1_1163215_remsing.dvi EUROPEAN JOURNAL OF PURE AND APPLIED MATHEMATICS Vol. 5, No. 1, 2012, 1-15 ISSN 1307-5543 – www.ejpam.com SPECIAL ISSUE FOR THE INTERNATIONAL CONFERENCE ON APPLIED ANALYSIS AND ALGEBRA 29 JUNE - 02 JULY 2011, ISTANBUL TURKEY Single-Input Control Systems on the Euclidean Group SE(2) Ross M. Adams, Rory Biggs, Claudiu C. Remsing∗ Department of Mathematics (Pure and Applied), Rhodes University, Grahamstown 6140, South Africa Abstract. We consider a general single-input left-invariant control affine system, evolving on the Eu- clidean group SE(2). Any such controllable control system is (detached feedback) equivalent to one of two typical cases. In each case, we consider an optimal control problem (with quadratic cost) which is then lifted, via the Pontryagin Maximum Principle, to a Hamiltonian system on the dual space se(2)∗. The reduced Hamilton equations are derived and the stability nature of all equilibrium states is then investigated. Finally, these equations are explicitly integrated by elliptic functions. 2000 Mathematics Subject Classifications: 49J15, 93D05, 22E60, 53D17 Key Words and Phrases: Left-invariant control system, (detached) feedback equivalence, Lie-Poisson structure, elliptic function, Lyapunov stability 1. Introduction In recent decades, attention has been drawn to invariant control systems evolving on ma- trix Lie groups of low dimension. Such systems arise, for instance, in the airplane landing problem, the motion planning for wheeled robots, and the control of underactuated under- water vehicles (see, e.g., [31, 21, 20, 22, 15] and the references therein). An arbitrary left-invariant control affine system on the Euclidean group SE(2) has the form ġ = g � A+ u1B1 + · · ·+ uℓBℓ � , where A, B1, . . . , Bℓ ∈ se(2), 1 ≤ ℓ ≤ 3. (The elements B1, . . . , Bℓ are assumed to be linearly independent.) Specific (left-invariant) optimal control problems on the Euclidean group SE(2), associated with above mentioned control systems, have been studied by several authors (see, e.g., [11, 10, 29, 24, 23, 28]). In this paper, we consider only single-input control systems (i.e., systems of the form ġ = g (A+ uB)). Such a system is controllable if and only if it has full rank. Moreover, ∗Corresponding author. Email addresses: dros�webmail. o.za (R. Adams), rorybiggs�gmail. om (R. Biggs), . .remsing�ru.a .za (C. Remsing) http://www.ejpam.com 1 c© 2012 EJPAM All rights reserved. R. Adams, R. Biggs, C. Remsing / Eur. J. Pure Appl. Math, 5 (2012), 1-15 2 any such controllable control system is (detached feedback) equivalent to exactly one of the following control systems: Σ1 or Σ2,α (α > 0) with trace Γ1 = E1+ 〈E3〉 or Γ2,α = αE3 + 〈E1〉, respectively. (Here E1 and E3 denote elements of the standard basis for se(2).) In each typical case, we consider an optimal control problem (with quadratic cost) of the form ġ = g (A+ uB) , g ∈ SE(2), u ∈ R g(0) = g0, g(T ) = gT J = 1 2 ∫ T 0 u2(t)d t →min . Each problem is lifted, via the Pontryagin Maximum Principle, to a Hamiltonian system on the dual of the Lie algebra se(2). Then the (minus) Lie-Poisson structure on se(2)∗ is used to derive the equations for extrema (cf. [11, 1, 13]; see also [25, 26] for similar computations on the rotation group SO(3)). The (Lyapunov) stability nature of all equilibrium states is then investigated (by the energy-Casimir method). Finally, these equations are explicitly integrated by elliptic functions. 2. Preliminaries 2.1. Invariant Control Systems Invariant control systems on Lie groups were first considered in 1972 by Brockett [8] and by Jurdjevic and Sussmann [12]. A left-invariant control system Σ is a (smooth) control system evolving on a (real, finite-dimensional) Lie group G, whose dynamics Ξ : G×U → TG is invariant under left translations. (The tangent bundle TG is identified with G× g, where g is the Lie algebra of G). For the sake of convenience, we shall assume that (the state space of the system) G is a matrix Lie group. For the purposes of this paper, we may also assume that (the input space) U = Rℓ. Such a control system is described as follows (cf. [11, 1, 27]) ġ = Ξ(g,u), g ∈ G, u ∈ Rℓ (1) where Ξ(g,u) = gΞ(1,u) ∈ TgG. Admissible controls are bounded and measurable maps u(·) : [0, T] → Rℓ. We further assume that the parametrisation map Ξ(1, ·) : Rℓ → g is an embedding. Hence, the trace Γ = imΞ(1, ·) is a submanifold of g. We have that Γ = ¦ Ξu = Ξ(1,u) : u ∈ Rℓ © (cf. [5, 6]). A trajectory for an admissible control u(·) : [0, T]→ Rℓ is an absolutely continuous curve g(·) : [0, T]→ G such that ġ(t) = g(t)Ξ(1,u(t)) for almost every t ∈ [0, T]. A left-invariant control system Σ is said to be controllable if for any g0, g1 ∈ G, there exists a trajectory g(·) : [0, T] → Rℓ such that g(0) = g0 and g(T ) = g1. Controllable systems on connected (matrix) Lie groups must have full rank; this means that the Lie algebra generated by the trace of the system, Lie(Γ), is g. The following result is well known (see, also, [30]). Theorem 1 ([7]). A left-invariant control system on the Euclidean group SE(n) is controllable if and only if it has full rank. R. Adams, R. Biggs, C. Remsing / Eur. J. Pure Appl. Math, 5 (2012), 1-15 3 We shall denote a (left-invariant control) system Σ by (G,Ξ) (see, e.g., [5, 6]). We say that a system Σ = (G,Ξ) is connected if its state space G is connected. Let Σ = (G,Ξ) and Σ′ = (G′,Ξ′) be two connected full-rank systems with traces Γ ⊆ g and Γ′ ⊆ g ′, respectively. We say that Σ and Σ′ are (locally) detached feedback equivalent if there exist open neighbour- hoods N and N ′ of (the unit elements) 1 and 1′, respectively, and a (local) diffeomorphism Φ = φ ×ϕ : N ×Rℓ → N ′ ×Rℓ such that φ(1) = 1′ and Tgφ · Ξ(g,u) = Ξ′(φ(g),ϕ(u)) for g ∈ N and u ∈ Rℓ. Two detached feedback equivalent systems have the same trajectories (up to a diffeomorphism in the state space), which are parametrised differently by admissible controls. We recall the following result. Theorem 2 ([6]). Σ = (G,Ξ) and Σ′ = (G′,Ξ′) are (locally) detached feedback equivalent if and only if there exists a Lie algebra isomorphism ψ : g→ g ′ such that ψ · Γ = Γ′. 2.2. Invariant Optimal Control Problems Consider a left-invariant control system (1) evolving on some matrix Lie group G ≤ GL(n,R) of dimension m. In addition, it is assumed that there is a prescribed (smooth) cost function L : Rℓ → R (which is also called a Lagrangian). Let g0 and g1 be arbitrary but fixed points of G. We shall be interested in finding a trajectory-control pair (g(·),u(·)) which satisfies g(0) = g0, g(T ) = g1 (2) and in addition minimizes the total cost functional J = ∫ T 0 L(u(t))d t among all trajectories of (1) which satisfy the same boundary conditions (2). The terminal time T > 0 can be either fixed or it can be free. The Pontryagin Maximum Principle is a necessary condition for optimality which is most naturally expressed in the language of the geometry of the cotangent bundle T ∗G of G (cf. [1, 11]). The cotangent bundle T ∗G can be trivialized (from the left) such that T ∗G = G×g∗, where g ∗ is the dual space of the Lie algebra g. The dual space g ∗ has a natural Poisson structure, called the “minus Lie-Poisson structure”, given by {F, G}− (p) = −p �� dF(p), dG(p) �� for p ∈ g∗ and F, G ∈ C∞(g∗). (Note that dF(p) is a linear function on g ∗ and so is an element of g.) The Poisson manifold (g∗, {·, ·}) is denoted by g ∗ −. Each left-invariant Hamiltonian on the cotangent bundle T ∗G is identified with its reduction on the dual space g ∗ −. To an optimal control problem (with fixed terminal time) ∫ T 0 L(u(t))d t →min (3) subject to (1) and (2), we associate, for each real number λ and each control parameter u ∈ Rℓ, a Hamiltonian function on T ∗G = G× g ∗ : Hλu (ξ) = λL(u) + ξ � gΞ(1,u) � = λL(u) + p (Ξ(1,u)) , ξ= (g, p) ∈ T ∗G. The Maximum Principle can be stated, in terms of the above Hamiltonians, as follows. R. Adams, R. Biggs, C. Remsing / Eur. J. Pure Appl. Math, 5 (2012), 1-15 4 Maximum Principle. Suppose the trajectory-control pair ( ḡ(·), ū(·)) defined over the interval [0, T] is a solution for the optimal control problem (1)-(2)-(3). Then, there exists a curve ξ(·) : [0, T] → T ∗G with ξ(t) ∈ T ∗ ḡ(t) G, t ∈ [0, T], and a real number λ ≤ 0, such that the following conditions hold for almost every t ∈ [0, T] : (λ,ξ(t)) 6≡ (0,0) (4) ξ̇(t) = ~Hλ ū(t) (ξ(t)) (5) Hλ ū(t) (ξ(t)) =max u Hλu (ξ(t)) = constant. (6) An optimal trajectory ḡ(·) : [0, T] → G is the projection of an integral curve ξ(·) of the (time-varying) Hamiltonian vector field ~Hλ ū(t) defined for all t ∈ [0, T]. A trajectory-control pair (ξ(·),u(·)) defined on [0, T] is said to be an extremal pair if ξ(·) satisfies the conditions (4), (5) and (6). The projection ξ(·) of an extremal pair is called an extremal. An extremal curve is called normal if λ = −1 and abnormal if λ = 0. In this paper, we shall be concerned only with normal extremals. Suppose the maximum condition (6) eliminates the parameter u from the family of Hamiltonians � Hu � , and as a result of this elimination, we obtain a smooth function H (without parameters) on T ∗G (in fact, on g ∗ −). Then the whole (left-invariant) optimal control problem reduces to the study of trajectories of a fixed Hamiltonian vector field ~H. The following result holds. Theorem 3 ([13]). For the left-invariant control problem ġ = g � A+ u1B1 + · · ·+ uℓBℓ � , g ∈ G, u ∈ Rℓ g(0) = g0, g(T ) = gT J = 1 2 ∫ T 0 � c1u2 1(t) + · · ·+ cℓu 2 ℓ(t) � d t →min (T is fixed ) every (normal) extremal control is given by ui(t) = 1 ci p(t)(Bi), i = 1, . . . ,ℓ where p(·) : [0, T]→ g ∗ is an integral curve of the Hamiltonian vector field ~H on g ∗ − correspond- ing to the reduced Hamiltonian H(p) = p(A) + 1 2 � 1 c1 p(B1) 2 + · · ·+ 1 cℓ p(Bℓ) 2 � . Remark 1. In coordinates on g ∗ −, the (components of the) integral curves satisfy ṗi = − m ∑ j,k=1 ck i j pk ∂ H ∂ p j , i = 1, . . . , m. Here, ck i j denote the structure constants of g with respect to a basis � Ek � 1≤k≤m for g (i.e., [Ei, E j] = ∑m k=1 ck i j Ek) and pi = p(Ei). R. Adams, R. Biggs, C. Remsing / Eur. J. Pure Appl. Math, 5 (2012), 1-15 5 2.3. Jacobi Elliptic Functions Given the modulus k ∈ [0,1], the basic Jacobi elliptic functions sn(·, k), cn(·, k) and dn(·, k) can be defined as sn(x , k) = sin am(x , k) cn(x , k) = cosam(x , k) dn(x , k) = p 1− k2 sin2 am(x , k) where am(·, k) = F(·, k)−1 is the amplitude and F(ϕ, k) = ∫ ϕ 0 d tp 1−k2 sin2 t · (For the degener- ate cases k = 0 and k = 1, we recover the circular functions and the hyperbolic functions, respectively.) Nine other elliptic functions are defined by taking reciprocals and quotients; in particular, we get ns(·, k) = 1 sn(·,k) · Simple elliptic integrals can be expressed in terms of appropriate inverse (elliptic) functions. The following formulas hold true for b < a ≤ x and b ≤ x ≤ a, respectively (see [2] or [14]): ∫ ∞ x d t p (t2 − a2)(t2 − b2) = 1 a ns−1 � 1 a x , b a � (7) ∫ a x d t p (a2− t2)(t2 − b2) = 1 a dn−1 � 1 a x , p a2−b2 a � · (8) 2.4. The Energy-Casimir Method The energy-Casimir method [9] gives sufficient conditions for Lyapunov stability of equi- librium states for certain types of Hamilton-Poisson dynamical systems (cf. [16, 19]). The method is restricted to certain types of systems, since its implementation relies on an abun- dant supply of Casimir functions. The standard energy-Casimir method states that if ze is an equilibrium point of a Hamilto- nian vector field ~H (associated with an energy function H) and if there exists a Casimir func- tion C such that ze is a critical point of H+C (on the whole state space) and d2(H +C)(ze) is (positive or negative) definite, then ze is Lyapunov stable. Ortega and Ratiu have obtained a generalisation of the standard energy-Casimir method (cf. [18, 17]). This extended version states that if C = λ1C1+· · ·+λkCk, where λ1, . . . ,λk ∈ R and C1, . . . , Ck are conserved quantities (i.e., they Poisson commute with the energy function H), then definiteness of d2(λ0H + C)(ze),λ0 ∈ R is only required on the intersection (sub- space) W = ker dH(ze)∩ ker dC1(ze)∩ · · · ∩ dCk(ze). 3. The Euclidean Group SE(2) The Euclidean group SE(2) = ¨� 1 0 v R � : v ∈ R2×1, R ∈ SO(2) « R. Adams, R. Biggs, C. Remsing / Eur. J. Pure Appl. Math, 5 (2012), 1-15 6 is a (real) three-dimensional connected matrix Lie group. The group is solvable and unimod- ular. The associated Lie algebra is given by se(2) =       0 0 0 x1 0 −x3 x2 x3 0    : x1, x2, x3 ∈ R    . Let E1 =    0 0 0 1 0 0 0 0 0    , E2 =    0 0 0 0 0 0 1 0 0    , E3 =    0 0 0 0 0 −1 0 1 0    be the standard basis of se(2) with the following table for the bracket operation [·, ·] E1 E2 E3 E1 0 0 −E2 E2 0 0 E1 E3 E2 −E1 0 With respect to this basis, the group Aut(se(2)) of Lie algebra automorphisms of se(2) is given by       x y v −ςy ςx w 0 0 ς    : x , y, v, w ∈ R, x2+ y2 6= 0,ς= ±1    . (9) As se(2) is not semisimple, the Killing form is degenerate. Moreover, it can be shown that there does not exist any non-degenerate invariant scalar product on se(2). Therefore, we use the non-degenerate bilinear form *    0 0 0 x1 0 −x3 x2 x3 0    ,    0 0 0 y1 0 −y3 y2 y3 0    + = x1 y1 + x2 y2 + x3 y3 (on se(2)) to identify se(2) with se(2)∗ (cf. [11]). That is, we identify P ∈ se(2) with 〈P, ·〉 ∈ se(2)∗. Then each extremal curve p(·) in se(2)∗ is identified with a curve P(·) in se(2) via the formula 〈P(t), X 〉 = p(t)(X ) for all X ∈ se(2). Thus P(t) =    0 0 0 P1(t) 0 −P3(t) P2(t) P3(t) 0    where Pi(t) = 〈P(t), Ei〉 = p(t)(Ei) = pi(t), i = 1,2,3. Now consider a Hamiltonian H on the (minus) Lie-Poison structure for se(2)∗. The equa- tions of motion take the following form ṗi = −p( � Ei, dH(p) � ), i = 1,2,3 R. Adams, R. Biggs, C. Remsing / Eur. J. Pure Appl. Math, 5 (2012), 1-15 7 or, explicitly,              ṗ1 = ∂ H ∂ p3 p2 ṗ2 = − ∂ H ∂ p3 p1 ṗ3 = ∂ H ∂ p2 p1 − ∂ H ∂ p1 p2· (10) We note that C : se(2)∗→ R, C(p) = p2 1 + p2 2 is a Casimir function. 4. Classification of Systems Consider a general single-input left-invariant control affine system Σ with trace Γ = A+ 〈B〉 ⊂ se(2). We shall assume that Σ has full rank (i.e., Lie{A, B} = se(2)). This means (by proposition 1) that Σ is precisely a controllable system. Note that the Lie algebra rank condition is equivalent to the conditions (i) A and B are linearly independent and (ii) {A, B} 6⊂ 〈E1, E2〉 (cf. [30]). The following result gives a classification of all such control systems under the detached feedback equivalence (see, also, [6, 4, 3]). Theorem 4. Any controllable single-input (left-invariant control affine) system Σ is (locally) detached feedback equivalent to exactly one of the following systems: Σ1 or Σ2,α (α > 0) with respective parametrisations Ξ1(1,u) = E1 + uE3, Ξ2,α(1,u) = αE3 + uE1. Proof. Throughout, we use the algebraic characterisation from proposition 2. Let the trace of the system Σ be given by Γ = ∑3 i=1 ai Ei + D ∑3 i=1 bi Ei E . First, consider the case b3 6= 0. Then Γ = (a1 − b1a3 b3 )E1 + (a2− b2a3 b3 )E2+ D b1 b3 E1 + b2 b3 E2 + E3 E = a′1E1+ a′2E2 + ¬ b′1E1 + b′2E2 + E3 ¶ for some corresponding constants a′i, b′i ∈ R, i = 1,2. Hence, ψ =    a′1 −a′2 b′1 a′2 a′1 b′2 0 0 1    is a Lie algebra automorphism mapping Γ1 to Γ. (Note that detψ= 0 if and only if a′1 = a′2 = 0, a contradiction.) Next, consider the case b3 = 0. Since a3 6= 0 (as Σ is of full rank), and either b1 6= 0 or b2 6= 0, we get that ψ=    b1 −sgn(a3)b2 a1 α b2 sgn(a3)b1 a2 α 0 0 sgn(a3)    R. Adams, R. Biggs, C. Remsing / Eur. J. Pure Appl. Math, 5 (2012), 1-15 8 is a Lie algebra automorphism. Let α = |a3|, then ψ·Γ2,α = Γ. A simple argument shows that Σ1 is not equivalent to any system Σ2,α and that Σ2,α is not equivalent to Σ2,β for any α 6= β , α,β > 0. Left-Invariant Control Problems Henceforth, we consider only the systems Σ1 and Σ2,α. In each of these typical cases, we investigate an optimal control problem (with quadratic cost): ġ = g � E1 + uE3 � g(0) = g0, g(T ) = gT J = 1 2 ∫ T 0 u2(t)d t →min        LiCP(1) and ġ = g � αE3 + uE1 � g(0) = g0, g(T ) = gT J = 1 2 ∫ T 0 u2(t)d t →min        LiCP(2) The following two results follow easily from proposition 3. Theorem 5 ([31]). For the LiCP(1), the extremal control is given by u= p3, where H(p) = p1 + 1 2 p2 3 and    ṗ1 = p2p3 ṗ2 = −p1p3 ṗ3 = −p2. (11) Theorem 6. For the LiCP(2), the extremal control is given by u = p1, where H(p) = 1 2 p2 1 +αp3 and    ṗ1 = αp2 ṗ2 = −αp1 ṗ3 = −p1p2. (12) 5. Stability The equilibrium states for (11) are e µ 1 = (µ, 0,0) and eν2 = (0,0,ν) where µ,ν ∈ R,ν 6= 0. Theorem 7. The equilibrium states have the following behaviour. R. Adams, R. Biggs, C. Remsing / Eur. J. Pure Appl. Math, 5 (2012), 1-15 9 (i) The equilibrium state e µ 1 is stable if µ < 0 and unstable if µ≥ 0. (ii) Each equilibrium state eν2 is stable. Proof. The linearization of the system is given by    0 p3 p2 −p3 0 −p1 0 −1 0    . (i) Assume µ > 0. The linearization of the system (at e µ 1 ) has eigenvalues λ1 = 0, λ2,3 = ±pµ. Thus e µ 1 is unstable. Now, assume µ = 0. Then the linearization of the system has eigenvalues λ1,2,3 = 0. Thus, as the geometric multiplicity is strictly less than the algebraic multiplicity, e µ 1 is unstable. Assume µ < 0. Let Hχ = H +χ(C) be an energy-Casimir function, i.e., Hχ(p1, p2, p3) = 1 2 p2 3 + p1 +χ � p2 1 + p2 2 � , where χ ∈ C∞(R). The derivative dHχ = � 1+ 2p1χ̇ � p2 1 + p2 2 � 2p2χ̇ � p2 1 + p2 2 � p3 � vanishes at e µ 1 if and only if χ̇ � µ2 � = − 1 2µ · Then, the Hessian (at e µ 1 ) d2Hχ(µ, 0,0) = diag � 4µ2χ̈ � µ2 � − 1 µ , − 1 µ , 1 � is positive definite if and only if χ̈ � µ2 � > 1 4µ3 · The function χ(x) = − 1 4µ3 x2 satisfies these requirements. Hence, by the standard energy-Casimir method, e µ 1 is stable. (ii) Let Hλ = λ0H +λ1C , where λ0 = 0, λ1 = 1. Then we get dHλ(0,0,ν) = � 2p1 2p2 0 � |(0,0,ν) = 0 and d2Hλ(0,0,ν) = diag (2, 2, 0). Also, ker dH(eν2)∩ ker dC(eν2) = span {(−ν , 0,1), (0,1,0)} and so d2Hλ(0,0,ν) � � W×W = diag � 2ν2, 2 � is positive definite. Hence, by the extended energy-Casimir method, eν2 is stable. The equilibrium states for (12) are e µ 3 = (0,0,µ), µ ∈ R. Again, using the extended energy-Casimir method (as in theorem 7), we obtain the following result. Theorem 8. Each equilibrium state e µ 3 is stable. R. Adams, R. Biggs, C. Remsing / Eur. J. Pure Appl. Math, 5 (2012), 1-15 10 6. Explicit Integration First, let us consider the invariant control problem LiCP(1). There are three typical cases for the reduced extremal equations (11), corresponding to H > p C , H = p C and −pC < H < p C . (Note that H = −pC and C = 0 correspond to constant solutions, whereas the situation H < p C is impossible.) In figure 1, we graph the level sets of H and C and their intersection. We also graph the stable equilibrium points (illustrated in blue) and unstable equilibrium points (illustrated in red), as presented in theorem 7. The reduced Hamilton -2 0 2 E1 * -2 0 2 E2 * -2 0 2 E3 * -2 0 2 E1 * -2 0 2 E2 * -2 0 2 E3 * (a) H > p C -2 0 2 E1 * -2 0 2 E2 * -2 0 2 E3 * -2 0 2 E1 * -2 0 2 E2 * -2 0 2 E3 * (b) H = p C -2 0 2 E1 * -2 0 2 E2 * -2 0 2 E3 * -2 0 2 E1 * -2 0 2 E2 * -2 0 2 E3 * (c) −pC < H < p C Figure 1: Typical cases of reduced extremals of LiCP(1). equations (11) can be integrated by Jacobi elliptic functions. In each of the typical cases, we obtain explicit expressions for the integral curves of ~H. We start by considering the case H > p C . Theorem 9. Suppose p(·) : (−ǫ,ǫ)→ se(2)∗ is an integral curve of ~H such that H(p(0)) = h0, C(p(0)) = c0 > 0 and h2 0 − c0 > 0. Then there exists t0 ∈ R and σ ∈ {−1,1} such that R. Adams, R. Biggs, C. Remsing / Eur. J. Pure Appl. Math, 5 (2012), 1-15 11 p(t) = p̄(t + t0) for t ∈ (−ǫ,ǫ), where              p̄1(t) = p c0 k− sn (Ωt, k) 1− k sn (Ωt, k) p̄2(t) = σ p c0 k′cn (Ωt, k) 1− k sn (Ωt, k) p̄3(t) = −σ p 2δ dn (Ωt, k) 1− k sn (Ωt, k) · Here δ = p h2 0 − c0, Ω = p h0 + δ, k = q h0−δ h0+δ and k′ = q 2δ h0+δ · Proof. We start by explaining how the expression for p̄(·) can be found. Assume p̄(·) is an integral curve of ~H satisfying H(p̄(0)) = h0, C(p̄(0)) = c0 > 0 and δ2 = h2 0 − c0 > 0. Then, as p̄(·) solves (11), we get that d d t p̄1(t) = ± p 2 � c0 − p̄1(t) 2 �� h0 − p̄1(t) � . (13) This (separable) differential equation is transformed into standard form (see [2]) and for- mula (7) is then applied. After further simplification, this yields p̄1(t) as specified. Then, as C(p̄(t)) = c0, we get that p̄2(t) 2 = c0 − p̄1(t) 2 = 2c0δ � 1− sn (Ωt, k)2 � � h0+ δ � (1− ksn (Ωt, k))2 = c0(k ′)2cn (Ωt, k)2 (1− ksn (Ωt, k))2 yielding p̄2(t) as specified, for some σ ∈ {−1,1}. Finally, as d d t p̄3(t) = −p̄2 and −σpc0 ∫ k′cn (Ωt, k) 1− k sn (Ωt, k) d t = −σpc0 k′ kΩ dn (Ωt, k) 1− k sn (Ωt, k) we get p̄3(t) as prescribed. This motivates p̄(·) as a prospective integral curve of ~H. Now notice, as δ2 = h2 0 − c0 > 0, that 0< k < 1 and so 1− k sn (Ωt, k) > 0. Hence p̄(t) is well defined and smooth for all t ∈ R (and all h0, c0 such that h2 0− c0 > 0, c0 > 0). We now verify that p̄(·) is a solution to (11). We get that d d t p̄1(t)− p̄2(t)p̄3(t) = � −pc0(k ′)2Ω+ p 2c0δ � cn (Ωt, k)dn (Ωt, k) (1− ksn (Ωt, k))2 · Substitution and simplification then yields d d t p̄1(t) = p̄2(t)p̄3(t). Likewise, we get d d t p̄2(t) = −p̄1(t)p̄3(t) and d d t p̄3(t) = −p̄2(t). Hence p̄(·) : R→ se(2)∗ is a periodic integral curve of ~H. Any integral curve p(·) developing on H−1(h0)∩ C−1(c0) must be of the form p(t) = p̄(t + t0) for some σ ∈ {−1,1} and t0 ∈ R (see figure 1a). We now prove this fact. R. Adams, R. Biggs, C. Remsing / Eur. J. Pure Appl. Math, 5 (2012), 1-15 12 Let σ = sgn(p3(0)). We may assume σ 6= 0. Next we note that (p̄1(t), p̄2(t)) parametrises the circle S = {(x , y) : x2+ y2 = c0}. But p1(0) 2+ p2(0) 2 = c0, i.e., (p1(0), p2(0)) ∈ S. Therefore, there exists t0 ∈ R such that p̄1(t0) = p1(0) and p̄2(t0) = p2(0). Then we have that p3(0) 2 = 2(h0 − p1(0)) = 2(h0 − p̄1(t0)) = p̄3(t0) 2. Hence, as sgn(p3(t0)) = σ = sgn(p3(0)), we get that p3(0) = p̄3(t0). Thus the integral curves t 7→ p(t) and t 7→ p̄(t + t0) solve the same Cauchy problem, and therefore are identical. (Throughout this proof we used Mathematica to facilitate calculations.) Remark 2. Note that, for any h0, c0 ∈ R, h2 0 − c0 > 0, c0 > 0 and σ ∈ {−1,1}, we have that p̄(·) : R → se(2)∗ is a periodic integral curve of ~H. Consequently, any integral curve p(·) of ~H (satisfying the conditions of theorem 9) has maximal domain R and is periodic (on R). We now proceed to the case −pC < H < p C . Theorem 10. Suppose p(·) : (−ǫ,ǫ)→ se(2)∗ is an integral curve of ~H such that H(p(0)) = h0, C(p(0)) = c0 > 0 and h2 0 − c0 < 0. Then there exists t0 ∈ R such that p(t) = p̄(t + t0) for t ∈ (−ǫ,ǫ), where                  p̄1(t) = − p c0 − 2δ+ �p c0 + 2δ � dn (Ωt, k) 1+ dn (Ωt, k) p̄2(t) = 4 4 p c0δ pp c0 + δ p k′+ dn (Ωt, k)sn (Ωt, k) p (1+ dn (Ωt, k))3 p̄3(t) = 4δ pp c0 + δ cn (Ωt, k) p k′+ dn (Ωt, k) p 1+ dn (Ωt, k) · Here δ = 1p 2 4 p c0 p h0 + p c0, Ω = p c0+δ 4pc0 , k = 2 pp c0δp c0+δ and k′ = p c0−δp c0+δ · Remark 3. The proof of this theorem is similar to that of theorem 9. The crucial difference is that before solving equation (13), one needs to deinterlace the roots of the two quadratics involved. (This leads to a somewhat more involved computation which relies on formula (8).) Again, we note that p̄(·) is always a periodic integral curve and that any integral curve p(·) has maximal domain R, on which it is periodic. We finish with the case H = p C . Theorem 11. Suppose p(·) : (−ǫ,ǫ)→ se(2)∗ is an integral curve of ~H such that H(p(0)) = h0, C(p(0)) = c0 > 0 and h0 = p c0. Then there exists t0 ∈ R and σ ∈ {−1,1} such that p(t) = p̄(t + t0) for t ∈ (−ǫ,ǫ), where      p̄1(t) = 1 2 h0(cosh(2 p h0 t)− 3)sech2( p h0t) p̄2(t) = σh0sinh(2 p h0 t)sech3( p h0 t) p̄3(t) = 2σ p h0sech( p h0 t). REFERENCES 13 Remark 4. This result can be obtained by limiting h0 → pc0 from the left (i.e., using theorem 10) and adding possible changes in sign. Note however that this cannot be done from the right (i.e., using theorem 9). Finally, for the invariant control problem LiCP(2), there is only one typical case. As before, we graph the level sets of H and C and their intersection in figure 2. A simple computation then gives the solutions in this case. -2 0 2 E1 * -2 0 2 E2 * -2 0 2 E3 * -2 0 2 E1 * -2 0 2 E2 * -2 0 2 E3 * Figure 2: Typical reduced extremal of LiCP(2). Theorem 12. The reduced Hamilton equations (12) have the solutions    p1(t) = p c0 sin � αt + t0 � p2(t) = p c0 cos � αt + t0 � p3(t) = h0 α − c0 2α sin2 � αt + t0 � where c0 = C(p(0)), h0 = H(p(0)) and t0 ∈ R. References [1] A Agrachev and Y Sachkov. Control Theory from the Geometric Viewpoint. Springer- Verlag, Berlin, 2004. [2] J Armitage and W Eberlein. Elliptic Functions. Cambridge University Press, Cambridge, 2006. [3] R Biggs and C Remsing. Control affine systems on solvable three-dimensional Lie groups, II (submitted). [4] R Biggs and C Remsing. A note on the affine subspaces of three-dimensional Lie algebras (submitted). [5] R Biggs and C Remsing. A category of control systems. To appear in An. Şt. Univ. Ovidius Constanţa, 20(1), 2012. REFERENCES 14 [6] R Biggs and C Remsing. On the equivalence of control systems on Lie groups. To appear in Balkan J. Geometry Appl., 17(1), 2012. [7] B Bonnard, V Jurdjevic, I Kupka, and G Sallet. Transitivity of families of invariant vector fields on the semidirect product of Lie groups. Trans. Amer. Math. Soc., 271(2):525–535, 1982. [8] R Brockett. System theory on group manifolds and coset spaces. SIAM J. Control, 10(2):265–284, 1972. [9] D Holm, J Marsden, T Ratiu, and A Weinstein. Nonlinear stability of fluid and plasma equilibrium. Phys. Rep., 123:1–116, 1985. [10] V Jurdjevic. Non-Euclidean elastica. Amer. J. Math., 117(1):93–124, 1995. [11] V Jurdjevic. Geometric Control Theory. Cambridge University Press, Cambridge, 1997. [12] V Jurdjevic and H Sussmann. Control systems on Lie groups. J. Diff. Equations, 12:313– 329, 1972. [13] P Krishnaprasad. Optimal control and Poisson reduction. Technical Research Report T.R.93-87, Inst. Systems Research, Univ. of Maryland, 1993. [14] D Lawden. Elliptic Functions and Applications. Springer-Verlag, New York, 1989. [15] N Leonard. Stability of a bottom-heavy underwater vehicle. Automatica, 33(3):331–346, 1997. [16] J Marsden and T Ratiu. Introduction to Mechanics and Symmetry. Springer-Verlag, New York, second edition, 1999. [17] J-P Ortega, V Planas-Bielsa, and T Ratiu. Asymptotic and Lyapunov stability of con- strained and Poisson equilibria. J. Diff. Equations, 214:92–127, 2005. [18] J-P Ortega and T Ratiu. Non-linear stability of singular relative periodic orbits in Hamil- tonian systems with symmetry. J. Geom. Phys., 32:160–188, 1999. [19] M Puta. Hamiltonian Mechanical Systems and Geometric Quantization. Kluwer, Dor- drecht, 1993. [20] M Puta. Optimal control problems on matrix Lie groups. Quad. Sem. Top. Alg. e Diff., Univ. di Roma “La Sapienza”, 1996. [21] M Puta. Stability and control in spacecraft dynamics. J. Lie Theory, 7:269–278, 1997. [22] M Puta, P Birtea, C Lăzureanu, C Pop, and R Tudoran. Control, integrability and stability in some concrete mechanical problems on matrix Lie groups. Quad. Sem. Top. Alg. e Diff., Univ. di Roma “La Sapienza”, 1998. REFERENCES 15 [23] M Puta, S Chirici, and A Voitecovici. An optimal control problem on the Lie group SE (2,R). Publ. Math. Debrecen, 60:15–22, 2002. [24] M Puta, G Schwab, and A Voitecovici. Some remarks on an optimal control problem on the Lie group SE (2,R). An. Şt. Univ. “A.I. Cuza” Iaşi, ser. Mat., 49(2):249–256, 2003. [25] C Remsing. Control and integrability on SO (3). In Lect. Notes Eng. Comp. Sci., pages 1705–1710, London, U.K., 2010. [26] C Remsing. Integrability and optimal control. In 19th Int. Symp. Math. Theory of Net- works & Syst., pages 1749–1754, Budapest, Hungary, 2010. [27] C Remsing. Optimal control and Hamilton-Poisson formalism. Int. J. Pure Appl. Math., 59(1):11–17, 2010. [28] C Remsing. Control and stability on the Euclidean group SE (2). In Lect. Notes Eng. Comp. Sci., pages 225–230, London, U.K., 2011. [29] Y Sachkov. Maxwell strata in the Euler elastic problem. J. Dynam. Control Syst., 14(2):169–234, 2008. [30] Y Sachkov. Control theory on Lie groups. J. Math. Sci., 156(3):381–439, 2009. [31] G Walsh, R Montgomery, and S Sastry. Optimal path planning on matrix Lie groups. In 33rd Conf. Decision & Control, pages 1258–1263, Lake Buena Vista, FL, U.S.A., 1994.