Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

EARTHQUAKE ENGINEERING & STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2016; 45:193–211


Published online 17 September 2015 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/eqe.2622

Modeling viscous damping in nonlinear response history analysis of


buildings for earthquake excitation

Anil K. Chopra*,† and Frank McKenna


1
Department of Civil Engineering, University of California, Berkeley, CA, USA

SUMMARY
The Rayleigh damping model, which is pervasive in nonlinear response history analysis (RHA) of buildings,
is shown to develop ‘spurious’ damping forces and lead to inaccurate response results. We prove that a vis-
cous damping matrix constructed by superposition of modal damping matrices—irrespective of the number
of modes included or values assigned to modal damping ratios—completely eliminates the ‘spurious’
damping forces. This is the damping model recommended for nonlinear RHA.
Replacing the stiffness-proportional part of Rayleigh damping by the tangent stiffness matrix is shown to
improve response results. However, this model is not recommended because it lacks a physical basis and has
conceptual implications that are troubling: hysteresis in damping force–velocity relationship and negative
damping at large displacements. Furthermore, the model conflicts with the constant-damping model that
has been the basis for fundamental concepts and accumulated experience about the inelastic response of
structures.
With a distributed plasticity model, the structural response is not sensitive to the damping model; even the
Rayleigh damping model leads to acceptable results. This perspective on damping provides yet another rea-
son to employ the superior distributed plasticity models in nonlinear RHA.
OPENSEES software has been extended to include a damping matrix defined as the superposition of modal
damping matrices. Although this model leads to a full populated damping matrix, the additional computa-
tional demands are demonstrated to be minimal. Copyright © 2015 John Wiley & Sons, Ltd.

Received 27 April 2015; Revised 16 July 2015; Accepted 27 July 2015

KEY WORDS: buildings; concentrated plasticity; damping models; distributed plasticity; MODAL
damping; nonlinear analysis; Rayleigh damping; response history analysis; spurious
damping forces

1. INTRODUCTION

Nonlinear response history analysis (RHA) of structures is being increasingly employed in


performance-based earthquake engineering. Realistic modeling of structures is obviously essential to
achieve reliable results. Several aspects of modeling of buildings remain challenging; these include
the following: structural element behavior at large deformations, contributions of gravity framing
and nonstructural elements that are usually ignored; accidental torsion, especially for unsymmetric-
plan buildings; and P-delta analysis to include gravity loads in the presence of torsion. The
modeling of damping, which remains unresolved, is the subject of this paper.
The standard equations that are solved in nonlinear RHA of structures are

u þ cu_ þ f s ðuÞ ¼ mιx €u gx ðt Þ  mιy €u gy ðt Þ  mιz €u gz ðt Þ þ pgr


m€ (1)

*Correspondence to: Anil K. Chopra, Department of Civil Engineering, University of California, Berkeley, CA, USA.

E-mail: ijeesd@berkeley.edu

Copyright © 2015 John Wiley & Sons, Ltd.


194 A. K. CHOPRA AND F. MCKENNA

where u is the vector of degrees of freedom (DOFs), m and c are the mass and damping matrices,
respectively, and the vector fs(u) represents the nonlinear relation between resisting forces and
deformations, which includes both material and geometric nonlinearities. (For linear systems,
fs = k u, where k is the stiffness matrix.) The right side represents the dynamic excitation: ügx(t), ügy
(t), and ügz(t) are the x, y, and vertical components of earthquake ground acceleration; ιx, ιy, and ιz
are the corresponding influence vectors; and pgr represents gravity loads.
The damping matrix is not required for linear analysis of classically damped systems because these
systems are amenable to classical modal analysis where modal damping ratios enter directly in the
modal equations of motion. However, the damping matrix is explicitly required if classical modal
analysis is not applicable, as for nonlinear RHA of structures.
Whenever appropriate, it is convenient to consider damping matrices that lead to classical natural
modes with prescribed modal damping ratios selected based on values ‘measured’ for similar
structures. System identification methods applied to structural motions recorded during earthquakes
provide estimates of damping ratios. For this purpose, earthquake records are appropriate only if they
come from structures that did not deform into the inelastic range because the damping matrix is
intended to model energy dissipation other than that associated with yielding of structural elements.
Energy dissipation in yielding is accounted for separately through nonlinear hysteretic force–
deformation relations for structural elements, as represented by fs(u) in Equation (1).
The viscous damping matrix that is consistent with estimated values of the modal damping ratios may
be interpreted as equivalent viscous damping in the sense that the vibrational energy it dissipates is
equivalent to the energy dissipated in all the damping mechanisms present in the structure including
thermal effects from repeated elastic straining of structural materials, repeated straining of
nonstructural elements—partition walls, mechanical equipment, fire proofing, and so on—friction at
steel connections, opening and closing of microcracks in concrete, friction between the structure itself
and nonstructural elements, and other mechanisms, some of which are even difficult to identify.
It is standard in earthquake dynamics of structures to model the various energy dissipating
mechanisms mentioned previously by the damping forces f D ¼ c u_ in Equation (1). Although
calibrated against modal damping ratios estimated from structural motions recorded within the pre-
yielding range of response, it is customary to use the same model for nonlinear RHA of
structures. This extension is based on the tacit assumption that the previously mentioned energy
dissipating mechanisms continue unchanged even after the structure has yielded. Because
experimental evidence is not available to support or refute this assumption, the damping model in
Equation (1) is likely to continue for the foreseeable future.
Rayleigh damping is the most common—almost pervasive—model for viscous damping in
nonlinear RHA of buildings. The damping matrix is expressed as c = a0m + a1k where the
coefficients a0 and a1 are determined from damping ratios specified in two natural modes of
vibration of the structure or at two selected frequencies. Rayleigh damping is computationally
attractive because the damping matrix preserves the sparsity pattern of the stiffness matrix and does
not require formation of additional matrices.
However, nonlinear RHA of buildings modeled with Rayleigh damping may lead to responses that
are physically not plausible [1–4]; for example, large damping moments appear at the beam–column
joints when the structural elements are yielding, with the corollary effect that bending moments
computed from fs(u) in beams and columns framing into a joint are not in equilibrium. This problem
was first identified by Chrisp [1] and subsequently by Bernal [3]. As will be demonstrated later, this
unbalanced moment is entirely associated with the damping force from the stiffness-proportional
part of the Rayleigh damping model; we will refer to these damping forces as ‘spurious’ for reasons
that will become apparent later.
Several ideas to alleviate or reduce these ‘spurious’ damping forces have been proposed: limit the
damping forces [4], use the tangent stiffness instead of the initial stiffness in Rayleigh damping [5–8],
or assign zero damping to the yielding component of each structural member [6, 9]. Furthermore, the
requirement on the damping matrix to eliminate the ‘spurious’ damping forces was identified, and two
methods to achieve such a damping matrix were proposed [3].
The purpose of this paper is to identify a damping model that is rooted in fundamental theory, is
practical to implement, and completely eliminates the ‘spurious’ damping forces. After a section on

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
MODAL DAMPING 195

preliminaries for convenient reference, we synthesize various ideas in the literature while providing
historical context and prove that a damping matrix constructed by superposition of modal damping
matrices—irrespective of the number of modes included or values assigned to modal damping ratios
—completely eliminates the ‘spurious’ damping forces. The computational implications of using this
damping model are discussed. With the goal of being effective in contributing towards an improved
damping model, we have adopted a textbook-like style for this paper.

2. PRELIMINARIES: DAMPING MODELS FOR LINEAR SYSTEMS

2.1. Rayleigh damping


A classical damping matrix can be assembled as a linear combination of the mass and stiffness
matrices [10]:

c ¼ a0 m þ a1 k (2)

Known as Rayleigh damping, this is the most common—almost pervasive—model in nonlinear


RHA of structures. The damping ratio for the nth mode of such a system is

a0 1 a1
ζn ¼ þ ωn (3)
2 ωn 2

The coefficients a0 and a1 can be determined from specified damping ratios ζ i and ζ j for the ith and
jth modes, respectively. Expressing Equation (3) for these two modes in matrix form leads to

   ( )
1 1=ωi ωi a0 ζi
¼ (4)
2 1=ωj ωj a1 ζj

These two algebraic equations can be solved to determine the coefficients a0 and a1.
The damping matrix is then known from Equation (2), and the damping ratio for any other mode, given
by Equation (3), varies with natural frequency, as shown in Figure 1. In applying this procedure to a

Figure 1. Variation of modal damping ratios with natural frequency in Rayleigh damping; in this figure,
ζi=ζj=ζ.

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
196 A. K. CHOPRA AND F. MCKENNA

practical problem, the modes i and j with specific damping ratios should be chosen to ensure that the
damping ratios in all the modes contributing significantly to the response are close to the target value.
Observe that the damping ratio in the modes with frequencies lower than ωi increases rapidly.
Similarly, the damping ratios in the modes with frequencies higher than ωj increase monotonically
without bound; the asymptotic value at high frequencies is ζ n = (a1/2)ωn.

2.2. Caughey damping


To specify values for damping ratios in more than two modes requires starting with the general form
for a classical damping matrix given by the Caughey series [11, 12]:

N1  l
c ¼ m ∑ a1 m1 k (5)
l¼0

where N is the number of DOFs in the system and al are constants. The summation in Equation (5)
could be over any set of N integer values of l, positive or negative; the range shown is perhaps the
most common. If damping ratios for J modes of an N-DOF system are to be specified, then J terms
need to be included in the Caughey series; in principle, these could be any J terms. If the first J
terms in Equation (5) are included,

J1  l
c ¼ m ∑ al m1 k (6)
l¼0

and the modal damping ratio ζ n is given by

1 J1
ζn ¼ ∑ al ω2l1
n (7)
2 l¼0

The coefficients al (l = 0 to J  1) can be determined by solving the system of algebraic equations:


2 38 9 8 9
1 ω21 ω41 ⋯ ω12J2 > a0 > > 2ω1 ζ 1 >
6 7>> >
> > > >
>
61 ω22 ω42 ⋯ ω22J2 7< a1 = < 2ω2 ζ 2 =
6 7 ¼ (8)
6⋮ ⋮ 7
4 ⋮ ⋮ ⋱ 5>>
> ⋮ >>
>
>
>
> ⋮ > >
>
: ; : ;
1 ω2N ω4N ⋯ ωN2J2 aJ1 2ωJ ζ J

With al determined, the damping matrix c is known from Equation (6), and the damping ratios for
modes n = J + 1, J + 2, …, N are given by Equation (7).
Although Caughey damping is attractive in that it can provide better control of damping ratios over a
wide range of modal frequencies, it has not been popular in practical analyses perhaps because it is a
fully populated matrix and its construction requires (i) vibration frequencies of the structure, (ii) the
solution of a potentially ill-conditioned system of linear equations to determine the al coefficients
[13], and (iii) care in ensuring that the unprescribed damping ratios in higher modes do not take on
negative values [14].

2.3. Superposition of modal damping matrices


Consider the square matrix

C ¼ ΦT c Φ (9)

where Φ = [ϕ1, ϕ 2, …, ϕN] is the collection of all the natural vibration modes. For classically damped
systems, C is a diagonal matrix. An alternative procedure to determine a classical damping matrix from
modal damping ratios can be derived from Equation (9), resulting in [15]

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
MODAL DAMPING 197

N
2ζ ωn
c ¼ m ∑ n ϕn ϕTn m (10)
n¼1 M n

The nth term in this summation is the contribution of the nth mode with its damping ratio ζ n to the
damping matrix c; if this term is not included, the resulting c implies zero damping ratio in the nth mode.
The damping matrix defined by Equation (10) has not been popular in practical analyses because it
is a fully populated matrix and its construction requires all frequencies and modes.

3. SYSTEM AND GROUND MOTION

All response results are presented for a variant of the Seattle 20-story moment-resisting steel frame building
designed for the SAC project, as described in the Appendix. Modal damping ratios are assumed to be 2%.
This value is assigned to the first and third modes to determine a0 and a1 in Rayleigh damping [Equation
(2)]; the same a0 and a1 are used in Equation (16) to define Rayleigh damping using the tangent stiffness
matrix. When the damping matrix is defined as superposition of modal damping matrices, all 20 modes are
included in Equation (10), with ζ n = 2% in all modes, but some results are presented with fewer modes
included. The initial stiffness of the concentrated plastic hinge element is defined as HM (6EI/L), where
E, I, and L refer to the beam and HM is the hinge multiplier. Most results are presented for HM = 1000.
A few results are presented for other values of HM, as will be noted.
All response results presented are for the ground motion (GM) defined by the SE30 record from the
1985 Valparaiso, Chile, earthquake. It was one of the 20 records with an exceedance probability of 2%
in 50 years selected for the Seattle site as part of the SAC project.

4. SPURIOUS DAMPING FORCES FROM RAYLEIGH DAMPING MODEL

Consider a structural model that has mass associated with DOFs ut, zero inertia associated with DOFs u0,
and external dynamic forces applied only along DOFs ut. This situation is typical in earthquake analysis of
building frames, where the joint rotation DOFs are usually assigned zero inertia and the effective external
forces associated with horizontal GM exist along lateral floor displacements, ut, but not in rotational DOFs.
Specializing Equation (1) for an undamped, linearly elastic system and writing in partitioned form gives
       
mtt 0 €t
u ktt kt0 ut p ðt Þ
þ ¼ (11)
0 0 €0
u k0t k00 u0 0
Because no inertia terms or external forces are associated with DOFs u0, the second equation permits a
static relation between u0 and ut:

u0 ¼ k1
00 k0t ut (12)

We demonstrate next that for a linear system with Rayleigh damping, the damping forces in u0
DOFs will be zero. The damping forces are
    
f Dt ctt ct0 u_ t
¼ (13)
f D0 c0t c00 u_ 0

Substituting Equation (12), the damping forces in u0 DOFs are given by

f D0 ¼ a1 ½k0t u_ t þ k00 u_ 0  (14)

Differentiating Equation (12) with respect to time gives a corresponding equation in velocities that is
substituted in Equation (14) to obtain


f D0 ¼ a1 k0t u_ t  k00 k1 _t ¼ 0
00 k0t u (15)

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
198 A. K. CHOPRA AND F. MCKENNA

Although the damping forces in u0 DOFs do not exist during linear response of a structure, they
develop after the structure deforms into the inelastic range. Because the stiffness-proportional part of
the Rayleigh damping matrix is based on the initial stiffness matrix, even after yielding of the
structure, the damping forces are given by Equation (14). However, the internal elastic resisting
forces are no longer given by fs = ku, but by the nonlinear hysteretic function fs(u). This implies that
the tangent stiffness matrix varies with time, Equation (12) is no longer valid, and the damping
forces fD0 defined by Equation (14) will no longer be zero. This seems anomalous in the sense that
viscous damping moments develop because of yielding of the structural elements, a mechanism that
has no causal relationship to viscous damping. It is for this reason that we refer to them as spurious
damping forces.
These spurious damping forces are demonstrated in Figure 2 where the damping moment—
normalized by the yield moment My of the beam—at a joint on the 18th floor of the selected
building (see Appendix) is presented as a function of time. This damping moment is equal to the
unbalanced moment Mu at the joint if member forces are computed from fs(u). Also shown in
Figure 2 are intervals of yielding of the plastic hinge. Observe that moment imbalance exists at a joint
if the structural elements framing into the joint are yielding but decreases quickly when they return to
linearly elastic state. The unbalanced moment is very large, almost three times the yield moment of the
beam; thus, the computed response of the structure is expected to be in error, as confirmed later.
To reduce the unbalanced moments at joints arising from Rayleigh damping, several researchers [5–8]
have proposed to modify the stiffness-proportional term by replacing the initial stiffness matrix by the
tangent stiffness matrix, that is,

c ¼ a0 m þ a1 kT (16)

This option is available in widely used software, for example, PERFORM-3D, OPENSEES, and
RUAUMOKO; the latter software had included this feature as early as 1980. The spurious damping
forces are reduced by using the tangent stiffness instead of the initial stiffness. This is apparent by
comparing the unbalanced moment at the same joint when damping is modeled by Equation (16),
Figure 3a, versus Equation (10), Figure 2.
The unbalanced moment Mu increases with initial stiffness of the hinge and with the intensity of
GM. The peak (or maximum absolute) value over time of Mu is plotted over the height of the
building; at each floor level, this represents the largest value among the six beam–column joints.
Presented in Figure 4a are such results for three choices of the initial stiffness of the hinge
element = HM (6 EI/L), defined by HM = 1000, 100, and 10; the basis for these choices is presented
in the Appendix. It is evident that the unbalanced joint moment is larger for the stiffer hinge.
Similar results are presented in Figure 4b for GMs of three different intensities; the same GM
modified by scale factors, SFs = 0.1, 0.5, and 1.0. When subjected to the least intense of the three
excitations, the building does not yield; that is, it responds within the linear range, and the
unbalanced moments disappear; this is a numerical confirmation of Equation (15). With increasing
SF, the structure responds farther into the inelastic range, resulting in a larger unbalanced moment.

Figure 2. Unbalanced moment and yielding histories at a beam–column joint on the 18th floor of a 20-story
frame with Rayleigh damping.

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
MODAL DAMPING 199

Figure 3. Unbalanced moment at same joint as in Figure 2 for two damping models: (a) Rayleigh damping
with tangent stiffness and (b) superposition of modal damping matrices.

Figure 4. Peak (or maximum absolute) value over time of the unbalanced moment; at each floor, it is the
largest value among the six beam–column joints: (a) results for three values of the initial stiffness of the
hinge defined by HM = 1000, 100, and 10; SF = 1; and (b) results for GMs of three different intensities de-
fined by scale factors, SF = 0.1, 0.5, and 1.0; HM = 1000.

5. RECOMMENDED DAMPING MODEL

5.1. Superposition of modal damping matrices


We recommend the damping model defined by superposition of modal damping matrices for nonlinear
RHA because this model eliminates the spurious damping forces, as will be demonstrated in this
section. Consider the damping matrix defined by Equation (10), written in partitioned form:

       
ctt ct0 mtt 0 N 2ζ ωn ϕnt mtt 0
c¼ ¼ ∑ n ϕ Tnt ϕTn0 (17)
c0t c00 0 0 n¼1 M n ϕ n0 0 0

where ϕnt and ϕ n0 are the subvectors of the nth mode corresponding to the ut and u0 DOFs.
Multiplying out the matrices on the right side gives


N 2ζ n ωn
ctt ¼ mtt ∑ ϕ nt ϕnt mtt
T
(18a)
n¼1 M n

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
200 A. K. CHOPRA AND F. MCKENNA

c00 ¼ c0t ¼ ct0 ¼ 0 (18b)

It is clear that the damping matrix defined by Equation (10) is of the special form identified in [3]:
 
ctt 0
c¼ (19)
0 0

With such a damping matrix, the damping forces in the DOFs u0 with no mass will always be zero—
during both elastic and inelastic regimes of response. This conclusion, confirmed by the numerical
results presented in Figure 3b, is valid independent of the number of modes included in Equation
(10) and irrespective of values specified—no matter how small or large—for the modal damping ratios.
However, the implications of not including a particular term in the series of Equation (10) should be
recognized. Recall that the nth term in the summation is the contribution of the nth mode with its
damping ratio ζ n to the damping matrix. If this term is not included in the series, the resulting
damping matrix implies zero damping in the nth mode. Obviously, damping ratios should be
specified for all modes that are expected to contribute significantly to structural response, say the
first J modes, but deciding on this number of modes is difficult in the context of inelastic response.
Much of the existing understanding of modal contributions, as exemplified by statements such as
lower modes provide dominant response, comes from studies of global responses—such as floor
displacements and story drifts—in the linear range [14]. However, a larger number of modes will be
required to represent local deformations, such as plastic-hinge rotations.
Clearly, solution strategies used for numerical integration of equations of motion should be robust
enough to work in the presence of no damping in modes J + 1 to N. Unconditionally stable time
stepping procedures should be employed for such applications.

5.2. Early work in RUAUMOKO

The idea of using superposition of modal damping matrices to construct the damping matrix for
nonlinear RHA goes back to Chrisp’s master’s thesis [1] under the supervision of Athol Carr [2].
However, their motivation for introducing this damping model was different than presented in this
paper. Our goal was to conform to the special structure of the damping matrix given by Equation
(19), which is achieved by this model [Equation (10)], independent of the number of modes and
modal damping values. In contrast, their goal was to limit the damping ratio in the very high modes.
For this purpose, in 1982, Carr added two implementations of the damping matrix in his signature
software RUAUMOKO [16]; modal damping ratios are (i) constant at all frequencies, and (ii) a
trilinear function of frequency that is constant at ω lower than ω1 varies linearly between ω1 and ωJ
and then stays constant at ω higher than ωJ. It was demonstrated that these damping models greatly
reduce the spurious damping moments at joints [1, 2].
Unfortunately, the work of Chrisp and Carr did not find widespread adoption, perhaps because it
never appeared as a published paper in a prominent journal. As a result, researchers continued to
look for a way of reducing the spurious damping forces associated with the Rayleigh damping
model. We mention some of these contributions in a later section.

5.3. Earlier proposals for damping matrix of special form


Two ways of constructing a damping matrix that conforms to Equation (19) were proposed in 1994
[3]. The first solution was to use the Caughey series with l ≤ 0 terms in Equation (5). The second
proposal was to first construct the condensed damping matrix with reference to DOFs ut with
mass and then expand this c to the full set of DOFs with columns and rows of zeros
corresponding to DOFs that have no mass. The condensed damping matrix could be modeled by
Rayleigh damping with initial-stiffness-proportional damping and coefficients a0 and a 1 determined
in the usual manner. Neither of these two approaches seem to have been adopted widely, perhaps
because they result in a damping matrix that is fully populated, and because the condensed
stiffness matrix required in the second approach is not computed as part of the standard
computation in most general purpose computer programs.

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
MODAL DAMPING 201

6. INFLUENCE OF DAMPING MODEL ON RESPONSE: CONCENTRATED PLASTICITY

Concentrated plasticity is the most common approach to modeling structural elements typical of
multistory buildings. This model is based on the assumption that inelastic deformation will be
concentrated as plastic hinges at the two ends of the structural element, and the remainder of the
element remains linearly elastic (Appendix Figure A.1).
The earthquake response of a concentrated plasticity model of a building responding into the
inelastic range can be greatly influenced by how damping is modeled. Response results will be
presented for three damping models: (i) RI, Rayleigh damping based on initial stiffness, Equation (2);
(ii) RT, Rayleigh damping using tangent stiffness, Equation (16); and (iii) MODAL, superposition of
modal damping matrices with damping specified in all 20 modes of dominantly lateral vibration of
the 20-story building.
We first demonstrate that if the structure remains within the linear range, its response is essentially
unaffected by the damping model. Although the RI and RT models are identical for a linear system, the
MODAL damping model is different. Damping ratio of 2% was specified at the first and third modal
frequencies to determine a0 and a1 in Rayleigh damping, and the damping ratios in other modes are
defined by Equation (3), which will give lower damping in the second mode and higher damping in
modes 4 and above (Figure 1). In contrast, the damping ratios at all modal frequencies are 2% in the
MODAL model. These differences in modal damping values have essentially no effect on the
structural response to a scaled-down version of a GM record, as shown in Figures 5 and 6. The roof
displacement history (Figure 5) and the peak (maximum absolute) values over time of floor

Figure 5. Response history of roof displacement due to GM record scaled by SF = 0.1 for three damping
models: RI, RT, and MODAL.

(a) (b) (c)


20

16

12
Floor

4 RI
RT
MODAL
G
0 0.1 0.2 0 0.5 1 0 0.005 0.01
Displacement / height, % Story drift, % Plastic rotation, rad

Figure 6. Peak values of floor displacement, story drifts, and plastic-hinge rotations due to GM record scaled
by SF = 0.1 for three damping models: RI, RT, and MODAL.

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
202 A. K. CHOPRA AND F. MCKENNA

Figure 7. Response history of roof displacement for three damping models: RI, RT, and MODAL (with
damping specified in all 20 modes): (a) HM = 1000 and (b) HM = 100.

displacements, story drifts, and hinge rotations (Figure 6) are all unaffected by the damping model if
the structure does not yield.
The unscaled GM drives the building significantly into the inelastic range, and its response is greatly
influenced by the choice of damping model. This is evident from Figure 7, where the roof displacement
response of the building with its damping defined by three models and two values of initial stiffness of
the hinge (HM = 1000 and 100) is presented. We argue that the MODAL damping model with damping
specified in all modes—which, recall, also eliminates the spurious damping moments at all joints—
provides a benchmark against which results using other damping models may be evaluated. The
roof displacement of the building with the RI damping model oscillates about the zero-displacement
axis, presumably because of the much larger damping resisting force arising from the large initial
stiffness of the hinge. In contrast, significant permanent drift is observed in the case of the other two
damping models. As expected based on the preceding observations, the plastic rotations are greatly
underestimated—relative to the benchmark results—if the RI damping model is used. This is shown
in Figure 8 where the height-wise variation of the peak values of plastic rotations is plotted. The
larger the initial stiffness of hinge elements, the worse is this underestimation as is apparent by
comparing Figure 8a and b, corresponding to HM = 1000 and 100.
Use of the RT damping model leads to results that display the right trends, including lateral drift
away from the undeformed configuration. The response is still underestimated, but now to a much
smaller degree (Figures 7 and 8).

Figure 8. Peak values of plastic-hinge rotations—at each floor is the largest value among all hinges—for
three damping models: RI, RT, and MODAL (with damping specified in all 20 modes): (a) HM = 1000
and (b) HM = 100.

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
MODAL DAMPING 203

7. NUMBER OF MODES INCLUDED IN MODAL DAMPING

Superposition of modal damping matrices is the recommended damping model for nonlinear RHA of
buildings. As evident from Equation (10), the nth term in the summation is the contribution of the nth
mode with its damping ratio ζ n to the damping matrix c; if the first J terms are included, the resulting c
implies zero damping ratio in modes J + 1 to N.
Damping should be specified in how many modes? To answer this question, we compare the
responses of the structure with modal damping of 2% specified in the first J modes, with J = 1, 5,
10, and 20. The last case implies that the damping ratio is specified in all the lateral vibration modes
of this 20-story building, leading to the benchmark result against which results with damping
specified in fewer modes can be evaluated. Presented in Figures 9 and 10 is the roof displacement
history, and the peak values of floor displacements, interstory drifts, and hinge rotations plotted over
the height of the building. These results suggest that damping specified in the first 10 modes gives
essentially the benchmark results.
We recognize that the discrepancy in results for J = 1, 5, and 10 relative to J = 20 is simply because
modes J + 1 to N are undamped in the first three cases. This is the only source of discrepancy because,
as demonstrated earlier, the MODAL damping model eliminates spurious damping forces at joints
independent of the number of modes in which damping is specified.
Roof displacement / height, %

J=1
1 5
10
0
20

−1

−2 J=1

0 20 40 60 80 100
Time, sec

Figure 9. Response history of roof displacement for MODAL damping model with damping ratios specified
in the first J modes, J = 1, 5, 10, and 20.

(a) (b) (c)


20

16

12
Floor

J=1
4 5
10
20
G
0 1 2 3 0 2 4 6 0 0.01 0.02 0.03 0.04
Displacement / height, % Story drift, % Plastic rotation, rad

Figure 10. Peak values of floor displacements, story drifts, and plastic-hinge rotations for MODAL damping
model and damping ratios specified in the first J modes, J = 1, 5, 10, and 20.

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
204 A. K. CHOPRA AND F. MCKENNA

8. INFLUENCE OF DAMPING MODEL ON RESPONSE: DISTRIBUTED PLASTICITY

All the previous results were obtained by nonlinear RHA of the building using concentrated plasticity
model for joints—that is, concentrated plastic hinges at ends of beams and columns. These results
demonstrated that structural response is generally influenced significantly by the choice of the
damping model, and the Rayleigh damping model leads to unacceptable results. Another influencing
factor is the initial stiffness assumed for plastic hinges.
In contrast, the choice of damping model has much less influence on the response when the
structural elements are modeled as force-based elements [17] in OPENSEES, implying a fiber model,
allowing distributed plasticity instead of concentrated plastic hinges. This is demonstrated by the
response results for the three damping models—RI, RT, and MODAL—presented in Figures 11 and
12. The three models provide similar roof displacement histories as well as peak values of floor
displacements, story drifts, and curvatures of beam sections adjacent to the joints. The RI model
now displays the right trends of permanent lateral drift because it is no longer encumbered by the
large initial stiffness of a concentrated hinge.
Thus, there is no intrinsic problem with Rayleigh damping; the issue is really with the plasticity
model. This perspective—gleaned from the study of damping models—provides yet another reason
to employ the superior distributed plasticity models in nonlinear RHA.

Figure 11. Response history of roof displacement for distributed plasticity model of the building and three
damping models: RI, RT, and MODAL.

Figure 12. Peak values of floor displacements, story drifts, and section curvatures adjacent to the joint for
distributed plasticity model of the building and three damping models: RI, RT, and MODAL.

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
MODAL DAMPING 205

9. IMPLEMENTATION OF MODAL DAMPING IN OPENSEES

All structural response results presented in this paper were obtained using the OPENSEES software.
Perhaps the most widely used software in earthquake engineering research, OPENSEES had its
genesis in Frank McKenna’s PhD thesis [18] and can be downloaded from the PEER website.
Since its inception, OPENSEES has included the Rayleigh damping model and its variation based
on the tangent stiffness [Equation (16)]. The computational advantage of both these models lies in
the sparsity pattern (the location of the non-null coefficients) of the c matrix being the same as the
k matrix. This feature is important because the sparsity of the matrix equation is exploited in order
to reduce memory overhead and computational time when setting up and solving the equations of
motion.
OPENSEES has now been extended to include a damping matrix defined as the superposition of
modal damping matrices [Equation (10)]. Two ways to deal effectively with the resulting fully
populated damping matrix c have been implemented in OPENSEES. They are described next with
reference to the numerical steps involved in solving the transient equation [Equation (1)] using
Newmark’s integration method, a Newton–Raphson solution algorithm, and an iterative refinement
technique [19] for dealing with the damping terms:

1.0 Calculations for each time step i = 1, 2, 3…

1.1 j = 1
ð jÞ
1.2 uiþ1 ¼ ui
   
ð jÞ
1.3 viþ1 ¼ 1  βγ vi þ Δt 1  2βγ ai
 
ð jÞ
1.4 aiþ1 ¼  βΔt
1
vi  2β
1
 1 ai

2.0 For each iteration j = 1, 2, 3… until convergence:


ð jÞ ðjÞ ðjÞ ðjÞ
2.1 Riþ1 ¼ piþ1  ðf S Þiþ1 cviþ1  maiþ1
 ð j Þ  
^T ¼ ð Þ ð jÞ
þ γ 
βΔt2 m þ βΔt c
1
2.2 k iþ1
kT iþ1

 ð j Þ
^T
2.3 Solve k
ð jÞ
Δu¼Riþ1 ⇒Δu
iþ1

ð jþ1Þ ð jÞ
2.4 uiþ1 ¼ uiþ1 þ Δu
ð jþ1Þ ð jÞ
γ
2.5 viþ1 ¼ viþ1 þ βΔt Δu
ð jþ1Þ ð jÞ
2.6 aiþ1 ¼ aiþ1 þ βΔt
1
Δu
ð jÞ ð jÞ ð jÞ
where γ and β are the Newmark constants; the vectors ui , vi , and ai represent the displacements,
ð jÞ ð jÞ
velocities, and accelerations at the DOFs at iteration j in time step i; and ðf s Þi and ðkT Þi are the in-
ð jÞ ð jÞ ð jÞ
elastic resisting force vector and tangent matrix given ui , vi , and ai ; the matrix c* in the
 calculation

of the tangent (step 2.2) represents that part of the damping matrix c that is included in k ^ T ð jÞ .
iþ1
In OPENSEES, the c* matrix has the property that we shall describe as c* ⊆ c; that is, not all the
coefficients in the full c matrix are necessarily included in the c* matrix. It is the use of the
approximate c* in the calculation of the tangent (step 2.2) and the use of the exact damping force,
ð jÞ
cviþ1 , in the calculation of the residual (step 2.1) that combines the iterative refinement process with
the Newton–Raphson process. Iterative refinement is commonly used to obtain solutions to linear
systems of equations where the solution cannot be computed accurately because of ill-conditioned
problems or where the tangent matrix is not known or stored exactly, for example, in case of mixed
precision linear equation solvers [19]. As mentioned earlier, two options have been implemented in

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
206 A. K. CHOPRA AND F. MCKENNA

OPENSEES to deal with a fully populated c, and they differ in their choice of the c* matrix. The first
option, which is the same as that employed in PERFORM-3D [20], is to include in c* only those terms
 ð j Þ
of c that fall in the storage scheme of the k ^T matrix. In passing, we note that common storage
i
schemes for finite element analysis include Banded, Skyline, Compressed Row, and Compressed
Column. We shall denote this approximation of c as c* = cks. Note that should the OPENSEES user
specify a full general storage scheme, then there would be no approximation, that is, c* = cks = c. The
second option available in OPENSEES is c* = 0.
Both options can yield accurate results—accurate in relation to the results for c* = c—at the expense
of additional iterations per step: option 1 requiring on average one additional iteration per time step and
option 2 requiring two additional iterations. For the two finite element models—concentrated plasticity
and distributed plasticity—and for two GMs (se30 GM with SFs = 0.1 and 1.0), Table I shows the total
time, the number of time steps, the average number of iterations per step, and the error in peak value of
roof displacement. As illustrated in Table I, despite the additional iterations, the methods using the
‘approximate’ c* are faster to compute. This is because they can take advantage of the sparsity
 ð j Þ
pattern of the k ^T matrix. This speedup is more pronounced for the concentrated plasticity model
i
because this model has more DOFs and less element state determination overhead compared with
the distributed plasticity model. Note that the concentrated plasticity model requires considerably
more time to complete than the distributed model for the nonlinear case, SF = 1.0. This is because
the Newton–Raphson solution scheme struggles at times to find the solution in certain iteration steps
because of large changes in stiffness terms when the joints go from elastic to yielding and vice versa
that result in very different search directions. The consequence, as shown in Table I, is an increase
in the number of subdivisions of time steps to achieve convergence (shown by the larger number of
time steps), and an increase in the number of iterations per step (those that failed were also included
in the total count). The distributed plasticity model does not have such sudden and intense changes
in the search direction resulting in fewer subdivisions and average number of iterations per time step.
The fact that the second option, c* = 0, is faster than the first, c* = cks, for the larger concentrated
plasticity model, despite the additional iterations per step, shows the importance of efficiently
computing and adding the c* matrix.
In OPENSEES, the additional memory requirements associated with the damping model of Equation
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
(10) are the storage of J (=number of modes included) vectors, dn ¼ 2ζ n ωn =M n mϕ n , each of size
equal to the number of DOFs in the model. Using the dn vectors, computations in steps 2.1 and 2.2
are implemented using a series of vector–vector and vector–scalar operations without even forming
c or c*. Implementation of the previously described procedures in OPENSEES was verified against an
independent MATLAB code in which the full damping matrix is computed.
Before closing, we compare computational demands for the two damping models, RT and
MODAL. The RI model is excluded because its results are unacceptable, as documented earlier in
the paper. For the RT model and a GM scale factor of 1.0 (SF = 1.0), the computation time,
number of time steps, and number of iterations per time step are 31, 4026, and 2.9, respectively,

Table I. Performance of various modal damping options.


SF = 0.1 SF = 1.0

No. of No. of
Plasticity Computation No. of iterations/ % Computation No. of iterations/ %
Model c* time (s) time steps step error time (s) time steps step error

Distributed c 27 4000 1.0 0.00 47 4000 1.8 0.00


cks 23 4000 1.6 0.00 33 4000 2.0 0.00
0 24 4000 2.7 0.07 33 4000 3.1 0.00
Concentrated c 89 4000 1.0 0.00 1416 9178 8.1 0.00
cks 36 4000 1.6 0.00 331 10134 9.4 1.65
0 15 4000 2.7 0.00 121 9895 10.0 4.80

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
MODAL DAMPING 207

for the distributed plasticity model, and 869, 9460, and 150 for the concentrated plasticity model. In
the latter case, the large average number of iterations (150) per time step was required because
Modified Newton–Raphson steps utilizing the initial stiffness were necessary to achieve
convergence in many of the time steps. These data, when compared with the data in Table I,
demonstrate that the MODAL damping model as implemented is comparable in computational
time with the RT model for nonlinear RHAs that require numerical iterations.

10. ALTERNATIVE REMEDIES FOR RAYLEIGH DAMPING

The implausible results presented in Figure 2b arise, in part, because the yielding mechanism limits the
resisting forces fs, but no limiting mechanism exists in the Rayleigh damping model for damping forces
fD. Motivated by this observation, it was proposed that a bound be imposed on the magnitude of the
damping forces [4]. This remedy is not appealing because no objective basis exists for deciding on
this bound. As mentioned earlier, to limit the damping forces after yielding of the structure, several
researchers [5–8] have proposed replacing initial stiffness by the tangent stiffness in Rayleigh
damping [Equation (16)].
However, defining damping as proportional to the tangent stiffness matrix lacks a physical basis
and has conceptual implications that are troubling. In nonlinear RHA, the structural stiffness
matrix is reformulated at each time step from the element tangent stiffness matrices corresponding
to the present deformations and their path dependence—whether the element is on the initial
loading, unloading, or reloading branches of the element force–deformation relation—resulting in
the tangent stiffness matrix kT that enters into c [Equation (16)]. Consequently, (i) the damping
force–velocity relationship would exhibit hysteresis because the element force–deformation
relations are hysteretic, and (ii) damping can become negative when the structural stiffness at large
deformations with the effects of gravity loads included may become negative. Both of these
implications suggest that the use of the tangent stiffness matrix in defining Rayleigh damping is
fraught with conceptual issues. Furthermore, this approach seems to lack a physical basis to
construct a nonlinear, hysteretic damping matrix when the ‘measured’ values for modal damping
ratios come from structures vibrating in their linear range. Because of these conceptual limitations,
the use of the tangent stiffness model is not recommended, although it leads to improved response
results.
Another reason to stay away from damping proportional to tangent stiffness is that this model
contradicts the constant-damping assumption used since the inception of research in earthquake
dynamics of structures; by constant damping, we imply that c is a constant in f D ¼ c u. _ Fundamental
concepts and accumulated experience about inelastic response spectra and relationship between peak
deformations of inelastic and corresponding linear systems, which are now part of the foundation of
earthquake engineering, are all based on the constant-damping model.
Despite these limitations, Rayleigh damping based on the tangent stiffness matrix has continued to
attract the interest of researchers, with the goal of reducing spurious damping forces, resulting in
several proposals. One group of proposals is focused on modifications to a0 and a1 in Equation (16).
A number of researchers have suggested using time-varying coefficients a0(t) and a1(t) updated at
each time step to maintain modal damping ratios in the inelastic regime of response constant at the
values assigned at two frequencies in the linearly elastic range; this requires computing vibration
frequencies of the structure at each time step [5–7]. Other researchers [6, 8] have advocated keeping
a0 and a1 constant.
A second group of proposals is focused on modifying the initial stiffness matrix before it is used in
the stiffness-proportional part of the damping matrix. These proposals include the following: (i) assign
zero stiffness to structural elements that are initially very stiff and are likely to yield [8]; (ii) model each
structural element as a combination of an elastic element with initial-stiffness-proportional damping
and yielding rotational springs at the member ends that are undamped [9]. In the commercial
program, PERFORM-3D, for fiber-type reinforced concrete elements, only 15% of the total fiber area
stiffness is used when calculating the stiffness-proportional part of the damping matrix [20].

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
208 A. K. CHOPRA AND F. MCKENNA

These and other methods that modify the stiffness-proportional part of the damping matrix with
the goal of controlling spurious damping forces have not, as of now, coalesced into a consensus
solution. Some of them are also not amenable to convenient adoption in general purpose
computer programs, especially in commercial software. Both these aspects become apparent
from the extensive discussions among leading researchers [21] and explanations occupying
many pages in several documents: the user’s manual for RUAUMOKO [16], Chapter 18 in the
users’ guide for PERFORM-3D [20], and Chapter 2 in PEER/ATC guidelines for tall buildings [22].
We believe that many of the issues discussed in these forums become moot in light of the
results presented in this paper.

11. CONCLUSIONS

This investigation of modeling viscous damping in nonlinear RHA of building models with
concentrated plasticity (plastic hinge) at beam–column joints has led to the first four conclusions:
(1) Rayleigh damping may cause spurious damping forces during regimes of inelastic response, but
they disappear during regimes of elastic response. These spurious forces cause unbalance in the
moments at beam–column joints. Furthermore, with this damping model, the structural response
is unreasonably sensitive to the assumed value of the initial stiffness when plastic-hinge ele-
ments are employed.
(2) A damping matrix constructed by superposition of modal damping matrices—irrespective of the
number of modes included or values assigned to modal damping ratios—completely eliminates
the spurious damping forces. We recommend this damping model for nonlinear RHA of
buildings.
(3) The inelastic earthquake response of a concentrated plasticity model of a building can be
greatly affected by how damping is defined. This was demonstrated by response results
presented for three damping models: (i) RI, Rayleigh damping based on initial stiffness;
(ii) RT, Rayleigh damping using tangent stiffness; and (iii) MODAL, superposition of modal
damping matrices. The presented results demonstrate that the RI model may lead to
unacceptable results; in particular, it fails to reveal the permanent lateral drift and grossly un-
derestimates the plastic-hinge rotations. The RT damping model leads to much improved
results.
(4) However, Rayleigh damping with its second term proportional to the tangent stiffness matrix is
not recommended because it lacks a physical basis and has conceptual implications that are
troubling: hysteresis in damping force–velocity relationship and negative damping at large
displacements with gravity load effects. Furthermore, it contradicts the constant-damping model
that has been the basis for fundamental concepts and accumulated experience about inelastic
response of structures.
Additional conclusions are as follows:
When the element plasticity model is more sophisticated—in that a fiber model is used to represent
structural elements, allowing distributed plasticity—the structural response is not sensitive to the
damping model. Even the Rayleigh damping model leads to acceptable results. Thus, there is no
intrinsic problem with this damping model, provided that plasticity is modeled properly. This is yet
another reason to abandon concentrated plasticity models.
OPENSEES software has been extended to include a damping matrix defined as the superposition of
modal damping matrices. Although this model leads to a fully populated damping matrix, the software
implementation does not require the creation of a full dense tangent matrix. Numerical techniques such
as iterative refinement lead to accurate solutions without the need for even adding any damping
coefficients to the tangent matrix. The cost of these numerical techniques is an additional iteration or
two per time step, depending on how much of the damping matrix is included in the tangent matrix,
compared with a full dense matrix for elastic problems. The additional computational demands when
included as part of a nonlinear strategy, which require iterations, are minimal.

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
MODAL DAMPING 209

APPENDIX A

The system considered is a variant of the post-Northridge 20-story Seattle steel moment-resisting frame
building designed as part of the SAC Steel Project (Figure A.1). The building dimensions, section
properties, and loading are outlined in Appendix B of the SAC Steel report [23]. For this work, a num-
ber of 2D models were created and analyzed using the OPENSEES software application:
(1) Concentrated plasticity model. For this model, each beam and column element was modeled
as three independent members in series: zero-length rotational springs at each beam end and
an elastic member in between. As shown in Figure A.1(b), this model requires the addition
of two extra nodes per element. For the rotational springs, the moment–rotation curve was a
multi-linear hysteretic model created using the properties outlined in Section 3.2.2 of the
PEER/ATC-72 report [22]. As described therein, the hinge properties are typically modified
to ensure that the vibration periods of the structure modeled by the concentrated hinge
approach remain the same for the elastic case. Three different hinge modification factors
(HM) were employed 10, 100, and 1000 (1000 being the factor internally employed by
PERFORM-3D) [20]. The effect of these on hinge properties used to define the moment–
rotation response at the rotational hinges is described by Ibarra and Krawinkler [24]. The
rotational springs in this model contributed to the damping forces. This is the default mode
in the FEMA beam and plastic-hinge beam models in PERFORM-3D [20], although not in
OPENSEES.
(2) Distributed plasticity element model. For this model, a force-based nonlinear element was used
[17] for the beam and column elements. The properties of the force-based elements are com-
puted using a number of fiber-discretized sections located at the Gauss points along the element
length. For this model, a five-point Gauss–Lobatto integration was used, as shown in Figure A.1(c),
with each section along the length of these elements comprising 25 steel fibers.

Figure A.1. (a) Overall model with P-delta column to model internal gravity columns, (b) concentrated plas-
ticity element, and (c) distributed plasticity element.

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
210 A. K. CHOPRA AND F. MCKENNA

For both models, each column element was assigned a P-delta transformation to account for the
P-delta gravity load effects on the lateral frame. A gravity column, as shown in Figure A.1(a), was
added to account for the P-delta effects of the interior loading on the lateral-resisting frame. Mass
was applied to all nodes in the models based on tributary area, and all horizontal DOFs at a floor were
constrained to move together. In both models, the subterranean part of the model was neglected, and
the columns on the ground floor were fully constrained.
As a consequence of the modeling approaches, the two models had different numbers of DOFs: The
concentrated plasticity model had 720 DOFs, and the distributed plasticity model had only 300 DOFs.
Other model simplifications included creating only a 2D model, modeling the beams centerline to cen-
terline of the columns with no rigid offsets, no panel-zone deformation was considered, and the top and
bottom flange plates were not considered. The vibration periods for the concentrated plasticity models
were as follows: T1 = 3.79 s, T2 = 1.41 s, and T3 = 0.82 s for the model with HM = 1000; T1 = 3.81 s,
T2 = 1.42 s, and T3 = 0.82 s for the model with HM = 100; and T1 = 3.95 s, T2 = 1.48 s, and T3 = 0.86 s
for the model with HM = 10.

ACKNOWLEDGEMENTS
The authors are grateful to Professor Filip Filippou for implementing the MODAL damping model in his
FEDEASlab code in MATLAB, where the full damping matrix is computed. The implementation in
OPENSEES was validated against this MATLAB code. Comments by Professors Dionisio Bernal and Athol
Carr on a draft of the manuscript are appreciated as they led to improvements.

REFERENCES
1. Chrisp D. Damping models for inelastic structures. Master’s Thesis, University of Canterbury, Christchurch, New
Zealand, 1980.
2. Carr AJ. Damping models for inelastic analysis. Proceedings Asia-Pacific Vibration Conference, Kyonju, Korea,
1997; 42.
3. Bernal D. Viscous damping in inelastic structural response. Journal of Structural Engineering (ASCE) 1994;
120(4):1240–1254.
4. Hall JF. Problems encountered from the use (or misuse) of Rayleigh damping. Earthquake Engineering and
Structural Dynamics 2006; 35:525–545.
5. Leger P, Dussault S. Seismic-energy dissipation in MDOF structures. Journal of Structural Engineering (ASCE)
1992; 118(6):1251–1267.
6. Charney FA. Unintended consequences of modeling damping in structures. Journal of Structural Engineering
(ASCE) 2008; 134(4):581–592.
7. Jehel P, Leger P, Ibrahibegovic A. Initial versus tangent-stiffness based Rayleigh damping in inelastic time history
analysis. Earthquake Engineering and Structural Dynamics 2014; 43:476–484.
8. Petrini L, Maggi C, Priestley N, Calvi M. Experimental verification of viscous damping modeling for inelastic time
history analyses. Journal of Earthquake Engineering 2008; 12(1):125–145.
9. Zareian F, Medina RA. A practical method for proper modeling of structural damping in inelastic plane structural
systems. Computers and Structures 2010; 88:45–53.
10. Rayleigh L. Theory of Sound, 1. Dover: New York, NY, 1945.
11. Caughey TK. Classical normal modes in damped linear dynamic systems. Journal of Applied Mechanics (ASME)
1960; 27(2):269–271.
12. Caughey TK, O’Kelly MEJ. Classical normal modes in damped linear dynamic systems. Journal of Applied
Mechanics (ASME) 1965; 32(3):583–588.
13. Luco JE. A note on classical damping matrices. Earthquake Engineering and Structural Dynamics 2008; 37:
615–626.
14. Chopra AK. Dynamics of Structures: Theory and Applications to Earthquake Engineering (4th edn). Prentice Hall:
Englewood Cliffs, New Jersey, 2012.
15. Wilson EL, Penzien J. Evaluation of orthogonal damping matrices. International Journal for Numerical Methods in
Engineering 1972; 14:5–10.
16. Carr AJ. RUAUMOKO Manual, Vol 1: Theory and User’s Guide to Associated Programs. University of Canterbury:
Christchurch, New Zealand, 1982.
17. Spacone E, Filippou FC, Taucer F. Fiber beam–column model for nonlinear analysis of R/C frames: I.: formulation.
Earthquake Engineering and Structural Dynamics 1996; 25(7):711–725.
18. McKenna F. Object-oriented finite element programming: frameworks for analysis, algorithms, and parallel
computing. PhD thesis, Department of Civil Engineering, University of California, Berkeley, December, 1997.
19. Wilkonsin JH. Rounding Errors in Algebraic Processes. Prentice Hall: Englewood Cliffs, New Jersey, 1963.
20. CSI. PERFORM-3D User’s Guides, Version 5. Computers and Structures Inc., Walnut Creek, CA, 2011.

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe
MODAL DAMPING 211

21. Powell G. Damping models for nonlinear dynamic analysis: survey results, analysis and conclusions—including ad-
dendums 1–4, unpublished papers, April to November, 2008.
22. PEER. Modeling and acceptance criteria for seismic design and analysis of tall buildings. Technical Report PEER
2010/111 or PEER/ATC-72, Pacific Earthquake Engineering Research Center, University of California, October
2010.
23. FEMA. State of the Art Report on Systems Performance of Steel Moment Frames Subjected to Earthquake Ground
Shaking, FEMA-355C. Federal Emergency Management Agency: Washington, D.C., 2000.
24. Ibarra LF, Krawinkler H. Global collapse of frame structures under seismic excitations. John A. Blume Earthquake
Engineering Center Technical Report 152, Stanford Digital Repository, Stanford University, Stanford, CA, 2005.

Copyright © 2015 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2016; 45:193–211
DOI: 10.1002/eqe

You might also like