)ϕ−2n 22p−1  < ϕ + 2 + ϕ + 2 + 4
ϕ2 ϕ2n/5 2ϕ2n ϕk/2
2 2 4
< 2n/5 + 2n + k/2 ,
ϕ ϕ ϕ
and since n  k + 2, we get
  8
1 − (ϕ + 2)ϕ−2n 22p−1  < . (3.6)
ϕ2k/5
Here we again apply Matveev’s result on the left-hand side of (3.6) with the data t := 3 and
γ1 := ϕ + 2, γ2 := ϕ2 , γ3 := 2, b1 := 1, b2 := −n, b3 := 2p − 1.
√
Note that K = Q( 5) has the degree D := 2 and contains γ1 , γ2 , γ3 . The left-hand side of (3.6) is not zero.
Indeed,
√ if this were zero, then we would get that ϕ2n = (ϕ + 2)22p−1 . Conjugating the above relation in
Q( 5), we get ϕ = (ϕ + 2)22p−1 , and thus F2n = 22p−1 . But this is possible only if n = 3 and p = 2,
2n
contradicting our assumption that n  5.
Note also that h(γ1 ) = (log 5)/2, h(γ2 ) = log ϕ, and h(γ3 ) = log 2. Hence we can take A1 := log 5,
A2 := 2 log ϕ, and A3 := 2 log 2. Further, recalling (3.1), we can also take B := 2n − 1. Then, Matveev’s
theorem gives
   
1 − (ϕ + 2)ϕ−2n 22p−1  > exp −4.2 · 1012 log n .

By comparing this with (3.6) we get k < 2.2 · 1013 log n. From this, using the upper bound for n from
Lemma 5, we obtain, after some calculations, that k < 2.7 · 1014 log k. In the above, we used the inequality
34 + 5 log k + 3 log log k < 12 log k for k > 270. This gives k < 1016 , and by Lemma 5 and (3.1) we get
n < 1.6 · 1099 and p < 1.3 · 1099 . We record what we have proved so far as a lemma.

Lith. Math. J., Online First, 2020

10 J.J. Bravo and J.L. Herrera

Lemma 6. If (n, k, p) is a solution of Eq. (1.1) with k > 400, then

k < 1016 , n < 1.6 · 1099 , and p < 1.3 · 1099 .

3.4 Lowering the bounds

We need to find a better bound on k and then those implied by Lemma 5 and (3.1) for n and p, respectively. We
begin with the following simple facts of the exponential function. We list it as a lemma for further reference.
Lemma 7. For any nonzero real number x, we have the following:
(a) 0 < x < |ex − 1|;
(b) If x < 0 and |ex − 1| < 1/2, then |x| < 2 |ex − 1|.
We put
z := (2p − 1) log 2 − 2n log ϕ + log(ϕ + 2)

and observe that (3.6) can be written as

 z  8
e − 1 < .
ϕ2k/5
Note that z = 0. If z > 0, then we can apply Lemma 7(a) to obtain |z| < |ez − 1| < 8/ϕ2k/5 . If, on
the contrary, z < 0, then |ez − 1| < 8/ϕ2k/5 < 1/2 for all k > 400. Thus by Lemma 7(b) we have that
|z| < 2|ez − 1| < 16/ϕ2k/5 . In any case, we get that |z| < 16/ϕ2k/5 for all k > 400. Replacing z in the above
inequality by its formula and dividing it across by 2 log ϕ, we get that
   
 log 2 log(ϕ + 2) 
0 < (2p − 1) −n+  < 17 . (3.7)
2 log ϕ 2 log ϕ  ϕ2k/5

Putting
log 2 log(ϕ + 2)
τ := , μ := , A := 17, and B := ϕ2/5 ,
2 log ϕ 2 log ϕ
inequality (3.7) implies
 
0 < (2p − 1)τ − n + μ < AB −k . (3.8)
Note that τ clearly is an irrational number. We also put M := 2.6 · 1099 , which is an upper bound on 2p − 1
by Lemma 6. Applying Lemma 4 to inequality (3.8), we obtain k < 1230. Then Lemma 5 tells us that
2p − 1 < 9.8 · 1032 . With this new upper bound for 2p − 1 we repeated the process, that is, we now take
M := 9.8 · 1032 to obtain k < 440. A third application gives k < 400, contradicting our assumption that
k > 400. We thus deduce that the possible solutions (n, k, p) of Eq. (1.1) all have k ∈ [3, 400].

3.5 The case of small k

Suppose now that k ∈ [3, 400]. Let us now work a little bit on (3.2) to find an improved upper bound on p. For
this purpose, let
 
z := n log γ − (2p − 1) log 2 + log gk (γ) .
Then from (3.2) we have
 z 
e − 1 < 2 .
2p
 Even perfect numbers in generalized Pell sequences 11

Here the same argument used for proving (3.7) shows that
 
 log γ log(gk (γ)) 

0 < n − (2p − 1) + < 6 · 2−p . (3.9)
log 2 log 2 
Putting now
log γ log(gk (γ))
τ = τ (k) := , μ = μ(k) := , A := 6, and B := 2,
log 2 log 2

inequality (3.9) implies

 
0 < nτ − (2p − 1) + μ < AB −p . (3.10)

It is clear that τ is an irrational number because τ > 1 is a unit in OK , the ring of integers of K = Q(γ).
So γ and 2 are multiplicatively independent. We also put Mk := 3.1·1014 k5 log3 k , which is an upper bound
on n from Lemma 5. Applying Lemma 4 to inequality (3.10) for all k ∈ [3, 400], we obtain that

log(Aq/ε)
p< ,
log B

where q = q(k) > 6Mk is the denominator of a convergent of the continued fraction of τ such that ε =
ε(k) = μq − Mk τ q > 0. A computer search with Mathematica revealed that if k ∈ [3, 400], then the
maximum value of log(Aq/ε)/ log B is  107. Hence p  107, and since 2p − 1 is prime, we get that
p ∈ {3, 5, 7, 13, 17, 19, 31, 61, 89, 107}. Furthermore, by (3.1), k + 2  n  2p  214, and then k  212.

3.6 The final computation

As we saw in the preceding subsection, it suffices to look for solutions to Eq. (1.1) in the range

3  k  212 and k + 2  n  214 with p ∈ {3, 5, 7, 13, 17, 19, 31, 61, 89, 107}.

(k)
What we did is generating Pn for all 3  k  212 and k + 2  n  214. Let L1 be the set of such numbers.
Next, we created a new list L2 with the numbers 2p−1 (2p − 1) for p ∈ {3, 5, 7, 13, 17, 19, 31, 61, 89, 107}.
Finally, we computed L1 ∩ L2 and checked that there are no solutions to Eq. (1.1) in this range. This completes
the analysis in the case k ∈ [3, 400] and therefore the proof of Theorem 1.

Acknowledgments. We thank the referee for suggestions which improved the quality of this paper. J.L. Her-
rera thanks the Universidad del Cauca for support during his Ph.D. studies.

References
1. A. Baker and H. Davenport, The equations 3x2 − 2 = y 2 and 8x2 − 7 = z 2 , Q. J. Math., Oxf. Ser., 20(2):129–137,
1969.
2. J.J. Bravo, C.A. Gómez, and F. Luca, Powers of two as sums of two k-Fibonacci numbers, Miskolc Math. Notes,
17(1):85–100, 2016.
3. J.J. Bravo, J.L. Herrera, and F. Luca, On a generalization of the Pell sequence, Math. Bohem., 2020, available from:
https://doi.org/10.21136/MB.2020.0098-19.
4. J.J. Bravo and F. Luca, Perfect Pell and Pell–Lucas numbers, Stud. Sci. Math. Hung., 56(4):381–387, 2019.

Lith. Math. J., Online First, 2020

12 J.J. Bravo and J.L. Herrera

5. A. Dujella and A. Pethő, A generalization of a theorem of Baker and Davenport, Q. J. Math., Oxf. II Ser., 49(195):
291–306, 1998.
6. V. Facó and D. Marques, Even perfect numbers among generalized Fibonacci sequences, Rend. Circ. Mat. Palermo,
63:363–370, 2014.
7. E. Kiliç, On the usual Fibonacci and generalized order-k Pell numbers, Ars Comb., 88:33–45, 2008.
8. T. Koshy, Fibonacci and Lucas Numbers with Applications, Pure Appl. Math., Wiley-Intersci. Ser. Texts Monogr.
Tracts, John Wiley & Sons, New York, 2001.
9. F. Luca, Perfect Fibonacci and Lucas numbers, Rend. Circ. Mat. Palermo, 49:313–318, 2000.
10. F. Luca and F. Nicolae, φ(Fn ) = Fm , Integers, 9:A30, 2009.
11. F. Luca and P. Stanica, Equations with arithmetic functions of Pell numbers, Bull. Math. Soc. Sci. Math. Roum.,
Nouv. Sér., 57(105)(4):409–413, 2014.
12. E.M. Matveev, An explicit lower bound for a homogeneous rational linear form in the logarithms of algebraic num-
bers. II, Izv. Ross. Akad. Nauk, Ser. Mat., 64(6):125–180 (in Russian), 2000. English transl.: Izv. Math., 64(6):1217–
1269, 2000.