Professional Documents
Culture Documents
i!1
be the
Fibonacci sequence given by 1
1
1. 1
2
1 and
1
i2
1
i1
1
i
for all i ! 1. It is well-known
that 1
2
i
1
2
i1
1
2i1
. Hence, 1
:
i
1
:
i1
is a
Fibonacci number for all i ! 1 when : 2 f1. 2g. In
the paper [4], the following result was proved:
Theorem A. If : ! 1 is an integer such that
1
:
i
1
:
i1
is a Fibonacci number for all suciently
large i, then : 2 f1. 2g.
Note that this doesnt say much about the
Diophantine equation
1
:
i
1
:
i1
1
i
1
in integers i ! 1, i ! 1, : ! 3. It merely says that
if : ! 3 is xed, then there are innitely many
positive integers i for which 1
:
i
1
:
i1
is not a
Fibonacci number. Thus, the main result from [4]
cannot answer to the question of whether the
Diophantine equation (1) has nitely or innitely
many integer solutions i. i, even for xed : ! 3.
Let us briey describe the proof of this result
from [4]. Put c : 1
5
p
2, u : 1
5
p
2.
Then it is known that
1
i
c
i
u
i
5
p holds for all i ! 1. 2
Hence, for xed : we have that
1
:
i
c
i
u
i
5
p
:
c
i:
5
:2
O
:
c
i:2
.
The constant implied by the above O depends on :.
Applying the above estimate with i and i 1 and
adding them up we get
1
:
i
1
:
i1
c
i:
c
i1:
5
:2
O
:
c
i:2
c
i:
1 c
:
5
:2
O
:
c
i:2
.
Comparing the above estimate with 1
i
given by (2)
we get right away that
c
i:i
1 c
:
5
:12
1 O
:
c
i:2i
c
2i
.
and the righthand side above is of smaller order of
magnitude than c
i:i
for large i and i. Hence, the
left hand side must be zero for large i, so we must have
i i: t, where t is some xed integer such that
1 c
:
5
:12
c
t
. 3
From here, the authors of [4] proceed to prove that the
above equation (3) has no integer solutions :. t with
: ! 3 by using linear forms in logarithms and calcu-
lations. We note in passing that there is no need for
linear forms in logarithms in order to solve (3). Indeed,
conjugating it and multiplying the two relations we get
1 c
:
1 u
:
1
:1
5
:1
cu
t
1
:1t
5
:1
.
The lefthand side is positive and smaller than
21 c
:
< 21 2
:
. Hence, we get
21 2
:
5
:1
.
which has no solution for any : ! 3 anyway.
Our main result is the following.
Theorem 1. Equation (1) has no solutions
i. i. : with i ! 2 and : ! 3.
2. The proof of Theorem 1.
2.1. The plan of attack. Our method is
roughly as follows. We use linear forms in loga-
doi: 10.3792/pjaa.87.45
#2011 The Japan Academy
2000 Mathematics Subject Classication. Primary 11B39,
11J86.
5
p 1
:
i1
1
:
i
u
i
5
p . 4
The righthand side above is a number in the
interval 1
:
i
1. 1
:
i
1. In particular, it is posi-
tive. Dividing both sides of equation (4) by 1
:
i1
,
we get
c
i
5
12
1
:
i1
1
< 2
1
i
1
i1
:
<
2
1.5
:
. 5
where we used the fact that 1
i
1
i1
23 for all
i ! 2. This last inequality is equivalent to the
inequality 21
i1
! 31
i
. Replacing the number
1
i1
by 1
i
1
i1
, the last inequality above is
seen to be equivalent to 21
i1
! 1
i
1
i1
1
i2
, which is equivalent to 1
i1
! 1
i2
, which
is true for all i ! 2.
We shall use several times a result of Matveev
(see [5] or Theorem 9.4 in [1]), which asserts that
if c
1
. c
2
. . . . . c
/
are positive real algebraic numbers
in an algebraic number eld K of degree 1, /
1
.
/
2
. . . . . /
/
are rational integers, and
: c
/
1
1
c
/
2
2
c
/
/
/
1
is not zero, then
jj exp C
/.1
1 log 1
1
/
6
where
C
/;1
: 1.4 30
/3
/
4.5
1
2
1 log 1.
1 ! maxfj/
1
j. j/
2
j. . . . . j/
/
jg.
and
i
! maxf1/c
i
. j log c
i
j. 0.16g. i 1. 2. . . . . /.
Here, for an algebraic number i we write /i for its
logarithmic height whose formula is
/i :
1
d
log o
0
d
i1
log maxfji
i
j. 1g
.
with d being the degree of i over Q and
1A : o
0
d
i1
A i
i
2 ZA
being the minimal primitive polynomial over the
integers having positive leading coecient and i as
a root.
In a rst application of Matveevs theorem, we
take / : 3, c
1
: c, c
2
:
5
p
, c
3
: 1
i1
. We also
take /
1
: i, /
2
: 1, /
3
: :. We thus take
1
: c
i
5
12
1
:
i1
1. 7
The absolute value of
1
appears in the lefthand
side of inequality (5). To see that
1
6 0, observe
that imposing that
1
0 yields c
i
5
p
1
:
i1
,
so c
2i
2 Z, which is false for positive i. Thus,
1
6 0.
The algebraic number eld containing
c
1
. c
2
. c
3
is K : Q
5
p
which is quadratic, so
we can take the degree 1 : 2. Since the height of
c
1
satises /c
1
log c2 0.240606 . . . , it
follows that we can take
1
: 0.5 2/c
1
. Since
the height of c
2
satises /c
2
log 52
0.804719 . . . , it follows that we can take
2
:
1.61. Since the inequality 1
< c
1
holds for all
integers ! 1, it follows that /c
3
log 1
i1
<
ilog c, therefore we can take
3
: 2ilog c.
Finally, observe that
c
i1
1
i
1
:
i
1
:
i1
! 1
:
i1
c
i1
:
.
c
i2
< 1
i
1
:
i
1
:
i1
< 1
i
1
i1
:
1
:
i2
< c
i1:
.
In particular, i ! i 1: ! 2:, so we can take
1 : i. It is also the case that
1 i i 1: 1 < i 2:.
46 F. LUCA and R. OYONO [Vol. 87(A),
Applying inequality (6) to get a lower bound for
j
1
j and comparing this with inequality (5), we get
exp C
3.2
1 log i 0.5 1.61 2ilog c
<
2
1.5
:
.
where
C
3.2
1.4 30
6
3
4.5
2
2
1 log 2 < 10
12
.
Hence, we get
: <
log 2
log 1.5
10
12
0.5 1.61 2 log c
log 1.5
1
i1 log i
< 2 10
12
i1 log i < 3 10
12
ilog i.
where we also used the fact that log i ! log 10 2.
Together with the fact that i < i 2:, we get
that
: < 3 10
12
ilogi 2:. 8
2.3. The case of small m. We next treat the
cases when i 2 3. 150. In this case,
: < 3 10
12
ilogi 2: 4.5 10
14
log152:.
giving : < 1.92 10
16
. Thus,
i < i 2: 152: < 3 10
18
.
We next take another look at
1
given by
expression (7). Put
1
: ilog c log
5
p
: log 1
i1
.
Thus,
1
c
1
1.
Observe that
1
is positive since
1
is positive
because the righthand side of equation (4) is
positive. Thus,
0 <
1
< c
1
1
1
<
2
1.5
:
.
so
0 < i
log c
log 1
i1
:
log
5
p
log 1
i1
<
2
1.5
:
log1
i1
<
2
1.5
:
<
2
1.5
1
152
i
. 9
We now apply the following result due to
Dujella and Peth} oo [3].
Lemma 2. Let ` be a positive integer, let
jc be a convergent of the continued fraction of the
irrational such that c 6`, and let j be some real
number. Let : kjck `kck. If 0, then there
is no solution to the inequality
0 < i : j < 1
i
in positive integers i and : with
logc
log 1
i `.
For us, inequality (9) is
0 < i : j < 1
i
.
where
:
log c
log 1
i1
. j:
log
5
p
log 1
i1
.
: 2. 1: 1.0026 < 1.5
1
152
.
We take ` : 3 10
18
. For each of our numbers i,
we take c : c
99
to be the denominator of the 99th
convergent to . Note that c depends on i. The
minimal value of c for i 2 3. 150 exceeds the
number 10
45
6`. Thus, we may apply Lemma 2
for each such c, and j. The maximal value of
`kck computed is smaller than 10
28
, whereas
the minimal value of kcjk is 1.1 10
17
. Thus,
we can take : kcjk `kck 10
17
. Also, the
maximal value of c is < 2 10
62
. Hence, by
Lemma 2, all solutions i. : of inequality (9)
have
i <
logc
log 1
<
log2 2 10
62
10
17
log1.0026
< 71000.
Next, since i 1: i, we have
: ii 1 < 71000i 1.
A computer search with Mathematica revealed
in less than one hour that there are no solu-
tions to the equation (1) in the range i 2
3. 150, i 2 10. 71000 and : 2 3. 71000i 1.
This completes the analysis in the case i 2
3. 150.
2.4. An upper bound on s in terms of m.
From now on, we assume that i ! 151. Recall that,
by (8), we have
: < 3 10
12
ilogi 2:. 10
Next we give an upper bound on : depending only
on i. If
: i 2. 11
then we are through. Otherwise, that is if i
2 < :, we then have
No. 4] Sums of powers of two Fibonacci numbers 47
: < 3 10
12
ilog:
2
6 10
12
ilog :.
which can be rewritten as
:
log :
< 6 10
12
i. 12
Since the function r 7! r log r is increasing for all
r c, it is easy to check that the inequality
r
log r
< yields r < 2log .
whenever ! 3. Indeed, for if not, then we would
have that r 2log c, therefore
r
log r
2log
log2log
.
where the last inequality follows because 2 log <
holds for all ! 3. This is a contradiction.
Taking : 6 10
12
i in the above argument, we
get that inequality (12) implies that
: < 26 10
12
i log6 10
12
i
12 10
12
ilog6 10
12
log i
< 12 10
12
i30 log i
< 12 10
12
i 7 log i
< 10
14
ilog i. 13
In the above inequalities, we used the fact that
log i ! log 151 5. From (11) and (13), we con-
clude that the inequality
: < 10
14
ilog i 14
holds for all i ! 151.
2.5. An absolute upper bound on s. Let us
look at the element
r :
:
c
2i
.
From the above inequality (14), it follows that
r <
10
14
ilog i
c
2i
<
1
c
i
. 15
where the last inequality holds for all i ! 80. In
particular, r < c
151
< 10
31
. We now write
1
:
i
c
i:
5
:2
1
1
i
c
2i
:
.
1
:
i1
c
i1:
5
:2
1
1
i1
c
2i1
:
.
If i is odd, then
1 < 1
1
i
c
2i
:
1
1
c
2i
:
< e
r
< 1 2r
because r < 10
31
is very small, while if i is even,
then
1 1
1
i
c
2i
:
exp log 1
1
c
2i
:
e
2r
1 2r.
again because r < 10
31
is very small. The same
inequalities are true if we replace i by i 1. Thus,
we have that
max 1
:
i
c
i:
5
:2
. 1
:
i1
c
i1:
5
:2
<
2rc
i1:
5
:2
. 16
We now return to our equation (1) and rewrite it as
c
i
u
i
5
12
1
i
1
:
i
1
:
i1
c
i:
5
:2
c
i1:
5
:2
1
:
i
c
i:
5
:2
1
:
i1
c
i1:
5
:2
.
or
c
i
5
12
c
i:
5
:2
1 c
:
u
i
5
12
1
:
i
c
i:
5
:2
1
:
i1
c
i1:
5
:2
<
1
c
i
1
:
i
c
i:
5
:2
1
:
i1
c
i1:
5
:2
<
1
c
i
2r
c
i:
1 c
:
5
:2
.
Thus, dividing both sides by c
i1:
5
:2
, we
conclude that
c
ii1:
5
:12
1 c
:
j 17
<
5
:2
c
ii1:
2r1 c
:
<
1
2c
i
5r
2
<
3
c
i
.
where we used the fact that
5
:2
c
i1:
5
p
c
152
:
<
1
2
. i ! i 1: i.
and c
:
! c
3
4, as well as inequality (15). Hence,
we conclude that
c
ii1:
5
:12
1
<
1
c
:
3
c
i
4
c
. 18
48 F. LUCA and R. OYONO [Vol. 87(A),
where we put : minfi. :g. We now set
2
: c
ii1:
5
:12
1. 19
and observe that
2
6 0. Indeed, for if
2
0, then
c
2i1:i
5
:1
2 Z, which is possible only when
i 1: i. But if this were so, then we would get
0
2
5
:12
1, which leads to the conclusion
that : 1, which is not possible. Hence,
2
6 0. Next,
let us notice that since : ! 3 and i ! 151, we have that
j
2
j
1
c
3
1
c
151
<
1
2
. 20
so that c
ii1:
5
:12
2 12. 2. In particular,
i 1: i <
1
log c
: 1 log 5
2
log 2
< :
log 5
2 log c
< 1.7:;
i 1: i
1
log c
: 1 log 5
2
log 2
1.6: 4. 21
It follows from (21) that i 1: i 1.6: 4 0
because : ! 3. We lower bound the lefthand side of
inequality (18) using again Matveevs theorem. We take
/ : 2. c
1
: c. c
2
:
5
p
. We also take the exponents
/
1
: i i 1: and /
2
: : 1. As in the previous
application of Matveevs result, we can take 1 : 2.
1
: 0.5.
2
: 1.61. Also, we can take 1 : 1.7:
maxfj/
1
j. j/
2
jg by inequality (21). We thus get that
expC
2.2
1 log1.7: 0.5 1.61 <
4
c
.
where
C
2.2
1.4 30
5
2
4.5
2
2
1 log 2 < 5.3 10
9
.
This leads to
<
log 4
log c
5.3 10
9
0.5 1.61 log c
1
1 log1.7:
< 9 10
9
1.6 log : < 9 2.6 10
9
log :
< 3 10
10
log :.
Here, we used the fact that 1 log1.7 < 1.6.
If :, we then get that : < 3 10
10
log :, so
: < 10
12
.
If i, then using also (14), we get that
i < 3 10
10
log : < 3 10
10
log 10
14
ilog i
.
The last inequality above leads to i < 2 10
12
, so,
by (14) once again, we get that
: < 10
14
2 10
12
log2 10
12
< 6 10
27
.
So, at any rate, we have that
: < 6 10
27
. 22
2.6. A better upper bound on s. Next, we
take
2
: : 1 log
5
p
i 1: i log c. Ob-
serve that
2
c
2
1, where
2
is given by (19).
Since j
2
j < 12 (see inequality (20)), we have that
c
j
2
j
< 2. Hence,
j
2
j c
j
2
j
jc
2
1j < 2j
2
j <
2
c
:
6
c
i
.
This leads to
log
5
p
log c
i 1: i
: 1
<
1
: 1 log c
2
c
:
6
c
i
. 23
Note rst that c
i
! c
151
3 10
31
10
3
: by esti-
mate (22).
Assume next that : 100. Then c
:
1000:.
Hence, we get that
1
: 1 log c
2
c
:
6
c
i
<
8
:: 11000 log c
<
1
60: 1
2
. 24
Estimates (23) and (24) lead to
log
5
p
log c
i 1: i
: 1
<
1
60: 1
2
. 25
By a criterion of Legendre, inequality (25) implies
that the rational number i 1: i: 1 is a
convergent to : log
5
p
log c. Let o
0
. o
1
. o
2
.
o
3
. o
4
. . . . 1. 1. 2. 19. 2. 9. . . . be the continued
fraction of , and let j
/
c
/
be its /th convergent.
Assume that i 1: i: 1 j
t
c
t
for some
t. Then : 1 dc
t
for some positive integer d, which
in fact is the greatest common divisor of i 1:
i and : 1. We have the inequality c
54
1.4
10
28
: 1. Thus, t 2 f0. . . . . 53g. Furthermore,
o
/
29 for all / 0. 1. . . . . 53. From the known
properties of continued fractions, we have that
i 1: i
: 1
j
t
c
t
1
o
t
2c
2
t
!
d
2
31: 1
2
!
1
31: 1
2
.
which contradicts inequality (25). Hence, : 100.
No. 4] Sums of powers of two Fibonacci numbers 49
2.7. The nal step. To nish, we go back to
inequality (17) and rewrite it as
c
ii1:
5
:12
1 c
:
1
1
<
3
c
i
1 c
:
<
3
c
i
.
Recall that : 2 3. 100 and
1.6: 4 < i 1: i < 1.7:.
Put t : i 1: i. We computed all the num-
bers jc
t
5
:12
1 c
:
1
1j for all : 2 3. 100
and all t 2 b1.6: 4c. b1.7:c. None of them ended
up being zero, since if it were we would get the
Diophantine equation c
:
1 5
:12
c
:t
, which
has no integer solution :. t with : ! 3 by the
arguments from the beginning. The smallest of
these numbers is 310
3
. Thus, 310
3
< 3c
i
, or
c
i
< 10
3
, so i < 15, which is false.
The theorem is therefore proved.
Acknowledgements. This work started
during a visit of Florian Luca at the University
of French Polynesia in Tahiti in November of
2010. Florian Luca thanks the Mathematics De-
partment for their hospitality. During the prepara-
tion of this paper, Florian Luca was also supported
in part by Grants SEP-CONACyT 79685 and
PAPIIT 100508.
References
[ 1 ] Y. Bugeaud, M. Mignotte and S. Siksek, Classical
and modular approaches to exponential Diop-
hantine equations. I. Fibonacci and Lucas
perfect powers, Ann. of Math. (2) 163 (2006),
no. 3, 9691018.
[ 2 ] Y. Bugeaud, F. Luca, M. Mignotte and S. Siksek,
Fibonacci numbers at most one away from a
perfect power, Elem. Math. 63 (2008), no. 2,
6575.
[ 3 ] A. Dujella and A. Peth} oo, A generalization of a
theorem of Baker and Davenport, Quart. J.
Math. Oxford Ser. (2) 49 (1998), no. 195, 291
306.
[ 4 ] D. Marques and A. Togbe , On the sum of powers
of two consecutive Fibonacci numbers, Proc.
Japan Acad. Ser. A Math. Sci. 86 (2010), 174
176.
[ 5 ] E. M. Matveev, An explicit lower bound for a
homogeneous linear form in logarithms of
algebraic numbers. II, Izv. Ross. Akad. Nauk
Ser. Mat. 64 (2000), no. 6, 125180; translation
in Izv. Math. 64 (2000), no. 6, 12171269.
50 F. LUCA and R. OYONO [Vol. 87(A),