Undrained Stability of Braced Excavations in Clay

Boonchai Ukritchon, M.ASCE1; Andrew J. Whittle, M.ASCE2; and Scott W. Sloan, M.ASCE3

Abstract: Short-term undrained stability often controls the design of braced excavations in soft clays. This paper summarizes the
formulation of numerical limit analyses that compute rigorous upper and lower bounds on the exact stability number and include
anisotropic yielding, typical of K 0 -consolidated clays and bending failure of the wall. Calculations for braced cuts bound the actual failure
conditions within ⫾5%, and highlight limitations of existing basal stability equations. The analyses clarify how wall embedment and
bending capacity improve the stability of well braced excavations. Careful selection of mobilized strengths at shear strains in the range
0.6 –1.0% are necessary to match the predictions of anisotropic limit analyses with nonlinear finite-element predictions of failure for the
embedded walls. Two example applications from recent projects in Boston highlight the practicality of the numerical limit analyses for
modeling realistic soil profiles and lateral earth support systems, but also focus attention on the need for careful selection of undrained
strength parameters. Credible estimates of stability have also been obtained in reanalyzing a series of case studies reported in literature
using isotropic strength parameters derived from field vane or laboratory simple shear tests.
DOI: 10.1061/共ASCE兲1090-0241共2003兲129:8共738兲
CE Database subject headings: Braced excavation; Stability; Clays; Finite-element method.

Introduction
For deep excavations in soft clay, the design of the lateral earth
support system is often controlled by stability requirements. If the
factor of safety is below an acceptable level, expensive ground
modification schemes may be necessary to stabilize the soil below
the final excavated grade 共e.g., O’Rourke and O’Donnell 1997兲.
Hand calculations of basal stability are also used for preliminary
estimation of ground deformations 共after Mana and Clough 1981兲
and may form the basis for deciding whether or not to underpin
adjacent structures. In current practice, there are two methods
used to perform stability calculations for braced excavations: 共1兲
limit equilibrium methods; and 共2兲 nonlinear finite-element meth-
ods.
Limit equilibrium methods are widely used in design practice
and include separate calculations of basal stability 共based on fail-
ure mechanisms proposed by Terzaghi 1943; Bjerrum and Eide
1956兲 or overall slope stability 共using circular or noncircular arc
mechanisms兲 based on well established methods 共Bishop 1955;
Spencer 1967; Morgenstern and Price 1967兲. It is often difficult to
assess the accuracy of these solutions due to ad hoc assumptions:
共1兲 in selecting the shape of the failure surface; 共2兲 in the search
1
Assistant Professor, Dept. of Civil Engineering. Chulalongkorn
Univ., Bangkok, Thailand.
2
Professor, Dept. of Civil & Environmental Engineering, Massachu-
setts Institute of Technology, Cambridge, MA 02139. E-mail:
ajwhittl@mit.edu
3
Professor, Dept. of Civil, Surveying, and Environmental Engineer-
ing, The Univ. of Newcastle, NSW 2308, Australia.
Note. Discussion open until January 1, 2004. Separate discussions
must be submitted for individual papers. To extend the closing date by
one month, a written request must be filed with the ASCE Managing
Editor. The manuscript for this paper was submitted for review and pos-
sible publication on September 6, 2000; approved on November 6, 2002.
This paper is part of the Journal of Geotechnical and Geoenvironmental
Engineering, Vol. 129, No. 8, August 1, 2003. ©ASCE, ISSN 1090-
0241/2003/8-738 –755/$18.00.
procedures used to locate the critical surface; and 共3兲 the approxi-
mations used to solve equilibrium calculations 共i.e., interslice
force assumptions兲. Further complications arise in analyzing soil–
structure interactions for embedded support walls, tieback an-
chors, etc.
Nonlinear finite-element methods provide a comprehensive
framework that can evaluate multiple facets of excavation perfor-
mance ranging from the design of the wall and support system, to
the prediction of ground movements, and the effects of construc-
tion activities such as dewatering, ground improvement, etc. They
are indispensable for predicting the distribution of ground move-
ments caused by excavations, and for simulating processes where
there is partial drainage within the soil. Excavation stability can
be assessed by either exploring the depth required to generate
failure or by factoring the strength parameters of the soil 共e.g.,
Brinkgreve and Bakker 1991兲. However, finite-element methods
are rarely used as a primary source of information on stability, or
during the initial phases of design. This situation reflects the rela-
tive simplicity of limit equilibrium methods, the uncertainties in
site characterization, and the difficulties in selecting appropriate
constitutive models and input parameters for the finite-element
analyses.
An alternative method for calculating stability is through
upper and lower bound limit analyses. Rigorous upper and lower
bound collapse loads are solved numerically by linear program-
ming methods, while spatial discretization and interpolation of the
field variables 共i.e., lower bound stresses and upper bound veloci-
ties兲 are accomplished using finite-element methods. This ap-
proach combines the advantages of finite-element methods for
handling complex geometric and loading conditions, with the
power of the plastic limit theorems for bounding the exact col-
lapse load. The numerical limit analyses assume rigid perfectly
plastic soil behavior 共i.e., same input strength parameters as the
limit equilibrium methods兲, but do not require user-defined search
procedures to locate the critical failure mechanisms, and can
readily handle mechanisms involving combined failure of the
soil and structural elements. Thus, they are more flexible and

738 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003

Fig. 1. Conventional basal stability mechanisms and notation: 共a兲 Without wall embedment 共principal stress directions Clough and Hansen 1981兲;
and 共b兲 with wall embedment
operationally more efficient than the existing limit equilibrium
methods.
The numerical limit analyses described in this paper are based
on upper and lower bound formulations presented by Sloan and
Kleeman 共1995兲 and Sloan 共1988a兲, respectively. The application
of these methods for braced excavations is summarized briefly,
focusing on two new features 共Ukritchon 1998兲: 共1兲 implementa-
tion of structural elements; and 共2兲 representation of undrained
strength anisotropy. Upper and lower bound solutions provide an
independent method for evaluating the accuracy of existing basal
stability calculations. The main focus of the current analyses is
deep excavations in clay where the support wall does not extend
into an underlying bearing stratum. Recent nonlinear finite-
element analyses 共Hashash and Whittle 1996兲 provide reference
solutions to these problems by incorporating realistic modeling of
the anisotropic effective stress–strain–strength properties of typi-
cal clays 共Whittle et al. 1994兲. Comparisons with these finite-
element calculations provide guidance on the selection of aniso-
tropic strength parameters in the numerical limit analyses.
Parametric studies are then used to develop design charts for
evaluating the stabilizing effects of wall embedment. Practical
application of the method is demonstrated through case studies
using data published in literature.
Calculation of Basal Stability
Fig. 1 summarizes the failure mechanisms assumed in conven-
tional limit equilibrium calculations of basal stability after Terza-
ghi 共1943兲 and Bjerrum and Eide 共1956兲. For excavations in an
homogeneous clay, the stability of the excavation can be most
conveniently expressed in terms of the stability number, N
⫽␥H/s u , where ␥ and s u ⫽the average total unit weight and un-
drained shear strength in the retained soil. Table 1 shows that
there are subtle differences in these two basic solutions associated
with assumed values of the bearing capacity factor, N c , the loca-
tion of the vertical shear surface in the retained soil, and the
inclusion of shear tractions along this plane. Simple modifications
are widely used to account for the proximity of an underlying
bearing layer 关 d b below the excavated grade, Fig. 1共a兲兴 and for
contrasts in undrained shear strength above and below the exca-
vated grade (s uu , s ub ; Table 1兲. Clough and Hansen 共1981兲 pro-
JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003 / 739
posed further refinements to account for undrained strength an-
isotropy, a well known feature of soft clay behavior 共e.g., Ladd
1991兲, based on reference shear strengths measured at three ori-
entations of the major principal stress (s u0 , s u45 , and s u90 ; Table
1兲.
The effects of wall embedment are usually computed follow-
ing the approach proposed by Terzaghi 共1943兲, Fig. 1共b兲, assum-
ing that failure occurs below the base of the wall and is resisted
by the weight of the interior soil plug and adhesion ( f s ⫽␣s u )
acting along the plug–wall interface. A similar approach is used
by Eide et al. 共1972兲, Table 1, and both methods implicitly as-
sume that the wall is rigid 共i.e., does not yield兲. O’Rourke 共1993兲
assumes that wall embedment does not alter the basal failure
mechanism in the soil, but does contribute to the stability due to
the elastic strain energy stored in flexure. The resulting stability
numbers are functions of the yield moment and assumed bound-
ary conditions at the base of the wall 共Table 1兲.
Numerical Limit Analyses
Fig. 2 gives a schematic summary of the numerical limit analyses
used for braced excavations. The lower bound formulation gener-
ates a statically admissible stress field that satisfies equilibrium,
stress boundary conditions, and does not violate the yield criteria
of either the soil or structural components. In these plane strain
analyses, the soil mass is discretized into three-noded triangular
elements 关Fig. 2共b兲兴 assuming that the unknown stresses
(␴ x ,␴ y ,␶ xy ) vary linearly within each element
3 3 3
␴ x⫽ 兺 N i ␴ xi ;
i⫽1
␴y⫽ 兺 N i ␴ yi ;
i⫽1
␶ xy ⫽ 兺 N i ␶ xyi
i⫽1
(1)
where N i ⫽standard linear shape functions 共e.g., Zienkiewicz
1983兲; and (␴ xi ,␴ yi ,␶ xyi )⫽stress components at node i.
In contrast to conventional displacement-based finite-element
methods, each node is unique to a particular element such that
stress discontinuities can occur along shared edges between ele-
ments 关defined by pairs of nodes, e.g., 1, 4 and 2, 6; Fig. 2共a兲兴.
The wall is modeled using a series of two-noded beam and one-
noded joint elements, Fig. 2共c兲. Each node has two unknown
forces, F x , F y and one moment, F z . It should be noted that joint
Table 1. Stability Numbers from Limit Equilibrium Calculations of Basal Stability
Case Stability number, N Notes Reference

1
␥H
su
⫽Nc1⫹冑2
H
B 冉冊 Base case—homogeneous deep clay
N c1 ⫽5.7; d b ⭓B/ 冑2
Terzaghi 共1943兲

␥H Base case—homogeneous deep clay

2 ⫽Nc2 Bjerrum and Eide 共1956兲
su N c2 ⫽ f (H/B)⫽5.1⫺⬎7.5

3
␥H
su
⫽Nc1⫹冉冊
H
db
Effect of underlying bearing layer
d b ⬍B/ 冑2
Terzaghi 共1943兲

4
␥H
suu
⫽Nc1冉 冊 冑冉 冊
sub
suu
⫹ 2
H
B
Two clay layers:
s uu , s ub —strengths in retained soil and below excavated grade
Terzaghi 共1943兲

冉 冊 冑冉 冊
Anisotropic shear strengths in clay
␥H 1 s u0 ⫹s u90 H
5 ⫽N c⬘ ⫹ 2 s u0 , s u45 , s u90—strengths at ␦⫽0°, 45°, 90°, Fig. 1 Clough and Hansen 共1981兲
su45 2 s u45 B
N ⬘c ⫽ f (s u 关 ␦ 兴 )⫽4.0⫺⬎5.5

冑 冉 冊 冉冊
␥H H⫹D D
6 ⫽Nc1⫹ 2 ⫹2 Effect of embedment—rigid wall Terzaghi 共1943兲
su B B

7
␥H
su
⫽Nc2⫹2␣冉冊
D
B
Effect of embedment—rigid wall
␣—adhesion between interior soil plug and wall
Eide et al. 共1972兲

Effect of embedment—elastic strain energy for bending of wall:

␥H ␲2 M y M y —yield moment of wall;
8 ⫽Nc2⫹x O’Rourke 共1993兲
su D共D⫹h兲su h—excavation depth below lowest support 共Fig. 1b兲; and
x⫽1/8, 9/32, 1/2 for free, sliding, and fixed-end conditions
Note: Bearing capacity factors N c2 and N c⬘ obtained from Skempton 共1951兲 and Davis and Christian 共1971兲.

elements have no lateral dimension and the beam nodes are Lower Bound Formulation
unique to each particular element. The formulation allows beam
For plane strain analyses, the lower bound stress field must satisfy
elements to carry linear normal and shear tractions, while the joint
the following equations of equilibrium:
elements carry point forces, including moments. The bracing sys-
tem is represented by a series of rigid supports with pin-ended ⳵␴ x ⳵␶ xy ⳵␶ xy ⳵␴ y
connections to the wall 共i.e., allowing transmission of shear and ⫹ ⫽0; ⫹ ⫽␥ (3)
⳵x ⳵y ⳵x ⳵y
axial force only兲.
The upper bound formulation equates the power dissipated in a where ␥⫽total unit weight of the soil, and compressive stresses
kinematically admissible velocity field with the power expended are positive.
by the external loads. A kinematically admissible velocity field is Equilibrium of the nodal point stresses can be achieved within
one which satisfies the compatibility equations, velocity boundary each element by differentiating Eq. 共1兲 and substituting into Eq.
conditions, and the flow rule. During plastic flow, power is dissi- 共3兲, generating a set of linear equalities in the form:
pated by the plastic yielding of the soil mass, and by sliding along A1␴⫽B1 (4)
velocity discontinuities, where jumps in the normal and tangential
velocities can occur. The soil mass is discretized into three-noded where ␴⫽vector of stress components at each node in the mesh
triangular elements as shown in Fig. 2共b兲, assuming that the un- and the unit weight of soil.
known soil velocities (u, v ) vary linearly within each element The shear and normal tractions must also be in equilibrium at
all points along the discontinuities between adjacent soil ele-
3 3 ments. This condition can be satisfied by matching traction com-
u⫽ 兺 N iu i ;
i⫽1
v⫽ 兺 N iv i
i⫽1
(2) ponents (␴ n ,␶) at the nodal pairs 关e.g., 1, 4 and 2, 6 in Fig. 2共b兲兴,
generating a second set of constraints
where (u i , v i )⫽components of the nodal point velocities; and ␴ n1 ⫽␴ n4 and ␶ 1 ⫽␶ 4 (5a)
N i ⫽linear shape functions.
Velocity discontinuities can occur along shared edges between ␴ n2 ⫽␴ n6 and ␶ 2 ⫽␶ 6 (5b)
two adjacent elements, and are modeled by assuming that each For each interface, these relations can be rewritten as a series
node is unique to its element. Plastic deformations can occur both of equality constraints on the nodal point stresses by introducing
within the elements and along the velocity discontinuities be- standard transformation equations between surface tractions and
tween elements. The wall is modeled by two-noded rigid beam stress components (␴ x ,␴ y ,␶ xy ), Table 2. These same types of
elements and one-noded joint elements, as shown in Fig. 2共c兲. constraint can be used to represent all stress controlled boundary
Each node of the beam element has two unknown velocity com- conditions.
ponents (u, v ) that vary linearly along the element and are unique The present analyses assume that the undrained shear strength
to a given element. Each joint element, j, has an additional degree of K 0 -consolidated clays can be represented by the anisotropic
of freedom, the angular velocity w j 关Fig. 2共c兲兴, such that plastic yield criterion proposed by Davis and Christian 共DC兲 共1971兲,
yielding can occur as a hinge at the joint element. Fig. 3

740 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003

Fig. 2. Summary of numerical limit analyses for plane strain and structural elements: 共a兲 Problem discretization; 共b兲 plane elements of limit
analysis; 共c兲 structural element of limit analysis; and 共d兲 modeling for soil–structure interfaces

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003 / 741

Table 2. Linear Programming Forms for Solution of Numerical Lower and Upper Bounds
Lower bound method Upper bound method
Min.
Objective Function: Max. CT␴ Objective Function: (CT1 ␭˙ 1 ⫹CT2 ␭˙ 2 ⫹CT3 ␭˙ 3 ⫺CT41u1)
Subject to: Subject to:
Equilibrium within elements A1␴⫽B1 Linearized flow rule—soil elements A11u1⫹A12␭˙ 1 ⫽0
Equilibrium—stress discontinuities A2␴⫽B2 Flow rule—velocity discontinuities 共soil–soil兲 A21u1⫹A22␭˙ 2 ⫽0
Linearized DC yield criterion A3␴⭐B3 Kinematics—rigid beam elements
Equilibrium structural elements A4F⫽B4 Flow rule—structural elements A3u2⫽0
Equilibrium soil–structure interfaces A51F⫹A52␴⫽B5 Flow rule—velocity discontinuities A41u2⫹A42␭˙ 3 ⫽0
共soil structure兲
Yield of structural elements A6F⭐B6 Velocity boundary conditions A51u1⫹A52u2⫹A53␭˙ 2 ⫽0
Sign requirements A61u1⫹A62u2⫽B6
␭˙ ⭓0; ␭˙ ⭓0; ␭˙ ⭓0
1 2 3
Note: ␴—global vector of nodal stresses and body forces 共soil兲. F—global vector of nodal forces and moments 共structural elements兲. DC—Davis–
Christian 共1971兲 anisotropic yield function. u1, u2—global vectors of nodal velocities for soil and structural elements. ␭˙ 1 , ␭˙ 2 , ␭˙ 3 —global vectors of plastic
multipliers for triangular elements, velocity discontinuities 共soil-soil, soil-structure兲, and structural elements, respectively.

F⫽ 冉 共 ␴ y ⫺␴ x 兲 /2⫺h
a 冊 冉 冊
2
⫹
␶ xy
b
2
⫺1⫽0 (6)
A k ⫽cos共 2␲k/p 兲 ;
C k ⫽2 共 a/b 兲 sin共 2␲k/p 兲 ;
B k ⫽⫺cos共 2␲k/p 兲
(8b)
D k ⫽2a cos共 ␲/p 兲 ⫹2h cos共 2␲k/p 兲
where h⫽1/2关 s u0 ⫺s u90兴 ; a⫽1/2关 s u0 ⫹s u90兴 ; and b
where F k ⫽yield function for the kth side of the polygon.
⫽as u45 / 冑su0 su90.
Thus, the linearized yield surface introduces p constraints on
Eq. 共6兲 presents the DC yield criterion as a circular locus in the
the stresses at each nodal point, written in matrix form as shown
transformed stress space (X,Y )
in Table 2.
F⫽X 2 ⫹Y 2 ⫺R 2 ⫽0 (7a) Equilibrium of the structural elements is readily achieved by
balancing the forces and moments within each beam element and
using the variables:
between the joint and beam elements
X⫽
1 1
再共 ␴ ⫺␴ y 兲 ⫺h ,
a 2 x 冎 Y⫽
␶ xy
b
, R⫽1 (7b)
Fx j⫽ 兺 共 F x,i 兲 (9a)
In the lower bound formulation, this nonlinear function is ap-
proximated by an interior polygon, with p sides of equal length
关after Lysmer 共1970兲兴, such that statically admissible stress states
F y j⫽ 兺 共 F y,i 兲 (9b)
satisfy a series of linear inequality constraints
F k ⫽A k ␴ x ⫹B k ␴ y ⫹C k ␶ xy ⫺D k ⭐0, k⫽1,2,...,p (8a) Fz j⫽ 兺 共 F z,i 兲 (9c)

Fig. 3. Anisotropic yield criterion in numerical limit analyses: 共a兲 Anisotropic yield criterion Davis and Christian 共1971兲; and 共b兲 anisotropic
shear resistance along discontinuities 共parameters for BBC at OCR⫽1.0兲

742 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003

where (F x j ,F y j ,F z j ) and (F xi ,F yi ,F zi )⫽nodal forces and mo- to satisfy conditions of kinematic admissibility which require that
ments of the joint element, j, and adjacent beam elements, i. the plastic strain rates within each element must be compatible
The unknown forces and moments at each joint element must with the velocity field and satisfy the associated flow rule
also satisfy prescribed force and moment boundary conditions. p p
⳵u ˙ ⳵F ⳵F k
All of the equilibrium equations for beam and joint elements can
be combined to form a set of equality constraints shown in Table
␧˙ x ⫽⫺
⳵x
⫽␭ ⫽ ␭˙ 兺 ⫽ ␭˙ A
⳵␴ x k⫽1 k ⳵␴ x k⫽1 k k 兺 (15a)
2 共F is the vector of unknown forces and moments at all beam and
p p
joint nodes兲. ⳵ v ˙ ⳵F ⳵F k
Fig. 2共d兲 summarizes the modeling of interaction between a ␧˙ y ⫽⫺
⳵y
⫽␭ ⫽ ␭˙ 兺 ⫽ ␭˙ B
⳵␴ y k⫽1 k ⳵␴ y k⫽1 k k 兺 (15b)
beam element and two adjacent soil elements (L,R). The contact

冉 冊
tractions from these triangular elements are (␴ nL ,␶ L ) and p p
⳵u ⳵ v ⳵F ⳵F k
(␴ nR ,␶ R ) 共and vary linearly along the elements兲. The lower ␥˙ xy ⫽⫺ ⫹
⳵y ⳵x
⫽␭˙ ⫽ ␭˙ ⫽ ␭˙ C
⳵␶ xy k⫽1 k ⳵␶ xy k⫽1 k k 兺 兺
bound formulation requires that the net shear traction, ⌺␶⫽␶ L
⫺␶ R , and the net normal tractions, ⌺␴ n ⫽␴ nL ⫺␴ nR , are applied (15c)
to the adjacent beam element. These net tractions must also be in where ␧˙ x , ␧˙ y , ␥˙ xy ⫽plastic strain rates 共positive in compression兲;
equilibrium with the unknown nodal forces and moments of the and ␭˙ k ⫽non-negative plastic multiplier rates associated with F k .
beam element leading to a set of linear constraints By combining Eqs. 共2兲 and 共15兲, a kinematically admissible
A51F⫹A52␴⫽B5 (10) velocity field can be represented by a series of linear constraints
on the nodal point velocities and unknown plastic multiplier rates
Note that smooth interfaces are readily modeled by setting ␶ L written in a compact matrix form in Table 2 as
⫽␶ R ⫽0.
The current application for braced excavations assumes that A11u1⫹A12␭˙ 1 ⫽0 (16)
failure of the wall is governed by its plastic moment capacity,
M p , leading to the two linear inequalities that must be satisfied at where u1 and ␭˙ 1 ⫽global vectors of nodal point velocities and
the joints and at all points along the beam plastic multipliers for triangular elements.
The internal power dissipation in each triangular element can
F M 1 ⫽M ⫺M p ⭐0; F M 2 ⫽⫺M ⫺M p ⭐0 (11) also be computed from the strain rate in Eq. 共15兲

冕 冕兺
Linear variations in the normal contact traction from the adja- p
cent soil produce internal bending moments that vary as cubic W ele⫽ 共 ␴ x ␧˙ x ⫹␴ x ␧˙ y ⫹␶ xy ␥˙ xy 兲 dA⫽ 关 ␭˙ k 共 A k ␴ x ⫹B k ␴ y
functions along each beam element. Therefore, linearization of A A k⫽1
the yield criterion is achieved by subdividing the beam elements

冕兺
p p
into n-internal nodes and satisfying a set of inequality constraints
in the form ⫹C k ␶ xy 兲兴 dA⫽
A k⫽1
关 ␭˙ k D k 兴 dA⫽A 兺 关 ␭˙ k D k 兴 ⫽CT1 ␭˙ 1
k⫽1
A6F⭐B6 (12)
(17)
Ukritchon 共1998兲 gives full details of the equation assembly
where A⫽area of each triangular element.
for the structural elements, together with linearization methods
The upper bound formulation allows velocity discontinuities to
for yielding due to combined bending, axial and shear forces.
occur along shared edges between elements. These discontinuities
For braced excavations, the vertical pressure of the retained
are defined in terms of tangential and normal velocity jumps
soil, ␥H, is the driving force which causes failure of the excava-
(⌬u i j,t ,⌬u i j,n ) between nodal pairs (i, j), Fig. 2
tion. Thus, the objective function for the numerical lower bound
analysis is to minimize this vertical pressure, Min. 兵 ␥H 其 共or ⌬u i j,t ⫽ 共 u j ⫺u i 兲 cos ␪⫹ 共 v j ⫺ v i 兲 sin ␪ (18a)
Min. 兵 兺 ␥ i H i 其 , for a profile with i sublayers兲. The unit weight ␥ is
linked to the unknown nodal stresses through the equilibrium ⌬u i j,n ⫽⫺ 共 u j ⫺u i 兲 sin ␪⫹ 共 v j ⫺ v i 兲 cos ␪ (18b)
equations 关Eqs. 共3兲 and 共4兲兴. Hence, the same objective function where (u i ,u j ) and ( v i , v j )⫽nodal velocities at nodes i and j in
can be restated as the x and y directions, respectively; the sign convention for ␪ is
Max.兵 CT␴其 (13) shown in Fig. 4.
The shear resistance along velocity discontinuities can be
Table 2 summarizes the formulation of the numerical lower found by considering the associated flow directions. The aniso-
bound analyses in a standard linear programming form which can tropic DC yield criterion is first written in the local (n,t) coordi-
be solved efficiently by an active set algorithm 共Sloan 1988b兲. nate frame

Upper Bound Formulation F⫽ 冉 q tn cos 2␪⫺␶ tn sin 2␪⫺h

a 冊 冉
2
⫹
q tn sin 2␪⫹␶ tn cos 2␪
b 冊 2
⫺1
In order to ensure a strict upper bound on the true collapse load,
the DC yield function 关Eq. 共7兲兴 is approximated by an exterior ⫽0 (19)
polygon with p sides and p vertices where the shear stress components q tn ⫽(␴ tt ⫺␴ nn )/2 (␴ tt ,
F k ⫽A k ␴ x ⫹B k ␴ y ⫹C k ␶ xy ⫺D k ⫽0, k⫽1,2,...,p (14a) ␴ nn ⫽normal stresses in the t and n directions兲 and ␶ tn can be
found by standard frame rotation.
A k ⫽cos共 2␲k/p 兲 ; B k ⫽⫺cos共 2␲k/p 兲 Eq. 共19兲 represents an ellipse rotated counterclockwise
(14b)
through 2␪ with respect to the (q tn ,␶ tn ) frame. For undrained
C k ⫽2 共 a/b 兲 sin共 2␲k/p 兲 ; D k ⫽2a⫹2h cos共 2␲k/p 兲
analyses, the incompressibility condition with associated flow im-
This linearized form of the yield function can then be applied poses an additional constraint on the normal velocity jump

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003 / 743

 Fig. 4. Velocity discontinuities in upper bound analysis: 共a兲 Notation; and 共b兲 positive sign convention for slip (⌬ut)

⌬u i j,n ⫽␧˙ nn ⫽0 (20a) The internal dissipation of work along a velocity discontinuity
can then be computed as follows:
␧˙ nn ⫹␧˙ tt ⫽0; ␧˙ tt ⫽0
冕 冕
(20b) L L
W dis⫽ ␶ tn ⌬u t dt⫽ ␶ tn 共 ␭˙ 1 ⫺␭˙ 2 兲 dt
共 ␧˙ tt ⫺␧˙ nn 兲 ˙ ⳵F 0 0
⫽␭ ⫽0 (20c)

2 ⳵q tn
where ␧˙ tt , ␧˙ nn ⫽normal strain rates in the t and n directions,
respectively.
Eq. 共20c兲 represents a condition for defining the shear resis-
tance along a velocity discontinuity. By solving this equation and
substituting back in Eq. 共19兲, anisotropic yield introduces two
possible values of shear resistance along a given discontinuity
关Fig. 3共b兲兴: 共1兲 ␶ tn ⫽ 兩 ␶ 1 兩 , corresponds to the slip in the positive
direction, while 共2兲 ␶ tn ⫽⫺ 兩 ␶ 2 兩 , corresponds to slip in the nega-
tive direction, with sign convention shown in Fig. 4共b兲:
冕
L
⫽ 共 兩 ␶ 1 兩 ␭˙ 1 ⫹ 兩 ␶ 2 兩 ␭˙ 2 兲 dt⫽CT2 ␭˙ 2 (25)
0
where L⫽length of the discontinuity.
The upper bound formulation treats the wall using a combina-
tion of two-noded rigid beam and one-noded joint elements, Fig.
2共c兲. The beam nodes have two unknown velocity components
(u, v ) that are unique to each element, while the joints have three
degrees of freedom (u j , v j ,w j ), where w j is the angular velocity.
Velocities vary linearly along the beam elements, and plastic

冏 冏
bending failure can occur in the adjacent joints.
兩 ␶ 1 兩 ⫽ h sin 2␪⫺ 冑 a 2 ⫹b 2 ⫺a 2 cos 4␪⫹b 2 cos 4␪
2
(21a)
Rigid motions of a beam element 关e.g., with nodes 1 and 2,
Fig. 2共c兲兴, with general orientation ␪, are ensured by specifying
that there is no change in their longitudinal velocity 共see Table 2兲:
冏
兩 ␶ 2 兩 ⫽ h sin 2␪⫹ 冑 a 2 ⫹b 2 ⫺a 2 cos 4␪⫹b 2 cos 4␪
2
冏
(21b) 共 cos ␪u 1 ⫹sin ␪ v 1 兲 ⫺ 共 cos ␪u 2 ⫹sin ␪ v 2 兲 ⫽0 (26)

Thus, two yield envelopes associated with these states are while the angular velocity between nodes 1 and 2 can be com-
puted as
Envelope No. 1: F 1 ⫽␶ tn ⫺ 兩 ␶ 1 兩 ⫽0 (22a)
1
Envelope No. 2: F 2 ⫽⫺␶ tn ⫺ 兩 ␶ 2 兩 ⫽0 (22b) w 1 – 2 ⫽ 兵 ⫺u 2 sin ␪⫹ v 2 cos ␪⫹u 1 sin ␪⫺ v 1 cos ␪ 其 (27)
L
The flow directions of these yield envelopes are given by
where L⫽length of the beam element.
⳵F i ˙ In order to satisfy the conditions of kinematic admissibility,
⌬u t ⫽ 兺 ␭˙ i
⳵␶ tn
⫽␭1 ⫺␭˙ 2 (23a) the jump in velocities between beam–joint nodal pairs 关e.g., j, 2,
Fig. 2共c兲兴 must satisfy the associated flow rule
⌬u n ⫽0 (23b)
⌬u 2 j,t ⫽u 2t ⫺u jt ⫽0 (28a)
where ␭˙ 1 , ␭˙ 2 ⫽non-negative plastic multiplier rates associated
with the yield envelopes, F 1 and F 2 , respectively. ⌬u 2 j,n ⫽u 2n ⫺u jn ⫽0 (28b)
Eqs. 共18兲 and 共23兲 can be combined to form a series of linear
constraints, written in matrix form as 2
⳵F M k
A21u1⫹A22␭˙ 2 ⫽0 (24)
⌬w 2 j ⫽w 1⫺2 ⫺w j ⫽ 兺 ␭˙ k
k⫽1 ⳵M
(28c)

where ␭˙ 2 ⫽vector of plastic multipliers for velocity discontinuities where ⌬u 2 j,t , ⌬u 2 j,n , and ⌬w 2 j ⫽normal, tangential, and angular
共i.e., soil–soil and soil–structure interfaces兲. velocity jumps between the joint–beam nodal pair; F M k and ␭ k

744 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003

(⭓0)⫽yield criterion 关positive and negative plastic moments, Eq.
共11兲兴 and associated plastic multipliers that can occur at the joint
element.
Eqs. 共26兲–共28兲 can be expressed in a compact matrix form by
a set of linear equality constraints
A3u2⫽0
A41u2⫹A42␭˙ 3 ⫽0
where u2⫽vector of velocity components for the beam and joint
elements; and ␭˙ 3 ⫽vector of plastic multipliers associated with
bending failure of these elements.
The interfaces between structural and soil elements are treated
as velocity discontinuities, following the same assumptions de-
scribed between soil elements 共no restriction on sign of the tan-
gential velocity jump兲. Table 2 shows that kinematic admissibility
for these interfaces can be achieved through a set of constraints
analogous to Eq. 共24兲
A51u1⫹A52u2⫹A53␭˙ 2 ⫽0
where u1 and u2⫽vectors of velocity components for the soil and
structural elements, respectively; while ␭˙ 2 ⫽vector of plastic mul-
tipliers for soil–soil and soil–structure interfaces.
Finally, the plastic work due to hinge formation in the joint
elements can be computed from
W s ⫽M ⌬w 2 j ⫽M p 共 ␭˙ 1k ⫹␭˙ 2k 兲 ⫽CT3 ␭˙ 3
The total internal power, W i , can be computed by summing
the dissipation that occurs due to plastic deformation within each
of the soil elements, plastic shearing along each of the velocity
discontinuities 共soil–soil and soil–structure interfaces兲, and in the
formation of plastic hinges at joint elements
W i ⫽W ele⫹W dis⫹W s ⫽CT1 ␭˙ 1 ⫹CT2 ␭˙ 2 ⫹CT3 ␭˙ 3
External work, W e , is done by external surface tractions and
the 共vertical兲 gravitational body forces that can be defined at el-
emental and global levels as follows:
Element level: W e ⫽ 冕L
T i u i dL⫺ 冕
A
␥ v dA
Global level: W e ⫽CT41u1⫺␥C T42u 1
An upper bound on the true collapse load can then be obtained
by equating the total internal power dissipation and the external
work
⫺␥CT42u1⫽W i ⫺CT41u1
Stability problems for braced excavations 共no applied sur-
charge兲 can be solved by setting the objective function to mini-
mize the unit weight of the soil. This is accomplished by intro-
ducing the constraint:
CT42u1⫽⫺1
Hence, the objective function can be stated as Min.兵 W i
⫺CT41u1其 as shown in Table 2, while Eq. 共35兲 can be combined
with other velocity boundary conditions for soil and structural
elements to form a set of constraints on the nodal point velocities
JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003 / 745
A61u1⫹A62u2⫽B6 (36)
The solution of numerical upper bound analyses is solved by
using the active set algorithm to the standard linear programming
form listed in Table 2.
Fig. 5 shows typical finite-element meshes used for computing
(29a) upper and lower bounds on the stability of braced excavations
共connections between soil and structural elements are more
clearly shown in Fig. 2兲. The lower bound analyses 关Fig. 5共b兲兴 use
(29b)
a very high density of elements around the base of the wall where
rotations of principal stress directions are expected. Extension
elements are introduced in order to ensure statically admissible
solutions at all points within a deep clay layer 共half-space兲. These
elements require special forms of the constraint equations for
equilibrium and yield 共Sloan et al. 1990兲. The current analyses
use two limiting values of adhesion along the soil–wall inter-
faces, ␣⫽0 and 1 corresponding to smooth and rough walls, but
place no constraints on the normal tractions. The mesh for the
upper bound analysis 关Fig. 5共a兲兴 is based on a uniform density of
rectangular elements each subdivided into four triangular ele-
(30) ments. There is apparently little advantage in more complex geo-
metric arrangements, presumably due to the large number of ve-
locity discontinuities permitted in the analysis. The dimensions of
the discretized domain must only be sufficient to contain all po-
tential failure mechanisms, enabling reductions in the problem
size by truncating the far-field 共zero-velocity兲 boundaries, Fig.
5共a兲. Both upper and lower bound calculations assume a constant
value, p⫽24, for the yield surface linearization. Typical compu-
(31) tation times for the analyses reported in this paper range from
10–30 min of CPU on a DEC workstation 共Alpha 3000-300x兲.
Results
Homogeneous Isotropic Clay
Fig. 6 summarizes upper and lower bound predictions of the sta-
(32)
bility number, N⫽␥H/s u , for well braced cuts 共i.e., no failure
considered for bracing system兲 in a deep homogeneous deposit of
clay over a wide range of width to depth aspect ratios, B/H.
These analyses describe isotropic yielding of the clay using the
Tresca criterion 关i.e., h⫽0, a⫽b⫽0.5 in Eqs. 共6兲 and 共7兲兴, and
(33a)
include calculations for rough 共␣⫽1兲 and smooth 共␣⫽0兲 soil–
wall interfaces. In all cases, the numerical limit analyses are able
to bound the stability number within ⫾5%. The results show that
for narrow excavations (H/B⭓3), the interface roughness causes
(33b) a 5–10% increase in the predicted value of N, but has minimal
effect on the stability of wide excavations (B/H⭓5). Fig. 6 also
shows the stability numbers computed using the basal stability
equations of Terzaghi 共1943兲 and Bjerrum and Eide 共1956兲 共Table
1兲. The basal stability mechanism proposed by Terzaghi 共1943兲
(34) corresponds to an upper bound on the actual stability of wide
excavations, but significantly overestimates the stability of nar-
row excavations. In contrast, the assumptions used by Bjerrum
and Eide 共1956兲 are conservative for all excavation aspect ratios.
Differences between the numerical limit analyses and the conven-
tional basal stability calculations are not entirely surprising. In-
spection of the upper bound results shows that the failure mecha-
(35) nisms in a deep clay layer depend on both the excavation depth,
H, and its width 关in contrast, Fig. 1共a兲 shows mechanisms that
depend only on B兴, while increases in the interface adhesion also
increase the size of the failure zone in the retained soil.
Fig. 7 presents further comparisons for cases where a rigid
bedrock layer is located at a depth, d b /H⫽1.0, below the exca-
Fig. 5. Finite-element meshes used in upper and lower bound limit analyses of braced excavations: 共a兲 Typical upper bound mesh; and 共b兲 typical
lower bound mesh

vated grade. The numerical limit analyses show that the proximity
of the bedrock increases the stability number by up to 10–20%
for wide excavations (B/H⫽2 – 5), but has no effect on the sta-
bility of narrow excavations. These effects are readily explained
by the constraints of the bedrock on the extent of the failure zone
beneath the excavation. The lower bound solutions coincide with
the modified stability number computed by Terzaghi 共1943兲 for
B/H⭓ 冑 2 共case 3; Table 1兲.
Effects of Wall Embedment
Fig. 8 illustrates the effects of wall embedment for an excavation
with aspect ratio, B/H⫽2.66 in a deep homogeneous clay layer
共assuming full adhesion along the wall–soil interface, ␣⫽1兲. Nu-
merical limit analyses are shown for two characteristic embed-
ment depths, D/H⫽2/3 共‘‘shallow’’兲 and 2.0 共‘‘deep’’兲. In these
analyses, the failure of the embedded wall is controlled by the
plastic bending moment, M p . Dimensional analyses show that
the stability number of the excavation can be expressed as a func-
tion of the relative strength parameter, M p /(s u D 2 ), and the em-
bedment ratio, D/H. Upper and lower bound predictions cover a
wide range of the relative strength parameter and, in all cases,
bound the true stability number within ⫾5%. The results show
that the stability number increases with embedment ratio D/H
and with the relative strength parameter, M p /(s u D 2 ). The solu-
tions converge to a unique stability number, similar to the zero-

746 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003

 Fig. 6. Effect of wall adhesion and excavation aspect ratio on stability of braced vertical cuts in clay

embedment case 共Fig. 6兲, for highly flexible walls, and reach a
maximum value for M p /(s u D 2 )⭓4. This limiting condition cor-
responds to the case where there is no bending failure of the wall.
Fig. 8 shows that the upper bound predictions for these rigid walls
closely match the simple empirical equation proposed by Terzaghi
共1943兲 共case 6, Table 1兲. However, the limit analyses suggest
inaccuracies in the solutions proposed by O’Rourke 共1993兲 as
described in Table 1 共case 8兲, assuming input parameters x
⫽0.125 共free-end condition兲 and M p ⫽1.5M y 共rectangular wall
cross section兲 and N c ⫽5.7 共for selected B/H). This method does
not predict an increase in stability due to the embedment ratio,
D/H. It also underestimates the stability for low values of the
relative strength parameter and can greatly overestimate N for
large values of M p /(s u D 2 ). The solutions of Terzaghi 共case 6兲
represent an upper limit on N values considered in practice and,
hence, mitigate intrinsic limitations of this method.
Figs. 9 and 10 summarize the upper and lower bound predic-
tions for walls embedded with D/H⫽2 and three selected relative
strength parameters, M p /(s u D 2 )⫽0.05, 0.5, and 6.0. The upper
bound results for the flexible wall, M p /(s u D 2 )⫽0.05, in Fig. 9共a兲
show that the zone of plastic shear distortion within finite ele-
ments 共dark shading in Fig. 9兲 extends to a depth y/B⫽1.5 below
the excavated grade and laterally up to x/B⭐2.5 in the retained
soil. The embedded wall forms plastic hinges at two elevations,
one at the excavated grade and the second at a depth y/B⬇0.2. As
the relative strength increases to 0.5 关Fig. 9共b兲兴, only one plastic
hinge occurs at the excavated grade and there is a small increase
in the extent of the failure mechanism in the soil. For
M p /(s u D 2 )⫽6.0, the upper bound results show that the wall re-
mains rigid, and basal failure occurs with the flow of soil around
the toe of the wall, causing upward displacement of a soil plug
关Fig. 9共c兲兴. As a result, the plastic zone extends to a depth y/B
⬇2.5 and laterally to x/B⬇3.0. It is interesting to note that much
of the retained soil translates as a rigid body 共except in the region
close to the wall itself兲 while most of the plastic shear distortion
occurs below the toe of the wall.
Figs. 10共a and b兲 summarize the lateral earth pressures,
␴ h /s u , and bending moment diagrams for the wall obtained from
the lower bound analyses. For M p /(s u D 2 )⫽0.05, the lower
bound analysis predicts that there are two elevations where bend-
ing failure occurs 关i.e., 兩 M /M p 兩 ⫽1.0, Fig. 10共b兲兴, the first at the
excavated grade and the second at a depth y/B⬇0.2. For the

Fig. 7. Effect of bedrock on stability of braced cuts with no wall embedment

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003 / 747

 Fig. 8. Effects of wall embedment on stability of braced excavations in deep clay layer

intermediate case (M p /(s u D 2 )⫽0.5), bending failure occurs at
the excavated grade only, while there is no failure for the case
with M p /(s u D 2 )⫽6.0 (M max /M p⬇0.7). These results are consis-
tent with the locations of plastic hinges computed in the corre-
sponding upper bound analyses 关Figs. 9共a–c兲兴. Bending failure of
the embedded wall causes a large reduction in the lateral pres-
sures transmitted from the retained soil 关Fig. 10共a兲兴, but has little
effect on the resisting pressures which are similar in magnitude to
conventional Rankine passive pressures 共i.e., assuming ␴ hp
⫽ 关 ␥y⫹2s u 兴 where y is the depth below the excavated grade兲.
Development of Design Charts
The principal advantage of the proposed numerical analyses lies
in their ability to compute stability for more complex layered soil
profiles and lateral earth support systems. The computational ef-
ficiency of the technique also makes it a very attractive method
for developing design charts. For example, the preceding calcu-
lations consider the effects of wall embedment in a deep
homogeneous clay layer. There are three dimensionless param-
eters that govern the stability number for this situation: The
aspect and embedment ratios and the relative strength parameter
共full adhesion is assumed along the wall–soil interfaces兲. Upper
and lower bound analyses have been performed over wide
ranges of these parameters, H/B⫽0.2– 3.0, D/H⫽0 – 2, and
M p /(s u D 2 )⫽0.001– 6 共i.e., all cases involving bending
failure兲. In all cases, the numerical solutions bound the
exact stability number within ⫾5% 共and in many cases are
more accurate than this兲. The average of the upper and lower
bound solutions provides the best estimate of the stability
number and can be interpreted using empirical curve fitting
techniques based on dimensional analyses. In these
examples, the resulting empirical equations require seven
coefficients:

␥H
共 2⫹␲ 兲 冉冊
B
H
⫹b 2 ⫹b 3 冑 冉 冊 冉 冊 冉 冊 再 冉 冊冉 冊 冑 冎
4 Mp
s uD 2
D
H
⫹
s uD
Mp
2
D
H
2
c2
D
H
H
B
2
⫹c 3
共 D/H 兲
共 H/B 兲
⫽

冑 冉 冊 再 冉 冊 冑冉 冊 冎
(37)
su B Mp D D D
⫹a 2 ⫹ a3 ⫹a 4
H s uD 2 H H H

Although Eq. 共37兲 is rather cumbersome, Fig. 11 shows that it produces an excellent fit to the numerical data and reduces to a simple
format for the zero-embedment case (D/H⫽0):

␥H 共 2⫹␲ 兲共 B/H 兲 ⫹b 2
⫽ (37a)
su 共 B/H 兲 ⫹a 2

where b 2 ⫽6.10 and a 2 ⫽0.61.

748 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003

Fig. 9. Upper bound failure mechanisms predicted for walls with deep embedment, D/H⫽2 in homogeneous, isotropic clay 共with H/B
⫽0.375) 共arrows indicate velocity vectors, shaded region indicates plastic distortion within elements兲

Role of Anisotropy
The preceding results are based on the assumption that the un-
drained shear strength of the clay is isotropic. In reality, most soft
clays exhibit undrained strength anisotropy due to their
K 0 -consolidation history. For example, Ladd 共1991兲 reports mea-
surements of undrained strength ratios measured in laboratory
triaxial 共compression and extension; s uTC , s uTE) and direct simple
shear tests (s uDSS⫽␶ h max) on a wide range of K 0 -normally con-
solidated clays. The results show that typical strength ratios
s uTE /s uTC⫽0.50 to 0.75 and s uDSS /s uTC⫽0.65 to 0.85. More de-
tailed measurements of undrained strength anisotropy can be ob-
tained using sophisticated laboratory shear devices such as the
torsional shear hollow cylinder or directional shear cell 共DSC兲.
Fig. 12 shows data from a series of five DSC undrained plane
strain shear tests on K 0 -normally consolidated Boston blue clay
关共BBC兲, Whittle et al. 1994兴. The specimens were first consoli-
dated under K 0 conditions (K 0 ⫽0.53; ␴ ⬘v c is the consolidation
vertical effective stress兲 and then sheared by imposing principal
stress increments oriented at angle, ␦ inc , to the original direction
of deposition. Tests performed at ␦ inc⫽0 and 90° correspond to
the undrained plane strain active and passive modes of shearing,
while intermediate values of ␦ inc cause continuous rotation of
principal stress directions. The data are presented in a normalized
shear stress space (␶,q⫽0.5兵 ␴ ⬘v ⫺␴ h⬘ 其 ) such that the undrained
shear strength, s u ⫽max兵冑q 2 ⫹␶ 2 其 . The undrained strength ratios
range from a maximum, s u0 /␴ ⬘v c ⫽0.34, in plane strain active
shearing to a minimum, s u90 /␴ ⬘v c ⫽0.17, in the plane strain pas-
sive mode 共␦⫽90°兲. The peak shear strength in the active mode
occurs at small shear strains (␥⫽ 关 ␧˙ 1 ⫺␧˙ 3 兴 ⬇0.6%), while rela-
tively large strains are required to mobilize the passive resistance
共␥⬎5%兲. Although the locus of peak strength conditions is very
well described by the three-parameter DC yield function 共Fig.
12兲, it is not clear how the variations in strength mobilization with
shear direction should be considered in the numerical limit analy-
ses. One possibility is to consider the mobilized strengths at a
selected level of shear strain 关similar to the ‘‘strain compatibility

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003 / 749

Fig. 10. Summary of lower bound predictions for deep wall embedment, D/H⫽2 in homogeneous, isotropic clay 共with H/B⫽0.375): 共a兲
Contact normal traction, ␴ h /s u ; and 共b兲 moment diagram, M /M p

approach’’ proposed by Ladd 共1991兲兴. For example, Fig. 12 shows
the measured data and corresponding best-fit DC yield loci
corresponding to shear strain levels, ␥⫽0.6 and 1.0% 共see also
Seah 1990兲. Finally, one should note the large difference between
the three anisotropic yield loci in Fig. 12 and the isotropic yield
surface selected as the ‘‘average’’ undrained strength ratio mea-
sured in laboratory direct simple shear tests, s uDSS /␴ ⬘v c ⫽0.21.
to describe the anisotropic stress–strain–strength properties of
the clay 共BBC parameters were presented by Whittle et al. 1994兲.
Their calculations consider an excavation with width, B⫽40 m,
in a deep layer of clay. The excavation is supported by a
heavily reinforced 0.9 m thick diaphragm wall with total depth,
L(⫽H⫹D)⫽12.5 m– 60 m and braced by rigid struts. The cur-

Comparisons With Nonlinear Finite-Element Analyses

One method for evaluating the selection of anisotropic shear
strengths in the numerical limit analyses is through comparison
with more comprehensive nonlinear finite-element analyses. For
example, Hashash and Whittle 共1996兲 have presented results of
finite-element analyses for idealized braced excavations, using the
MIT-E3 effective stress soil model 共Whittle and Kavvadas 1994兲

Fig. 12. Anisotropic yield of K 0 normally consolidated Boston Blue

Fig. 11. Parametric representation of embedment depth effects Clay in undrained plane strain shearing

750 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003

 Fig. 13. Nonlinear finite-element analyses of idealized excavation using the MIT-E3 model 共Hashash and Whittle 1996兲

rent comparisons focus on finite-element calculations where: 共1兲
the struts are installed at grade with a vertical spacing, h
⫽2.5 m; and 共2兲 the soil profile is characterized by a constant
overconsolidation ratio of 1.0, such that the undrained shear
strength varies linearly with depth as shown in Fig. 13 共i.e., the
undrained strength ratio is constant for a given direction of load-
ing, ␦, and mobilized shear strain, ␥兲. In the original finite-
element analyses, the excavation was performed as a series of
discrete 2.5 m steps and the diaphragm wall modeled as an elastic
material. Failure was defined when there was inadequate numeri-
cal convergence 共due to large incremental soil deformations兲
within a specified excavation step, and was refined by considering
the maximum bending moment in the wall. Given these approxi-
mations, the failure depths, H f , can only be reported to a vertical
resolution of 2.5 m.
Numerical limit analyses of these same idealized excavations
have assumed that the diaphragm wall is heavily reinforced with
M p ⫽2.0 MNm/m. Upper and lower bound predictions of H f have
been performed for four undrained strength profiles: 共1兲 peak an-
isotropic shear strengths; 共2兲 anisotropic strengths mobilized at
␥⫽0.6%; 共3兲 anisotropic strengths mobilized at ␥⫽1%; and 共4兲
isotropic strengths based on undrained strength ratios measured in
direct simple shear 共DDS兲 tests (s uDSS /␴ v c ⫽0.21; Fig. 12兲. Fig.
14 summarizes the computed failure depths as functions of the
wall length, L. For the shorter walls (L⬍20 m), there is very
good matching between the finite-element results and the numeri-
cal limit analyses based on strength mobilized at ␥⫽0.6%. How-
ever, for the longer walls, the same assumption leads to an under-
estimate of the finite-element failure depths, and better agreement
is achieved at a mobilized strain ␥⫽1.0%. This behavior is con-
sistent with characteristic shear strains reported in the finite-
element analyses by Hashash 共1992兲 since larger shear strains are

Fig. 14. Comparison of numerical limit analyses and finite-element predictions of failure depths as functions of diaphragm wall length in
normally consolidated Boston blue clay

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003 / 751

Table 3. Numerical Limit Analyses of Basal Stability for Braced Excavations Published in the Literature
Davidson Ave. HDR-4 Chicago
Vaterland
Parameters No. 3 Mexico City No. 1 No. 2 CD 152 CD 162 Taipei
H (m) 9.0 9.0 9.1 7.3 7.9 12.2 13.45
B (m) 11.0 7.0 7.6 7.6 12.2 12.2 25.8
L (m) 11.0 18.2 13.7 13.7 19.2 19.2 24.0
d b (m) 15.0 — — — 14.1 9.8 —
Wall: SPW SPW SPW SPW SPW SPW DW
PZ32 Z section PZ32 PZ32 PZ40 PZ40 0.9 m
M p (MNm/m) 0.56 0.45 0.64 0.64 0.73 0.73 1.50
Factor of safety
共lower bound
to upper bound兲
Isotropic 1.27–1.38 1.20–1.31 1.15–1.12 0.88 –0.96 1.09–1.18 1.12–1.22 0.95–1.03
Anisotropic 1.12–1.21 1.13–1.23 1.24 –1.32
Source for s u : Field vane CK 0 UDSS UU UU Field vane Field vane
共Covarrubias 1994兲 CK 0 UC/E CK 0 UC/E
共Finno and Chung 1992兲
Observed behavior No failure No failure No failure Failure No failure No failure Failure
␦ wall (cm) 10–13 cm 10–15 cm 10–25 cm 15–17 cm 15–17 cm
References Norwegian Technical Institute Rodriguez and Flamand Clough and Reed Finno et al. Su et al.
共1962兲 共1969兲 共1984兲 共1989兲 共1998兲

mobilized prior to failure as wall length increases. Fig. 14 also
shows that calculations of failure depths based on the peak aniso-
tropic strength envelope grossly overestimate the finite-element
predictions. The isotropic limit analyses slightly overestimate the
reference values of H f from these finite-element analyses. How-
ever, this happy coincidence does not substantiate the general
application of isotropic strength approximations!
Fig. 14 includes predictions of H f based on the empirical limit
equilibrium calculations proposed by Terzaghi 共1943兲. These cal-
culations use depth-averaged shear strengths above and below the
excavated grade, for reference undrained strength ratios in DSS
and plane strain passive modes of shearing (s uDSS and s u90). It is
not surprising that these calculations overestimate the failure
depths for very long walls 共where wall flexibility and bending
failure are ignored兲. However, the results also show substantial
overestimation for all wall lengths based on the DSS strengths.
Overall, the results in Fig. 14 highlight the difficulties in se-
lecting undrained strength parameters for anisotropic stability
analyses of braced excavations. The results confirm that realistic
predictions can be obtained by considering the strengths mobi-

Fig. 15. Cross section and undrained strength profile of braced excavation in South Boston

752 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003


Fig. 16. Upper bound failure mechanism for braced excavation in
South Boston, isotropic strength profile
lized at a characteristic level of shear strain 共␥⫽0.6 –1.0% in
these examples兲. More extensive calculations 共over a range of soil
profiles and bracing systems兲 are required before these results can
be generalized for design.
Practical Application of Limit Analyses
Published Case Studies
Numerical limit analyses have been used to reinterpret the stabil-
ity of braced excavations in clay from five projects reported in the
literature as shown in Table 3. Four of these projects involved
excavations supported by sheet pile walls and prestressed cross-
lot struts, while the fifth used a perimeter diaphragm wall. Al-
though relatively large wall movements 共exceeding 10 cm兲 and
surface settlements occurred in each of these projects, only two of
the seven cross sections actually failed 共Davidson Ave. No. 2 and
Taipei兲. The selection of undrained shear strength profiles is con-
ditioned by the availability of data 共and the prevailing local site
investigation practice兲. Field vane data were used to provide iso-
tropic strength profiles for the Vaterland-3, HDR-4, and Taipei
Fig. 17. Cross section and strength parameters for raker supported excavation on Massachusetts Institute of Technology campus
JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003 / 753
cases, while UU triaxial tests were the only information available
to reanalyze the two sections on Davidson Ave. in San Francisco.
Calculations for the Mexico City case were based on recent high
quality DSS tests on undisturbed samples 共Covarrubias 1994兲 in
preference to the more scattered UU tests published originally by
Rodriguez and Flamand 共1969兲. Finno and Chung 共1992兲 and Su
et al. 共1998兲 published 共peak兲 anisotropic strength parameters
from laboratory triaxial tests (CK 0 UC and CK 0 UE) and these
have been used to define the DC yield criterion in the current
plane strain limit analyses 共i.e., assuming s uTC⫽s u0 and s uTE
⫽s u90 , and s uFV⫽s u45).
The factor of safety 共FS兲 against basal stability is based on the
definition conventionally defined in the slope stability calculation:
FS⫽s u /s um (⫽tan ␾⬘/tan ␾m⬘ ⫽c⬘/cm⬘ , where c ⬘ , ␾⬘⫽the apparent
cohesion and internal friction angle, c m⬘ , ␾ m⬘ ⫽the mobilized val-
ues of these parameters for drained soil layers兲. Numerical limit
analyses optimize the loading factor, L f for vertical pressure ex-
erted by the retained soil, P⫽L f 关 兺 ␥ i h i 兴 , where ␥ i , h i are the
unit weight and thickness of layer, i, in the retained soil 共Fig. 1兲,
respectively. Iterative analyses are carried out using trial values of
FS in order to obtain the condition L f ⬇1.00.
The computed FS values for isotropic strength profiles in
Table 3 are consistent with the field observations regarding fail-
ure. The calculations show FS⫽1.1–1.4 for the five sections
which remained stable, while FS⫽0.88 –0.96 and 0.95–1.03 for
the two failed sections 共Davidson Ave. No. 2 and Taipei兲. Apart
from Davidson Ave. No. 2 共where the wall rotates about a single
level of bracing兲 the predicted mechanisms of failure all involve
formation of plastic hinges in the wall at approximately the eleva-
tion of the lowest level of bracing.
Both sections of the HDR-4 project in Chicago had very low
margins of safety 共FS⬇1.1–1.2兲 against basal failure. It is also
interesting to note that calculations using the peak anisotropic
strength parameters give very similar values of FS as the isotropic
cases 共field vane profile兲. These results contrast with previous
example calculations for normally consolidated BBC, where peak
anisotropic strength predicted substantially higher stability num-
bers. This reflects the relative magnitudes of field vane and un-
drained triaxial shear strengths 共for HDR-4, s uFV⫽0.5关 s uTC
⫹s uTE兴 ; while for BBC the previous example uses s uDSS
⫽0.4关 s u0 ⫹s u90兴 ). Anisotropic analysis for the Taipei failure
共Table 3兲 show FS⫽1.23–1.32 and are clearly producing an un-
 Fig. 18. Upper bound failure mechanism for raker supported excavation, Massachusetts Institute of Technology campus
safe estimate of basal stability. This result is consistent with be-
havior described in the previous section for the idealized excava-
tions in BBC 共see also Whittle and Ukritchon 2000兲.
Design Applications
Fig. 15 shows the cross section and proposed support system for
an 18 m deep, 33 m wide excavation in Boston. The support
system includes a diaphragm wall that extends 6 m below the
final grade elevation, while cross-lot struts are installed at a maxi-
mum vertical spacing of 4 m. The soil profile consists of approxi-
mately 11 m of fill materials and silty sand overlying a 31 m deep
layer of BBC. Fig. 15 lists the shear strength parameters for the
overlying soils and compares profiles of undrained shear strength
from an extensive program of laboratory CK 0 UDSS and CK 0 UC
tests 共Ladd et al. 1999兲 using both SHANSEP 共Ladd and Foott
1974兲 and reconsolidation test procedures together with strength
profiles derived using the MIT-E3 model from the measured
stress history profile. The data consistently show minimum shear
resistance at 26 m–28 m depth corresponding to the base of the
overconsolidated clay crust 共just below the toe of the diaphragm
wall兲.
Limit analyses have been performed using a variety of inter-
pretations for the undrained strength profile. Fig. 16 illustrates
typical upper bound results using an isotropic strength criterion
that matches s uDSS of the MIT-E3 model. The results show a
typical failure mechanism which extends to the base of the clay
with FS⫽1.18 –1.23, and a plastic hinge in the wall at the eleva-
tion of the fourth level of bracing. Further calculations based on
anisotropic peak strengths predict FS⫽1.29–1.36 and are likely to
overestimate stability, while calculations based on ␥⫽1.0% mo-
bilized strengths may be more realistic and predict only a small
margin of safety with FS⫽1.05–1.13. In practice, it is difficult to
achieve further resolution using the two-dimensional 共2D兲 stabil-
ity calculations given the uncertainties in representing undrained
strength anisotropy. The limit analyses can provide useful guid-
ance on the effectiveness of berms, soil mix walls, etc. 共Ukritchon
1998兲, but other strategies to improve stability, such as excavation
in short panel lengths, require much more computationally de-
manding three-dimensional analyses 共e.g., Lyamin 1999兲.
Fig. 17 shows the cross section through a 12.5 m deep exca-
vation planned on the MIT campus 共Olsen 2001兲. Part of this very
wide 共110 m兲 excavation is to be supported by three levels of
inclined rakers reacting against a 1.2 m thick central concrete
base slab. The design includes a perimeter diaphragm wall em-
bedded almost 9 m below the final grade and within the crust of a
26 m thick layer of BBC. The undrained shear strengths of the
754 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003
lower clay at this site are notably higher than in the preceding
example 共a consequence of local variability in the stress history;
cf., Ladd et al. 1999兲. Fig. 18 shows that the predicted upper
bound failure mechanism again extends to the base of the clay
layer but is constrained inside the excavation by the presence of
the base slab. The analyses predict a relatively high factor of
safety, FS⫽1.70–2.05, which is consistent with independent re-
sults of nonlinear finite-element analyses 共using c-␾ reduction
methods; Brinkgreve and Bakker 1991兲. Much lower factors of
safety 共FS⫽1.2–1.4兲 are computed by limit equilibrium methods
based on Terzaghi basal heave calculations and circular arc 共rigid
body兲 rotation mechanisms that do not explicitly consider the
constraints imposed by the proposed bracing system.
Conclusions
This paper has described the formulation of 2D plane strain, nu-
merical limit analyses for modeling the stability of braced exca-
vations in clay. Upper and lower bound estimates of collapse are
formulated as linear programming problems based on earlier pub-
lications by Sloan and Kleeman 共1995兲 and Sloan 共1988a兲, re-
spectively. The current application includes new contributions for
incorporating structural elements and modeling undrained
strength anisotropy of clays using the yield criterion proposed by
Davis and Christian 共1971兲.
Results of the limit analyses provide an independent check on
the accuracy of long established empirical equations for comput-
ing the stability number of braced excavations. For each geometry
and set of material properties, the analyses are able to bound the
exact stability number within ⫾5%. The analyses show clearly
how mechanisms of failure for embedded walls are governed by
the ratio of their plastic moment capacity to the undrained
strength of the clay, through the dimensionless group,
M p / 关 s u D 2 兴 , where D is the depth of wall below the excavated
grade.
In principle, the undrained strength anisotropy of clays is
readily handled within the numerical limit analyses using appro-
priate anisotropic yield functions. However, the selection of suit-
able strength parameters is a difficult task due to different rates of
strength mobilization with the direction of shearing and level of
shear strain. This issue is most clearly seen in comparisons be-
tween the limit analyses and nonlinear 共displacement based兲
finite-element analyses that incorporate a realistic constitutive
model of the anisotropic stress–strain–strength properties. The
limit analyses predict similar failure depths as the finite-element
calculations when using mobilized anisotropic strengths corre-
sponding to shear strain levels in the range, ␥⫽0.6 –1.0%. Further
studies are needed to develop more general guidelines on param-
eter selection for anisotropic calculations.
Limit analyses, using available undrained strength data 共as-
sumed isotropic兲, provide credible estimates of the stability for a
series of braced excavations reported in literature, and correctly
predict failure that occurred in two of these cases. Recent practi-
cal applications in Boston highlight the capabilities of the numeri-
cal limit analyses for computing stability with realistic soil pro-
files and lateral earth support systems. Robustness and efficiency
of the analyses also serve to refocus attention on the measurement
and interpretation of undrained strength parameters.
Acknowledgment
The upper and lower bound codes used in this paper were origi-
nally developed by the third writer with support from the Austra-
lian Research Council.
References
Bishop, A. W. 共1955兲. ‘‘The use of the slip circle in the stability analysis
of slopes.’’ Geotechnique, 5, 7–17.
Bjerrum, L., and Eide, O. 共1956兲. ‘‘Stability of strutted excavations in
clay.’’ Geotechnique, 6, 115–128.
Brinkgreve, R. B. J., and Bakker, H. L. 共1991兲. ‘‘Nonlinear finite-element
analysis of safety factors.’’ Proc., 7th Int. Conf. on Comp. Methods
and Advances in Geomechanics, Cairns, Balkema, 2, 1117–1122.
Clough, G. W., and Hansen, L. A. 共1981兲. ‘‘Clay anisotropy and braced
wall behavior.’’ J. Geotech. Eng. Div., Am. Soc. Civ. Eng., 107共7兲,
893–913.
Clough, G. W., and Reed, M. W. 共1984兲. ‘‘Measured behavior of braced
wall in very soft clay.’’ J. Geotech. Eng., 110共1兲, 1–19.
Covarrubias, S. 共1994兲. ‘‘Characterization of the engineering properties
of Mexico City clay.’’ SM thesis, Dept. of Civil Engineering, MIT,
Cambridge, Mass.
Davis, E. H., and Christian, J. T. 共1971兲. ‘‘Bearing capacity of anisotropic
cohesive soil.’’ J. Soil Mech. Found. Div., Am. Soc. Civ. Eng., 97共5兲,
753–769.
Eide, O., Aas, G., and Josang, T. 共1972兲. ‘‘Special application of cast-in-
place walls for tunnels in soft clay.’’ Proc., 5th European Conf. on
Soil Mechanics and Foundation Engineering, Madrid, Spain, 1, 485–
498.
Finno, R. J., Atmatzidis, D. K., and Perkins, S. B. 共1989兲. ‘‘Observed
performance of a deep excavation in clay.’’ J. Geotech. Eng., 115共8兲,
1045–1064.
Finno, R. J., and Chung, C. K. 共1992兲. ‘‘Stress–strain–strength responses
of compressible Chicago glacial clays.’’ J. Geotech. Eng., 118共10兲,
1607–1625.
Hashash, Y. M. A. 共1992兲. ‘‘Analysis of deep excavations in clay.’’ PhD
thesis, Dept. of Civil and Environmental Engineering, MIT, Cam-
bridge, Mass.
Hashash, Y. M. A., and Whittle, A. J. 共1996兲. ‘‘Ground movement pre-
diction for deep excavations in soft clay.’’ J. Geotech. Eng., 122共6兲,
474 – 486.
Ladd, C. C. 共1991兲. ‘‘Stability evaluation during staged construction.’’ J.
Geotech. Eng., 117共4兲, 540– 615.
Ladd, C. C., and Foott, R. 共1974兲. ‘‘New design procedure for stability of
soft clays.’’ J. Geotech. Eng. Div., Am. Soc. Civ. Eng., 100共7兲, 763–
786.
JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / AUGUST 2003 / 755
Ladd, C. C., Young, G. A., Kraemer, S. R., and Burke, D. M. 共1999兲.
‘‘Engineering properties of Boston Blue Clay from special testing
program.’’ Proc., ASCE Geo-Congress’98, GSP 91, 1–24.
Lyamin, A. V. 共1999兲. ‘‘Three dimensional lower bound limit analysis
using nonlinear programming.’’ PhD thesis, Dept. of Civil, Surveying
and Environmental Engineering, Univ. of Newcastle, NSW.
Lysmer, J. 共1970兲. ‘‘Limit analysis of plane problems in soil mechanics.’’
J. Soil Mech. Found. Div., Am. Soc. Civ. Eng., 96共SM4兲, 1311–1334.
Mana, A. I., and Clough, G. W. 共1981兲. ‘‘Prediction of movements for
braced cuts in clay.’’ J. Geotech. Eng. Div., Am. Soc. Civ. Eng.,
107共6兲, 759–777.
Morgenstern, N. R., and Price, V. E. 共1965兲. ‘‘The analysis of the stability
of general slip surfaces.’’ Geotechnique, 15共1兲, 79–93.
Norwegian Geotechnical Institute. 共1962兲. ‘‘Measurements at a strutted
excavation, Oslo Subway, Vaterland 3.’’ Tech. Rep. No. 8, Norwegian
Geotechnical Institute. Blindern, Norway.
Olsen, M. B. 共2001兲. ‘‘Measured performance of a large excavation on
the MIT campus.’’ SM thesis, Dept. of Civil and Environmental En-
gineering, MIT, Cambridge, Mass.
O’Rourke, T. D. 共1993兲. ‘‘Base stability and ground movement prediction
for excavations in soft clay.’’ Retaining structures, Thomas Telford,
London, 131–139.
O’Rourke, T. D., and O’Donnell, C. J. 共1997兲. ‘‘Deep rotational stability
of tieback excavations in clay.’’ J. Geotech. Geoenviron. Eng., 123共6兲,
506 –515.
Rodriguez, J. M., and Flamand, C. L. 共1969兲. ‘‘Strut loads recorded in a
deep excavation in clay.’’ Proc., 7th Int. Conf. on Soil Mechanics and
Foundations Engineering, Mexico City, 2, 459– 467.
Seah, T. H. 共1990兲. ‘‘Anisotropy of normally consolidated Boston blue
clay.’’ ScD thesis, Dept. of Civil Engineering, MIT, Cambridge, Mass.
Skempton, A. W. 共1951兲. ‘‘The bearing capacity of clays.’’ Proc., Build-
ing Research Congress, London, 180–189.
Sloan, S. W. 共1988a兲. ‘‘Lower bound limit analysis using finite elements
and linear programming.’’ Int. J. Numer. Analyt. Meth. Geomech.,
12共1兲, 61–77.
Sloan, S. W. 共1988b兲. ‘‘A steepest edge active set algorithm for solving
sparse linear programming problems.’’ Int. J. Numer. Analyt. Meth.
Geomech., 26共12兲, 2671–2685.
Sloan, S. W., and Kleeman, P. W. 共1995兲. ‘‘Upper bound limit analysis
using discontinuous velocity fields.’’ Comput. Methods Appl. Mech.
Eng., 127, 293–314.
Sloan, S. W., Assadi, A., and Purushothaman, N. 共1990兲. ‘‘Undrained
stability of a trapdoor.’’ Geotechnique, 40共1兲, 45– 62.
Spencer, E. 共1967兲. ‘‘A method of analysis of the stability of embank-
ments assuming parallel interslice forces.’’ Geotechnique, 17共1兲, 11–
26.
Su, S. F., Liao, H. J., and Lin, Y. H. 共1998兲. ‘‘Base stability of deep
excavations in anisotropic soft clay.’’ J. Geotech. Geoenviron. Eng.,
124共9兲, 809– 819.
Terzaghi, K. 共1943兲. Theoretical soil mechanics, Wiley, New York.
Ukritchon, B. 共1998兲. ‘‘Application of numerical limit analyses for und-
rained stability problems in clay.’’ PhD thesis, Dept. of Civil and
Environmental Engineering, MIT, Cambridge, Mass.
Whittle, A. J., DeGroot, D. J., Ladd, C. C., and Seah, T. H. 共1994兲.
‘‘Model prediction of anisotropic behavior of Boston blue clay.’’ J.
Geotech. Eng., 120共1兲, 199–224.
Whittle, A. J., and Kavvadas, M. 共1994兲. ‘‘Formulation of the MIT-E3
constitutive model for over-consolidated clays.’’ J. Geotech. Eng.,
120共1兲, 173–198.
Whittle, A. J., and Ukritchon, B. 共2000兲. ‘‘Discussion of paper by Su
et al.’’ J. Geotech. Geoenviron. Eng., 126共8兲, 757–760.
Zienkiewicz, O. C. 共1983兲. The finite-element method, McGraw-Hill,
New York.