EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS, VOL. 21.

1059- 1070 (1992)

DYNAMIC CHARACTERISTICS OF MULTIPLE SUBSTRUCTURES

WITH CLOSELY SPACED FREQUENCIES

KANGMING XU AND TAKERU IGUSA

Department of Civil Engineering. Northwestern University, Evanston. Illinois 60208, U.S.A.

SUMMARY
This paper considers a main structure supporting a large number of substructures.The substructures have closely spaced
natural frequencies, and the combined main structure/multiple substructures system is subjected to harmonic or
wide-band forces. The goal is to characterize the effects of the substructures on the response of the main structure.
A special, fundamental case is studied in detail, where the substructures are oscillators with equal stiffnesses and
equally spaced natural frequencies.The exact response expressions for the combined system are in terms of a complicated
rational polynomial. However, by taking the limit where the number of substructures becomes large, the response
expressions reduce to simple, physically meaningful results. It is found that the multiple substrutures are equivalent to
a single viscous damping which is added to the damping of the main structure.An example illustrates how the results can
be applied to passive vibration control of large structures.

INTRODUCTION
Consider a system consisting of a main structure supporting one or more substructures. When a dynamic
load is applied to the system, the substructures produce forces which interact with the main structure. In
some cases, the response of the main structure will be less than that of the main structure without
substructures. Thus, with proper design, substructures can act as passive vibration controllers.
To understand the interaction between a main structure and its substructures, researchers have studied
several idealized, yet fundamental configurations. The most basic configuration is the two-degree-of-freedom
(2-DOF) structure consisting of a main oscillator supporting a single sub-oscillator. Den Hartog’ and
Crandall and Mark2 developed analytical expressions for the response of the 2-DOF system for harmonic
and wide-band inputs, respectively. These results were applied to the passive vibration control problem.
Optimal parameters values were determined for the sub-oscillator for both harmonic1. and wide-band4*
loads. For both cases, the sub-oscillator natural frequency was close to that of the main oscillator; thus, such
vibration controllers were termed tuned mass dampers (TMD’s).
The present paper examines structures which support many substructures. The goals are: to determine the
dynamic characteristics of such systems, to compare such systems with the 2-DOF system, and to show how
the results can be applied to passive vibration control problems.
The system configuration consists of a main oscillator supporting n sub-oscillators, as shown in Figure 1.
A harmonic or wide-band force is applied to the combined system. The displacement response of the main
oscillator is of interest. The system configuration is not new in vibration control of civil engineering
structures.6 However, the sub-oscillators are usually all tuned to the main oscillator. Such systems are
essentially identical to 2-DOF systems, and are therefore designed using the standard theory of TMDs.
This paper investigates the case where the natural frequencies of the sub-oscillators are equally spaced over
a frequency range. It is found that such systems are fundamentally different from 2-DOF systems. The paper
also shows that, under certain conditions, multiple sub-oscillators can have better vibration control
properties than a single sub-oscillator of equal mass. The basic ideas behind this paper were first presented by
the authors in References 7 and 8.

0098-8897/92/121059-12$11~00 Received 22 February 1992

0 1992 by John Wiley & Sons, Ltd. Revised I1 May 1992
1060 K. XU AND T. IGUSA

0 1 . Cl

I main oscillator I sub-oscillators I

Figure 1. System configuration

The analysis begins with the exact expression for the response of the main structure. Then, simple,
approximate expressions are derived by examining the limit for large numbers of sub-oscillators, and using
analytical techniques from mean-value theory', and asymptotic modal analysis.' The response expres-
sions for harmonic and wide-band input are surprisingly simple, considering the complexity of the coupled
system. It is found that multiple sub-oscillators are equivalent to a single damper which reduces the response
of the main oscillator. The results are applied to a vibration control problem.

EXACT RESPONSE EXPRESSIONS FOR THE MAIN OSCILLATOR

The properties of the system in Figure 1 are as follows. The main oscillator has natural frequency wo,
damping ratio lo and mass mo. Connected to the main oscillator are a set of n sub-oscillators, where thejth
oscillator has natural frequency wj, damping ratio l j and mass mi, for j = 1, . . . , n. The system is subjected
to a base acceleration or a force applied directly on the main oscillator mass. The response of interest is the
relative displacement of the main oscillator mass with respect to the base.
First, the steady-state harmonic response of the main oscillator is examined. For a harmonic excitation of
frequency w, the excitation is expressed by the complex exponential e -'Or and the main oscillator response is
given by y(w)e-'O', where y(w) is the complex harmonic amplitude of the response.
Appendix I derives the exact expressions for y(w) for the two loading cases. For the base acceleration
excitation, the harmonic response is

where
- 2iwwjCj)
Z(o)= - io 1 ornj(wf
j= j' 2iwwjcj w 2
- -

For the excitation applied on the main oscillator mass, the harmonic response is
y(w) = [cog - 2iwwO[, - o2- iwrn;'Z(w)]-' (3)
If the excitation is a wide-band stationary process, then the mean-square of the main oscillator response
is12

where C(o) is the one-sided power spectral density of the excitation.

 MULTIPLE SUBSTRUCTURES 1061

Figure 2. Configurationfor impedance definition

Hereafter, attention is only on base acceleration excitations. However, the results of this paper can be used
for excitations applied to the main oscillator mass by using equation (3) instead of equation (1).
The frequency function Z(o)has an important physical interpretation. Consider the configuration in
Figure 2, where the n sub-oscillatorsare connected to a common base without the main oscillator. If the base
is subjected to a unit harmonic velocity, then Z(o) is the amplitude of the harmonic force required to produce
the base motion. The function Z(o)is termed the impedance of the sub-oscillators.
Two special, elementary cases are briefly noted. The first case is n = 1; it is the 2-DOF system described in
the introduction. The second case is where n > 1, but the damping ratios and natural frequencies of the
sub-oscillators are all equal, i.e. i j = isand wj = 0,. It can be shown by substitution that the harmonic
response amplitude y ( o ) in equation (1) is equal to that of an equivalent 2-DOF system. The main oscillator
of the equivalent system is identical to that of the original system. The sub-oscillator of the equivalent system
has damping ratio is, natural frequency w, and mass m, = ml + +
. . . m,.

SIMPLE APPROXIMATIONS FOR A SPECIAL CLASS OF SUB-OSCILLATORS

In this paper, interest is on large values of n and sub-oscillators with unequal, but closely spaced natural
frequencies. The harmonic and wide-band responses of such systems can always be evaluated using the exact
expressions in equations (l), (2) and (4).However, since Z ( w ) is a high-order rational expression in o,it is
difficult to obtain a physical understanding of the dynamic characteristics of the systems. Moreover, when
the sub-oscillatorshave arbitrary properties, the system response can be erratic and complicated. Therefore,
in this section a special class of sub-oscillators is examined and approximations are developed to derive
simple, physically meaningful response expressions.

Properties of the sub-oscillators

The natural frequencies of the sub-oscillators are equally spaced, and their average is equal to the natural
frequency of the main oscillator, coo. The stiffness and damping constants are all equal. These properties are
mathematically expressed by
co.=coo[l
J + ( j - nT+ )l p ]
1062 K. XU AND T. IGUSA

f o r j = 1,2, . . . , n. The dimensionless parameter B measures the spacing of natural frequencies and is given
by

By combining equations (5)-(7), it can be shown that the average damping ratio is
is

The properties given in equations (5)-(7) were chosen because the main oscillator/sub-oscillators system
has well-behaved dynamic characteristics, as will be shown, shortly. In addition, the properties are parti-
cularly suitable for the design of passive vibration controllers. The sub-oscillators can be manufactured using
identical stiffnesses and damping constants; only the masses are varied. This is far simpler than manufactur-
ing sub-oscillators with varying stiffnesses or damping constants.
In the following, the response of the main oscillator is examined when the sub-oscillators have the
properties given in equations (5)-(7). Two approximations are developed. The first is valid when the natural
frequencies of the sub-oscillators are closely spaced. The second approximation, which has a very simple
form, is valid under the additional condition that the natural frequencies span a relatively wide range.

Approximate results for closely spaced natural frequencies

The impedance of the n sub-oscillators is obtained by substituting equations (5)-(7) into equation (2). The
result is

Z ( w ) = - iwk, 1 (03 - a2w2)-l

j= 1

where a is a function of the forcing frequency,

x = (1 - 2ic,

An approximation is developed for the impedance by replacing the summation in equation (10) with an
equivalent definite integral. This idea has been successfully used by Skudryzk', l o and Dowel1 and Kubotal'
to analyse a single structure with many closely spaced natural frequencies. The present paper uses this
technique to analyse many sub-oscillators.
First, equation (10) is written as

Z(w)=-2
iok
c Amj
mop i = 1 Of - x 2 02
where Amj is the frequency increment, Amj = mj+ - w j = boo.( F o r j = n, let Am,, = Pwo.) The summation
in equation (12) is approximated by an integral where the variable of integration is the sub-oscillator natural
frequency, w j . The result is

Since x 2 is a complex number, the integrand is a continuous, bounded function of s. From the theory of
Riemann integration, it is known that the accuracy of the approximation in equation (13) improves as the
frequency increment, Amj, becomes small. A quantitative criterion for accuracy is as follow^:'^ the errors in
the approximation will be less than 10 per cent when the frequency spacing, B, is less than the average
damping, is.
Approximations for the harmonic and mean-square response of the main oscillator are obtained by
substituting the impedance result in equation (13) into equations (1) and (4).
 MULTIPLE SUBSTRUCTURES 1063

Simplijied results for a wide span of natural frequencies

The approximation in equation (13) simplifies further when the span of sub-oscillator natural frequenciesis
wide. Define the non-dimensional parameter w = (w, - w l ) / w o ,which measures the span of the natural
frequencies. The condition that the natural frequencies span a wide frequency range is mathematically given
by two relations:

When these relations are satisfied, it is shown in Appendix I1 that the impedance simplifies to

Here, zI(w) is a linear function of w which is of the same order of magnitude as unity. Substituting equation
(15) into equation (1) yields an approximation for the harmonic response of the main oscillator

where

cE cE
If is of order unity, then the result for y(w) in equation (16)cannot be further simplified. However, if is
an order of magnitude less than unity, then two additional approximations are possible. The numerator in
equation (16) can be approximated by neglecting the second term. The term zI(w) in the denominator causes
a small shift of the resonance peak and only a small change in the height and width of the peak. Thus, this
term can also be neglected.
The final result for the harmonic response of the main oscillator is

The corresponding mean-square response is obtained by substituting this expression into equation (4). For
most wide-band inputs with slowly varying power spectral densities, the integral for the mean-square
response can be determined accurately using a white-noise approximation.” The result is

The expressions for the harmonic and mean-square responses in equations (18) and (19) are far simpler
than either the exact expressions in equations (l), (4) and (12) or the integral approximation given by
equations (l), (4) and (13). The computational effort for the mean-square response result in equation (19) is
minimal since it does not require the numerical integration in equation (4).
The results in equations (17), (18) and (19) also provide physical insight into the system. The expressions for
the responses are identical to those for an oscillator with the same natural frequency as the main oscillator,
but with a damping ratio that is increased by the amount cE. This implies that the effect of the multiple
sub-oscillators on the main oscillator response can be approximated by a simple increase in damping. This
additional damping value, CE, which is defined in equation (17),is called the equivalent damping ratio.
There are several properties of the equivalent damping ratio.

1. Independence with respect to the sub-oscillator damping ratios, cj.

When Ti are sufficiently small
[equation (14)], the equivalent damping, CE, does not depend on the damping of the sub-oscillators.
This may appear counter-intuitive since the damping of a structural system usually depends on the
1064 K. XU AND T. IGUSA

damping of its components. It is shown in the next section how this property is useful in vibration
control problems where the damping of the sub-oscillators is limited.
2. Independence with respect to the excitation frequency, w. When the conditions in equation (14) are
satisfied, iE does not change with w. It is shown later how this independence makes multiple
sub-oscillators a relatively robust vibration controller.
3. Proportionality to k , / ( w i m o ) . For fixed natural frequencies, the ratio k , / ( o $ m o ) is proportional to the
mass of the sub-oscillators. Therefore, as the sub-oscillators become more massive, they have a greater
damping effect on the main oscillator.
4. Proportionality to 1/8. The parameter B measures the spacing of sub-oscillator natural frequencies.
Smaller B, which corresponds to a larger number of sub-oscillators in any fixed band of natural
frequencies, also increases the equivalent damping.

The results of this section are illustrated by several numerical examples. In all examples, the main oscillator
has natural frequency w0/2n: = 1 Hz and damping ratio Co = 0.01.
Figure 3 compares harmonic responses of the main oscillator with and without sub-oscillators. Twenty-
one sub-oscillators are used with average damping ratio [, = 0.02 and natural frequencies spanning from
w1/2n = 0.8 Hz to ~ ~ ~ /=2 1.2 7 1Hz. According to equation (8), this yields the value f? = 0.020 for the
frequency spacing parameter. Also, from equation (6),the total mass of the sub-oscillators,mT, is given by the
sum
n n i
mT= 1 mj=k, 1
j= 1 j = 1 Wf

0.25

0.w' ' I ' ' I ' I

'. -..
0.25
0

--__
0.00
0.90 0.95 1.00 1.05 1 .I0
excitation frequency q427r) Hz

Figure 3. Normalized frequency response amplitudes: (a) mT = O.O1mo; (b) mT = 0.02mo; (c) m, = O.lrno: ~ exact and integral
approximation, - ~ ~simple approximation, . . - without sub-oscillators
~
 MULTIPLE SUBSTRUCTURES 1065

Substituting n = 21 and the values for the sub-oscillator natural frequencies yields the following relationship
between the sub-oscillators' total mass, mT, and their common stiffness, k,:
mT = 21*966ks (21)
which is valid for the particular examples studied in Figure 3.
The response amplitudes y ( w ) for the main oscillator with sub-oscillators are computed using the exact
results [equations (1) and (2)] and the integral [equations (1) and (1 3)] and simple [equations (17) and (18)]
approximations. The absolute values of the response amplitudes are normalized by the resonant response
amplitude of the main oscillator without sub-oscillators.
Figures 3(aHc) give the results when ?nT is 1, 2 and 10 per cent of the mass of the main oscillator,
respectively. For comparison, the figures also give the normalized responses for the main oscillator without
sub-oscillators. The plots show that the sub-oscillators reduce significantly the response of the main
oscillator. In addition, the plots show that the integral approximation is indistinguishable from the exact
results and that the simple approximation follows closely the exact results.
The simple approximation is in terms of an equivalent damping, which gives a physically meaningful
measure of the vibration reduction effect of the sub-oscillators.This equivalent damping is computed easily
using equations (17) and (21); as discussed earlier, the equivalent damping is proportional to mT.Thus, when
mT is 1,2 and 10 per cent of the mass of the main oscillator, the equivalent damping is CE = 0-018,0-036and
018, respectively. The increase in damping is clearly shown by the reduction of the resonant peak response in
Figures 3(aHc).
Figure 4 shows how changes in the span of the sub-oscillator natural frequenciesaffects the response of the
main oscillator when the input is wide band. Twenty-one sub-oscillatorsare used with average damping ratio
5, = 0.02 and total mass which is 1 per cent of the mass of the main oscillator. The input is a white-noise base
acceleration. The root-mean-square response, D,,, of the main oscillator is computed using the exact results
[equations (l), (2) and (4)] and the integral [equations (l), (4) and (1 3)] and simple [equations (17) and (19)]
approximations. The responses are normalized by the root-mean-square response of the main oscillator
without sub-oscillators.
The normalized responses are plotted with respect to the span of natural frequencies (021 - o,)/2n in
Figure 4. The smallest span is 0 Hz, which corresponds to sub-oscillators with identical natural frequencies,
oj/27r = 1 Hz. This limiting case is equivalent to a single sub-oscillator with a mass equal to 1 per cent of the
mass of the main oscillator. The largest span is 0 8 Hz, which corresponds to sub-oscillators with natural
frequencies ranging from co1/2n = 0.6 Hz to w,,/2n = 1.4 Hz.
Figure 4 shows that, as the span of natural frequencies increases, the response first decreases, then steadily
increases. The increasing portion of the response curve can be explained in terms of the equivalent damping
concept. According to the discussion after equation (19), the equivalent damping is inversely proportional to

I
0.M) L - L l h
0.0 0.2 0.4 0.6 0.8
(@I - mM2N Hz

- wl)/2n: -exact,
Figure 4. Normalized root-mean-square response with respect to the span of natural frequencies, (02,
- - _ _ - integral approximation, - -- simple approximation
~
1066 K. XU AND T.IGUSA

the frequency spacing parameter, /I. Since the number of sub-oscillators, n, is fixed, /I is proportional to the
span of natural frequencies. Therefore, as the span increases, the following occurs: fl increases, the equivalent
damping decreases and the response increases. The initial decreasing portion of the response curve does not
follow the equivalent damping concept. This is because equivalent damping applies only when the span of
natural frequencies is sufficiently wide.
The integral approximation nearly coincides with the exact results for all frequency spans. The simple
approximation follows the exact results when the span of natural frequencies is greater than about 0 3 Hz,
and gives inaccurate results for smaller spans. This is expected because the simple approximation is valid
only for wide spans, as specified in equation (14).

APPLICATION TO PASSIVE CONTROL

The preceding results suggest that multiple sub-oscillators can be used to control passively vibrations of the
main oscillator. This section explores this possibility and compares the effectiveness of multiple sub-
oscillators with that of a single TMD. Since the multiple sub-oscillators are nearly tuned to the main
oscillator, they are called multiple TMD’s.
Response to harmonic input is best controlled by a single TMD. This is because the input consists of only
a single frequency, and the TMD can be designed so that its natural frequency coincides with the input
frequency. In cases where the input frequency varies slowly with time, self-tuning TMDs have been
developed.l4 These adaptive devices continuously adjust their natural frequency so that it always coincides
with the input frequency.
Response to wide-band input is more difficult because the force includes contributions from a wide range
of frequencies. The following analysis shows that, under certain conditions, multiple TMDs can be more
effective than a single TMD in controlling such responses.
The same main oscillator used in Figures 3 and 4 is used again. For the multiple TMDs, twenty-one
sub-oscillators are used with total mass equal to 1 per cent of the mass of the main oscillator. The same total
mass is also used for the single TMD. The input is a white-noise base acceleration. The root-mean-square
response is normalized as before, and only exact calculations are used.
The natural frequencies of the multiple TMDs are determined using the results in Figure 4.The figure
shows that the response of the main oscillator is minimized when the span of natural frequencies is 0-16 Hz.
This corresponds to natural frequencies ranging from w1/27c = 0.92 Hz to w2,/27c = 1.0 Hz. For the single
TMD, the optimal natural frequency is 099 H z . ( ~ )
The remaining unspecified parameter is the damping, is, of the TMD’s. Figure 5 shows how the main
oscillator response varies with is. At sufficiently high damping, the two control devices give similar results:
the main oscillator response slowly increases with damping. However, at low damping, the two devices give
markedly different results.

0251

L l I , I
O Too 0.05 0.10 0.15 0.20
damping of TMD Cs
Figure 5. Normalized root-mean-square response with respect to TMD damping, <.:-21 TMDs. ~ ~
single TMD
 MULTIPLE SUBSTRUCTURES 1067

The single TMD minimizes the main oscillator response at approximately 5 per cent damping. This agrees
with the theoretical results given by Warburtons At lower damping, the single TMD rapidly loses all of its
effectiveness: the main oscillator response approaches unity, which corresponds to the response of the main
oscillator without passive control.
The multiple TMDs are significantly less sensitive than the single TMD at low damping. To explain this
phenomenon, the equivalent damping concept and the associated simple approximation are examined
briefly. The normalized response obtained from the simple approximation is 0.42 1. Since the equivalent
damping is independent of the sub-oscillator damping, the result would be a straight line in Figure 5. As
described in Figure 4, the simple approximation becomes inaccurate for the frequency span used in Figure 5.
Nevertheless, the equivalent damping concept gives physical insight into the relatively insensitive response
for multiple TMDs at low damping.
Figure 5 shows that multiple TMDs can be more effective than a single TMD. Although the difference is
slight for most damping values, it becomes large at low damping. Thus, multiple TMDs would provide
greater vibration control in cases where the damping of the sub-oscillators is small. One important
application in civil engineering is in the design of liquid tuned dampers (LTDs).
LTDs are containers partially filled with a liquid such as water. Such devices can have resonant behaviour,
and the liquid sloshing produces a damping effect. For preliminary analysis, LTDs can be modelled
approximately as a single-degree-of-freedom oscillator.6The most significant limitation of LTDs is their low
level of damping. A limited increase of damping is possible by two methods: increasing the free surface area
by using many small LTDs and placing additives in the liquid. The resulting net damping value is about
2 per cent.”
The usual design is to make the natural frequencies of all of the LTDs equal to that of the main structure.6
However, this is equivalent to a single TMD with mass equal to the sum of masses of the LTD’s. The results
of this section demonstrate that a single TMD design is not suitable for low damping values. The alternative
is to make the natural frequencies of the LTD’s equally spaced over a frequency range. Figure 5 shows how
this design can reduce effectively the response of the main structure.

SUMMARY AND CONCLUSIONS

This paper shows how multiple sub-oscillators can affect the response of a main oscillator. The results are
applied to vibration control. The key properties of multiple sub-oscillators are summarized in the following:

1. They have simple limiting behaviour when the number of sub-oscillators becomes large and the natural
frequencies become closely spaced.
2. They can be represented by an equivalent damping when their natural frequencies span a sufficiently
wide range. This equivalent damping is proportional to the sub-oscillator masses, inversely propor-
tional to the spacing of natural frequencies and independent of the damping.
3. They effectivelyreduce the response of the main oscillator. The equivalent damping concept gives
a physical interpretation of the vibration reduction effect.
4. They are nearly as effective as a single tuned mass damper for most damping values. They are
significantly more effectiveat low damping values. Thus, multiple sub-oscillators are appropriate for
cases where the damping of the oscillators is limited to low values. An important civil engineering
application is the design of tuned liquid dampers.

ACKNOWLEDGMENTS
The research work in this paper was partially supported by the National Science Foundation under Grant
No. BCS-8 858 549 and by the Ohsaki Research Institute of Shimizu Corporation. This support is gratefully
acknowledged. The authors would also like to thank Professor G. B. Warburton for his critical review and
valuable suggestions.
1068 K.XU AND T. ICUSA

APPENDIX I
Derivation of the harmonic response function
The equation of motion of the system in Figure 1 is
Mji(t) + Cir(t)+ K x ( t ) = f w ( t ) (22)
+
Here, x is the ( n 1)-vector for the displacement responses relative to the base, where the first component
corresponds to the main oscillator and the remaining components correspond to the n sub-oscillators. The
( n + 1)x (n + 1) matrices M , C and K are the mass, damping and stiffness matrices, respectively, which are
given by
M = diag{mo, ml,. . . , m,> (23)

c1 - cz ... -
C1 0 ...
c2 ...

k, - k2 . . . -
kl 0 ...
K= k2 . . .

Here, diag {. ) denotes the diagonal matrix with diagonal elements given in braces, and (sym) indicates
symmetric matrices. The matrices C and K contain scalars c j = 2wjl;imj and k j = wfmj which are the
damping constant and stiffness of the jth oscillator.
The right hand side of equation (22) is the external excitation, which is expressed as a product of a constant
+
( n 1)-vector f = [fo fi . . .f.lTand a scalar time function w ( t ) .If the excitation is a force w ( t ) applied only
on the main oscillator, thenfo = 1, and f i = 0 for i 2 1. If the excitation is an acceleration w ( t ) at the base of
the system, then f i = mi for all i.
The steady-state harmonic response of the system is obtained by substituting w ( t ) = e-ior and x ( t ) = X(w)
- i o t into equation (22) and solving for X(o):

X(w) = ( 0 2 M- iwC + K)-'f (26)

Since the matrices M, C and K have non-zero terms only along the diagonal, the first row and the first
column, the inverse in equation (26) can be evaluated explicitly using Cramer's rule. The result is given for the
first element of X(w), which is the harmonic response amplitude y(w) for the main oscillator:

For the base acceleration excitation, f i = mi and the two summations become identical. Therefore, the
expression for y ( w ) can be rewritten
mo - (iw)-'Z(w)
y ( w ) = ko - iwco - u Z m O- iwZ(w)
where
mj(kj - iocj)
Z(o)=-iw C
j=
.
k j - twcj - w2mj
 MULTIPLE SUBSTRUCTURES 1069

For an excitation applied at the main oscillator mass, the first bracketed term in equation (27) reduces to
unity. Thus, the expression for y(w) reduces to
y ( o ) = [k, - ioc, - d m o - iwZ(w)]-’ (30)
The final results in equations (28)-(30) are rewritten in terms of the natural frequencies,damping ratios and
masses of the sub-oscillators in equations (1) and (3) of the main text.

APPENDIX I1
Approximation for the impedance
The conditions in equation (14) are used to develop an approximation for the complex logarithm
expression for Z ( w ) in equation (13).
The first condition in equation (14) implies that the argument of the logarithm in equation (13) has a real
part which is negative and large in comparison with the imaginary part. Therefore, the complex-valued
=
parameter a can be approximated by a 1, and the impedance can be rewritten

Here, the first logarithm is in. The second logarithm is simplified by using the second condition in equation
(14).
It is useful to define another parameter, p = (w - wo)/wo.This non-dimensional parameter measures how
close the excitation frequency is to the central natural frequency of the sub-oscillators, o0.The second
condition in equation (14) becomes p a w. The second logarithm in equation (31) is rewritten in terms of
w and p, yielding

+ 4 + 2p)
z(w) = -
iwk,
-[in
O O P
+ log1(-(w -w 2-p 2p)(w
)(- w + 4 + 2p) I1
The argument of the logarithm is expanded in terms of p. Since p is small, only the first-order term is retained

Finally, the approximation log (1 + x) = x , which is valid for small x , is applied to equation (33). The final
result is given by
nk,
Z(o)z - [l + izI(o)] (34)
2WOP
where

REFERENCES
1. J. P. Den Hartog, Mechanical Vibrations, 4th edn, McGraw-Hill, New York, 1956.
2. S. H. Crandall and W. D. Mark, Random Vibration in Mechanical Engineering, Academic Press, New York, 1963.
3. G . B. Warburton and E. 0.Ayorinde, ‘Optimum absorber parameters for simple systems’, Earthquake eng. struct. dyn. 8, 197-217
(1 980).
4. P. H. Winching and G . C. Campbell, ‘Minimal structural response under random excitation using the vibration absorber’,
Earthquake eng. stmct. dyn. 2, 303-312 (1974).
5. G. B. Warburton, ‘Optimum absorber parameters for various combinations of response and excitation parameters’, Earthquake eng.
struct. dyn. 10, 381-401 (1982).
6. L. M. Sun, Y. Fujino, B. Pacheco and P. Chaiseri, ‘Modelling of tuned liquid damper (TLD)’, J . wind eng. indus. aerodyn. (to appear).
7. T. Igusa and K. Xu,‘Wide-band response of multiple subsystems with high modal density’, Proc. 2nd int. con5 stochasf.struct. dyn.
Stochasr. strucr. dyn. 1: New theor. developments (Eds. Y. Lin and I. Elishakoff), Florida, U.S.A. (1990).
1070 K. XU AND T.IGUSA

8. T. Igusa and K. Xu, ‘Vibration reduction characteristics of distributed tuned mass dampers’, Proc. 4th int. con$ recent ado. strut. dyn.
Southampton, U.K. (1991).
9. E. Skudrzyk, Simple and Complex Vibratory Systems, Pennsylvania State University Press, State College, Pennsylvania, 1968.
10. E. Skudrzyk, ‘The mean-value method of predicting the dynamic response of complex vibrators’, J. acoust. soc. Am. 67, 110551135
(1980).
11. E. H. Dowel1 and Y.Kubota, ‘Asymptotic modal analysis and statistical energy analysis of dynamical systems’, J . appl. mech. A S M E ,
52,949-957 (1985).
12. Y. K. Lin, Probabilistic Theory of Structural Dynamics, Krieger, Huntington, New York, 1967.
13. K. Xu, ‘Dynamic analysis and control of structures with closely spaced natural frequencies’, Ph.D. Thesis, Department of Civil
Engineering, Northwestern University, Evanston, Illinois, 1992.
14. R. G. Loewy, ‘Helicopter vibrations: a technological perspective’, J. Am. helicopter SOC. 29, 4-30 (1984).
15. Y. Fujino, B. Pacheco, P. Chaiseri, L. M. Sun and K. Koya, ‘Understanding of TLD properties based on TMD analogy’ (in
Japanese), J. struct. eng JSCE MA, 577-590 (1990).