On a Diophantine equation involving powers of Fibonacci numbers
By Krisztia´n GUETH,1ÞFlorian LUCA2Þ,3Þ,4Þand La´szlo´ SZALAY5Þ
(Communicated by Shigefumi MORI,M.J.A., March 12, 2020)
Abstract: This paper deals with the diophantine equationF1pþ2F2pþ þkFkp¼Fnq, an equation on the weighted power terms of Fibonacci sequence. For the exponentsp; q2 f1;2gthe problem has already been solved in ad hoc ways using the properties of the summatory identities appear on the left-hand side of the equation. Here we suggest a uniform treatment for arbitrary positive integerspandqwhich works, in practice, for small values. We obtained all the solutions for p; q10by testing the new approach.
Key words: Fibonacci number; diophantine equation; weighted sum.
1. Introduction. As usual, fFmgm0 de- notes the sequence of Fibonacci numbers F0¼0, F1¼1 and Fmþ2¼Fmþ1þFm for all m0. Its companion sequencefLmgm0is the Lucas sequence given byL0¼2,L1¼1and Lmþ2¼Lmþ1þLm for all m0. We assume that the reader is familiar with their Binet formula.
In this paper, we determine the solutions to the Diophantine equation
F1pþ2F2pþ þkFkp¼Fnq ð1Þ
in positive integers ðp; q; k; nÞ where p and q are small. This equation was first investigated by Ne´methet al.in [5] for the four possibilitiesfp; qg f1;2g, and all the solutions in these particular cases were obtained in elementary ad hoc ways. The purpose of this paper is to provide a uniform treatment independently from the values of p and q. As a particular case, we solve the above equation for all values ofp; qwhich do not exceed the upper bound 10.
We consider
F1p¼1¼F1q¼F2q; and F1pþ2F2p¼3¼F4
as trivial solutions to (1). The authors in [5] have made the following
Conjecture 1. Equation (1) has only the three non-trivial solutions
ðp; q; k; nÞ ¼ ð1;1;4;8Þ; ð1;2;3;4Þ; ð3;3;3;4Þ:
The above conjecture says, in particular, that there exist only finitely many solutions. Since the equation is not a standard equation, the finiteness of its number of solutions does not seem to follow in an easy way. Note that the first two quadruples were obtained in [5], while the last one is justified here in the sense that our present work confirms the conjecture by solving the equation for maxfp; qg 10. The result is recorded in
Theorem 2. Conjecture 1 is true whenever maxfp; qg 10.
Problems having similar flavour appear in the extensive literature of Fibonacci sequence. For instance, the sumFnsþFnþ1s (n0) gives Fibonacci numbers when s2 f1;2g. For larger exponents s, Marques and Togbe´ [4] proved that ifFnsþFnþ1s is a Fibonacci number for all sufficiently large n, then s¼1 or 2. Afterwards, Luca and Oyono [2] com- pleted the solution of the question by showing that apart from F1sþF2s¼F3 there is no solutions3 to the equationFnsþFnþ1s ¼Fm.
A naturally arising question is what would happen if we replace the Fibonacci numbers by other linear recurrence? In the case of non degen- erate binary recurrences with real roots it is likely our approach works. On the other hand, we do not think that the method extends to Tribonacci numbers or to other recurrences of order higher
doi: 10.3792/pjaa.96.007
#2020 The Japan Academy
2020 Mathematics Subject Classification. Primary 11B39;
Secondary 11D45.
1Þ University Eo¨tvo¨s Lora´nd, Savaria Centre, Ka´rolyi Ga´spa´r te´r 4, 9700 Szombathely, Hungary.
2Þ University of the Witwatersrand, School of Mathematics, 1 Jan Smuts Ave, Johannesburg, 2000, South Africa.
3Þ Research Group of Algebraic Structure & Applications, King Abdulaziz University, Jeddah, Saudi Arabia.
4Þ Department of Mathematics, Centro de Ciencias Mate- ma´ttics UNAM, Morelia, Mexico.
5Þ Jan Selye University, Institute of Mathematics and Informatics, Elekra´renska´ cesta 2, 94501 Koma´rno, Slovakia.
than 2 although we have made no efforts in this direction.
Now we collect some preliminary results we will use in the proof of Theorem 2. In what follows, logb denotes the logarithm to baseb, whereb >1 is any real number, while :¼ ð1þ ffiffiffi
p5
Þ=2 is the dominant root of the Fibonacci sequence. Since the following three lemmata are widely known, we present them without proof.
Lemma 3. For n1, we have n2 Fnn1, andn1 Ln.
Lemma 4. The inequality Fnþ1=Fn3=2 holds forn2.
Lemma 5. Assume that n is divisible by 4.
ThenFnF4¼Fðn4Þ=2Lðnþ4Þ=2.
At some stage of the proof of the theorem we will use the following estimates.
Lemma 6. Equation(1) implies . ðk2Þp <ðn1Þqif k2, and . ðn2Þq <ðk1Þpþlogð4kÞifk3.
Proof. Combining Lemma 3 andFkp< Fnq(pro- vided by (1) and k2) leads immediately to the first statement.
For the second statement, Lemma 4 yields
Fk
Fki¼Yi1
j¼0
Fkj
Fkj1
3 2
i1
for alli¼1;2;. . .; k1, wherek3. Note that the lower bound could be improved toð3=2Þiifik2 because we avoid the quotientF2=F1¼1. Put:¼ 2=3. Recalling Lemma 3, observe that
ðn2Þq< Fnq¼kFkpXk1
j¼0
kj k
Fkj Fk
p
ð2Þ
< kFkpð1þ1þþ2þ þk2Þ
<4kðk1Þp:
Then the statement follows by taking logarithms.
Lemma 7. Suppose thatkandnare positive integers. Then5kkFn if and only if 5kkn.
Proof. See Lemma 1 in [1].
Lemma 8. Assume thatk,p andqare posi- tive integers,p is odd. If
Lpk2þ ðLp2Þk1¼ 5pqL2p holds, thenðp; q; kÞ ¼ ð1;1;1Þ;ð1;1;2Þ.
Proof. Since Lp is never a multiple of 5, and 5pqL2p is an integer, it follows that qp. Put r¼pq. If p¼1, then k2k1¼ 1 gives the
two solutions above. The condition p3 entails that the left-hand side is positive so the sign in the right-hand side must be +. Suppose now thatp¼3 or p¼5. The left-hand side is a quadratic poly- nomial in k with leading coefficient Lp6¼0. A verification with q2 f1;. . .; pg provides no more solutions.
In the sequel, we may assume p7. We will show that this assertion contradicts the equality in the lemma. Reducing
Lpk2þ ðLp2Þk1¼5rL2p ð3Þ
moduloLp, it leads to Lpj2kþ1. Thus, Lp is odd, sop is not a multiple of 3. Moreover 2kþ1¼aLp
holds for some odd integera. Thus,k¼ ðaLp1Þ=2.
Substituting this into (3), after some manipulations we obtain
a2L2p ð4aþ1Þ ¼45rLp: ð4Þ
On one hand, this gives Lpj4aþ1, therefore a ðLp1Þ=47. On the other hand, since Lp29, we have4aþ1<5a < a2L2p=2. Consequently
45rLp¼a2L2p ð4aþ1Þ>a2L2p 2 ; therefore
a < 23=25r=2 L1=2p
:
This implies
ðLp1ÞL1=2p <27=25r=2: ð5Þ
On the other hand, rewriting (4) as ðL2pÞa24a ð45rLpþ1Þ ¼0;
and treating it as a quadratic ina, its discriminant is a perfect square. Subsequently,
4þL2pð45rLpþ1Þ ¼y2
holds for some positive integer y. So, 45rL3pþ ðL2pþ4Þ ¼y2. Since L2pþ4¼5Fp2 (because p is odd), we get
45rL3pþ5Fp2¼y2:
Letc; dbe such that5ckFpand5dky. ThenFp¼5cu, y¼5dv for some integers u; v with gcdðuv;5Þ ¼1 and
45rL3pþ52cþ1u2¼52dv2:
Since 5-Lp, 5r is the exact power of 5 in the
factorisation of 45rL3p. From the above equation, we have that since2cþ16¼2d, eitherr¼2cþ1, or r¼2d, and in the last case r2cþ1. Clearly, in both cases r2cþ12 log5pþ1, where in the second inequality we used Lemma 7. Thus, 5r 5p2. Returning to (5), we obtain
L1=2p ðLp1Þ 27=25r=227=251=2p;
and sinceLpp1 (see Lemma 3), we conclude ðp1Þ=2ðp11Þ 27=251=2p;
an inequality false for anyp6.
2. Proof of the theorem. Because we al- ready accounted for the trivial solutions, we may assume k3. First we handle the case k¼3 separately. In fact we will exploit k4 only in (12), but without this assumption, one has more difficulties in our argument after (11). With k¼3 we find Fnq¼F1pþ2F2pþ3F3p¼3ð1þ2pÞ, so 3jFn
and Fn is odd, so 4jn and 3-n. If q¼1, then 3 2p¼Fn3¼FnF4¼Fðn4Þ=2Lðnþ4Þ=2 by Lemma 5. Thus Fðn4Þ=2 has its largest prime factor at most 3. The Primitive Divisor Theorem implies ðn4Þ=212, so n28 and the remaining possi- bilities can be verified by hand. If q¼2, we have 3ð1þ2pÞ ¼. Thus, 2pþ1¼3. Distinguishing between p 0;1;2 (mod 3Þ, we get the equation 3y2 ¼1þ2rx3, where r2 f0;1;2g, which one can solve with MAGMA [3]. If q¼3, we handle similarly the equation 1þ2p¼9y3. Distinguishing between p even and odd one has 9y3¼1þ2rx2, r2 f0;1g, and all integer solutions ðx; yÞ to these equations can again be computed with MAGMA [3].
Assume now that q4. Then 34jFnq. Hence, 33j1þ2p, so p is odd and 9jp. In particular, 19j29þ1j2pþ1jFnq, so19jFn. Thus18jn, con- sequently3jn, a contradiction.
So, from now onk4. Let the integerp1be fixed (not necessarily inf1;2;. . .;10g), and consider the term Fjp with j1. Since we have Fj¼ ðj jÞ= ffiffiffi
p5
, where¼ ð1 ffiffiffi p5
Þ=2, it follows that Fjp¼ðjjÞp
5p=2 ¼ jp 5p=2þp;j; where
jp;jj<2pðp1Þj
5p=2 < jðp1Þ: Thus,
Xk
j¼1
jFjp ¼ 1 5p=2
Xk
j¼1
jjp
! þR1; where jR1j<Pk
j¼1jjðp1Þ< k2kðp1Þ. The inner sum is
Xk
j¼1
jxj¼x d dx
Xk
j¼1
xj
!
¼x d
dx xxk1 x1
¼kxkþ2 ðkþ1Þxkþ1þx ðx1Þ2
with x:¼p. Thus, Xk
j¼1
jFjp ¼kp ðkþ1Þ 5p=2ðp1Þ2 pðkþ1Þ þ p
5p=2ðp1Þ2þR1: Letqbe a positive integer. Writing also
Fnq¼ nq 5q=2þR2; where
jR2j 2 51=2 q
nðq1Þ< nðq1Þ;
the above formulas lead to 5ðqpÞ=2ðkp ðkþ1ÞÞ
ðp1Þ2 pðkþ1Þqn
ð6Þ
¼5q=2R25q=2R15ðqpÞ=2p ðp1Þ2: Thus, in the right-hand side of (6) we see that
5q=2jR1j 5q=2k2kpk; ð7Þ
and
5ðqpÞ=2p
ðp1Þ2 5q=235q=2kpkþ3: ð8Þ
In the latter case, we used the facts that p1 p=2 for all p2, and =ð1Þ2 ¼3 (to include the casep¼1). Bounding5q=2R2 takes a bit longer.
Clearly, in the exponent of the upper bound onjR2j we have nðq1Þ ¼ ðn2Þqþ2qn and further 5q=22q¼ ð ffiffiffi
p5
2Þq<6q. Combining these and (2), we obtain
5q=2jR2j<5q=2nðq1Þ ð9Þ
<5q=24kðk1Þpþ2qn
<6qþ1kðk1Þpn:
If kn, then the last term is not larger then 6qþ1kðk1Þpk. Assume nowk > n. Obviously,pq leads to Fnq< Fkp, which contradicts (1). Hence, p < q. Putb¼maxfp; qg ¼q. The first statement of Lemma 6 provides
n >ðk2Þp
qþ1k2 b þ1;
which together with (9) implies
5q=2jR2j<6qþ1kðk1Þpðk2Þ=b1
<6qþ1kkpk=b:
Comparing it with the estimate obtained forkn, we get a general upper bound on5q=2jR2j. Obvious- ly, in (7) and in (8) we can replace k in the exponents byk=b to unify the two upper bounds obtained for the casesknandk > n, respectively.
Letp:¼ ðp1Þ2, and
zqðkÞ:¼5q=23þ6qþ1kþ5q=2k2:
Putting all the above estimates together, we get that
5ðqpÞ=2ðkpk1Þ
p qnpðkþ1Þ
< zqðkÞ k=bþp: ð10Þ
The next goal is to analyse the exponent :¼ qnpðkþ1Þ. We distinguish two situations. Sup- pose first that
5ðqpÞ=2 3p : ð11Þ
Here observe that fork3andp1 5ðqpÞ=2
p
ðkpk1Þ ð12Þ
5ðqpÞ=2 p
ðkðp1Þ 11=3Þ>5ðqpÞ=2 p
:
It now follows from (10) and (12), that k=bþp<5ðpqÞ=2pzqðkÞ;
ð13Þ
which gives us an upper bound on k for fixed p; q.
Later, we will see that the other branch provides also a bound which is larger.
In the sequel, we suppose that the opposite of (11) is true. First consider the case when the left- hand side in (10) is zero. After rearranging the corresponding relation, we take the norms inQ½ ffiffiffi
p5 and get to
k2ðÞp ðkþ1ÞkðpþpÞ þ ðkþ1Þ2
¼5pqðÞqnpðkþ1ÞNQpffiffi5ðpÞ:
By¼ 1,pþp¼Lp, and
NQ½pffiffi5ðpÞ ¼ ðð1ÞpLpþ1Þ2
the previous equality simplifies
ðð1ÞpLpþ1Þk2þ ð2LpÞkþ1
¼ 5pqðð1ÞpLpþ1Þ2:
Ifp is even, the corresponding equation is
ðLp2Þk2þ ðLp2Þk1¼ 5pqðLp2Þ2: Hence,ðLp2Þ j1, leading toLp¼3, sop¼2. Now k2þk1¼ 52q;where the eligible values forqis 1 or 2. Each case leads to a trivial solution.
If p is odd, then Lemma 8 handles the equation
Lpk2þ ðLp2Þk1¼ 5pqL2p
and provides only ðp; q; kÞ ¼ ð1;1;1Þ;ð1;1;2Þ, deriv- ing trivial solutions to (1).
Assume now that the left-hand side in (10) is nonzero. We then have
jkp ðkþ1Þ 5ðpqÞ=2pj ð14Þ
<5ðpqÞ=2pzqðkÞ
k=bþp :
Changingtoin the left above we get an amount jkp ðkþ1Þ 5ðpqÞ=2ðpÞj
ð15Þ
<ð2kþ1Þ þ95pqp;
whereðpÞ ¼ ðp1Þ2<3. We also used the fact that the opposite of (11) is true, therefore
jj¼<35ðpqÞ=2p:
The product of the left-hand sides of (14) and (15) is the norm of a nonzero algebraic integer inKso it is 1. We thus get that
k=bþp<5ðpqÞ=2pzqðkÞðð2kþ1Þ þ95pqpÞ:
ð16Þ
Note that (13) is weaker in (16), which therefore gives a general bound for kirregardless of whether (11) holds or not.
Taking maxfp; qg 10, (16) gives k1104.
Now one can easily check when Xk
j¼1
jFjp
!1=q
is a Fibonacci number for positive integer variables p; q2 f1;. . .;10g and k1104 getting only the
solutions from the statement of the theorem. The proof is complete.
Acknowledgements. The work on this pa- per started when the last author visited School of Mathematics of the Wits University. He thanks this Institution for support, and also thanks Kruger Park for excellent working conditions. F. L. was supported in part by the Number Theory Focus Area Grant of CoEMaSS at Wits (South Africa).
Part of this work was done when this author visited the Max Planck Institute for Mathematics in Bonn, Germany from September 2019 to February 2020.
He thanks this Institution for hospitality and support. For L. Sz. the research was supported by Hungarian National Foundation for Scientific Re- search Grant No. 128088. This presentation has been made also in the frame of the ‘‘EFOP-3.6.1-16- 2016-00018 – Improving the role of the research + development + innovation in the higher education
through institutional developments assisting intel- ligent specialization in Sopron and Szombathely’’.
References
[ 1 ] T. Lengyel, The order of the Fibonacci and Lucas numbers, Fibonacci Quart. 33 (1995), no. 3, 234–239.
[ 2 ] F. Luca and R. Oyono, An exponential Diophantine equation related to powers of two consecutive Fibonacci numbers, Proc. Japan Acad. Ser. A Math. Sci.87(2011), no. 4, 45–
50.
[ 3 ] MAGMA Handbook, http://magma.maths.usyd.
edu.au/magma/handbook/.
[ 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), no. 10, 174–176.
[ 5 ] G. Soydan, L. Ne´meth and L. Szalay, On the Diophantine equation Pk
j¼1jFjp¼Fnq, Arch.
Math. (Brno)54(2018), no. 3, 177–188.