MONOTONE SOLUTIONS OF DYNAMIC SYSTEMS ON TIME SCALES L. ERBE, A. PETERSON, AND C. C. TISDELL Lynn Erbe and Allan Peterson Department of Mathematics University of Nebraska - Lincoln Lincoln, NE 68588-0323 USA lerbe@math.unl.edu apeterso@math.unl.edu Christopher C. Tisdell School of Mathematics The University of New South Wales Sydney, NSW 2052 AUSTRALIA cct@maths.unsw.edu.au Abstract. We are concerned with proving that solutions of certain dynamical systems on time scales satisfy some monotoneity conditions. These results then give important results for nth order linear scalar equations. We then give a related result for a third order nonlinear (Emden–Fowler type) dynamic equation. Key words: time scales, dynamic equations, monotone solutions. AMS Subject Classification: 39A10. This paper is dedicated to the memory of Bernd Aulbach. 1. Introduction First we give some introductory definitions and results concerning the time scale calculus that will be used in this paper. For more detailed information see the books [2], [3], and [8] and the papers [1], [7]. The set T is a time scale provided it is a nonempty closed subset of the real numbers R. The forward jump operator σ and the backward jump operator ρ are defined by σ(t) := inf{τ > t : τ ∈ T} ∈ T, and ρ(t) := sup{τ < t : τ ∈ T} ∈ T, 1 2 ERBE, PETERSON, AND TISDELL for all t ∈ T, where inf ∅ := sup T and sup ∅ := inf T, where ∅ denotes the empty set. We assume thoughout that T has the topology that it inherits from the standard topology on the real numbers R. If σ(t) > t, we say t is right-scattered, while if ρ(t) < t we say t is left-scattered. If σ(t) = t and t < sup T we say t is right-dense, while if ρ(t) = t and t > inf T we say t is left-dense. The function x : T → R is said to be right-dense continuous (rd-continuous) and we write x ∈ Crd provided x is continuous at each right-dense point in T and at each left-dense point in T left-hand limits exist (finite). The function x : T → R is said to be regressive provided the regressivity condition 1 + µ(t)x(t) 6= 0, t ∈ T holds. Let R denote the set of all functions x : T → R such that x is right-dense continuous and regressive and let R+ := {x ∈ R : 1 + µ(t)x(t) > 0, t ∈ T}. The set R is called the set of regressive functions and the set R+ is called the set of positively regressive functions. Thoughout this paper we make the blanket assumption that a < b are points in T. We define the time scale interval [a, b]T := {t ∈ T such that a ≤ t ≤ b} and other types of time scale intervals are defined similarly. Time scale calculus unifies continuous and discrete calculus and is much more general as T can be any nonempty closed subset of the reals R. For example, it includes quantum calculus [5], which is time scale calculus on the time scales q Z ∪ {0} := {0, 1, q ±1 , q ±2 , q ±3 , · · · }, q > 0, q 6= 1, and hZ := {0, ±h, ±2h, ±3h, · · · }. Definition 1. Assume x : T → R and fix t ∈ Tκ , then we define x∆ (t) to be the number (provided it exists) with the property that given any > 0, there is a neighborhood U of t such that |[x(σ(t)) − x(s)] − x∆ (t)[σ(t) − s]| ≤ |σ(t) − s|, for all s ∈ U. We call x∆ (t) the (delta) derivative of x at t. n n We write x ∈ Crd if the n-th delta derivative function, denoted by x∆ , is rdcontinuous. The following theorem concerning (delta) differentiation is due to Hilger [7]. See also [2, Theorem 1.16]. Theorem 1. Assume that g : T → Rn and let t ∈ Tκ . (i) If g is differentiable at t, then g is continuous at t. (ii) If g is continuous at t and t is right-scattered, then g is differentiable at t with g ∆ (t) = g(σ(t)) − g(t) . σ(t) − t (iii) If g is differentiable and t is right-dense, then g ∆ (t) = lim s→t g(t) − g(s) . t−s MONOTONE SOLUTIONS 3 (iv) If g is differentiable at t, then g(σ(t)) = g(t) + µ(t)g ∆ (t). (1) Note that g ∆ (t) = g 0 (t), if T = R, and g ∆ (t) = ∆g(t) := g(t + 1) − g(t) if T = Z, where ∆ is the forward difference operator. If T = q N0 := {1, q, q 2 , q 3 , · · · }, q > 1, then one gets the so-called q derivative (quantum derivative) [5] g ∆ (t) = Dq g(t) := g(qt) − g(t) . (q − 1)t See [5] for some important applications of this quantum derivative. This q derivative is called the Hahn derivative in orthogonal polynomial theory (where it is usually assumed that 0 < q < 1 with a related time scale). 2. Main Results Our first main result concerns the first-order linear vector dynamic equation x∆ = A(t)xσ , (2) where xσ denotes the composite function x ◦ σ. Let us recall some notation. We write A(t) ≤ 0 provided each element aij (t) of A(t) satisfies aij (t) ≤ 0. We say A is right-dense continuous on T provided each element of A is right-dense continuous on T. Finally we say A is a regressive matrix function on T provided I + µ(t)A(t) is invertible for t ∈ T. In the proof of the next theorem we will use the fact [2, Chapter 5] that if the n × n matrix function is regressive and right-dense continuous on T, t0 ∈ Tκ , and x0 ∈ Rn , then the IVP x∆ = A(t)xσ , x(t0 ) = x0 has a unique solution defined on all of T. Throughout the remainder of the paper we assume ω := sup T = ∞ or that ω ∈ T is left-dense and we will be concerned with the behavior of solutions on [a, ω)T . If ω < ∞, then we do not assume that the matrix function A is defined at ω. Theorem 2. Assume that the n × n matrix function A is regressive and right-dense continuous on [a, ω)T , with A(t) ≤ 0 on T. Then the linear dynamic system (2) has a nontrivial solution x satisfying x(t) ≥ 0, x∆ (t) ≤ 0, t ∈ [a, ω)T . Proof. Assume τ ∈ (a, ω)T , y0 ∈ Rn with y0 > 0, and let y(t, τ ) be the solution of the IVP y ∆ = A(t)y σ , y(τ ) = y0 . We claim that y(t, τ ) > 0 on [a, τ ]T . Assume not, then there is a t1 ∈ [a, τ )T such that either σ(t1 ) = t1 , y(t, τ ) > 0 on (t1 , τ ]T and at least one component of y(t, τ ) is zero at t1 or σ(t1 ) > t1 , y(t, τ ) > 0 on [σ(t1 ), τ ]T , and at least one component of y(t, τ ) is nonpositive at t1 . In either case y ∆ (t, τ ) = A(t)y σ (t, τ ) ≤ 0 4 ERBE, PETERSON, AND TISDELL for t ∈ [t1 , τ )T . It follows from this that y(t1 , τ ) ≥ y(τ, τ ) = y0 > 0, which is a contradiction. Hence y(t, τ ) > 0 on [a, τ ]T for each τ ∈ (a, ω)T . Let {τn }∞ n=1 ⊂ (a, ω)T with limn→∞ τn = ω and let xn (t) := y(t, τn ) , ky(a, τn )k t ∈ [a, ω)T , n ≥ 1. Then for each n ≥ 1, xn is a solution of (2) with kxn (a)k = 1. It follows that there is a subsequence {xnk (a)}∞ k=1 such that lim xnk (a) = x0 , k→∞ where kx0 k = 1. Let x be the solution of the limit IVP x∆ = A(t)xσ , x(a) = x0 . Then x(t) = lim xnk (t) ≥ 0, t ∈ [a, ω)T x∆ (t) = A(t)xσ (t) ≤ 0, t ∈ [a, ω)T . k→∞ and so it follows that We next give the corresponding result for an alternative form of a first order linear system, x∆ = B(t)x. (3) Corollary 3. Assume that B is a regressive and right-dense continuous matrix function on [a, ω)T . If (I + µ(t)B(t))−1 B(t) ≤ 0 (or B(t)(I + µ(t)B(t))−1 ≤ 0) on [a, ω)T . Then the linear dynamic system (3) has a nontrivial solution x satisfying x(t) ≥ 0, x∆ (t) ≤ 0, t ∈ [a, ω)T . Proof. Using xσ (t) = x(t) + µ(t)x∆ (t) (see part (iv) of Theorem 1) is easy to see that the vector dynamic equation (3) is equivalent to the vector dynamic equation x∆ = (I + µ(t)B(t))−1 B(t)xσ . Also it is easy to varify that (I + µ(t)B(t))−1 B(t) = B(t)(I + µ(t)B(t))−1 . This corollary then follows from Theorem 2. We now can use Theorem 2 to prove the analogous result (Theorem 4) for the nth order scalar linear dynamic equation. (4) n u∆ + pn−1 (t)u∆ n−1 σ + pn−2 (t)u∆ n−2 σ + · · · + p0 (t)uσ = 0. We say that equation (4) is regressive on [a, ω)T in case pn−1 ∈ R([a, ω)T ) and pi ∈ Crd ([a, ω)T ), 0 ≤ i ≤ n − 1. Under these conditions all initial value problems for (4) have unique solutions that exist on [a, ω)T (see [2, Section 5.5]). MONOTONE SOLUTIONS 5 Theorem 4. Assume (4) is regressive and that the coefficient functions pi in (4) satisfy (−1)n+i pi−1 (t) ≥ 0 on [a, ω)T , 1 ≤ i ≤ n. Then (4) has a solution satisfying (5) i (−1)i u∆ (t) ≥ 0, u(t) > 0, 1 ≤ i ≤ n, t ∈ [a, ω)T . Proof. Let u be a solution of (4) on [a, ω)T and set u(t) u∆ (t) , x(t) = D ··· ∆n−1 (t) u t ∈ [a, ω)T , where D is the diagonal matrix D := diag{1, −1, 1, · · · , (−1)n−1 }. (6) Then u∆ (t) u∆2 (t) x∆ (t) = D ··· , n u∆ (t) t ∈ [a, ω)T , Using the formula (1) we get that i i u∆ σ (t) = u∆ (t) + µ(t)u∆ (7) i+1 (t), 1≤i≤n−1 and since u is a solution of (4) we have (8) n u∆ (t) = −p0 (t)uσ (t) − p1 (t)u∆σ (t) − · · · − pn−1 (t)u∆ n−1 σ (t). The equations (7) and (8) can be written in the vector form (9) 0 1 0 0 .. . 0 .. . 1 .. . = ··· .. . .. . 0 ··· ··· 0 −p0 −p1 −p2 · · · ··· 0 . .. . .. 0 1 µ .. . . . ... ... 0 . 0 ··· ··· 1 µ 0 0 ··· 0 1 1 µ 0 0 .. . 0 1 −pn−1 u∆ 2 u∆ 3 u∆ ··· n u∆ . uσ u∆σ 2 u∆ σ ··· ∆n−1 σ u 6 ERBE, PETERSON, AND TISDELL It follows, after a simple calculation of the inverse of the matrix appearing on the right hand side of (9), that = · · · (−1)n−1 µn−1 . 1 −µ . . (−1)n−2 µn−2 .. .. .. .. . . . . ··· 0 1 −µ 0 ··· 0 1 1 −µ 0 .. . 0 0 u∆ 2 u∆ 3 u∆ ··· n u∆ µ2 0 1 0 0 .. . 0 .. . 1 .. . ··· ... .. 0 .. . . 0 1 −pn−1 0 ··· ··· 0 −p0 −p1 −p2 · · · Hence x∆ = D u∆ 2 u∆ 3 u∆ ··· n u∆ = BC · D uσ u∆σ 2 u∆ σ ··· ∆n−1 σ u = BCxσ , where B = D 1 = 0 .. . 0 0 · · · (−1)n−1 µn−1 . 0 1 −µ . . (−1)n−2 µn−2 .. . . .. .. .. . . . . . 0 ··· ··· 1 −µ 0 0 ··· 0 1 −µ µ2 · · · (−1)n−1 µn−1 .. . (−1)n−1 µn−2 −1 µ .. ... ... ... . ··· ··· 0 (−1)n−1 µ 0 ··· 0 (−1)n−1 1 −µ µ2 uσ u∆σ 2 u∆ σ ··· ∆n−1 σ u . MONOTONE SOLUTIONS 7 and 0 1 C = 0 .. . 0 .. . 0 ··· −p0 −p1 ··· 0 .. .. . 1 . .. .. . . 0 ··· 0 1 ··· −pn−1 0 0 −1 0 = 0 .. . 0 ... 1 ... 0 ··· ··· −p0 p1 −p2 D ··· .. . ... 0 .. . .. . 0 (−1)n−1 · · · (−1)n pn−1 . Since the sign of every element in the ith column of B(t) is (−1)i−1 and from the sign assumptions on the coefficient functions pi we see that the sign of every element in the ith row of C is (−1)i it follows that A(t) := B(t)C(t) ≤ 0 on [a, ω)T . Therefore, from Theorem 2 there is a nontrivial solution u of (4) on [a, ω)T satisfying u(t) u∆ (t) ≥ 0, t ∈ [a, ω)T , x(t) = D ··· ∆n−1 u (t) and u∆ (t) u∆2 (t) ∆ x (t) = D · · · ≤ 0, n u∆ (t) t ∈ [a, ω)T . It follows that u satisfies (5). A second important form of an nth order linear scalar equation (see [2, Section 5.5]) is (10) n u∆ + qn−1 (t)u∆ n−1 + · · · + q0 (t)u = 0. We say the dynamic equation (10) is regressive provided the coefficient functions qi (t), 0 ≤ i ≤ n − 1, are rd-continuous on T and the regressivity condition R(t) := 1 + n−1 X (−µ(t))n−j qj (t) 6= 0, t∈T j=0 holds. It follows ([2, Corollary 5.90]) that if the dynamic equation (10) is regressive on [a, ω)T , then every initial value problem has a unique solution and all solutions exist on [a, ω)T . 8 ERBE, PETERSON, AND TISDELL Corollary 5. Assume that the dynamic equation (10) is regressive and (11) R(t) i−1 X (−1)n−j−1 µi−j−1 (t)qj (t) ≥ 0, j=0 for t ∈ [a, ω)T , 1 ≤ i ≤ n. Then the dynamic equation (10) has a nontrivial solution u satisfying (5). Proof. This follows from the fact that (see [2, Theorem 5.99]) the dynamic equations (4) and (10) are equivalent if i pi (t) := 1 X (−µ(t))i−j qj (t), R(t) j=0 0 ≤ i ≤ n − 1. Hence we have from (11) that i−1 n+i (−1) 1 X pi−1 (t) = (−1)n−j−1 µi−j−1 (t)qj (t) ≥ 0 R(t) j=0 for t ∈ [a, ω)T , 1 ≤ i ≤ n. The result then follows from Theorem 4. In the next theorem we see that we can relax the sign condition on the coefficient function pn−1 in Theorem 4 and get a slightly different conclusion. In Theorem 6 we consider the generalized exponential function eq (t, t0 ) for q ∈ R. See [2, Section 2.2] for an elementary development of this generalized exponential function. Theorem 6. Assume pn−1 ∈ R+ and that the coefficient functions pi satisfy (−1)n+i pi−1 (t) ≥ 0 on [a, ω)T , 1 ≤ i ≤ n − 1. Then the dynamic equation (4) has a solution u satisfying (12) u(t) > 0, i (−1)i u∆ (t) ≥ 0, 1 ≤ i ≤ n − 1, (−1)n (pxn−1 )∆ (t) ≥ 0, for t ∈ [a, ω)T , where p(t) := epn−1 (t, a). Proof. Since pn−1 ∈ R+ , we have (by [2, Theorem 2.48]) that p(t) := epn−1 (t, a) > 0 for all t ∈ T. Letting u be a solution of (4) and multiplying both sides of (4) by the integrating factor p(t) we get that u is a solution of (13) (pu∆ n−1 )∆ + qn−2 (t)u∆ n−2 σ + · · · + q0 (t)uσ = 0, where qi (t) := p(t)pi (t), 0 ≤ i ≤ n − 2. Note that (14) (−1)n+i qi (t) = p(t) (−1)n+i pi (t) ≥ 0, t ∈ [a, ω)T , 1 ≤ i ≤ n − 2. Let u be a solution of (4), then u is a solution of (13). Setting u(t) u∆ (t) , t ∈ [a, ω)T , x(t) = D ··· ∆n−1 ∆ (pu ) (t) MONOTONE SOLUTIONS 9 where D is given by (6) we get using an argument very similar to that in the proof of Theorem 4 that u(t) solves a system of the form x∆ = A(t)xσ , where A(t) = B(t)C(t) 1 −µ 0 −1 . . .. .. B= 0 0 0 0 0 0 where in this case (surpressing arguments) n−1 n−2 (−1)n−1 µ p µ2 · · · (−1)n−3 µn−3 (−1)n−2 µ p n−3 n−2 µ · · · (−1)n−3 µn−4 (−1)n−2 µ p (−1)n−1 µ p .. .. .. .. . ··· . . . n−1 µ2 n−3 n−2 µ (−1) 0 ··· (−1) (−1) p p n−2 1 (−1)n−1 µp 0 ··· 0 (−1) p 0 ··· 0 0 (−1)n−1 p1 and −1 0 .. . 0 0 .. . 0 1 .. . ··· ··· 0 0 .. . 0 0 .. . C= ··· 0 ··· ··· ··· 0 (−1)n−1 n−2 −q0 q1 −q2 · · · (−1) qn−2 0 . Using the sign conditions (14) on the coefficient functions qi (t) the rest of the proof is similar to the end of the proof of Theorem 4. We can now slightly improve Corollary 5. Corollary 7. Assume that the dynamic equation (10) is regressive and q ∈ R+ , where n−1 1 X (−µ(t))n−1−j pj (t). q(t) := R(t) j=0 Further assume that (15) R(t) i−1 X (−1)n−j−1 µi−j−1 (t)qj (t) ≥ 0, j=0 for t ∈ [a, ω)T , 1 ≤ i ≤ n−1. Then the dynamic equation (10) has a nontrivial solution u satisfying u(t) > 0, i (−1)i u∆ (t) ≥ 0, t ∈ [a, ω)T , 1 ≤ i ≤ n − 1, and (−1)n (eq (t, a)u∆ n−1 (t))∆ ≥ 0, t ∈ [a, ω)T . 3. A Third Order Nonlinear Dynamic Equation In this section we will be concerned with the third order nonlinear dynamic equation (16) x∆∆∆ + p(t)x∆σ + r(t)xγσ = 0, where γ is the quotient of odd integers and p, q are rd-continuous functions on [a, ω)T . This may be considered as an analogue of the third order Emden–Fowler equation. 10 ERBE, PETERSON, AND TISDELL These results are related to some results of Erbe [4] dealing with monotonicity properties of a third order nonlinear differential equation. To help us prove our main result concerning the dynamic equation (16) we first prove two important lemmas. Lemma 8. If x is a solution of (16) and b ∈ [a, ω)T , then b Z γ (17) ∆ b Z σ [(x∆∆ (t))2 − p(t)(x∆σ (t))2 ]∆t − x∆ (t)x∆∆ (t)]ba . r(t)(x (t)x (t)) ∆t = a a Proof. Assume x is a solution of (16), then multiplying both sides of equation (16) by x∆σ (t) and integrating from a to b we get b Z x ∆σ (t)x ∆∆∆ b Z p(t)(x (t)∆t + ∆σ 2 (t)) ∆t + a a Z b r(t)(xγ (t)x∆ (t))σ ∆t = 0. a After an integration by parts on the first term one easily gets the desired result (17). In connection with the third order dynamic equation (16), we will be concerned with the second order dynamic equation y ∆∆ + p(t)y σ = 0. (18) Definition We say that (18) is right-disfocal on [a, ω)T provided if y is a solution of (18), with y(s) = 0, y ∆ (s) > 0, then y ∆ (t) > 0 on (s, ω)T , for all s ∈ T. 2 1 Lemma 9. Assume v(t) > 0 with v ∈ Crd and assume y ∈ Crd . Then Z b (19) a 2 b Z b v ∆∆ (t) σ y (t)v ∆ (t) 2 2 (y (t)) + σ (y (t)) ∆t = F (t)∆t + , v (t) v(t) a a ∆ 2 q ∆ √y(t)v σ(t) . where F (t) := y (t) vv(t) σ (t) − ∆ v(t)v (t) MONOTONE SOLUTIONS 11 Proof. Consider (here we surpress arguments) Z b v ∆∆ σ 2 ∆ 2 (y ) + σ (y ) ∆t v a 2 σ Z b (y) ∆∆ ∆ 2 ∆t v = (y ) + v a 2 ∆ # 2 b Z b" y y ∆ ∆ 2 ∆ (y ) − = v ∆t + v (integrating by parts) v v a a 2 b Z b v(y 2 )∆ − y 2 v ∆ y ∆ ∆ 2 ∆ (y ) − = v ∆t + v (quotient rule) σ vv v a a 2 b Z b 2vyy ∆ + µv(y ∆ )2 − y 2 v ∆ y ∆ ∆ 2 ∆ = (y ) − (product rule) v v ∆t + σ vv v a a 2 b Z b 2yy ∆ v ∆ y 2 (v ∆ )2 µ(y ∆ )2 v ∆ y ∆ ∆ 2 + − v = (y ) − ∆t + σ σ σ v vv v v a a 2 b Z b ∆ 2 ∆ ∆ 2 ∆ 2 2yy v y (v ) (y ) σ y ∆ = (v − µv ∆ ) − + v ∆t + σ σ σ v v vv v a a 2 b Z b ∆ ∆ 2 ∆ 2 v ∆ 2 2yy v y (v ) y ∆ = (y ) − + v ∆t + σ σ σ v v vv v a a 2 b 2 Z b r ∆ y ∆ yv v √ = − ∆t + v y∆ vσ v vv σ a a 2 b Z b y ∆ F 2 ∆t + = . v v a a With the aid of Lemmas 8 and 9 we can now easily prove the following theorem. Theorem 10. If (18) is right-disfocal on [a, ∞)T and r(t) ≤ 0 on [a, ∞)T and not identically zero on any nondegenerate time scale subinterval of [a, ∞)T , then (16) has a solution x satisfying x(t) > 0, x∆ (t) > 0, x∆∆ (t) > 0 on (σ(a), ∞)T . Proof. Let x be a solution of (16) satisfying x(a) = x∆ (a) = 0, x∆∆ (a) > 0. The claim is that x(t) > 0, x∆ (t) > 0, and x∆∆ (t) > 0 on (σ(a), ∞)T . Assume this is not the case. Then there is a first b ∈ (σ(a), ω)T such that x∆∆ (b) ≤ 0. From Lemma 8, using x∆ (a) = 0, we get 12 ERBE, PETERSON, AND TISDELL Z b γ ∆ Z σ b [(x∆∆ (t))2 − p(t)(x∆σ (t))2 ]∆t − x∆ (t)x∆∆ (t)]ba r(t)(x (t)x (t)) ∆t = a a Z b = [(x∆∆ (t))2 − p(t)(x∆σ (t))2 ]∆t − x∆ (b)x∆∆ (b). a Since (16) is right-disfocal it is easy to see that there is a solution v of (16) satisfying v ∆ (t) > 0, v(t) > 0, t ∈ [a, b]T . Then by Lemma 9, with y(t) := x∆ (t), we have that Z b Z b γ ∆ σ r(t)(x (t)x (t)) ∆t = [(x∆∆ (t))2 − p(t)(x∆σ (t))2 ]∆t − x∆ (b)x∆∆ (b) a a Z b v ∆∆ (t) σ (y (t))2 ]∆t − x∆ (b)x∆∆ (b) [(y ∆ (t))2 + σ = v (t) a 2 b Z b y ∆ 2 F (t)∆t + = − x∆ (b)x∆∆ (b) v v a a ∆ 2 b (x ) ∆ v ≥ − x∆ (b)x∆∆ (b) v a ∆ 2 (x (b)) ∆ = v (b) − x∆ (b)x∆∆ (b). v(b) Since the left hand side is strictly negative to get a contradiction it suffices to show that the right hand side D := (x∆ (b))2 ∆ v (b) − x∆ (b)x∆∆ (b) ≥ 0. v(b) We know that x∆∆ (b) ≤ 0. If x∆∆ (b) = 0, then D= (x∆ (b))2 ∆ v (b) ≥ 0. v(b) Next assume that x∆∆ (b) < 0. In this case ρ(b) < b and since x∆∆ (ρ(b)) > 0 implies x∆ (b) > 0 we get that D := (x∆ (b))2 ∆ v (b) − x∆ (b)x∆∆ (b) ≥ 0 v(b) and this is the desired contradiction. Acknowledgement: This research was supported by the Australian Research Council’s Discovery Project DP0450752. References [1] R. Agarwal and M. Bohner, Basic calculus on time scales and some of its applications, Results Math., 35 (1999), 3–22. [2] M. Bohner and A. Peterson, Dynamic Equations on Time Scales:An Introduction With Applications, Birkháuser, Boston, 2001. MONOTONE SOLUTIONS 13 [3] M. Bohner and A. Peterson, Editors, Advances in Dynamic Equations on Time Scales, Birkháuser, Boston, 2003. [4] L. Erbe, Oscillation, nonoscillation, and asymptotic behavior for third order nonlinear differential equations, Annali di Mat. Pura ed Applicata, 100 (1976), 373–391. [5] V. Kac and P. Chueng, Quantum Calculus, Springer, Universitext, New York, 2002. [6] P. Hartman, Ordinary Differential Equations, Wiley, New York, 1964. [7] S. Hilger, Analysis on measure chains-a unified approach to continuous and discrete calculus, Results Math., 18 (1990), 18–56. [8] B. Kaymakçalan, V. Laksmikantham, and S. Sivasundaram, Dynamical Systems on Measure Chains, Kluwer Academic Publishers, Boston, 1996.