See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/229890386

Finite element implementation of non-

linear elastoplastic constitutive laws using
local global explicit...

Article in International Journal for Numerical Methods in Engineering · February 2001

DOI: 10.1002/1097-0207(20010220)50:5<1191::AID-NME73>3.0.CO;2-T

CITATIONS READS

25 143

3 authors, including:

Juan M. Pestana R. L. Taylor

University of California, Berkeley University of California, Berkeley
78 PUBLICATIONS 1,279 CITATIONS 325 PUBLICATIONS 27,238 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Juan M. Pestana on 22 January 2015.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
Int. J. Numer. Meth. Engng 2001; 50:1191–1212

Finite element implementation of non-linear elastoplastic

constitutive laws using local and global explicit
algorithms with automatic error control‡

Laurent X. Luccioni, Juan M. Pestana∗ ; † and Robert L. Taylor

Department of Civil and Environmental Engineering; University of California; Berkeley; CA 94720; U.S.A.

SUMMARY

Implicit stress integration algorithms have been demonstrated to provide a robust formulation for nite element
analyses in computational mechanics, but are dicult and impractical to apply to increasingly complex non-
linear constitutive laws. This paper discusses the performance of fully explicit local and global algorithms with
automatic error control used to integrate general non-linear constitutive laws into a non-linear nite element
computer code. The local explicit stress integration procedure falls under the category of return mapping
algorithm with standard operator split and does not require the determination of initial yield or the use of
any form of stress adjustment to prevent drift from the yield surface. The global equations are solved using
an explicit load stepping with automatic error control algorithm in which the convergence criterion is used to
compute automatically the coarse load increment size. The proposed numerical procedure is illustrated here
through the implementation of a set of elastoplastic constitutive relations including isotropic and kinematic
hardening as well as small strain hysteretic non-linearity. A series of numerical simulations con rm the
robustness, accuracy and eciency of the algorithms at the local and global level. Published in 2001 by John
Wiley & Sons, Ltd.

KEY WORDS: elastoplastic relations; small strain non-linearity; anisotropy; nite element method; explicit
integration algorithms

1. INTRODUCTION

There has been a signi cant e ort in computational geomechanics to describe the discrete consti-
tutive equations using fully implicit local and global stress integration algorithms [1; 2]. The most
popular is, perhaps, the fully implicit return mapping algorithm in which the return directions
are computed by the closest point projection method [1; 3–5], and has the advantage of being
amenable to consistent linearization [6]. Recent advances in classical computational plasticity have

∗ Correspondence to: Juan M. Pestana. Department of Civil and Environmental Engineering; University of California at
Berkeley; 440 Davis Hall 1710; Berkeley; CA 94720-1710; U.S.A.
† E-mail: [email protected]
‡ This article is a U.S. Government work and is in the public domain in the U.S.A.

Contract=grant sponsor: National Science Foundation; contract=grant number: CMS 9612136

Received 6 September 1999

Published in 2001 by John Wiley & Sons, Ltd. Revised 19 April 2000
1192 L. X. LUCCIONI, J. M. PESTANA AND R. L. TAYLOR

established the superiority of fully implicit algorithms to solve boundary value problems using the
nite element method for relatively simple plasticity models [7; 8]. However, as the constitutive
laws become more complex, the attractiveness of the fully implicit algorithm decreases signi cantly.
First, the formulation requires the second Frechet derivative of the yield function with respect to the
state variable tensors. Second, the solution of the local set of non-linear equations for highly non-
linear yield functions is by no means trivial, as the local Newton scheme may not converge. It is
important to note that fully implicit algorithms do not guarantee non-linear stability (or B-stability),
even though linear stability is automatically achieved [8; 9]. Thus, as of today, none of the existing
stress integration, either implicit or explicit, algorithms is able to guarantee stability and conver-
gence for general incremental elastoplastic constitutive laws under general conditions. Finally, an
explicit expression for the algorithmic consistent tangent may not be available, hence at best, only
a quasi-Newton method may be used and quadratic rate of convergence can not be expected. For
instance, Luccioni et al. [10] presented fully implicit local and global algorithms for the Bear–
Clay model using a quasi-Newton technique with a numerical tangent computed every load step
by nite di erence and optimized with iterative updating procedures since an explicit expression
of the algorithmic consistent tangent could not be determined.
For classical explicit integration schemes the discrete constitutive equations are much simpler
to formulate, but their accuracy depends signi cantly on the selected step size. Sloan [11] rst
proposed the application of an automatic stepping with error control numerical algorithm for the
prediction of collapse load using a relatively simple constitutive model. This numerical technique,
widely used in the eld of numerical analysis, is based on extrapolation procedures. The use of
automatic substepping and load stepping with error control algorithm overcomes the main limitation
of explicit techniques, since the system adapts as the estimated error changes.
Constitutive modelling and numerical implementation applied to geomaterials is by no means
trivial since it involves a subtle balance between the complexity associated with realistically
describing soil response and the use of both robust and accurate numerical algorithms. The Cam–
Clay family of models based on the critical state soil mechanics framework [12] is perhaps the most
widely used plasticity model for clays used to perform geotechnical (i.e. boundary value problem)
analyses. The advantages of this family of models derive from three main considerations: (1) their
ability to capture some aspects of soil behaviour [13], (2) their ecient numerical implementa-
tion into non-linear nite element codes [2; 14] and (3) the availability of a large database for
material ‘input’ parameters for di erent materials [15]. Nevertheless, these models do not incor-
porate key elements of material response, such as anisotropic stress–strain–strength behaviour or
small strain non-linearity. In particular, soils exhibit a high degree on non-linearity in the ‘elastic’
(i.e. recoverable strains) regime which is better described by a perfectly hysteretic formulation.
The main characteristic of this formulation is that the tangent sti ness decreases monotonically
with continued straining and is a function of a (typically dimensionless) stress (or strain) measure
describing the distance of the current state of stress (or strain) to the last reversal state. The
perfectly hysteretic model predicts fully recoverable strains in a closed unload–reload cycle, but
dissipates energy according to the prescribed sti ness reduction law (cf., Gsec =Gmax as a function of
shear strain, Figure 1). Since this non-linearity is strongly strain level dependent, the proposed ex-
plicit integration algorithm must be able to vary the ‘step size’ in order to maintain accuracy while,
at the same time, be computationally ecient. In contrast, general subincrementation procedures
with no step control must use very small incremental strains in order to accurately capture large
changes in material response (i.e. change in sti ness) following stress reversal which becomes com-
putationally inecient for many other practical conditions. The e ect of small strain non-linearity

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
 NON-LINEAR ELASTOPLASTIC CONSTITUTIVE LAWS 1193

Figure 1. Summary of Bear–Clay model formulation: (a) anisotropic plasticity and generalization of yield
surface (after Luccioni et al., 2000); (b) hysteretic elastic model for volumetric and shear response.

and anisotropy has been proven to be of signi cant importance in the prediction of deformations
in soil–structure interaction problems [16; 17], such as excavations [18] and tunneling activities
[19]. This problem is particularly relevant for dynamic problems for which soil non-linearity and
the associated energy dissipation is the most important aspect of material response.
Previous evaluations of explicit techniques have included relatively simple models such as linear
elastic-perfectly plastic Mohr–Coulomb or Tresca-strain hardening models [11; 19]. The following
sections present details of a fully explicit automatic substepping scheme with error control to

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
1194 L. X. LUCCIONI, J. M. PESTANA AND R. L. TAYLOR

integrate a non-linear model with anisotropic plasticity, small strain ‘hysteretic’ non-linearity and
dependence of the yield and plastic ow rules on the third invariant of the stress tensor following
the Matsuoka–Nakai generalization (cf. Figure 1(a)). Hysteretic non-linearity is mathematically
analogous to the bounding surface plasticity formulation where the tangent sti ness degrades as a
function of the proximity of the ‘current state’ to the bounding surface using appropriate mapping
rules. For the particular case of monotonic loading, this formulation is also similar in concept to
the sti ness degradation resulting from a ‘damage-related’ process and it is therefore applicable to
a more general class of models. The local stress integration algorithm falls under the category of
return mapping algorithm with standard operator split procedure and does not require the determi-
nation of initial yield or drift correction techniques. The discrete ‘local’ equations are integrated
using numerical techniques that preserve the incremental nature of the continuum formulation,
while the global system of equations are solved using an explicit automatic load stepping with
error control algorithm as proposed by Abbo and Sloan [20]. In contrast to previous derivations,
the explicit scheme must be used to verify the integration error even for ‘elastic states’, since there
is no general analytical expression of the elastic sti ness for the perfectly hysteretic formulation.
The proposed numerical procedure is illustrated here by implementing a recently developed
constitutive model for lightly overconsolidated clays [21; 22] that has been used in the solution of
boundary value problems such as the prediction of deformations around ground openings in soft
clays [23; 24]. The model, referred to as Bear–Clay, is based on the theory of incrementally lin-
earized plasticity which is extensively documented in the literature [7; 25; 26]. Model formulation
includes three important components to describe the observed clay response: (a) an elastoplastic
framework for normally consolidated clays with a single anisotropic yield function with depen-
dence on the third invariant of the stress tensor to represent accurately the e ect of consolidation
stress history, (b) equations describing the small strain non-linearity and hysteretic stress–strain
response in unload–reload cycles, and (c) non-associated ow and hardening rules to describe the
evolution of anisotropic stress–strain properties. The elastic sti ness tensor is assumed isotropic
allowing the decomposition of the elastoplastic relations into volumetric and deviatoric compo-
nents and they are brie y summarized in Appendix A. It should be emphasized that the numerical
algorithm proposed here is independent on the particular set of constitutive expressions used to
illustrate the procedure and can be directly extended to other material models without conceptual
changes. The numerical simulations of single element tests as well as a boundary value problem
con rm the robustness, accuracy, and eciency of the proposed algorithm at the local and global
levels. Finally, the implicit technique proposed by Luccioni et al. [10] and the explicit technique
presented here are compared in terms computational e ort as re ected by CPU time.

2. NUMERICAL IMPLEMENTATION

2.1. Local stress integration algorithm

The elastoplastic ow problem at the Gauss point level can be recast in term of a system of
di erential=algebraic equations that have to be integrated numerically. From the point of view of
integration techniques, this system of di erential=algebraic equations has the property of being
‘in nitely sti ’ [27]. This type of sti system is usually best integrated using methods such as
backward di erentiation formula (BDF), where convergence and accuracy can be formally estab-
lished [27; 28]. Gear [27] shows results indicating that fully explicit algorithms are, in general,

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
 NON-LINEAR ELASTOPLASTIC CONSTITUTIVE LAWS 1195

unstable and do not converge, suggesting that these algorithms should not be used to integrate
this type of system. However, in the context of numerical implementation into a computer code
one has to think in terms of nite precision. Indeed, we do not require to guarantee yield condi-
tions exactly (i.e. ’ = 0) but only in an approximated way, |’| 6 TOL where TOL may be the
machine precision or an even less restrictive constraint. This concept may be viewed as a penalty
regularization of the yield condition used for viscoplastic materials [29; 30].
Following Sloan [11], we de ne two pseudo-time quantities,
T and T , to be used in the
substepping scheme. When entering for the rst time the algorithm,
T is set to one, whereas T
is set to zero. The total incremental strain tensor,
, is translated into subincrement:

k+1 =
Tk
n+1 (1)
where k represents the kth substep and n represents the nth load or coarse step. For the remainder
of the paper, strains are assumed subincremental strains, unless otherwise indicated and tensorial
quantities are represented by boldface roman characters. The return mapping algorithm is used to
integrate the continuum rate equations using the rst-order accurate forward Euler scheme. The
trial state, represented by the superscript tr, is characterized as follows:

1 tr
pk+1 = pk + Kk (
p )k+1 (2a)
1 tr tr
sk+1 = sk+1 + 2Gk (
Us )k+1 (2b)

where p; s are the mean e ective stress and the deviatoric components of the stress tensor,
A;
p ;
Us are the volumetric and deviatoric components of the strain tensor,
U, and the left
superscript ‘1’ denotes the rst-order integration method. The model introduces expressions to
model the small strain non-linearity through the tangent elastic sti ness parameters Kk and 2Gk
describing the bulk and shear moduli, respectively, and summarized in Appendix A (cf. Equations
(A1) – (A3)). The loading=unloading conditions are determined by the sign of the yield function
at the trial (predictor) state: ’(1 Ak+1
tr
; qk ), where q is a vector containing the plastic (i.e. memory)
variables. This procedure has been demonstrated to be algorithmically consistent with the forward
Euler integration scheme [31] while Papadopoulos and Taylor [32] discusses its limitations for
implicit stress integration schemes. For ‘loading’ states, the (plastic) corrector step is invoked:

1
pk+1 = pk + Kk [(
p )k+1 −
(Pp )k ] = 1pktr − Kk
(Pp )k (3a)
1
sk+1 = sk + 2Gk [(
s )k+1 −
(Ps )k ] = 1 sktr − 2Gk
(Ps )k (3b)
1
qk+1 = qk +
q(
) (3c)

’k+1 = ’k+1 (1pk+1 ; 1 sk+1 ; k+1 ;

1
bk+1 ) = 0 (3d)

where Pp and Ps are the volumetric and deviatoric components of the ow rule describing the
direction of the plastic strains (cf. Appendix A),
q represents the change in the plastic variables
with continued plastic deformation (i.e. hardening) and
 is the incremental elastoplastic param-
eter, also referred to as the consistency parameter, obtained from the solution of this system of
equations. For the particular model used to illustrate the numerical procedure, the plastic variables
are given by and b representing the size of the yield surface and the traceless tensor describing
the orientation of the yield surface in a generalized stress space, respectively. It must be noted

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
1196 L. X. LUCCIONI, J. M. PESTANA AND R. L. TAYLOR

that the numerical algorithm proposed here is not dependent on the particular set of constitutive
laws used to illustrate the procedure and therefore it can be extended to other material models
without conceptual changes.
In contrast to previous models elastoplastic models for clays, the Bear–Clay model treats the
isotropic hardening variable, , as a dependent variable. The coupling between isotropic and kine-
matic hardening existing in the continuum equations can be written as = (b; p ) where is
updated from converged values of anisotropy, b, as follows:
    
(
p )k+1 1 + ek+1 @
k+1 = k exp exp :
b (4a)
c ek+1 @b
1
bk+1 = bk +
b(
) (4b)

where e is the void ratio of the soil (= volume of voids=volume of solids), c is the slope of
the hydrostatic limiting compression curve, H-LCC, for normally consolidated clays in a log(e)-
log(p) space,
b represents the change in anisotropy (i.e. kinematic hardening) resulting from
plastic loading and ‘:’ represents the double contraction of tensor multiplication. The term @ =@b
describes the coupling resulting from the existence of the limiting compression boundary surface,
LCBS, as described by Pestana and Luccioni [22] or the use of a spacing function [33; 34]. The
existence of the LCBS satis es the robustness of the drained response (i.e. consolidation behaviour)
under constant shear stress ratio, W( = s=p), conditions for models incorporating both kinematic and
density hardening as suggested by Pestana [33]. The system of 12 non-linear implicit equations
(cf. Equations (3a)–(3d)) has only one common unknown,
. Then, it follows that only one
single scalar equation needs to be solved instead of the inversion of a 12 × 12 Jacobian as in the
case of fully implicit algorithm [10]. The consistency equation, ’k+1 = 0, which insures that the
converged state of stress is on the yield surface, can be regarded as a function of
 only:
’k+1 = ’k+1 (A(
); q(
)) = ’k+1 (
) (5)
The consistency equation is then solved using the Newton’s method

’k+1 (
m )

m+1 =
m − (6a)
(@’k+1 \@(
))|m

where

@’k+1 @’k+1 @Ak+1 @’k+1 @qk+1

= + : (6b)
@(
) @Ak+1 @(
) @qk+1 @(
)
For Bear–Clay, the change in the yield function ’ resulting from the change in the elastoplastic
parameter can be derived making use of the chain rule as follows:
@’k+1 @’k+1 @pk+1 @’k+1 @sk+1 @’k+1 @bk+1
= + : + : (7a)
@(
) @pk+1 @(
) @sk+1 @(
) @bk+1 @(
)
where
@’k+1 @’k+1 @’k+1 @ k+1
= + (7b)
@bk+1 @bk+1 k+1
@ k+1 @bk+1

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
 NON-LINEAR ELASTOPLASTIC CONSTITUTIVE LAWS 1197

A summary of the constitutive equations for the Bear–Clay model is presented in Appendix A. In
order to solve Equation (6) with Newton’s method, a starting value for
 is required. Although it
is not uncommon to use
 = 0 as an initial guess=value, it is best to start with a value of
 as
close as possible to the converged one in order to recover quickly the asymptotic quadratic rate of
convergence of Newton’s method. The ‘continuum’ expression of
 is used here, since for small
steps, this value is very close to the
 obtained once Newton’s method has converged. Note
that in the context of explicit integration techniques, subincrement steps used to reach a converged
state are small, which is the basis for the development and validity of the in nitesimal linearized
theory of plasticity. The initial value for the elastoplastic parameter,
(0) is then given by
(K(@’=@p) + (@’=@e)(1 + e)) (
p )k+1 + 2G(@’=@s) : (
Us )k+1

(0) = (8)
−(@’=@q)(@q=@) + K(@’=@p)Pp + 2G(@’=@s) : Ps
where all the quantities, except the strain increment, are evaluated at kth step. The error control
scheme used in the proposed formulation is based on local extrapolation [11; 27], and requires
that the constitutive laws be reintegrated with a second-order method. The modi ed Euler (i.e.
predictor–corrector) method was chosen for this purpose. The modi ed Euler method uses the
converged state achieved from the forward Euler method as base values (known quantities). The
discrete equations can be written in the following form:
Predictor step:

2 tr
pk+1 = pk + 1 Kk+1 (
p )k+1 (9a)
2 tr 1
sk+1 = sk + 2 Gk+1 (
Us )k+1 (9b)

where the sti ness coecients, K and G have been evaluated at the converged state from the
forward Euler procedure.
Corrector step:

2




pk+1 = pk + 1 Kk+1
p k+1 −
 1 Pp k+1 = 2pk+1 tr
− 1 Kk+1
 1 Pp k+1 (10a)
2




sk+1 = sk + 21 Gk+1
Us k+1 −
 1 Ps k+1 = 2 sk+1 tr
− 21 Gk+1
 1 Ps k+1 (10b)
2


qk+1 = qk + 1
q
; 1pk+1 ; 1 sk+1; 1 b (10c)


’k+1 = ’k+1 2pk+1 ; 2 sk+1 ; k+1 ; 2 bk+1 = 0 (10d)

where superscript 1 indicates converged state from the forward Euler method, superscript 2 indi-
cates converged state from the modi ed Euler method, 1
b is the kinematic hardening evaluated
with the converged state from the forward Euler procedure, and the isotropic parameter, , is
evaluated as follows:
    
2 (
p )k+1 1 + ek+1 @ 1
k+1 = k exp exp :
b (11)
c ek+1 @b
The system of equations derived from the corrector step is solved by using the exact same technique
as the one proposed above for the rst-order integration method. Subtracting Equation (10) from

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
1198 L. X. LUCCIONI, J. M. PESTANA AND R. L. TAYLOR

Equation (9), a second-order accurate estimate of the local truncation error for the stress tensor,
A, is obtained:


Ek+1 = 2 Ak+1 −1Ak+1 =2 (12)
The relative error for a substep is written as [27; 11]
Rk+1 = kEk+1 k=kAk+1 k (13)
where Ak+1 = (1 Ak+1 + 2 Ak+1 )=2, which is second-order accurate. During the integration process,
each substep size is continually updated to insure that Rk+1 6STOL. The value of STOL is problem
and constitutive model dependent. On one hand, if the value set for STOL is too large, the solution
obtained may be completely inaccurate and therefore not satisfactory. On the other hand, if STOL
is chosen too small, the problem may not converge. Sloan [11] reported values of STOL between
10−1 and 10−5 and the same range of values are investigated here for the Bear–Clay model. A
parametric study on the satisfactory values of STOL is reported through the numerical simulations
presented in the last section.
In the case where Rk+1 6STOL, the subincrement is declared successful and all variables are
updated as follows:

1


Ak+1 = Ak+1 + 2 Ak+1 =2 (14a)


qk+1 = 1 qk+1 + 2 qk+1 =2 (14b)
T = T +
Tk (14c)

The local extrapolation is used to compute the next subincrement size. Hence,
Tk+1 = Â
Tk ,
where  is given by
 = [STOL=Rk+1 ]1=2 (15)
Heuristic bounds on  must be introduced to prevent the extrapolation to be carried too far, resulting
on unstable results. For the Bear–Clay model, it was found that the range 0:26Â62 gives good
results which is in agreement with values reported by Sloan. In the case where Rk+1 ¿STOL, the
subincrement has failed and the size of the subincrement has to be reduced and the algorithm
is restarted from the last converged value with a smaller subincrement size. An extrapolation is
used to compute the next subincrement size, following the same procedure as shown previously.
In order to reduce the number of unsuccessful substep and simultaneously keep
Tk ¿0:01, the
magnitude of k
n+1 k has to be controlled which is achieved by the global solution algorithm as
described in the following paragraphs.

2.2. Global solution algorithm

The numerical techniques used for solving the global discrete equations can be broadly classi ed
as either incremental or iterative procedures. In the iterative procedure, the discretization leads to
a system of non-linear equations to be solved by methods such as Newton–Raphson, or quasi-
Newton. These type of methods have the advantage of satisfying equilibrium equations at the end
of each converged time step, and if the consistent tangent is used an asymptotic rate of convergence
is recovered. In addition, the stability theorem can be proven for certain cases [35]. On the other
hand, when the material behaviour is strongly non-linear the iterations may not converge as all of

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
 NON-LINEAR ELASTOPLASTIC CONSTITUTIVE LAWS 1199

these methods have a nite radius of convergence. Moreover, if the algorithmic consistent tangent
is not used, the inaccuracy of the substitute tangent may result in intermediate strain increments
that may be too large, causing the local stress algorithm to either use too many subincrements or
diverge altogether.
Incremental procedures, on the other hand, treat the governing equations as a system of ordi-
nary di erential equations (ODE). Thus, the solution consists of a series of piecewise linear steps
that attempts to approximate the load–deformation behaviour of the system. This class of methods
guaranties that small load increment will be used which is an advantage when explicit stress inte-
gration techniques are used at the Gaussian (i.e. ‘local’) level. Nevertheless, this class of methods
tends to ‘drift’ from equilibrium as the solution proceeds, leading to very doubtful interpretations.
Abbo and Sloan [20] presented an incremental algorithm based on an automatic load stepping
scheme with error control. This algorithm tries to minimize the ‘drift’ from equilibrium by calcu-
lating the residual forces at the end of each load increment and adding these to the applied forces
for the next increment. The authors choose to implement a slightly modi ed version of the Abbo
and Sloan [20] algorithm into a non-linear nite element code, referred to as FEAP. This nite
element code was developed at the University of California, Berkeley for teaching and research,
and complete documentation is available [36]. The elastoplastic continuum tangent is assembled,
and used by the explicit algorithm to solve for the displacement. In contrast to the implicit al-
gorithm, the elastoplastic continuum tangent is not signi cantly di erent from the algorithmic
consistent tangent as step increments are much smaller. The elastoplastic continuum tangent, C ep ,
is given by
(Ce : Q + @’=@p ) ⊗ Ce : P
Cep = Ce − (16)
H + Q : Ce : P
where Q is the gradient to yield surface, P describes the ow rule indicating the direction of plastic
strain increments, C e is the continuum elastic sti ness tensor, and H is the hardening modulus
(cf. Appendix A). In contrast to the traditional formulation of incrementally linearized plasticity,
Equation (16) introduces the rate of change of the shape and=or size of the yield surface with
respect to the total volumetric strain (i.e. void ratio) as described by Pestana [33].
The proposed modi cation to the algorithm consists of using a convergence criterion to compute
automatically the coarse load increment size, making the global algorithm entirely automatic from
the Gauss point level to the coarse load level. If more than two consecutive unsuccessful substeps
are detected at any level, the current coarse load level is divided by two, whereas if more than two
consecutive successful coarse steps are detected, the current coarse load level is multiplied by 1.2.
These bounds are purely heuristic based on multiple numerical simulations performed using the
Bear–Clay model, hence caution should be used in generalizing these numbers to others model or
situations. The previous remark showcases some drawbacks in using incremental techniques, but
it also constitutes their strengths as these explicit techniques are exible enough to accommodate
very strong non-linear behaviour.

3. NUMERICAL SIMULATIONS

This section describes numerical simulations used to validate the proposed approach. Both local
and global algorithms are coded into a series of Fortran90 subroutines and they are linked to the

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
1200 L. X. LUCCIONI, J. M. PESTANA AND R. L. TAYLOR

Figure 2. Plane strain nite element mesh.

main non-linear nite element code, FEAP [36]. All computations were performed with double-
precision arithmetic on a 32-bit architecture DEC 3000 workstation at the University of California
at Berkeley. The convergence criterion used in the acceptance or rejection strategy of incremental
displacements, u, at the global level is based on the local extrapolation procedure [20], where

2


En+1 =
un+1 −1
un+1 =2 (17a)
Rk+1 = kEk+1 k=kuk+1 k6DTOL (17b)

Similarly to STOL discussed previously, DTOL is a user-de ned tolerance that is model and
problem dependent. For the Bear–Clay model and within the context of the numerical simulations
presented below, values of DTOL ranging between 10−1 and 10−5 have been investigated.

3.1. Single-element tests

The rst numerical example investigates the local stress integration algorithm and its implemen-
tation. These simulations consist of single element (cf. Figure 2) undrained plane strain com-
pression and extension tests, where the sample is initially anisotropically (i.e. K0 = h0 0 =v0 0) nor-
mally consolidated or 1-D unloaded to an overconsolidation ratio, OCR, of 2. The ‘simulated
soil’ corresponds to Boston Blue Clay a low-plasticity clay widely documented in the geotechni-
cal literature. Its model speci c material parameters are described in detail by Luccioni [24] and
Pestana and Luccioni [22]. The axial strain for the undrained test is applied in 100 nite steps of
size
A up to a total axial strain |A | = 10 per cent (i.e.
A = 0:1 per cent). Measure of accuracy
is based on the algorithmic consistency property, which implies that the solution accuracy increases
as the number of steps increased. Figures 3 and 4 show the e ective stress path and shear strain
response for normally K0 consolidated and overconsolidated (OCR = 2) samples, respectively.
For each past consolidation history, a parametric study of the user-de ned local tolerance STOL
is presented. Figure 3 demonstrates that the integration scheme is reasonably accurate for STOL
values ranging from 10−5 to 10−3 , whereas for STOL equal to 10−1 , the algorithm looses accuracy

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
 NON-LINEAR ELASTOPLASTIC CONSTITUTIVE LAWS 1201

Figure 3. Parametric study on local algorithm for normally consolidated specimen. (OCR = 1).

Figure 4. Parametric study on local algorithm for overconsolidated specimen (OCR = 2).

in the compression mode. The same conclusions are reached from Figure 4 where not only a loss
of accuracy in the compression mode is observed, but also the extension test did not converge for
STOL equal to 10−1 . These results seem to suggest that values of STOL¡10−3 are acceptable.
Once the value of STOL is xed, the algorithm computes automatically the number of substeps
needed to achieve the speci ed tolerance. Table I reports the number of steps, N , associated with

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
1202 L. X. LUCCIONI, J. M. PESTANA AND R. L. TAYLOR

Table I. Summary of the strain subincrement for OCR = 1 simulation.

Average No. of strain subincrements per step

Successful Failed
Tolerance STOL Compression Extension Compression Extension

10−5 100 80 1 0
10−3 20 14 2 3
10−1 4 3 4 n=a

n=a: The algorithm did not converge.

Table II. Rate of convergence of local Newton algorithm for

OCR = 1 simulation.
Rate of convergence
Tolerance STOL 10−5 10−1

Step 2 1:186 × 10−3 0.3168

6:283 × 10−6 0.1331
1:799 × 10−10 5:978 × 10−2
1:245 × 10−3
3:452 × 10−4
Step 100 9:459 × 10−4 5:888 × 10−2
5:009 × 10−6 8:560 × 10−3
1:462 × 10−10 3:600 × 10−4
7:687 × 10−7

a given value of STOL. Intuitively, the normally consolidated plane strain compression test is the
most critical because the response is entirely elasto-plastic and exhibits strain softening behaviour.
This intuitive argument is validated by numerical simulations, as for a given STOL, this mode of
shearing uses the largest number of steps. Table II shows the rate of convergence of the local
Newton–Raphson algorithm for two typical iterations. The results indicate that for a strict tolerance
(STOL = 10−5 ) the asymptotic rate of convergence is quickly achieved, attesting of the quality of
the initial guess for the consistency parameter,
(0) . Loosening the tolerance (STOL = 10−1 )
results in larger strain subincrement, which leads to a progressive breakdown of the explicit
integration technique. Comparisons of the stress–paths and stress–strain curves between the ex-
plicit and implicit integration techniques, shows excellent agreement in terms of accuracy. The
explicit technique is more costly than the implicit one in term of number of iterations (10 000 vs
1500), but each explicit iteration requires at the most the solution of a single scalar non-linear
equation compared to the inversion of a 12 × 12 Jacobian in the case of the implicit algorithm [10].
As a result, the two techniques are in the same order of magnitude in terms of the computational
e ort (i.e. CPU time).

3.2. Patch tests

The second series of numerical simulations focus on the performance of the global solution algo-
rithm. Abbo and Sloan [20] applied this algorithm to predict collapse load for various boundary
value problems with a cohesive-frictional constitutive law based on a rounded Mohr–Coulomb

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
 NON-LINEAR ELASTOPLASTIC CONSTITUTIVE LAWS 1203

Figure 5. One-dimensional stress–strain law Figure 6. Finite element patch test mesh.
for the J2 ow model.

Table III. J2 ow model properties used for testing

global algorithm—patch test.

Parameter Value

Shear modulus 100 000

Yield stress 10
Yield increment 10
Exponent isotropic 300
Hardening modulus 30

yield surface. They integrated the constitutive law using an explicit technique developed by Sloan
[11]. Herein, we evaluated the global explicit algorithm performance separately, for a di erent
‘quality’ of the Jacobian (i.e. continuum vs consistent). For this purpose, the J2 ow model with
isotropic and kinematic hardening laws was selected as the constitutive law (cf. Figure 5). The
J2 ow model is a popular associative model in the structure mechanics eld, which is simple
enough to be consistently linearized under the return map algorithm such that both continuum
and consistent tangent are available [8]. A four-element patch test, shown in Figure 6, is used to
compare the performance of the di erent algorithms. A summary of the material properties for the
J2 ow model is presented in Table III. The surface load applied on the external boundary has a
magnitude of 15 kN, which is higher that the value of the yield stress, Y = 10 kN. The local stress
integration for the J2 ow model consists of the ecient implicit return mapping algorithm. In
the following numerical simulations the global solution algorithm is varied between the proposed
incremental algorithm and the iterative algorithm based on Newton–Raphson technique, for the
continuum and consistent Jacobian. A summary of the rst simulation program is presented in
Table IV. The parameter RES represents the residual force vector whereas Ux and Uy are the
displacements of node 5 (cf. Figure 6) in the horizontal and vertical direction, respectively. From
Table IV, it is clear that the global incremental algorithm is accurate for DTOL smaller than 10−3 .
One important and interesting feature is that for values of DTOL between 10−3 and 10−5 , the

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
1204 L. X. LUCCIONI, J. M. PESTANA AND R. L. TAYLOR

Table IV. Summary of global solution algorithm simulation—patch test.

Global incremental algorithm

Consistent Jacobian Continuum Jacobian

No. of Coarse CPU Ux Uy No. of Coarse CPU Ux Uy

DTOL RES increment (s) (cm) (cm) increment (s) (cm) (cm)

10−5 6 × 10−7 1980 49.7 1.720 2.007 1990 50.2 1.720 2.007
10−4 1 × 10−7 890 25.3 1.720 2.007 920 26.4 1.720 2.007
10−3 1 × 10−7 310 9.8 1.720 2.007 330 11.3 1.720 2.007
10−2 5 × 10−2 50 4.1 2.857 3.333 120 6.4 3.012 2.998
10−1 10−1 10 1.9 1.980 2.310 35 5.1 3.450 3.237

Figure 7. Comparison of performance of global incremental algorithm.

algorithm performed equally well using the consistent or continuum tangent. This result is further
illustrated through Figure 7 where CPU time and the number of coarse iterations are plotted as
a function of DTOL. Note that as the tolerance DTOL is becoming looser, using the consistent
tangent leads to a more accurate and ecient solution algorithm. These results demonstrate that

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
 NON-LINEAR ELASTOPLASTIC CONSTITUTIVE LAWS 1205

Table V. Comparison between iterative and incremental global solution algorithm.

Iterative algorithm (RES¡10−8 ) Incremental algorithm (DTOL = 10−3 )

No. of CPU Ux Uy No. of coarse CPU Ux Uy

Jacobian Iteration (s) (cm) (cm) increment (s) (cm) (cm)

Consistent 8 0.87 1.720 2.007 310 10.03 1.720 2.007

Continuum 13 2.13 1.720 2.007 330 11.12 1.720 2.007

in the context of the incremental technique, continuum tangent may be used as long as the user
de ned tolerance is kept strict enough (less than 10−3 in this case). Table V reports a com-
parison between the iterative and incremental global solution algorithm using both consistent and
continuum tangent. The iterative algorithm is 10 and ve times faster than the incremental tech-
nique using the consistent and continuum tangent, respectively. These results point out the superior
eciency of the iterative technique for this simple case, but also illustrates the sensitivity of the
iterative technique to the ‘quality’ of the Jacobian used in the solution. Results also show that
both algorithms are accurate regardless of the Jacobian used.

3.3. Boundary value problem

Finally, the third numerical example focuses on the performance of the overall implementation
technique into the non-linear nite element code, FEAP. To allow a direct comparison between
explicit and implicit techniques, the application problem presented here is identical to that shown
by Luccioni and Pestana [10] for the implicit technique. In this example, we consider the plane
strain problem of a vertically loaded exible strip footing of width B on a uniform ‘ nite’ deposit
of normally consolidated clay. The initial geostatic stresses were generated within FEAP using
a body force command to simulate the gravity load at each Gauss point and a prescribed value of
K0 of 0.53 describing the ratio of horizontal to vertical stress. The nite element mesh is shown
in Figure 8, and it is composed of 200, four-noded quadrilateral ‘soil’ elements with a 2 × 2
Gaussian integration rule employed for each element. The soil pro le is underlying by a rigid
and rough bedrock. The load is applied using di erent global tolerance DTOL, whereas the local
user tolerance is xed to a value of 10−3 . As described previously, for each value of DTOL, the
algorithm computes automatically the number of coarse load increments as well as the number
of subincrements necessary to achieve the prescribed user tolerance. Figure 8 shows the load-
settlement curve at the centerline of the exible foundation for di erent values of DTOL. Again,
the results indicate that the algorithm is accurate for values of DTOL less than 10−3 . The lack of
accuracy observed when using DTOL equal to 10−2 leads to a 25 per cent overestimation of the
settlement. Table VI summarizes the cost-related variables, showing that although the di erence in
accuracy for DTOL values of 10−3 and 10−4 is less than 1 per cent, the CPU time is multiplied
by a factor larger than two and the number of coarse steps quadruple. The load–settlement curve
corresponding to DTOL equal to 10−3 compares very well with the one computed using the global
implicit-iterative algorithm. A maximum error of less than 5 per cent is observed for a pressure
of p=pa ∼ 2 (where pa is the atmospheric pressure). Figure 9 compares the load–settlement curve
for the proposed explicit algorithm with DTOL = 10−4 and the corresponding implicit algorithm
with 20 steps [10]. The results from both procedures are in very good agreement and for most
practical purposes identical. As can be seen, the cost for the two techniques is quite similar with

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
1206 L. X. LUCCIONI, J. M. PESTANA AND R. L. TAYLOR

Figure 8. Performance of numerical integration algorithm: undrained loading of a exible footing:

(a) undeformed and deformed mesh con guration; (b) soil pro le and load settlement curves.

Table VI. Iterations summary for the global algorithms.

Tolerance No. of coarse CPU time

DTOL steps (s)

10−2 50 65
10−3 345 375
10−4 1210 780

432 s CPU time for the 5 steps iterative solution and 375 s for the incremental solution with DTOL
equal to 10−3 .

4. CONCLUSIONS

Fully implicit stress integration algorithms are attractive from the computational point of view but
they are typically very cumbersome, and in many cases impractical, to implement for complex

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
 NON-LINEAR ELASTOPLASTIC CONSTITUTIVE LAWS 1207

Figure 9. Comparison of predicted settlement curve using implicit and explicit stress integration algortihm.

non-linear constitutive soil models. For these models, the choice of an explicit automatic sub-
stepping with error control algorithm represents a computationally ecient alternative. The paper
has presented detailed information of the application of fully explicit local and global algorithms
with automatic error control for the integration of non-linear elastoplastic constitutive laws. The
proposed explicit local stress integration algorithm falls under the category of return mapping
algorithm and does not require determination of the initial yield or any drift correction techniques.
The iterations for the explicit algorithm require, at most, the solution of a single scalar non-linear
equation compared to the inversion of a 12 × 12 Jacobian as in the case of the implicit algorithm.
The proposed global explicit technique is found to be computationally ecient and accurate using
the continuum tangent as long as the user tolerance DTOL is tight enough. The numerical pro-
cedure is illustrated here by the implementing a recently developed constitutive model for lightly
overconsolidated clays including anisotropic behaviour with small strain non-linearity in shear into
a nite element computer code. The numerical algorithm proposed here can be directly extended to
other models without conceptual changes and is not dependent on the selected set of constitutive
expressions used to illustrate the procedure. Numerical simulations of single-element tests as well
as a boundary value problem con rm the robustness, accuracy, and eciency of the proposed
algorithm at the local and global level. Preliminary comparisons suggest that the proposed explicit
and implicit algorithms have similar computational performances in terms of CPU time, in spite
of the fact that the number of steps is 3–4 times larger for the explicit technique.

APPENDIX A: SUMMARY OF BEAR–CLAY MODEL FORMULATION

The following sections summarize the equations de ning the formulation of the Bear–Clay model
for lightly overconsolidated soils. Tensorial quantities are represented by boldface roman characters.

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
1208 L. X. LUCCIONI, J. M. PESTANA AND R. L. TAYLOR

A.1. Hypo-elasticity formulation

The tangent elastic sti ness tensor is assumed isotropic and it is decomposed into the volumetric
and deviatoric components, K and 2G, respectively, describing the bulk and shear moduli:

 −1    
K r0 (1 − ) e 3 1 − 2 (1 + !s s )
= + (A1a)
pa (p=pa ) 1+e 2 1+ (Gmax =pa )

2G 3(1 − 2)
= (A1b)
K (1 + )
 1−1:3c
Gmax Gb p
= 1:3 (A1c)
pa e pa

where Gb is a constant parameter describing the magnitude of the small strain shear modulus,
Gmax , c is the slope of the compression curve of normally consolidated clays in a log(e) –log(p)
space, r0 is the slope of the compression curve at OCR¿10, parameter !s controls the small
strain non-linearity in shear,  is a constant elastic Poisson’s ratio chosen to match the measured
unloading from normally consolidated states to moderately overconsolidated states (OCR 3–4) and
pa is the atmospheric pressure. The parameters  and s are memory parameters describing the
amount of unloading from the reversal state:

s ={(W − Wrev ): (W − Wrev )}1=2 ; = min(prev =p; p=prev ) ( only activated for drained testing)

where W (= s=p) is the current shear stress ratio tensor, Wrev is the shear stress ratio corresponding
to the last reversal point, prev is the mean e ective stress at last reversal, and ‘:’ represents the
double contraction of tensor multiplication.

A.2. Plastic states—yield surface

The yield surface prescribing ‘plastic state conditions’ is described by a function ’:

’ = W : W − c2 + 2 (p= )m = 0 (A4a)

where

2 = c2 + (1 − a)b : b − (2 − a)W : b (A4b)

where controls the size of the yield surface, b is a second-order tensor describing the orientation
of the yield surface in e ective stress space, parameter m controls the slenderness of the yield

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
 NON-LINEAR ELASTOPLASTIC CONSTITUTIVE LAWS 1209

surface, parameter ‘a’ de nes the shape of the yield surface near the tip (i.e. p ∼ ) and controls
the ratio of horizontal to vertical stress (i.e. K0 ) for one-dimensional strain consolidation. The
parameter c2 describes the aperture of the yield surface at the origin (i.e. p ∼ 0) and follows the
Matsuoka–Nakai generalization given by the maximum angle 0m :



−1
c2 = ca2 + 3 − ca2 =2 J3Á ; ca2 = 8 sin2 m 3 + sin2 m (A5)

where J3Á is the third invariant of the stress ratio tensor (i.e. J3Á = det[W])

A.3. Flow rule and large strain failure conditions

Failure conditions are represented by an isotropic function of the form proposed by Matsuoka and
Nakai (1974):

hf = k 2 − W : W = 0; k 2 = ka2 + (3 − ka2 =2)J3Á ; ka2 = 8 sin2 cs =(3 + sin2 cs ) (A6)

The tensor P de nes the direction of the plastic strain increment with volumetric and deviatoric
components Pp and Ps , respectively (i.e. P = Pp I + Ps , where I is the identity tensor):

@’
Pp = 0:5 (k2 − Á : Á); Ps = 2(p=pa ) (A7)
@s

A.4. Hardening laws

The isotropic and kinematic hardening are described by the following expressions:

(1 + e) @
d = dp − : db (A8)
c e @b

db = (W − b) d (A9)

where dp is the incremental change in total volumetric strain and @ =@b represents the coupling
between the density and kinematic hardening controlled by the limiting compression boundary
surface, LCBS (after Pestana and Luccioni [22]):

s " # 
@  3d2 r 6 @J 
3b
= : − cos 3Â:b − b (A10a)
@b (ÿd2 − b : b)  8 b : b @b 

24 √ J3b
d2 = where cos 3Â = 3 6 (A10b)
(3= sin cs − cos 3Â)2 (b : b)1:5

where ÿ is a material parameter describing the shape of LCBS and J3b is the third invariant of
the anisotropy tensor (i.e. J3b = det[b]).

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
1210 L. X. LUCCIONI, J. M. PESTANA AND R. L. TAYLOR

Table A1. Bear–Clay input material parameters used in the analysis.

Input Boston blue

Test type parameter Physical interpretation Clay

1. Hydrostatic or 1-D c Compressibility of Clays LCC 0.178

compression test r0 Unloading behaviour in 1D compression 0.035
2. 1-D compression with  Average Poisson’s ratio 0.26
lateral stress measurement
3. Undrained triaxial m Geometry of the yield surface 1.0
compression and extension ÿ Geometry of LC boundary surface 1.25
cs Critical state friction angle 33:50
Rate of evolution of anisotropy 10.0
!s Small strain non-linearity in shear 4.0
4. Elastic shear wave velocity Gb Small strain shear sti ness 250

A.5. Gradient of yield surface and elastoplastic modulus

Herein, all quantities are evaluated at k + 1 unless speci ed otherwise. The gradient of the yield
surface is decomposed into the volumetric and deviatoric components, respectively

@’ 1 n  p m   p m o
Qp = = (m2 + (2 − a)W : b) − 2W : W + 3cb2 J3Á 1 − (A11)
@p p
  p m 
@’ 1   p m  @J3Á 
Qs = = 2W − (2 − a) :b − cb2 1− (A12)
@s p @W
@J3Á
= det[W]:W−T = J3Á :W−T (A13)
@W
The change in the yield surface with respect to changes in the memory parameters, and b, can
be determined as
m2
@’=@ = − :(p= )m ; @’=@b = {2(1 − a):b − (2 − a)W}:(p= )m (A14)
As a result the elastoplastic modulus can be determined as
   
1 @’ @ @’ @b 1 @’ @ @’ @b
H =− p
+ : p
= − + : p (A15)
d @ @ @b @ d @ @b @b @
where p is the plastic strain tensor and d is the elastoplastic parameter obtained from
(KQp + @’=@e(1 + e)) dp + 2GQs : dUs
d = (A16)
H + KQp Pp + 2GQs : Ps

ACKNOWLEDGEMENTS

Support for this research was provided by the National Science Foundation Grant No. CMS 9612136 with
the University of California, Berkeley and by the National Science Foundation CAREER award to the second

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
 NON-LINEAR ELASTOPLASTIC CONSTITUTIVE LAWS 1211

author. This support is gratefully acknowledged. The authors thank Miss Lynn Salvati for careful review of
the nal manuscript.

REFERENCES
1. Borja RI, Lee SR. CAM-CLAY plasticity, Part I: implicit integration of elasto-plastic constitutive relations. Computer
Methods in Applied Mechanics and Engineering 1990; 78:49 –72.
2. Borja RI. CAM-CLAY plasticity, Part II: implicit integration of constitutive equations based on a nonlinear stress
predictor. Computer Methods in Applied Mechanics and Engineering 1991; 88(2):225 –240.
3. Lee JH, Zhang Y. On the numerical integration of a class of pressure-dependent plasticity models with mixed hardening.
International Journal for Numerical Methods in Engineering 1991; 32:419 – 438.
4. Zhang ZL. Explicit consistent tangent moduli with a return mapping algorithm for pressure-dependent elastoplasticity
models. Computer Methods in Applied Mechanics and Engineering 1995; 121:29 – 44.
5. Manzari MT, Nour MA. On implicit integration of bounding surface plasticity models. Computers and Structures
1997; 63:385 – 395.
6. Simo JC, Taylor RL. Consistent tangent operator for rate independent elasto-plasticity. Computer Methods in Applied
Mechanics and Engineering 1985; 48:79 –116.
7. Simo JC, Ortiz M. A uni ed approach to nite deformation elastoplastic analysis based on the use of hyperelastic
constitutive equations. Computer Methods in Applied Mechanics and Engineering 1985; 49:221– 245.
8. Simo JC, Hughes TJR. Computational Inelasticity, Interdisciplinary Applied Mathematics (IAM). Springer: Berlin,
1998.
9. Simo JC, Govindjee S. Non-linear B-stability and symmetry preserving return mapping algorithms for plasticity and
viscoplasticity. International Journal for Numerical Methods in Engineering 1991; 31:151–176.
10. Luccioni XL, Pestana JM, Rodriguez-Marek A. An implicit integration algorithm for the nite element implementation
of a nonlinear anisotropic material model including hysteretic nonlinearity. Computer Methods in Applied Mechanics
and Engineering 2000, to appear.
11. Sloan SW. Substepping schemes for the numerical integration of elastoplastic stress–strain relations. International
Journal for Numerical Methods in Engineering 1987; 24:893 – 911.
12. Wood DM. Soil Behavior and Critical State Soil Mechanics. Cambridge University Press: Cambridge, 1990; 462.
13. Randolph MF, Wroth CP. Application of the failure state in undrained simple shear to the shaft capacity of driven
piles. GÃeotechnique, London 1981; 7(1):19 – 38.
14. Borja RI, Tamagnini C, Amorosi A. Coupling plasticity and energy-conservating elasticity models for clays. ASCE,
Journal of Geotechnical Engineering 1997; 123(10):948 – 956.
15. Nakase A, Kamei T, Kusakabe O. Constitutive parameters estimated by plasticity index. ASCE, Journal of Geotechnical
Engineering 1998; 114(7):844 – 858.
16. Hight DW, Higgins KG. An approach to the prediction of ground movements in engineering practice: background and
application. In Pre-failure Deformation of Geomaterials, Shibuya S, Mitachi T, Miura S (eds). Balkema: Rotterdam,
1995; 909 – 945.
17. Jardine RJ, Potts DM, Fourie AB, Burland JB. Studies of the in uence of non-linear stress–strain characteristics in
soil–structure interaction. GÃeotechnique 1986; 36(3):377 – 396.
18. Finno RJ, Harahap IS. Finite element analysis of HDR-4 excavation. ASCE, Journal of Geotechnical Engineering
1991; 117(10):1590 –1609.
19. Addenbrooke TI, Potts DM, Puzrin AM. The in uence of pre-failure soil sti ness on the numerical analysis of tunnel
construction. GÃeotechnique 1997; 47(3):693–712.
20. Abbo AJ, Sloan W. An automatic load stepping algorithm with error control. International Journal for Numerical
Methods in Engineering 1996; 39:1737 –1759.
21. Pestana JM, Luccioni L. Description of drained and undrained behaviour of soft marine clay deposits. ASCE, 12th
Engineering Mechanical Conference, San Diego, CA, 1998; 1013 –1016.
22. Pestana JM, Luccioni LX. A simpli ed constitutive model for lightly overconsolidated clays. Report UCB=GT 99-23,
Department of Civil and Environmental Engineering, University of California, Berkeley, 1999.
23. Luccioni LX, Pestana JM, Koutsoftas D. Modeling of deformations around deep excavations in soft soils. In Proceedings
of Geo-Engineering for Underground Facilities, Fernandez, Bauer (eds), Geotechnical Special Publication No. 90.
ASCE: New York, 1999; 231– 242.
24. Luccioni LX. Numerical development and implementation of a constitutive model for clays with application to
deformations around a deep excavation. Ph.D. Thesis, Department of Civil and Environmental Engineering, University
of California, Berkeley, 1999.
25. Prevost JH. Plasticity theory for soil stress–strain behaviour. ASCE, Journal of Engineering Mechanics 1978;
104(EM5):1177 –1194.
26. Hashigushi K. Constitutive equations of elastoplastic materials with elastic-plastic transition. Journal of Applied
Mechanics, ASME 1980; 47:266 –272.

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212
 1212 L. X. LUCCIONI, J. M. PESTANA AND R. L. TAYLOR

27. Gear CW. Numerical Initial Value Problems in Ordinary Di erential Equations. Prentice-Hall, Englewood Cli s, NJ,
1971.
28. Petzold LR. Recent developments in the numerical solutions of di erential=algebraic systems. Computer Methods in
Applied Mechanics and Engineering 1989; 75:77 – 89.
29. Duvaut G, Lion JL. Les Inequations en Mecanique et en Physique. Dumod: Paris, 1972.
30. Ortiz M, Simo JC. Analysis of a new class of integration algorithms for elastoplastic constitutive relations. International
Journal for Numerical Methods in Engineering 1986; 23:353 – 366.
31. Papadopoulos P, Taylor RL. On the loading=unloading conditions of in nitesimal discrete elastoplasticity. Engineering
Computations 1995; 12:373 – 383.
32. Papadopoulos P, Taylor RL. On the application of multi-step integration methods to in nitesimal elastoplasticity.
International Journal for Numerical Methods in Engineering 1994; 37:3169 – 3184.
33. Pestana JM. A uni ed constitutive model for clays and sands. Sc.D. Thesis. Department of Civil and Environmental
Engineering, Massachusetts Institute of Technology, Cambridge, MA, 1994.
34. Pestana JM, Whittle AJ. Formulation of a uni ed constitutive model for clays and sands. International Journal for
Numerical and Analytical Methods in Geomechanics 1999; 23(12):1215–1243.
35. Ogden RW. Non-Linear Elastic Deformations. Ellis Horwood: West Sussex, England, 1998.
36. Taylor RT. FEAP Manual. Department of Civil and Environmental Engineering, University of California, Berkeley.

Published in 2001 by John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2001; 50:1191–1212

View publication stats