DYNAMIC ANALYSIS BY

NUMERICAL INTEGRATION
Normally, For Earthquake Loading
Direct Numerical Integration Is Very Slow

20.1 INT RODUCT ION

The most general approach for the solution of the dynamic response of structural
systems is the direct numerical integration of the dynamic equilibrium equations.
This involves, after the solution is defined at time zero, the attempt to satisfy
dynamic equilibrium at discrete points in time. Most methods use equal time
intervals at ∆t , 2 ∆t , 3∆t ........ N∆t . Many different numerical techniques
have previously been presented; however, all approaches can fundamentally be
classified as either explicit or implicit integration methods.

Explicit methods do not involve the solution of a set of linear equations at each
step. Basically, these methods use the differential equation at time “ t ” to predict a
solution at time “ t + ∆t ”. For most real structures, which contain stiff elements, a
very small time step is required in order to obtain a stable solution. Therefore, all
explicit methods are conditionally stable with respect to the size of the time step.

Implicit methods attempt to satisfy the differential equation at time “ t ” after the
solution at time “ t − ∆t ” is found. These methods require the solution of a set of
linear equations at each time step; however, larger time steps may be used.
Implicit methods can be conditionally or unconditionally stable.
2 STATIC AND DYNAMIC ANALYSIS

There exist a large number of accurate, higher-order, multi-step methods that have
been developed for the numerical solution of differential equations. These multi-
step methods assume that the solution is a smooth function in which the higher
derivatives are continuous. The exact solution of many nonlinear structures
requires that the accelerations, the second derivative of the displacements, are not
smooth functions. This discontinuity of the acceleration is caused by the nonlinear
hysteresis of most structural materials, contact between parts of the structure, and
buckling of elements. Therefore, only single-step methods will be presented in this
chapter. Based on a significant amount of experience, it is the conclusion of the
author that only single-step, implicit, unconditional stable methods be used for the
step-by-step seismic analysis of practical structures.

20.2 NEWMARK F AMILY OF MET HODS

In 1959 Newmark [1] presented a family of single-step integration methods for the
solution of structural dynamic problems for both blast and seismic loading. During
the past 40 years Newmark’s method has been applied to the dynamic analysis of
many practical engineering structures. In addition, it has been modified and
improved by many other researchers. In order to illustrate the use of this family of
numerical integration methods consider the solution of the linear dynamic
equilibrium equations written in the following form:

Mu
&& t + Cu
& t + Ku t = Ft (20.1)

The direct use of Taylor’s series provides a rigorous approach to obtain the
following two additional equations:

∆t 2 ∆t 3
u t = u t-∆t + ∆t u& t-∆t + && t- ∆t +
u &&t- ∆t +......
&u (20.2a)
2 6

∆t 2
u& t = u& t-∆t + ∆t u
&& t- ∆t + &&t- ∆t + ......
&u (20.2b)
2
DIRECT INTEGRATION METHODS 3

Newmark truncated these equations and expressed them in the following form:

∆t 2
u t = u t-∆t + ∆t u& t- ∆t + && t- ∆t + β ∆t &u
3
u && (20.2a)
2

&& t- ∆t + γ ∆t &u
u& t = u& t-∆t + ∆t u 2
&& (20.2b)

If the acceleration is assumed to be linear within the time step, the following
equation can be written:

&& t − u
(u && t − ∆t )
&& =
&u (20.3)
∆t

The substitution of Equation (20.3) into Equations (20.2a and b) produces

Newmark’s equations in standard form

1
u t = u t-∆t + ∆t u& t-∆t + ( − β ) ∆t 2 u
&& t- ∆t + β ∆t u
2
&& t (20.4a)
2

u& t = u& t-∆t + (1 − γ ) ∆t u

&& t- ∆t + γ ∆t u
&& t (20.4b)

Newmark used Equations (20.4a, 20.4b and 20.1) iteratively, for each time step, for
each displacement DOF of the structural system. The term u &&t was obtained from

Equation (20.1) by dividing the equation by the mass associated with the DOF.

In 1962 Wilson [2] formulated Newmark’s method in matrix notation, added

stiffness and mass proportional damping, and eliminated the need for iteration by
introducing the direct solution of equations at each time step. This requires that
Equations (20.4a and 20.4b) be rewritten in the following form:

&& t = b1 ( u t − u t − ∆t ) + b2 u
u & t − ∆t + b3u
&& t − ∆t (20.5a)

u& t = b4 (u t − u t − ∆t ) + b5u& t − ∆t + b6u

&& t − ∆t (20.5b)

where the constants b1 to b6 are defined in Table 20.1. The substitution of

Equations (20.5a and20.5b) into Equation (20.1) allows the dynamic equilibrium of
the system at time “ t ”to be written in terms of the unknown node displacements
u t . Or,
4 STATIC AND DYNAMIC ANALYSIS

(b1M + b4 C + K )u t = Ft + M (b1u t − ∆t − b2u& t − ∆t − b3u

&& t − ∆t )
(20.6)
+ C(b4 u t − ∆t − b5u& t − ∆t − b6u
&& t − ∆t )

The Newmark direct integration algorithm is summarized in Table 20.1. Note that
the constants bi need be calculated only once. Also, for linear systems, the
effective dynamic stiffness matrix K is formed and triangularized only once.

20.3 ST ABIL IT Y OF NEWMARK’ S MET HOD

For zero damping Newmark’s method is conditionally stable if

1 1 1
γ ≥ , β ≤ and ∆t ≤ (20.7)
2 2
ω MAX γ 2 - β

where ω MAX is the maximum frequency in the structural system [1]. Newmark’s
method is unconditionally stable if

1
2β ≥ γ ≥ (20.8)
2

However, if γ is greater than ½, errors are introduced. These errors are associated
with “numerical damping” and “period elongation”.

For large multi degree-of-freedom structural systems the time step limit, given by
Equation (20.7), can be written in a more useable form as

∆t 1
≤ (20.9)
2π γ - β
TMIN
2

Computer models of large real structures normally contain a large number of

periods which are smaller than the integration time step; therefore, it is essential
that one select a numerical integration method that is unconditional for all time
steps.
DIRECT INTEGRATION METHODS 5

Table 20.1. Summary of the Newmark Method for Direct Integration

I. INITIAL CALCULATION

A. Form static stiffness matrix K , mass matrix M and damping matrix C

B. Specify integration parameters β and γ

C. Calculate integration constants

1 1 1
b1 = b2 = b3 = β − b4 = γ ∆t b1
β ∆t 2 β ∆t 2

b5 = 1 + γ ∆t b2 b6 = ∆t (1+ γ b3 − γ )

D. Form effective stiffness matrix K = K + b1M + b4 C

E. Triangularize effective stiffness matrix K = LDLT

F. Specify initial conditions u 0 , u& 0 , u

&& 0

II. FOR EACH TIME STEP t = ∆t, 2 ∆t, 3∆t - - - - - -

A. Calculate effective load vector

Ft = Ft + M (b1u t − ∆t − b2u& t − ∆t − b3u

&& t − ∆t ) + C(b4 u t − ∆t − b5u
& t − ∆t − b6 &u
& t − ∆t )

B. Solve for node displacement vector at time t

LDLTu t = Ft forward and back-substitution only

C. Calculate node velocities and accelerations at time t

u& t = b4 (u t − u t − ∆t ) + b5u& t − ∆t + b6u

&& t − ∆t

&& t = b1 ( u t − u t − ∆t ) + b2 u
u & t − ∆t + b3u
&& t − ∆t

D. Go to Step II.A with t = t + ∆t

6 STATIC AND DYNAMIC ANALYSIS

20.4 T HE AVERAG E ACCEL ERAT ION MET HOD

The average acceleration method is identical to the trapezoidal rule that has been
used to numerically evaluate second order differential equations for approximately
100 years. It can easily be derived from the following truncated Taylor’s series
expansion:

τ2 τ3
uτ = u t-∆t + τ u& t-∆t + && t- ∆t +
u &&t- ∆t +......
&u
2 6
(20.10)
τ u2
&& +u
&& t
≈ u t-∆t + τ u& t- ∆t + ( t-∆t )
2 2

where τ is a variable point within the time step. The consistent velocity can be
obtained by differentiation of Equation (20.10). Or,

&& t- ∆t + u
u && t
u& τ = u& t-∆t + τ ( ) (20.11)
2

If τ = ∆t

∆t 2 ∆t 2
u t = u t-∆t + ∆t u& t- ∆t + && t- ∆t +
u &u
&t (20.12a)
4 4

∆t ∆t
u& t = u& t- ∆t + && t - ∆t +
u u
&& t (20.12b)
2 2

These equations are identical to Newmark’s Equations (20.4a and b) with γ = 1 / 2

and β = 1 / 4 .

It can easily be shown that the average acceleration method conserves energy for
the free vibration problem, Mu && + Ku = 0 , for all possible time steps [4]. .
Therefore, the sum of the kinetic and strain energy is constant. Or,

2E = u& Tt Mu& t + u Tt Ku t = u& Tt-∆t Mu& t-∆t + u Tt-∆t Ku t-∆t (20.13)

DIRECT INTEGRATION METHODS 7

20.5 WIL SON’ S θ FACT OR

In 1973, the general Newmark method was made unconditionally stable by the
introduction of a θ factor [3]. The introduction of the θ factor is motivated by the
observation that an unstable solution tends to oscillate about the true solution.
Therefore, if the numerical solution is evaluated within the time increment the
spurious oscillations are minimized. This can be accomplished by a simple
modification to the Newmark method by using a time step defined by

∆t ′ = θ ∆t (20.14a)

and a load defined by

R t ′ = R t-∆t + θ ( R t − R t-∆t ) (20.14b)

where θ ≥ 10. . After the acceleration &u& t′ vector is evaluated by Newmark’s

method at the integration time step θ ∆t , values of node accelerations, velocities
and displacements are calculated from the following fundamental equations:

1
u && t- ∆t +
&& t = u && − u
(u && t- ∆t ) (20.15a)
θ t′

u& t = u& t-∆t + (1 − γ ) ∆t u

&& t- ∆t + γ ∆t u
&& t (20.15b)

∆t 2 (1 − 2β )
u t = u t-∆t + ∆t u& t- ∆t + && t- ∆t + β ∆t u
2
u && t (20.15c)
2

The use of the θ factor tends to numerically damp out the high modes of the
system. If θ equals 1.0 Newmark’s method is not modified. However, for
problems where the higher mode response is important, the errors that are
introduced can be large. In addition, the dynamic equilibrium equations are not
exactly satisfied at time t . Therefore, the author no longer recommends the use
of the θ factor. At the time of the introduction of the method, it solved all
problems associated with stability of the Newmark family of methods. However,
during the past twenty years new and more accurate numerical methods have been
developed.
8 STATIC AND DYNAMIC ANALYSIS

20.6 T HE USE OF ST IFFNESS PRO PORT IONAL DAMPING

Because of the unconditional stability of the average acceleration method, it is the

most robust method to be used for the step-by-step dynamic analysis of large
complex structural systems in which a large number of high frequencies, short
periods, are present. The only problem with the method is that the short periods,
which are smaller than the time step, oscillate indefinitely after they are excited.
The higher mode oscillation can be reduced by the addition of stiffness
proportional damping. The additional damping that is added to the system is of the
form

CD = δ K (20.16)

where the modal damping ratio, given by Equation (8.8) is defined by

1 π
ξn = δ ω n = δ (20.17)
2 Tn

One notes that the damping is large for short periods and small for the long periods
or low frequencies. It is apparent that periods which are greater than the time step
cannot be integrated accurately by any direct integration method. Therefore, it is
logical to damp these short periods to prevent them from oscillating during the
solution procedure. For a time step equal to the period, Equation (20.17) can be
rewritten as

∆T
δ = ξn (20.18)
π

Hence, if the integration time step is 0.02 seconds and we wish to assign a
minimum of 1.0 to all periods shorter than the time step, a value of δ = 0.0064
should be used. The damping ratio in all modes is now predictable for this
example from Equation (20.17). Therefore, the damping ratio for a 1.0 second
period is 0.02 and for a 0.10 second period is 0.2.
DIRECT INTEGRATION METHODS 9

20.7 T HE HILBER, HUG HES AND T AYLOR α MET HOD

The α method [4] uses the Newmark method to solve the following modified
equations of motion:

&& t + (1 + α ) Cu
Mu & t + (1 + α ) Ku t = (1 + α ) Ft
(20.19)
− αFt + α Cu& t − ∆t + α Ku t − ∆t

With α equals zero the method reduces to the constant acceleration method. It
produces numerical energy dissipation in the higher modes; however, it cannot be
predicted as a damping ratio as in the use of stiffness proportional damping. Also,
it does not solve the fundamental equilibrium equation at time t. However, it is
currently being used in many computer programs. The performance of the method
appears to be very similar to the use of stiffness proportional damping.

20.8 SELECT ION OF A DIRECT INT EGRAT ION MET HOD

It is apparent that a large number of different direct numerical integration methods

are possible by specifying different integration parameters. A few of the most
commonly used methods are summarized in 20.2.

Table 20.2. Summary of Newmark Methods Modified by the δ Factor

γ β ∆t
METHOD δ ACCURACY
T M IN

Central Difference 1/2 0 0 0.3183 Excellent for small ∆t

Unstable for large ∆t

Linear Acceleration 1/2 1/6 0 0.5513 Very good for small ∆t

Unstable for large ∆t

Average Acceleration 1/2 1/4 0 ∞ Good for small ∆t

No energy dissipation

Modified Average 1/2 1/4 ∆T ∞ Good for small ∆t

Energy dissipation for
Acceleration π large ∆t
10 STATIC AND DYNAMIC ANALYSIS

For single degree-of-freedom systems the central difference method is most

accurate; and the linear acceleration method is more accurate than the average
acceleration method. However, if only single degree-of-freedom systems are to be
integrated the piece-wise exact method, previously presented, should be used since
there is no need to use an approximate method.

It appears that the modified average acceleration method, with a minimum addition
of stiffness proportional damping, is a general procedure that can be used for the
dynamic analysis of all structural systems. Using δ = ∆T / π will damp out
periods shorter than the time step and introduces a minimum error in the long
period response.

20.9 NONLINEAR ANAL YSIS

The basic Newmark Constant acceleration method can be extended to nonlinear

dynamic analysis. This requires that iteration must be performed at each time step
in order to satisfy equilibrium. Also, the incremental stiffness matrix must be
formed and triangularized at each iteration or at selective points in time. Many
different numerical tricks, including element by element methods, have been
developed in order to minimize the computational requirements. Also, the
triangularization of the effective incremental stiffness matrix may be avoided by
the introduction of iterative solution methods.

20.10 SUMMARY

For earthquake analysis of linear structures, it should be noted that the direct
integration of the dynamic equilibrium equations is normally not numerically
efficient as compared to the mode superposition method using LDR vectors. If the
triangularized stiffness and mass matrices and other vectors cannot be stored in
high-speed storage, the computer execution time can be large.

After using direct integration methods for approximately forty years, the author can
no longer recommend the Wilson method for the direct integration of the dynamic
equilibrium equations. The Newmark constant acceleration method, with the
addition of very small amount of stiffness proportional damping, is recommended
for dynamic analysis nonlinear structural systems. For all methods of direct
DIRECT INTEGRATION METHODS 11

integration great care should be taken to make certain that the stiffness proportional
damping does not eliminate important high-frequency response. Mass proportional
damping cannot be justified because it causes external forces to be applied to the
structure that reduce the base shear for seismic loading.

In the area of nonlinear dynamic analysis one cannot prove that any one method
will always converge. One should always check the error in the conservation of
energy for every solution obtained. In future editions of this book it is hoped that
numerical examples will be presented in order that the appropriate method can be
recommended for different classes of problems in structural analysis.

20.11 REFERENCES

1. Newmark, N. M., “A Method of Computation for Structural Dynamics”, ASCE

Journal of the Engineering Mechanics Division, Vol. 85 No. EM3, (1959).

2. Wilson, E. L., “Dynamic Response by Step-By-Step Matrix Analysis”,

Proceedings, Symposium On The Use of Computers in Civil Engineering,
Labortorio Nacional de Engenharia Civil, Lisbon, Portugal, October 1-5, (1962).

3. Wilson, E. L., I. Farhoomand and K. J. Bathe, “Nonlinear Dynamic Analysis of

Complex Structures”, Earthquake Engineering and Structural Dynamics, 1, 241-
252, (1973).

4. Hughes, Thomas, “The Finite Element Method - Linear Static and Dynamic
Finite Element Analysis”, Prentice Hall, Inc., (1987).
12 STATIC AND DYNAMIC ANALYSIS