You are on page 1of 18

Oo~r molhods

In aMHIod
meohanlu and
onglnoorlng
ELqEVIER Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-240

Numerical computation of algorithmic (consistent) tangent


moduli in large-strain computational inelasticity*
Christian Miehe
Institut fiir Mechanik (Bauwesen) Lehrstuhl I, Universitdt Stuttgart, Pfaffenwaldring 7, Stuttgart, 70569, Germany

Received 20 April 1994; revised 7 October 1995

Abstract

An algorithm for the numerical computation of so-called algorithmic or consistent tangent moduli in finite inelasticity is
presented. These moduli determine the sensitivity of algorithmic expressions for stresses with respect to the change in total
deformation. They serve as iteration operators by application of Newton-type solvers for the iterative solution of non-linear
initial-boundary-value problems in finite inelasticity. The underlying concept of the numerical computation is a perturbation
technique based on a forward difference approximation which reduces the computation of the tangent moduli to a multiple stress
computation. The algorithmic procedure is material-independent and surprisinglysimple, it is outlined for a Lagrangian, as well
as an Eulerian geometric setting and applied to model problems of finite elasticity and finite elastoplasticity.

1. Introduction

This paper presents an algorithm for the nume~cal computation of the so-called algorithmic or
consistent tangent moduli in large-strain inelasticity. These moduli give the sensitivity of algorithmic
expressions for the stresses, resulting from incrementally objective implicit time-integration procedures,
with respect to the change in total deforma0on. Typically, the moduli serve as iteration operators by
application of Newton-type solvers for the iterative solution of non-linear initial-boundary-value
problems in finite inelasticity. They are furthermore needed in an accompanying stability analysis for
the detection of limit and bifurcation points.
The notion 'consistent linearization' can he traced back to Hughes and Pister [1] and is often used in
the context of geometrically exact (directional-derivative-based) linearization procedures and gradient
computations of tensor functions in non-linear continuum mechanics (see for exawple [2, p. 24, 3, p.
226]). These applications are restricted to the case of finite elasticity. Here, the so-called consistent
tangent moduli are identical with the elasticity moduli and represent gradients of continuous non-linear
stress functions with respect to tensorial deformation variables (see e.g. the applications in [4, 5]).
The terminology becomes more meaningful in the context of the linearization of incremental
stress-update algorithms in time-dependent problems of computational inelasticity. A typical class of
stress-update algorithm are the well-established return algorithms in convex elastoplasticity, which
result from incrementally objective integration procedures of constitutive evolution equations in
connection with the exploitation of operator-split techniques. Typical examples are documented in
[6-8], see also the references therein. In these applications, the phrase 'consistent' means 'consistent

"~Dedicated to Professor Erwin Stein on the occasion of his 65th birthday.

0045-7825/96/$15.00 ~) 1996 Elsevier Science S.A. All rights reserved


P I I S0045-7825(96)01019-5
224 C. Miehe I Comput. Methods" Appl. Mech. Engrg. 134 (1996) 223-240

with the stress-update algorithm' in the following sense: The consistent tangent moduli determine the
sensitivity of algorithmic expressions for the stresses in terms of the change in total deformation in finite
time steps. They are gladients of an incremental stress-update algorithm with respect to a tensorial
variable which describes the total deformation. In order to express this more precisely, in what follows
we will use the alternative notion of algorithmic tangent moduli. The algorithmic moduli converge in a
limiting process, if the time step tends to zero, towards the so-called continuous moduli which govern
constitutive equations for objective stress rates. Typical examples of the analytical computation of
algorithmic tangent moduli in large-strain computational polycrystal and monocrystal elastoplasticity
are, for instance, reported in [6-14], see also the references therein.
With respect to the computational speed, a code-up of the analytical closed-form expressions for the
moduli always gives the best performance. However, the analytical derivation of the algorithmic moduli
for complicated geometric and material non-linearities can be an extremely hard task. Gradients of
algorithmic stress functions for sophisticated large-strain constitutive models with anisotropy effects or
complex micromechanical substructures are almost impossibE to obtain. In many situations, the high
degree of effort required for the analytical set-up of the algorithmic moduli detracts somewhat from the
goal of computational ine!~sticity, nam,:ly the modelling of a certain physical phenomenon. Recall that
the algorithmic moduli serve only as an iteration operator which guarantees the asymptotically
quadratic rate of convergence of Newton's method. An approximation of the iteration operator, as used
for instance in quasi-Newton methods, does not change the physical result in any way, only the speed of
the iteration process. The only point where the moduli are really needed is an algorithmic stability
analysis for the detection of localized zones or global bifurcation and limit points.
This has motivated the development of a material-independent procedure for the numerical
computation of the algorithmic tangent moduli. The underlying concept is a perturbation technique
based on a forward difference approximation. Approaches of this type are often used in the context of
sensitivity analysis in non-linear programming (see for instance [15,16]). The objective of this work is
to apply these concepts to computational inelasticity for a set-up of consistent moduli within a
Lagrangian, as well as an Eulerian geometric setting. The procqdure proposed here reduces the
computation of the tangent moduli to N additional stress computations, where N denotes the dimensiun
of the stress space (N = 4 for 2-D problems and N = 6 for 3-D problems). The numerically computed
moduli can be applied in the situations mentioned above when analytical expressions are not available.
Though their set-up is more expensive, they provide the same quality with respect to the application of
Newton's method or a global and local stability analysis than their analytical counterparts.
The formulation of the perturbation technique within the large-strain framework is not trivial,
especially within the Eulerian geometric format. The following viewpoint plays a crucial role: Standard
integration algorithms for local constitutive inelastic equations are usually deformation-driven. That
means, the first computational step in a typical time-stepping process is the update of the deformation.
The following steps of the algorithm are then concerned with the update of internal variables at frozen
deformation. Based on this observation, we have chosen the deformation gradient as the primary
variable for the perturbation process. Furthermore, the observation underlines a fact which allows the
straightforward application of algorithms for the computation of elasticity moduli to the computation of
algorithmic tangent moduli in computational inelasticity: Every algorithmic setting of inelasticity can be
viewed within a typical time step as a formulation of non-linear elasticity, driven by the total
deformation.
Section 2 summarizes in a compact format some basic geometry of large-strain deformation
processes. The perturbation technique is outlined in Section 3, first for the numerical computation of
elasticity moduli in a Lagrangian, as well as the Eulerian format. These settings are generalized in
SecUon 4 to the case of the history-dependent finite inelasticity. Here, we investigate a representative
internal variable formulation and consider a typical deformation-driven time-integration algorithm.
Section 5 gives some comments on the finite element implementation within a Lagrangian, as well as a
Eulerian geometric setting. Finally, Section 6 treats model problems of finite elasticity and elasto-
plasticity which show the performance of the proposed algorithm. Here, we investigate the sensitivity of
the algorithm with respect to the perturbation parameter and relate its computational effort to the
analytical computation of the tangent moduli.
C. Miehe / Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-240 225

2. N u m e r i c a l c o m p u t a t i o n of elasticity moduli

W e first turn o u r a t t e n t i o n to non-linear elastic p r o b l e m s at large strains. A f t e r a s h o r t review o f


s o m e basic finite kinematics, we d e v e l o p algorithms for t h e n u m e r i c a l c o m p u t a t i o n o f t h e elasticity
m o d u l i in a L a g r a n g i a n , as well as an Eulerian g e o m e t r i c setting.

2. I. Basic kinematics o f large-strain deformations

Let ~ C ~ ~ be t h e ceference configuration of the b o d y of interest a n d ~p,: f ~ - - ~ R 3 t h e n o n - l i n e a r


d e f o r m a t i o n m a p at time t E ~ +. ,~,, m a p s points X E ~ of the reference configuration ~ o n t o p o i n t s
x= ¢,(X)~¢,(~) o f the c u r r e n t configuration. T h e d e f o r m a t i o n gradient F ( X ) : = V x ¢ , ( X ) with
J a c o b i a n J ( X ) : = det[F(X)] > 0 m a p s t a n g e n t vectors of material curves o n t o t a n g e n t v e c t o r s o f [he
d e f o r m e d material curves. Let, f u r t h e r m o r e , g a n d G d e n o t e t h e s t a n d a r d c o v a r i a n t m e t r i c t e n s o r s o n
t h e c u r r e n t configurational a n d the reference configuration, respectively. By application o f C a r t e s i a n
c o o r d i n a t e s we have the simple e v a l u a t i o n s g = t~h and G = BAH in t e r m s o f t h e K r o n e c k e r s y m b o l . W e
denote the tensors

C:= FTgF and c := F - r G F s (2.1)

o n t h e r e f e r e n c e configuration a n d the c u r r e n t configuration as c o n v e c t e d c u r r e n t m e t r i c a n d as


c o n v e c t e d r e f e r e n c e metric, respectively. F u r t h e r m o r e , let ~" d e n o t e t h e c o n t r a v a r i a n t E u l e r i a n
K i r c h h o f f stress tensor. T h e n

S : = F -I-rF - g (2.2)

is t h e L a g r a n g i a n s y m m e t r i c Piola stress t e n s o r Fig. 1 visualizes t h e s e geometrical relationships.

G. C, H, S g.c.h. T

1) Reference Metric b G ~ e
2) Current Metricb (:7 .-~ g
3) Internal Metrics H ,--4 h
4) Seres~l S ~-~ 1"

Fig. I. (I) Reference metric: Let G-----fiAAdenote the standard covariant metric on the reference configuration. Then,
c := F -TGF -' is called the convected reference metric. (2) Current metric: Let g ---#oh denote the standard covariant metric on
the current configuration. Then, C:= F TgF is called the convected current metric. (3) Internal metric: Let h denote a covariant
Eulerian internal variable tensor. Then, H := FThF is the associated Lagrangian internal variable tensor. (4) Stress tensor: Let ¢
denote the contravariant Kirchhoff stress tensor on the curren~ configuration work-conjugate to g. Then, the symmetric Piola
stress tensor S := F ' ¢ F - T is work-conjugate to C.
226 C. Miehe / Comput. Methods Appl. Mech. Engr8. 134 (1996) 223-240

2.2. Lagrangian setting of large-strain elasticity

Assume an elastic constitutive equation for the symmetric Piola stresses of the form

I s=e(,~)=~(¢;G) 1 (2.3)
We denote S as the Lagrangian stress function. It has the symmetry property ~,~B = ,{BA.~ is a reduced
form which satisfies a priori the principle of material frame invariance S ( Q F ) = S ( F ) V Q E S O ( 3 ) .
Here, SO(3) denotes the proper orthogonai group. The linear increment of (2.3) reads
1
AS = C : -~ [FTg AF + AF T gF] (2.4)

where

C := 2~cS(C; G) (2.5)
denote the fourth-order Lagrangian elasticity moduli. They are related to the gradient of the
Lagrangian stress function with respect to the convected current metric and have the symmetry
properties C aacn= C BAch= C ABnc. Let AC = FVg AF + AFT gF denote the increment of the con-
vected current metric. Then, (2.4) can be recast in the well-known form

I 'L
AS = C : ~- AC

The Lagrangian moduli C determine the sensitivity AS of the symmetric :"iola stresses $ with respect to
(2.6)

the increment AC of the convected current metric.


For the numerical computation of the moduli C we consider the forward difference approximation

AS(co) ~ S(F~co)) - S (2.7)

of the symmetric Piola stresses where

F~cn) = F + AF~cn) (2.8)

denotes a particular perturbed defnrmation gradient. The perturbation is assumed to have the form

AF~c D) := ; [(g-tF -*Ec) ® E n + ( g - i F -TEn) @Ec] (2.9)

where {E,~}a= t.3 denotes a covariant Lagrangian base. The choice of the perturbation parameter ~ is
commented on in Section 5.1. Insertion of (2.7) and (2.9) into (2.4) gives the representation
E
S(F ~cn )) - S ~ C : "~ (E c ® E n + E n ® E c)

By exploitation of the symmetry properties we end up with the numerical approximation

1 ^a~( r ; c n 0 - S AB] [
c A ~ ' C n ) ~ ¥ [s (2.10)
I

of the Lagrangian moduli (2.5) in component representation.


Thus, we obtain the column (CD) of the Lagrangian moduli C with row indices ( A B ) = (11), (22),
(33), (12), (23), (13) by means of ;-7~additional stress computation based on the perturbed deformation
gradient F~co) in (2.8), In the 3-D case, N = 6 stress computations are necessary for the determination
of the columns (CD) = (11), (22), (33), (12), (23), (13).
C. Miehe / Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-240 227

2.3. Eulerian setting o f large-strain elasticity

The Eulerian equations are obtained by a straightforward transformation of the Lagrangian equations
to the current configuration based on compositions with the linear tangent map F and normal map F -T.
The transformation of (2.3) gives the elastic constitutive equation for the Kirchhoff stresses

I ,=~F):~g,e) ] (2.11)
We denote ~-= F S F T as the Eulerian stress function. It has the symmetry property ?ab = ~0a. ~. satisfies
a priori the principal of material frame invariance ~(QF) = Q ~ ( F ) Q r V Q E SO(3). The transformation
of the incremental form (2.4) to the current configuration yields

A'r - (AF F - 1)~. _ 1"(AF F - l )T = c : 1 [g(AF F - ' ) + (AF F -')Wg I (2.12)

where
¢ := 20~f-(g; c) (2.13)

denote the fourth-order Eulerian elasticity moduli. They are related to the gradient of the Eulerian
stress function with respect to the current metric and have the symmetry properties c abed = c aacd = c abdc.
The Enlerian elasticity moduli are related to the Lagrangian moduli defined in (2.5) by the full
composition
cabcd = cABCD Fa A Fb BFc c Fd D

with the tangent map F. Note that AF F - : = ~ is the spatial gradient of the field ~ of incremental
deformations. Furthermore, let ~al-: = F AS F T = M" -- ~ r --rvT~o and .~Ag: = F -T AC/,,-I = g ~ +
TA
V ~ g denote the Oldroyd-type or Lie-type derivative of the Kirchhoff stress 1" and the current metric g
with respect to the incremental field ~, respectively. Then, (2.12) can be written in the well-known
form

i '
~"= c : ~" ~ a g
I (2.14)

The Eulerian moduli c determine the relative (Lie-type) sensitivity .~Ar of the Kirchhoff stresses ~ in
terms of the relative (Lie-type) increment ..~',g of the current metric.
For the numerical computation of the Eulerian moduli c, we now consider the forward difference
approximation

AT(cd) ~" T(F ~cd)) -- 7 (2.15)

of the Kirchhoff stresses based on the particular pertmbed deformation gradient


F~d) = F + AF~d) (2.16)

The perturbation is assumed to have the form

AF~e) : = 2 [(g-:e~) ® (edF) + ( g - l e d ) ® (ecF)] (2.17)

where {ea}affiL3 denotes a covariant Eulerian base. See Section 5.1 for a definition of e. Insertion of
(2.15) and (2.17) into (2.12) gives the representation

d ,
--'~ [e ~(e¢*l') + ( ' r e c ) ~ e a] ...~c :-~ [e~ ® e d + ea®e~]
228 C. Miehe / Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-240

where {ea}a= L3 denotes a covariant Eulerian base. By exploitation of the symmetry properties we end
up with the numerical approximation

cab(cd)

1
[ec ~ (ea'r) + ('rea) ~ec] -- 2 led ~ (ecT) + ('re,.) ® e a] ( 2.18 )

of the Eulerian moduli (2.13) in component representation.


We obtain the column (cd) of the Eulerian moduli c with row indices (ab) = (11), (22), (33), (12),
(23), (13) by means of an additional stress computation based on the perturbed aeformation gradient
F~ca) in (2.16). In the 3-D case, N = 6 stress computations are necessary for the determination of the
columns (cd) = (11), (22), (33), (12), (23), (13). In contrast to the Lagrangian form (2.10), onc has to
take into account in the Eulerian format (2.18) a purely geometric contribution in addition to the finite
difference term.

3. Numerical computation of algorithmic moduli

This section extends the results pointed out above to the case of finite inelasticity. Here, the main
additional effect is the history-dependence of the stress response. The first subsection gives some
general comments on the treatment of history dependence in the framework of an internal variable
formulation. Then, the numerical computation of the algorithmic tangent moduli within a Lagrangian,
as well as an Eulerian geometric setting is outlined.

3.1. History-dependence, internal variable formulation

The history dependence of the inelastic stress response is usually described by takirg into account
so-called internal variables in addition to the primary deformation variables. The evolution of these
internal variables is then determined by additional constitutive functions. Typical examples of internal
variables are the plastic strain and the back stress for the description of Bauschinger's effect in plasticity
theory or tensorial damage variables in the framework of continuous damage mechanics. A typical
example of an evolution equation is the flow rule in plasticity theory which determines the evolution of
the plastic strain.
We do not wish to restrict our con3iderations to a particular theory and consider in what follows only
a representative structure of inelastic constitutive equations. In particular, we restrict our considerations
to the presence of one tensorial internal variable. Let h denote a covariant Eulerian tensoriai internal
variable. Then its Langrangian counterpart
H:=FThF (3.t)

is obtained by a composition with the linear tangent map F. Fig. 1 visualizes this geometrical
transformation. Observe that the internal variable h transforms in the same way as the current metric g.
Thus, it makes sense to view h as an internal metric whose evolution is determined by a constitutive
evolution equation.

3.2. Lagrangian setting of large-strain inelasticity

Assume an elastic constitutive equation f6r the symmetric Piola stresses of the form
S = S(F, H) = S(C, H; G) (3.2)
which in contrast to (2.3) now depends additionally on the Lagrangian internal variable H. S satisfies
S(QF, H) = S(F, H ) V Q ~-SO(3) due to the dependence on F through the converted current metric C.
C. Miehe / Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-2449 229

As already mentioned above, the internal variable H has to be determined by an evolution equation.
Consider as a model type the first-order evolution equation

H=F.(F,H)=ff.(C,H;G) with HI, = H , = g i v e n (3.3)

which coverssimple models of non-linear viscoelasticity theory. /~ satisfies the objectivity requirement
k(QF, H) = E(F, H).
In order to obtain the algorithmic counterpart of the continuous formulation (3.2) and (3.3), we
consider a discrete time interval It,, t,,+ i] E •+ and assume all variables at time tn as known. Then

!!,+ I = H . + At[(1 - a)F.(F.,tl~) + aE'(F . . . . H . + , ) I (3.4)


is a typical integration algorithm of the first-order evolution equation (3.3). Here, At:=tn+ I --tn
denotes the time step. a • [0, 1] is an integration parameter where a = 0, 1 characterizes the first-order
explicit and implicit Euler scheme, respectively, a = 0.5 gives the second-order trapezoidal rule. The
algorithm (3.4) can be recast into the symbolic format

Ho +, = Ba'~°(F,.I, (Fn, Ho)) (3.5)


We denote /.)a~go as the deformation-driven Lagrangian state-update-algorithm. It maps, for a given
deformation gradient Fn + 1, the given history variables {Fn. H, } onto the updated internal Lag rangian
variable H n +1 at time t,+ 1. Eq. (3.5) follows trivially from (3.4) for linear evolution operators E. In the
case of non-linear evolution operators, (3.5) stands for a box which gives the result Hn+ I based on a
local iteration process.
The stress update algorithm for the symmetric Piola stresses then follows simply by insertion of the
result (3.5) into (3.2)
i ^algo - - algo
I S.+,=S (F~+t;{F.,H.})=S(F . ., U . . (F,,+ ; {Fn, n,,})) I (3.6)

We denote (3.6) as the deformation-driven Lagrangian stress-update algorithm. Clearly, it satisfies


- algo - algo
S (QFn+I; { F n , H , } ) = S (F~+,; {Fn,H,,}) due to the dependence on Fn+ I through C,,+l. The
Lagrangian stress-update algorithm is said to be incrementally objective, a notion which goes back to
Hughes and Winget [8].
A comparison of (2.3) and (3.6) reveals the following fact. In a given time interval [t,, tn+l] E R+,
the update-algorithm (3.6) has the structure of a non-linear elastic constitutive equation for the stresses
S. + i at time t~ + I. It depends only on the deformation gradient F,, +1 while the history variables {F~, H. }
are frozen. Thus, the results obtained in Section 2.2 for the numerical computation of the Lagrangian
elasticity moduli can be immediately generalized to the computation of the algorithmic moduli in
Lagrangian finite inelasticity. In particular we consider as the generalization of (2.4) the equation

E 1
AS. +, = C,~'~'; : ~- AC~+ I
1 (3.7)

for the increment of the algorithmic Piola stress where


Ca,~
o+ := 2Oco. S. ~,go(F,,+ I, {Fn,H~) ) (3.8)

denote the fourth-order Lagrangian algorithmic tangent moduli. They determine the sensitivity of the
Lagrangian stress-update algorithm (3.6) with respect to the deformation variable C,,.I.
With regard to the numerical computation of the moduli C~,t+g'~,we consider the forward difference
approximation described in :~ection 2.2 and end up with the formulation

A~<co) l [ ~ a g o A n t F ,
C~lS°.+l -e ..... ,co,,+I;(F,,H,})-S~I] (3.9)
230 C. Miehe / Comput, Methods Appl. Mech. Engrg. 134 (1996) 223-240

of the Lagrangian algorithmic tangent moduli (3.8) in component representation. Here the perturbed
deformation gradient F~col,,+t has already been specified in (2.8) and (2.9). The perturbation
parameter • is defined in (5.5). The algorithm for the computation of the algorithm moduli is
summarized in Box 1.

3.3. Ealerian setting of large-strain inelasticity

The Eulerian equations are obtained by a straightforward transformation of the Lagrangian equations
to the current configuration based on compositions with the linear tangent map F. The transformation
of (3,2) results in the elastic constitutive equation for the Kirchhoff stresses

• = ~(F, h) = ; ( g , h; c) (3A0)

depending on the Euclidean internal variable h. Due to the transformations (2.2) and (3.1), this
function obviously satisfies the requirement ~(QF, QhQ T) = Q ~ ( F , h ) Q T V Q E SO(3). The Eulerian
counterpart of the evolution equation (3.3) has the form

.~h=i(F,h)=~(g,h;c) with hit = h . = g i v e n (3.21)

•~ h := F - T H F - ' = II + hi + ITh denotes the Lie derivative or OIdroyd rate ~f the covariant Eulerian
tensor field h, formulated in terms of the spatial velocity gradient ! := # F -t = W . The Eu!erian

Box 1
Numerical computation of algorithmictangent moduli
Lagrangian geometric setting

Gradient F~. ~, history iFn, H.I and Piola stresses 5.., are given. Compote for (CD) = (I 1). (22), i33), (12), (23), (13):
1. Perturbed deformation gradient

2. Perturbed Piola stresses

s~...... = ~'"°(F~,..... : {Fn,H,))


3. Column (CD) of Langrangian moduli

Eulerian geometric setting

Gradient F~. ~. history iF., h. } and Kirchhoffstres~s s'., are given. Compute for (cd) = (11), (22), (33), (12), (23)° i 13):
1. Perturbed deformation gradient

~.~,.., "A= ~'.., "A+ Fig" F.., ~ + g'~F~.,'A


2, Perturbed Kirchhoffstresses

*~...... = ~'""(FL~,~. , ; iF.. h. } )


3. Columt~(cd) of Eolerian moduli

f I ~ 1 1 [
,:'.*; ...... ~ ~ I,~,; . . . . - ,:~.,I - ~ [ g " ' ? . , + , : % g " l - ~ Ig'~,~% + , : . , g ' ~ l
C+ Miehe / Compute Methods Appl. Mech. Engrg. 134 (1996) 223-240 231

evolution operator + is an objective function satisfying i(QF, QhQ T) = Q+(F, h)Q x. This is a result of
the geometrically well-defined relationship ~ = F-T~/'F ~ to its Lagrangian counterpart (3.3).
The algorithmic counterpart of (3.10) and (3. I l ) is obtained as follows: A composition of (3.4) with
the tangent map F,,+ j at time t,,+, yields the incrementally objective Eulerian algorithm

h,,+, = F~,(F~h,,F,,)F;,,J+,
+ At[(1 - a)F,-,++,(F~i,(F.,, h.))F,,).F,~, + aJ(F . . . . h .... )] (3.12)
This algorithm can be recast in the form
h.+~ =/~g°(F,,+~; {Fn,h.)) (3.13)
We denote/~"~g° as the deformation-driven Eulerian state-update algorithm+ For a given deformation
gradient F.+~ it maps the given history variables {F,,, h. } onto the updated Eulerian history variable
hn+,.
The stress update algorithm for the Kirchhoff stresses then follows simply by insertion of the result
(3.13) into (3.10)
[
II
algo
[ 'r,,+, = .r (F,,+,; { F . , h , , } ) = 'ffF . . . . h (F++,, {~',,, h,,})) (3.14)

We denote (3.14) as the deformation-driven Eulerian stress-update algorithm. ÷"~+° satisfies


÷+'g°(QF+ +~; {F+, h+}) = Q~'"~+°(F,+ ~; {F,,, h,, })QT. The Eulerian stress-update algorithm is said to be
incrementally objective.
A comparison of (2.11) and (3.14) reveals the fact that has already been pointed out in the
Lagrangian formulation. In a given time interval [t,, t++~] E [~+, the update algorithm (3.14) has the
structure of a non-linear elastic constitutive equation for the Kirchhoff stresses ~rn+I at time tn+ r It
depends only on the deformation gradient F,,+l while the history variables {Fn, h,} are frozen. Thus,
the results obtained in Section 2.3 for the numerical computation of the Eulcrian elasticity moduli can
be immediately transferred to the computation of the algorithmic Eulerian moduli in finite inelasticity.
The generalization of (2.4) gives the expression

I+ .....
.. , 1 : ~ .~e+g (3.15)

for the Lie-type relative increment of the Kirchhoff stress where


c~,t+g~:-- 2c~,+"'+°(F~+,; {F,, h+}) (3.16)
denote the fourth-order Eulerian algorithmic tangent moduli. They determine the sensitivity of the
stress-update algorithm (3.14) with respect to the current metric g. The Eulerian algorithmic moduli are
related to the Lagrangian moduli defined in (3.8) by the full composition
algoabcd=d-,algoABCDiu a ~. b K'
Cn+, Vn+l " n+l A" n+l B = n+lC('Fn+ldD

with the tangent map F~ + ~ at time t. +~.


With regard to the numerical computation of c~+~'~we consider the forward difference approximation
described in Section 2.3 and end up with the formulation

algo
c.,+,
ab(cd) ~ 1, - algo ahzm,+e
--'~" t+",,+,
x
t..'r0:+d).+, I -- "r,~b+,]

- [e ® (e++-,+, ) + (+'n+,e+)®eCl - +1 le + ®(e+7o+,) + (+'=+,++)®+ai (3.17)

of the algorithmic Eulerian tangent moduli (2.13) in component representation with e defined in (5.5).
232 C. Miehe / Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-240

The perturbed deformation gradient F~¢d;,+l has been specified in (2.16) and (2.17). The algorithmic
steps are summarized in Box 1.

4. R e m a r k s on the finite element implementation

This section gives some remarks on the implementation of the numerically computed algorithmic
moduli in standard displacement finite elements. We first review the governing equation of the
boundary val',e ~reblem of quasi-static finite inelasticity. Then the governing variational formulation is
outlined within the Lagrangian, as well as the Eulerian format.

4.1. Boundary-value problem of quasi-static finite inelasticity

The central field equation of quasi-static finite inelasticity is the balance of linear momentum which
assumes, within the Lagrangian parametrization in terms of the coordinates X • ~ , the form
divx[P]+~=0 in~. (4.1)

P := F$ denotes the nominal stress or the non-symmetric Piola stress. The Eulerian counterpart in
terms of a parametrization by x E ~0(~ ) reads

J div~[~-/J] + ,~ = 0 in ~0(9~). (4.2)

The symmetric Piola stress S and the Kirchhoff stress lr are determined by the algorithmic expressions
(3.6) and (3.14), "~ is a given body force field. Note that we have dropped the time index n + 1 in order
to obtain a more compact representation. Thus, in what follows all variables are understood as being
evaluated at time tn+l.
The essential boundary conditions for the deformation field and the boundary conditions for the
tractions t are

~p = ~ on F~ and t := ~-. n = t on ¢p(F,) (4.3)

where n denotes the outward normal on ~(Ft). Note that the surface of the reference configuration is
d¢composgd according to F = F~ 13 Ft and F~ f'l Ft = $.
Weak forms of the field equations (4.1) and (4.2) are obtained by a standard Galerkin procedure.
They have the general structure
Gin t - Gex, = 0 (4.4)

The so-called internal part Gin t is associated with the divergence terms in (4.1) and (4.2). The external
part Go, t results from the given body forces ~ and surface tractions i'. In what follows we turn our
attention only to the internal part.
The standard finite element displacement model discretizes G~nt and its linear increment AG,,, in a
straightforward manner on the element domain ~e C ~ . This results in a linearized discrete finite-
element equation of the form
* A
~0, + A C L =d*{_~ + ~ t ) • (4.5)
* A
Here, d and d denote the virtual and incremental element displacement vector, respectively. _rd and k dd
are called the element residual vector and elemental tangent matrix. A typical global Newton iteration
step assembles these element arrays to form the global residual vector and tangent matrix, solves the
associated linear algebraic system and updates the incremental displacements.
In what follows we take a look at the structure of the element residual vector £d and the element
tangent matrix k_~d within both the Lagrangian, as well as the Eulerian geometric format.
C. Miehe I Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-240 233

4.2. Lagrangian variational formulation

The divergence term in the weak form of the Lagrangian balance of momentum (4.1) takes the form

G,,, := S~o (gVx~o):P dV= S~ (F TgVx~O)~y


* m :S dV (4.6)
• , . .
Here '. Vxg,---~F. is the material

gradient
.
of a field ¢, of test functions , parametnzed
,
m terms of. the
material coordinates X E ~ . ~, sausfies the homogeneous boundary conditions ~o = 0 on F~. The linear
increment of (4.6) takes the form

AGim = f~e (gVx~°) :A: (gVx~°) dV

= f.~ {(Vx~ogVx~O)sym:S+(F
TA * TgVx~0)sym:C:(F
* TgVx~O)~ym)dV
L~ (4.7)

A
Vx~O~- ~LF is the gradient of a field ~ of incremental deformations, parametrized in terms of the material
coordinates X E ~ . The fourth-order tensor

[ A. . . . : = C . . . . F'*AFCc + SBDg '~c ] (4.8)

is calledz~the Lagrangian nominal modulus. It determines the increment of the nominal stress AP =
A: (gVx¢,) .
T h e finite element discretization of (4.6)2 and (4.7)2 yields the Lagrangian structure

r.:=f~ ~+SdV and k.ea:=f$,{G_+B'CB}dV (4.9)

of the element residual and the element tangent matrix in (4.5). The algorithmic expression~ of the
Lagrangian stresses _S_ and the moduli _C are given in Box 1. The matrix B_ is associated with the
discretization of the differential operators in (4.6)2 and (4.7)e, i.e.
a ~ (F T.+V,A.I)+",.. (4.1o)
G_ results from the discretization of the first p~Jrely geometric term in (4.7)2.

4.3. Eulerian variational formulation


The divergence term in the weak form of the Eulerian balance of m o m e n t u m (4.2) takes the forra

Gin,:= (gV+~o):~'dV= (gV, cp)sym :7 d V . (4.11)


• -1 *
Here., Vz~o --- ~F
. F is the spatial
, gradient
. of a field ~o of test functions , parametrized
, in terms of the
spaUal coordinates x • ~o(~ ). ~o sausfies the homogeneous boundary conditions ~, = 0 on ~0(r+). The
Lie-type increment of (4.11) takes the form

AG= S+(gV.*~o) :a: (gV.~o) dV


= f~ {(Vx~OgVx~)sy
TA * m :'r + (gVx¢)sy
* m :c:(gVx¢)sym)
A dV (4.12)
A -I A
Here, Vx~p~- AF F is the spatial gradient of a field ~, of incremental deformations, parametrized in
terms of the spatial coordinates x E ~p(~ ). The fourth-order tensor

l aabCd:----cabcd'FTbdgaCl (4.13)
234 C. Miehe / Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-240
A
is called the Eulerian nominal modulus. It governs the Eulerian nominal stress rate ~ ¢ + V.¢¢ =
a." ( g V ~ 0 ) .
The finite element discretization of (4.1 I)2 and (4.12)2 yields the Eulerian structure

F_d :'= f3aebT~r d]/ and kdd :-- +~Tcb} dV (4.14)

of the element residual and the element tangent matrix in (4.5). The algorithmic expressions of the
Eulerian stresses _¢ and moduli _care given in Box 1. The matrix _b is associated with the discretization of
the gradient operators in (4.11)2 and (4.12):, i.e.
_b~ (gVx['])hym • (4.15)

g_ results horn the discretization of the first purely geometric term in (4.12)2.
While the geometric matrix g has the identical structure to its Lagrangian counterpart in (4.9), the
matrix b does not. This results-froul a comparison of (4.15) and (4.10). In contrast to its Lagrangian
counterpart, the Eulerian b-matrix is sparse and has the identical structure in the geometric linear
theory. This is usually considered the main advantage of the Eulerian formulation.

5. N u m e r i c a l examples

The formulation described above has been implemented in the general-purpose finite-element
program FEAP, developed by R.L. Taylor and partly documented in Chapter 24 of [17]. The objective
of this section is to assess the performance of the perturbation technique for the numerical computation
of the elastic and elastoplastic tangent moduli outlined in Sections 2 and 3 by means of some typical
examples of finite strain elasticity and inelasticity.
As a model problem we apply the Eulerian stress-space formulation of multiplicative elastoplastieity
developed by Simo and Miehe [11]. The set of constitutive equations is summarized in Table 1. Here,
.~vb ~ := F O/Ot(C p- ')F T, with initial condition CP-I(t = to) = 1, is the Oldroyd rate of the elastic Finger
tensor b ~.
The macroscopic part of the free energy is assumed to be a quadratic function of the logarithmic
elastic strains, i.e. em.... (b e) = ½K In2[j e] 4- 2/~"2( ½ In[g~]) with the definitions J" := (det[b']) 1~2 and
b" := J~-'~3b'. Here, 3-,(T) := ½T : T is a second invariant of the tensor T. The microscopic part of the
free energy is assumed to he associated with a saturation-type hardening. We postulate the derivative
o~,o(a) = ( y = - y o ) ( 1 - e x p [ - a / a o ] ) + hff. The macroscopic part of the flow criterion function is
assumed to have the standard v.Mises form d~. . . . . (¢) = 3"2(dev[¢]) formulated in terms of the Kirchhoff
stress. Finally, the microscopic part is set to ~mi~o(/3)= --½(Yo +/3);.
The material parameters for the subsequent computations are summarized in Table 2. The
algorithmic counterpart of the model outlined above has been described in detail in [7]. Its basic
features are the application of an exponential integrator, yielding the standard radial-return map and

Table 1
Eulerian isotropic elastoplasticity
Elasticity
1. freeenergy 0 = 0m,c,~(b')+ ~m,,,,(¢)
2. macrostress ~"= 2 0,,0 . . . . . . (b')b"
3. micro-stress /3 = 0~0,,c,o(a)

Pla.~ticRy:
4. flow crY.teflon ~ = ~ ..... 0") + ~,~,o(/3)
s. llowrule - ',( :£ob')b: - ' = Ao ,e:..... (~)
6. evolution --~ = Af~#0m,c,o(~)
7. loading A>~0. & ~<0, A~ = 0
C. Miehe I Comput. MethodsAppL Mech. EnlJrg. 134 (1996)223-240 235

Table 2
Material parameters
Bulk modulus K 164.206 kN/mm"
Shear modulus p 80.1938 kN/mm:
Flowstress y,, 0.450 kN/mm"
Saturation stress y, 0.715 kN/mm2
Saturation parameter a0 0.0591 -
Linear hardening h 11.12924 kN/mm:

the associated algorithmic elastoplastic tangent moduli in the eigenvalue space of the elastic. Finger
tensor b ". This formulation has been implemented in a Q I / P 0 mean dilatation mixed finite element as
described in detail, for example, in [11], see also the references therein. All computations were
performed on an IBM RISC 6000-340 workstation under the UNIX operating system.

5.1. The choice of the perturbation parameter

The success of the procedure outlined in the Sections 2 and 3 depends crucially on the choice of the
perturbation parameter e in Eqs. (2.5), (2.13), (3.8) and (3.16). Clearly, the parameter e depends on
the maschine precision, (see, for instance, [15]). The commonly used maschine epsiion is defined as the
smallest positive number ~" such that we have 1 + ¢ > 1 on the computer. In a double precision code on
the IBM maschine mentioned above, we expect an accurately store up to 16 decimal digits. Thus, we
have in this situation
mascheps = 10-16. (5.1)

Now, consider a one-dimensional model problem. For the numerical computation of the gradient of a
function jr: R--, R we apply the forward difference approximation

O J ( x ) ~ g : = l [f(x + e ) - f ( x ) ] with eER+. (5.2)

A general analysis of the effect of the perturbation e on the convergence rate of Newton's method has
been outlined in [18, p. 29].
In order not to loose most of the significant digits when evaluating f(x + e) - f ( x ) in (5.2), a common
rule for the case where f is assumed to be computed accurately to maschine precision is to perturb
roughly half the digits of x. We therefore set
• = max(Ixl, typx)~/-m-~c~ps (5.3)
where typx is a user-supplied value for the typical size of x. For numerical computation of derivatives of
tensor functions f : Rn---~R by vectors x, with j = 1 . . . . . n, one defines in analogy to (5.3) a
perturbation ej = max(Ix, l, ts,px) ~ for each vector component.
With respect to the application to the gradient computations of algorithmic stress functions outlined
in Sections 2 and 3, xj is associated with a component of the current metric in the Lagrangian or
Eulerian representation, see (2.5), (2.13), (3.8) and (3.16). Thus, for standard applications in
large-strain metal plasticity and rubber elasticity, the components xj are of the same order
~7(xi) = 1. (5.4)

As a consequence, we set typx = l and combine for the problem under consideration (5.4) and (5.3) to
the constant perturbation parameter

I • "~ep,- 10 -s I (5.5)
valid for all vector components. To the experience of the author, this choice of the perturbation
parameter works. The performance is demonstrated for two model problems in the subsections below.
236 c. Miehe / Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-240

5.2. Elasticity moduli and Cook's membrane problem

In the first example we test the performance of the algorithm for the numerical computation of
Eulerian elasticity moduli outlined in Section 2. For this purpose, the part of the constitutive model in
Table 1 associated with the plasticity has been excluded.
The so-called Cook's membrane problem has been used by many authors in the context of finite
element tests. A tapered panel, clamped on one side, is subjected to a shearing load at the other end.
Fig. 2 depicts the initial geometry of the plane strain problem. In order to investigate the sensitivity of
the CPU time with respect to the scale of the problem, the panel was diseretized by two different finite
element meshes. The load F depicted in Fig. 2 was successively increased in 10 equal steps until the
final value F = 200 is reached. In each time step a Newton iteration was performed with an average of 5
iteration steps by application of the analytical elasticity moduli. Fig. 3 shows the deformed configura-
tion, discretized by a 16 x 16 and a 32 x 32 finite element mesh, respectively.
Fig. 4 depicts the load-deflection curve. As indicated in Fig. 4, in the last increment a comparison
between the analytical and the numerical computation of the elasticity moduli was carried out,
Therefore, we report in Table 3 the Euclidean norm of the residual for this typical load increment.
Here, we have investigated the sensitivity of the rate of convergence with respect to a change in the
perturbation parameter ~. The parameter has been set to • -- 10 -4, 10 -6, 10-8,10 -~° and 10 -~2. The
associated rate of convergence based on the numerical computation of the Eulerian elasticity moduli
developed in Section 2 is documented in Table 3. Observe that the quadratic rate of convergence is
obtained for the value E = 10 -8 as defined in (5.5). The comparative study of the rate of convergence in
Table 3 indicates furthermore that the iteration behaviour is of comparable quality over a wide range of
the perturbation parameter ~. Thus, the perturbation technique shows a robust performance for the
problem under consideration.
Finally, we have compared the CPU times for computations based on the analytical and the
numerical build-up (for ~ = 10 -a) of the elasticity moduli. Table 4 shows the CPU times for both
meshes depicted in Fig. 3. Clearly, the numerical computation of the elasticity moduli needs more C P U
time than the analytical computation. But the difference between the two algorithms is by no means
dramatic. Observe furthermore that the difference in the CPU time decrease with the ;.ncrease in mesh
density, because then more computational effort goes into the solving of the linear algebraic equations.

_~m r

.=

4 8......
,

Fig. 2. Cook's membrane. Geometry and boundaD' conditions.


C. Miehe / Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-240 237

~6e ..... Convergence 'rest --

"~o
~ ~ ...... "~l. . . . . ',,. . . . . . . . .Ji. . . . . . .

: '1 ,
M~h, " ~ " ' i ...... t~ . . . . . .
displmcen~nt u [mr.j

Fig. 3. Cook's membrane. Deformed configuration discretized by 16 x 16 and 32 x 32 finite element meshes, respectively.
Fig. 4. Cook's membrane problem: Load-deflection curve.

Table 3
Cook's membrane. Euclidean norm of residual
Analytical •
1 . 0 - 1 0 -4 1 . 0 . 1 0 -~ 1 . 0 . 1 0 -8 1.0.10 .,o 1 . n . 10-12

3.5078038.10 +~ 3.5078038-10"~° 3.5078038.10 ''~° 3.5078038-10"'~ 3.5078038.10÷°° 3.507-3038.10÷~"


2.9448719.10 "°° 2.8713398.10"°° 2.9441182.10"°° 2.9448610.10.oQ 2.9450461.10~°° 2.9898887-10÷°°
7.1645440.10 -°~ 4.9709660.10-03 7.1375777.10-'; 7.1641664.10-03 7.1731111• 10-03 1.1792197.10-°~
2.7729332.10 -°° 7.7668552.10-°6 2.T~10182• 10-~ 2.7720605.10 '~ 3.3730424.10-°° 3.4475860.10-04
7.5457091 - 10T M 1.1067431• 10-°~ 7.9744094-10-H 8.5647449-10-s' 2.8263690"10-1° 5.7435167"10-°°
1.9552439-10 -09 8.3269421" 10-~1 2.4247265"10-°~
8.67815"/5" 10-" 5.4063335" 10-°9
1.1850386" 10- Io

Table 4
Cook's membrane. Comparison of CPU time
Analytical Numerical Factor
Mesh 1 21 s 31 s 1.46-
Mesh 2 106 s 145 s 1.36-

5.3. Consistent elastoplastic tangent moduli and necking o f a circular bar

T h e second e x a m p l e s h o w s the p e r f o r m a n c e of the algorithm for the numerical c o m p u t a t i o n of


E u l e r i a n consistent elastoplastic tangent moduli as outlined in Section 3. H e r e , we have c h o s e n as a
s t a n d a r d b e n c h m a r k the necking of a circular bar. Numerical simulations o f this class o f p r o b l e m have
b e e n r e p o r t e d by a n u m b e r of authors. T h e g e o m e t r y depicted in Fig. 5 coincides with that considered
in Sim6 [7], and is characterized by the relation r a d i u s / l e n g t h = 6.413/53.334. T h e b a r is c l a m p e d at
the two e n d s and has a linear geometric imperfection as indicated in Fig. 5. In o r d e r to investigate the
sensitivity of the C P U time with respect to the m e s h density, we c o m p a r e again different finite-element
m e s h e s . By exploiting s y m m e t r y properties, a q u a r t e r of the specimen has b e e n discretized by 400 and
1600 4-node m e a n dilatation finite elements, respectively. Fig. 6 s h o w s the d e f o r m e d m e s h e s w h e r e the
238 C. Miehe / Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-240

i.m R

it.,
aa

--'2'6.67 ... ~ . 26.67, . , . ~ t . -


Fig. 5. Necking of a Bar. Geometry and boundary conditions.

,2 .... I-- -~- .... r ....

Mesh 1 ~ 6 ~ - ' ' ; ' .... i .... i-- ~ I


o . D i i i *

; : .... ! ~-~ M,,b ~. _1


....... .... .... - ;.

•, .... ~ .... ~ .... ~ .... , .... ~ ....

Mesh 2 d i . ~ l a c e m e . t v Imml
Fig. 6. Necking of a Bar. Deformed configuration discrelized by 400 and 1600 finite elements for a quarter, respectively.
Fig. 7. Necking of a Bar. Load-deflection curve.

necking has been developed. T h e b a r is s t r e t c h e d in a d e f o r m a t i o , i - c o n t r o l l e d numerical simulation in


60 equal time steps up to an elongation Au = 6 at both ends.
F i g . 7 d e p i c t s t h e l o a d - d e f l e c t i o n c u r v e . I n a n a l o g y t o t h e p r e v i o u s e x a m p l e w e p e r f o r m in a t y p i c a l
increment a comparison between the analytical and the numerical computation of the Eulerian
consistent tangent moduli. This has been performed at u = 5.1. In Table 5 the Euclidean norm of the
r e s i d u a l is r e p o r t e d . W e h a v e i n v e s t i g a t e d t h e s e n s i t i v i t y o f t h e r a t e o f c o n v e r g e n c e w i t h r e s p e c t t o a
c h a n g e in t h e p e r t u r b a t i o n p a r a m e t e r E. T h e r e f o r e , t h e v a l u e s E = 10 - 5 , 10 - 6 , 10 - s , 10 - I ° a n d 1 0 - H

Table 5
Necking of a Bar. Euclidean norm of residual
Analy' cal •
1.0" 10 -5 1.0-10 ,6 1.0-10 -s 1.0-10 -~° 1.0.10 -~t
2.1570763-10 .02 2.1570723-10 *°2 2.1570759" 10 .02 2.1570763" 10 "°- 2.1570762' 10 *°z 2.1570792-10 +o2
5.4938768" 10 -°* 2.9264975" 10 "°° 5.1059855" 10 -°~ 5.4898057" 10 -°' 5.5209713-10 -°~ 5.8956062-10 -°]
1.2014516-10 '': 3.3537173-10'c° 1.3243887-10 -°l 1.2642307.10 -02 1.2623314" 10 -02 7.5685545.10 -02
1.0441766.10 -03 9.1795933" 10 -°j 6.4469867-10 -03 1.0441490-10 -03 1.0438109.10 -03 4.1717881 - 10 -03
1.5968715-10 -05 3.0458427" 10 -°~ 1.0427058" 10 -03 1.5968251" 10 -05 1.5994691" 10 -05 1.0397134-10 -°3
6.8235590' 10 T M 4.4147493.10 -02 1.3701855" 10 -°s 7.4568860.10 " 2.6604728.10 -'~ 1.6476999.10 -05
4.8842518.10 -°s 3.1746328.10 -°~ 1.2958148.10 -*t 1.5266026.10 -°7
8.6479103-10 -°4 2.9501753-10 -II 2.2629079.10 -09
1.4074464.10 -as 3.2074082.10- H
8.4850021.10 .07
5.7739691.10 -°s
3.9643778-10 -°~
2.7373384.10- Io
2.2410024-10 -~
C. Miehe / Comput, Methods Appl. Mech. Engrg. 134 (1996) 223-240 239

Table 6
Necking of a Bar. Comparison of CPU time
Analytical Numerical Factor
Mesh 1 265 s 484 s 1.83-
Mesh 2 1232 s 23(~)s 1.86-

have b e e n considered. T h e associated residuals are reported in Table 3. T h e o p t i m u m value is again


= 10 -s as defined in (5.5), where the rate of convergence is quadratic and almost identical with the
analytical computation. T h e comparative study of the rate of convergence in Table 5 shows a robust
p e r f o r m a n c e for the problem over a wide range of the p a r a m e t e r E.
C P U times for computations based on the analytical and the numerical c o m p u t a t i o n (¢ = 10 - s ) o f the
algorithmic tangent moduli are d o c u m e n t e d in Table 6 for the two m e s h e s depicted in Fig. 6. T h e
numerical computation o f the consistent tangent moduli needs m o r e C P U time than the analytical
c o m p u t a t i o n . T h e factor is almost invariant for the case of the two m e s h e s u n d e r consideration.

6. Conclusion

A n algorithm for the numerical computation o f the so-called consistent tangent moduli in finite-strain
inelasticity has b e e n presented. These moduli d e t e r m i n e the sensitivity o f algorithmic expressions for
the stresses with respect to the change in total d e f o r m a t i o n and serve as iteration o p e r a t o r s by
application o f N e w t o n - t y p e solvers. T h e algorithmic p r o c e d u r e ~3 m a t e r i a l - i n d e p e n d e n t and can b e
applied if the gradients o f complicated algorithmic stress functions are not available. This can be t h e
case for extremely complicated constitutive functions or during the d e c e l o p i n g process o f a finite
e l e m e n t for a particular material model. T h e underlying concept of the numerical computation is a
perturbation technique based on a forward difference approximation. This p r o c e d u r e reduces the
c o m p u t a t i o n o f the tangent moduli to a multiple stress computation. T h e algorithm has b e e n f o r m u l a t e d
for the Lagrangian, as well as the Eulerian geometric setting and tested for a m o d e l p r o b l e m s o f finite
elasticity and elastoplasticity.

References

[l I T.J.R. Hughes and K.S. Pister, Consistent linearization in mechanics of solids and structures, Comput. Struct. 8 (1978)
391-397.
[2] C. Truesdell and W. Nofi, The nonlinear field theo6es of mechanics, in: S. F10gg¢, ed., Handbuch der Physik Bd. Ill/3
(Springer-Verlag, Berlin, 1965).
[3] J.E. Marsden and T.J.R. Hughes, Mathematical Foundations of Elasticity (Prentice Hall, Englewood Cliffs, NJ, 1983).
[4] C. Miehe, Aspects of the formulation and finite element implementation of large strain isotropic elasticity, Int. J. Numer.
Methods Engrg. 37 (1994) 1981-2004.
[5] C. Miehe, Entropic thermoelasticity at finite strains. Aspects of the formulation and numerical implementation, Comput.
Methods Appl. Mech. Engrg. 120 (1995) 243-269.
[6] J.C. Sire6 and C. Ortiz, A unified approach to the finite deformation elastoplastic analysis based on the use of hyperelastic
constitutive equations, Comput. Methods Appl. Mech. Engrg. 49 (1985) 221-245.
[7] J.C. Sire6, Algorithms for static and dynamic muhiplicative plasticity that preserve the classical return mapping schemes of
the infinitesimal theory, Comput. Methods Appl. Mech. Engrg. 99 (1992) 61-112.
[8] A.M. Cuiti~o and M. Ortiz, Computational modeling of single crystals, in: D.R.J. Owen, E. Onate and E. Hinton, eds.,
Computational Plasticity--Fundamentals and Applications (Pineridge Press. Swansea, 1992) 57-76.
[9] J.C. Nagtegaal, On the implementation of inelastic constitutive equations with special reference to large deformation
problems, Comput. Methods Appl. Mech. Engrg. 33 (1982) 469--484.
[10] D. Peric, D.R.J. Owen and M.E. Honner, A model for finite strain elasto-plasticity based on logarithmic strains:
Computational Issues, Comput. Methods Appl. Mech. Engrg. 94 (1992) 35-61.
[11] J.C. Sire6 and C. Miehe, Associative coupled thermoplasticity at finite strains: Formulation numerical analysis and
implementation, Comput. Methods Appl. Mech. Engrg. 98 (1992) 41-104.
240 C. Miehe / Comput. Methods Appl. Mech. Engrg. 134 (1996) 223-240

[12] A.M. Cuitifio and M. Orliz, A material-independent method for extending stress update algorithms from small-strain
plasticity to finite plasticity with multiplicative kinematics, Engrg. Comput. 9 (1992) 437-451.
[13] C. Miehe, Multisurface thermoplasticity for single crystals at large strains in terms of eulerian vector updates, lot. J. Solids
Struct. (1994), in press.
[14] C. Miehe, Exponential map algorithm for stress updates in anisotropic multiplicarive elastoplasticity for single crystals, Int.
J. Numer. Methods Engrg. (1994), in press.
[15] J.E. Dennis and R.B. Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations (Prentice
Hall, Englewood Cliffs, NJ, 1978).
[161 D.G. Luenberger, Linear and Nonlinear Programming, 2nd edition (Addison-Wesley, Reading, MA, 1984).
[17] O.C. Zienkiewicz and R.L. Taylor, The Finite Element Method, Vols. 1 and 2 (McGraw-Hill, London, 1989).
[18] T..I.R. Hughes and J. Winget, Finite rotation effects in numerical integration of rate constitutive equations arising in
large-deformation analysis, Int. J. Numer. Methods Engrg. 15 (1980) 1862-1867.

You might also like