Documente Academic
Documente Profesional
Documente Cultură
FIBONACCI NUMBERS
Abstract. Recently, several authors considered the problem of estimating expressions of the following type
where (Fn)n0 is the classical Fibonacci sequence given by F0 = 0, F1 = 1 and Fn = Fn1+Fn2 for all n
2. Most of the previous work focused on the relatively simple case s = 1 and s = 2. In the present
paper, we find the asymptotic expansions for F1(n,s) for all 1 s 6. To give the reader an idea
how these expansions look like, we state the result in the case s = 3.
.
1
1. The Problem.
Let Fn denote the n-th term of the Fibonacci sequence defined by the initial conditions F1 = F2
= 1 and the recurrence relation Fn+2 = Fn+1 + Fn. There is a huge body of literature on Fibonacci
numbers; there are thousands of research papers dedicated to the topic. A quick internet search
with keyword Fibonacci returns close to three million links. It is therefore somewhat
surprising that there are still things we do not know about this sequence.
Recently, the following question concerning the Fibonacci sequence has been considered.
, if n is odd.
Similar results were proved by Liu and Zhao [8] for the sequence {Gn}n1, the hyperfibonacci
for all n 1.
1
.
, if n is even
Earlier, Backstrom [1] considered the problem of estimating where a, b and c are
fixed integers. Komatsu and Laohakosol [7] consider the generalized Fibonacci sequence
instead; this is defined by the recurrence un = un1 + un2 + ... + unm, with ui 0 for all
1 i m 1. Generalizing some earlier results of the first author (see [6, 5]), Komatsu and
Theorem 1.4. [7] For n n0, where n0 large enough the following equalities hold:
Here denotes the nearest integer function, namely x = x + 1/2 . Related problems
A few questions seem natural. First, all papers cited above estimate F1(n,s) (or some of its
variants) only for s = 1 and (eventually) for s = 2. It is reasonable to inquire what happens for
other values of s? Second, why stop at finding the (nearest) integer part of F1(n,s) and not try
to find a more precise expression? The goal of this project is to answer these questions.
Theorem 1.5.
2
.
Theorem 2.1.
(1) .
3
The proof of the claim is a matter of straightforward algebraic verification. Using the iden-
whichfurthergives
0 fork 4.
This proves the right inequality in (1). For the left inequality we proceed in a similar manner.
fromwhich
0 fork 4.
This completes the proof of claim 2.2. Summing now (1) over all k from n to we obtain for
every n 4 that
Theorem 2.3.
(2) .
For the right inequality we need to show that 0. Since the denominators
are all positive this is equivalent to showing that (bk+1 bk)Fk2 bk bk+1 > 0. Indeed, by using the
, that is,
(3) .
This proves the right inequality in (2). For the left inequality it suffices to prove that
.
Using (3), the left term of the above inequality can be written as
Summing now (3) over all k from n to we obtain for every n 2 that
5
from which .
In this section we prove estimates for F1(n,3) and F1(n,4). While the main approach stays
the same, the computations become more complicated. For this reason we employ the computer
algebra software MAPLE to perform and simplify some of these calculations. For completeness,
we initially decided to include an appendix containing these details. However, the paper became
longer since we wanted to also explain how we determined these expansions. In order to meet
to page limit we ended up leaving the appendix out. The interested readers can check these
computations themselves; also, a copy of this appendix can be made available upon request.
Theorem 3.1.
Proof. Let
(4) .
It is easy to check that ck > 0 for all k 1. The result stated in the theorem will be a
(5) .
6
As noticed earlier, the denominator of the above expression is positive so it suffices to study
only the numerator. As before, we use the identities Fk+2 = 2Fk + Fk1, Fk+1 = Fk + Fk1, Fk2 = Fk
Fk and Fk1 only. We skip the intermediate computations and present the final result
(6) ( 0
forallk 1,
and the right inequality in (5) is proved. For the left inequality it suffices to show that
Using (6), the left term of the above inequality can be written as
0 for all k 2.
As above, the above equality steps can be easily verified using the identities Fk+1 = Fk + Fk1
Summing now (5) over all k from n to we obtain for every n 2 that
from which ,
Theorem 3.3.
.
7
Proof. Define
(7) .
(8) .
It is easy to check that for all k 1 we have that dk > 751Fk2 hence the denominators in (8) are
As noticed earlier, the denominator of the left hand side term in the inequality above is positive.
It thus suffices to prove that the numerator is positive as well. Consider the following identities:
3Fk2 + 4FkFk1 + Fk21,F2k3 = Fk21 + Fk22 = Fk2 2FkFk1 + 2Fk21, F2k1 = Fk2 + Fk21,
F2k+1 = Fk2+1 + Fk2 = 2Fk2 + 2FkFk1 + Fk21, F2k+3 = Fk2+2 + Fk2+1 = 5Fk2 + 6FkFk1 + 2Fk21.
Using these identities in conjunction with (7) and the additional identity
(1)k+1 leads to the following equality
8
(9) + (1)k(750Fk6 + 750Fk3Fk+1 + 375Fk4 1190FkFk+1 749Fk2 + 47).
It is easy to notice that the dominant term in the second expression in the equality above is
Fk5(825Fk+1 (1)k750Fk) and it is positive. A quick computer check shows that the second term
in (9) is positive for all k 3. This proves the right inequality in (8).
For the left inequality in (8) we proceed in a very similar manner. The inequality to prove is
As above, focus on the numerator and use the same set of identities to simplify it. We get
Another computer check shows that the second term in the above equality is negative for all k
1. This proves the left inequality in (8) and with it the entire Claim 3.4. Summing now the
.
9
To complete the proof of Theorem 3.3 all that remains to show is the following
Claim 3.5.
Employing now the identities F2n+1 = Fn2+1 + Fn2 = 2Fn2 + 2FnFn1 + Fn21, F2n1 = Fn2 + Fn21
For proving (10) we will use the closed form solution for Fn, known as Binets formula. Let
Fn = (n n)/5. Using this formula, the left side term in (10) becomes
, as desired.
This finishes the proof of Claim 3.5 and with this, the proof of Theorem 3.3 is complete.
As it can be easily seen from the stated expressions of 1(n,5) and 1(n,6), things get
10
Theorem 4.1.
(11)
It is easy to check that ek > 1/Fk for all k 3. As in the previous proofs, we start with the
following
(12)
Using the identities F3k+4 = Fk3+2 + 3Fk+2Fk+1 Fk3+1, F3k+2 = Fk3+1 + 3Fk2+1Fk + Fk3, F3k+1 =
we obtain that
11
A straightforward computer check shows that the right hand term of the above equality is
negative for all k 1. This proves the left inequality in (12). For proving the right inequality we
proceed in a similar manner. Using the same set of identities mentioned earlier we obtain that
As above, we can easily check that the right hand term of the above equality is positive for all k
1. This completes the proof of Claim 4.2. Summing now (12) for all k from n to we obtain
which gives .
The proof of Theorem 4.1 is now complete.
Theorem 4.3.
(13)
It is a straightforward matter to check that gk > 0 for all k 2. As it was the case with the earlier
12
Claim 4.4. For every k 3 we have that
(14) .
The left inequality is equivalent to 0. Since the denominators are all positive
this reduces to showing that (gk+1 gk)Fk6 gkgk+1 < 0. We are going to need the following
identities: ;
;
.
It is obvious that by using the identities above we should be able to express (gk+1 gk)Fk6
gkgk+1 in terms of Fk and Fk1 only. Again, we use MAPLE to perform the simplifications and we
This proves the left inequality in (14). For the right inequality we need to show that
Making use of (15) the left term of the above inequality can be written as
Here we used the (easy to check) fact that for all k 3. This proves the
right inequality in (14) and thus completes the proof of Claim 4.4. Summing now (14) for all k
from n to we obtain that for every n 3 the following double inequality holds.
13
which gives .
The proof of Theorem 4.3 is now complete.
5. How were these formulae discovered?
It is apparent that the expansion of F1(n,s) becomes more complicated as the values of s get
higher. We believe it would be interesting to say a few words about how we found these
expressions. Since the process is the same in all the cases we will present the details for the case
s = 4 only.
First, let us introduce two notations. Let xn and yn be two sequences of nonzero terms. If
It can be easily shown that for every s > 0 we have that where
b = (1 5)/2. It follows that the leading term in the expansion of F1(n,4) can be chosen to be
(1 b4)Fn4. However, we seek a nicer leading term, preferably one which has rational
coefficients only. The reason is not purely aesthetic: we want to be able to use various identities
involving Fibonacci numbers in performing certain calculations; if the coefficients are rational
we have a much better chance to obtain simpler final expressions. Let a = (1 + 5)/2 and
recall that Fn = (an bn)/5 that is, Fn an/5. Notice that a + b = 1, a b = 5 and ab = 1. Then
we have
14
So rather than using (1 as the leading term in the expansion of F1(n,4) we use
is close to 1/Fk4 (in a sense to be made precise a bit later). The reason behind this
choice is that we intend to obtain a double inequality similar to (8) from which we can later
and we obtain
As mentioned earlier, we want the left hand term of the above equality to be small. This
roughly means that the terms of highest degree in the right hand term cancel out, that is,
2
(16) xk = F2k2Fk 5 F2k3.
15
Now look again at the difference F1(n,4)xn. Computer data suggests that 1(n,4)xn c where
c is about 0.2596.... To find the exact value of c we proceed in a manner similar to the one above.
close to 1/Fk4 for the correct value of c. To this end express the numerator of in
terms of Fk and Fk1 only; of course, we will need to use the appropriate identities. Here is what
we obtain:
(yk+1 yk)Fk4 yk yk+1 =(75c 25)Fk4 + (150c 30)Fk3Fk1 + (25c2 10c 8)+
k1 = 0 which gives.
This describes our approach for estimating F1(n,4). Similar techniques were used for the
other cases. In principle, for a given s this process can be carried out for as many steps as one
wishes. For instance, we present stronger versions of the results proved above. The proofs use
the same ideas presented earlier; of course, the computations are going to be more complicated
but for the most part this is not an issue as we can use MAPLE to perform these calculations.
16
.
A general expansion formula for F1(n,s) is highly desirable. However, the equalities in the
theorem above shed very little light on how such a formula may look like. If there is a pattern to
be discovered, it still eludes us. It may be that rewriting these equalities in terms of powers of
Fn only may be useful. But if we do that the coefficients are going to be irrational. Recall that our
main concern was to obtain expansions with rational coefficients; this way, we had an easier
17
References
[1] R. Backstrom, On reciprocal series related to Fibonacci numbers with subscripts in arithmetic progression,
[2] N. N. Cao and F. Z. Zhao, Some properties of hyperfibonacci and hyperlucas Numbers, J. Integer Sequences 13
[3] C. Elsner, S. Shimomura and I. Shiokawa, Algebraic relations for reciprocal sums of Fibonacci numbers, Acta
[4] C. Elsner, S. Shimomura and I. Shiokawa, Algebraic relations for reciprocal sums of odd terms in Fibonacci
[5] S. H. Holliday and T. Komatsu, On the sum of reciprocal generalized Fibonacci numbers, Integers 11A
(2011).
[6] T. Komatsu, On the sum of reciprocal sums of tribonacci numbers, Ars Combinatoria 98 (2011), 447459.
[7] T. Komatsu and V. Laohakosol, On the sum of reciprocals of numbers satisfying a recurrence relation of order
[8] R. Liu and F. Z. Zhao, On the sums of reciprocal hyperfibonacci numbers and hyperlucas numbers, Journal of
[9] H. Ohtsuka and S. Nakamura, On the sum of reciprocal sums of Fibonacci numbers, Fibonacci Quarterly