J.
Indones.
Math.
Soc.
Vol.
No.
, pp.
1Ae17.
TWO ASPECTS OF A GENERALIZED FIBONACCI
SEQUENCE
Tuwankotta Analysis and Geometry Group.
FMIPA.
Institut Teknologi Bandung.
Ganesha no.
Bandung.
Indonesia theo@math.
Abstract.
In this paper we study the so-called generalized Fibonacci sequence:
xn 2 = xn 1 xn , n OO N.
We derive an open domain around the origin of the parameter space where the sequence converges to 0.
The limiting behavior on the boundary of this domain are: convergence to a nontrivial limit, k-periodic .
OO N), or quasi-periodic.
We use the ratio of two consecutive terms of the sequence to Oo construct a rational approximation for algebraic numbers of the form: r, r OO Q.
Using a similar idea, we extend this to higher dimension to construct a rational Oo Oo approximation for 3 a b c 3 a Oe b c d.
Key words and Phrases: Generalized Fibonacci sequence, convergence, discrete dynamical system, rational approximation.
Abstrak.
Dalam makalah ini barisan Fibonacci yang diperumum: xn 2 = xn 1 xn , n OO N, dipelajari.
Di sekitar titik asal dari ruang parameter, sebuah daerah .
impunan terhubung sederhan.
yang buka diturunkan.
Pada daerah ini barman tersebut konvergen ke 0.
Perilaku barisan untuk n yang besar, pada batas daerah juga diturunkan yaitu: konvergen ke titik limit tak trivial, periodik-k .
OO N), atau quasi-periodic.
Dengan menghitung perbandingan dari dua suku berturutan dari Oo barisan, sebuah hampiran dengan menggunakan bilangan rasional untuk: r, r OO Q Ide yang serupa digunakan untuk mengkonstruksi hampiran rasional Oo Oo untuk 3 a b c 3 a Oe b c d.
Kata kunci: Barisan Fibonacci diperumum, kekonvergenan, sistem dinamik diskrit, hampiran rasional.
Introduction Arguably, one of the most studied sequence in the history of mathematics is the Fibonacci sequence.
The sequence, which is generated using a recursive 2000 Mathematics Subject Classification: 11B39.
27Exx.
11J68.
Received: 10-05-2014, revised: 01-10-2014, accepted: 07-10-2014.
Tuwankotta xn 2 = xn 1 xn , n OO N, with: x1 = 0 and x2 = 1, can be found in the Book of Calculation (Liber Abbac.
by Leonardo of Pisa .
, pp.
as a solution to the problem: AyHow many pairs of rabbits can be bred in one year from one pair?Ay.
This sequence has beautiful properties, among others: its ratio:
, n > 1.
Oo converges to the golden ratio: 12 1 5 .
Generalization of the Fibonacci sequence has been done using various approaches.
One usually found in the literature that the generalization is done by varying the initial conditions: x1 and x2 .
ee for example: .
Another type of generalization is the one that can be found already in 1878 .
or more recent literature see .
), by considering the so called: a two terms recurrence, i.
xn 2 = xn 1 xn , n OO N.
A wonderful exposition about properties of the generalized Fibonacci sequence can be found in .
, .
The application of this type of generalized Fibonacci sequence can be found in various fields of science, among others: in computer algorithm .
and in probability theory .
A further generalization of the two terms recurrence can be found in .
, where the authors there use different formula for the even numbered and the odd numbered term of the sequence.
Another interesting variation can be found in .
, 9, .
where the authors there consider the two terms recurrence modulo an One of the most striking results is that a connection with FermatAos Last Theorem has been discovered.
Summary of the results.
In this paper, we follow the generalization in .
, .
, the two terms recurrence, and the sequence shall be called the generalized Fibonacci sequence.
Following .
, we concentrate on the issue of convergence of the generalized Fibonacci sequence for various value of and .
We rewrite the generalized Fibonacci sequence as a two dimensional discrete linear dynamical system:
v n 1 = A(, )v n , n OO N.
There are three cases: the case where A(, ) has two real eigenvalues, has one real eigenvalue and a pair of complex eigenvalues.
For the first case, the iteration can be represented as linear combination of the eigenvectors of A(, ).
For the second case, we rewrite A(, ) as a sum to two commuting matrices: a semi-simple matrix and a nilpotent matrix.
For the third case.
A(, ) is represented as V RV Oe1 where R is a rotation matrix.
These are all standard techniques in linear algebra.
Using this, we derive an open domain in (, )-plane where the generalized Fibonacci sequence converges to 0.
It turns out that the boundary of this open domain is also interesting to analyze.
On this boundary, the limiting behavior of the Fibonacci sequence could be quite different.
The boundary consists of three Two aspects of a generalized Fibonacci sequence On one of those lines the generalized Fibonacci sequence converges to a nontrivial limit.
On the other two lines the behavior is either 2-periodic, or k-periodic or quasi-periodic.
Another aspect that we discuss in this paper is the ratio of generalized Fibonacci As is mentioned, in the classical Fibonacci sequence, the ratio converges to the golden ratio.
For the generalized Fibonacci sequence, the limit of the ratio is an algebraic number of the form:
for some and .
For rational and , the ratio of the generalized Fibonacci sequence is rational.
Thus, we can use this iteration to construct a rational approximation of radicals.
The rate of convergence of this iteration is also discussed in this paper.
We present also a generalization to higher dimension, i.
by looking at three terms recurrent.
The ratio of this generalization can be used to approximate a more sophisticated algebraic number.
Apart from a detail analysis and explicit computation in applying the knowledge in linear algebra, in this paper we also describe a few examples which are in confirmation with the analytical results.
A Generalized Fibonaci Sequence Consider a sequence of real numbers which is defined by the following recursive formula xn 2 = xn 1 xn , n OO N .
where and are real constants, and x1 and x2 are two real numbers called the initial conditions.
This sequence is called: generalized Fibonaci sequence.
setting yn = xn 1 , we can write .
as a two dimensional discrete dynamical system v n 1 = A(, )v n .
and A(, ) = Given, the initial condition: v 1 = .
1 , x2 )T , then the iteration .
can be written v n = A(, )nOe1 v 1 , n OO N.
We are interested in the limiting behavior of .
nd consequently .
), as n Ie O.
It is clear that an asymptotically stable fixed point of the discrete system .
is a limiting behavior of .
as n goes to infinity.
However, an asymptotically stable fixed point of .
is an isolated point in R2 , which is the origin: .
, .
T OO R2 .
In the next section we will use this technique to derive the domain E in (, )-plane Tuwankotta where .
converges to 0.
Another interesting limiting behavior of .
will be found on the boundaries of E.
Definition 2.
A sequence .
n } OC R is called p-periodic .
r, periodic with period .
, if there exists p OO N such that xn = xn p .
OAn OO N.
Furthermore, if m OO N satisfies: xn = xn p .
OAn OO N, then m is divisible by p.
A p-periodic solution will be denoted by:
1 , x2 , .
, xp } .
Clearly, a p-periodic sequence .
ith p > .
does not satisfy the Cauchy criteria for convergence sequence, thus the sequence diverges.
In the next definition we will describe what we mean by converging to a periodic sequence.
Definition 2.
A sequence .
n } OC R converges to .
1 , x2 , .
, xp } if, for every A > 0, there exists M OO N such that, for every k > N .
M k Oe x1 ) .
M 1 k Oe x2 ) .
M p k Oe xp ) < A.
In the next section, we will look at the eigenvectors of the matrix A(, ) to derive the limiting behavior for .
Furthermore, a periodic behavior will also be Limiting behavior of the generalized Fibonaci sequence Consider the equation for the eigenvalues of A(, ), i.
2 Oe Oe = 0.
There are three cases to be considered, i.
the case where: 2 4 > 0, the case where: 2 4 = 0, and the case where: 2 4 < 0.
The case where 2 4 > 0.
Let us consider the situation for 2 4 > 0, where the matrixp A(, ) has two different real p eigenvalues.
Those of eigenvalues are: 1 = 21 12 2 4 and 2 = 12 Oe 12 2 4, where their corresponding eigenvectors are:
, k = 1, 2, k These eigenvectors are linearly independent, so we can write:
v 1 = 1 1 x1 Oe y1 2 x1 Oe y1 and 2 = Oe p 1 = p Then, nOe1 v n = 1 1 The iteration .
converges to .
, .
T if .
| < 1 and .
| < 1.
nOe1 Two aspects of a generalized Fibonacci sequence Consider: .
| < 1, or equivalently:
Oe2 < 2 4 and 2 4 < 2.
From Oe2 < 2 4 we derive:
Oe.
) < 2 4.
If > Oe2, then the inequality is satisfied for every .
henever 2 4 > .
< Oe2 then .
) < 2 4 which implies Oe > 1.
From 2 4 < 2 we derive:
2 4 < .
Oe ).
Then, there are no possible solutions for the inequality, if > 2.
If < 2, then 2 4 < .
Oe )2 which implies < 1.
These results are presented geometrically in Figure 1.
Figure 1.
In this Figure, we have plotted the area in the (, )Oeplane where .
| < 1.
The domain is bounded by the line = 1 which is plotted using a dashed and dotted line, the parabola: 2 4 = 0 which is plotted using a dashed line, and Oe = 1 which is plotted using a solid line.
The previously described analysis can be repeated to derive the domain where .
| < 1.
However, we will use a symmetry argument to derive that domain.
Note that, by writing = Oe then 2 = Oe 4 = Oe Oe 4.
Tuwankotta Thus, the domain where .
| < 1 can be achieved by taking the mirror image of the domain in Figure 1 with respect to the -axis.
Taking the intersection between the two domains, we derive the domain in (, )-plane where .
converges to .
, .
for 2 4 > 0.
The domain is plotted in Figure 2 Figure 2.
In this Figure, we have plotted the area in the (, )Oeplane where .
| < 1 and .
| < 1.
The domain is bounded by the line = 1 which is plotted using a dashed and dotted line, the parabola: 2 4 = 0 which is plotted using a dashed line, and Oe = 1 which is plotted using a solid line.
Let us now explore the boundaries of this domain.
The boundaries are:
= 1, for 0 O O 2, .
Oe = 1, for Oe2 O < 0, and .
2 4 = 0 for Oe2 < < 2.
In 3.
1 and 3.
2 we will analyze the first two boundaries while the third will be treated in Subsection 3.
The case where = 1.
Let = 1 Oe .
Then one of the eigenvalues of A(, 1 Oe ) is 1.
The other eigenvalue is Oe 1.
The eigenvectors corresponding to this eigenvalue are:
, 0 6= OO R.
Oe1 Clearly, for 6= 2, the set of vectors:
Oe1 Two aspects of a generalized Fibonacci sequence is linearly independent.
Then for any x1 OO R and y1 OO R, we can write:
= 1
Oe1 1 = x1 y1 Oe x1 y1 Oe x1 and 2 = Oe 2Oe 2Oe Since v n = AnOe1 v 1 , we have:
v n = 1 2 ( Oe .
nOe1 1Oe Thus, the iteration .
for 0 < < 2 and = 1 Oe converges to:
x2 Oe x1 2Oe .
ince y1 = x2 ).
When = 0, then .
xn 2 = xn .
Thus, .
n } is either constant, that is when x1 = x2 , or 2-periodic when x1 6= x2 .
When = 2 then .
xn 2 = 2xn 1 Oe xn .
If x1 = x2 = OO R, then x3 = .
Oe .
= .
Then xn = , for all n OO N.
Consider:
n 2 Oe xn 1 | = .
xn 1 Oe xn Oe xn 1 | = .
n 1 Oe xn |.
OAn OO N.
Then the sequence .
does not satisfy the Cauchy criteria for convergent sequence.
Thus, the sequence .
converge for = 2 and = Oe1 if and only if x1 = x2 .
Remark 3.
Let us now consider the cases where < 0 or > 2.
In these cases, | Oe .
> 1.
Then clearly v n in .
diverges as n goes to infinity, except for 2 = 0.
Then, in these cases, the sequence .
diverges except if x1 = x2 .
The case where Oe = 1.
Let = 1 .
The eigenvalues of A(, 1 ) 1 , and Oe 1, with their corresponding eigenvectors:
Oe1 Writing:
v 1 = 1 Oe1 y1 Oe .
)x1 x1 y1 and 2 = 2 Tuwankotta Then v n = 1 .
) nOe1 2 (Oe.
nOe1 Oe1 For Oe2 < < 0 the iteration .
converges to a periodic sequence:
Oe1 2 (Oe.
nOe1 For = 0, the situation is similar with the case: = 1 for = 0.
For = Oe2, then the eigenvalue of A(Oe2.
Oe.
is Oe1 with algebraic multiplicity 2 but geometric We will address this situation in the next subsection.
Remark 3.
For < Oe2 or > 0, the term: .
)nOe1 grows without bound which implies that the iteration in .
nd hence, .
) diverges.
The case where 2 4 = 0.
For 2 4 = 0, the eigenvalue of A(, ) is 2 with algebraic multiplicity two and geometric multiplicity one.
Let
= I
and N = A(, ) Oe S.
Clearly:
S N = I A(, ) Oe I = A(, ) I Oe I = A(, ) Oe I I = N S.
Furthermore:
N2 = 2 2 =
= 0.
Since A(, ) = S N then nOe1 v n = (S N ) Lemma 3.
For arbitrary n OO N (S N ) = S n nS nOe1 N.
Proof.
The proof is done by induction on n.
For n = 2, (S N ) = S 2 SN N S N 2 = S 2 2SN.
If (S N ) nOe1 = S nOe1 .
Oe .
S nOe2 N , then (S N ) This ends the proof.
nOe1
(S N ) (S N )
(S N ) S nOe1 .
Oe .
S nOe2 N S n .
Oe .
S nOe1 N N S nOe1 .
Oe .
N S nOe2 N S n nS nOe1 N .
Oe .
S nOe2 N 2 S n nS nOe1 N.
Two aspects of a generalized Fibonacci sequence Using Lemma 3.
3 we conclude that:
nOe2 I .
Oe .
N v 1 .
Convergence of this iteration to .
, .
= .
, .
is achieved when Oe2 < < 2.
The situation for = 2 has been treated in Subsection 3.
For = Oe2, this iteration diverges, as .
Oe .
N grows without bound.
The case where 6= 1 and 2 4 < 0.
For 2 4 < 0, then we have a pair of complex eigenvalues and , which can be represented as:
= A .
os i sin ) , 2 p Oe2 Oe 4 = Oe.
From linear algebra, we know that: if 0 6= v OO C2 , satisfies:
A(, ) = v, then: V = (Re.
Im.
) OO R 2y2 , satisfies:
A(, ) = AV R()V Oe1 .
R() = cos Oe sin sin cos Then we have:
v n = A(, )nOe1 v 1 = AnOe1 V R (.
Oe .
) V Oe1 v 1 .
This iteration converges to .
, .
as n goes to infinity, if 0 < A < 1.
For A > 1, the iteration diverges as n goes to infinity.
Periodic and quasi-periodic behavior.
Let us look at the case where A = 1.
Since A = Oe then = Oe1.
Moreover:
= i Oe Oe 4 = cos i sin , then = 2 cos .
= , p OO Z, q OO N with gcd(.
, .
= 1, then .
, and hence .
n }, is q-periodic.
Thus, for p OO Z and q OO N with gcd(.
, .
= A xn 1 Oe xn , n OO N xn 2 = 2 cos is q-periodic.
6OO Q,
Tuwankotta then .
n } remains bounded for all n but the sequence diverges.
The sequence .
n } fills up densely a close curve in R2 , see Figure 4 for an example.
This behavior is known as quasi-periodic.
Conclusion.
We conclude our discussion on the convergence of .
with the following proposition.
In Figure 3 we have presented the Proposition 3.
4 geometrically.
Figure 3.
In this Figure, we have plotted the area in the (, )Oeplane where the sequence .
converges to zero.
This is done by analyzing the magnitude of 1 and 2 .
Proposition 3.
Given x1 , x2 , and real numbers.
Consider the sequence xn 2 = xn 1 xn , n OO N.
Then, if and is an element of an open domain E defined by:
Oe<1 > Oe1 At the boundary: = 1, for 0 < < 2, the sequence converges to:
x2 Oe x1 At the boundary: Oe = 1, for Oe2 < O 0, the sequence goes to a periodic y1 Oe .
)x1 .
)x1 Oe y1 Two aspects of a generalized Fibonacci sequence At the boundary: = Oe1, for Oe2 < < 2, the sequence .
takes the form:
xn 2 = 2 cos .
OA) xn 1 Oe xn , n OO N.
the sequence is periodic whenever O OO Q and quasi-periodic whenever O 6OO Q.
Examples Let us look at a view examples of iteration .
for various combination of and .
These examples are presented in Table 1.
In the first column indicates the iteration number n.
At the rows where n = 1 and n = 2, we put the initial conditions, namely x1 = 4 and x2 = 3.
For n Ou 3, xn is computed using the formula .
To give an idea on the limiting behavior of the iteration, we listed the first 13th iterations, and a few later iterations .
rom n = 64 to .
Indeed, for the purpose of illustration we have used only four decimals to represent real numbers.
Convergence to 0.
The first example is for = 0.
2500 and = 0.
This is an example for the situation where (, ) OO E, where .
converges to 0.
this example: 1 = 0.
8431 and 2 = Oe0.
This explains the slow convergence of the iteration.
See 1 column two.
The second example is for = 1.
6500 and = Oe0.
ee Table 1 column thre.
while the third example is for = Oe1.
and = Oe0.
ee Table 1 column fou.
These examples satisfy 2 4 = 0.
In the second example: 1 = 2 = 0.
8200 while in the third: 1 = 2 = Oe0.
In both of these examples, .
converges to 0, just as the first example.
The rate of the convergence in the third example is slightly better than in the first two, since the absolute value of the eigenvalue is smaller than in the first two examples.
Quasi-periodic behavior, periodic behavior, and eventually periodic behavior.
Let us consider the fifth column and the sixth column of Table 1.
In these examples, and satisfy: 2 4 < 0.
thus the eigenvalues of A(, ) are complex valued, i.
:A .
os i sin ).
In both examples.
A = 1.
The argument is approximately 0.
8956 for the example in the fifth column while, for the example in the sixth column:
OO 0.
Thus, in the fifth column a quasi-periodic behavior is displayed, while in sixth:
periodic behavior.
If the periodic behavior exactly repeating it self after seven iterations, the quasi-periodic behavior shows that after seven iterations, it becomes closed to the starting point, but not exactly the same.
Another periodic behavior is displayed in column eight.
In contrast with the periodic behavior in column sixth.
In column eight, the iterations does not seems to be periodic at the beginning.
After ten iterations it becomes periodic.
This type of behavior is called: eventually periodic.
This behavior is displayed for the choice of and that satisfy: Oe = 1.
Tuwankotta Oe0.
Oe0.
Oe0.
Oe0.
Oe1.
Oe0.
Oe6.
Oe8.
Oe7.
Oe6.
Oe4.
Oe2.
Oe1.
Oe1.
Oe0.
Oe3.
Oe3.
Oe1.
Oe0.
Oe3.
Oe3.
Oe1.
Oe1.
Oe0.
Oe3.
Oe3.
Oe1.
Oe0.
Oe3.
Oe3.
Oe1.
Oe0.
Oe1.
Oe1.
Oe1.
Oe1.
Oe1.
Oe1.
Oe1.
Oe0.
Oe0.
Oe0.
Oe0.
Oe0.
Oe3.
Oe3.
Oe1.
Oe0.
Oe3.
Oe3.
Oe1.
Oe1.
Oe1.
Oe1.
Oe1.
Table 1.
A few samples of iteration .
for various combination of (, ).
The first column indicates the iteration number n.
The second column is for (, ) that satisfies: 2 4 > 0, and both .
| < 1 and .
| < 1.
The third and fourth columns are for 2 4 = 0.
The fifth column is an example of quasi-periodic behavior, while the sixth column is 7-periodic behavior.
These are for 2 4 < 0.
The seventh column is for = 1, while the eighth is for Oe = 1.
Convergence to a nontrivial limit.
As the last example to be discussed, we have choose: = 0.
3000 and = 0.
Clearly, the choice of and in this example, satisfy: = 1.
In this case, a nontrivial limit exists, i.
x2 Oe x1 3Oe4 OO 3.
2Oe 2 Oe 0.
Two aspects of a generalized Fibonacci sequence Oe1 Oe2 Oe3 Oe4 Oe5 Oe5 Oe4 Oe3 Oe2 Oe1 Figure 4.
In this figure we have plotted an ellipse which is formed by 5000 iterations of .
for = 1.
25 and = Oe1.
The initial conditions are x1 = 4 and x2 = 3.
This iterations densely fill up an ellipse.
The horizontal axis is xn while the vertical axis is xn 1 .
On the ellipse, there are five 7-periodic sequences .
ee Definition .
which are obtained by 150 iterations of .
for = 2 cos 2A and = Oe1, but five different sets of initial conditions: .
1 = 4589, x2 = 0.
(), .
1 = Oe2.
2150, x2 = Oe4.
(),
1 = Oe3.
3212, x2 = Oe0.
(?), .
1 = 2.
4215, x2 = 4.
), and .
1 = 3.
1707, x2 = 3.
().
The eigenvalues of A(, ) are 1 and Oe0.
thus the convergence is similar to the first three examples.
Sequence of Ratios of the Generalized Fibonaci Sequence Let us now look at the ratios of the Generalized Fibonaci sequence .
Dividing .
by xn 1 we have:
assuming xn 6= 0 for all n.
Writing: An = xxn 1 , we have the following recursive , n OO N.
An 1 = An Note that if .
converges, then the limiting point satisfies:
A2 Oe A Oe = 0.
Tuwankotta which is also the equation for eigenvalues of A(, ).
Thus, if .
converges, it converges to an eigenvalue of A(, ).
This is not surprising, since the ratio: An AymeasuresAy the growth of xn , while the eigenvalue measures the growth of v n .
As a sequence of real numbers, it might be interesting to discuss the convergence of .
for general value of and .
However, we are interested on using .
as an alternative way to construct rational approximation for radicals.
Thus, our interest is restricted to the cases where: 2 4 > 0.
In particular, for > 0, > 0, x0 > 0 and y0 > 0, then .
defines a monotonic increasing sequence.
As a consequence.
An > 1 for all n.
Then we have, the sequence .
converges if < 1.
Let < 1, then the sequence .
converges to a positive root of: A2 Oe A Oe , i.
for every > 0.
A1 > 1 as long as 0 < < 1.
The sequence Oo .
can be used for constructing an approximation for an algebraic number: r, 1 < r OO Q, by setting: r = 2 4, where and are rationals.
To do this, we translate the sequence .
by setting rn = 2An Oe .
Then .
rn 1 = rn which converges to: 2 4.
Oo For example, we want to construct an approximation for 7.
We need to find a rational solution for: 7 = 2 4.
Then we can set:
= and = 2.
This solution is clearly not unique.
By remembering that: 7 can be written as , 2 , or 2 , we can choose to be , or while is:
, , or The result is presented in Table 2.
The results in Table 2 is an agreement with our analysis.
The convergence of .
depends on the value of .
When is closer to 0.
75 then it takes 18 iteration to converge, while when = 0.
06 then it takes only 8 iteration.
Furthermore, the convergence seems to be independent with the distance of the initial condition to the actual value of the radical.
Two aspects of a generalized Fibonacci sequence r1 = 8 r2 = 1.
4 = 0.
16 = 0.
18 OO 0.
50 = 0.
Iteration Error 10Oe15 10Oe15 10Oe16 10Oe17 Iteration Error 10Oe15 10Oe15 10Oe16 10Oe17 Table In this table we present some results on approximating Oo 7 by .
The rate of convergence.
Let f .
= x2 Oe x Oe , then f 0 .
= 2x Oe , and f 00 .
= 2.
Suppose that A is the limit of .
and let en = A Oe An , n OO N.
Then = f (A) = f .
n An ) = f (An ) f 0 (An )en f 00 (An )en 2 O.
n 3 ) Thus, f (An ) 2An Oe =Oe en Oe en O.
n ) = Oe 2 Oe en O.
n 2 ).
An An An An Since:
en 1 = A Oe An 1 = A Oe Oe f (An ) f (An ) = A Oe An = en An An An Oe 1 en O.
n 2 ).
An Thus, in contrast with the quadratic convergence of Newton iteration, the convergence of .
is linear.
en 1 = Generalization to higher dimension.
Interesting generalization to higher dimension is as the following.
Let us consider:
xn 3 = xn 2 xn 1 xn , .
Tuwankotta where n OO N.
Let us assume that the initial conditions are positive, i.
: x1 > 0, x2 > 0 and x3 > 0.
Furthermore: , and are also positive.
If at least one of them is greater than one, then .
is monotonically increasing.
Dividing .
by xn 2 , gives:
, we have:
An = xxn 1 An 2 = An 1 An An 1 Thus, the limit of .
, if exists, satisfies:
A3 Oe A2 Oe A Oe = 0.
Let p = Oe Oe , and q = Oe Oe Oe , and furthermore:
P =
Oe and Q = Oe Oe Then, the solution for .
P Q 3 ,
P Q
Oei 3 (P Q) , and P Q 1Oo 3 (P Q) .
Oe i This is known as CardanoAos solution for cubic equation.
OI=
> 0,
then the equation .
has a unique real solution.
Let us look at an example, where: = 3, = 13 , and = 91 .
For these , while OI = 729 Then, the Real solution choices, we have: p = Oe 23 and q = Oe 27 for .
, is Oo Oo r = 4 2 2 4 Oe 2 2 OO 1.
By generating the sequence .
, and then computing the ratio, after 22 iterations the same accuracy is achieved for approximating r.
Two aspects of a generalized Fibonacci sequence Concluding remarks We have studied the convergence of the two terms recurrent sequence, also known as the generalized Fibonacci sequence.
We have derived an open domain E around the origin of (, )-plane, where the sequence converges to 0.
On the boundary of that domain, interesting behavior, such as: convergent to a nontrivial limit, period behavior or quasi-periodic behavior have been analyzed.
There are still possibilities that for (, ) which are not in E the sequence converges.
However, we have to choose carefully the initial conditions.
Interesting application that we have proposed is an Oo alternative for constructr, where r OO Q, i.
ing a rational approximation to algebraic numbers:
A slight generalization is by considering .
to approximate algebraic numbers of the form:
Oo Oo a b c aOeb c d In contrast with the previous formula .
), the computation is less cumbersome if one construct the sequence using .
, and then directly construct the ratios.
The proof for the convergence of iteration .
is still open.
In the next study, our attention will be devoted on proving the convergence of .
, for some value of , and .
References