EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2006; 35:851866

Published online 14 February 2006 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/eqe.560

A simple displacement control technique for

pushover analyses
Armelle Anthoine;
ELSA; IPSC; Joint Research Centre; European Commission; I-21020 Ispra (VA); Italy

SUMMARY
This paper explains how to control in displacement any force proportional loading. Such a procedure
makes it possible to derive the complete (i.e. including the possible softening branch) response curve
of a structure along any radial loading path in the force space. This is exactly what is required in
the so-called pushover analysis used in the seismic assessment of structures. The proposed procedure
is simple in the sense that it can be easily implemented in any classical (displacement-based)
nite
element code through a standard displacement control loading process. Furthermore, it leads to an
interesting de
nition of the controlled degree-of-freedom, which, in the case of the pushover analysis,
could substitute the classical roof displacement. Copyright ? 2006 John Wiley & Sons, Ltd.
KEY WORDS:

proportional loading;
nite element method; pushover analysis

1. INTRODUCTION
In the pushover analysis, a non-linear
nite element model of a given structure (e.g. a building frame) subjected to gravity loads, is laterally loaded until either a prede
ned target
displacement is met, or the model collapses. The lateral load has a prescribed distribution
(e.g. uniform, inverse triangular, modal, etc.) and the control node is a particular point of
the structure (usually the centre of mass at the roof level). The result of the analysis can be
thus expressed in terms of a base shear versus control node displacement relationship, i.e. a
so-called pushover curve. Depending on the mechanical behaviour of the structure and also on
the controlled degree of freedom, this pushover curve may exhibit limit points and softening
branches or even snap-backs and bifurcation points.
Owing to the prescribed pattern of the lateral forces, a force control solution scheme is
straightforward but operates only along the ascending branch of the pushover curve. At a
limit point, the applied forces cannot anymore increase either because of second-order eects
Correspondence

to: Armelle Anthoine, ELSA, IPSC, Joint Research Centre, European Commission, I-21020
Ispra (VA), Italy.
E-mail: [email protected]

Copyright ? 2006 John Wiley & Sons, Ltd.

Received 11 May 2005

Revised 12 October 2005
Accepted 14 November 2005

852

A. ANTHOINE

(e.g. buckling, P-delta eect) or because a negative-sloped portion of a behaviour curve has
been reached somewhere in the structure. The lateral forces must therefore be decreased but the
structure should not undergo a simple elastic unloading. During this phase, the displacement
of the control node will either increase (softening branch) or decrease (snap-back). To follow
a softening branch, a displacement control technique should be used (e.g. Reference [1])
while only more powerful schemes (e.g. arc-length or work control methods) will be able to
follow also snap-back branches and detect bifurcation points. Actually, these latter schemes
have been developed appositely for non-linear analyses involving unstable responses due to
softening and=or buckling for example.
It is worth mentioning that force control strategies based on a partial=total unloading of
the structure (or a part of it) followed by a reloading with damaged characteristics, as the
methods proposed in SAP 2000 NL [2], are not always able to pass a limit point and, if
successful, do not necessarily produce the same result. In fact, the solution found beyond
the limit point depends on the whole path followed previously. Since each method follows
a dierent unloading=reloading path and not a monotonic push, the obtained solutions might
dier from one another and from the monotonic pushover curve. Furthermore, they usually
lead to a stepwise decreasing pushover curve, each vertical drop corresponding to a single unloadingreloading path aiming at bypassing a softening or a snap-back portion of the
monotonic pushover curve.
The simple displacement control method presented here is able to follow the monotonic
pushover curve also along a softening branch. It does not resort to any arti
ce such as the
addition of
ctitious springs as proposed by Archer [1] but simply relies on an appropriate
de
nition of the displacement variable to be used for controlling the loading process.

2. PRINCIPLE OF THE PROPOSED METHOD

The method will be introduced through a simple example, a cantilever beam subjected to
xed
gravity loads QG and increasing lateral load F with an inverse triangular pattern (Figure 1(a)).
The beam is composed of three elements of equal length and the same lumped mass m
F

mg

mg
F

mg

mg

2
F

3
3

mg

mg

(a)

(b)

Figure 1. (a) Pushover analysis on a cantilever beam; and (b) equivalent isostatic loading system.
Copyright ? 2006 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:851866

DISPLACEMENT CONTROL TECHNIQUE FOR PUSHOVER ANALYSES

853

is aected to the three nodes above the ground, numbered 1, 2 and 3 (the base node is
numbered 0). The inverse triangular pattern consists in horizontal nodal forces proportional
to the mass and to the perpendicular height of the node considered. In the case considered, if
F is the total base shear, this pattern gives F=6 at node 1, F=3 at node 2 and F=2 at node 3.
No assumption is made on the mechanical behaviour of the beams elements and thus, the
failure mode is not known and can be dierent from the one expected when the three beam
elements are homogeneous and mechanically identical (failure at the base).
The aim of the pushover analysis is to compute the response of the structure when the
lateral loads are increasing (ascending branch of the pushover curve) and, if a limit point is
reached, to pursue the computation until complete failure (descending branch of the pushover
curve). During the second phase, the lateral loads are decreasing but this is not a simple
unloading since the damage caused to the structure is still increasing.
Let us imagine how such a proportional loading could be imposed experimentally.
A possible solution consists in superposing two simply supported rigid beams (Figure 1(b)).
The
rst beam of length rests on nodes 1 and 2 of the structure while the second beam
of length 4=3 rests on node 3 of the structure and on the
rst beam at a distance 5=3
above the ground (node 4). Finally, the lateral force F is applied at the middle of the second
beam (node 5). This loading system is statically determinate and the forces transmitted to the
structure are exactly F=6 at node 1, F=3 at node 2 and F=2 at node 3. This simply results
from the equilibrium of each of the rigid beams. The applied force F gives rise to equal
forces F=2 at nodes 3 and 4. On its turn, the force F=2 transmitted at node 4, gives rise to a
force F=6 at node 1 and a force F=3 at node 2.
In this experimental set-up, F happens to be the reaction force associated to the horizontal
displacement u5 of node 5. Thus, if u5 , instead of F, is made increasing monotonously, the
pushover analysis can be pursued also beyond the possible limit point independently of the
failure mode, even in case of local mechanisms leading to partial unloading of the structure. In
other words, u5 is the suitable control displacement and the pushover curve should therefore
be expressed in terms of F versus u5 , and not in terms of F versus a particular horizontal
displacement (e.g. u3 ). A posteriori, the Fu5 pushover curve can always be expressed in
terms of any horizontal displacement by simply replacing the monotonic evolution of u5 by
the computed evolution of u3 (roof displacement), u1 or u2 .
In a
nite element code environment, this experimental set-up could be numerically reproduced as it is, together with the structure. The Fu5 pushover curve would thus result from a
standard displacement control loading process (imposed horizontal displacement of node 5).
However, the numerical modelling of many perfectly rigid beams may quickly become cumbersome. Mechanically speaking, each rigid beam is nothing else than a linear relationship
between the horizontal displacements of the three nodes involved. For the two rigid beams in
Figure 1(b), one has:
3u4 = u1 + 2u2
2u5 = u3 + u4

(1)

Eliminating u4 in Equation (1), one obtains:

u5 =
Copyright ? 2006 John Wiley & Sons, Ltd.

u1 u 2 u 3
+
+
6
3
2

(2)

Earthquake Engng Struct. Dyn. 2006; 35:851866

854

A. ANTHOINE

The quantity u5 happens to depend exclusively on the lateral load distribution, which appears
explicitly in Equation (2), and not on the chosen loading set-up, which is not unique. Here,
two rigid beams can be installed in three dierent ways, depending on which structural nodes
the
rst beam rests on. As a matter of fact, Equation (2) can be derived directly on the original
structure (Figure 1(a)), through the expression of the virtual work done by the external forces
(horizontal Hi and vertical Vi ) in any cinematically admissible displacement
eld, i.e.





u1
u2
u3
+
+
mg(
v1 +
v2 +
v3 )
(Hi
ui + Vi
vi ) = F
(3)

Wext =
6
3
2
i=03
where
ui (resp.
vi ) stands for the horizontal (resp. vertical) virtual displacement of node i
and g is the acceleration due to gravity. The gravity loads being constant, Equation (3) proves
that the structure is subjected to a one-parameter loading mode according to the terminology
introduced by Halphen and Salenon [3] and that F is actually the load parameter, whereas the
corresponding displacement parameter ueq is, by duality, the same quantity as in Equation (2),
i.e.
u1 u 2 u 3
+
(4)
ueq = +
6
3
2
Therefore, the loading set-up can be removed provided that the displacement control is shifted
from a single nodal displacement of the loading set-up (u5 ) to a linear combination of the
nodal displacements of the structure (u1 , u2 and u3 ). In practice, if the original force control
elastic problem reads
K:u = Fp + M:g

(5)

where p is the spatial distribution of force, M the mass matrix and F a scalar increasing
from 0, it can be replaced by the following displacement control problem:
K:u = M:g

submitted to pT :u = ueq

(6)

where ueq is now the scalar increasing from 0. General-purpose
nite element codes usually
oer the possibility of imposing linear constraints on nodal displacements such as Equation (4),
by resorting to dierent numerical methods (e.g. elimination, penalty, Lagrange multiplier).
Among them, the Lagrange multiplier method is one of the most ecient and is briey
recalled in Appendix A. Other methods may be found in Reference [4].

3. GENERALIZATION OF THE PROPOSED METHOD

The proposed method can be generalized to the case of any structure having N degrees
of freedom (DOFs) generically denoted (ui )i=1N . The term ui may stand for a 2D or 3D
translation or rotation DOF. The kinematical admissibility (boundary conditions) imposes M
linear constraints on the DOFs
(cki ui = hk )k=1M

(7)

which, in most cases, reduce to (uk = 0)k=1M . In the case of a pushover analysis, the loading
is composed of two parts, the
xed loads (Qi )i=1N (usually gravity loads) and the varying
Copyright ? 2006 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:851866

DISPLACEMENT CONTROL TECHNIQUE FOR PUSHOVER ANALYSES

855

loads (Fi )i=1N (lateral seismic equivalent loads), applying on the N
rst nodes (N 6N )
according a given pattern:
i = 1N ;

Fi = Fpi

(8)

In Equation (8), F is a scalar value that may coincide with the resultant load if the pattern
(pi )i=1N is normalized appositely. For example, in the case of the cantilever beam presented
earlier, the scalar F coincides with the
lateral resultant (and the base shear) because the
pattern (pi )i=13 = ( 16 ; 13 ; 12 ) is such that i=13 pi = 1.
The key-point is to express the virtual work of the external forces in any cinematically
admissible displacement
eld,
Wext . With the adopted notations, one obtains:







Wext =
(Fi
ui + Qi
ui ) = F
pi
ui +
Qi
ui
(9)
i=1N
i=1N
i=1N
Equation (9) proves that the structure is subjected to a one-parameter loading mode and that
F is the load parameter, whereas the corresponding displacement parameter ueq is, by duality,
the following linear combination:


pi ui
(10)
ueq =
i=1N
Therefore, the Fueq pushover curve can be obtained by increasing monotonically the linear
combination of the DOFs as shown on the right-hand side of Equation (10). The Lagrange
multiplier method described in Appendix A may be applied. However, one should keep in
mind that the corresponding computing time and=or memory might be substantially increased
if the force pattern involves most of the DOFs (e.g. in fully distributed loading). The number
of additional unknowns is only 1 (or 2 if the Lagrange multiplier is duplicated) but the
bandwidth of the augmented stiness matrix is at least N .
Again, it is possible to build a statically determined loading system able to impose
Equation (10). Generalizing what has been done in paragraph 2, (N 1) new DOFs can
be introduced through the following (N 1) relations:
uN +1 = p1 u1 + p2 u2
uN +i = uN +i1 + pi+1 ui+1

if 2 6 i 6 N 1

(11)

Here, each relation stands for a rigid link between the three DOFs involved (e.g. simply
supported rigid beam in the case of three parallel displacements) and the last added DOF
uN +N 1 coincides with the control DOF ueq de
ned in Equation (10). It is worth noting that
Equation (11) can be used as a numerical strategy for implementing Equation (10). Instead
of considering one linear relation involving N existing DOFs, one can introduce (N 1)
new DOFs and consider (N 1) relations involving only three DOFs at a time. With the
Lagrange multiplier method, the number of additional unknowns is 2(N 1) (or 3(N 1) if
the Lagrange multipliers are duplicated) but the bandwidth of the augmented stiness matrix
remains roughly unchanged.
Independently of the numerical method chosen for imposing Equation (10), it is important
to check the consistency of the problem to be solved, i.e. the orthogonality between the force
pattern and the boundary conditions. For example, any force (resp. moment) applied on a
blocked translation (resp. rotation) DOF should be removed. More generally speaking, the
Copyright ? 2006 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:851866

856

A. ANTHOINE

loading pattern should aect only free DOFs and=or free combinations of DOFs. In the force
control formulation, incompatibilities between imposed forces and imposed displacements are
not an issue because the displacement conditions naturally prevail on the force conditions.
Conversely, in the displacement control formulation, such incompatibilities cause the problem
to have no solution because the displacement constraint equation (10), which derives from
the loading pattern equation (8), is incompatible with the boundary conditions equation (7).
It is therefore fundamental to start from a well-de
ned problem, after having eliminating any
possible inconsistency.

4. PROPERTIES OF THE EQUIVALENT DISPLACEMENT ueq AND OF THE

ASSOCIATED CURVE Fueq
This de
nition of the displacement parameter ueq coincides with the energy-based displacement proposed by Hernandez-Montes et al. [5]. As a matter of fact, from Equations (9)
and (10), the incremental work done by the lateral force F is

WF = F
ueq

(12)

Equation (12) means that the area beneath the Fueq pushover curve coincides with the energy
provided to the structure by the lateral force. This last property has precisely been used by
Hernandez-Montes et al. to compute ueq from the results of a conventional pushover analysis,
through a step-by-step integration based on Equation (12). For being a
xed linear combination
of structural displacements, Equation (10) is undeniably a simpler way of computing ueq , but
its main advantage is to allow the pushover analysis to be controlled directly by ueq and thus
to obtain directly the Fueq curve.
Compared to other displacement control methods based on a particular displacement (e.g.
Reference [1]), the proposed method is much simpler since it relies on a standard displacement
control loading process where the displacement parameter is a linear combination of several
node displacements. In particular, it is potentially compatible with any numerical solver without requiring the addition of
ctitious springs.
However, the proposed method is still unable to follow a snap-back branch. To do so,
a more advanced numerical strategy is required (e.g. arc-length method). The question thus
arises as whether the Fueq curve exhibits more or less frequently snap-backs than a particular
Fui0 curve. Since the original Fueq curve is nothing else than a particular weighted average
of all possible pushover Fui curves, the following property can be easily established: If
the Fueq curve exhibits a snap-back, then at least one of the Fui curve does the same.
However, a snap-back in at least one Fui curve does not necessarily imply a snap-back in
the Fueq curve. Therefore all four combinations are possible: snap-back in Fueq but not in
Fui0 , snap-back in Fui0 but not in Fueq , snap-back in both or in none. The most frequent
combination depends on the structural typology and only an extensive numerical investigation
on a large class of structures could give a statistical answer.
A snap-back in the Fueq pushover curve may be given an interesting energetic interpretation. For a while, let us assume that the gravity loads are zero. Then, the energy WF provided
by the lateral load coincides with the sum of the stored elastic energy We and of the dissipated
Copyright ? 2006 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:851866

DISPLACEMENT CONTROL TECHNIQUE FOR PUSHOVER ANALYSES

857

energy Wd . Incrementally, this equality reads:

WF =
We +
Wd

(13)

Here, only
Wd should be positive (or zero) whereas no restriction holds on the other two
quantities. A snap-back is characterized by the fact that the release of elastic energy due to
the decrease of the lateral force is higher than the energy dissipated in the structure:

We 0

and

|
We |
Wd
WF 0

(14)

The energy provided by the lateral load is therefore negative, which means that the excess
of energy is restored. In particular, the extreme case of a perfectly brittle structure (
Wd = 0)
appears equivalent to a simple elastic unloading (
WF =
We 0). This last property is precisely at the basis of the
rst method (Unload Entire Structure) in SAP 2000 NL but holds
true only for drops of strength (strength degradation without dissipation). Snap-backs are inherent to the static character of the pushover curve and do not appear in a dynamic analysis
where the excess of released energy is mostly transformed into kinematical energy.
Now, if the gravity loads are also present, their incremental work
Wg should be added on
the left-hand side of Equation (13) or, equivalently, on the right-hand side with a negative
sign. Then,
Wg is the variation of the gravitational potential energy Wg , which can be
added to the elastic energy We , so as to form the total potential energy of the structure. Thus,
a snap-back is characterized by the fact that the release of the total potential energy due to
the decrease of the lateral force, i.e.
(We Wg ), is higher than the energy dissipated in
the structure. The gravity loads will therefore increase (resp. decrease) the snap-back eect if
their incremental work
Wg happens to be positive (resp. negative).
It is worth mentioning also the eventuality for the pushover curve to be early terminated
(without snap-back) for a non-zero lateral force because the gravity loads cannot be equilibrated anymore. Last but not least, the pushover curve may also exhibit bifurcation points.
These two latter cases as well as the softening and snap-back cases will now be illustrated
on an analytical example.

5. AN ILLUSTRATIVE EXAMPLE
The irregular frame sketched in Figure 2 is not intended to be realistic since it has been
designed appositely to exhibit all the possible diculties that can be encountered in real cases,
namely limit point, softening branch, snap-back, early termination and even bifurcation. Three
hinges (white disks in Figure 2) have been introduced to keep it as simple as possible. Two
equal masses m are lumped at the middle of the storey beams. Besides gravity, an inverse
triangular lateral load is considered. Therefore, the equivalent displacement corresponding to
the load pattern is simply:
ueq = uI =3 + 2uII =3

(15)

where uI (resp. uII ) is the horizontal displacement of the
rst (resp. second) oor. All beams
are linear elastic (same Young modulus E and inertia I ) until the bending moment reaches
Copyright ? 2006 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:851866

858

A. ANTHOINE

2F

mg

mg
1

Figure 2. Pushover analysis on an irregular frame.

a given limit 0 in some section i. Then a plastic hinge forms, which is characterized by a
linear isotropic softening
i = 0 (1 + i |ip |)

(16)

where i is the reduced yielding moment of the section, ip is the plastic rotation and i 0 is
a constant softening coecient, which depends on the section. A section that has completely
soften (i.e. i = 0) is thus replaced by a free hinge. Given the loading, any section coinciding
with a beam extremity and=or a point of load application, is one of the six potential plastic
hinges (numbered black points in Figure 2) that might form during the loading because the
bending moment varies linearly between these critical sections.
The structure is twice hyperstatic. If mi denotes the bending moment in the critical section i,
the elastic solution can be found as the minimum of the potential bending energy of the
structure We :
We = (2m21 + 2m1 m2 + 3m22 + m2 m3 + 2m23 + 3m24 + 5m25 + 4m5 m6 + 4m26 )l=12EI

(17)

under the following conditions of equilibrium:

Fl
3
4Fl
2m4 m5 + m6 =
3
mgl
m2 + 2m3 =
2
mgl
m4 m5 =
2

m1 + m2 m4 =

Copyright ? 2006 John Wiley & Sons, Ltd.

(18)

Earthquake Engng Struct. Dyn. 2006; 35:851866

DISPLACEMENT CONTROL TECHNIQUE FOR PUSHOVER ANALYSES

859

Each equilibrium condition is here derived from a particular virtual displacement
eld featuring
a frame or beam mechanism. The solution is
67
16
mgl Fl
728
39
16
237
mgl + Fl
m2 =
1820
65
8
673
mgl + Fl
m3 =
3640
65
21
809
mgl + Fl
m4 =
3640
65
1011
21
mgl Fl
m5 =
3640
65
71
607
mgl +
Fl
m6 =
3640
195

m1 =

(19)

When F is increasing (or decreasing) from 0, Equation (19) proves that the location of the

rst plastic hinge depends on the value of the ratio mgl=0 . In the following, the structure
will be assumed elastic for F = 0, which is true if mgl=0 3640
1011  3:60, and three particular
cases will be considered. In case 1 (F0 and mgl=0 = 0:25) the
rst plastic hinge appears
in section 1 while, in case 2 (F0 and mgl=0 = 3), it appears in section 4. Finally, in case 3
(F0 and mgl=0 = 24
53  0:453), two plastic hinges appear simultaneously in sections 1 and 6.
This example can be solved analytically: at each step, the elasto-plastic solution minimizes


the function G = We + 6i=1 mi ip under the equilibrium conditions (18) and the hypothesis of
yielding (|mi | = i and mi pi 0) or non-yielding (|mi |6i and pi = 0) in each section. More
generally, this method can be applied to any beam system in which plasticity is con
ned to
pre-selected critical sections and is characterized by a piece-wise linear constitutive law. The
reader is referred to Reference [6] who applied recent results of mathematical programming
to this particular class of problems.
5.1. Case 1: F0 and mgl=0 = 0:25
The elasto-plastic response is conditioned by the ratio l0 i =EI of the activated hinge(s).
Assuming l0 1 =EI = 0:85, l0 6 =EI = 1:5 and l0 2 =EI = 0:75, the pushover analysis
is composed of eight successive steps (Table I). Each step is characterized by a dierent
combination of damage (Di = 1 i =0 = i |ip |) and damage evolution (D i = i |pi |)
in the three activated hinges (1, 6 and 2) as shown in the last three columns of Table I:

D = 0 (and D = 0): no hinge.

0D1 and D0:

yielding hinge.

D = 1 (and D = 0): destroyed hinge.

0D1 and D = 0: elastic unloadingreloading of a damaged hinge.

Once the three hinges are destroyed, a free mechanism forms (Figure 3(a)) and the displacements are indeterminate (last row of Table I). Nonetheless, the equilibrium equations (18) can
Copyright ? 2006 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:851866

Copyright ? 2006 John Wiley & Sons, Ltd.

8937

+ d F()
3584
127 557

+ d F()
52 192
5 321 091

+ d F(
)
3 705 632
15 027

+ d F()
7456
10 107

+ d F()
5152
1893

+ d F()
3808
3
()
56

0 + d F(
)

F = lF=0

uII EI=l2 0
29
158
+
dF
3120
585
18 353
9038

dF
26 880
4815
15 185
38

dF
19 572
501
1 184 383
305
+
dF
1 389 612
648
9437
2015

dF
8388
1296
84 053
517
+
dF
69 552
3240
100 223
4
+
dF
102 816
81
1921
+ 2
2016

7
56
+
dF
3120
585
227
20 216

dF
960
4815
4973
98
+
d F
11 184
1503
100 297
49
+
dF
264 688
432
4973
2527

dF
11 184
864
3985
763

dF
6624
2160
10 957
14

dF
9792
27
259
+ 3
192
17
124
+
dF
3120
585
7177
12 764

dF
13 440
4815
52 097
130

dF
78 288
4509
11 581 301
2284
+
dF
16 675 344
7101
90 415
15 641

dF
100 656
7776
419 897
221

d F
417 312
19 440
37 117
34

dF
36 288
243
13 123
7
+ 
12 096
3

ueq EI=l2 0

Table I. Case 1 analysis.

uI EI=l 0
2

1904
dF
321
2193
7456
2193
7456
2193
595

dF
7456
144
1139
119

dF
2208
360

D1

161
dF
360
1069
7
d F
1632
9

497
dF
501
1

D2

D6

860
A. ANTHOINE

Earthquake Engng Struct. Dyn. 2006; 35:851866

861

DISPLACEMENT CONTROL TECHNIQUE FOR PUSHOVER ANALYSES

3
(a)
3

ueq
u II
uI
ueq (SAP )

Force

0
0.0

0.5

(b)

1.0

1.5

2.0

Displacement

Figure 3. Case 1: (a) failure mechanism; and (b) analytical pushover curve in terms of ueq , uII and uI
(continuous lines) compared with SAP 2000 NL Restart Using Secant Stiness (dotted line).

still be satis
ed provided that the lateral force does not change and is thus able to equilibrate
the gravity loads, but an unloading is impossible.
The displacements being normalized to 0 l2 =EI and the lateral force to 0 =l, the complete
Fueq pushover curve is shown in Figure 3(b) together with its two other possible forms
FuII and FuI (continuous lines). These curves do not begin at the origin because the chosen
reference state is the totally unloaded structure so that, for F = 0, the displacements are
those induced by the gravity loads. Furthermore, the
nal horizontal branches do not coincide
with the x-axis because the gravity loads are involved in the failure mechanism. Here, the work
of the gravity loads in the mechanism is opposed to the work of the lateral force. If
it were of the same sign, the
nal horizontal branch would fall below the x-axis. Only when
the work of the gravity loads in the failure mechanism is zero, does the
nal horizontal branch
of the pushover curve coincide with the x-axis.
The slope of the pushover curve depends on the ratio l0 i =EI of the active hinge(s) and
also on the chosen control DOF (ueq , uI or uII ). In the present case, the Fueq curve does not
Copyright ? 2006 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:851866

862

A. ANTHOINE

present any snap-back whereas the two other curves do. The proposed method is therefore
able to follow completely the pushover curve whereas a displacement control method based
either on uII or on uI fails at some stage. Of course, the proposed method would also fail
in case of a snap-back of the Fueq curve. By varying the softening ratios, it is possible to
derive cases where the best control DOF would be uI , other cases where it would be uII , but
if the hinges soften quickly enough, all pushover curves exhibits a snap-back. In particular
for l0 i =EI = , i.e. for perfectly brittle hinges, the snap-back branch coincides with the
elastic unloading branch.
For comparison purposes, the example has also been run with SAP 2000 NL. The numerical
solution turned out to depend not only on the member unloading method adopted but also
on the monitored DOF (uII or uI ) and on the control method (use of conjugate displacement
or not). The
rst method (Unload Entire Structure) always stops prematurely when the
second plastic hinge appears in section 6. The second method (Apply Local Redistribution)
is able to give a complete solution only if the conjugate displacement control is not used.
However, the solution then depends on the monitored DOF. Only the third method (Restart
Using Secant Stiness) gives the same and complete solution, independently of the monitored
DOF and on the control method. The solution is however quite dierent from the analytical
one (dotted line in Figure 3(b)). In particular, although the collapse mechanism is right, the
corresponding ultimate displacements are largely overestimated because the order of failure
of the three sections is incorrect. This is obviously due to the SAP 2000 NL third method,
which forces any hinge that has become non-linear to reach directly the bottom end of a
negatively sloping segment of its moment-rotation curve. Since the hinges are rigid plastic with
a unique softening segment down to zero, the method gives the same solution independently
of the i values. Unfortunately, the present example proves that such a solution is not always
conservative.

5.2. Case 2: F0 and mgl=0 = 3

Assuming l0 4 =EI = 0:2 and 5 0, the pushover analysis is composed of two successive
steps (see Table II) but is interrupted without snap-back when the second plastic hinge appears
in section 5. This is because the gravity loads cannot be equilibrated anymore. On the one
hand, as soon as yielding begins in section 5, d5p 0 and m5 0 . On the other hand,
owing to the previous yielding in section 4, m4 64 = 0 =2. Under these two conditions, the

Table II. Case 2 analysis.

n

F = lF=0

0 + d F()

2
3

1213

+ d F(
)
1176
353

1176

uI EI=l2 0

uII EI=l2 0

ueq EI=l2 0

7
56
+
dF
260
585
95
287

+
dF
756
1935
13

756

29
158
+
dF
260
585
221
2734

dF
1323
1935
1588

1323

17
124
+
dF
260
585
811
1727

dF
5292
1935
4265

5292

Copyright ? 2006 John Wiley & Sons, Ltd.

D4
0
147
dF
215
1
2

Earthquake Engng Struct. Dyn. 2006; 35:851866

863

DISPLACEMENT CONTROL TECHNIQUE FOR PUSHOVER ANALYSES

fourth equilibrium equation in Equation (18), which reads

m4 m5 =

30
2

(20)

cannot be veri
ed. This means that the pushover curve terminates at this point (third row
of Table II) and the reader will easily check that even an unloading is impossible. More
generally, the early termination of the pushover curve occurs when the work of the lateral
force in the failure mechanism is zero. Here, the failure mechanism is a beam mechanism at
the second storey (Figure 4(a)). The complete Fueq pushover curve is shown in Figure 4(b)
together with its two other possible forms FuII and FuI . Again, the proposed method is
able to follow completely the pushover curve, as a displacement control method based on
uII would do. However, nothing would indicate that the
nal point is not a snap-back point,
unless an advanced numerical strategy is used (e.g. arc-length).
For case 2, the
rst method of SAP 2000 NL gives the right answer. However, the program
sends the same warning message as in case 1 so that nothing indicates that no solution exists
beyond. The second method either stops prematurely or goes beyond the
nal point, which is

(a)

0
-0.2

Force

-0.4
-0.6
-0.8
-1
-1.2
-1.4

ueq
u II
uI

-1.2

-1

(b)

-0.8

-0.6

-0.4

-0.2

0.2

Displacement

Figure 4. Case 2: (a) failure mechanism; and (b) pushover curve in terms of ueq ; uII and uI .
Copyright ? 2006 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:851866

864

A. ANTHOINE

Table III. Case 3 analysis.

n

F = lF=0

0 + d F(
)

942

+ d F()
371
942

2b
+ d F()
371
942

2c
+ d F()
371
2a

uI EI=l2 0

uII EI=l2 0

ueq EI=l2 0

14
56
+
dF
3445
585
38
8344

dF
159
20 835
38
49
+
d F
159
1710
38
1043

dF
159
6030

58
158
+
dF
3445
585
782
458
+
d F
1113
20 835
782
419

dF
1113
855
782
1367

dF
1113
3015

34
124
+
dF
3445
585
610
2476

dF
1113
20 835
610
1627

dF
1113
5130
610
6511

dF
1113
18 090

D1

D6

1456
dF
1389

497
dF
285
1421

dF
1005

91
dF
201

ueq
uII
uI
ueq(SAP)

2.5

Force

2
1.5
1
0.5
0
0

0.2

0.4

0.6
Displacement

0.8

1.2

Figure 5. Case 3: bifurcation point of the analytical pushover curve expressed in terms
of ueq , uII and uI (continuous lines) and comparison with SAP 2000 NL Restart
Using Secant Stiness (dotted line).

obviously wrong. The third method gives an absurd answer, which was foreseeable because
the gravity load is preponderant.
5.3. Case 3: F0 and mgl=0 = 24
53
Assuming l0 1 =EI = 1:3 and l0 6 =EI = 1:1, three solutions are possible for the second
step of the pushover analysis, namely yielding in section 1 and elastic unloading in section 6,
yielding in section 6 and elastic unloading in section 1 or yielding in both sections 1 and 6.
The end of the elastic segment of the pushover curve is therefore a bifurcation point. The
three possible solutions are given in Table III and shown in Figure 5 (continuous lines).
For each hypothesis, the pushover curve has been computed only until the complete failure
of the activated plastic hinge. In this case, the proposed method will follow one among the
possible solutions that do not exhibit a snap-back if any, otherwise it will fail. If at least two
Copyright ? 2006 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:851866

DISPLACEMENT CONTROL TECHNIQUE FOR PUSHOVER ANALYSES

865

possible solutions do not exhibit a snap-back, the selection of the solution will probably be
random because of the limited numerical precision. In particular, the bifurcation will not be
detected.
For case 3, the two
rst methods of SAP 2000 NL stop prematurely at the bifurcation point,
independently of the monitored DOF and of the control method. As in case 1, only the third
method gives a unique and complete solution, which is again independent of the i values.
In particular, it diers from any of the three analytical pushover curves, at least immediately
after the bifurcation point (dotted line in Figure 5).

6. CONCLUSIONS
The proposed method is very simple and fully operational in the case of pushover curves
(monotonic or cyclic) exhibiting limit points and softening branches. It is based on an appropriate de
nition of the displacement variable ueq to be used for controlling the loading. The
proposed equivalent displacement ueq is related by duality to the base shear F in the sense
that, the energy WF provided to the structure by the lateral load is exactly the surface below
the Fueq pushover curve, i.e.
WF = F
ueq .
The method is however unable to follow snap-back branches of the Fueq curve and to
detect bifurcation points. Snap-backs and bifurcation are likely to occur as soon as the softening phenomenon is fast enough, which, in the given example, corresponds to low value
of the parameter . In particular, a sudden drop of strength (i.e.  = ) always implies
a snap-back. A possible solution would be to implement an arc-length method in which the
maximum increment of plastic strain would play a preponderant role in the measure of the
arc along the Fueq curve.
In its present state, the proposed method could therefore be added to the three methods
already available in SAP 2000 NL, as the
rst method to be tried.
Finally, the proposed method naturally suggests the use of ueq as the control displacement
variable to be used in seismic displacement-based design methods instead of the roof displacement. The implications of such a choice are nevertheless beyond the scope of this paper
and the reader is referred to Reference [5] who eectively used the Fueq pushover curve
(modal load) for seismic design.

APPENDIX A: THE LAGRANGE MULTIPLIER METHOD

To keep it as simple as possible, the method is presented in the linear elastic framework. For
the non-linear case, the reader is referred to Reference [4].
Assuming that the node displacements involved in the boundary conditions have been
eliminated, the force control pushover problem reads
K:u = M:g + Fp

(A1)

whereas the displacement control pushover problem takes the form

K:u = M:g
Copyright ? 2006 John Wiley & Sons, Ltd.

submitted to pT :u = ueq

(A2)

Earthquake Engng Struct. Dyn. 2006; 35:851866

866

A. ANTHOINE

The constraint in Equation (A2) may be appended to the classical potential function W with
a Lagrange multiplier  so that the solution (u; ) is the stationary point of
WL = 12 uT :K:u uT :M:g + (pT :u ueq )
Hence, the derivatives of WL with respect to u and  must vanish:

@WL

 

= K:u M:g + p = 0

K p
u
M:g
@u

@WL

ueq
pT 0 
T

= p :u ueq = 0
@

(A3)

(A4)

Compared to the force control system equation (A1), the displacement control system
equation (A4) has one additional unknown , which is the reaction force associated to the
constraint, i.e. F. This clearly results from comparing the original problem equation (A1)
to the
rst derivative in Equation (A4). However, if the original stiness matrix K is positive
de
nite, the augmented matrix is only symmetric. If this is incompatible with the numerical solver in use, an alternative solution consists in duplicating the Lagrange multiplier by
considering the function
WLL = 12 uT :K:u uT :M:g + 1 (pT :u ueq ) + 2 (pT :u ueq ) + 12 (1 2 )2

(A5)

and the associated system

T
p

pT

M:g

1
1 = ueq

(A6)

ueq

2

The augmented matrix is now positive de
nite. At the stationary point, the two additional
unknowns are obviously equal and their sum coincides with the reaction force F.
It is worth mentioning that the Lagrange multiplier method can also handle the displacement
boundary conditions, which are only particular restraints on the node displacements.
REFERENCES
1. Archer GC. A constant displacement iteration algorithm for nonlinear static push-over analyses. Electronic Journal
of Structural Engineering 2001; 2:120134.
2. Computers & Structures Incorporated (CSI). SAP 2000 NL, Berkeley, CA, U.S.A., 2000.
3. Halphen B, Salencon J. Elasto-plasticit
e. Presses des Ponts et chaussees: Paris, 1987.
4. Belytschko T, Liu WK, Moran B. Nonlinear Finite Elements for Continua and Structures. Wiley: England,
2000.
5. Hernandez-Montes E, Kwon OS, Aschheim MA. An energy-based formulation for
rst and multiple-mode nonlinear static (pushover) analyses. Journal of Earthquake Engineering 2004; 8(1):6988.
6. Cocchetti G, Maier G. Elasticplastic and limit-state analyses of frames with softening plastic-hinge models by
mathematical programming. International Journal of Solids and Structures 2003; 40:72197244.

Copyright ? 2006 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:851866