June 9, 2008 18:54 WSPC/165-IJSSD 00264

International Journal of Structural Stability and Dynamics

Vol. 8, No. 2 (2008) 257–270

c World Scientific Publishing Company

IMPROVING STABILITY IN THE TIME-STEPPING

ANALYSIS OF STRUCTURAL NONLINEAR DYNAMICS

S. LOPEZ∗ and K. RUSSO†

Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com

Dipartimento di Strutture, Università della Calabria

Rende, 87030, Italy
∗[email protected]
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

†[email protected]

Received 8 December 2006

Accepted 19 June 2007

A change in the representation of discrete motion equations for nonlinear structural

dynamics of two-dimensional bodies is developed. The objective is to write the motion
equation in a less nonlinear form. This leads to a significant increase in the range of sta-
bility of the time integration process and a reduction in the number of Newton iterations
required in the time integration step.

Keywords: Numerical stability; nonlinear structural dynamics.

1. Introduction
Numerical instability frequently appears in nonlinear regimes, so the main interest
of numerical dynamics is in the numerically stable algorithm. In order to obtain
stable solutions, schemes which demand the conservation of or a decrease in the
total energy of the Hamiltonian system within each time step are used extensively.
For these nonlinear models Belytschko and Schoeberle1 and Hughes et al.2 proved
that the discrete energy is bounded if it remains positive.
Nevertheless, large instabilities can arise, leading to divergence in the numeri-
cal simulation. The energy–momentum method introduced by Simo and Tarnow3
preserves energy as well as linear and angular momentum in the time interval.
This algorithm was extended to shells by Kuhl and Crisfield,4 where conservation
properties are enforced in the equation of motion via Lagrange multipliers. Armero
and Petöcz5 presented a numerically dissipative version of the energy–momentum
method. The application of this unconditionally stable energy decaying algorithm
to the nonlinear dynamics of three-dimensional beams was presented by Crisfield
et al.6 Adaptive time-stepping procedures7 and controllable numerical dissipation8,9
can also be introduced to permit larger time steps and, consequently, obtain better
computational efficiency.

257
June 9, 2008 18:54 WSPC/165-IJSSD 00264

258 S. Lopez & K. Russo

Although finite difference methods appear to prevail in the literature on the

numerical treatment of initial value problems, a number of alternative finite ele-
ment methods have been developed for the temporal discretization process (see
the textbook by Crisfield10 ). A weighted residual statement and the Galerkin finite
element approach to the numerical solution of the equations of motion have been
employed by Zienkiewicz et al.11 and by Lasaint and Raviart.12 More recently,
time finite elements, where the Newmark family formulas can be recovered by
the choice of representative constants and the algorithmic energy conservation is
implicitly preserved, have been carried out (see Refs. 13 and 14 and the related
bibliography).
Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com

In all cases it seems that appropriate representation and discretization of the

motion equation leads to an appreciable difference in the time integration schemes
with regard to stability. Argyris et al.,15 in particular, noted a significant increase
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

in the range of stability of the classical Newmark method if the natural mode
finite element discretization and lumped mass matrices are used. Of course, all the
above-mentioned time-stepping algorithms can take additional advantage of these
opportune representations.
Following the idea of Lopez,16 this paper reports, for two-dimensional elastic-
ity problems, how the average acceleration Newmark scheme is applied to motion
equations represented in two different sets of state variables: the first representa-
tion makes use of the typical Cartesian coordinates; the second one is obtained by
adding other local variables. In the latter case, we can write the motion equations
in a less nonlinear form since the rational expressions of the internal forces in the
Cartesian coordinates are substituted by analytical trigonometric versions. In fact,
the choice of additional local variables is strictly linked to the representation of
the internal energy and the related internal forces that are generically governed by
measurement strain. So, through the introduction of new unknowns and related
motion equations, the internal strains can be expressed more simply, being written
in their intrinsic reference system. This leads to a significant increase in the range
of stability of the time integration process.
The increase in computational effort due to the introduction of new unknowns
and the loss of symmetry in the iteration matrix is balanced by a reduction in the
number of Newton iterations required in the time integration steps.
This paper is organized as follows. In Sec. 2, the discrete motion equations are
described. Section 3 presents the finite element formulation and the change in the
representation of the internal forces. Section 4 contains representative numerical
examples of application of the proposed method. Some final conclusions are drawn
in Sec. 5.

2. Discrete Motion Equations

The topic has arise from the study of the evolution of a structure submitted to
forces inherent in the deformations; for example, forces involved in the temporal
June 9, 2008 18:54 WSPC/165-IJSSD 00264

Improving Stability in the Time-Stepping Analysis of Structural Nonlinear Dynamics 259

evolution of the same body, such as inertia forces. We schematize the steps to obtain
the solution of the system of the motion equations.
We consider the vector of the displacements u of three spatial components u, v, w
which depend on the original position of the body x = (x, y, z) and on the time t,
i.e. u = u(x, t).
The principal energetic quantities involved are the potential energy,


1
V (u) = εT A ε dΩ, (2.1)
2 Ω
in which ε = ε(u) is the deformation vector of the body and Ω the two-dimensional
domain of the body; the kinetic energy,
Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com

1
T (u̇(x, t)) = ρ u̇2 dΩ, (2.2)
2 Ω
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

which through the mass density ρ is tied to the derivative of the displacements with
respect to the time t (the velocities); and the external energy, of which the external
loads p and the displacements u appear:


L(u(x))(t) = pT u dΩ. (2.3)
Ω

By means of a finite element approach the displacements in the body are located
in n displacements in discrete form. With this spatial discretization and including
the boundary conditions, we obtain the following semidiscrete formulation of the
equations of the motion:

 Mü(t) + N(u(t)) − P(t) = 0,

u(0) = u0 , (2.4)


u̇(0) = u̇0 ,
in which ü denotes the accelerations; and u0 and u̇0 represent, respectively, the
initial displacements and initial velocities. The inertia forces Mü, the internal forces
N(u(t)) and the external forces P(t) are defined as follows:
 
∂ ∂T (u̇(t))
Mü(t) = ,
∂t ∂ u̇
∂V (u(t))
N(u(t)) = , (2.5)
∂u
∂L(u(t))
P(t) = .
∂u
We proceed now to the time integration of the nonlinear semidiscrete initial
value problem (2.4). In the following, we assume that the time step ∆t = tn+1 − tn
is constant and that the displacement, velocity and acceleration vectors at the
time tn , denoted by un , u̇n , ün , respectively, are known. The time integration is
restricted to the successive solution of the state variables at the end for each step
June 9, 2008 18:54 WSPC/165-IJSSD 00264

260 S. Lopez & K. Russo

un+1 , u̇n+1 , ün+1 . In order to realize this step-by-step integration, the set of vari-
ables is reduced to the displacement un+1 alone by the Newmark approximations:
   
γ γ γ
u̇n+1 = (un+1 − un ) + 1 − u̇n + 1 − ∆tün ,
β∆t β 2β
  (2.6)
γ 1 1
ün+1 = (un+1 − un ) − u̇n + 1 − ün .
β(∆t)2 β∆t 2β
In the following we use the average acceleration scheme by adopting γ = 1/2
and β = 1/4. By replacing the relations (2.6) in the initial value problem (2.4),
we arrive at the nonlinear system of algebraic equations defined at the time tn+1 ,
where the vector unknowns are un+1 of the form
Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com

F(un+1 ) = 0. (2.7)
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

We use the Newton-like iteration scheme for the solution of nonlinear equations
(2.7) by linearization:
(k)
(k+1) (k) ∂F(un+1 ) (k+1) (k)
F(un+1 ) = F(un+1 ) + (un+1 − un+1 ) + O(∆t2 ). (2.8)
∂un+1
(0)
The initial conditions in this iteration are obtained by choosing un+1 as the linear
extrapolation of the previously computed un and un−1 vectors when n > 0, while
(0)
the formula u1 = u0 + ∆tu̇0 is used when n = 0. The iteration is continued
until an appropriate convergence criterion is satisfied, in this case making the fixed
tolerance η = 10−8 , until the formula
(k+1) (k)

un+1 − un+1

(k+1)
≤η (2.9)

un+1 − un

is satisfied.
At this point, with nu indicating the number of components of the state vector
u(t), we define nq as the number of components of the new state vector q(t). In the
following we assume that the components of u(t) follow the Cartesian coordinates
while the components of q(t) follow a suitable local frame. We require that q(t) =
q(u(t)). The adopted definition of internal force is now

∂V (u(t)) 
N(q(t)) =  , (2.10)
∂u 
u(t)=u(q(t))

where we assume the well-posed inverse relation u(t) = u(q(t)). We denote with
(u) the Cartesian space, (q) the local space and (u,q) the nuq -dimensional new
representation. Note that (u) ⊂ (u, q), (q) ⊂ (u, q), and in the case (u) ∩ (q)= ∅,
nuq ≤ nu + nq . The system (2.4) is then defined by nuq unknowns, where nu
equations are
g(u, q) = Mü(t) + N(q(t)) − P(t) = 0, (2.11)
with the remaining nuq − nu equations aimed at satisfying the q(t) = q(u(t))
relation.
June 9, 2008 18:54 WSPC/165-IJSSD 00264

Improving Stability in the Time-Stepping Analysis of Structural Nonlinear Dynamics 261

Initial conditions are defined by

∂q(u) ∂ q̇(u)
q(t) = q(u(t)), q̇(t) = u̇(t), q̈(t) = u̇(t),
∂u ∂u
while boundary conditions remain imposed on the u components. In the following
we refer to internal energy V (q) computed in the (q) representation and classical
T (u̇) kinetic energy in the (u) representation, and we use the trapezoidal rule to
calculate the work of the external forces:
1
Ln+1 = Ln + ∆Ln , ∆Ln = (un+1 − un )T (Pn+1 − Pn ). (2.12)
2
The time integration scheme, finally, follows as outlined above in the nuq
Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com

space (u(t), q(t)), where the approximations (2.6) are also used for the q̇n and
q̈n definitions.
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

3. Finite Element and Change of Representation

We have discretized the two-dimensional body by using two types of finite elements.
In the first case a four-node element with bilinear interpolation of u(x, y) and v(x, y)
is considered:
u(x, y) = a00 + a10 x + a01 y + a11 xy,
(3.1)
v(x, y) = b00 + b10 x + b01 y + b11 xy.
In the second case a six-node element interpolation of u(x, y) and v(x, y) is as
follows:
u(x, y) = a00 + a10 x + a01 y + a20 x2 + a11 xy + a21 x2 y,
(3.2)
v(x, y) = b00 + b10 x + b01 y + b20 x2 + b11 xy + b21 x2 y.
By imposing the compatibility conditions along the edges of the element boundaries,
the coefficients aij and bij in (3.1) and (3.2) are calculated. In this way the previous
expansions are in the usual form of interpolation functions related to the respective
nodal displacements. By using the classical Green deformations the potential energy
V (u(x, t)) is obtained.
The main aim of our study is the use of variables, like the polar change, in
order to represent the finite element. In particular, in the case of the four-node
element, eight new variables are defined and the evolution of the element is described
observing the evolution of these new variables. We have chosen the new variable
vector q(t) = (L1 , α1 , L2 , α2 , L3 , α3 , uG , vG ) as follows (see Fig. 1):
L1 : length of the segment that combines the nodal points 1 and 2;
α1 : angle between L1 and the initial position;
L3 : length of the segment that combines the nodal points 3 and 4;
α3 : angle between L3 and the initial position;
L2 : length of the segment that combines the midpoints of the segments L1 and L3 ;
α2 : angle between L2 and the initial position;
uG , vG : displacement of the center of mass.
June 9, 2008 18:54 WSPC/165-IJSSD 00264

262 S. Lopez & K. Russo

Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

Fig. 1. Change of representation of the four-node element.

The new variables depend in a nonlinear way on the old ones, as the following
relationships show:

 (u3 − u1 ) + L1 sin α1 = 0,





 (v3 − v1 ) − L1 cos α1 + 2hy = 0,



 1

 (u2 + u4 − u1 − u3 ) − L2 cos α2 + 2hx = 0,

 2



 1


 2 (v2 + v4 − v1 − v3 ) − L2 sin α2 = 0,
(3.3)

 (u 4 − u 2 ) + L 3 sin α 3 = 0,





 (v4 − v2 ) + L3 cos α3 + 2hy = 0,




 1
 (u1 + u2 + u3 + u4 ) − uG = 0,



 4



 1 (v + v + v + v ) − v = 0,
1 2 3 4 G
4
where 2hx and 2hy are the length and width of the element, respectively. The
subscripts denote the vector q(t) = (L1 , α1 , L2 , α2 , L3 , α3 , L4 , α4 , L5 , α5 , uG , vG ),
which collects 12 new variables of the six-node element (see Fig. 2):

L1 : length of the segment that combines the nodal points 1 and 2;

α1 : angle between L1 and the initial position;
L2 : length of the segment that combines the nodal points 3 and 4;
α2 : angle between L2 and the initial position;
L3 : length of the segment that combines the nodal points 5 and 6;
α3 : angle between L3 and the initial position;
June 9, 2008 18:54 WSPC/165-IJSSD 00264

Improving Stability in the Time-Stepping Analysis of Structural Nonlinear Dynamics 263

Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

Fig. 2. Change of representation of the six-node element.

L4 : length of the segment that combines the midpoints of the segments L1 and L2 ;
α4 : angle between L4 and the initial position;
L5 : length of the segment that combines the midpoints of the segments L2 and L3 ;
α5 : angle between L5 and the initial position;
uG , vG : displacement of the center of mass.
Also in this case, the new variables depend in a nonlinear way on the old ones
as the following relationships show:


 (u2 − u1 ) + L1 sin α1 = 0,



 (v2 − v1 ) − L1 cos α1 + 2hy = 0,




 (u4 − u3 ) + L2 sin α2 = 0,




 (v4 − v3 ) − L2 cos α2 + 2hy = 0,



 (u6 − u5 ) + L3 sin α3 = 0,




 (v 6 − v5 ) − L3 cos α3 + 2hy = 0,



 1
 (u3 + u4 − u1 − u2 ) − L4 cos α4 + hx = 0,



2
1 (3.4)
 (v3 + v4 − v1 − v2 ) − L4 sin α4 = 0,

 2



 1

 (u5 + u6 − u3 − u4 ) − L5 cos α5 + hx = 0,

 2



1

 (v5 + v6 − v3 − v4 ) − L5 sin α5 = 0,



 2



 1

 (u1 + u2 + 4u3 + 4u4 + u5 + u6 ) − uG = 0,

 12




 1 (v1 + v2 + 4v3 + 4v4 + v5 + v6 ) − vG = 0.
12
June 9, 2008 18:54 WSPC/165-IJSSD 00264

264 S. Lopez & K. Russo

One can see by standard calculation that the systems (3.3) and (3.4) are written
as u(t) = u(q(t)) in both cases. Now, in the initial system (2.4), there are terms
that depend on the old variables u(t) only and the contribution of the internal
forces which depends on the new variables only.
In order to resolve the system (2.11), it is necessary to introduce a set Γ of
nuq − nu extra equations obtained by double derivation with respect to the time of
Eqs. (3.3) and (3.4), respectively. The equations in Γ are in the form

h(u, q) = b(u) + a(q) + r = 0. (3.5)

Finally, we proceed to solve the system

Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com

g + M∆u + K∆q = 0,
(3.6)
h + B∆u + A∆q = 0,
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

∂Ni
where the matrix K, defined by Kij = ∂qj , is a stiffness matrix; the matrix B is
∂hi
defined by Bij = and A, defined by Aij = ∂h
∂uj ; ∂qj , is an easily invertible 3 × 3
i

block matrix.
The following iterative scheme for determining ∆u and ∆q is proposed:
∆qk+1 = −A−1 (h + B∆uk+1 ),
(3.7)
(M − KA−1 B)∆uk+1 + g − KA−1 h = 0.
The iteration sequence terminates when the relative errors
g
and
h
are less
than the prefixed tolerance η.

4. Numerical Test
In the numerical tests, we refer to values of the energies T , L and V computed
by the expressions (2.2), (2.12) and (2.1), respectively. Furthermore, let steps be
the number of time steps effected by the integration process so as to analyze the
behavior for t = 0, . . . , T .
We report the mean value of the N wi Newton iterations with the ith step,
step
N wm = N wi /steps, (4.1)
i=1

which are carried out, unless it becomes nonstable (div), in the process in the (u)
and (u, q) representations.

4.1. Motion of an L-shaped block

This problem appears in the work of Betsch and Steinmann.14 The initial mesh
configuration of a two-dimensional L-shaped block is shown in Fig. 3. The spa-
tial discretization relies on 36 displacement-based isoparametric four-node finite
elements.
June 9, 2008 18:54 WSPC/165-IJSSD 00264

Improving Stability in the Time-Stepping Analysis of Structural Nonlinear Dynamics 265

Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com

Fig. 3. Initial mesh configuration of an L-shaped block.

by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

Fig. 4. Time history of external loading.

The block is subjected to spatially fixed external distributed loads p which are
applied according to the load history depicted in Fig. 4. The material constants are
E = 107 and ν = 0.3, and the reference mass density is assumed to be ρ = 1.
In the classical representation (u) the iterative process produces exact defor-
mations if ∆t < 0.05 and becomes nonstable if ∆t ≥ 0.05, while in the (u, q)
representation we obtain exact deformations if the time step ∆t is also slightly
greater than 0.05. Moreover, the value N wm of Newton iterations is less than in
the classical case. The integration process that describes the motion is computed
with different time steps, and Table 1 reports the respective N wm values. Figure 5
reveals the deformations in the (u, q) representation when the time interval [0, 10]
is plotted for ∆t = 0.05. The unstable behavior in the (u) representation, in the
case ∆t = 0.05, is reported in Fig. 6, where near t = 4 s an unphysical chaotic
motion of the system occurs, while in the (u, q) representation the total energy of
the system is conserved, as shown in Fig. 7.

Table 1. L-shaped block: mean value N wm of the

Newton iterations.

∆t 0.01 0.02 0.05 0.1 0.2

(u) 3.000 3.725 div
(u, q) 2.089 2.098 2.105 2.178 2.294
June 9, 2008 18:54 WSPC/165-IJSSD 00264

266 S. Lopez & K. Russo

Fig. 5. L-shaped block: sequence of deformed configurations calculated in the (u, q) representa-
tion and with ∆t = 0.05.
Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com

5000
4000
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

3000
2000
1000
0

0 2 4 6 8

Fig. 6. L-shaped block: time history of the energies for the (u) representation with ∆t = 0.05.

3000

2000

1000

0 2 4 6 8 10
Fig. 7. L-shaped block: time history of the energies for the (u, q) representation with ∆t = 0.05.

4.2. Toss rule

The characteristics of the time integration scheme applied to different state variable
representations will be shown here for the example of the plane movement of a toss
rule (see Ref. 17 for a solution to this dynamic problem).
The spatial discretization relies on 15 six-node finite elements; the geometry
and position of the nodal concentrated loads P and the load function of the rule
are shown in Fig. 8. The material constants are E = 1.236 · 1010 and ν = 0.3, and
the reference mass density is assumed to be ρ = 4.68 · 102 .
In the classical representation (u) the iterative process produces exact defor-
mations if ∆t < 0.0001 and becomes nonstable if ∆t ≥ 0.0001, while in the (u, q)
representation we obtain exact deformations if the time step ∆t is also slightly
June 9, 2008 18:54 WSPC/165-IJSSD 00264

Improving Stability in the Time-Stepping Analysis of Structural Nonlinear Dynamics 267

Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com

Fig. 8. Initial mesh configuration of the toss rule and time history of external loading.
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

greater than 0.0001. Moreover, the value N wm of Newton iterations is less than in
the classical case. The integration process that describes the motion is computed
with different time steps, and Table 2 reports the respective N wm values. Figure 9
reveals the deformations in the (u, q) representation when the time interval [0, 0.1]
is plotted for ∆t = 0.0001. The unstable behavior in the (u) representation, in
the case ∆t = 0.0001, is reported in Fig. 10, where near t = 0.043 s an unphysical

Table 2. Toss rule: mean value N wm of the Newton iterations.

∆t 0.00001 0.00002 0.00005 0.0001 0.0002 0.0005 0.001

(u) 2.013 2.477 3.072 div
(u, q) 2.000 2.051 2.970 3.000 3.826 4.360 5.290

Fig. 9. Toss rule: sequence of deformed configurations calculated in the (u, q) representation and
with ∆t = 0.0001.

400

300

200

100

0 0.01 0.02 0.03 0.04 0.05

Fig. 10. Toss rule: time history of the energies for the (u) representation with ∆t = 0.0001.
June 9, 2008 18:54 WSPC/165-IJSSD 00264

268 S. Lopez & K. Russo

120

80

40

0 0.02 0.04 0.06 0.08 0.1

Fig. 11. Toss rule: time history of the energies for the (u, q) representation with ∆t = 0.0001.
Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

Fig. 12. Toss rule in the (u, q) representation, with ∆t = 0.0001: representation of the deformed
configurations obtained with 9, 15 and 21 elements for some values of the time sequence.
June 9, 2008 18:54 WSPC/165-IJSSD 00264

Improving Stability in the Time-Stepping Analysis of Structural Nonlinear Dynamics 269

chaotic motion of the system occurs, while in the (u, q) representation the total
energy of the system is conserved, as shown in Fig. 11.
In order to show the behavior of the integration scheme as the mesh is refined,
in Fig. 12, the configurations of the model are represented for some values of the
time sequence. In particular, we refer to meshes made up of 9, 15 and 21 elements.
As is shown, the sequences of deformed configurations of the coarser mesh are not
acceptable, while a fast convergence is obtained in the case of the refined meshes.

5. Conclusion
Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com

We have presented an approach based on a change of representation in motion

equations for nonlinear dynamics of two-dimensional bodies. The main character-
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

istic of the proposed method is the change of representation in the internal forces
only. In particular, an improvement in the stability of the time integration scheme
is observed. Arbitrary implicit one-step time integration schemes with Newmark
approximations are suitable for use as basic algorithms for the proposed procedure.
Finite element assembling procedures, moreover, are unchanged, maintaining their
typical local nature.
The good convergence of the internal Newton iteration shows that despite the
increase in the number of equations and the nonsymmetry, the overall computa-
tional effort is similar to that required in the classical representation. The total
energy of a system in the (u, q) representation is constant. While this has been
demonstrated in numerical examples, no proof has yet been found and so further
work is clearly required.

References
1. T. Belytschko and D. F. Schoeberle, On the unconditional stability of an implicit
algorithm for nonlinear structural dynamics, J. Appl. Mech. 42 (1975) 865–869.
2. T. J. R. Hughes, T. K. Caughy and W. K. Liu, Finite-element methods for nonlinear
elastodynamics which conserve energy, J. Appl. Mech. 45 (1978) 366–370.
3. J. C. Simo and N. Tarnow, A new energy and momentum conserving algorithm for
the non-linear dynamics of shells, Int. J. Numer. Meth. Eng. 37 (1994) 2527–2549.
4. D. Kuhl and M. A. Crisfield, Energy-conserving and decaying algorithms in nonlinear
structural dynamics, Int. J. Numer. Meth. Eng. 45 (1999) 569–599.
5. F. Armero and E. Petöcz, Formulation and analysis of conserving algorithms for
frictionless dynamic contact/impact problems, Comput. Meth. Appl. Mech. Eng. 158
(1998) 269–300.
6. M. A. Crisfield, U. Galvanetto and G. Jelenić, Dynamics of 3-D co-rotational beams,
Comput. Mech. 20 (1997) 507–519.
7. D. Kuhl and E. Ramm, Generalized energy–momentum method for non-linear adap-
tive shell dynamics, Comput. Meth. Appl. Mech. Eng. 178 (1999) 343–366.
8. C. Hoff and P. J. Pahl, Development of an implicit method with numerical dissipation
for generalized single step algorithm for structural dynamics, Comput. Meth. Appl.
Mech. Eng. 67 (1988) 367–385.
June 9, 2008 18:54 WSPC/165-IJSSD 00264

270 S. Lopez & K. Russo

9. J. Chung and G. M. Hulbert, A time integration algorithm for structural dynamics

with improved numerical dissipation: The generalized-α method, J. Appl. Mech. 60
(1993) 371–375.
10. M. A. Crisfield, Non-linear Finite Element Analysis of Solids and Structures:
Advanced Topics, Vol. 2 (John Wiley & Sons, Chichester, 2003).
11. O. C. Zienkiewicz, W. L. Wood, N. W. Hine and R. L. Taylor, A unified set of single
step algorithms. Part I: General formulation and applications, Int. J. Numer. Meth.
Eng. 20 (1984) 1529–1552.
12. P. Lasaint and P. A. Raviart, On a finite element method for solving the neutron
transport equation, in Mathemetical Aspects of Finite Elements in Partial Differential
Equations, ed. C. de Boor (Academic Press, New York, 1974), pp. 89–123.
13. P. Betsch and P. Steinmann, Conservation properties of a time FE method. Part I:
Int. J. Str. Stab. Dyn. 2008.08:257-270. Downloaded from www.worldscientific.com

Time-stepping schemes for N -body problems, Int. J. Numer. Meth. Eng. 49 (2000)
599–638.
14. P. Betsch and P. Steinmann, Conservation properties of a time FE method, Part II:
by UNIVERSITY OF IOWA on 03/13/15. For personal use only.

Time-stepping schemes for non-linear elastodynamics, Int. J. Numer. Meth. Eng. 50

(2001) 1931–1955.
15. J. Argyris, M. Papadrakakis and Z. S. Mauroutis, Nonlinear dynamic analysis of shells
with the triangular element TRIC, Comput. Meth. Appl. Mech. Eng. 192 (2003) 3005–
3038.
16. S. Lopez, Improving stability by change of representation in time-stepping analysis of
non-linear beams dynamics, Int. J. Numer. Meth. Eng. 69 (2007) 822–836.
17. D. Kuhl and E. Ramm, Constraint energy momentum algorithm and its application to
non-linear dynamics of shells, Comput. Meth. Appl. Mech. Eng. 136 (1996) 293–315.