Jump to ContentJump to Main Navigation
Show Summary Details
More options …

Open Physics

formerly Central European Journal of Physics

Editor-in-Chief: Seidel, Sally

Managing Editor: Lesna-Szreter, Paulina

IMPACT FACTOR 2018: 1.005

CiteScore 2018: 1.01

SCImago Journal Rank (SJR) 2018: 0.237
Source Normalized Impact per Paper (SNIP) 2018: 0.541

ICV 2017: 162.45

Open Access
See all formats and pricing
More options …
Volume 14, Issue 1


Volume 13 (2015)

Stability and Analytic Solutions of an Optimal Control Problem on the Schrödinger Lie Group

Remus-Daniel Ene / Camelia Pop / Camelia Petrişor
Published Online: 2016-12-30 | DOI: https://doi.org/10.1515/phys-2016-0060


The nonlinear stability and the existence of the periodic solutions for an optimal control problem on the Schrödinger Lie group are discussed. The analytic solutions via optimal homotopy asymptotic method of the dynamics and numerical simulations are presented, too.

Keywords: optimal control problem; nonlinear stability; periodic solutions; optimal homotopy asymptotic method

PACS: 02.20.Sv; 02.60.-x; 37J25; 65P40; 74H10; 37N20

1 Introduction

The role of the Schrödinger group in physics was always important and the literature on the subject is immense. The Schrödinger group and the Schrödinger algebra were first introduced by Niederer [13] and Hagen [14] as a nonrelativistic limit of the vector-field realization of the conformal algebra. The Schrödinger group could play the space configuration role in some concrete mechanical problems, like other important Lie groups do: SO(3) for aircraft dynamics, SE(3) for underwater dynamics, SO(2)× R2 for the rolling-penny dynamics, R2 × SO(3) for ball-plate problem, and so on. All these dynamics are modeled as invariant optimal control problems on the mentioned Lie groups. The study of the invariant optimal control problem on the Schrödinger Lie group naturally completes the short list from above. More details can be found in [5] or [9].

This article is concerned with some important comments on the behavior of a dynamical system arising from an optimal control problem on the Schrödinger Lie group presented in [10]. More specifically, we are interested in improving the stability results, proving that some equilibrium states are nonlinearly stable and finding the periodic orbits around them, and finally, we propose an alternative method for finding analytic solution of the system.

The system, which can be put into the equivalent form: x1˙=x2x3x2˙=2x12+2x32x42x3˙=x1x2x4x5x4˙=x3x5x5˙=x1x4+kx4,(1)

k ∈ ℝ, has the Hamilton-Poisson realization (see [10]): (R5,Π,H),

where Π=02x1x20x42x102x3x4x5x22x30x500x4x50kx4x50k0

is the minus Lie-Poisson strcture and H(x1,x2,x3,x4,x5)=12(x12+x32+x42)

is the Hamiltonian.Also, there exists only one functionally independent Casimir of our Poisson configuration, given by C : ℝ5 → ℝ, C(x1,x2,x3,x4,x5)=2kx1x3+k2x22+x1x52+x2x4x5x3x42.

Following [10], the equilibrium states of the dynamics (1) are:

  • e1P=(P,0,P,0,0) which are unstable for any P ∈*;

  • e2P=(P,0,P,0,0) which are spectrally stable for Pk>0;

  • e3PQ=(0,P,0,0,Q) which are spectrally stable for any P,Q ∈*.

The paper is organized as follows: the second section is dedicated to the nonlinear stability results; also, the existence of the periodical orbits around the nonlinear stable states is proved. The existence of the Hamilton-Poisson realization offers the necessary geometric framework in order to obtain this results. Specific tools, such as the energy methods and the Moser Theorem, belong to this framework.

The Optimal Homotopy Asymptotic Method is presented in the third section. Using this method, the analytical approximate solution of the proposed dynamics is found in the fourth paragraph. A comparison with the Runge-Kutta 4th steps integrator obtained using MATHEMATICA 6.0 and numerical simulations are the subjects of the last part.

2 Nonlinear Stability Problems and Periodical Orbits

Let us present very briefly, the concept of nonlinear stable equilibrium point of a dynamical system. For more details, see [7].

An equilibrium state xe is said to be nonlinear stable if for each neighborhood U of xe in D there is a neighborhood V of xe in U such that trajectory x(t) initially in V never leaves U.

This definition supposes well-defined dynamics and a specified topology. In terms of a norm ‖ ‖. nonlinear stability means that for each ε > 0 there is δ > 0 such that if x(0)xe<δ

then x(t)xe<ε,()t>0.

It is clear that nonlinear stability implies spectral stability; the converse is not always true.

The equilibrium states e2P are nonlinearly stable if Pk>0,P,kR.

We will use Arnold’s technique. Consider the function Fλ(x1,x2,x3,x4,x5)=H(x1,x2,x3,x4,x5)+λC(x1,x2,x3,x4,x5)=12(x12+x32+x42)+2kλx1x3+k2λx22+λx1x52+λx2x4x5λx3x42.

The following conditions hold:

  1. Fλ(e2P)=0ifλ=12k;

  2. Taking now W=ker[dC(e2P)]=Span01000,00010,00001,

    then, for all v ∈ W, i.e. v = (0,a,0,b,c), a, b, c ∈ ℝ, we have v2F12k(e2P)vt=12a2+1+Pkb2+Pkc2

    which is positive definite under the restriction Pk>0, and so 2F12k(e2P)|W×W

    is also positive definite.

Hence, via Arnold’s technique, the equilibrium states e2PwithPk>0,P,kR are nonlinear stable, as required.

As a consequence, we can find the periodic orbits of the equilibrium points e2PforPk>0,P,kR. More exactly, we can prove:

IfPk>0,P,kR the reduced dynamics to the coadjoint orbit 2kx1x3+k2x22+x1x52+x2x4x5x3x42=2kP2(2)

has near the equilibrium point e2P at least two periodic solution whose period is close to π22|P|andπkP+P2.

We will use Moser-Weinstein theorem with zero eigenvalue, see [3] for details.

  1. The restriction of our dynamics (1) to the coadjoint orbit (2) gives rise to a classical Hamiltonian system.

  2. The matrix of the linear part of our reduced dynamics to (1) has purely imaginary roots at the equilibrium of interest: λ1=0,λ2,3=±2i2Pandλ4,5=±ikP+P2.

  3. Span[C(e2P)]=ker[A(e2P)]=Span10100.

  4. IfPk>0,P,kR, then the reduced Hamiltonian has a local minimum at the equilibrium state e2P.

The equilibrium states e3PQ=(0,P,0,0,Q) are nonlinearly stable for any P,Q ∈*.

We will use energy-Casimir method, see [2] for details. Let Fφ(x1,x2,x3,x4,x5)=H(x1,x2,x3,x4,x5)+φ[C(x1,x2,x3,x4,x5)]=12(x12+x32+x42)+φ(2kx1x3+k2x22+x1x52+x2x4x5x3x42)

be the energy-Casimir function, where φ : ℝ → ℝ is a smooth real valued function.

Now, the first variation of Fφ is given by δFφ(x1,x2,x3,x4,x5)=x1δx1+x3δx3+x4δx4+φ˙(2kx1x3+k2x22+x1x52+x2x4x5x3x42)[(2kx3+x52)δx1+(kx2+x4x5)δx2+(2kx1x42)δx3+(x2x52x3x4)δx4+(2x1x5+x2x4)δx5]

so we obtain δFφ(e3PQ)=φ˙k2P2(Q2δx1+kPδx2+PQδx4)

that is equals zero for any P,Q ∈ ℝ* if and only if φ˙k2P2=0.(3)

The second variation of Fφ at the equilibrium of interest is given by δ2Fφ(e3PQ)=(δx1)2+(δx3)2+(δx4)2+φ¨k2P2(Q2δx1+kPδx2+PQδx4)2.

If we choose now φ such that the relation (3) is valid and φ¨k2P2>0,

then the second variation of Fφ at the equilibrium of interest is positive definited and so our equilibrium states e3PQ are nonlinearly stable.

3 The Optimal Homotopy Asymptotic Method

Let us pass now to find the analytical approximate solutions for the nonlinear differential system given by the Eq. (1) with the boundary conditions xi(0)=Ai,i=1,5¯,(4)

using the Optimal Homotopy Asymptotic Method (OHAM) (see [11], [12] for details). In the beginning, we briefly present the description of this method.

For equations of the general form: L(F(η))+N(F(η))=0,(5)

subject to the boundary/initial conditions of the type: B(F(η),dF(η)dη)=0.(6)

we construct (see [11] and [12]) the homotopy given by: H[L(F(η,p)),H(η,Ci),N(F(η,p))],(7)

where p ∈ [0, 1] is the embedding parameter, L is a linear operator and H(η, Ci), (H ≠ 0) is an auxiliary convergence-control function, depending on the variable η and on the parameters C1, C2, …, Cs.

One can prove that the following properties hold:

  1. H[L(F(η,0)),H(η,Ci),N(F(η,0))]=L(F(η,0))=L(F0(η))(8)

  2. H[L(F(η,1)),H(η,Ci),N(F(η,1))]=H(η,Ci)N(F(η,1)).(9)

If the function F is giving by: F(η,p)=F0(η)+pF1(η,Ci),(10)

then, by substituting (10) in (7), the following relation is obtained: H[L(F(η,p)),H(η,Ci),N(F(η,p))]=0.(11)

Consider the operator 𝓗 of the specific expression: H[L(F(η,p)),H(η,Ci),N(F(η,p))]=L(F0(η))++p[L(F1(η,Ci))H(η,Ci)N(F0(η))],(12)

where the governing equations of F0(η) and F1(η, Ci) can be obtained by equating the coefficients of p0 and p1, respectively: L(F0(η))=0,B(F0(η),dF0(η)dη)=0(13)


The expression of F0(η) can be found solving the linear equation (13). Also, to compute F1(η, Ci) we solve the equation (14), by taking into consideration that the nonlinear operator N presents the general form: NF0(η)=i=1mhi(η)gi(η),(15)

where m is a positive integer and hi(η) and gi(η) are known functions depending both on F0(η) and N.

Although equation (14) is a nonhomogeneous linear equation, in most cases its solution can not be found.

In order to compute the function F1(η, Ci) we will use the third modified version of OHAM (see [12] for details), consisting in the following steps:

– First we consider one of the following expressions for F1(η, Ci): F1(η,Ci)=i=1mHi(η,hj(η),Cj)gi(η),j=1,s¯,(16)

or F1(η,Ci)=i=1mHi(y,gj(η),Cj)hi(η),j=1,s¯,BF1(η,Ci),dF1(η,Ci)dη=0.(17)

These expressions of Hi(η, hj(η), Cj) contain both linear combinations of the functions hj and the parameters Cj,j=1,s¯. The summation limit m is an arbitrary positive integer number.

– Next, by taking into account the equation (10), the first-order analytical approximate solution of the equations (5) - (6) is: F¯(η,Ci)=F(η,1)=F0(η)+F1(η,Ci).(18)

– Finally, the convergence-control parameters C1, C2, …,Cs, which determine the first-order approximate solution (18), can be optimally computed by means of various methods, such as: the least square method, the Galerkin method, the collocation method, the Kantorowich method or the weighted residual method and so on. The first option should be minimizing the square residual error: J(C1,C2,...,Cs)=(D)R2(η,C1,C2,...,Cs)dη(19)

where the residual R is given by R(η,C1,C2,...,Cs)=Nf¯(η,Ci).(20)

The unknown parameters C1, C2, … , Cs can be identified from the conditions: JC1=JC2=...=JCs=0.(21)

With these parameters known (called optimal convergence-control parameters), the first-order approximate solution given by Eq.(18) is well-determined.

It should be emphasized that our procedure contains the auxiliary functions Hi(η, fi, Cj), i = 1, …, m, j = 1, …, s which provides us with a simple way to adjust and control the convergence of the approximate solutions. It is very important to properly choose these functions Hi(η, fi, Cj) which appear in the construction in the first-order approximation.

4 Application of Optimal Homotopy Asymptotic Method for Solving the Nonlinear Differential System (1)

Let us now apply the method described above to solve the nonlinear differential system given by the equations (1). We choose the linear operators: Lx1(t)=x1(t)+K1x1(t)Lx2(t)=x2(t)+K1x2(t)Lx3(t)=x3(t)+K1x3(t)Lx4(t)=x4(t)+K1x4(t)Lx5(t)=x5(t)+K1x5(t)ax4(t)(22)

where K1 > 0 is unknown parameter.

The corresponding linear equations for the initial approximationsxi0, i=1,5¯, becomes Lxi0(t)=0,xi0(0)=Ai,i=1,5¯,(23)

whose solutions are x10(t)=A1eK1tx20(t)=A2eK1tx30(t)=A3eK1tx40(t)=A4eK1tx50(t)=(A5+aA4t)eK1t.(24)

The nonlinear operators Nxi(t),i=1,5¯, are obtained from the equations (1) as follows: Nx1(t)=K1x1(t)+x2(t)x3(t)Nx2(t)=K1x2(t)+2x12(t)2x32(t)+x42(t)Nx3(t)=K1x3(t)x1(t)x2(t)+x4(t)x5(t)Nx4(t)=K1x4(t)x3(t)x5(t)Nx5(t)=K1x5(t)x1(t)x4(t).(25)

Substituting Eqs. (24) into Eqs. (25), we obtain Nx10(t)=K1x10(t)+x20(t)x30(t)Nx20(t)=K1x20(t)+2x102(t)2x302(t)+x402(t)Nx30(t)=K1x30(t)x10(t)x20(t)+x40(t)x50(t)Nx40(t)=K1x40(t)x30(t)x50(t)Nx50(t)=K1x50(t)x10(t)x40(t).(26)

a) We have a large freedom to choice the linear operator L, such as: Lx1(t)=x1(t)+K1x1(t)+g1(t)

and therefore, the nonlinear operator N becomes Nx1(t)=K1x1(t)+x2(t)x3(t)g1(t)

where g1(t) is an arbitrary continuous function that convergence-control the solution.

b) The function g1(t) can be chosen as g1(t) = B1 cos ωt + C1 sin ωt or g1(t) = B1 cos ωt + C1 sin ωt + B2 cos ω2 t + C2 sin ω2 t or g1(t) = B1 cos ωt + C1 sin ωt + B2 cos 3ωt + C2 sin 3ωt and so on.

The auxiliary convergence-control functions Hi are chosen such that the product Hi · N[xi0(t) has the same form like N[xi0(t). Then, the first approximation is: xi1=B1cosωt+C1sinωt+B2cos3ωt+C2sin3ωt+B3cos5ω1t+C3sin5ω1t+B4cos7ω1t+C4sin7ω1t+B5cosω2t+C5sinω2t+B6cos3ω2t+C6sin3ω2t+B7cos5ω3t+C7sin5ω3t+B8cos7ω3t+C8sin7ω3teK1t.(27)

Using now the third-alternative of OHAM and the equations (18), the first-order approximate solution can be put in the form x¯i(t,Cj)=xi0(t)+xi1(t,Cj),i=1,5¯(28)

where xi0(t) and xi1(t, Cj) are given by (24) and (27), respectively.

For the nonlinear differential system given by Eq. (1) the corresponding residual functions given by Eq. (20) become: R1(t,Bj,Cj)=x¯˙1+x¯2x¯3R2(t,Bj,Cj)=x¯˙2+2x¯122x¯32+x¯42R3(t,Bj,Cj)=x¯˙3x¯1x¯2+x¯4x¯5R4(t,Bj,Cj)=x¯˙4x¯3x¯5R5(t,Bj,Cj)=x¯˙5x¯1x¯4kx¯4,(29)

k ∈ ℝ.

5 Numerical Examples and Discussions

In this section, the accuracy and validity of the OHAM technique is proved using a comparison of our approximate solutions with numerical results obtained via the fourth-order Runge-Kutta method in the following case: we consider the initial value problem given by Eq. (1) with initial conditions Ai=0.1,i=1,5¯,k=50.

The convergence-control parameters K1,ω,ω1,ω2,ω3,Bi,Ci,i=1,8¯ are optimally determined by means of the least-square method.

For all unknown functions x¯i,i=1,5¯,we findK1=2.278126552620239,ω=1.3762803403795802,ω1=0.47813539122122517,ω2=1.7711316031320425,ω3=0.7512679521244802.

The first-order approximate solutions given by the Eq. (28) are respectively:

for x¯1:


–For x¯2:








Finally, Tables 15 emphasizes the accuracy of the OHAM technique by comparing the approximate analytic solutions x¯i,i=1,5¯ presented above with the corresponding numerical integration values.

Table 1

Comparison between OHAM results given by Eq. (30) and numerical results for a = 50 (relative error:εx1=|x¯1OHAMx1numeric|)

Table 2

Comparison between OHAM results given by Eq. (31) and numerical results for a = 50 (relative error: εx2=|x¯2OHAMx2numeric|)

Table 3

Comparison between OHAM results given by Eq. (32) and numerical results for a = 50 (relative error: εx3=|x¯3OHAMx3numeric|)

Table 4

Comparison between OHAM results given by Eq. (33) and numerical results for a = 50 (relative error: εx4=|x¯4OHAMx4numeric|)

Table 5

Comparison between OHAM results given by Eq. (34) and numerical results for a = 50 (relative error:εx5=|x¯5OHAMx5numeric|)

The Figures 15 present a comparison between the analytical approximate solutions obtained by OHAM method and numerical results obtained by Runge-Kutta 4th steps integrator. We can see that the analytical approximate solutions offer us the same numerical results as Runge-Kutta 4th integrator.

Comparison between the approximate solutions
x¯1 $\bar{x}_1 $  
given by Eq. (30) and the corresponding numerical solutions: ————— numerical solution,· · · · · · · · · · · OHAM solution.
Figure 1

Comparison between the approximate solutions x¯1 given by Eq. (30) and the corresponding numerical solutions: ————— numerical solution,· · · · · · · · · · · OHAM solution.

Comparison between the approximate solutions x¯2 $\bar{x}_2 $  given by Eq. (31) and the corresponding numerical solutions: ————— numerical solution,· · · · · · · · · · · OHAM solution.
Figure 2

Comparison between the approximate solutions x¯2 given by Eq. (31) and the corresponding numerical solutions: ————— numerical solution,· · · · · · · · · · · OHAM solution.

Comparison between the approximate solutions x¯3 $\bar{x}_3 $  given by Eq. (32) and the corresponding numerical solutions: ————— numerical solution,· · · · · · · · · · · OHAM solution.
Figure 3

Comparison between the approximate solutions x¯3 given by Eq. (32) and the corresponding numerical solutions: ————— numerical solution,· · · · · · · · · · · OHAM solution.

Comparison between the approximate solutions x¯4 $\bar{x}_4 $  given by Eq. (33) and the corresponding numerical solutions: ————— numerical solution,· · · · · · · · · · · OHAM solution.
Figure 4

Comparison between the approximate solutions x¯4 given by Eq. (33) and the corresponding numerical solutions: ————— numerical solution,· · · · · · · · · · · OHAM solution.

Comparison between the approximate solutions x¯5 $\bar{x}_5 $  given by Eq. (34) and the corresponding numerical solutions: ————— numerical solution,· · · · · · · · · · · OHAM solution.
Figure 5

Comparison between the approximate solutions x¯5 given by Eq. (34) and the corresponding numerical solutions: ————— numerical solution,· · · · · · · · · · · OHAM solution.

For the analytic approximate solutions given by Eqs. (30) - (34), the corresponding numerical value of the square residual are presented in Table 6.

Table 6

Numerical value of the square residual 0tmaxRi2(t,Bj,Cj)dt,i=1,5¯ given by Eq. (29) for a = 50

6 Conclusion

The main goal of this paper is to present some comments about the nonlinear stability and the periodic orbits for some equilibrium points of a dynamical system. Its solutions are the optimal controls that minimize a cost function which steer a left-invariant drift-free control system on the Schrödinger Lie group from an initial state to a final state (see [10]).

A lot of similar left-invariant drift-free control system on matrix Lie groups were studied so far in [6], [4], [1], [8].

In the last section the approximate analytic solutions of the considered controlled system (1) are established using the optimal homotopy asymptotic method (OHAM). %Numerical simulations via Mathematica 6.0 software and the approximations deviations are presented. Both numerical simulations via Mathematica 6.0 software and approximations deviations were presented. The accuracy of our results is pointed out by means of the approximate residual of the solutions. We can see that the proposed method, even if it does not use a Poisson integrator, offers better results that the Lie-Trotter integrator used in [10].


We would like to thank the referees very much for their valuable comments and suggestions.


  • [1]

    Aron A., Pop C., Puta M., An Optimal Control Problem on the Lie Group SE(2, ℝ) × SO(2), Boletin de la Sociedad Mexicana, 2009, 15(2). Google Scholar

  • [2]

    Birtea P., Puta M., Equivalence of energy methods in stability theory, Journal of Mathematical Physics, 2007, 48(4), 81-99. Google Scholar

  • [3]

    Birtea P., Puta M., Tudoran R. M., Periodic orbits in the case of a zero eigenvalue, Comptes Rendus Mathématique. Académie des Sciences, Paris, 2007, 344(12), 779-784. Google Scholar

  • [4]

    David D., Holm D., Multiple Lie-Poisson structures, reduction and geometric phases for the Maxwell-Bloch traveling wave equations, J. Nonlinear Sci, 1992, 2, 241-262. Google Scholar

  • [5]

    Jurdjevic V., Geometric Control Theory, Cambridge University Press, Cambridge, 1997. Google Scholar

  • [6]

    Leonard N. E., Averaging and motion control systems on Lie groups, PhD Thesis, University of Maryland, College Park, MD, 1994. Google Scholar

  • [7]

    Ortega J. P., Planas-Bielsa V., Ratiu T.S., Asymptotic and Lyapunov stability of constrained and Poisson equilibria, J. Differ. Eq., 2005, 214(1), 92–127.  CrossrefGoogle Scholar

  • [8]

    Pop C., A Drift-Free Left Invariant Control System on the Lie Group SO(3) × R3 × R3, Mathematical Problems in Engineering, 2015, vol. 2015, Article ID 652819, 9 pages. Google Scholar

  • [9]

    Puta M., Hamiltonian Mechanical Systems and Geometric Quantization, Kluwer Academic, Dordrecht, 1993, . CrossrefGoogle Scholar

  • [10]

    Sahoo S. R., Sahoo S. P., Jena A., Pati K. C., Optimal control on Schrödinger Lie group and the behavior of the dynamics, IMA Journal of Mathematical Control and Information, 2016, 1–17, . CrossrefGoogle Scholar

  • [11]

    Marinca V., Herişanu N., Bota C., Marinca B., An optimal homotopy asymptotic method applied to the steady flow of a fourth grade fluid past a porous plate, Applied Mathematics Letters, 2009, 22, 245-251. Google Scholar

  • [12]

    Marinca V., Herişanu N., The Optimal Homotopy Asymptotic Method: Engineering Applications, Springer Verlag, Heidelberg, 2015. Google Scholar

  • [13]

    Niederer U., The maximal kinematical invariance group of the free Schrödinger equation, Helv. Phys. Acta, 1972, 45, 802-810. Google Scholar

  • [14]

    Hagen C. R., Scale and conformal transformations in galilean-covariant field theory, Phys. Rev. D5, 1972, 377-388. Google Scholar

About the article

Received: 2016-08-27

Accepted: 2016-11-17

Published Online: 2016-12-30

Published in Print: 2016-01-01

Conflict of Interests: The authors declare that there is no conflict of interests regarding the publication of this paper.

Citation Information: Open Physics, Volume 14, Issue 1, Pages 549–558, ISSN (Online) 2391-5471, DOI: https://doi.org/10.1515/phys-2016-0060.

Export Citation

© 2016 Remus-Daniel Ene et al.. This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License. BY-NC-ND 3.0

Comments (0)

Please log in or register to comment.
Log in