Studies on composite implicit time

integration schemes for transient response

analysis of a structural model
Cite as: AIP Conference Proceedings 2336, 020011 (2021); https://doi.org/10.1063/5.0046834
Published Online: 26 March 2021

P. Deepak, and Satyananda Panda

ARTICLES YOU MAY BE INTERESTED IN

Analytical study of time-fractional porous medium equation using homotopy analysis method
AIP Conference Proceedings 2336, 020004 (2021); https://doi.org/10.1063/5.0046222

A delayed version of prey predator system with modified Holling-Tanner response

AIP Conference Proceedings 2336, 020005 (2021); https://doi.org/10.1063/5.0046306

A review on spectral element solver Nek5000

AIP Conference Proceedings 2336, 030001 (2021); https://doi.org/10.1063/5.0045709

AIP Conference Proceedings 2336, 020011 (2021); https://doi.org/10.1063/5.0046834 2336, 020011

© 2021 Author(s).
 6WXGLHVRQ&RPSRVLWH,PSOLFLW7LPH,QWHJUDWLRQ6FKHPHV
IRU7UDQVLHQW5HVSRQVH$QDO\VLVRID6WUXFWXUDO0RGHO
P
Deepaka)
and
Satyananda
Pandab)
Department
of
Mathematics
National
Institute
of
Technology,
Calicut,
Kerala,
India
–
673
601.
a) Corresponding author: deepak_p140109ma@nitc.ac.in
b) Electronicmail: satyanand@nitc.ac.in.
Abstract. We investigated composite implicit time integration schemes for structural applications. An unconditionally stable,
two-sub-step per time step, the Bathe scheme is analyzed. The transient response analysis of a one-dimensional linear model
with step excitation is presented. In the conventional Bathe method, the first sub-step is based on the Newmark scheme with
the trapezoidal rule with parameter control. The second sub-step employs a three-point Euler backward method. The modified
Bathe method uses a three-point trapezoidal rule with two new Newmark parameters β1 and β2 for the second sub-step. In this
contribution, we used both methods with a variable time splitting parameter to analyse numerical errors introduced during the time
integration process. The response of a clamped-free rod with excitation force applied at free-end is presented with various time
splitting ratios. The results are compared with the conventional Newmark method and suggest that the Bathe scheme with equal
time sub-step is suitable for linear structural dynamic applications.

Keywords: Composite time integration, Newmark parameters, Time splitting ratio

INTRODUCTION

In the past several decades, efficiency of time integration schemes for structural dynamic applications are the topic
of investigation for many researchers. The two primary types of direct time-integration methods that are thoroughly
discussed are explicit [1, 2, 3] and implicit methods [4, 5, 6]. Structural dynamic applications involve complex numer-
ical models with multiple Degrees of freedom (Dof) seeking linear and nonlinear solutions. Dynamic analysis, extract
eigenvalues and eigenvectors of these models, which represent frequencies and mode shapes of the structure. Wave
propagation problems require all the frequencies of the finite element model, where explicit methods are suitable. The
response of structural vibration applications depends on the extraction of the first few frequencies of the dominant
modes, hence focusing on implicit schemes.
The Newmark scheme [4] with varying parameters, α, and β for making it implicit are the most popular method
employed in the Finite Element packages. The scheme becomes unconditionally stable, implicit trapezoidal rule with
α = 0.5 and β = 0.25. For α > 0.5 and β > 0.5α introduces numerical dissipation into the system, which also
damps out lower frequencies. Bathe and Wilson [7] have done one of the earliest works on the effect of time step
size on the spectral radius. Subsequently, time integration with controllable parameters for numerical dissipation
was developed. Hilber et al. [8] introduced an unconditionally stable and low-frequency preserving, implicit method
popularly known as HHT-α. Wood et al. [9] developed a second-order accurate unconditionally stable method known
as WBZ-α scheme.
Chung and Hulbert [10] combined the Newmark method, HHT-α and WBZ-α schemes to introduce a Generalized-
α method. Bathe proposed a composite scheme [11, 12] to combine a highly dissipative numerical scheme with a
non-dissipative time integration method. The method consists of two equal sub-steps per time step with numerical
integration schemes, trapezoidal rule, and three-point backward Euler method. Silva and Bezerra [13] proposed
generalized sub-step size in the Bathe scheme with a time splitting parameter, γ, and Newmark values (α,β ) for
generalizing trapezoidal rule in the first sub-step.
Bathe introduced a modification [14] by changing the three-point backward Euler scheme with a three-point trape-
zoidal rule introducing a new set of Newmark parameters, β1 , β2 . In this paper, our objective is to discuss the numerical
errors of standard Bathe and Bathe-β1 /β2 composite time integration schemes for various time splitting parameters.
We studied the effect of Newmark parameters on the Stability, Amplitude Decay (AD), and Period Elongation (PE)
integration scheme. We also analysed the transient response of a Finite Element structural model and also discussed
the limitation of the Newmark method.
International Conference on Computational Sciences-Modelling, Computing and Soft Computing (CSMCS 2020)
AIP Conf. Proc. 2336, 020011-1–020011-11; https://doi.org/10.1063/5.0046834
Published by AIP Publishing. 978-0-7354-4080-7/$30.00

020011-1
 COMPOSITE DIRECT TIME INTEGRATION METHOD

The equilibrium equation of motion for a linear transient response problems for structural analysis is expressed as:

M ü(t) +Cu̇(t) + Ku(t) = F(t), (1)

where M is the mass matrix, K represents stiffness matrix, C stands for the damping matrix, and F(t) is the external
force vector at time t. The displacement at time t is represented by u(t) and corresponding velocity and acceleration
vectors are represented by the first and second order differentiation, u̇(t), and ü(t) respectively. In direct time integra-
tion method, Equation (1) is satisfied only at discrete time intervals, Δt. The solution vectors, ut , u̇t and üt are found
at n + 1 discrete time instants, t = { t0 , t0 + Δt, t0 + 2Δt, ...t0 + nΔt}.
Equation (1) being a second order differential equation in time t, the two initial conditions for the discrete solution
are:

u0 = u(0); and u̇0 = u̇(0), (2)

where u0 and u̇0 are vectors of initial displacement and velocity respectively. Direct time integration scheme proceeds
by numerical time-stepping procedure to compute the structural dynamic response. The solution progressively calcu-
lates time-varying displacement, velocity, and acceleration at each time step. Accuracy and stability of the selected
numerical method depends on the displacement, velocity, and acceleration assumptions within the time interval.
In composite time integration schemes, the time interval, Δt used in a single step is divided into two sub-intervals
γΔt and (1 − γ)Δt, where γ ∈ (0, 1) as shown in Figure 1.

FIGURE
1.
Generalized
composite
time
integration
sub-step
sizes

Generalized-Bathe
0ethod

In
the
first
sub-step
the
assumptions
of
velocity,
u̇t+γΔt
and
displacement
ut+γΔt
at
t
+
γΔt
follows
trapezoidal
rule,

which
is
a
specific
case
of
the
Newmark
method
(α
=
0.5,
β
=
0.25),
i.e.,
1
u̇t+γΔt = u̇t + (üt + üt+γΔt )γΔt (3)
2

1
ut+γΔt = ut + (u̇t + u̇t+γΔt )γΔt, (4)
2
where ut , u̇t and üt as shown in Figure 1 are the approximate values of displacement vector, velocity vector and
acceleration vector at time t. Here, üt+γΔt is the approximate acceleration at time t + γΔt.
Using dynamic equilibrium Equation (1) at time t + γΔt, we have

M üt+γΔt +Cu̇t+γΔt + Kut+γΔt = F t+γΔt , (5)

020011-2
where F t+γΔt is the external force vector at time t + γΔt.
Combining the relationship between unknown vectors, ut+γΔt , u̇t+γΔt and üt+γΔt in Equation (3) and (4) and substi-
tuting it in dynamic equilibrium Equation 5, we get the solution at time t + γΔt. Velocity vector, u̇t+Δt and acceleration
vector, üt+Δt at time t + Δt uses a three-point backward Euler method. This approximation is applied over the sub-
interval (1 − γ)Δt, which introduces some algorithmic damping and shrink spectral radius inside unit circle.
The approximation of the velocity and acceleration vectors at time t + Δt using three-point backward Euler scheme
are

u̇t+Δt = a1 ut + a2 ut+γΔt + a3 ut+Δt (6)

üt+Δt = a1 u̇t + a2 u̇t+γΔt + a3 u̇t+Δt , (7)

where the constants are given as

1−γ −1 2−γ
a1 = , a2 = , and a3 = .
γΔt (1 − γ)γΔt (1 − γ)Δt

Again, substituting the velocity vector given in Equation (6) in the expression for acceleration (Equation (7)), we
obtain

üt+Δt = a1 u̇t + a2 u̇t+γΔt + a3 a1 ut + a3 a2 ut+γΔt + a3 a3 ut+Δt (8)

Now, the equilibrium Equation (1) can be written at a time t + Δt as

M üt+Δt +Cu̇t+Δt + Kut+Δt = F t+Δt . (9)

Combining Equations (6)), (7), and (8) and substituting the result in Equations(1), the three unknown vectors ut+Δt ,
u̇t+Δtand üt+Δt can be solved at time t + Δt. Since, ut , u̇t and üt are required only for the time step Δt, it remains as a
single-step method. The controlling parameter γ is assumed as 0.5 for the standard Bathe composite scheme, which
makes the sub-steps following trapezoidal and three-point backward Euler schemes.
The increased computational and memory requirement with the standard Bathe scheme is due to the two stiffness
matrix factorization matrix in a single time step. Setting the splitting parameter, γ = 0.568 results in the similar
stiffness matrices from both sub-steps and results only in one factorization [14] for linear analysis. The recursive
numerical solution at nth time step of a single Degree of freedom system is written as:
⎡ ⎤ ⎡ t⎤
üt+nΔt ü
⎢ ⎥
X nΔt = ⎣u̇t+nΔt ⎦ = An ⎣u̇t ⎦ + La F t+γΔt + Lb F t+Δt , (10)
ut+nΔt ut

where A is the amplification matrix and La and Lb represent load operators, at nth time step t + nγΔt and t + nΔt
respectively.
The amplification matrix can be derived from Equation (5) and (9). Spectral radius, ρ(A) of the amplification
matrix characterizes the time integration scheme’s stability and accuracy by neglecting load vector components. The
characteristic polynomial for the amplification matrix is defined by its invariant, d1 = (1/2)trace(A), d2 = ∑ principle
minors(A) and d3 = determinant(A).

p(λ ) = λ 3 − 2d1 λ 2 + d2 λ − d3 (11)

where λ is the eigenvalue of A. If the matrix A can be diagonalized, then the general solution of Equation (10) can be
written as

X nΔt = a1 λ1n + a2 λ2n + a3 λ3n , (12)

where a1 , a2 and a3 are constants depend on initial conditions. The complex conjugate eigenvalues λ1,2 are the
principal roots of A and the spurious root, λ3 satisfies the relation |λ3 | < |λ1,2 |. The numerical stability is defined
for a given time period T , where T = 2π/ω, and step size Δt/T . The computed response at time instant t + nΔt

020011-3
 1.5
Houbolt method
Wilson−θ method,θ=1.4
Newmark method, α=1/2,β=1/2
Newmark method, α=1/2,β=1/4
Newmark method, α=11/20,β=3/10
Spectral radius, ρ(A) Bathe method(γ=1/2)

0.5

0
−2 −1 0 1 2 3
10 10 10 10 10 10
Time ratio (Δt/T)

FIGURE 2. Spectral radius, ρ(A) for an undamped system for various time integration schemes

is assumed not to amplified artificially in Equation (10). Accordingly, we have considered, A ≤ 1, or, using the
spectral radius of the matrix A, we have, ρ(A) = max| λi | ≤ 1, for i = 1, 2, 3.
The behaviour of the spectral radius, ρ(A) as a function of Δt/T , shows the errors associated with various numeri-
cal integration schemes. Figure. 1 compares the ρ(A) of the Bathe schemes for a time splitting parameters, γ = 1/2
with Houbolt method [15], Wilson-θ method [7], and with other popular Newmark integration schemes. It shows the
high degree of numerical damping for larger values of Δt/T for the Bathe method, a desirable property for structural
dynamic response. The technique ensures high accuracy and unconditional stability up to Δt/T value equal to 0.1, and
for Δt/T = 0.3, the ρ(A) decreases rapidly. In the Houbolt method, which is not a self-starting scheme, the numer-
ical dissipation appears for a minimum value of Δt/T , which is not a desirable in computation efficiency point of view.

The complex conjugate roots of Equation 12 are λ1 = a + i b; λ2 = a − i b; b ≥ 0. The discrete solution,

similar to exact solution given in [16] and parameters are comparable, i.e.,

X nΔt = e−ξ̄ Ω c1 cos(Ω̄D n) + c2 sin(Ω̄D n) + c3 λ3n (13)

where Ω = ωΔt, and other parameters in Equation 13 are defined as:

Ω̄D = tan−1 (b/a) (14)

1
ξ̄ = − ln(a2 + b2 ) (15)
2Ω

T̄D − T Ω
P= = −1 (16)
T Ω̄D

T̄D = 2π/ω̄D ; T = 2π/ω (17)

ω̄D = Ω̄D /Δt, (18)

where ξ̄ is the algorithmic damping, and P denote period elongation. These parameters measure the dispersion and
dissipation of the numerical method.

020011-4
 (a) Percentage period elongations (b) Percentage amplitude decay
20 20
Newmark method, α=1/2,β=1/2 Newmark method, α=1/2,β=1/2

Percentage Period Elongation,(PE x 100%)

Percentage Amplitude Decay,(AD x 100%)

18 Newmark method, α=1/2,β=1/4 18 Newmark method, α=1/2,β=1/4
Newmark method, α=11/20,β=3/10 Newmark method, α=11/20,β=3/10
16 16
Bathe method(γ=1/2) Bathe method(γ=1/2)
14 14

12 12

10 10

8 8

6 6

4 4

2 2

0 0
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2
Δt/T Δt/T

FIGURE
3.
Period
Elongation(PE)
and
Amplitude
Decay(AD)
errors
of
Bathe
and
Newmark
schemes

Figure
3
shows
the
comparison
of
numerical
errors
in
the
Bathe
method
with
various
Newmark
schemes.
The

trapezoidal
rule
(α
=
1/2, β
=
1/4)
does
not
show
amplitude
decay
and
display
the
minimal
period
of
elongation.

The
numerical
errors
in
the
Bathe
method
are
acceptable
for
a
reasonably
small
time
step
interval.
For
a
step
size
of

Δt/T
=
0.1,
the
Bathe
method
results
in
the
same
amplitude
decay
within
10-time
steps
for
which
trapezoidal
rule

takes
14-time
steps.
This
increases
by
around
40%
computational
time
for
the
Bathe
method
for
solving
the
same

problem
with
a
reduced
period
elongation.

β1/β2-Bathe
6cheme

This
method
is
a
modification
of
the
standard
Bathe
scheme
and
retains
unconditional
stability.
The
first
sub-step

is
the
same
as
the
Bathe
method,
whereas
the
second
sub-step
follows
a
three-point
trapezoidal
rule
with
Newmark

parameters
β1
and
β2 .

u̇t+Δt = u̇t + (γΔt) (1 − β1 )üt + β1 üt+γΔt + ((1 − γ)Δt) (1 − β2 )üt+γΔt + β2 üt+Δt (19)

ut+Δt = ut + (γΔt) (1 − β1 )u̇t + β1 u̇t+γΔt + ((1 − γ)Δt) (1 − β2 )u̇t+γΔt + β2 u̇t+Δt (20)

We get the trapezoidal rule with the Newmark parameters, β1 = β2 = 1/2 in Equations (19), and (20). It can also
be shown that for β1 = 1/3, β2 = 2/3 and γ = 0.5, we arrive at the standard Bathe method. The vectors üt+Δt and
u̇t+Δt are given as:

1 
üt+Δt = u̇t+Δt − u̇t − ((γΔt)(1 − β1 ))üt − ((γΔt)β1 + (1 − γ)Δt(1 − β2 ))üt+γΔt (21)
(1 − γ)(Δt)β2

1 
u̇t+Δt = ut+Δt − ut − ((γΔt)(1 − β1 ))u̇t − ((γΔt)β1 + (1 − γ)Δt(1 − β2 ))u̇t+γΔt (22)
(1 − γ)(Δt)β2

Substituting Equation (22) in Equation (21), we have

1 γ(1 − β1 ) + β2 (1 − γ) t
üt+Δt = (ut+Δt − ut ) − u̇
((1 − γ)(Δt)β2 ) 2 (β2 (1 − γ))2 Δt
(23)
γβ1 + (1 − β2 )(1 − γ) t+γΔt γ(1 − β1 ) t γβ1 + (1 − β2 )(1 − γ) t+γΔt
− u̇ − ü − ü
(β2 (1 − γ))2 Δt β2 (1 − γ) β2 (1 − γ)

020011-5
Rewriting Equation 22, we have

1 γ(1 − β1 ) t γβ1 + (1 − β2 )(1 − γ) t+γΔt

u̇t+Δt = (ut+Δt − ut ) − u̇ − u̇ (24)
β2 (1 − γ)Δt β2 (1 − γ) β2 (1 − γ)

Substituting, üt+Δt and u̇t+Δt in the Equation (9) at t + Δt, we obtain the equilibrium equation in terms of ut+Δt . The
solutions are obtained by solving the governing finite element equations at the distinct time points.
The standard Bathe method is obtained by using the values β1 = 1 + 2γ(γ−2) 1
, and β2 = (2−γ)
1
for any values of
2β2
γΔt. Using the γ = 1+2β2 for any β1 , the stiffness matrices are same for both sub-steps, and only one factorization is
needed.

1.5
Spectral radius, ρ(A)

β =0.50, β =0.50
1 2
0.5 β =0.47, β =0.53
1 2
β =0.44, β =0.56
1 2
β1 =0.41, β2 =0.59
β1 =0.37, β2 =0.63
β =0.33, β =0.67
1 2
0
10-2 10-1 100 101 102 103
Time ratio ( Δt/T)

FIGURE 4. Spectral radius, ρ(A) for an undamped system for various β1 /β2 values

(a) Percentage period elongations (b) Percentage amplitude decay

20 20
β =0.50, β =0.50 β =0.50, β =0.50
Percentage Period Elongation,(PE x 100%)

1 2 1 2
Percentage Amplitude Decay,(AD x 100%)

18 β =0.47, β =0.53 18 β =0.47, β =0.53

1 2 1 2
β =0.44, β =0.56 β =0.44, β =0.56
1 2 1 2
16 16
β =0.41, β =0.59 β =0.41, β =0.59
1 2 1 2
14 β =0.37, β =0.63 14 β =0.37, β =0.63
1 2 1 2
β1 =0.33, β2 =0.67 β1 =0.33, β2 =0.67
12 12

10 10

8 8

6 6

4 4

2 2

0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Time ratio (Δt/T) Time ratio (Δt/T)

FIGURE 5. Period Elongation (PE) and Amplitude Decay (AD) for various β1 /β2 values

Figure. 4 shows the smooth variation of spectral radius for the Bathe-β1 /β2 scheme for various β1 and β2 values.
As β1 values reduces, the numerical damping is increased and the method reduces to the standard Bathe method,
i.e., when β1 = 0.33 and β2 = 0.67. Figure 5 gives the Amplitude Decay and Period Elongation variations for time
step ratios Δt/T . As shown in Figure 4, for a small increase of β1 from 0.33 to 0.37 with β2 = 1 − β1 , the results
in ρ(A) = 0 for large values of Δt/T . The numerical damping is reduced, and higher (spurious) modes will not be
eliminated as rapidly.

020011-6
 TIME SPLITTING RATIO OF COMPOSITE DIRECT INTEGRATION

Standard Bathe and modified Bathe-β1 /β2 method by default uses an equal time interval for every sub-steps. For
computational efficiency and to facilitate single factorization in each time step, Newmark parameters are varied. In
this work, we are focusing on time splitting parameter γ to analyze numerical accuracy of the composite method.

1.2
Standard Bathe method
Split ratioγ=0.2
1 Split ratioγ=0.4
Split ratioγ=0.6
Spectral radius, ρ(A)

Split ratioγ=0.8
0.8

0.6

0.4

0.2

0
10-2 10-1 100 101 102 103
Time ratio ( Δt/T)

FIGURE 6. Spectral radius, ρ(A) of general Bathe method for various γ values.

(a) Percentage period elongations (b) Percentage amplitude decay

20 20
Standard Bathe method Standard Bathe method
Percentage Period Elongation,(PE x 100%)

Percentage Amplitude Decay,(AD x 100%)

18 Split ratioγ=0.2 18 Split ratioγ=0.2

Split ratioγ=0.4 Split ratioγ=0.4
16 Split ratioγ=0.6 16 Split ratioγ=0.6
Split ratioγ=0.8 Split ratioγ=0.8
14 14

12 12

10 10

8 8

6 6

4 4

2 2

0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Time ratio (Δt/T) Time ratio (Δt/T)

FIGURE 7. Period Elongation(PE) and Amplitude Decay(AD) of general Bathe method for various γ values.

Figure 6 and 7 show the spectral radius and numerical error estimates of general Bathe method for various time
splitting ratios, γ = 0.2, 0.4, 0.6 and 0.8. The Period Elongation(PE) is minimum, when γ = 0.6, which ensures max-
imum Amplitude Decay(AD). Time splitting parameter, γ = 0.2 shows the maximum error and minimum algorithmic
damping.
The Bathe-β1 /β2 method with Newmark parameters, β1 = 0.33 and β2 = 1 − β1 is studied for time splitting ratio,
γ = 0.1, 0.2, 0.3 and 0.4. Figure 8 and 9 show the variation in spectral radius, period elongation and amplitude decay.
The minimum error in period elongation is achieved for the value γ = 0.4. But damping is high for the Bathe-β1 /β2
method with β1 = 0.33, and β2 = 1 − β1 .

020011-7
 1.5
Bathe−β1/β2
Split ratio,γ=0.1
Split ratio,γ=0.2
Split ratio,γ=0.3
Split ratio,γ=0.4

Spectral radius, ρ(A)

1

0.5

0
−2 −1 0 1 2
10 10 10 10 10
Time ratio (Δt/T)

FIGURE 8. Spectral radius, ρ(A) of Bathe-β1 /β2 method for various γ values.

(a) Percentage period elongations (b) Percentage amplitude decay

20 20
Bathe-β /β
Percentage Period Elongation,(PE x 100%)

Percentage Amplitude Decay,(AD x 100%)

1 2
18 Split ratio,γ=0.1 18
Split ratio,γ=0.2
16 Split ratio,γ=0.3 16
Split ratio,γ=0.4
14 14
Bathe-β /β
1 2

12 12 Split ratio,γ=0.1
Split ratio,γ=0.2
10 10 Split ratio,γ=0.3
Split ratio,γ=0.4

8 8

6 6

4 4

2 2

0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Time ratio (Δt/T) Time ratio (Δt/T)

FIGURE 9. Period Elongation(PE) and Amplitude Decay(AD) of Bathe-β1 /β2 method for various γ values.

TRANSIENT RESPONSE OF A 1D-STRUCTURAL MODEL

We consider a clamped bar excited by a step load, as shown in Figure 10. The bar is modelled with N finite elements
of equal length l = L/N. The global stiffness matrix K, and mass matrix M of the model are given by
⎡ ⎤ ⎡ ⎤
2 −1 2
⎢−1 2 −1 ⎥ ⎢ 2 ⎥
⎢ ⎥ ⎢ ⎥
EA ⎢ ⎢ −1 2 −1 ⎥ ρAl ⎢ ⎢ 2 ⎥
K= =⎢ ⎥ , M= =⎢ ⎥ (25)
l . . . ⎥ 2 . ⎥
⎢ ⎥ ⎢ ⎥
⎣ . . −1 ⎦ ⎣ . ⎦
−1 2 2
N×N N×N

The material and physical properties assumed for the model are the modulus of elasticity, E = 30MPa, density ρ =
0.00073Kg/m3 , cross sectional area,A = 1m2 , and the length of the bar, L = 200m. Model is assumed to be undamped
and applied step load at the free end is F(t) = 10, 000N. The number of finite elements used are, N = 1000.

020011-8
 FIGURE 10. A clamped–free bar excited by a step load.

The predicted response using the Newmark method, standard Bathe method and β1 /β2 -Bathe method is shown in
Figure 11- 13, respectively. The γ value is assumed to be 0.5, and the time step size Δt depends on the CFL value.
Here, the time step is calculated as, Δt = (9.88×10−7 )×CFL. The response is sought at node 700(x = 140) for
displacement, velocity, and acceleration. The response shows that displacement is predicted almost accurately by all
the three-time integration schemes. But, the velocity and acceleration of the Newmark scheme show some oscillations.
It may be concluded that the Bathe methods are accurate and do not introduces over-shoot in velocity and acceleration
compared to the trapezoidal rule.

0.1
Displacement(m)

NewMark( =1/2, =1/4)

Analytical

0.05

0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Time
100
NewMark( =1/2, =1/4)
Velocity(m/s)

50 Analytical

-50

-100
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Time

FIGURE 11. Displacement and Velocity of node 700 using Newmark method(δ = 1/2, α = 1/4).

CONCLUSION

The composite time integration scheme is studied with the objective of numerical error analysis. Composite direct
integration splits each time step into two sub-steps with a time splitting parameter, γ. Standard Bathe method works
with three control parameters with α = 1/2, β = 1/4 with the first sub-step to have the trapezoidal rule. The time
splitting ratio, γ = 0.5, is used to split the sub-steps into equal sub-intervals. The second sub-step uses a 3-point
backward Euler method.
The β1 /β2 -Bathe method uses trapezoidal method for the first sub-step and a 3-point Newmark scheme in the
second sub-step and introduces two extra parameters, β1 and β2 . With β1 = 0.33 and β2 = 0.67, the method is
reduced to the standard Bathe method. The parameter values, β1 = 0.5, and β2 = 05, are used for the trapezoidal rule
for the second sub-step.
The time splitting parameter, γ, also decides the numerical errors introduced in each time step. We studied this factor

020011-9
 0.1

Displacement(m)
Bathe- =0.50, CFL=1.0
Analytical

0.05

0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Time
100
Bathe- =0.50, CFL=1.0
Velocity(m/s)

50 Analytical

-50

-100
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Time

FIGURE 12. Displacement and Velocity of node 700 using Standard Bathe method(γ = 0.50).

0.1
Displacement(m)

Bathe- =0.50, CFL=1.0

Analytical

0.05

0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Time
100
Bathe- =0.50, CFL=1.0
Velocity(m/s)

50 Analytical

-50

-100
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Time

FIGURE 13. Displacement and Velocity of node 700 using β1 /β2 -Bathe method(β1 = 0.33, β2 = 1 − β1 ).

concerning the Bathe composite integration scheme. The errors are presented for various γ values. The displacement
and velocity response of a clamped-free bar with step load excitation is studied using 1000 Finite elements and results
compared with the Newmark method.
Standard Bathe method with variable time splitting ratio, γ = 0.5 and γ = 0.6 enforce high frequency dissipation.
The appropriate values of γ to be used in Finite Element packages lies in this range. The effect of varying γ on
Bathe-β1 /β2 method results in reduced numerical damping and hence not recommended for structural applications.

REFERENCES
1. R. Mullen and T. Belytschko, “An analysis of an unconditionally stable explicit method.” Computers and Structures. 16, 691–696 (1983).
2. M. A. Dokainish and K. Subbaraj, “A survey of direct time-integration methods in computational structural dynamics-I: explicit methods.”
Computers and Structures. 32, 1371–1386 (1989).
3. J. Chung and J. M. Lee, “A new family of explicit time integration methods for linear and non-linear structural dynamics.” International Journal
for Numerical Methods in Engineering. 37, 3961–3976 (1994).
4. N. M. Newmark, “A method of computation for structural dynamics.” Journal of the Engineering Mechanics Division: Proc. ASCE. 83, 67–94
(1959).
5. K. Subbaraj and M. A. Dokainish, “A survey of direct time-integration methods in computational structural dynamics-II: implicit methods.”
Computers and Structures. 32, 1387–1401 (1989).
6. E. Rougier, A. Munjiza, and N. W. M. John, “Insight into an implicit time integration scheme for structural dyanmics.” International Journal
for Numerical Methods in Engineering. 61(6), 856–879 (2004).
7. E. L. Wilson, I. Farhoomand, and K. J. Bathe, “Nonlinear dynamic analysis of complex structures,” International Journal of Earthquake
Engineering and Structural Dynamics. 1, 241–252 (1973).
8. H. Hilber, T. Hughes, and R. Taylor, “Improved numerical dissipation for time integration algorithms in structural dynamics.” Earthquake
Engineering and Structural Dynamics. 5, 283–292 (1977).
9. W. Wood, M. Bossak, and O. Zienkiewicz, “An alpha modification of newmark’s method.” International Journal for Numerical Methods in
Engineering. 5, 1562–1566 (1980).

020011-10
10. J. Chung and G. M. Hulbert, “A time integration algorithm for structural dynamics with improved numerical dissipation: the generalized-α
method,” Journal of Applied Mechanics. 60, 371–375 (1993).
11. K. Bathe and M. M. I. Baig, “On a composite implicit time integration procedure for nonlinear dynamics.” Computers and Structures. 83,
2513–2524 (2005).
12. K. J. Bathe, “Conserving energy and momentum in nonlinear dynamics: a simple implicit time integration scheme.” Computers and Structures.
85(7-8), 437–445 (2007).
13. W. T. M. Silva and L. M. Bezerra, “Performance of composite implicit time integration scheme for nonlinear dynamic analysis.” Mathematical
Problems in Engineering. 2008 (2008).
14. K. J. Bathe and G. Noh, “Insight into an implicit time integration scheme for structural dyanmics.” Computers and Structures 98/99, 1–6
(2012).
15. K. J. Bathe and E. L. Wilson, “Stability and accuracy analysis of direct integration methods,” International Journal of Earthquake Engineering
and Structural Dynamics. 1, 283–291 (1973).
16. P. Paultre., Dynamics of Structures. (Wiley., 2011).
17. H. Hilber, T. Hughes, and R. Taylor, “Collocation, dissipation and overshoot for time integration schemes in structural dynamics.” Earthquake
Engineering and Structural Dynamics. 6, 99–117 (1978).
18. Y. M. Xie, “An assessment of time integration schemes for non-linear equations.” Journal of Sound and Vibration. 192, 321–331 (1996).
19. G. M. Hulbert and J. Chung, “Explicit time integration algorithms for structural dynamics with optimal numerical dissipation.” Computer
Methods in Applied Mechanics and Engineering. 137, 175–188 (1996).
20. T. Fung, “Numerical dissipation in time-step integration algorithms for structural dynamic analysis.” Progress in Structural Engineering and
Materials. 5.
21. R. D. Cook, D. S. Malkus, M. E. Plesha, and R. J. Witt., Concepts And Applications of Finite Element Analysis. (Wiley., 2003).
22. S. Y. Chang and W. I. Liao, “An unconditionally stable explicit method for structural dynamics.” Journal of Earthquake Engineering. 9(3),
349–370 (2005).
23. S. Krenk and J. R. Hogsberg, “Properties of time integration with first order filter damping.” Int. J. Numerical Methods Eng. 64, 547–566
(2005).
24. S. Krenk, “State-space time integration with energy control and fourth-order accuracy for linear dynamic systems.” Int. J. Numerical Methods
Eng. 65, 595–619 (2006).
25. S. Krenk, “Extended state-space time integration with high-frequency energy dissipation.” Int. J. Numerical Methods Eng. 73, 1767–1787
(2008).
26. T. Liu, C. Zhao, Q. Li, and L. Zhang, “An efficient backward euler time-integration method for nonlinear dynamic analysis of structures,”
Computers and Structures. 106-107, 20–28 (2012).
27. G. Noh, S. Ham, and K. J. Bathe, “Performance of an implicit time integration scheme in the analysis of wave propagations.” Computers and
Structures. 123, 93–105 (2013).
28. G. Noh and K. J. Bathe, “An explicit time integration scheme for the analysis of wave propagations.” Computers and Structures. 129, 178–193
(2013).
29. S. Rostami, S. Shojaee, and A. Moeinadini, “A parabolic acceleration time integration method for structural dynamics using quartic b-spline
functions.” Applied Mathematical Modelling. 36(11), 5162–5182 (2012).
30. W. B. Wen, K. L. Jian, and S. M. Luo, “An explicit time integration method for structural dynamics using septuple b-spline functions.”
International Journal for Numerical Methods in Engineering. 97(9), 629–657 (2014).
31. W. B. Wen, K. L. Jian, and S. Luo, “2d numerical manifold method based on quartic uniform b-spline interpolation and its application in thin
plate bending.” Applied Mathematics and Mechanics (English Edition). 34, 1017–1030 (2013).
32. W. B. Wen, S. Luo, and K. L. Jian, “A novel time integration method for structural dynamics utilizing uniform quintic b-spline functions.”
Archive of Applied Mechanics. 85, 1743–1759 (2015).

020011-11