Volume 2017, Issue 1 7163809
Research Article
Open Access

Nonnegative Periodic Solutions of a Three-Term Recurrence Relation Depending on Two Real Parameters

Yen Chih Chang

Yen Chih Chang

Department of Industrial Engineering and Industrial Management, National Tsing Hua University, Hsinchu 30013, Taiwan nthu.edu.tw

Search for more papers by this author
Sui Sun Cheng

Corresponding Author

Sui Sun Cheng

Department of Mathematics, National Tsing Hua University, Hsinchu 30013, Taiwan nthu.edu.tw

Search for more papers by this author
Wei Chang Yeh

Wei Chang Yeh

Department of Industrial Engineering and Industrial Management, National Tsing Hua University, Hsinchu 30013, Taiwan nthu.edu.tw

Search for more papers by this author
First published: 06 September 2017
Academic Editor: Zhan Zhou

Abstract

Simple dynamic systems representing time varying states of interconnected neurons may exhibit extremely complex behaviors when bifurcation parameters are switched from one set of values to another. In this paper, motivated by simulation results, we examine the steady states of one such system with bang-bang control and two real parameters. We found that nonnegative and negative periodic states are of special interests since these states are solutions of linear nonhomogeneous three-term recurrence relations. Although the standard approach to analyse such recurrence relations is the method of finding the general solutions by means of variation of parameters, we find novel alternate geometric methods that offer the tracking of solution trajectories in the plane. By means of this geometric approach, we are then able, without much tedious computation, to completely characterize the nonnegative and negative periodic solutions in terms of the bifurcation parameters.

1. Introduction

Simple dynamic systems representing time varying states of interconnected compartments or “neurons” may exhibit extremely complex behaviors when bifurcation parameters are switched from one set of values to another. An example has been given in several of our previous studies [1] and a slightly modified model of which is described as follows. Let n neurons be placed on the vertices of a regular n-gon and let the time dependent state values of the neurons be denoted by ui(t), for i = 1,2, …, n. Suppose the rate of change is determined by the resultant effect of a constant multiple of u2(t), plus a magnified on-off (or bang bang) state dependent control mechanism as well as its two near neighbors u1(t), u3(t) in the form
(1)
Then assuming “uniformity” among the neuron interactions, we have
(2)
where u0(t) is identified with un−1(t) while un+1(t) with u1(t) for compatibility reasons. Here H(x) is the on-off Heaviside step function defined by
(3)

When the real parameters ρ and μ are switched among different real values, it is expected (and verified by simulations) that complex dynamic behaviors will be more than abundant. Some of these behaviors can be explained (see [13]) but some not (at least to the best of our knowledge). Similar models of piecewise constant dynamic systems which exhibit similar behaviors with parameters can be found in many recent investigations; see for examples [410] and the references therein.

In this note, we discuss the steady state solutions of the above dynamic system (i.e., those that satisfy uk(t) ≡ φk for all t and for each k = 1,2, …, or n). Then we will face the existence problem of periodic solutions of the following three-term recurrence relation:
(4)
In [13], we are lucky to obtain complete information about the periodic solutions of (4) when μ = 1 and ρ = 1, 0, or −1 by breaking the solutions into two sequences, a companion and an error (more specifically, for a solution of (4) with μ = 1 and ρ ∈ {−1,0, 1}, the companion sequence is which is an even integral sequence so that λk = 2[(1/2)φk], where 2[x/2] is the greatest even integer that is less than or equal to xR, and the error sequence is defined by ϵk = φkλk,   kZ).

Yet, for other values of μ and ρ, simulations show complex periodic behaviors beyond our present comprehension, except when the periodic solutions are also “nonnegative” (or “negative”). Such exceptional results, when examined more closely, can be explained.

In this paper, we will devote ourselves to explaining the behaviors of these nonnegative (or negative) and periodic solutions of (7).

First, a real sequence is a solution of (4) if it renders (4) into an identity after substitution. It is nonnegative (or negative) if all its terms are nonnegative (respectively negative). If is a nonnegative solution of (4), then clearly ζ satisfies
(5)
while if ξ is a negative solution of (4), then
(6)
Hence it seems that we are back to the usual nonhomogeneous linear second-order difference equations studied in elementary theory of difference equations. Since μR in (5) and (6), we can further restrict ourselves in the sequel to the following difference equation:
(7)
with arbitrary μR. By means of the techniques of general solutions plus the method of variation of parameters, it may be argued that the existence of nonnegative (respectively negative) periodic solutions can be handled completely. Such an assertion may be true in theory, but the general solution here is ρ- as well as μ-dependent and hence actual attempts lead to many complications.
That said, in this paper, we will handle our equation from a novel approach and the crux of which is based on representing each pair of two consecutive terms (φk, φk+1) of a solution as a point in the plane and invent a geometric method to track the movements of these points. First, the totality of such pairs is called the orbit of this solution. More precisely, let be a solution of (4). We define the orbit of φ by
(8)
and the “positive” orbit by
(9)
As examples, let μ = 1 and let ζ and ξ be solutions of (4) with ρ = 1 or 0, respectively. The orbits with (ζ0, ζ1) = (9,1) and with (ξ0, ξ1) = (15/2,1/2) are depicted in Figures 1 and 2, respectively. These figures clearly suggest that ζ is “28-periodic,” that ξ is “30-periodic,” and that there are accompanying “distinctive” features which can be exploited further.
Details are in the caption following the image
The orbit of ζ with ρ = 1.
Details are in the caption following the image
The orbit of ξ with ρ = 0.
Several sets will be encountered in the ensuing discussions and we denote them as follows:
  • (i)

    R≔(−, ) is the entire set of real numbers.

  • (ii)

    Z≔{…, −1,0, 1, …}  is the set of integers

  • (iii)

    Z+≔{1,2, 3, …}  is the set of positive integers.

  • (iv)

    N≔{0,1, 2, …}  is the set of nonnegative integers.

  • (v)

    Q≔{±m/nmN, nZ+}  is the set of rational numbers.

Before we enter into discussions on the necessary and sufficient conditions for a solution of (7) to be nonnegative and periodic, we take a note that any solution of our three-term equation (7) is uniquely determined by two of its consecutive terms. Furthermore, in view of the fact that (7) can be written as
(10)
as well as
(11)
it is both “symmetric” (or “reflection invariant”) and “translation invariant.” More precisely, let be the solution of (7) that satisfies φ0 = α and φ1 = β. Then the solution of (7) determined by ξ0 = α and ξ−1 = β, called the reflection of φ and denoted by ξ = Uφ, will satisfy
(12)
while the solution of (7) determined by ζk = α and ζk+1 = β for any kZ, called the k-translation of φ and denoted by ζ = Ekφ, will satisfy
(13)
In view of these invariances, to study (7), we may simply concentrate our attention on the positive orbits of its solutions! Indeed, let be the solution of (7) with (ψ0, ψ1) = (σ, τ). We investigate {ψ0, ψ1, ψ2, …} by observing the positive orbit defined by (σ, τ). On the other hand, we analyze {ψ1, ψ0, ψ−1, ψ−2, …} by studying the positive orbit Note that for all , where n < 0, (ψn+1, ψn) = Q|n|, where In the sequel, if no doubt arises, we frequently let be a solution of (7) which is defined by (φ0, φ1) = (α, β) and we denote
(14)
so that
(15)

Proposition 1. Let be a solution of (7) defined by (φ0, φ1) = (α, β) and be the “positive” orbit of φ. Then ω is a period of φ if and only if (xω, yω) = (α, β). Furthermore, if limk|xk| = , then φ is aperiodic and if limkxkyk < 0, then φ is neither a nonnegative nor a negative solution.

The statements of Proposition 1 can be verified directly from the definitions, as well as the invariant properties of (7), and, thus, the proof is omitted here.

2. Some Basic Results of (7)

Periods and the prime periods of sequences are defined as usual. Furthermore, let be a nonnegative solution of (7). Note that for some distinct m, nN, if (φm, φm+1) = (φn, φn+1), then |mn| is a period of φ and the least one ω among the periods is the least or prime period of φ (we also say that φ is ω-periodic). A constant solution of (7) is 1-periodic. We first find the necessary and sufficient conditions for φ to be a constant solution.

Proposition 2. Let be a solution of (7) with (φ0, φ1) = (α, β).

  • (i)

    Suppose ρ ≠ −2. Then φ is a constant solution if and only if α = β = μ/(2 + ρ).

  • (ii)

    Suppose ρ = −2. Then φ is a constant solution if and only if μ = 0 and α = β.

Proof. Suppose ρ ≠ −2. If φ is a constant solution of (7), then α = β = φ2 and

(16)
so that
(17)
Accordingly, α = μ/(2 + ρ) = β. The converse can be verified by direct iteration. Next, suppose ρ = −2. If φ is a constant solution of (7), then α = β and by (7)
(18)
which leads to μ = 0. Conversely, if μ = 0 and α = β, then by iteration φ is clearly a constant solution.

The proof is complete.

Corollary 3. Let be a solution of (7) with φ0 = φ1 = γ. Let γ = μ/(2 + ρ) where ρ ≠ −2. Then φ is 1-periodic. φ is a nonnegative solution if and only if γ ≥ 0; and φ is a negative solution if and only if γ < 0.

The results can be obtained from Proposition 2 directly and, hence, we omit the proof.

In view of Proposition 2, it is easy to see that the orbit of a constant solution is just the set containing the only point (μ/(2 + ρ), μ/(2 + ρ)).

Proposition 4. Let be a solution of (5) with (φ0, φ1) = (α, β). Then φ is 2-periodic if and only if ρ = 2, α + β = μ/2, and αβ.

Proof. Suppose φ is 2-periodic. Then φ2 = α as well as φ3 = β and by (5), φ2 = μρβα = α and φ3 = μραβ = β which lead us to 2α + ρβ = μ and ρα + 2β = μ, respectively. Accordingly, we have (2 + ρ)(α + β) = 2μ and (2 − ρ)(αβ) = 0. Note that since φ is 2-periodic, we can be sure that αβ. Hence, if ρ ≠ 2, then αβ = 0 which is a contradiction; if ρ = 2, then α + β = μ/2 and αβ. The converse may be checked by direct substitution into (5) and this completes the proof.

Corollary 5. Let be a solution of (7) with ρ = 2 and (φ0, φ1) = (α, β). Suppose α + β = μ/2. Then φ is periodic with period 2.  φ is nonnegative if and only if α, β, μ ≥ 0; and φ is negative if and only if α, β, μ < 0.

Proof. Suppose α + β = μ/2. First of all, if α = β = μ/4, then, by Proposition 2, φ is 1-periodic; otherwise, by Proposition 4, φ is 2-periodic. Hence, φ is periodic with period 2. If φ is a nonnegative solution, then α,  β ≥ 0 which implies μ ≥ 0. Furthermore, since φ is periodic with period 2, we can see that φk ≥ 0 for all kZ. If α, β, μ ≥ 0, then α + β = μ/2 is satisfied and φ is a nonnegative solution. The case where φ is a negative solution can be handled similarly and this completes the proof.

In this paper, since we are interested in the periodicity of the solutions of (4) which are also nonnegative (or negative), the following results will be useful.

Proposition 6. Let be a nonnull solution of (7). If μ ≥ 0 and ρ ≥ 0, then φ cannot be a nonpositive solution; if μ ≤ 0 and ρ ≤ 0, then φ cannot be a nonnegative solution.

Proof. Suppose μ ≥ 0 and ρ ≥ 0. If φ is nonpositive, then 0 ≥ φ2 = −ρφ1φ0 + μ ≥ 0 and 0 ≥ φ3 = −ρφ2φ1 + μ ≥ 0. Hence φ2 = φ3 = 0, which implies φ is null, contrary to our assumption.

Suppose μ ≤ 0 and ρ ≤ 0. If φ is nonnegative, then 0 ≤ φ2 = −ρφ1φ0 + μ ≤ 0 and 0 ≤ φ3 = −ρφ2φ1 + μ ≤ 0 so that φ2 = φ3 = 0. A contradiction is arrived and this completes the proof.

Proposition 7. φ is a solution of (5) if and only if −φ is a solution of (6).

Proof. Suppose that is a solution of (5) with (φ0, φ1) = (α, β). By (5), we have φ2 = μρβα. Let ψ = −φ such that (ψ0, ψ1) = (−α, −β). Accordingly, ψ2 = −φ2 and we see that

(19)
which satisfies (6). By induction, it follows that ψ is a solution of (6) and this completes the proof.

Next, we discuss nonconstant solutions of (5) (although 2-periodic solutions are discussed in Proposition 4, we may include them in the following discussions). The behavior of solution φ of (7) with least period 1 (or 2) was quite easy to analyze directly, yet it is difficult to conduct similar analyses for φ with larger least period (or φ is aperiodic) by similar manner. Hence, we will investigate the behavior of φ from a new perspective.

3. The Orbits of Solutions of (7)

We will need the following quadratic function:
(20)
and its f-level curve Γμ,ρ,f is defined by
(21)
Since Γμ,ρ is a quadratic function, its level curves are plane conic sections. The properties of these conic sections are well known. First, we see that the x = y line is one of the principal axes of Γμ,ρ,f and if ρ ≠ −2; then (μ/(2 + ρ), μ/(2 + ρ)) is the center of Γμ,ρ,f and the other principal axis is the x + y = 2μ/(2 + ρ) line. In particular (see, e.g., [11]) Γμ,ρ,f can be classified by the value of its discriminant ρ2 − 4 and the values (ρ − 2)μ2 as well as f:
  • (1)

    If |ρ| > 2, then Γμ,ρ,f is a hyperbola (see the green and red curves in Figures 3 and 4) when f ≠ −μ2/(2 + ρ) or a degenerate hyperbola consisting of two intersecting lines when f = −μ2/(2 + ρ) (see the blue lines in Figures 3 and 4).

  • (2)

    If |ρ| < 2, then Γμ,ρ,f is an ellipse (see Figure 5) when f ≠ −μ2/(2 + ρ) or a degenerate point ellipse when f = −μ2/(2 + ρ).

  • (3)

    If |ρ| = 2, then Γμ,ρ,f is a parabola when ρ = −2 or a degenerate parabola consisting of a single or two distinct parallel lines when ρ = 2 (see the green and blue lines in Figure 6).

Details are in the caption following the image
The hyperbolae with ρ > 2.
Details are in the caption following the image
The hyperbolae with ρ < −2.
Details are in the caption following the image
The ellipses with −2 < ρ < 2.
Details are in the caption following the image
The parabolae with |ρ| = 2.

Unless indicated by the adjective “degenerate,” a conic section is meant to be nondegenerate. The conic sections Γμ,ρ,f are symmetric with respect to the x = y line. This property is a simple consequence of the fact that Γρ,f(x, y) is a symmetric function in x and y. Hence, the x = y line is one of the principal axes of the conic sections.

To discuss the principal axes further, we define the two variable functions:
(22)
and the associated line Also, for ρ ≠ −2, we let
(23)
and Furthermore, let f = Γμ,ρ(α, β), where (α, β) ∈ R2 and for |ρ| > 2. We define the following two variable functions:
(24)
as well as
(25)
with and Note that in view of (24) and (25), as well as are independent from f when |ρ| > 2.
Let SR2. Define
(26)
Then S and S are symmetric with respect to the x = y line, while S and S are symmetric with respect to the x + y = 0 line. The boundary of the plane set S will be denoted by S.

The conic sections will be of great help in the analysis of solutions of (4). The most significant result is the following.

Theorem 8. Let be a solution of (5). If f = Γμ,ρ(φm, φm+1) for some mZ, then Furthermore, if (α, β) ∈ Γμ,ρ,f, where f = Γμ,ρ(α, β), then

Proof. Without loss of generality, we let m = 0 and (φ0, φ1) = (α, β). Let f = Γμ,ρ(α, β). Then (α, β) ∈ Γμ,ρ,f. By (21), we see that

(27)
and by (7), it follows (φ1, φ2) = (β, μρβα) so that
(28)
which leads to (φ1, φ2) ∈ Γμ,ρ,f. By induction, for any nN, (φn, φn+1) ∈ Γμ,ρ,f which implies Next, we consider (φn, φn+1), where nZN. Let such that (ψ0, ψ1) = (β, α) and φn = ψn+1. Since Γμ,ρ,f is symmetric with respect to the x = y line, (β, α) ∈ Γμ,ρ,f and by the previous discussions, we can be sure that which implies , where Accordingly, and by (15), it follows that Furthermore, by the definitions, if (α, β) ∈ Γμ,ρ,f, where f = Γμ,ρ(α, β), then and since itself is symmetric with respect to the x = y line, it is clear as desired. The proof is complete.

Let be a solution of (5). Then φ is uniquely determined by two consecutive terms φm and φm+1. Although we can calculate the next term φm+2 directly, the conic sections allow us to easily “track” the movements of the points from (φm, φm+1) to (φm+1, φm+2). Indeed, by Theorem 8, we see that , where f = Γμ,ρ(φm, φm+1). Then by means of this conic section, we may plot the corresponding orbit as follows.

Tracking Procedure. Input (α, β) ∈ R2.
  • (1)

    Plot Γμ,ρ,f where f = Γμ,ρ(α, β) on the plane.

  • (2)

    Take k = 0 and start at the point Pk = (xk, yk) = (α, β).

  • (3)

    Let be the point on Γμ,ρ,f which is symmetric to Pk with respect to the x = y line.

  • (4)

    Draw a vertical line Lk through

  • (5)

    The line Lk and the conic section Γμ,ρ,f can intersect at and possibly at another point If , then let ; otherwise, take

  • (6)

    If Pk+1 = P0, then stop; otherwise, let Pk = Pk+1 and go to step (3).

Note that by the Principal Axes Theorem, the level curve Γμ,ρ,f is defined by setting Γμ,ρ = f, which is a polynomial of degree 2. The conic section has one principal on the x = y line which implies any vertical line through meets Γμ,ρ,f at least once and at most twice through two points or If , then is on the ρx + 2y = μ line and this can be verified directly by (7). In view of the Tracking Procedure, we can generate each Pk = (xk, yk) as (φk, φk+1) by initializing P0 = (x0, y0) = (φ0, φ1) and it is not difficult to see that the above algorithm can yield the “positive” orbit ; on the other hand, the “negative” orbit of can be plotted by the “positive” orbit which is generated by the Tracking Procedure with input (φ1, φ0). In Figures 7 and 8, we illustrate the orbits of Pk on ellipses; Figures 9 and 10 depict the orbits on a parabola and a degenerate parabola correspondingly; and in Figures 11 and 12, we show the orbits on a degenerate hyperbola and a hyperbola, respectively.

Details are in the caption following the image
The orbit on .
Details are in the caption following the image
The orbit on .
Details are in the caption following the image
The orbit on .
Details are in the caption following the image
The orbit on .
Details are in the caption following the image
The orbit on .
Details are in the caption following the image
The orbit on .

As we have shown previously, the f-level curve Γμ,ρ,f of Γμ,ρ is a conic section and it can be classified into hyperbola, ellipse, or parabola (or the degenerate curves). Let be a solution of (7) defined by (φ0, φ1) = (α, β). By means of the Tracking Procedure, we can easily analyze the asymptotic behavior of φk when k. In the sequel, we will discuss the sequence which is generated by the Tracking Procedure with input (x0, y0) = (α, β) depending on |ρ| > 2, |ρ| = 2, and |ρ| < 2.

We first consider the case where |ρ| > 2. By the previous discussions, the points generated by the Tracking Procedure with input (x0, y0) move on Γμ,ρ,f where f = Γμ,ρ(x0, y0) and Γμ,ρ,f is a hyperbola (or a degenerate one made up of two lines). Since Γμ,ρ,f is a hyperbola, we first locate the asymptotes of Γμ,ρ,f. Note that Pc = (μ/(2 + ρ), μ/(2 + ρ)) is the center of Γμ,ρ,f. In view of (21), we rewrite the equation Γμ,ρ,f as
(29)
and suppose y = m±x + n± are the oblique asymptotes of Γμ,ρ,f. Then we have
(30)
As for n±, since lines y = m±x + n± intersect at Pc, we apply Pc to the lines and obtain n± = (μμm±)/(2 + ρ). Accordingly, we see that and can be expressed by ymxn = 0 as well as ym+xn+ = 0, respectively. For |ρ| > 2, one relation between Pk and as well as defined in (24) and (25), respectively, is revealed by
(31)
as well as
(32)
Moreover, with the oblique asymptotes and , we may illustrate the asymptotic behaviors of Γμ,ρ,f where |ρ| > 2.

Theorem 9. Suppose |ρ| > 2 and let (α, β) ∈ R2 such that f = Γμ,ρ(α, β). Let be generated by the Tracking Procedure with input (x0, y0) = (α, β). Then Γμ,ρ,f is a hyperbola (or a degenerate one made up of two lines and ) and we have the following.

  • (i)

    Suppose Γμ,ρ,f is a degenerate hyperbola. If , then for all mN; if , then for all nN.

  • (ii)

    Suppose Γμ,ρ,f is a hyperbola. If ρ ≥ 0, then (xm, xm+1) and (xm+1, xm+2) are sitting on different branches for all mN; if ρ < 0, then (xn, xn+1) and (xn+1, xn+2) lie on the same branch for all nN.

Proof. Note that x0 = α and x1 = y0 = β. By Theorem 8 and (7), we see that y1 = μρy0x0 = μρβα = γ. Suppose or In the former case, and by (31), as well as (32), which implies ; in the latter case, by similar arguments, we may see that Note that if , then (α, β) = (μ/(2 + ρ), μ/(2 + ρ)) and by the Tracking Procedure, (x1, y1) = (μ/(2 + ρ), μ/(2 + ρ)) which implies (xi, yi) = (μ/(2 + ρ), μ/(2 + ρ)) for all iN. Suppose Then Γμ,ρ,f is a hyperbola and (α, β) is on one branch of Γμ,ρ,f. By the Tracking Procedure, we may see that if ρ ≥ 0, then P1 “jumps” to another branch (see Figure 3), while if ρ < 0, then P1 will stay on the same branch (see Figure 4). By induction, we may see that (ii) holds. This completes our proof.

In view of Theorem 9, we may follow the movements of (xk, yk) generated by the Tracking Procedure when |ρ| > 2. This enables us to see the asymptotic behavior of (xk, yk) when k in Theorem 10. Before introducing Theorem 10, we denote the distance between two points P1 and P2 on R2 by |P1P2| and the distance from a point P to the line L by d(P, L).

Theorem 10. Suppose |ρ| > 2 and let (α, β) ∈ R2 such that f = Γμ,ρ(α, β). Let be generated by the Tracking Procedure with (x0, y0) = (α, β). Then Γμ,ρ,f is a hyperbola (or a degenerate one made up of two lines and ) and we have the following.

  • (i)

    Suppose If (α, β) = (μ/(2 + ρ), μ/(2 + ρ)), then (xk, yk) = (α, β) for all kN; if and (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)), then

    (33)

  • and if and (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)), then

    (34)

  • (ii)

    If ρ > 2 and Γμ,ρ,f is a hyperbola, then ; and if ρ < −2 and Γμ,ρ,f is a hyperbola, then

Proof. For the sake of convenience, we let Pc = (μ/(2 + ρ), μ/(2 + ρ)) and Pk = (xk, yk) so that P0 = (α, β) and f = Γμ,ρ(α, β). Let y1 = γ = μρβα. By Proposition 2, if P0 = Pc, then φ is a constant solution of (7) which implies Now, we suppose P0Pc. Since |ρ| > 2, Γμ,ρ,f    is a hyperbola (or a degenerate one). Hence, we discuss the two cases where P0 is on a degenerate hyperbola or a hyperbola.

Suppose and P0Pc. Then by Theorem 9, and are on the same line. In view of (24) and (25), we let m+ = ρ + δ,  m = ρδ, Δ+ = μ + δ/(2 + ρ), and Δ+ = μδ/(2 + ρ). For the sake of convenience, we also let

(35)
Since Γμ,ρ,f is symmetric with respect to the x = y line, we see that if ρ > 2, then κ < 1; if ρ < −2, then κ > 1. Suppose ρ > 2. If and P0Pc, then and so that α + m+β + Δ+ = 0 and γ + mβ + Δ = 0, respectively. Also, by the symmetry of Γμ,ρ,f with respect to the x = y line, we can be sure that so that
(36)
On the other hand, we calculate
(37)
and by (35) and (36), as well as (37), it follows that
(38)
By induction, we have , where κ < 1 and iN. Accordingly, (34) holds. The case where ρ < −2 and can be handled similarly with , where κ > 1 and jN. By similar arguments, (33) is also true.

Suppose We first consider the case where ρ > 2. Then κ < 1 and by (24)

(39)
By induction, it follows that , where iN. Consequently, case (ii) is true when ρ > 2 and by similar arguments, we also see that ρ < −2 holds with , where jN and κ > 1.

The proof is complete.

Corollary 11. Suppose |ρ| > 2 and let (α, β) ∈ R2 such that f = Γμ,ρ(α, β). Let be generated by the Tracking Procedure with input (x0, y0) = (α, β). Then Γμ,ρ,f is a hyperbola (or a degenerate one made up of two lines and ). Suppose (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)). If , where ρ > 2, then xkxk+1 < 0 and |xk| → when k; if , where ρ < −2, then xkxk+1 > 0 and |xk| → when k.

The argument of Corollary 11 can be verified directly from Theorem 10 based on the facts that the slope of is characterized by −ρ + δ < 0 and the slope of is characterized by −ρδ > 0, where ρ > 2 and ρ < −2, respectively, and, thus, the proof is omitted.

In Theorem 10, we showed the asymptotic behaviors of generated by the Tracking Procedure. We illustrate the movements of when k in Figures 13, 14, and 15 where the arrows indicate the moving directions of Pk as k.

Details are in the caption following the image
The asymptotic behavior of {Pk} plotted by Tracking Procedure with ρ > 2.
Details are in the caption following the image
The asymptotic behavior of {Pk} plotted by Tracking Procedure with ρ < −2.
Details are in the caption following the image
The asymptotic behavior of {Pk} plotted by Tracking Procedure with ρ = −2.

Next, we consider the asymptotic behaviors of with ρ = 2 or ρ = −2.

Theorem 12. Suppose ρ = 2 and let (α, β) ∈ R2 such that f = Γμ,2(α, β). Let be generated by the Tracking Procedure with input (x0, y0) = (α, β). Then f ≥ −μ2/4 and Γμ,ρ,f is a degenerate parabola made up of and . If f = −μ2/4, then and (xk, yk) = (xk+2, yk+2) for all kN; otherwise, (xk, yk) as well as (xk+1, yk+1) are not on the same line. Furthermore, if f > −μ2/4, then and xkyk < 0 when k.

Proof. For the sake of convenience, we let Pc = (μ/(2 + ρ), μ/(2 + ρ)) and Pk = (xk, yk). First of all, we see that

(40)
which implies f ≥ −μ2/4. Furthermore,
(41)
which leads to Γμ,2,f being two parallel lines. Consequently, if f = −μ2/4, then ; otherwise, and are two distinct parallel lines. Suppose f = −μ2/4. Then In view of the Tracking Procedure, it is found that if P0 = Pc, then P1 = P0 which implies Pm = P0 for all mN; otherwise, which leads to Pn = Pn+2 for all nN. Next, suppose f > −μ2/4.    By the Tracking Procedure, we can directly observe (see Figure 10) that for an arbitrary nN, if , then and vice versa. Now, we discuss the asymptotic behaviors of By the Tracking Procedure, we first observe that
(42)
and clearly, the converses are also true. Note that d(Pn, Π) ≠ d(Pn+1, Π) for any nN. Furthermore, by the Tracking Procedure again, if , then for all kN
(43)
otherwise, we see that for some jN,
(44)
In view of (44), since ,  , and is a positive constant, there is iN such that as well as and, thus, by the previous discussions, Accordingly, by the previous discussions again, when k and since the slopes of and are −1, xkyk < 0 as k. The proof is complete.

Next, we consider the case where ρ = −2.

Theorem 13. Suppose ρ = −2 and let (α, β) ∈ R2 such that f = Γμ,−2(α, β). Let be generated by the Tracking Procedure with input (x0, y0) = (α, β). Then we have the following.

  • (i)

    Suppose μ = 0. Then f ≥ 0 and Γ0,−2,f is a degenerate parabola made up of two parallel lines and If f = 0, then and α = β such that (xk, yk) = (α, α) for all kN; otherwise, (xk, yk) and (xk+1, yk+1) are on the same line and d((xk, yk), P0) → when k.

  • (ii)

    Suppose μ ≠ 0. Then Γμ,−2,f is a parabola and when k.

Proof. Suppose μ = 0. By (20), Γ0,−2(x, y) = x2 − 2xy + y2 = (xy)2 and, thus, f = Γ0,−2(α, β) = (αβ)2 ≥ 0. Accordingly, Γ0,−2,f is a degenerate parabola which is made up of two parallel lines and as defined in (24) and (25), respectively. More precisely, if α = β, then f = 0 which leads to ; if αβ, then f > 0 which implies and are two distinct parallel lines. By the Tracking Procedure, if α = β, then (xk, yk) = (α, α) for all kN; if αβ, then (xk, yk) and (xk+1, yk+1) are on the same line. Furthermore, if αβ, then we have and by the Tracking Procedure again, d((xk, yk), (α, β)) = kΔtan⁡(π/4) = kΔ, where kN. Accordingly, d((xk, yk), P0) → when k. Next, suppose μ ≠ 0. Then Γμ,−2,f is parabola. By (7), for kN, we have yk = xk+1 such that xk+2xk+1 = μ + xk+1xk and thus

(45)
which leads to
(46)
Hence, when k. Note that, if μ > 0, then Γμ,−2,f is concave upward (see Figure 15), while if μ < 0, then Γμ,−2,f is concave downward. Accordingly, (ii) is true and this completes the proof.

Corollary 14. Suppose ρ = −2 and let (α, β) ∈ R2 such that f = Γμ,−2(α, β). Let be generated by the Tracking Procedure with input (x0, y0) = (α, β). Then we have the following.

  • (i)

    Suppose μ = 0. Then f ≥ 0 and Γ0,−2,f is a degenerate parabola with two parallel lines, and If α = β, then f = 0 such that and xk = α for all kN; otherwise, f > 0 such that as well as are two distinct parallel lines and

    (47)

  • (ii)

    Suppose μ ≠ 0. Then Γμ,−2,f is a parabola. If μ > 0, then Γμ,−2,f is concave upward and xk when k; otherwise, Γμ,−2,f is concave downward and xk → − when k.

The results stated in Corollary 14 can be derived directly from Theorem 13 by observing the asymptotic behavior of the points (xk, yk).

Let be generated by the Tracking Procedure with inputs (x0, y0) = (α, β) and f = Γμ,ρ(α, β). In Theorems 10, 12, and 13, we have shown the asymptotic behaviors of φ with |ρ| ≥ 2. Next, we consider the case where |ρ| < 2 and (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)). Suppose |ρ| < 2. Then the conic section Γμ,ρ,f is an ellipse which implies xk is both bounded above and below. Moreover, by the Principal Axes Theorem, the nondegenerate Γμ,ρ,f is an ellipse which is rotated by ϑ ∈ {−π/4, π/4} and translated by (xc, yc) = (μ/(2 + ρ), μ/(2 + ρ)) from a standard ellipse with major axis a as well as minor axis b. Thus, Γμ,ρ,f where f ≠ −μ2/(2 + ρ)2   can be expressed parametrically by
(48)
where the polar angle θ ∈ [0,2π]. For the sake of convenience, we let
(49)
where By the previous discussions, if α = β = μ/(2 + ρ), then where f = Γμ,ρ(μ/(2 + ρ), μ/(2 + ρ)) is a point ellipse and in view of (49), we may arrive at the following result.

Theorem 15. Suppose |ρ| < 2 and (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)) such that f = Γμ,ρ(α, β). Let be generated by the Tracking Procedure with input (α, β) = (x(θ0), y(θ0)). For any mN,

(50)

Proof. In view of (49), we first consider the case where ϑ = π/4. By (49), it follows that

(51)
as well as
(52)
Note that xc = yc = μ/(2 + ρ). By (7) and (52), we also have
(53)
In view of (52), with respect to xk+2, it follows that
(54)
which leads to
(55)
The case where ϑ = −π/4 can be handled similarly. Hence, (50) is true and this completes the proof.

By Theorem 15, we can find the necessary and sufficient condition for a solution φ of (7) with |ρ| < 2 to be a periodic sequence.

Corollary 16. Suppose |ρ| < 2 and (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)) such that f = Γμ,ρ(α, β). Let be generated by the Tracking Procedure with input (x0, y0) = (α, β). Then is periodic if and only if

(56)
Furthermore, suppose π−1arccos(−ρ/2) = (m/n) ∈ Q such that m/n is irreducible and 0 < m < n. If m is even, then is n-periodic; otherwise, is 2n-periodic.

Proof. Note that if ρ ≠ −2 and (α, β) = (μ/(2 + ρ), μ/(2 + ρ)), then by the Tracking Procedure and Proposition 1, (x1, y1) = (α, β) which implies (xk, yk) = (α, β) for all kN. Suppose |ρ| < 2 and (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)). By (49), we have

(57)
By Theorem 15, for any kN, θΔ = θk+1θk is constant such that cos⁡(θΔ) = −ρ/2. Hence θk = θ0 + kθΔ. If is periodic with period nZ+, then we see that there is some mZ+ such that θn = θ0 + 2mπ. Accordingly, nθΔ = 2mπ which implies
(58)
On the other hand, if (56) holds, then, by the previous discussions and (58), is periodic with period 2n. Note that by Propositions 2 and 4, if is ω-periodic with (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)), then ω > 2. Next, suppose rQ, where r = π−1arccos⁡(−ρ/2). Since 2jπ where jZ are periods of cos⁡(x), we can first restrict ourselves to 0 < r < 2, where rπ. Also, if π < x < 2π, then cos⁡(x) = cos⁡(−x) = cos⁡(2πx) and this allows us to further concentrate on 0 < r < 1. Note that cos⁡(x) is bijective on (0, π). Hence, for each ρ such that rQ, where 0 < r < 1, there is a unique pair (m, n) such that r = (m/n), where m and n are coprime. In view of (58), it is easy to see that if m is even, then n is a period of ; if m is odd, then is periodic with period 2n. In the former case, we claim that is q-periodic. To this end, if q < q is a period of , then, by the previous discussions, there is pZ+ such that p as well q are coprime and r = p/q. Hence, (p/q) = (p/q) which is not possible. The latter case can be handled similarly and this completes the proof.

In view of Corollary 16, we have found the necessary and sufficient condition for , which is generated by the Tracking Procedure with input (x0, y0) ≠ (μ/(2 + ρ), μ/(2 + ρ)) such that f = Γμ,ρ(x0, y0) and |ρ| < 2, to be periodic. Let ρ = −1, 0, or 1; for instance, by (58), we take (m/n) = 1/3, 1/2, or 2/3, correspondingly, so that is 6-, 4-, or 3-periodic, respectively. For another example, by (58) again, if (m/n) = 2/5 or 4/5, then is 5-periodic where or , respectively.

For the analysis of nonnegative or negative solutions of (7) with |ρ| < 2, we need one more result.

Theorem 17. Let rQ. Suppose ϕk ≡ 2πkr mod 2π where kN. Then {ϕkkN} is dense in [0,2π].

In view of Theorem 17, let ηkkr mod 1, where kN. Then we see that ηk is dense in [0,1] (see the details in [12] in pages 280 and 281). Suppose s ∈ [0,2π]. Since (s/2π) ∈ [0,1], there exists a subsequence of such that , when j, from which it follows that
(59)
Accordingly, ϕk is dense in [0,2π] as desired.

Recall that by the Principal Axes Theorem, and are the principal axes of Γμ,ρ,f, where ρ ≠ −2 and f = Γμ,ρ(α, β), (α, β) ∈ R2. For −2 < ρ < 0 and (α, β) ∈ R2 such that f = Γμ,ρ(α, β), one property of which is generated by the Tracking Procedure with input (x0, y0) = (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)) is revealed as follows.

Corollary 18. Suppose −2 < ρ < 0 and (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)) such that f = Γμ,ρ(α, β). Let be generated by the Tracking Procedure with input (x0, y0) = (α, β). The set {(xm, ym)∣mN} cannot lie entirely in one of the half planes divided by

Proof. Since −2 < ρ < 0 and (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)), Γμ,ρ,f is an ellipse. By Corollary 16 and Propositions 2 and 4, if π−1arccos⁡(−ρ/2) ∈ Q, then φ is ω-periodic, where ω ≥ 3; otherwise, φ is aperiodic. Suppose π−1arccos⁡(−ρ/2) ∈ Q. Then is ω-periodic, where ω ≥ 3 such that (xω, yω) = (α, β). First of all, by the Tracking Procedure, we see that (xk, yk) moves on Γμ,ρ,f in a clockwise manner. Secondly, Γμ,ρ,f is symmetric with respect to x = y. Hence, without loss of generality, we suppose Π(α, β) ≥ 0. Since (xk, yk) move on Γμ,ρ,f in a clockwise manner; by the Tracking Procedure, we can be sure that for some i ∈ {0,1, 2, …, ω − 1}, H(Π(xi, yi))H(Π(α, β)) < 0. Note that if H(Π(α, β)) = 0, then we may investigate the orbit or so that H(Π(β, μρβα)) ≠ 0 or H(Π(μραβ, α)) ≠ 0, respectively. Suppose π−1arccos⁡(−ρ/2) ∉ Q. Without loss of generality, we suppose Π(α, β) < 0 and (s, t) ∈ Γμ,ρ,f such that Π(s, t) > 0. Then by Theorem 17, there is a subsequence such that as required and this completes the proof.

The previous discussions about the asymptotic behaviors of the solution of (7) help us to understand the periodicity of φ. To further analyze the positivity of solutions of (7), we require one specific solution with μ ≠ 0,  ρ < 2, and We note that f = Γμ,ρ(μ/2,0) = −μ2/4. Also, some properties of ρμχ are useful for the ensuing analysis and they are as follows.

Proposition 19. For μ, ρR, the lines y = 0 and x = 0 intercept with at (μ/2,0) and (0, μ/2), respectively. Furthermore, if ρ < 2 and μ ≠ 0, then is a nondegenerate conic section and the lines y = 0 and x = 0 are tangents of at the points (μ/2,0) and (0, μ/2), respectively.

Proof. First of all, we show that meets the y = 0 line at (μ/2,0). To this end, note that

(60)
and, thus, x = μ/2 is the unique solution to and y = 0. Accordingly, the y = 0 line intercept with at (μ/2,0). By the previous discussions, if ρ = 2, then is a degenerate parabola made up of the x + y − (1/2)μ = 0 line; if ρ = −2 and μ = 0, then Γ0,−2,0  is a degenerate parabola made up of the x + y = 0 line. Note that the is a parabola with μ ≠ 0. Accordingly, if ρ < 2 and μ ≠ 0, then is not a degenerate conic section which implies is tangent to x = 0 and y = 0 at (0, μ/2) as well as (μ/2,0), respectively. The proof is complete.

We illustrate Proposition 19 by several plots of , namely, ellipses with 0 < ρ < 2 plotted in Figure 16, ellipses with −2 < ρ < 0 plotted in Figure 17, and hyperbola with ρ < −2 plotted in Figure 18.

Details are in the caption following the image
The Γ1, ρ,−1/4, where 0 < ρ < 2.
Details are in the caption following the image
The Γ1,ρ,−1/4, where −2 < ρ < 0.
Details are in the caption following the image
The plot of Γ1,ρ,−1/4, where ρ = −3.

In the following section, we establish the relation between the nonnegative (or negative) solution of (7) and a special solution in the geometric setting.

4. A Special Positive Orbit

Let be the solution of (7) with By (12) and (13), we see that U(ρμχ) = E−2(ρμχ) which implies for all mZN. Hence, it is found that and by (15), we have
(61)
Note that (61) allows us to obtain from plotting only. In the sequel, we focus on the the “positive” orbit of when the orbit is discussed. By the Tracking Procedure with , we can generate the positive orbit Next, we establish the connection between the nonnegative solutions and the specific solution ρμχ.
Let
(62)
Let of (7) with (φm, φm+1) = (α, β) for some mZ. Then by the definitions of Φμ,ρ and Ωμ,ρ, we can be sure that
  • (1)

    φ is a nonnegative solution if and only if (α, β) ∈ Φμ,ρ;

  • (2)

    φ is a negative solution if and only if (α, β) ∈ Ωμ,ρ.

Now, it is our objective to find Φμ,ρ and Ωμ,ρ and in several special cases where ρ < 2 and μ > 0, we will show that Φμ,ρ and Ωμ,ρ can be expressed as the intersections of certain half planes in R2.

Proposition 20. Let and be solutions of (7) with (φ0, φ1) = (α, β) and (c0, c1) = (0,0). Let and be solutions of (7) with μ = 0, with (a0, a1) = (0,1), and with (b0, b1) = (1,0). Then we have

(63)

Proof. For the sake of convenience, we suppose kN. First of all, if (a0, b0, c0) = (0,1, 0) and (a1, b1, c1) = (1,0, 0), then (φ0, φ1) = (α, β) which implies (63) is true when k ≤ 1. Next, we show that (63) holds for k ≥ 2. To this end, by (7), it follows that

(64)
(65)
(66)
(67)
as required. Note that in view of (66), as well as (67), we have
(68)
Next, we consider the case where kZN. Let ψ = UE−1φ such that (ψ0, ψ1) = (β, α). Thus, we may reformulate (63) for ψm, where mN as
(69)
Furthermore, in view of (68), UE−1a = b and UE−1c = c which implies (63) is true and this completes the proof.

With (ak, bk, ck) defined in Proposition 20, we define the 2-variable function
(70)
and the set
(71)
Then we construct a set which is an intersection of half planes:
(72)
We claim that
(73)
Let be a solution of (7). Note that by Corollaries 3 and 5 and the definition of (72), we see that Φμ,ρ is nonempty which implies If (φm, φm+1) ∈ Φμ,ρ, for some mZ, then, by the definition, φ is nonnegative which implies φn ≥ 0 for all nZ. Since each hμ,ρ,k is characterized by φk, in view of (72), we see that On the other hand, if (φi, φi+1) ∉ Φμ,ρ, then φ is not nonnegative which implies there is some jZ such that φj < 0. Accordingly, we have hμ,ρ,j(x, y) < 0 and by (72), it follows that which is not possible.
Theoretically, we can calculate hμ,ρ,k(x, y) for all kZ to determine (73) explicitly. Suppose is an ω-periodic solution of (7) with given μ and ρ. Then we only have to calculate hμ,ρ,k(x, y), where k ∈ {0,1, 2, …, ω − 1}, to determine Φμ,ρ (or Ωμ,ρ). Even though this is feasible for us, it is still difficult to illustrate Φμ,ρ explicitly on R2 when ω is sufficiently large. In the sequel, we find Φμ,ρ, where ρ < 2 and μ ≠ 0, by one specific Note that the cases where Φμ,ρ, with ρ ≥ 2 or μ = 0, can be handled separately. By means of investigating the specific and where ρ < 2 and μ ≠ 0, Φμ,ρ can be found in an explicit way. Recall that if ρ < 2 and μ ≠ 0, then the x-axis and the y-axis are tangent to at (0, μ/2) as well as (μ/2,0), respectively. Let be a solution of (7) with The tangent line of at (xk, yk) is given by lμ,ρ,k(x, y) = 0, where
(74)
In addition, we also let be the line so that and are symmetric with respect to the x = y line. Now, we establish the connection between and the set of intersection of half planes divided by

Theorem 21. Given ρ,   μR, where ρ < 2 and μ > 0, then we have

(75)

Proof. First of all, we consider the case where k ≥ 0 and we show that hμ,ρ,k(x, y) = rlμ,ρ,k(x, y), where r = μ(ρ − 2)/2 < 0. To this end, we show the following results,

(76)
are true for all kZ. Let be with Since (a0, b0, c0) = (0,1, 0)    and (a1, b1, c1) = (1,0, 0), by (70) and (74), it follows that (75) holds for m ≤ 1. Now, we suppose that for some jN such that (75) holds for j and j + 1. Then we have seen that for i ∈ {j, j + 1}
(77)
For cj+2, by (68), (76), and (7), it follows that
(78)
as required. Next we consider aj+2 and bj+2. By (68), (77), and (7) again, in the former case,
(79)
as desired; the latter case can be handled similarly. By induction, we see that for all mN, hμ,ρ,k(x, y) = rlμ,ρ,k(x, y) is true. Now, we consider the case where k < 0. By the reflection and translation invariances, we transform ψ = UE−1φ in Proposition 20 and ζ = UE−1(ρμχ) and by similarly applied arguments for k ≥ 0, we see that hμ,ρ,k(x, y) = rlμ,ρ,k(x, y), for all k < 0. Note that U(ρμχ) = E−2(ρμχ) which implies , where k < 0. Accordingly, if k ≥ 0, then hμ,ρ,k(x, y) = rlμ,ρ,k(x, y); if k < 0, then Finally, since r < 0, we see that (75) holds and this completes the proof.

Corollary 22. Let be a solution of (7) with (φm, φm+1) = (σ, τ), for some mZ. If , then there is some nZ such that φn = 0.

Proof. By the definitions of in (75), if , then (σ, τ) is on the tangent line at , for some jZ. By (75) again, without loss of generality, we may further suppose that , for some iN. For the sake of convenience, we let δ = μρτσ and such that Since , by (74), we have

(80)
Next, we show that To this end, we first have
(81)
We apply (x, y) = (τ, δ) = (τ, μρτσ) and u = μρts to (81) and by simplifying, it is found that li+1(τ, δ) = li(σ, τ) = 0 which implies as desired. Since and are the y = 0 and x = 0 lines, respectively, by the previous discussions, there is some nZ such that φn = 0 and this completes the proof.

In this section, we introduced the definitions of Φμ,ρ (or Ωμ,ρ) so that the necessary and sufficient conditions for a solution φ of (7) to be nonnegative (or negative) can be obtained, respectively. Also, in Theorem 21, we discussed the specific solution of (7) with and the intersection of the closed half planes generated by the tangent lines at , for all kZ. For the case where ρ < 2 and μ > 0, we will show that some Φμ,ρ can be expressed explicitly in terms of in the sequel. Note that by Proposition 7 and Theorem 8, some Φμ,ρ and Ωμ,ρ with ρ < 2 and μ < 0 can be derived from the results of the cases where ρ < 2 and μ > 0. Hence, with all the results derived in the previous discussions together, we are ready to discuss the necessary and sufficient conditions for φ to be nonnegative (or negative).

5. Nonnegative and Negative Solutions when |ρ| > 2

Let be a solution of (7) defined by (φ0, φ1) = (α, β). In the last section, we have completed the discussions about the periodicity of φ and the asymptotic behaviors of which is generated by the Tracking Procedure with input (α, β). Note that (φk, φk+1) = (xk, yk) = (xk, xk+1), where is generated by the Tracking Procedure with input (x0, y0) = (α, β). To facilitate the ensuing discussions, we define two sets Θμ,ρ and Υμ,ρ on R2 where
(82)
Note that Θμ,ρ is defined by four closed half planes and Υμ,ρ is defined by two open as well as two closed half planes. Also note that by the definitions in (24) and (25), and are both independent of f. Now, we are able to apply these results to derive necessary and sufficient conditions for φ to be nonnegative (or negative).

Theorem 23. Let be a solution of (7) with ρ > 2. Suppose (φm, φm+1) ≠ (μ/(2 + ρ), μ/(2 + ρ)), for some mZ. Then φ is neither a nonnegative solution nor a negative solution.

Proof. Without loss of generality, we let m = 0. Let (φ0, φ1) = (α, β) such that f = Γμ,ρ(α, β). Let and be generated by the Tracking Procedure with inputs (x0, y0) = (α, β) and (s0, t0) = (β, α), respectively. Then Γμ,ρ,f is a hyperbola (or a degenerate one made up of two lines and ) such that For the sake of convenience, we let Pc = (μ/(2 + ρ), μ/(2 + ρ)). If (α, β) = Pc, then, by Proposition 2, (xk, yk) = (sk, tk) = Pc, for all kN. Suppose (α, β) ≠ Pc. By the previous discussions, we first consider the case where If , then and by Corollary 11, it is found that xkxk+1 = φkφk+1 < 0 when k; if , then we take and we see that so that sksk+1 < 0 when k    as required. Next, suppose Γμ,ρ,f is a hyperbola. Then by Corollary 11 again, xkxk+1 = φkφk+1 < 0 when k. Accordingly, if (α, β) ≠ Pc, then φ is neither a nonnegative nor a negative solution and this completes the proof.

Next, we consider the case where ρ < −2 and μR. This case is a little more complicated than the one described in Theorem 23, and we give the result of μ = 0 first.

Theorem 24. Let be a solution of (7) with ρ < −2 and μ = 0. Suppose (φm, φm+1) = (α, β), for some mZ such that f = Γμ,ρ(α, β). Then φ is a nonnegative solution if and only if (α, β) ∈ Θ0,ρ; and φ is a negative solution if and only if (α, β) ∈ Υ0,ρ.

Proof. Note that in view of (82), Θ0,ρ is symmetric with respect to the x = y line and Θ0,ρ∖{(0,0)} and Υ0,ρ are symmetric with respect to the x + y = 0 line. Let and be generated by the Tracking Procedure with inputs (x0, y0) = (α, β) and (s0, t0) = (β, α), respectively. Here, Γ0,ρ,f, where f = Γ0,ρ(α, β), is a hyperbola (or a degenerate one made up of two lines and ). Note that Θ0,ρ is a cone with two rays and such that u, v ≥ 0, where (u, v) ∈ Θ0,ρ. First of all, we show that Φ0,ρ = Θ0,ρ. To this end, we investigate (xk, yk) and (sk, yk) directly, and by Theorem 10, it is found that (xk, yk), (sk, tk) ≥ 0 for all kN if and only if (x0, y0) ∈ Θ0,ρ. Accordingly, we have Φ0,ρ = Θ0,ρ and the argument about Υ0,ρ can be handled similarly. Note that both Θ0,ρ and Θ0,ρ are symmetric with respect to the x = y line. Hence, the cases where kZN can be handled by the invariances of translation and reflection. The proof is complete.

Next, we discuss the case where ρ < −2 and μ ≠ 0.

Theorem 25. Let be a solution of (7) with ρ < −2 and μ ≠ 0. Suppose (φm, φm+1) = (α, β) for some mZ such that f = Γμ,ρ(α, β).

  • (i)

    Suppose μ > 0. Then φ is a nonnegative solution if and only if which is the intersection of the closed half planes defined in (75). Furthermore, φ is a negative solution if and only if (α, β) ∈ Υμ,ρ.

  • (ii)

    Suppose μ < 0.  φ is a nonnegative solution if and only if (α, β) ∈ Θμ,ρ; and φ is a negative solution if and only if , where is the intersection of the closed half planes defined in (75).

Proof. Without loss of generality, we let m = 0 and Pc = (μ/(2 + ρ), μ/(2 + ρ)). Since f = Γμ,ρ(α, β), Γμ,ρ,f is a hyperbola (or a degenerate one, two lines and ). Let and be generated by the Tracking Procedure with inputs (x0, y0) = (α, β) and (s0, t0) = (β, α), respectively. Suppose μ > 0. Then μ/(2 + ρ) < 0. By Theorem 21, it follows that as desired. Next, by Corollary 11, if (α, β) ∈ Υμ,ρ, then xk, sk < 0, for all kN. Conversely, if xk, sk < 0 for all kN, then, by Theorem 10 and Corollary 11, (α, β) ∈ Υμ,ρ. Now, we consider the case where μ < 0. From the previous discussions, Proposition 7 and Theorem 8, we can be sure that Φμ,ρ = Θμ,ρ. Next, we claim that , where is given by (75). From the previous discussions and Corollary 22, we can be sure that φ is a positive solution if and only if Thus, from Proposition 7 and Theorem 8, it follows that and this completes the proof.

Corollary 26. Let be a solution of (7) with |ρ| > 2.  φ is a nonnegative and periodic solution if and only if μ ≥ 0 and (φm, φm+1) = (μ/(2 + ρ), μ/(2 + ρ)) for some mZ; φ is a negative and periodic solution if and only if μ < 0 and (φn, φn+1) = (μ/(2 + ρ), μ/(2 + ρ)) for some nZ.

From Proposition 2, Theorem 10, Corollary 11, and Theorems 23, 24, and 25, we can easily see that the statement of Corollary 26 holds.

This section shows that both Φμ,ρ and Ωμ,ρ are nonempty and if φ is not a constant solution of (7) with |μ| > 2, then φ is not periodic. Next, we will discuss Φμ,ρ and Ωμ,ρ where |ρ| = 2.

6. Nonnegative and Negative Solutions when |ρ| = 2

Let be a solution of (7) with ρ = |2| and (φ0, φ1) = (α, β). Let f = Γμ,ρ(α, β). Recall that is on the conic section Γμ,ρ,f which is a parabola or degenerate parabola that is made up of either one line or two distinct parallel lines. In this section, we now discuss the nonnegative and negative solutions of (7) by investigating ρ = 2 and ρ = −2.

Theorem 27. Let be a solution of (7) with ρ = 2. Suppose (φm, φm+1) = (α, β) for some mZ.

  • (i)

    Suppose μ ≥ 0. Then φ cannot be a negative solution; furthermore, φ is a nonnegative solution if and only if

    (83)

  • (ii)

    Suppose μ < 0. Then φ cannot be a nonnegative solution; furthermore, φ is a negative solution if and only if

    (84)

Proof. From the previous discussions, Γμ,2,f, where f = Γμ,2(α, β) ≥ −μ2/4, is a degenerate parabola made up of two lines, and defined by (24) and (25), respectively. Moreover, if f = −μ2/4, then ; otherwise, and are two distinct parallel lines. We first consider the case where μ ≥ 0. By Theorem 12 and the Tracking Procedure, if α, β ≥ 0 such that f = Γμ,2(α, β) = −μ2/4, then φ is periodic with period 2 and φ2 ≥ 0 which implies φ is a nonnegative solution. Conversely, if φ is a nonnegative solution, then α, β ≥ 0 and by Theorem 12, f = −μ2/4 which implies Accordingly, we see that

(85)
On the other hand, by Theorem 12 again, we can be sure that φ cannot be a nonpositive solution which implies Ωμ,ρ is empty. Next, we suppose μ < 0. From the previous discussions and Proposition 7, if φ is a nonnegative solution then −φ is a solution of (7) with −μ > 0 which is not possible. Also, and are symmetric with respect to line x + y = 0 and from Proposition 7 and Theorem 8, it follows that
(86)
and this completes the proof.

Theorem 28. Let be a solution of (7) with ρ = −2. Suppose (φm, φm+1) = (α, β) for some mZ.

  • (i)

    Suppose μ = 0. Then φ is a nonnegative solution if and only if α, β ≥ 0 and which is the x = y line; furthermore, φ is negative solution if and only if and α, β < 0.

  • (ii)

    Suppose μ > 0. Then φ cannot be a negative solution; furthermore, φ is a nonnegative solution if and only if which is the intersection of the closed half planes defined by (75).

  • (iii)

    Suppose μ < 0. Then φ cannot be a nonnegative solution; furthermore, φ is a negative solution if and only if

    (87)

  • where is the intersection of the closed half planes defined by (75).

Proof. Recall that is the line x = y which is one principal axis of Γμ,ρ,f, where f = Γμ,ρ(α, β). Suppose μ = 0. By Theorem 13, if , then φ is neither a nonnegative nor a nonpositive solution. If , then φ is 1-periodic and we see that (i) holds by direct verification. Suppose μ > 0. Then by Theorem 21, we can be sure that as required and is depicted in Figure 19. For the case of (iii), from Proposition 7 and Theorem 8, it follows that (87) is true and this completes the proof.

Details are in the caption following the image
The set Φ1, −2.

In view of Theorem 28. we have shown that which is the intersection of the half planes defined by the tangent lines at all the points of on the parabola Figure 19 illustrates Φ1,−2 where the red-dashed curve is the parabola Γ1,−2,−1/4. The discussions of the cases where |ρ| ≥ 2 are done and move forward to analyze the last cases where |ρ| < 2.

7. Nonnegative and Negative Solutions when |ρ| < 2

In the previous discussions, we have analyzed the cases where |ρ| ≥ 2 and derived the necessary and sufficient conditions for a solution φ to be nonnegative (or negative). Let be a solution of (7). From our analyses of the case where |ρ| ≥ 2, the periodicity of φ has no significant role in the analysis (albeit φ may be periodic with least period ω ≤ 2); yet the periodicity of φ is of great importance when discussing the case where |ρ| < 2. In Corollary 16, we illustrated the necessary and sufficient conditions for φ to be a periodic solution of (7) with |ρ| < 2 and we will discuss the case where μ = 0,  μ > 0, and μ < 0 in the sequel.

Theorem 29. Let be a solution of (7) with |ρ| < 2 and μ = 0. Then φ cannot be a negative solution; furthermore, φ is a nonnegative solution if and only if φ is the null solution.

Proof. Let (φ0, φ1) = (α, β) such that f = Γ0,ρ(α, β). Let be generated by the Tracking Procedure with input (x0, y0) = (α, β). Suppose (α, β) = (0,0). From the previous discussions, f = Γ0,ρ(0,0) = 0 which implies Γ0,ρ,0 is a point ellipse. On the other hand, by Proposition 2, (φ0, φ1) = (0,0) which leads to φk = 0, for all kN and, thus, φ is a nonnegative solution with least period 1. Now, we suppose (α, β) ≠ (0,0). Then Γ0,ρ,f is an ellipse and we show that φ is neither a nonnegative nor a negative solution. To this end, we consider 0 ≤ ρ < 2 and −2 < ρ < 0. Suppose 0 ≤ ρ < 2. By Proposition 6, φ cannot be a nonpositive solution which implies φ is not negative. If φ is a nonnegative solution, then by Proposition 7, −φ is a nonpositive solution which is a contradiction. The case where −2 < ρ < 0 can be handled similarly and this completes the proof.

Next, we consider the corresponding results when |ρ| < 2 and μ ≠ 0.

Theorem 30. Let be a solution of (7) with |ρ| < 2 and μ ≠ 0. Suppose π−1arccos⁡(−ρ/2) ∉ Q and let (φm, φm+1) = (α, β) for some mZ.

  • (i)

    Suppose μ > 0. Then φ cannot be a negative solution; furthermore, φ is a nonnegative solution if and only if

    (88)

  • which is the interior of the ellipse

  • (ii)

    Suppose μ < 0. Then φ cannot be a nonnegative solution; furthermore, φ is a negative solution if and only if , where

    (89)

Proof. Without loss of generality, we let m = 0. Also, for the sake of convenience, let Pc = (μ/(2 + ρ), μ/(2 + ρ)) and r = π−1arccos⁡(−ρ/2). Also, let be generated by the Tracking Procedure with input (x0, y0) = (α, β). Then we consider two cases where μ > 0 and μ < 0. Here, if (α, β) = Pc, then, by the previous discussions, is a point ellipse so that (xk, yk) = Pc, for all kN. Note that is tangent to lines x = 0 and y = 0 at (0, μ/2) and (μ/2,0), respectively. Suppose μ > 0. We first show that φ cannot be a nonpositive solution. If ρ ≥ 0, then, by Proposition 6, φ is not a nonpositive solution. If φ is nonpositive with ρ < 0, then, by the previous discussions, (α, β) ≠ Pc and by Corollary 18, there is some mN such that xm ≥ 0 or ym ≥ 0. Furthermore, if α = β = 0, then, by (7), φ2 = μ > 0. Accordingly, φ is not a nonpositive solution when μ > 0.

Now, we show

(90)
where ρ > 0. Note that s, t ≥ 0, where (s, t) ∈ {(x, y) ∈ R2Γμ,ρ(x, y) ≤ (1/4)μ2}. If (88) holds, then, by Theorem 8, we can be sure that Γμ,ρ,fΦμ,ρ, where f = Γμ,ρ(α, β). Accordingly, φm ≥ 0 for all mZ. On the other hand, suppose (88) does not hold. Since rQ, by Theorem 17, there is some iN such that (or some jN such that ) which implies φ is not nonnegative. Hence, (90) is true. Next, suppose μ < 0. First of all, by the previous discussions and Corollary 22, we see that φ is a positive solution if and only if (α, β) ∈ Λμ,ρ. Thus, from Proposition 7 and Theorem 8, it follows that as required.

The proof is complete.

Theorems 30 and 17 and Proposition 7 help us to analyze the necessary and sufficient conditions for an aperiodic solution φ of (7) to be nonnegative (or negative). For the case when φ is periodic, we have to resort to the help of the specific solution ρμχ of (7).

Theorem 31. Let be a solution of (7) with |ρ| < 2 and μ ≠ 0. Suppose π−1arccos⁡(−ρ/2) ∈ Q and (φm, φm+1) = (α, β), for some mZ. Then φ is periodic and we have the following.

  • (i)

    Suppose μ > 0. Then φ cannot be a negative solution; furthermore, φ is a nonnegative solution if and only if which is the intersection of the closed half planes defined by (75).

  • (ii)

    Suppose μ < 0. Then φ cannot be a nonnegative solution; furthermore, φ is a negative solution if and only if

    (91)

  • where is the intersection of the closed half planes defined by (75).

Proof. Without loss of generality, we let m = 0. Also, let be generated by the Tracking Procedure with input (x0, y0) = (α, β). Suppose μ > 0. From the previous discussions, if (α, β) = (μ/(2 + ρ), μ/(2 + ρ)), then φ is a nonnegative solution with least period 1. If (α, β) ≠ (μ/(2 + ρ), μ/(2 + ρ)), then, by Corollary 16, φ is periodic and Γμ,ρ,f is an ellipse, where f = Γμ,ρ(α, β). Also, by Theorem 21, we can be sure that as desired. Next, we show that φ is not a nonpositive solution. First of all, by Proposition 6, φ is not nonpositive when ρ ≥ 0. Secondly, suppose φ is a nonpositive solution of (7) with ρ < 0. Then by Corollary 18, there is some mN such that xm or ym ≥ 0. If xm = ym = 0, then, by (7), xm+1 = μ > 0 which is a contradiction. Accordingly, φ is not a nonpositive solution.

Suppose μ < 0. By Corollary 22, Proposition 7, and Theorem 8, it follows that (91) holds. Also, if φ is a nonnegative solution of (7) with μ < 0, then, by Proposition 7, −φ is a solution of (7) with −μ > 0 which is not possible and this completes the proof.

Corollary 32. Let be a solution of (7) with |ρ| < 2 and μ > 0. Suppose r = π−1arccos⁡(−ρ/2) = (n/m) ∈ Q such that 0 < r < 1 and r is irreducible. Then Φμ,ρ is a closed polygon with ω sides which is circumscribed around the ellipse and ω is either 2n or n depending on whether m is even or odd, respectively.

Proof. In view of Theorem 21, which is defined by the tangent lines of Furthermore, all the lines are tangent to at the points which are generated by a specific solution ρμχ of (7) with Since π−1arccos⁡(−ρ/2) ∈ Q, by Corollary 16, ρμχ is ω-periodic where ω is dependent on n and the parity of m. Note that ω > 2. Accordingly, we see that Φμ,ρ is defined by ω distinct lines. To see Φμ,ρ is a closed polygon with ω sides which is circumscribed around , it is sufficient to show that Φμ,ρ is closed. If ρ = 0, then ρμχ is 4-periodic such that

(92)
if 0 < ρ < 2, then, by (70), hμ,ρ,2(x, y) = μρxy such that
(93)
Accordingly, Φμ,ρ is closed, where 0 ≤ ρ < 2. Suppose −2 < ρ < 0. To see Φμ,ρ is closed, it is sufficient to show that there is some iN such that Since , in view of Corollary 18, there is some jN such that as required and this completes the proof.

Corollary 32 gives an explicit description of Φμ,ρ, where |ρ| < 0, and two examples for the cases (ρ, μ) = (1,1) and (ρ, μ) = (−1,1) are in Figures 20 and 21 respectively.

Details are in the caption following the image
The set Φ1,1.
Details are in the caption following the image
The set Φ1,−1.

8. Conclusions

In the previous sections, we have analyzed the nonnegative (or negative) solution of (7) by investigating the pairs (φm, φm+1) for all mZ which are on a conic section (or degenerate one). Specifically, the orbit is the collection of (φm, φm+1) for all mZ which is a subset of a conic section and by the Tracking Procedure, (φn+1, φn+2) is uniquely determined by giving (φn, φn+1). By means of this geometric method, many qualitative properties of  (7) can be obtained. In particular, we obtain the necessary and sufficient conditions for φ to be a nonnegative (or negative) solution.

Now, we summarize the necessary and sufficient conditions for φ to be a nonnegative (or negative) periodic solution of (7) as follows (recall that for a set SR2, S is the set on R2 such that S and S are symmetric with respect to the x + y = 0 line).
  • (1)

    Suppose |ρ| > 2.  φ is a nonnegative and periodic solution if and only if ρμ ≥ 0, and for some iZ, (φi, φi+1) = (μ/(2 + ρ), μ/(2 + ρ)). Furthermore, such φ must satisfy φj = μ/(2 + ρ) for all jZ (and hence is 1-periodic).

  • (2)

    Suppose |ρ| < 2.  φ is a nonnegative and periodic solution if and only if μ > 0

    (94)

  • (where m/n is irreducible and 0 < m < n) and for some iZ, which, by Corollary 32, is a closed ω-side polygon where ω is either n or 2n depending on whether m is even or odd, respectively. Furthermore, such φ must either satisfy (φj, φj+1) = (μ/(2 + ρ), μ/(2 + ρ)) for all jZ (and hence is 1-periodic) or φ is ω-periodic.

  • (3)

    Suppose |ρ| = 2. If ρ = 2, then φ is a nonnegative and periodic solution if and only if μ ≥ 0, and for some mZ, φm, φm+1 ≥ 0 and (φm, φm+1) is on the x + y = μ/4 line; if ρ = −2, then φ is a nonnegative and periodic solution if and only if μ = 0, and for some nZ, φn, φn+1 ≥ 0, and (φn, φn+1) lie on the x = y line. Furthermore, such φ is either 1-periodic if φm = φm+1 (and 2-periodic if φmφm+1) when ρ = 2 or 1-periodic when ρ = −2.

As for the negative periodic solutions, we have the following.
  • (1)

    Suppose |ρ| > 2.  φ is a negative and periodic solution if and only if ρμ < 0 and for some iZ, (φi, φi+1) = (μ/(2 + ρ), μ/(2 + ρ)). Furthermore, such φ must satisfy φj = μ/(2 + ρ) for all jZ (and hence is 1-periodic).

  • (2)

    Suppose |ρ| < 2.  φ is a negative and periodic solution if and only if μ < 0

    (95)

  • (where m/n is irreducible and 0 < m < n) and for some iZ, , where by Corollary 32, is a closed ω-side polygon and ω is either n or 2n depending on whether m is even or odd, respectively. Furthermore, such φ must either satisfy (φj, φj+1) = (μ/(2 + ρ), μ/(2 + ρ)) for all jZ (i.e., φ is 1-periodic) or φ is ω-periodic.

  • (3)

    Suppose |ρ| = 2. If ρ = 2, then φ is a negative and periodic solution if and only if μ < 0, and for some mZ, φm, φm+1 < 0 and (φm, φm+1) is on the x + y = μ/4 line; if ρ = −2, then φ is a negative and periodic solution if and only if μ = 0, and for some nZ, φn, φn+1 < 0 and (φn, φn+1) lie on the x = y line. Furthermore, such φ is either 1-periodic if φm = φm+1 (and 2-periodic if φmφm+1) when ρ = 2 or 1-periodic when ρ = −2.

With the help of the Tracking Procedure, the behaviors of the solution φ of (7) can be visualized (instead of investigating each term of φ by the method of general solutions). By means of our geometric method, we may save lots of calculation and most properties of φ which may be obtained by the general solutions are also available from our geometric discoveries.

In [13], we conducted complete analyses on the periodic solutions (7) with μ = 1 and ρ ∈ {−1,0, 1}. By means of breaking a solution ψ into two subsequences, the companion sequence and the error sequence, we successfully show that both the companion sequence and error sequence are periodic so that ψ is periodic. Note that for ρ = −1, 0, and 1, we have arccos(−ρ/2) = π/3, π/2, and 2π/3, respectively.

The nice thing here is that by means of our geometric method, not only can we track the orbit of a solution φ on R2 easily but also we may obtain the conditions for φ to be nonnegative (or negative) quickly (especially for the case where |ρ| < 2 and φ is periodic). By [13], we have observed the importance of nonnegative solutions of (7) with ρ = 0, 1, and −1, respectively. However, we also realize the difficulties of analyzing the solution ζ of (7) with ρ ∉ {−1,0, 1} since the breaking of ζ into companion and error sequences is not applicable for all the solutions anymore.

Finally, in this paper we have only touched upon one class of steady state solutions of our original neural network. Much work has to be done before the dynamic behaviors of its solutions can be fully understood. Our contributions here, however, are that elementary geometric methods can be used to explain some of the complex phenomena obtained from simulations (see, e.g., Figures 1 and 2) and these methods are also expected to be applied in other piecewise smooth dynamic systems.

Conflicts of Interest

The authors declare that there are no conflicts of interest whatsoever regarding the publication of this paper.

Acknowledgments

The third author is partially supported by the Ministry of Science and Technology, R.O.C. under Grant MOST 104-2221-E-007-061-MY3 which also complies with the declaration of conflicts of interest stated above.

      The full text of this article hosted at iucr.org is unavailable due to technical difficulties.