See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/308184960

Diophantine triples of Fibonacci numbers

Article in Acta Arithmetica · January 2016

DOI: 10.4064/aa8259-6-2016

CITATIONS READS

10 767

3 authors:

Bo He Florian Luca
Aba Teachers University, China University of the Witwatersrand
46 PUBLICATIONS 372 CITATIONS 922 PUBLICATIONS 7,318 CITATIONS

SEE PROFILE SEE PROFILE

Alain Togbé
Purdue University Northwest
221 PUBLICATIONS 1,596 CITATIONS

SEE PROFILE

All content following this page was uploaded by Bo He on 28 September 2016.

The user has requested enhancement of the downloaded file.

 ⇤†
Diophantine triples of Fibonacci Numbers

B. He, F. Luca and A. Togbé

April 13, 2016

Abstract
Let Fm be the mth Fibonacci number. We prove that if F2n Fk +1 and F2n+2 Fk +1
are both perfect square, then k = 2n + 4, for n 1 or k = 2n 2, for n 2, except
when n = 2 case in which we can additionally have k = 1.

1 Introduction
As we know, the sequence of Fibonacci numbers {Fn }n 1 is given by

F1 = 1, F2 = 1, Fn+2 = Fn+1 + Fn , n 1.

There are many properties of Fibonacci numbers. A Diophantine m-tuple is a set of m pos-
itive integers {a1 , . . . , am } such that ai aj + 1 is a perfect square for all i 6= j in {1, . . . , m}.
In 1977, Hoggat and Bergum [4] proved that {F2n , F2n+2 , F2n+4 , 4F2n+1 F2n+2 F2n+3 } is a
Diophantine quadruple. In 1999, Dujella [2] proved that if d is a positive integer such that
{F2n , F2n+2 , F2n+4 , d} is a Diophantine quadruple, then d = 4F2n+1 F2n+2 F2n+3 . Here, we
take this one step further, by fixing a positive integer n and looking at positive integers
k such that {F2n , F2n+2 , Fk } is a Diophantine triple. Our result is the following.
Theorem 1. If {F2n , F2n+2 , Fk } is a Diophantine triple, then k 2 {2n + 4, 2n 2}, except
when n = 2, case in which we have the additional solution k = 1.
Note that the exception k = 1 in case n = 2 is not truly an exception but it appears
merely due to the fact that F1 = F2 .

2 Intersection of two sequences

For any fixed positive integer n, assume that there exist positive integers k, x, y such that

F2n Fk + 1 = x2 , F2n+2 Fk + 1 = y 2 . (1)

⇤
2010 Mathematics subject Classification: 11D09, 11D45, 11B37, 11J86.
†
Keywords: Diophantine m-tuple, Fibonacci numbers, Linear forms in logarithms

1
Eliminating Fk from the two relations above, we deduce the norm form equation

F2n · y 2 F2n+2 · x2 = F2n F2n+2 . (2)

Since F2n < F2n+2 < 4F2n , by Theorem 8 of [5], we have

p p p p p
y F2n + x F2n+2 = (± F2n + F2n+2 )(F2n+1 + F2n F2n+2 )j , j 0.

Define p p
V j + Uj F2n F2n+2 := (F2n+1 + F2n F2n+2 )j .
Then, we obtain
x = xj = Vj ± F2n Uj . (3)
Substituting the above formula (3) into the first equation of (1), we get

Fk = ±2Vj Uj + (F2n + F2n+2 )Uj2 . (4)

This is the main equation to be solved. Let

(±)
Cj := ±2Vj Uj + (F2n + F2n+2 )Uj2 , for j = 1, 2, . . . . (5)

Therefore, we have to solve the equation

(±)
Cj = Fk , (6)

for some positive integers j and k. Notice that the above equation has solutions such as
(+) ( )
C1 = F2n+4 and C1 = F2n 2 , (7)

which are the ones appearing in the statement of Theorem 1. We need to show that there
( )
are no other solutions (except when n = 2, for which C1 = F2·2 2 = F2 = F1 ). So, we
shall assume that j 2 in order to get a contradiction.
Let p p
1+ 5 1 5
↵= , and ↵
¯= .
2 2
We use the well-known Binet formula
↵k ↵¯k
Fk = p , for all k 1. (8)
5
We put q
2
n := F2n+1 + F2n+1 1.
We also put
j j j j
n + n
Vj := , and Uj = pn 2 n . (9)
2 2 F2n+1 1

2
Technically, Vj and Uj depend on n, but we shall assume that n is fixed throughout the
argument. Define
(±) 1 F2n + F2n+2
n := ± p 2 + 2
. (10)
2 F2n+1 1 4(F2n+1 1)
Formula (5) leads to:
2j 2j 2j
(±) 2 + n 2j
Cj = ± pn 2 n + (F2n + F2n+2 ) · n 2
2 F2n+1 1 4(F2n+1 1)
F2n + F2n+2
= n2j n(±) 2
+ n 2j n(⌥) . (11)
2(F2n+1 1)

Therefore, by (8) and (11), equation (6) becomes

2j (±) F2n + F2n+2 2j (⌥) ↵k ↵¯k

n n 2
+ n n = p . (12)
2(F2n+1 1) 5

3 A linear form in three logarithms

Lemma 1. We have
2n (+) 2n
(i) 1.46↵ < n < 1.66↵ ;
2n ( ) 2n
(ii) 0.08↵ < n < 0.13↵ .
(±)
Proof. From the definition of n , we deduce
0 1
q
(±) 1 1 @q 1 1 A.
2 n =p ±p = 51/4 ↵ n
± q (13)
F2n F2n+2 1 1
↵ 1 1
↵4n ↵4n+4

1/2
The Taylor series of (1 x) is
1 1 3 5
p = 1 + x + x2 + x3 + · · · ,
1 x 2 8 16

which implies
1 1 x
1+ x< p <1+ , for x 2 (0, 1).
2 1 x 2(1 x)

Therefore, we have
1 1 1 1
1± <q ± q < 1.1 ± . (14)
↵ 1 1
↵ 1 1 ↵
↵4n ↵4n+4

3
Using (13) and (14), we get
q
(±)
1 2 n 1
1 ± < 1/4 n < 1.1 ± .
↵ 5 ↵ ↵
Straightforward calculations give the results (i) and (ii) in the lemma.
Now, we define the following linear form in three logarithms:
p
⇤ := 2j log n k log ↵ + log( 5 · n(±) ). (15)
The next result determines an upper bound for ⇤.
2j
Lemma 2. If j 2, then 0 < ⇤ < 66 n .
Proof. Using equation (12), we have
↵k F +F ¯k
↵
2j (±)
n n p = 2n 2 2n+2 n
2j (⌥)
n p .
5 2(F2n+1 1) 5
2j (±)
p
Suppose first that n n  ↵k / 5. Then
p 2j 2j
5 n n
 (±)
 ( )
,
↵k n n

and
1 1 1 F2n + F2n+2 2j (⌥) ¯k
↵ 2j (+) 1
< + = 2
< n n +p  n n +p ,
F2n+2 2F2n 2F2n+2 2(F2n+1 1) 5 5 · ↵k
imply !
1 2j (+) 1
< n n + ( )
. (16)
F2n+2 5 n

This last inequality (16) and Lemma 1 give

!
j j 1
4j F2n F2n+2 < 2j
n < F2n+2 (+)
n + ( )
< F2n+2 (1.66↵ 2n
+ 2.5↵2n ),
5 n

so
j j 1
4j F2n F2n+2 < 1.66↵ 2n
+ 2.5↵2n . (17)
The above inequality (17) implies easily that j < 2. This contradicts the assumption. So,
(±) p
we have n2j n > ↵k / 5. Therefore, ⇤ > 0. Moreover, as
2j (±) 1 F2n + F2n+2 1 1 1
|↵k 5 1/2
n n 1| < 2
· 2j (±)
< · 2j ( )
< 33 n
2j
2(F2n+1 1) n n F2n n n
2j
and the right–most quantity above is < 1/2, we deduce that ⇤ < 66 n . Here, we used
the fact that
|⇤| < 2|e⇤ 1|, whenever |e⇤ 1| < 1/2. (18)
This completes the proof of the lemma.

4
 For anyQnon-zero algebraic number of degree d over Q, whose minimal polynomial
over Z is a dj=1 X (j)
, we denote by
d
!
1 X
(j)
h( ) = log a + log max 1,
d j=1

its absolute logarithmic height. We need the following result due to Matveev [8].
Lemma 3. Let ⇤ be a linear form in logarithms of N multiplicatively independent totally
real algebraic numbers ↵1 , . . . , ↵N with rational integer coefficients b1 , . . . , bN (bN 6= 0).
Let h(↵j ) denote the absolute logarithmic height of ↵j , for 1  j  N . Define the numbers
D, Aj (1  j  N ) and E by D := [Q(↵1 , . . . , ↵N ) : Q], Aj := max{Dh(↵j ), | log ↵j |},
E := max{1, max{|bj |Aj /AN ; 1  j  N }}. Then,

log |⇤| > C(N )C0 W0 D2 ⌦,

where
8
C(N ) := (N + 2)(2N + 3)(4e(N + 1))N +1 ,
(N 1)!
C0 := log(e4.4N +7 N 5.5 D2 log(eD)),
W0 := log(1.5eED log(eD)), ⌦ = A1 · · · AN .
In order to use Lemma 3 to the linear form in three logarithms
p
⇤ = 2j log n k log ↵ + log( 5 · n(±) ),

we take
p (±)
N = 3, D = 4, b1 = 2j, b2 = k, b3 = 1 and ↵1 = n, ↵2 = ↵, ↵3 = 5· n .

We needp to justify that ↵1 , p↵2 , ↵3 are multiplicatively independent. But note that
↵2 2 Q( 5) and ↵1 , ↵32 2 Q( F2n F2n+2 ). Let us show that F2n F2n+2 is neither a square,
nor 5 times a square. Indeed, otherwise, since gcd(F2n , F2n+2 ) = Fgcd(2n,2n+2) = F2 = 1, we
would get that one of F2n or F2n+2 is a square. It is well-known that the only squares in the
Fibonacci sequence are 1 and 144, leading to n = 1, 5, 6, but none of F2 F4 , F10 F12 , F12 F14
is a square or 5 times a square. Thus, if we write F2n F2n+2 = du2 for an integer u and
a square-free integer d, then d > 1 and d 6= 5. So, if ↵1 , ↵2 , ↵3 are multiplicatively
dependent, then ↵1 and ↵32 are multiplicatively
p dependent (because no p power of ↵2 of
a nonzero integer exponent is in Q( d)). Since ↵1 is a unit in Q( d), we get that
(±) (±)
↵32 = 5( n )2 is a unit, which is false since the norm of 5( n )2 is
4
(+) ( ) 2 25F2n+1
25( n n ) = 4 4
,
256F2n F2n+2

and the above fraction is not an integer for any n 1.

5
 One can see that
1 1
h(↵1 ) = h( n) = log n and h(↵2 ) = h(↵) = log ↵.
2 2
(+) ( )
As n , n are conjugate and roots of the quadratic polynomial
2 2
16F2n F2n+2 X2 2
8(F2n 2
F2n+2 + F2n F2n+2 )X + (F2n+2 F2n )2 ,

and furthermore, ✓ ◆2
(±) (+) 1 1 1
| n | | n | = p +p < 1,
4 F2n F2n+2
we get
1
(±) 2 2
h( n ) 
log 16F2n F2n+2 = log(4F2n F2n+2 ) < log(4/5) + (4n + 2) log ↵,
2
p
where we used the fact that F` < ↵` / 5, for ` 2 {2n, 2n + 2}. This implies that
⇣p p ⌘ 1
(±)
h(↵3 ) = h 5·  h( 5) + h( n(±) ) < log 5 + log(4/5) + (4n + 2) log ↵
n
p 2
= log(4/ 5) + (4n + 2) log ↵ < 4(n + 1) log ↵,
p
where we used the fact that 4/ 5 < ↵2 . We put

A1 = 2 log n, A2 = 2 log ↵, A3 = 16(n + 1) log ↵.

Relabeling for the purpose of computing E the three numbers (so making the substitution
↵2 $ ↵3 ), we get that we can take
⇢
2j log n
E = max , 8(n + 1), k  4j(n + 1).
log ↵

For the last inequality above we used, on the one hand, the fact that ↵` 2
 F`  ↵ ` 1 ,
for all ` 1, to deduce that

n < 2F2n+1 < 2↵2n < ↵2(n+1) ,

because ↵2 > 2, while on the other hand the fact that for n 2, we have

↵k 1
< 2↵k 2
 2Fk  4Uj Vj + 2(F2n + F2n+2 )Uj2
p p
= (Vj + Uj F2n F2n+2 )2 = (F2n+1 + F2n F2n+2 )2j < (2↵2n )2j
< ↵4j(n+1) .

while for n = 1, we have k  6 by a result of Robbins [9].

6
 Using Lemma 2 and Lemma 3, we get
8
C(3) = (3 + 2)(6 + 3)(42 e)4 < 6.45 ⇥ 108 ;
2!
C0 = log e4.4·3+7 35.5 42 log(4e) < 30;
W0 = log(1.5 · 4eE log(4e)) < log(156j(n + 1));
⌦ = (2 log n )(2 log ↵)(16(n + 1) log ↵);
so
2j log n log 66 < log |⇤|
< (6.45 ⇥ 108 ) · 30 · 42 log(156j(n + 1))(2 log n )(2 log ↵)(16(n + 1) log ↵),
leading to
j < 2.3 ⇥ 1012 (n + 1) log (156j(n + 1)) .
We record what we have obtained:
Proposition 1. If equation (4) has a positive integer solution (j, k) with j > 1, then
j < 2.3 · 1012 · (n + 1) · log(156 · j · (n + 1)). (19)

4 A linear form in two logarithms

From (7), when j = 1, one can see that equation (4) has the solutions
(
( )
2n 2, for C = C1 ,
k= (+) (20)
2n + 4, for C = C1 .

This leads us to define

p (±)
⇤0 = 2 log n ((2n + 1) ± 3) log ↵ + log( 5 · n ). (21)
2
Lemma 4. We have |⇤0 | < 66 n

Proof. For n = 1, this can be checked directly. Assume that n 2. Substituting (20)
into (12), we have

2 (±) ↵(2n+1)±3 F2n + F2n+2 2 (⌥) ↵ (2n+1)⌥3

n n p = 2 n n p .
5 2(F2n+1 1) 5
2 (±)
p p
If n n  ↵(2n+1)±3 / 5, then ↵ (2n+1)⌥3 / 5  1/(5 n2 (±)
) and hence
2 (⌥) (2n+1)⌥3
p (⌥) (±)
((2n+1)±3) 1/2 2 (±) n n +↵ / 5 n + 1/(5 n )
|↵ 5 n / n 1| < (±)
< (±)
2 4
n n n n
4n
20.75 + 31.25↵
< 4
.
n

7
 p
This inequality together with n >2+ 3, and n > ↵2n give

|↵((2n+1)±3) 5 1/2
n
2
/ (±)
n 1| < 32.8 n
2
.
2 (±)
p
On the other hand, if n n > ↵(2n+1)±3 / 5, then we have

1/(2F2n ) + 1/(2F2n+2 ) 1
|↵((2n+1)±3) 5 1/2
n
2
/ (±)
n 1| < < < 8.7 n
2
.
2 (±) F2n 2 (±)
n n n n

In both cases, we have

|↵((2n+1)±3) 5 1/2 n
2
/ (±)
n 1| < 32.8 n
2
. (22)
p 2
Since n 2, we have n 5 + 24, so n > 66, and the above inequality (22) implies
2
|⇤0 | < 66 n

via inequality (18).

Let K := (2j 1)(2n + 1) k ± 3 and

⇤1 = K log ↵ (j 1) log(5/4). (23)

4n 2
Lemma 5. We have |⇤1 | < (6j + 192)↵ .

Proof. We know that

q
2 1
n = F2n+1 + F2n+1 1 = 2F2n+1 p 2
F2n+1 + F2n+1 1
!
1
= 2F2n+1 1 p 2 (24)
2F2n+1 (F2n+1 + F2n+1 1)

and ✓ ◆
2 2 2n+1 1
2F2n+1 = p (↵2n+1 ↵
¯ 2n+1
)= p ↵ 1 + 4n+2 .
5 5 ↵
We define !✓ ◆
1 1
n = 1 p 2 1+ .
2F2n+1 (F2n+1 + F2n+1 1) ↵4n+2
From the above, we deduce that
p
log n = log(2/ 5) + (2n + 1) log ↵ + log n.

8
Using (15) and (21), we have
⇤ ⇤0 = (2j 2) log n (k (2n + 1) ⌥ 3) log ↵
2
= (2j 2) log p + (2j 2)(2n + 1) log ↵
5
+(2j 2) log n (k (2n + 1) ⌥ 3) log ↵
= (2j 2) log n + K log ↵ (j 1) log(5/4).
The above calculation and the definition of ⇤1 give
⇤1 = ⇤ ⇤0 (2j 2) log n.

One can see that Lemma 2, Lemma 4 and the inequalities

! ✓ ◆
1 1
| log n |  log 1 p 2 + log 1 + 4n+2
2F2n+1 (F2n+1 + F2n+1 1) ↵
1.2 1 3
< p 2 + 4n+2 < 4n+2 ,
2F2n+1 (F2n+1 + F2n+1 1) ↵ ↵
imply
132 6(j 1)
|⇤1 |  |⇤| + |⇤0 | + |(2j 2) log n| < 2
+ . (25)
n ↵4n+2
Clearly,
s !
1 ⇣ p ⌘ p p
n = F2n+1 1+ 1 2
F2n+1 1 + 3/2 > ↵2n+1 (1 + 3/2)/ 5,
F2n+1
so p
2 4n+2 (1 + 3/2)2 2↵4n+2
n >↵ > . (26)
5 3
From (25) and (26), we get the desired conclusion.
At this point, we recall the following result of Laurent [7].
Lemma 6. Let 1 > 1 and 2 > 1 be two real multiplicatively independent algebraic
numbers, b1 and b2 2 Z not both 0 and
⇤ = b2 log 2 b1 log 1.

Let D := [Q( 1 , 2) : Q]. Let

⇢
|log i | 1
hi max h( i ), , for i = 1, 2
D D
and
|b1 | |b2 |
b0 + .
D h2 D h1
Then ✓ ⇢ ◆2
4 0 30 1
log |⇤| 17.9 · D max log b + 0.38, , h1 h2 .
D 2

9
 To apply Lemma 6 to ⇤1 , we put

D = 2, 1 = 5/4, 2 = ↵, b1 = j 1, b2 = K.

The conditions of Lemma 6 are fulfilled for our choices of parameters. Furthermore, we
can take h1 = log 5, h2 = 1/2. By Lemma 5, we have
4n 2
(j 1) log(5/4) + (6j + 192)↵
K< < 0.47(j 1) + 0.7j + 22.3 < 1.2j + 22.
log ↵
So, we can take
K |b1 | |b2 |
b0 = 1.38j + 6.9 > (j 1) + = + .
2 log 5 D h2 D h1
Therefore, we get, by Lemma 6,

log |⇤1 | > 17.9 · 8 · log 5 · (max{log(1.38j + 6.9) + 0.38, 15})2 . (27)

From Lemma 5, we get

log |⇤1 | < (4n + 2) log ↵ + log(6j + 192). (28)

Combining the two bounds (27) and (28) on log |⇤1 |, we have

n < 120(max{log(1.38j + 6.9) + 0.38, 15})2 + 0.6 log(6j + 192). (29)

If
log(1.38j + 6.9) + 0.38  15,
then

j < 1619963, and n < 120 · 152 + 0.6 log(6 · 1619963 + 192) < 27010.

Otherwise,
n < 120(log(1.38j + 6.9) + 0.38)2 + 0.6 log(6j + 192). (30)
Substituting inequality (30) into Proposition 1, we have

j < 2.3 · 1012 · (120(log(1.38j + 6.9) + 0.38)2 + 0.6 log(6j + 192) + 1)

⇥ log(156j · (120(log(1.38j + 6.9) + 0.38)2 + 0.6 log(6j + 192) + 1)). (31)

A straightforward calculation gives j < 4·1019 , which together with (30) imply n < 252158.
Therefore, we get the following result.

Lemma 7. If equation (4) has a positive integer solution (j, k) with j > 1, then

j < 4 · 1019 and n < 252158.

10
5 Better bounds on j and n
From Lemma 5, we have
4n 2
|K log ↵ (j 1) log(5/4)| < (6j + 192)↵ .

Hence,
log(5/4) K 6j + 192
< . (32)
log ↵ j 1 (j 1)↵4n+2 log ↵
Assume first that
6j + 192 1
< . (33)
(j 1)↵4n+2 log ↵ 2(j 1)2
Then
log(5/4) K 1
< .
log ↵ j 1 2(j 1)2
This implies, by a criterion of Legendre, that K/(j 1) is a convergent in the simple
continued fraction expansion of log(5/4)/(log ↵). We know that

log(5/4)
= [0, 2, 6, 2, 1, 1, 3, 7, 1, 3, 1, 1, 22, 2, 1, 1, 4, 3, 1, 2, 1, 1, 1, 1, 4,
log ↵
1, 12, 6, 1, 1, 4, 1, 8, 2, 1, 49, 1, 10, 6, 1, 1, 3, 1, 1, 1, 5, 22, 1, 1, . . .]. (34)

The denominator of the 46th convergent

25158053660121411107
54253653513327093513
is greater than the upper bound 4 · 1019 on j. The 45th convergent
4460457560349832575
9619031832089360168
provides the lower bound

log(5/4) K 39
> 1.9 · 10 . (35)
log ↵ j 1

From (32) and (35), we get

39 6j + 192 4n 2
1.9 · 10 < < 204↵ (log ↵) 1 ,
(j 1)↵4n+2 log ↵

which implies that n < 49. It is known that (see [6]), if pr /qr is the rth such convergent
of log(5/4)/(log ↵), then

log(5/4) pr 1
> .
log ↵ qr (ar+1 + 2)qr2

11
where ar+1 is the (r + 1)st partial quotient of log(5/4)/(log ↵). We thus have
⇢
1 6j + 192
min 2
< , for 2  r  45. (36)
(ar+1 + 2)(j 1) (j 1)↵4n+2 log ↵
Since max{ar+1 : 2  r  45} = a36 = 49, from the above inequality (36) we get
↵4n+2 < 51(j 1)(6j + 192)(log ↵) 1 .
All this was when inequality (33) holds. On the other hand, if inequality (33) doesn’t
hold, then we have
↵4n+2  2(j 1)(6j + 192)(log ↵) 1 .
Both possibilities give
↵4n+2 < 51(j 1)(6j + 192)(log ↵) 1
< 636j(j + 32) < 20988j 2 .
Therefore, we deduce the following result.
Proposition 2. If equation (4) has positive integer solution (j, k) with j > 1, then
n < 1.04 log j + 4.7. (37)
This bound is better than that in (30). Combining Proposition 1 and Proposition 2,
we get
j < 2.3 · 1012 · (1.04 log j + 5.7) · log(156 · j · (1.04 log j + 5.7)),
which implies that j < 5 · 1015 . Using Proposition 2, we get the following result.
Lemma 8. If equation (4) has a positive integer solution (j, k) with j > 1, then
j < 5 · 1015 and n < 43.

6 The Baker-Davenport reduction method

In order to deal with the remaining cases, for 1  n  42 we used a Diophantine approxi-
mation algorithm called the Baker-Davenport reduction method. The following lemma is
a slight modification of the original version of Baker-Davenport reduction method. (See
[3, Lemma 5a]).
Lemma 9. Assume that  and µ are real numbers and M is a positive integer. Let P/Q
be the convergent of the continued fraction expansion of  such that Q > 6M and let
⌘ = kµQk M · kQk,
where k · k denotes the distance from the nearest integer. If ⌘ > 0, then there is no
solution of the inequality
0 < j k + µ < AB j
in integers j and k with
log (AQ/⌘)
 j  M.
log B

12
 As p (±) 2j
0 < 2j log n k log ↵ + log( 5 · n ) < 66 n ,
we apply Lemma 9 with
p (±)
2 log n log( 5 · n ) 66 2
= , µ= , A= , B= n, M = 5 · 1015 .
log ↵ log ↵ log ↵
The program was developed in PARI/GP running with 200 digits. For the computations,
if the first convergent such that q > 6M does not satisfy the condition ⌘ > 0, then we use
the next convergent until we find the one that satisfies the conditions. In one minute all
the computations were done. In all cases, we obtained j  17. From Proposition 2, we
deduce that n < 8. We set M = 17 to check again in the range 1  n  7. The second
run of the reduction method yields j  6 and then n  6. So we have the following result.
Lemma 10. If equation (4) has a positive integer solution (j, k) with j > 1, then

j6 and n  6.

We are now ready to prove Theorem 1

7 The Proof of Theorem 1

(±)
For 1  n  6, 2  j  6, we compute all Cj . None of them is a Fibonacci number.
This means that equation (4) has no positive integer solution (j, k) with j 2. When
j = 1, we have
(+)
C1 = 2V1 U1 + (F2n + F2n+2 )U12 = 2F2n+1 + F2n + F2n+2 = F2n+4 , for n 1

and
( )
C1 = 2V1 U1 + (F2n + F2n+2 )U12 = 2F2n+1 + F2n + F2n+2 = F2n 2 , for n 2.

Since F1 = F2 = 1, the additional solutions come from the triple {F1 , F4 , F6 } = {F2 , F4 , F6 } =
{1, 3, 8}.

8 Final remark
We close by o↵ering the following conjectures.
Conjecture 1. There are no positive integers p, q, m, and n such that {Fp , Fq , Fm , Fn } is
a Diophantine quadruple.
Conjecture 2. If there exist three positive integers p < q < r such that {Fp , Fq , Fr } is a
Diophantine triple, then (p, q, r) = (2n, 2n + 2, 2n + 4) for some n 1 with the additional
exception (p, q, r) = (1, 4, 6).

13
Acknowledgements
We thank the referee for comments which improved the quality of this manuscript. This
paper started when all authors attended Journées Arithmétiques in Debrecen, Hungary in
July 2015. They thank the organizers for the opportunity to attend this event. The first
author was supported by Natural Science Foundation of China (Grant No. 11301363),
and Sichuan provincial scientific research and innovation team in Universities (Grant No.
14TD0040), and the Natural Science Foundation of Education Department of Sichuan
Province (Grant No. 16ZA0371).

References
[1] J. Arkin, V. E. Hoggatt, and E. G. Strauss, On Euler’s solution of a problem of
Diophantus, Fibonacci Quart. 17 (1979), 333–339.

[2] A. Dujella, A proof of the Hoggatt-Bergum conjecture, Proc. Amer. Math. Soc. 127
(1999), 1999–2005

[3] A. Dujella and A. Pethő, A generalization of a theorem of Baker and Davenport,

Quart. J. Math. Oxford Ser. (2) 49 (1998), 291–306.

[4] V. E. Hoggatt and G. E. Bergum, A problem of Fermat and the Fibonacci sequence,
Fibonacci Quart. 15 (1977), 323–330.

[5] B. W. Jones, A second variation on a problem of Diophantus and Davenport, Fibonacci

Quart. 16 (1978), 155–165.

[6] A. Y. Khinchin, Continued Fractions, P. Noordho↵ Ltd., Gorningen, 1963, 3rd edition.

[7] M. Laurent, Linear forms in two logarithms and interpolation determinants II, Acta
Arith. 133 (4) (2008), 325–348.

[8] E. M. Matveev, An explicit lower bound for a homogeneous rational linear form in
logarithms of algebraic numbers II, Izv. Math 64 (2000), 1217–1269.

[9] N. Robbins, Fibonacci and Lucas numbers of the forms w2 1, w3 ± 1, The Fibonacci
Quarterly 19 (1981), 369–373.

Bo He
Institute of Mathematics
Aba Teachers University
Wenchuan, Sichuan, 623000, P. R. China
and
Department of Mathematics

14
 Hubei University for Nationalities
Enshi, Hubei, 445000, P.R. China
[email protected]

Florian Luca
School of Mathematics
University of the Witwatersrand
Private Bag X3, Wits 2050, South Africa
[email protected]

Alain Togbé
Department of Mathematics
Purdue University North Central
1401 S. U.S. 421
Westville, IN 46391 USA
[email protected]

15

View publication stats