Variational Principle Numerical Simulation Introduction A Lie Group Variational Integrator for the Attitude dynamics of a Rigid body Taeyoung Lee Student Seminar in Flight Dynamics and Control Department of Aerospace Engineering University of Michigan February 4, 2005 Lie Group Variational Integrator 1/28 Variational Principle Numerical Simulation Introduction Motivation Dynamics and Control of a System The motion of a system is described by a Differential Equation. Geometric feature is crucial for a dynamics and control problem. The analytic solution of a differential equation is hard to find. Numerical Integration The numerical solution is used to study dynamics of the system, to design/verify a controller. There has been little attention given to computational approaches. Generally, numerical integration methods DO NOT preserve the characteristics of the system. Geometric Numerical Integration Lie Group Variational Integrator 2/28 Variational Principle Numerical Simulation Introduction Motivation Dynamics and Control of a System The motion of a system is described by a Differential Equation. Geometric feature is crucial for a dynamics and control problem. The analytic solution of a differential equation is hard to find. Numerical Integration The numerical solution is used to study dynamics of the system, to design/verify a controller. There has been little attention given to computational approaches. Generally, numerical integration methods DO NOT preserve the characteristics of the system. Geometric Numerical Integration Lie Group Variational Integrator 2/28 Variational Principle Numerical Simulation Introduction Motivation Dynamics and Control of a System The motion of a system is described by a Differential Equation. Geometric feature is crucial for a dynamics and control problem. The analytic solution of a differential equation is hard to find. Numerical Integration The numerical solution is used to study dynamics of the system, to design/verify a controller. There has been little attention given to computational approaches. Generally, numerical integration methods DO NOT preserve the characteristics of the system. Geometric Numerical Integration Lie Group Variational Integrator 2/28 Variational Principle Numerical Simulation Introduction Motivation Numerical Example I : Two Body Problem Exact flow of the Two Body Problem Normalized Equation of motion q2 . (q21 + q22 )3/2 ! r 1+e q10 = 1 − e, q20 = 0, q̇10 = 0, q̇20 = 1−e q̈1 = − q1 , (q21 + q22 )3/2 q̈2 = − The trajectory is a conic section. The total energy and the angular momentum are conserved. H= 1 2 1 q̇ + q̇22 − 2 2 1 (q1 + q22 )1/2 h = q1 q̇2 − q2 q̇1 Lie Group Variational Integrator 3/28 Variational Principle Numerical Simulation Introduction Motivation Numerical Example I : Two Body Problem Numerical integration result Variational step-size Runge-Kutta method (Matlab ode45.m) 1 −0.49 Energy −0.5 0.5 −0.51 −0.52 q2 −0.53 −0.54 0 0 1 2 3 4 5 t/T 6 7 8 9 10 1 2 3 4 5 t/T 6 7 8 9 10 Angular Momentum 0.808 −0.5 −1 −1.5 −1 −0.5 q1 0 0.5 (Exact flow 0.806 0.804 0.802 0.8 0.798 0 , Runge-Kutta ) The trajectory is not closed orbit. (e = 0.6) The total energy and the angular momentum are not conserved. Lie Group Variational Integrator 4/28 Variational Principle Numerical Simulation Introduction Motivation Numerical Example II : Pendulum Exact flow of a Pendulum .... .... .... .... .... .... .... .... .... .... .... .... .... .... .... .... .... .... .... .... .... . q x Equation of motion q̈ = − sin q The total energy is conserved. H= 1 2 q̇ − cos q 2 The flow is symplectic. (Area conservation in phase plane) Lie Group Variational Integrator 5/28 Variational Principle Numerical Simulation Introduction Motivation Numerical Example II : Pendulum Numerical integration result 2 −0.45 1 −0.46 0 −0.47 −1 −0.48 −2 0 5 10 15 20 25 30 time (sec) 35 40 45 50 Velocity 1 −0.49 Energy Position Variational step-size Runge-Kutta method (Matlab ode45.m) −0.5 −0.51 0.5 −0.52 0 −0.53 −0.5 −0.54 −1 0 5 10 15 20 25 30 time (sec) 35 40 45 (Exact flow 50 −0.55 0 5 10 15 20 , Runge-Kutta 25 time 30 35 40 45 50 ) The position and velocity errors increase as t. The total energy is not conserved. Lie Group Variational Integrator 6/28 Variational Principle Numerical Simulation Introduction Motivation Numerical Example II : Pendulum Numerical integration result The numerical flow is not symplectic. Exact Flow Lie Group Variational Integrator Numerical Flow 7/28 Variational Principle Numerical Simulation Introduction Motivation Numerical Example II : Pendulum Numerical integration result The numerical flow is not symplectic. 2 1 0 −1 t = 50.0 (sec) −2 −3 −2 −1 Exact Flow Lie Group Variational Integrator 0 1 2 3 Numerical Flow 7/28 Variational Principle Numerical Simulation Introduction Geometric Numerical Integration Energy Momentum method [LaBudde 1976],[Simo 1992] The parameters of numerical integration is chosen such that energy and momentum are conserved. The states are projected to the constant energy momentum surface. Variational Integrator [Moser 1991], [Marsden 2001] A systematic method of constructing structure-preserving integrator It approximates the continuous dynamics by discretizing Hamilton’s principle. It inherits the properties of the continuous dynamics. Lie group method [Iserles 2000] Many practical differential equations evolve on a Lie group. It preserves the geometry of the configuration space. Lie Group Variational Integrator 8/28 Variational Principle Numerical Simulation Introduction Geometric Numerical Integration Energy Momentum method [LaBudde 1976],[Simo 1992] The parameters of numerical integration is chosen such that energy and momentum are conserved. The states are projected to the constant energy momentum surface. Variational Integrator [Moser 1991], [Marsden 2001] A systematic method of constructing structure-preserving integrator It approximates the continuous dynamics by discretizing Hamilton’s principle. It inherits the properties of the continuous dynamics. Lie group method [Iserles 2000] Many practical differential equations evolve on a Lie group. It preserves the geometry of the configuration space. Lie Group Variational Integrator 8/28 Variational Principle Numerical Simulation Introduction Geometric Numerical Integration Energy Momentum method [LaBudde 1976],[Simo 1992] The parameters of numerical integration is chosen such that energy and momentum are conserved. The states are projected to the constant energy momentum surface. Variational Integrator [Moser 1991], [Marsden 2001] A systematic method of constructing structure-preserving integrator It approximates the continuous dynamics by discretizing Hamilton’s principle. It inherits the properties of the continuous dynamics. Lie group method [Iserles 2000] Many practical differential equations evolve on a Lie group. It preserves the geometry of the configuration space. Lie Group Variational Integrator 8/28 Variational Principle Numerical Simulation Introduction Problem Definition Attitude Dynamics of a Rigid body under a Potential depends on the Attitude Assumption V ......... ..... .......... ... ... ...O .r... ........ . . .... ... .... .... .. .... .. ...... ......... .......... . ..... ........ρ ......... .... ... ... ... ...... ... ... ... ... ... ... q ... e2 .. ... CM .. ?.... ... . . . . . e3 ........ . . ............................. Lie Group Variational Integrator e1 The O − e1 e2 e3 is an inertial frame. V : SO(3) 7→ R Geometric Feature The configuration space is SO(3). The Hamiltonian is conserved. Other first integral may exist according to the symmetries of V. Goal A numerical integrator that respects the geometric features. The integrator updates ω, R. 9/28 Variational Principle Numerical Simulation Introduction Problem Definition Attitude Dynamics of a Rigid body under a Potential depends on the Attitude Assumption V ......... ..... .......... ... ... ...O .r... ........ . . .... ... .... .... .. .... .. ...... ......... .......... . ..... ........ρ ......... .... ... ... ... ...... ... ... ... ... ... ... q ... e2 .. ... CM .. ?.... ... . . . . . e3 ........ . . ............................. Lie Group Variational Integrator e1 The O − e1 e2 e3 is an inertial frame. V : SO(3) 7→ R Geometric Feature The configuration space is SO(3). The Hamiltonian is conserved. Other first integral may exist according to the symmetries of V. Goal A numerical integrator that respects the geometric features. The integrator updates ω, R. 9/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Variational Principle The basic idea of the variational integrator Continuous time Discrete time Configuration space Configuration space (q, q̇) ∈ TQ (qk , qk+1 ) ∈ Q × Q Lagrangian Lagrangian L(q, q̇) Ld (qk , qk+1 ) Action Integral G= Rt f t0 L(q, q̇) dt Variation δG = d G d =0 Action Sum Gd = PN−1 k=0 Ld (qk , qk+1 ) Variation δGd = d G d d =0 Equation of motion Equation of motion q̇ = f (q, q̇) qk+2 = fd (qk , qk+1 ) Lie Group Variational Integrator 10/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Variational Principle The basic idea of the variational integrator Continuous time Discrete time Configuration space Configuration space (q, q̇) ∈ TQ (qk , qk+1 ) ∈ Q × Q Lagrangian Lagrangian L(q, q̇) Ld (qk , qk+1 ) Action Integral G= Rt f t0 L(q, q̇) dt Variation δG = d G d =0 Action Sum Gd = PN−1 k=0 Ld (qk , qk+1 ) Variation δGd = d G d d =0 Equation of motion Equation of motion q̇ = f (q, q̇) qk+2 = fd (qk , qk+1 ) Lie Group Variational Integrator 10/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Lagrangian Two equivalent forms of Lagrangian L(R, ω) = ................. ... .q ..... - e1 ... ....... ..... ... ....... ...... ... ....... ..... .... .......... ........ ... ρ̃ ...... .... .... .. c ... ..... ... e2 ... . . ?. ... . . e3 ........ . .................... 1 2 Z kω × ρ̃k2 dm − V(R). Body Using kω × ρ̃k2 = kS(ρ̃)ωk2 , L(R, ω) = 1 T ω 2 Z S(ρ̃)T S(ρ̃)dm ω − V(R), Body 1 , ω T Jω − V(R). 2 (1) Using kω × ρ̃k2 = kS(ω)ρ̃k2 , L(R, ω) = 1 2 Z tr S(ω)ρ̃ρ̃T S(ω)T dm − V(R), Body 1 = tr S(ω) 2 Z Body 1 , tr S(ω)Jd S(ω)T − V(R), 2 Lie Group Variational Integrator ρ̃ρ̃T dm S(ω)T − V(R), (2) 11/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Lagrangian Two equivalent forms of Lagrangian L(R, ω) = ................. ... .q ..... - e1 ... ....... ..... ... ....... ...... ... ....... ..... .... .......... ........ ... ρ̃ ...... .... .... .. c ... ..... ... e2 ... . . ?. ... . . e3 ........ . .................... 1 2 Z kω × ρ̃k2 dm − V(R). Body Using kω × ρ̃k2 = kS(ρ̃)ωk2 , L(R, ω) = 1 T ω 2 Z S(ρ̃)T S(ρ̃)dm ω − V(R), Body 1 , ω T Jω − V(R). 2 (1) Using kω × ρ̃k2 = kS(ω)ρ̃k2 , L(R, ω) = 1 2 Z tr S(ω)ρ̃ρ̃T S(ω)T dm − V(R), Body 1 = tr S(ω) 2 Z Body 1 , tr S(ω)Jd S(ω)T − V(R), 2 Lie Group Variational Integrator ρ̃ρ̃T dm S(ω)T − V(R), (2) 11/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Lagrangian Relationship between J and Jd Using the relationship, S(ρ̃)T S(ρ̃) = ρ̃T ρ̃I3×3 − ρ̃ρ̃T , J and Jd are related as Z ρ̃T ρ̃dm J= I3×3 − Jd Body = tr[Jd ] I3×3 − Jd . If we express ρ̃ in coordinates (x, y, z), J and Jd are expressed as "R J= y2R+ z2 dm − R xy dm − xz dm R− R xy dm 2 R+ x dm − yz dm z2 R "R # − R xz dm R − yz dm , x2 + y2 dm Jd = 2 R x dm xy dm R xz dm R R xy dm 2 R y dm yz dm R # R xz dm R yz dm . z2 dm Furthermore, the following equation is satisfied for ω ∈ R3 . S(Jω) = S(ω)Jd + Jd S(ω). Lie Group Variational Integrator (3) 12/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Variation Action Integral Z t f G= Z t f L(R, ω)dt, = t0 t0 i 1 h tr S(ω)Jd S(ω)T − V(R) dt. 2 (4) Variation of R, ω The variation should respect the geometry of the configuration space. The varied rotation matrix R ∈ SO(3) can be expressed as R = Reη , (5) where ∈ R, η ∈ so(3) Using the kinematic relationship Ṙ = RS(ω), the variation of ω is induced from R . S(ω ) = RT Ṙ = e−η S(ω)eη + η̇, = S(ω) + [η̇ + S(ω)η − ηS(ω)] + O(2 ). Lie Group Variational Integrator (6) 13/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Equation of Motion Variation of Action Integral d δG = G = d =0 Z t f t0 1 ∂V tr η S(J ω̇ + ω × Jω) + 2RT 2 ∂R dt. Continuous Equation of motion From Hamilton’s principle, δG = 0. Then, we obtain S(J ω̇ + ω × Jω) = ∂V T ∂V R − RT , ∂R ∂R or equivalently, J ω̇ + ω × Jω = M, 3 where M ∈ R is determined by S(M) = ∂V T R ∂R − (7) RT ∂V . ∂R More explicitly, it can be shown that M = r1 × vr1 + r2 × vr2 + r3 × vr3 , where ri , vri ∈ R1×3 are the ith row vectors of R and Lie Group Variational Integrator ∂V , ∂R (8) respectively. 14/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Variational Principle The basic idea of the variational integrator Continuous time Discrete time Configuration space Configuration space (q, q̇) (qk , qk+1 ) Lagrangian Lagrangian L(q, q̇) Ld (qk , qk+1 ) Action Integral G= Rt f t0 L(q, q̇) dt Variation δG = d G d =0 Action Sum Gd = PN−1 k=0 Ld (qk , qk+1 ) Variation δGd = d G d d =0 Equation of motion Equation of motion q̇ = f (q, q̇) qk+2 = fd (qk , qk+1 ) Lie Group Variational Integrator 15/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Discrete Lagrangian Discrete Lagrangian Define new variable Fk ∈ SO(3) such that Rk+1 = Rk Fk . Using the kinematic relationship Ṙ = RS(ω), S(ωk ) can be approximated as S(ωk ) = RTk Ṙk ≈ RTk Rk+1 − Rk h = 1 (Fk − I3×3 ) . h (9) Consider the discrete Lagrangian Ld , h h L(Rk , ωk ) + L(Rk+1 , ωk+1 ), 2 2 h h 1 = tr[(I3×3 − Fk ) Jd ] − V(Rk ) − V(Rk+1 ). h 2 2 Ld (Rk , Fk ) ' Action Sum Since the discrete Lagrangian Ld approximates N−1 X Gd = N−1 X Ld (Rk , Fk ) = k=0 k=0 Rt k+1 tk (10) L(R, ω) dt, 1 h h tr[(I3×3 − Fk ) Jd ] − V(Rk ) − V(Rk+1 ). h 2 2 (11) Lie Group Variational Integrator 16/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Variation Variation of Rk , Fk The varied rotation matrix Rk ∈ SO(3) can be expressed as Rk = Rk eηk , (12) where ∈ R, ηk ∈ so(3). Using the definition of Fk , the variation of Fk is induced as −ηk T Fk = RT Rk Rk+1 eηk+1 = e−ηk Fk eηk+1 . k Rk+1 = e (13) Variation of Action Sum N−1 X 1 d h ∂V ∂V = Gd tr[(ηk Fk − Fk ηk+1 ) Jd ] + tr ηk RTk + ηk+1 RTk+1 , d =0 h 2 ∂R ∂R k k+1 k=0 N−1 X = tr ηk k=1 Lie Group Variational Integrator ∂V 1 (Fk Jd − Jd Fk−1 ) + hRTk h ∂Rk . (14) 17/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Discrete Equation of Motion Discrete Equation of Motion in Rk , Fk From Hamilton’s principle, δGd = 0. Then, we obtain 1 T Fk+1 Jd − Jd Fk − Jd Fk+1 + FkT Jd = hS(Mk+1 ), h Rk+1 = Rk Fk . (15) (16) This yields a discrete map (R0 , F0 ) 7→ (R1 , F1 ) and the process can be repeated. Discrete Equation of Motion in Rk , ωk Inertial and body-fixed angular momentum are related as S(Jω) = S(ω)Jd + Jd S(ω), S(RJω) = RS(Jω)RT = ṘJd RT − RJd ṘT . If we approximate Ṙk = Rk+1 −Rk , h we obtain 1 Rk+1 Jd RTk − Rk Jd RTk+1 , h 1 T S(Jωk ) = Rk S(Rk Jωk )Rk = Fk Jd − Jd FkT . h S(Rk Jωk ) = Lie Group Variational Integrator (17) 18/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Discrete Equation of Motion Variational Integrator for the Attitude dynamics of a Rigid Body Discrete Equation of Motion in Rk , ωk Using the above equations, the discrete equations of motion can be expressed as Jωk+1 = FkT Jωk + hMk+1 , 1 Fk Jd − Jd FkT , S(Jωk ) = h Rk+1 = Rk Fk , where Mk ∈ R3 is determined by S(Mk ) = ∂V T Rk ∂Rk (18) (19) (20) ∂V − RTk ∂R . More explicitly, k Mk = r1 × vr1 + r2 × vr2 + r3 × vr3 , where ri , vri ∈ R 1×3 are the ith row vectors of Rk and ∂V , ∂Rk (21) respectively. Given R0 , and ω0 , we can obtain F0 implicitly by solving (19), R1 is updated by (20), and the angular velocity ω1 is updated by (18). This yields a map (R0 , ω0 ) 7→ (R1 , ω1 ) and the process is repeated. Lie Group Variational Integrator 19/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Discrete Equation of Motion Properties of the Variational Integrator Preservation of conserved quantities Noether’s theorem Symmetries in the Lagrangian result in the conservation of the associated momentum map. Variational integrators exhibit a discrete analogue of Noether’s theorem. All the conserved momenta in the continuous dynamics are preserved in the discrete dynamics. Preservation of the configuration space The state is automatically evolves on the Rotation group. Since SO(3) is closed under multiplication, the attitude matrix Rk remains on SO(3). (Rk+1 = Rk Fk ) The attitude is globally defined without any projection or constrains. Lie Group Variational Integrator 20/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Discrete Equation of Motion Properties of the Variational Integrator Preservation of conserved quantities Noether’s theorem Symmetries in the Lagrangian result in the conservation of the associated momentum map. Variational integrators exhibit a discrete analogue of Noether’s theorem. All the conserved momenta in the continuous dynamics are preserved in the discrete dynamics. Preservation of the configuration space The state is automatically evolves on the Rotation group. Since SO(3) is closed under multiplication, the attitude matrix Rk remains on SO(3). (Rk+1 = Rk Fk ) The attitude is globally defined without any projection or constrains. Lie Group Variational Integrator 20/28 Variational Principle Numerical Simulation Continuous time model Variational Integrator Discrete Equation of Motion Computational Approach Solution of hS(Jω) = FJd − Jd F T . Using the Rodrigues’ formula, F = eS(f ) = I3×3 + sin kf k 1 − cos kf k S(f )2 , S(f ) + kf k kf k2 where f is a vector in R3 . Then the matrix equation can be transformed into hJω = sin kf k 1 − cos kf k Jf + f × Jf , G(f ). kf k kf k2 (22) Newton iteration method gives fi+1 = fi + ∇G(fi )−1 (hJω − G(fi )) , where ∇G(f ) can be expressed as ∇G(f ) = Lie Group Variational Integrator cos kf k kf k − sin kf k T sin kf k kf k − 2(1 − cos kf k) Jff + (f × Jf ) f T kf k3 kf k4 sin kf k 1 − cos kf k + J+ {−S(Jf ) + S(f )J} . kf k kf k2 21/28 Variational Principle Numerical Simulation 3D Pendulum Simulation Result 3D Pendulum Model Uniform gravity potential O.....r.............. ..... ......... ............ ...... ......... .......... e2 g ? e1 ... .... .... .... ... ..... ... ρ ..... ... .... ... ... ... .... ... .... ... ... .... ... ... .... ... ... ? .... ... ..q ... . e3 ..... ... C.M. ... . .. ... . ... .. . . . .... ................................ V = −mgeT3 Rρ, ∂V = −mge3 ρT . ∂R Equation of motion Since M = r3 × vr3 = −eT3 R × mgρT , Jωk+1 = FkT Jωk + hmgρ × RTk+1 e3 . Conserved quantities Since the Lagrangian is independent of time, and is invariant under the inertial rotation about e3 . H= 1 T ω Jω − mgeT3 Rρ, 2 π3 = eT3 RJω. are conserved. The equation of motion can be written as Rk+1 Jωk+1 − Rk Jωk = hmgRk+1 ρ × e3 , which shows the conservation of π3 directly. Lie Group Variational Integrator 22/28 Variational Principle Numerical Simulation 3D Pendulum Simulation Result Simulation Parameters Mass properties J = diag [1, 2.8, 2] kg m2 , m = 1kg, ρ = [0, 0, 1] m. Initial conditions 1 Small perturbation from the hanging equilibrium ω0 = [0.5, −0.5, 0.4] rad/s, R0 = I3×3 . 2 Small perturbation from the inverted equilibrium ω0 = [0.5, −0.5, 0.4] rad/s, R0 = diag [−1, 1, −1] . 3 Perturbation from an initially horizontal position 0 0 1 ω0 = [2.5, 2.5, 0] rad/s, R0 = 0 1 0 . −1 0 0 Lie Group Variational Integrator 23/28 Variational Principle Numerical Simulation 3D Pendulum Simulation Result Simulation Result (i) Small perturbation from the hanging equilibrium −9.174 0 −1 0 H (J) ω1 (rad/s) 1 5 10 15 20 25 30 −9.178 0 0 5 10 15 20 25 30 0.7995 0 4 0.4 || I−RTR || ω3 (rad/s) 10 15 20 25 30 5 10 15 20 25 30 5 10 15 times (s) 20 25 30 0.8 0.5 0.3 0.2 0 5 0.8005 π3 (Nms) ω2 (rad/s) 1 −1 0 −9.176 5 10 15 time (s) 20 25 30 Angular velocity (h = 10−3 (sec), Variational integrator Lie Group Variational Integrator x 10 −4 2 0 0 Conserved Quantities , Runge-Kutta ) 24/28 Variational Principle Numerical Simulation 3D Pendulum Simulation Result Simulation Result (ii) Small perturbation from the inverted equilibrium ω1 (rad/s) 10 10.5 −10 0 H (J) 0 5 10 15 20 25 10.4 0 30 0 5 10 15 20 25 30 15 20 25 30 5 10 15 20 25 30 5 10 15 times (s) 20 25 30 −0.8 −0.82 0 10 || I−RTR || ω3 (rad/s) 10 −0.78 5 0 −5 0 5 −0.76 π3 (Nms) ω2 (rad/s) 5 −5 0 10.45 5 10 15 time (s) 20 25 30 Angular velocity (h = 10−5 (sec), Variational integrator Lie Group Variational Integrator x 10 −3 5 0 0 Conserved Quantities , Runge-Kutta ) 25/28 Variational Principle Numerical Simulation 3D Pendulum Simulation Result Simulation Result (iii) Perturbation from an initially horizontal position ω1 (rad/s) 10 11.89 −10 0 H (J) 0 5 10 15 20 25 0 5 10 15 20 25 30 −2.51 0 4 || I−RTR || ω3 (rad/s) 10 15 20 25 30 5 10 15 20 25 30 5 10 15 times (s) 20 25 30 −2.5 5 0 −5 0 5 −2.49 π3 (Nms) ω2 (rad/s) 11.87 11.86 0 30 5 −5 0 11.88 5 10 15 time (s) 20 25 30 Angular velocity (h = 10−5 (sec), Variational integrator Lie Group Variational Integrator x 10 −3 2 0 0 Conserved Quantities , Runge-Kutta ) 26/28 Variational Principle Numerical Simulation 3D Pendulum Simulation Result Simulation Result Chaotic Motion : (iii) Perturbation from an initially horizontal position Lie Group Variational Integrator 27/28 Variational Principle Numerical Simulation 3D Pendulum Simulation Result Conclusion Variational Integrator for the Attitude dynamics of a Rigid body with the Potential depends on the Attitude The integrator is obtained from a discrete variational principle. It exhibits the symplectic properties and preserves conserved quantities. It adopts the approach of Lie group integrators. The discrete state evolves on a rotation group automatically without any reprojection and constraints. The general idea can be applied to other Lie groups. Numerical Simulation for the 3D Pendulum The Lie group variational integrator is used to study the dynamics of the 3D Pendulum. Some initial condition may lead to irregular or chaotic solutions. Lie Group Variational Integrator 28/28 Variational Principle Numerical Simulation 3D Pendulum Simulation Result Conclusion Variational Integrator for the Attitude dynamics of a Rigid body with the Potential depends on the Attitude The integrator is obtained from a discrete variational principle. It exhibits the symplectic properties and preserves conserved quantities. It adopts the approach of Lie group integrators. The discrete state evolves on a rotation group automatically without any reprojection and constraints. The general idea can be applied to other Lie groups. Numerical Simulation for the 3D Pendulum The Lie group variational integrator is used to study the dynamics of the 3D Pendulum. Some initial condition may lead to irregular or chaotic solutions. Lie Group Variational Integrator 28/28