Proceedings of the ASME 2011 International Design Engineering Technical Conferences &

Computers and Information in Engineering Conference

IDETC/CIE 2011
August 28-31, 2011, Washington, USA

DETC2011/MSNDC-48132

TWO LIE GROUP FORMULATIONS FOR DYNAMIC MULTIBODY SYSTEMS WITH

LARGE ROTATIONS

Olivier Bruls

Department of Aerospace and Mechanical Engineering
`
University of Liege
`
Liege,
4000
Belgium
Email: o.bruls@ulg.ac.be

Martin Arnold
NWF II - Institute of Mathematics
Martin Luther University Halle-Wittenberg
Halle (Saale), D-06099
Germany
Email: martin.arnold@mathematik.uni-halle.de

Alberto Cardona
CIMEC-INTEC
Universidad Nacional del Litoral - Conicet
Santa Fe, 3000
Argentina
Email: acardona@intec.unl.edu.ar

ABSTRACT
This paper studies the formulation of the dynamics of
multibody systems with large rotation variables and kinematic
constraints as differential-algebraic equations on a matrix Lie
group. Those equations can then be solved using a Lie group
time integration method proposed in a previous work. The general structure of the equations of motion are derived from Hamilton principle in a general and unifying framework. Then, in the
case of rigid body dynamics, two particular formulations are developed and compared from the viewpoint of the structure of the
equations of motion, of the accuracy of the numerical solution
obtained by time integration, and of the computational cost of
the iteration matrix involved in the Newton iterations at each
time step. In the first formulation, the equations of motion are
described on a Lie group defined as the Cartesian product of the
group of translations R3 (the Euclidean space) and the group
of rotations SO(3) (the special group of 3 by 3 proper orthogonal transformations). In the second formulation, the equations

Address

all correspondence to this author.

of motion are described on the group of Euclidean transformations SE(3) (the group of 4 by 4 homogeneous transformations).
Both formulations lead to a second-order accurate numerical solution. For an academic example, we show that the formulation
on SE(3) offers the advantage of an almost constant iteration
matrix.

INTRODUCTION
Current simulation tools in flexible multibody dynamics allow the study of complex and industrial systems. The motivation for this work comes from the need of simplified and more
efficient formulations for advanced studies including sensitivity
analysis, optimization, real-time simulation, experimental identification and control design. The description of large rotations
and the time integration problem are the main aspects reconsidered here for the development of simpler formulations in flexible
multibody dynamics.
The configuration of an articulated system composed of
rigid and flexible bodies can be represented by a set of absolute
1

c 2011 by ASME
Copyright

nodal translation and rotation variables, see e.g. [1]. Each translation variable belongs to the linear space R3 whereas each rotation variable belongs to the nonlinear group of special orthogonal
transformations SO(3). In classical parameterization-based approaches, rotation matrices, angular velocities as well as angular
accelerations are explicitly represented in a coordinate system.
The equations of motion are then written as differential-algebraic
equations (DAE) in the vector space Rk so that they can be solved
using standard DAE-solvers. Recently, a Lie group extension of
the generalized- method was proposed for the analysis of flexible multibody systems [2, 3]. In this approach, the integration
algorithm is directly formulated in a Lie group setting, with the
advantage that no rotation parameterization is required for the
formulation of the equations of motion. Compared to energyand momentum-preserving methods, see e.g. [4], this Lie group
method is not tailored to specific formulations of rigid bodies,
beams, or shell elements but it is general and applicable to any
mechanical system with large rotation variables.
Let us briefly review classical parameterization-based simulation strategies. A first family of methods is based on a minimal
parameterization, which means that three parameters are used to
represent each rotation variable. Due to singularity problems,
minimal parameterizations have a limited validity range and they
can only be used locally. However, this approach can be used
if a reparameterization strategy is implemented to avoid singularities, see e.g. the updated Lagrangian point of view proposed
in [1, 5]. A second family of methods is based on a global but
redundant parameterization of rotations, which allows to avoid
singularity problems at the cost of a larger set of equations to be
solved. For example, the 4 Euler parameters can be used to represent an arbitrarily large rotation but one additional constraint
should then be defined for each rotation variable [6]. Alternatively, the 9 components of the rotation matrix allow to represent
a large rotation provided the definition of 6 additional constraints
for each rotation variable [7]. The resulting equations of motion
have the structure of a DAE on the linear parameter space, which
can be solved using a suitable DAE solver.
Considering that the set of configuration variables of a flexible multibody system has a matrix Lie group structure, this paper develops a general framework based on Hamilton principle
for the formulation of the equations of motion on a Lie group.
The Lie group nature of rotational fields already played a major
role in the development of geometrically consistent models for
mechanical systems with large rotations [810]. Since kinematic
constraints are usually considered to represent rigid connections
as well as interconnections between bodies, the equations of motion have the structure of a DAE on a Lie group. These equations
can then be solved using the Lie group integration method [2, 3],
which is known to be globally second-order accurate in the DAE
case.
In the case of rigid body dynamics, two particular formulations of the equations of motion are developed and compared

from the viewpoint of the structure of the equations of motion,

of the accuracy of the numerical solution obtained by time integration, and of the computational cost of the iteration matrix involved in the Newton iterations at each time step. In the first formulation, the equations of motion are described on a Lie group
defined as the Cartesian product of the group of translations R3
(the Euclidean space) and the group of rotations SO(3) (the special group of 3 by 3 proper orthogonal transformations). In the
second formulation, the Lie group is defined as the group of Euclidean transformations SE(3) (the group of 4 by 4 homogeneous
transformations). A spinning top benchmark serves as a basis for
the comparison of the two formulations.

EQUATIONS OF MOTION
Kinematics of a multibody system
The configuration of a flexible multibody system can generally be described on a k-dimensional manifold G with a matrix
Lie group structure. From a mathematical viewpoint, a Lie group
G is a differentiable manifold for which the product (or composition) and inversion operations are smooth maps, see [11]. In an
absolute coordinate formulation, an element q G is composed
of several subsets of absolute nodal translations and rotations,
a priori considered as independent variables. The composition
operation G G G is written as a matrix product
qtot = q1 q2

(1)

with q1 , q2 , qtot G and the identity element e is simply the identity matrix. Tq G denotes the tangent space at a point q G and
the Lie algebra is defined as the tangent space at the identity
g = Te G. The Lie algebra is a vector space, which is isomorphic
to Rk by an invertible linear mapping
f : Rk g, v 7 e
()
v.

(2)

A tangent vector at any point q can be represented in the Lie

algebra using the left translation map Lq . Indeed, Lq is a diffeomorphism of G
Lq : G G, y 7 q y

(3)

and its derivative defines a diffeomorphism between Ty G and

Tq y G. In the particular case y = e, we thus have a bijection between Te G = g and Tq G :
e 7 q w
e
DLq (e) : g Tq G, w

(4)

e is the directional derivative of Lq evaluated at

where DLq (e) w
e g. Hence, a tangent vector w
eg
point e in the direction w
2

c 2011 by ASME
Copyright

where q Rk is the virtual displacement vector such that

defines a left invariant vector field on G which is constructed by

e to the tangent space at any point of G. For
left translation of w
example, at any configuration, the velocity vector e
v g or v Rk
is defined from the time derivative q such that

q = q fq,

(11)

g(q) is the vector of external and internal forces such that

q = qe
v.

(5)
DV (q) fq = qT g(q),

In a multibody system, the nodal translation and rotation

variables are generally not independent but they have to satisfy
a set of m kinematic constraints : G Rm , which restrict the
dynamics to the submanifold N of dimension k m
N = {q G : (q) = 0}.

and B is the m k matrix of constraint gradients such that

e = B(q)w.
D(q) w

(6)

e
v = fq + [e
v, fq]

(7)

v = q + b
v q

Z tf

(L (q, v) (q)T ) dt = 0

d
qT Mv = q T Mv + qT Mv
dt
= vT Mv + qT (Mv b
vT Mv).

(8)

[ qT Mv]tif

(17)

Z tf
ti

(( qT (Mv b
vT Mv + g(q) + BT (q) )

(18)

+ T (q)) dt = 0

where the first term vanishes due to the boundary conditions.

Finally, the stationarity conditions lead to a set of index-3
differential-algebraic equations
q = qe
v

(9)

where Rm is the vector of Lagrange multipliers associated

with the constraints . This equation becomes

ti

(16)

Integrating Equation (10) by part leads to

ti

Z tf

(15)

where b
is a linear operator of the Lie group which transforms a
k 1 vector into an k k matrix. Therefore, we have

where M is the k k symmetric mass matrix. Since the kinetic

energy is expressed as a function of the left-invariant velocity
vector, rotational inertia are defined in the body-attached frame
and the mass matrix remains constant during motion. This is
an advantage of using a left invariant representation of velocities
compared to a right invariant representation.
According to Hamilton principle and following an augmented Lagrangian method, the actual trajectory of the system
between two time instants ti and t f is such that the variation of
the augmented action integral is stationary provided that the initial and final configurations are fixed, i.e.

(14)

e =e
e be
ea is the matrix commutator. Since the comwhere [e
a, b]
ab
mutator is linear with respect to both arguments, the above expression can be written in terms of vectors in Rk

where K and V respectively denote the kinetic and potential

energies of the system. The kinetic energy is a quadratic form in
the velocity
1
K (v) = vT Mv
2

(13)

Combining Equations (5) and (11), one observes that

Dynamics of a Multibody System

For a conservative multibody system, the Lagrangian function can be written as a function of the configuration of the system q G and of the velocity vector v Rk
L (q, v) = K (v) V (q)

(12)

(19)

Mv b
v Mv + g(q) + B (q) = 0k1

(20)

(q) = 0m1

(21)

These equations of motion allow to represent the dynamics of a

general class of conservative flexible multibody systems. It is
also possible to extend the formulation in order to account for
non-conservative forces. We observe that no parameterization of
rotations is needed to formulate those equations.

( vT Mv qT g(q) qT BT (q) T (q)) dt = 0

(10)
3

c 2011 by ASME
Copyright

EXPONENTIAL MAP
As seen in the previous section, the time derivative of the
configuration variable q is conveniently represented in the Lie
algebra using the left translation map of the group. The exponential map, which maps any element of the Lie algebra to the
Lie group
exp : g G,

e 7 q = exp(e
q
q)

A comparison of the last two equations leads to a linear relationship between q and v
e = exp(e
Dexp(e
q) q
q)e
v.

In literature, this relation is sometimes denoted as the left trivialized derivative of the exponential map

(22)

dexpqe : g g,

is a useful tool for the design of Lie group integrators. Its mathematical definition is related with integral curves of left (or right)
invariant vector fields. Accordingly, the solution of the fundamental equation

e 7 e
q
v

q(0) = q0

(30)

and it can also be represented as a linear relation from Rk to Rk

v = T(q)q

e,
q(t)
= qw

(29)

(31)

(23)
where T(q) is the so-called k k tangent operator of the exponential map. It admits the series expansion [12]

e g is given by
for a constant w
e ).
q(t) = q0 exp(t w

(1)i i
b.
q
i=0 (i + 1)!

(24)

T(q) =

(32)

The exponential map of a matrix Lie group admits the series expansion
TIME INTEGRATION METHOD
The equations of motion defined on the Lie group G can
be solved using a Lie group time integrator without the need of
defining generalized coordinates. In this work, we exploit the
Lie group time integrator described and analyzed in [2, 3], which
is inspired by the index-3 formulation of the generalized- time
integration scheme [13] as well as by the work of Crouch and
Grossmann [14] and Munthe-Kaas [15, 16].
The Lie group time integrator is based on the discretized set
of equations:

1 i
e.
q
i=0 i!

exp(e
q) =

(25)

The exponential map allows the construction of a local parameterization of G about an arbitrary point q0 G. Indeed, the
relation
q = q0 exp(e
q)

(26)
Mvn+1 b
vTn+1 Mvn+1 = g(qn+1 ) BT (qn+1 ) n+1
(qn+1 ) = 0m1


fn
qn+1 = qn exp hq

defines a diffeomorphism between G and g and consequently between G and Rk . This diffeomorphism can be written as a coordinate map Rk G : q 7 q = q0 exp(e
q).
The derivative of the exponential map can be obtained by
differentiation of Equation (26) with respect to time
e
q = q0 (Dexp(e
q) q).

(35)

qn = vn + (0.5 )han + han+1 (36)

vn+1 = vn + (1 )han + han+1
(1 m )an+1 + m an = (1 f )vn+1 + f v n

(27)

(37)
(38)

f n may
where h = tn+1 tn is the time step size. The variable q
be interpreted as an average and frozen velocity field in the
time interval between tn and tn+1 . The particular form of Equation (35) makes this formulation applicable to dynamic systems
on matrix Lie groups. This method is second-order accurate for
index-3 differential-algebraic problems.

This equation defines a relation between q and the time derivative

Using Equations (5) and (26),
of the Lie algebra coordinates q.
q can also be written in terms of the left invariant derivative
q = q0 exp(e
q)e
v

(33)
(34)

(28)
4

c 2011 by ASME
Copyright

[uT T ]T with = [1 2 3 ]T . If v denotes a velocity, Eq. (5)

yields

The (k + m) (k + m) iteration matrix St is evaluated as

M 0 + Ct (v) 0 + Kt (q)T(hq)
St (q, hq, v) =
B(q)T(hq)


BT (q)
.
0mm
(39)
with the algorithmic parameters 0 = (1 m )/( h2 (1 f ))
and 0 = / h. In this expression, Ct is the tangent damping
matrix such that D(b
vT Mv) v = Ct v and Kt is the tangent
e = Kt w. For small time steps
stiffness matrix such that Dg(q) w
h, the matrix St becomes severely ill conditioned. This difficulty
can be eliminated by the implementation of a suitable scaling
strategy, see e.g. [13, 17, 18].


x = u
e

R = R

(43)
(44)

so that u is interpreted as the vector of translation velocities in

the inertial frame and is the vector of angular velocities in the
body-attached frame. We also have

b
v=

APPLICATION TO RIGID BODY DYNAMICS

The free motion of a single rigid body with combined 3D
translations and rotations can either be represented in the special
Euclidean group SE(3) as in [19] or in R3 SO(3) as in [2, 3].
These two options are investigated in the following and it is
shown that the formulation on R3 SO(3) leads to a representation of translation velocities in the inertial frame whereas the
formulation on SE(3) leads to a representation of translation velocities in the body-attached frame.


033 033
.
e
033

(45)

For a vector q = [xT T ]T , the exponential map is computed

as

expSO(3) () 033 031

expR3 SO(3) (x, ) = 033

I3 x
013
013 1

(46)

where expSO(3) is given by Rodrigues formula:

R3 SO(3)

System on
The configuration of a rigid body can be represented by the
pair q = (x, R) with the translation vector x R3 and the rotation matrix R SO(3). The set R3 SO(3) is a 6-dimensional
Lie group with the composition operation defined as (x1 , R1 )
(x2 , R2 ) = (x1 + x2 , R1 R2 ), which can be written as a matrix
product if the element (x, R) is represented by the 7 7 matrix

expSO(3) () = I3 +

I
033
TR3 SO(3) (x, ) = 3
033 TSO(3) ()

(40)

TSO(3) () = I3 +

(41)

(48)



e
e
cos 1
sin
e

+
1

.
2
2

(49)

System on SE(3)
The special Euclidean group SE(3) is also isomorphic to
R3 SO(3). However, the composition operation of two elements q1 = (x1 , R1 ) and q2 = (x2 , R2 ) is defined as the matrix
product between 4 4 homogeneous transformation matrices of
the form

e is a skew-symmetric matrix
where

0 3 2
e = 3 0 1 .

2 1 0

where TSO(3) () is given by the formula

The Lie algebra is the set of 7 7 matrices

033 031
e
v = 033 033 u
013 013 0

(47)

The axial vector R3 is the so-called Cartesian rotation vector

and = kk. The tangent operator of the exponential map is

R 033 031
.
q = 033 I3 x
013 013 1

1 cos
sin
e .
e
e+

(42)

The Lie algebra can be identified to R6 since any matrix e

v defined
according to Eq. (41) can be represented by the 6 1 vector v =


R x
.
013 1

(50)

c 2011 by ASME
Copyright

At any point q, the tangent space is noted Tq SE(3) and the

Lie algebra se(3) is the set of 4 4 matrices


U
e
v=
013 0


(51)

FIGURE 1.

e was defined in Eq. (42). Any matrix of the Lie algebra

where
e
v g(6) can be represented by the 6 1 vector v = [UT T ]T . If
v denotes a velocity, Eq. (5) yields
x = RU
e

R = R

CASE STUDY
As shown in Figure 1, the heavy top is a rotating body fixed
to the ground by a spherical joint. In principle, the motion could
be described in SO(3) if the equilibrium of momentum were expressed with respect to the fixed point. However, in order to test
our algorithms, the translation of the center of mass and the rotation of the body are considered here as independent variables.
The configuration of the system is thus represented by q = (x, R)
where x is the translation of the center of mass and R is the rotation matrix of the body. Due to the fixed point condition, the
variables x and R have to satisfy some kinematic constraints. The
kinetic energy, the potential energy and the constraints take the
following expressions

(52)
(53)

so that U is interpreted as the vector of translation velocities in

the body-attached frame and is the vector of angular velocities
in the body-attached frame. We also have
#
e
e

U
b
v=
e .
033

"

(54)

1
1 T
mx x + T J
2
2
V (q) = xT m

For a vector q = [xT T ]T , the exponential operator is computed as


expSE(3) (x, ) =

expSO(3) () A()x
013
1

K (x, ) =

(q) = R x + X


(55)


1  2
e +
e
e
I3 + (I3 expSO(3) ())
2

TSO(3) () C 0.5e
x
033
TSO(3) ()

(60)
(61)

(56)
Equations of motion on R3 SO(3)
Since v = [uT T ]T , the kinetic energy can be written as in
Eq. (8) with the constant mass matrix

and = kk. The tangent operator of the exponential map is

TSE(3) (x, ) =

(59)

where is the angular velocity in the body-attached frame, m is

the mass of the body, J is the tensor of inertia with respect to the
center of mass in the body-attached frame and X is the position
of the center of mass in the reference configuration R = I3 .

with
A() =

Spinning top example.


(57)


mI3 033
M=
.
033 J

with

(62)

The inertia forces are computed as

C=

1
1
T
e + e
e x)
e
e
x + 2 (e
x
( x)
2

2
1 1
3
e
e
+ 2
2 (1 ) ( T x)


mu
Mv b
v Mv =
.
e
+ J
J
T

(58)

(63)

According to Eq. (11), the virtual displacement is represented by the vector q = [ xT T ]T where is the angular

and = 2 sin( /2) cos( /2)/ , = 4 sin2 ( /2)/ 2 .

6

c 2011 by ASME
Copyright

The inertia forces are evaluated as

variation vector such that R = Rf

. The internal and external
forces are computed from the variations of the potential energy

"

V (q) = xT m

#
e
+ mU
mU
Mv b
v Mv =
.
e
+ J
J
T

(64)

so that

According to Eq. (11), the virtual displacement is represented by the vector q = [ DT T ]T where D and are
such that x = R D and R = Rf
. The internal and external
forces are computed from the variations of the potential energy


m
g(q) =
.
031

(65)

V (q) = xT m = DT RT m

The constraint gradient is evaluated from the variations of

the constraints
(q) = RT x RT x
= f
RT x RT x
g
T x RT x
= R

(76)

(77)

so that
(66)


(67)

g(q) =

(68)


RT m
.
031

(78)

The constraint gradient is evaluated from the variations of

the constraints

which leads to
B(q) = [RT

g
T x].
R

(69)

(q) = RT x RT x
g
T x D
= R

At equilibrium, this expression becomes

B(q) = [RT

e
X].

B(q) = [I3

In summary, the equations of motion on R3 SO(3) are

(71)

+ J + X = 031
J

(72)

R x + X = 031

g
T x].
R

B(q) = [I3

e
X].

(82)

(73)
This matrix of constraint gradients is constant during motion, in
other words, B is independent of the configuration of the body
q. This is a noticeable difference compared to the formulation on
R3 SO(3).
In summary, the equations of motion on SE(3) are

(74)

+ m U RT m = 031 ,
mU
+ J + X = 031 ,
J

with q = [xT T ]T .

RT x + X = 031

Equations of motion on SE(3)

Since v = [UT T ]T , the kinetic energy can be written as in
Eq. (8) with the constant mass matrix

mI3 033
.
M=
033 J

(81)

At equilibrium, this expression becomes

and the iteration matrix is

mI3 0
Re T(h)
R

e J)
f 0 X
e
St (q, hq, v) = 033 J 0 + (J

e
RT
XT(h)
033

(80)

which leads to

(70)

mu m R = 031

(79)

(83)
(84)
(85)

and the iteration matrix is

^
T mT(h) I
e 0 mU
e 0 R
mI3 0 + m
3

e J)
f 0
e
St =
. (86)
033
J 0 + (J
X
e
I3
XT(h)
033

(75)
7

c 2011 by ASME
Copyright

u3 U3 (m/s)

0.2

x3 (m)

10

R3xSO(3)
SE(3)

0.4
0.6
0.8
1
0

0.5

1.5

10
0

R3xSO(3)
SE(3)

0.5

time (s)
3

1
0
1

3
0

0.02

0.04

0.06

0.08

5434
5433

5431
0

0.1

time (s)
600

FIGURE 3.

0
200

0.02

0.04

0.06

1.5

0.08

Fast spinning top: velocity and energy.

In Figures 2 and 3, the numerical results are obtained using a large time step h = 0.002 s and the algorithmic parameters
are defined according to the Chung-Hulbert scheme [20] in order
to have a spectral radius at infinity = 0.8. The solution obtained with the two Lie group formulations are compared. The
vertical displacement of the center of mass x3 is shown in the
interval [0, 2] s and the fast rotation of the top is represented by
the second component of the Cartesian rotation vector 2 in the
interval [0, 0.1] s. Figure 2 also shows that the formulation on
R3 SO(3) leads to numerical oscillations in the multiplier 1
at the beginning of the simulation, which are then stabilized by
the time integrator. No oscillations are observed in the results
for the Lie group formulation on SE(3). From the comparison
of vertical displacements x3 in Fig. 2 and of the energy drifts in
Fig. 3, the formulation on SE(3) appears less accurate than the
formulation on R3 SO(3). Figure 3 also shows that the velocity U3 in the formulation on SE(3) is modulated by the spinning
frequency, which is not the case for the velocity u3 in the formulation on R3 SO(3). The presence of this high frequency
content may explain the lower accuracy of the formulation on
SE(3) in this particular case.
In order to analyze the convergence when the time step size
decreases, a reference solution was computed using a small time
step h = 1.25e-4 s. In agreement with the theoretical convergence
properties, second-order accuracy is observed both for differential and algebraic variables x and for the two algorithms. The

200

400
0

0.5

time (s)
R3xSO(3)
SE(3)

400

1 (N)

5432

0.1

time (s)
FIGURE 2.

1.5

R3xSO(3)
SE(3)

5435

Energy (J)

2 (rad)

5436

R3xSO(3)
SE(3)

time (s)

Fast spinning top: displacement, rotation and multipliers.

The constant nature of the constraint gradients has a direct consequence on the fluctuations of the iteration matrix during motion.
We shall see in the following example that the iteration matrix
remains nearly constant during motion if the dynamics is formulated on SE(3).
Results
In the numerical tests, the parameters of the model are
defined as m = 15 kg, J = diag(0.234375, 0.46875, 0.234375)
kg m2 , X = [0. 1. 0.]T m, = [0. 0. 9.81]T m/s2 and the initial
conditions are R(0) = I3 , (0) = [0. 150. 4.61538]T rad/s. The

initial values x (0), x (0) and (0)

are set to be consistent with the
constraints and the equations of motion. The initial values for u,
are derived from those data.
U, u and U
8

c 2011 by ASME
Copyright

10

R3xSO(3)
SE(3)

R3xSO(3)
SE(3)

10

u3 U3 (m/s)

error x (m)

10

10

10 4
10

10

10

4
0

h (s)

0.5

10

R3xSO(3)
SE(3)

0.548

10

10

0.544
0.542
0.54

10

10

0.538
0

h (s)
FIGURE 4.

1.5

R3xSO(3)
SE(3)

0.546

Energy (J)

error (N)

10

10 4
10

time (s)

0.5

1.5

time (s)

Fast spinning top: convergence analysis.

FIGURE 5.

results are shown in Fig. 4 and, again, the formulation on SE(3)

appears less accurate in this particular case.
The analysis was repeated with the initial velocity divided
by 100, i.e. (0) = [0. 1.50 0.0461538]T rad/s. The results
are shown in Figs. 5 and 6. In this case, the frequency content
of the variables u3 and U3 are similar and, unlike for the fast
spinning top, the formulation on SE(3) leads to more accurate
results than the formulation on R3 SO(3).
Finally, the required frequency of recalculation of the iteration matrix is analyzed. An algorithm was implemented where
the iteration matrix is updated only if the number of iterations at
the current time step is a multiple of 6. In the case h = 0.002 s
and after 1000 steps, the iteration matrix was updated 529 times
for the formulation on R3 SO(3), whereas a single evaluation
of the iteration matrix at the begining of the simulation was sufficient for the formulation on SE(3). This observation is explained
by the nearly constant nature of the iteration matrix in this case.

Slow spinning top: velocity and energy.

10

R3xSO(3)
SE(3)

error x (m)

10

10

10

10 4
10

10

10

h (s)
2

10

R3xSO(3)
SE(3)

error (N)

10

CONCLUSION
This paper studies the formulation of the dynamics of multibody systems with large rotation variables and kinematic constraints as differential-algebraic equations on a matrix Lie group.
The general structure of the equations of motion is derived from
Hamilton principle in a general and unifying framework. Then,
two particular formulations are developed and compared for rigid

10

10

10 4
10

10

10

h (s)
FIGURE 6.

Slow spinning top: convergence analysis.

c 2011 by ASME
Copyright

body dynamics: in the first one, the equations of motion are described on the Lie group R3 SO(3), and in the second one,
they are described on the Lie group SE(3). It is shown that the
formulation on R3 SO(3) leads to a representation of translation velocities in the inertial frame whereas the formulation on
SE(3) leads to a representation of translation velocities in the
body-attached frame.
A spinning top benchmark serves as a basis for the comparison of the two formulations. Second-order accuracy is observed
in all results. In the formulation on SE(3), the high speed spinning frequency modulates the response so that the frequency content is higher than in the formulation on R3 SO(3). Therefore,
for fast spinning speeds, the accuracy is better for the formulation on R3 SO(3), whereas the formulation on SE(3) leads to
more accurate results for slow spinning speeds. This aspect is related to the fact that the former formulation is built in the inertial
frame, while the latter is constructed in the body-attached frame.
In the formulation on SE(3), the constraint gradient is constant. We observe that the numerical solution is more stable in
this case and that the iteration matrix is nearly constant during
motion. If the number of evaluations and factorizations of the
iteration matrix can be reduced during a simulation, important
savings are expected in the computational cost.
In previous investigations [3], the formulation on R3
SO(3) was successfully applied for the analysis of flexible multibody systems including nonlinear elastic beam models. Future
work will address the comparison of the formulations on SE(3)
and R3 SO(3) for systems composed of several rigid bodies,
kinematic joints and flexible bodies.

[7] Betsch, P., and Steinmann, P., 2001. Constrained integration of rigid body dynamics. Computer Methods in Applied Mechanics and Engineering, 191, pp. 467488.
[8] Simo, J., and Vu-Quoc, L., 1988. On the dynamics in
space of rods undergoing large motions - A geometrically
exact approach. Computer Methods in Applied Mechanics
and Engineering, 66, pp. 125161.
[9] Simo, J., and Wong, K., 1991. Unconditionally stable algorithms for rigid body dynamics that exactly preserve energy and momentum. International Journal for Numerical
Methods in Engineering, 31, pp. 1952.
[10] Cardona, A., and Geradin, M., 1988. A beam finite element non-linear theory with finite rotations. International Journal for Numerical Methods in Engineering, 26,
pp. 24032438.
[11] Boothby, W., 2003. An Introduction to Differentiable Manifolds and Riemannian Geometry, 2nd ed. Academic Press.
[12] Iserles, A., Munthe-Kaas, H., Nrsett, S., and Zanna, A.,
2000. Lie-group methods. Acta Numerica, 9, pp. 215
365.
[13] Arnold, M., and Bruls, O., 2007. Convergence of
the generalized- scheme for constrained mechanical systems. Multibody System Dynamics, 18(2), pp. 185202.
[14] Crouch, P., and Grossman, R., 1993. Numerical integration of ordinary differential equations on manifolds. Journal of Nonlinear Science, 3, pp. 133.
[15] Munthe-Kaas, H., 1995. Lie-Butcher theory for RungeKutta methods. BIT, 35, pp. 572587.
[16] Munthe-Kaas, H., 1998. Runge-Kutta methods on Lie
groups. BIT, 38, pp. 92111.
[17] Cardona, A., and Geradin, M., 1994. Numerical integration of second order differential-algebraic systems in flexible mechanism dynamics. In Computer Aided Analysis of
Rigid and Flexible Mechanical Systems, J. A. C. Ambrosio
and M. F. Seabra Pereira, eds., Vol. 268 of NATO ASI series.
Kluwer Academic Publishers, Dordrecht, pp. 501529.
[18] Bottasso, C., Bauchau, O., and Cardona, A., 2007. Timestep-size-independent conditioning and sensitivity to perturbations in the numerical solution of index three differential algebraic equations. SIAM Journal on Scientific Computing, 29(1), pp. 397414.
[19] Park, J., and Chung, W., 2005. Geometric integration on
Euclidean group with application to articulated multibody
systems. IEEE Transactions on Robotics, 21(5), pp. 850
863.
[20] Chung, J., and Hulbert, G., 1993. A time integration algorithm for structural dynamics with improved numerical
dissipation: The generalized- method. ASME Journal of
Applied Mechanics, 60, pp. 371375.

REFERENCES
[1] Geradin, M., and Cardona, A., 2001. Flexible Multibody
Dynamics: A Finite Element Approach. John Wiley &
Sons, New York.
[2] Bruls, O., and Cardona, A., 2010. On the use of Lie group
time integrators in multibody dynamics. ASME Journal of
Computational and Nonlinear Dynamics, 5(3), p. 031002.
[3] Bruls, O., Arnold, M., and Cardona, A., 2010. Numerical
solution of daes in flexible multibody dynamics using lie
group time integrators. In Proceedings of the First Joint
International Conference on Multibody System Dynamics.
[4] Armero, F., and Romero, I., 2003. Energy-dissipating
momentum-conserving time-stepping algorithms for the
dynamics of nonlinear cosserat rods. Computational Mechanics, 31, pp. 326.
[5] Cardona, A., and Geradin, M., 1989. Time integration of
the equations of motion in mechanism analysis. Computers and Structures, 33, pp. 801820.
[6] Lang, H., Linn, J., and Arnold, M., 2011. Multibody dynamics simulation of geometrically exact Cosserat rods.
Multibody System Dynamics, 25, pp. 285312.
10

c 2011 by ASME
Copyright

All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately.