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

net/publication/43939530

Numerical approaches to rate-independent processes and applications in

inelasticity

Article in ESAIM Mathematical Modelling and Numerical Analysis · May 2009

DOI: 10.1051/m2an/2009009

CITATIONS READS

50 30

2 authors:

Alexander Mielke Tomas Roubicek

Weierstrass Institute for Applied Analysis and Stochastics Charles University in Prague
317 PUBLICATIONS 12,343 CITATIONS 269 PUBLICATIONS 4,666 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Rate-Independent Systems: Theory and Application View project

Computational nonlinear mechanics of solids View project

All content following this page was uploaded by Tomas Roubicek on 22 April 2015.

The user has requested enhancement of the downloaded file.

 Weierstraÿ-Institut
für Angewandte Analysis und Sto hastik
im Fors hungsverbund Berlin e.V.

Preprint ISSN 0946  8633

Numeri al approa hes to

rate-independent pro esses and

appli ations in inelasti ity

Alexander Mielke1,2 and Tomá² Roubí£ek3,4

1
Weierstraÿ-Institut für Angewandte Analysis und Sto hastik, Mohrenstraÿe 39,
10117 Berlin, Germany, E-Mail: mielkewias-berlin.de
2
Institut für Mathematik, Humboldt-Universität zu Berlin,
Rudower Chaussee 25, 12489 Berlin-Adlershof, Germany
3
Mathemati al Institute, Charles University, Sokolovská 83, CZ18675 Praha 8,
Cze h Republi , E-Mail: tomas.roubi ekm. uni. z
4
Institute of Information Theory and Automation, A ademy of S ien es,
Pod vodárenskou v¥ºí 4, CZ18208 Praha 8, Cze h Republi

submitted: 17th O tober 2006

No. 1169
Berlin 2006

W I A S
2000 Mathemati s Subje t Classi ation. 35K55, 35K85, 49J40, 49S05, 65J15, 74C05, 74F15,
74H15, 74H20, 74N10, 74R05, 74S05. .
Key words and phrases. Rate-independent evolution, energeti solution, approximation, plas-
ti ity, damage, debonding, magnetostri tion, martensiti transformation.
Edited by
Weierstraÿ-Institut für Angewandte Analysis und Sto hastik (WIAS)
Mohrenstraÿe 39
10117 Berlin
Germany

Fax: + 49 30 2044975
E-Mail: preprintwias-berlin.de
World Wide Web: http://www.wias-berlin.de/
 Abstra t
A general abstra t approximation s heme for rate-independent pro esses in
the energeti formulation is proposed and its onvergen e is proved under var-
ious rather mild data quali ations. The abstra t theory is illustrated on sev-
eral examples: plasti ity with isotropi hardening, damage, debonding, mag-
netostri tion, and two models of martensiti transformation in shape-memory
alloys.

1 Introdu tion

Rate independent pro esses o ur (after ertain, and usually ne essary, simpli-
ations) in various physi al (mainly me hani al but not only) systems exhibiting
hystereti response during isothermal evolution pro esses. Mathemati al analysis of
su h pro esses, based on the notion of energeti solutions introdu ed in [40, 42℄, has
been intensively s rutinized and develop in parti ular in [31, 35, 36, 37, 38, 41, 43, 53℄.
However, ex ept for some parti ular attempts [7, 18, 32, 54℄, there has been no nu-
meri al analysis developed for su h pro esses so far.
This paper lls the gap of a universally-appli able numeri al s heme in the ontext of
rate-independent pro esses and its analysis. After introdu ing the energeti formu-
lation in Se t. 2, a fairly general on eptual numeri al dis retization is proposed and
its onvergen e is analyzed in Se t. 3. Then, in Se t. 4, the generality is redu ed
to problems based on Bana h spa es and with dissipation distan es governed by
degree-1 homogeneous potentials, whi h in turn allows for various spe i onstru -
tions dire tly appli able in on rete situations. This is demonstrated in Se t. 5 on
various examples from ontinuum me hani s of deformable bodies, namely plasti ity
with hardening, two models of martensiti transformation, damage, debonding, and
magnetostri tion.
In parti ular, it a ompanies a large variety of existing models by on eptual nite-
element dis retizations supported by rigorous analysis as far as onvergen e on-
erns, and in some ases oers new results or improves known results as far as mere
existen e of solutions on erns.

2 An abstra t setting: energeti solution

We onsider a state spa e Q (independent of time) as a topologi al spa e. Typi ally,

it is subset of a lo ally onvex spa e. We will distinguish between a non-dissipative

1
 omponent u ∈ U and a dissipative omponent z ∈ Z of the state q = (u, z) ∈
Q := U × Z .
For a xed time horizon T > 0, we onsider a Gibbs-type stored energy E : [0, T ] ×
Q → R ∪ {+∞}. The further ingredient is a (time-independent but not ne essarily
symmetri ) dissipation distan e D : Z × Z → R ∪ {+∞} whi h will later determine
the dissipated energy and whi h is assumed to satisfy

∀z1 , z2 , z3 ∈ Z : D(z1 , z1 ) = 0 & D(z1 , z3 ) ≤ D(z1 , z2 ) + D(z3 , z3 ). (2.1)

Let us agree to write o asionally D(q1 , q2 ) with the meaning D(z1 , z2 ) for q1 =
(u1 , z1 ) and q2 = (u2 , z2 ).
In ase of Q having a linear stru ture, D(z1 , z2 ) := R(z2 − z1 ) (as in Se t. 4 below)
and onvexity of both E and R, we want to address an evolution of q = q(t) governed
by the doubly nonlinear in lusion
 ∂q 
∂R + ∂q E(t, q) ∋ 0 (2.2)
∂t
where  ∂  denotes the subdierential. Under some additional quali ation, it is
equivalent (see [36, 41℄) to the energeti formulation based on Denition 2.1 below
whi h, however, works under mu h weaker data quali ation where (2.2) loses any
sense. In fa t, this denition is based on a global-minimization hypothesis ompeting
with the maximum-dissipation prin iple (or rather Levitas' realizability prin iple
[33℄). In mathemati al terms, it means stability

(2.3)



∀q̃ ∈ Q : E t, q(t) ≤ E(t, q̃) + D q(t), q̃ ,

and energy equality

Z t
E(t, q(t)) + VarD (q; s, t) = E(s, q(s)) + P(r, q(r)) dr, (2.4)
s

where
∂
P(t, q) := E(t, q) and (2.5)
∂t
j
X
(2.6)


VarD (q; s, t) := sup D q(ti−1 ), q(ti )
i=1

with the supremum taken over all j ∈ N and over all partitions of [s, t] in the form
s = t0 < t1 < ... < tj−1 < tj = t. The parti ular terms in (2.4) represent the stored
energy at time t, the energy dissipated by hanges of the internal variable during
the time interval [s, t], the stored energy at the initial time s, and the work done by
external loadings during the time interval [s, t]; P is then the power.

Denition 2.1 The pro ess q : [0, T ] → Q is alled an energeti solution to the
initial-value problem given by the triple (E, D, q0) if

2
(i) it is stable in the sense that (2.3) holds for all t ∈ [0, T ],
(ii) the energy balan e (2.4) holds for any 0 ≤ s < t ≤ T , in parti ular
t 7→ P(t, q(t)) L1 (0, T ), and
is in
(iii) the initial ondition q(0) = q0 holds.

For the analysis of the rate-independent problems, it is onvenient to introdu e the

sets of stable states S(t) for any t ∈ [0, T ] by putting

(2.7)

S(t) := q ∈ Q; E(t, q) < +∞ & ∀q̃ ∈ Q : E(t, q) ≤ E(t, q̃) + D(q, q̃) .

This allows us to re ast the stability ondition (i) in Denition 2.1 in the form
q(t) ∈ S(t) for all t ∈ [0, T ]. Yet, more importantly, we may address losedness
properties of S(t).
In Se t. 4, we will spe ialize this setting by introdu ing an additional linear stru ture,
i.e. Q will be (a subset of) a Bana h spa e equipped with the weak or the norm
topology. This will allow us to make the abstra t properties more spe i .

3 An abstra t approximation

For an abstra t approximation, we onsider three positive parameters τ , h, and

ε. Here τ > 0 represents the neness of a time dis retization by a partition (not
ne essarily equidistant) of the time interval [0, T ]. The parameter h > 0 denotes a
spatial dis retization of the state spa e Q by a subset Qh again having the stru ture
Qh := Uh × Zh . Moreover, ε > 0 is used for a possible approximation of the
fun tionals E and D to be implemented more easily when restri ted on Qh (see also
Remark 3.10 below) or just to guarantee the onvergen e in some more ompli ated
ases. Typi ally, a penalization of some onstraints may be S involved by this way,
f. Se t.5.S These last approximations lead to Eε : [0, T ] × h>0 Qh → R ∪ {+∞}
and Dε : h>0 (Zh × Zh ) → R ∪ {+∞}.
Using the indi ator fun tion δQh : Q → {0, +∞}, i.e. δQh = 0 on Qh and δQh = +∞
on Q \ Qh , it will o asionally be onvenient to introdu e the restri tion to Qh also
by repla ing Eε and Dε respe tively by

Eε,h = Eε + δQh and Dε,h : (q, q̃) 7→ Dε (q, q̃) + δQh (q) + δQh (q̃). (3.1)

3.1 Basi assumptions

We rst olle t a few basi assumptions. We assume (2.1) also for ea h Dε , i.e. for
all ε > 0:

∀z1 , z2 , z3 ∈ Z : Dε (z1 , z1 ) = 0 & Dε (z1 , z3 ) ≤ Dε (z1 , z2 ) + Dε (z3 , z3 ). (3.2)

3
For proving existen e results we will need the following lower semi- ontinuity and
ompa tness results:

∀ ε, h > 0 : Dε : Qh ×Qh → R∞ are lower semi- ontinuous, (3.3)

∀ ε, h > 0 ∀ t ∈ [0, T ] ∀ a ∈ R :
the sublevels { q ∈ Qh ; Eε (t, q) ≤ a } are sequentially ompa t in Q. (3.4)

To pass to the limit will need a uniform inf- ompa tness of the olle tion
(Eε,h(t, ·))ε,h>0, t∈[0,T ] :

∀ a ∈ R ∀ ε, h > 0, θ ∈ [0, T ], qh,ε ∈ Qh : Eε (θ, qh,ε ) ≤ a

=⇒ ∃ q ∈ Q ∃ subsequen e {qhn ,εn }n∈N : q = lim qhn ,εn . (3.5)
n→∞

Next we need a  Γ-liminf estimate for the family (Dε )ε>0 on (Qh ×Qh )h>0 in the
limit ε, h → 0:

z∈Z, zh,ε ∈Zh , z = lim zh,ε 
(h,ε)→(0,0)
⇒ D(z, z̃) ≤ lim inf Dε (zh,ε , z̃h,ε ). (3.6)
z̃∈Z, z̃h,ε ∈Zh , z̃ = lim z̃h,ε  (h,ε)→(0,0)
(h,ε)→(0,0)

The limit fun tional D has to satisfy a positivity ondition:

)
∀ z ∈ Z ∀ K ⊂ Z sequentially ompa t ∀ zn ∈ K :
lim min{D(zn , z), D(z, zn )} = 0 ⇒ z = lim zn . (3.7)
n→∞
n→∞

Like for Dε we also need a  Γ-liminf estimate for the family (Eε (t, ·))ε>0, t∈[0,T ] on
(Qh )h>0 :

∀ q∈Q ∀ qh,ε ∈Qh with q = lim qh,ε : E(t, q) ≤ lim inf Eε (θ, qh,ε ). (3.8)
(h,ε)→(0,0) (h,ε,θ)→(0,0,t)

Note that (3.6) and (3.8) are only lower Γ-liminf estimates for (Dε,h )ε,h>0 and
(Eε,h(t, ·))ε,h>0, t∈[0,T ] . The orresponding upper estimates are onsequen es of the
entral ondition (3.16) whi h postulates the existen e of joint re overy sequen es.
So far all onditions above relate to stati on epts. The next three onditions
relate to the time dependen e, whi h involves the power of external for es Pε (t, q) =
∂
E (t, q). The rst assumption provides a uniform energeti ontrol of the power
∂t ε
Pε , viz.,
 
∃ c0 , c1 ∈ R ∀ ε, h > 0 ∀q ∈ Qh : ∃t0 ∈ [0, T ] : Eε (t0 , q) < +∞ =⇒
Eε (·, q) ∈ C 1 ([0, T ]) and (3.9a)
(3.9b)


∀ t ∈ [0, T ] : Pε (t, q) ≤ c1 Eε (t, q)+c0 .

Using a Gronwall estimate we immediately obtain the growth restri tions

∀ s, t ∈ [0, T ] : Eε (s, q) + c0 ≤ ec1 |t−s| Eε (t, q)+c0 . (3.10)

4
The se ond assumption is a onditioned (with respe t to sublevels of E ) equi- (with
respe t to q ) uniform (with respe t to t) ontinuity of P(·, q):

∀ a ∈ R ∀σ > 0 ∃ δ > 0 ∀s, t ∈ [0, T ] ∀q ∈ Q :

if E(0, q) ≤ a and |t−s| < δ, then P(s, q) − P(t, q) < σ. (3.11)

The third assumption on Pε,h on erns the onvergen e of Pε,h for ε, h → 0. It is a

 ontinuous onvergen e but onditioned by the fa t that the onsidered arguments
are in the asso iated sets of stable states

Sε,h (t) := q ∈ Qh ; Eε (t, q) < +∞ &
∀q̃ ∈ Qh : Eε (t, q) ≤ Eε (t, q̃) + Dε (q, q̃) , (3.12)

and that the energies are bounded:

If (εn , hn , tn ) → (0, 0, t), qn ∈ Sεn ,hn (tn ), qn → q, and

sup Eεn,hn (tn , qn ) < +∞, then lim Pεn (tn , qn ) = P(t, q). (3.13)
n∈N n→∞

Re all that Dε and D only depend on the z - omponent of q = (u, z) and we have
agreed to write o asionally, as e.g. in (3.12), Dε (q, q̃) in the meaning of Dε (z, z̃).
An essential ingredient for the onvergen e analysis is the abstra t version of Helly's
sele tion prin iple, whi h has been proved in the Appendix of [39℄ generalizing [35,
Theorem 3.2℄.

Lemma 3.1 (Abstra t Helly's sele tion prin iple [39℄.) Under the onditions
(2.1), (3.6) and (3.7), for every sequen e zn : [0, T ] → Z , n ∈ N satisfying

∃C > 0 ∀ n ∈ N : VarDεn ,hn (zn ; 0, T ) ≤ C, (3.14a)

∃K ⊂ Z sequentially ompa t ∀ n ∈ N ∀ t ∈ [0, T ] : zn (t) ∈ K, (3.14b)

there exists a subsequen e (znj )j∈N , a nonde reasing fun tion D : [0, T ] → R, and a
limit pro ess z : [0, T ] → Z su h that we have

∀ t ∈ [0, T ] : z(t) = lim znj (t), D(t) = lim VarDεn (znj ; 0, t), and (3.15a)
j→∞ j→∞ j ,hnj

∀ s, t ∈ [0, T ] with s≤t: VarD (z; s, t) ≤ D(t) − D(s). (3.15b)

Remark 3.2 (Weakening (3.13) on Bana h spa es.) In the appli ations presented
in this paper we will not make use of the full strength of the  onditioned ontinuous
onvergen e. However, we refer to [14℄, where a setting is onsidered where Q is a
Bana h spa e equipped with its weak topology. It is shown that the assumptions
in (3.13) rst imply the energy onvergen e Eεn ,hn (tn , qn ) → E(t, q). This, together
with the weak onvergen e qn ⇀ q , an then be used to improve the onvergen e
into the strong onvergen e. Hen e, in that ase the onditioning implies that only
strongly onvergent sequen es have to be onsidered for the ontinuous onvergen e
in (3.13).

5
3.2 Stability of sets of stable states

All the assumptions of the previous subse tion are either on the family (Dε,h)ε,h>0 or
on the family (Eε,h)ε,h>0. The nal ondition links the behavior of these two families
and thus provide the upper Γ-limit estimates whi h are needed to omplement the
lower Γ-limit estimate for D in (3.6) and for E in (3.8). Sometimes, in parti ular
when some holonomi -type onstraints are involved in E , it o urs that a onvergen e
riterion of the type h ≤ H(ε), for some H : R → R monotone and satisfying
+ +

H(ε) → 0 for ε → 0, is needed.

The following entral ondition states the existen e of a joint re overy sequen e
under suitable quali ations:
∀ q, q̃ ∈ Q ∀tn ∈ [0, T ] with tn → t ∀ εn , hn → 0+ with hn ≤ H(εn )
∀qn ∈ Sεn ,hn (tn ) with qn → q and supn∈N Eεn ,hn (tn , qn ) < +∞
∃ q̃n ∈ Qhn with q̃n → q̃ :
 
lim sup Eεn ,hn (tn , q̃n )+Dεn ,hn (qn , q̃n )−Eεn,hn (tn , qn )
n→∞
≤ E(t, q̃)+D(q, q̃)−E(t, q). (3.16)

The following assertion says, in other words, that the graph of the set-valued
mapping S : [0, T ] ⇉ Q ontains Kuratowski's limes superior of the graphs of
Sε,h : [0, T ] ⇉ Qh at least if restri ted to states with bounded energy as in (3.5) and
if h ≤ H(ε) is taken into a ount. This upper semi ontinuity result establishes a
ertain stability of sets of stable states that is ru ial for the onvergen e analysis.

Lemma 3.3 (Conditioned upper semi- ontinuity of the sets of stable

states.) Let (3.8) and (3.16) hold and tn , εn , hn , qn and q = lim qn be as in (3.16).
n→∞
Then q ∈ S(t).

Proof. By (3.8), we have

E(t, q) ≤ lim inf Eεn ,hn (tn , qn ) ≤ sup Eεn ,hn (tn , qn ) < +∞, (3.17)
n→∞ n∈N

where the last inequality is assumed in (3.16). Next, for q̃ ∈ Q arbitrary, hoose
q̃n ∈ Qhn as in (3.16). By denition (3.12), qn ∈ Sεn ,hn (tn ) says that Eεn,hn (tn , q̃n ) +
Dεn ,hn (qn , q̃n ) −Eεn,hn (tn , qn ) ≥ 0. Using now the limsup estimate in (3.16) we obtain
0 ≤ lim sup Eεn ,hn (tn , q̃n ) + Dεn ,hn (qn , q̃n ) − Eεn,hn (tn , qn )
n→∞
≤ E(t, q̃) + D(q, q̃) − E(t, q). (3.18)
Sin e q̃ was arbitrary, denition (2.7) gives q ∈ S(t). 2

Remark 3.4 (Weakening of (3.16).) In this proof the ondition q̃n → q̃ was not
used. Thus, in prin iple assumption (3.16) ould be weakened by dropping this ad-
ditional request. However, in doing so, the limsup estimate in (3.16) degenerates in

6
the sense that the two sides in this estimate no longer depend on ea h other. In fa t,
the best hoi e for making the left-hand side small is, by re alling stability, the hoi e
q̃n = qn , whi h makes ea h member in the sequen e identi al 0. Sin e this is inde-
pendent of q̃ , the weakened ondition (3.16) just means 0 ≤ E(t, q̃)+D(q, q̃)−E(t, q),
whi h is the desired stability of q . As we will see in the appli ations in Se tion 5,
the strengthened ondition is useful, sin e properly hosen joint re overy sequen es
allow us to prove


0 ≤ Eεn ,hn (tn , q̃n )+Dεn ,hn (qn , q̃n )−Eεn ,hn (tn , qn ) → E(t, q̃)+D(q, q̃)−E(t, q),

from whi h we then on lude stability. See [39℄ for more dis ussion of this point.

Example 3.5 Quite typi al way how the quali ation (3.16) an be ensured is the
situation when Dε,h onverges ontinuously to D in the sense

lim Dε (qh,ε , q̃h,ε ) = D(q, q̃) (3.19)

ε→0, h→0
qh,ε →q, q̃h,ε →q̃
qh,ε ,q̃h,ε ∈Qh

and, in addition,

∀q̃ ∈ Q ∀h, ε > 0 ∃q̃h,ε ∈ Qh : lim q̃h,ε = q̃ and

(h,ε)→(0,0)

lim sup Eε (θ, q̃h,ε ) ≤ E(t, q̃). (3.20)

h≤H(ε)
(h,ε,θ)→(0,0,t)

Then (3.16) holds: indeed, it su es to sum (3.20) used for q̃n = q̃hn ,εn with (3.19)
used for qn = qhn ,εn and q̃n = q̃hn ,εn and subtra t (3.8) used for qn = qhn ,εn , and
eventually estimate the sum of limits superior from below by limits superior of the
sum. Let us still remark that (3.20) together with (3.8) is just the onditioned
Γ- onvergen e (sometimes also alled epi- onvergen e) of the olle tion (Eε,h(θ, ·) +
δQh )h,ε>0, θ∈[0,T ] to E if (h, ε, θ) → (0, 0, t) onditioned by h ≤ H(ε).

3.3 Approximate solutions

We onsider now τ > 0, and a partition 0 = t0τ < t1τ < ... < tkτ τ = T with

tτi − tτi−1 ≤ τ for i = 1, ..., kτ . (3.21)

We do not assume this partition to be equidistant. Further, we onsider an ap-

proximation [q0 ]h,ε of the initial ondition q0 and the following re ursive in remental
formula: we put qτ,h,ε
0
= [q0 ]h,ε a given initial ondition, and, for k = 1, ..., kτ we
dene qτ,h,ε , an approximation of a solution at time tkτ , to be any solution of the
k

minimization problem

Minimize Eε,h(tkτ , q) + Dε,h (zτ,h,ε

k−1
)
, z)
(3.22)
subje t to q = (u, z) ∈ Qh .

7
We dene the approximate solution qτ,h,ε : [0, T ] → Q as a pie ewise onstant
approximation, namely
for tk−1
( k
qτ,h,ε τ < t ≤ tkτ , k = 1, ..., kτ ,
qτ,h,ε (t) := (3.23)
0
qτ,h,ε = [q0 ]h,ε for t = 0.
R : [0, T ] → Q with
We also need the retarded approximate solution qτ,h,ε

qτ,h,ε for tk−1

( k
R τ ≤ t < tkτ , k = 1, ..., kτ ,
qτ,h,ε (t) := (3.24)
kτ
qτ,h,ε for t = T,

Proposition 3.6 (Dis rete stability and energy inequalities.) Let (3.2), the
lower semi ontinuity (3.3)(3.4) of the approximate stored and the dissipated ener-
gies, and smoothness of external for ing (3.9a) hold. Then (3.22) has a solution
k
qτ,h,ε for any k = 1, ..., kτ and qτ,h,ε is stable in the sense

qτ,h,ε (t) ∈ Sε,h (tkτ ) for any t ∈ (tk−1

τ , tkτ ], k = 0, ..., kτ , (3.25)
and satises the dis rete upper energy inequality
s
∂Eε,h R (t)
dt
Z
Eε,h (s, qτ,h,ε(s)) + VarDε,h (qτ,h,ε ; r, s) − Eε,h (r, qτ,h,ε (r)) ≤ t, qτ,h,ε
r ∂t
(3.26)
for r = tkτ 1 and s = tkτ 2 with any k1 , k2 ∈ N ∪ {0}, 0 ≤ k1 ≤ k2 ≤ kτ , as well as a
similar dis rete lower energy inequality
s
∂Eε,h
Z


Eε,h (s, qτ,h,ε(s)) + VarDε,h (qτ,h,ε ; r, s) − Eε,h (r, qτ,h,ε (r)) ≥ t, qτ,h,ε (t) dt
∂t
r
(3.27)
for r = tkτ 1 and s = tkτ 2 but now only with k 1 , k 2 ∈ N, 1 ≤ k 1 ≤ k 2 ≤ k τ .

Proof.The existen e of qτ,h,ε

k
solving (3.22) follows from (3.3) and (3.4) via a re ursive
R exist, too.
argument for k = 1, ..., kτ . Hen e qτ,h,ε and qτ,h,ε
The dis rete stability ondition (3.25) follows by using su essively that qτ,h,ε
k
is a
solution to (3.22) and the triangle inequality (2.1) for Dε,h:
Eε,h (tkτ , qτ,h,ε
k
) ≤ Eε,h (tkτ , q̃) + Dε,h (qτ,h,ε
k−1 k−1
, q̃) − Dε,h (qτ,h,ε k
, qτ,h,ε )
≤ Eε,h (tkτ , q̃) + Dε,h (qτ,h,ε
k
, q̃) (3.28)
for any k = 1, ..., kτ .
As to (3.26), we again use that qτ,h,ε k
solves (3.22) and, omparing it with qτ,h,ε k−1
, we
get
Eε,h (tkτ , qτ,h,ε
k
) − Eε,h tk−1 k−1 k−1 k



τ , qτ,h,ε + Dε,h qτ,h,ε , qτ,h,ε
Z tkτ k−1
∂Eε,h (t, qτ,h,ε )
k k−1 k−1 k−1
(3.29)



≤ Eε,h tτ , qτ,h,ε − Eε,h tτ , qτ,h,ε = dt.
tk−1
τ
∂t

8
Now the estimate (3.26) follows after a summation for k = k1 +1, ..., k2. As to the
estimate (3.27), by the stability (3.28) written for qτ,h,ε
k−1 k
q̃ = qτ,h,ε , we nd

Eε,h tkτ , qτ,h,ε

k
− Eε,h tk−1 k−1 k−1 k




τ , qτ,h,ε + Dε,h qτ,h,ε , qτ,h,ε
Z tkτ k
∂Eε,h (t, qτ,h,ε )
k k k−1 k
(3.30)



≥ Eε,h tτ , qτ,h,ε − Eε,h tτ , qτ,h,ε = dt.
tk−1
τ
∂t

By a summation for k = k1 +1, ..., k2, we obtain (3.27). 2

Remark 3.7 (Approximation of initial onditions.) Note that (3.30) does not work
for k = 1 be ause we (intentionally) did not assume numeri al stability of the
approximate initial ondition, i.e. [q0 ]h,ε ∈ Sε,h (0) whi h would only very hardly be
guaranteed in on rete numeri al s hemes. This is also why (3.27) does not hold
with r = 0, unlike (3.26).

3.4 Convergen e of the approximate solutions

Now we investigate the asymptoti s for τ → 0, h → 0, and ε → 0. Like for

spa e dis retization, we do not assume the partition of the time interval [0, T ] to be
nested, but we assume that both time and spa e dis retization renes when τ → 0
and h → 0, respe tively. Namely (3.21) for the time dis retization while, for the
spatial dis retization, this renement requirement is impli itly ontained in (3.16);
later it will be assumed expli itly (4.2) to prove (3.16).

Theorem 3.8 Let the assumptions (2.1), (3.2) (3.9), (3.13), (3.16) and (3.21)
hold. Assume that the initial ondition q0 is stable, i.e.

q0 ∈ S(0), (3.31)

and is approximated by [q0 ]h,ε ∈ Qh in the sense

[q0 ]h,ε → q0 and Eε (0, [q0 ]h,ε ) → E(0, q0 ). (3.32)

Then, there exists a subsequen e {(τn , hn , εn )}n∈N with (τn , hn , εn ) → (0, 0, 0) for
n→∞ satisfying the onvergen e riterion hn ≤ H(εn ) from ondition (3.16) and
a pro ess q : [0, T ] → Q being an energeti solution a ording to Denition 2.1 su h
that the following holds:
(i) for all t ∈ [0, T ] we have Eεn (t, qn (t)) → E(t, q(t)),
(ii) for all t ∈ [0, T ] we have VarDεn (qn ; 0, t) → VarD (q; 0, t),
(iii) for all t ∈ [0, T ] we have zn (t) → z(t) in Z ,
(iv) ∂
E (·, qn (·)) → ∂t
∂t εn
∂
E(·, q(·)) in L1 (0, T ),
(v) for all t ∈ [0, T ] there is a subsequen e {nl }l∈N su h that liml→∞ unl (t) =
u(t) in U , hen e liml→∞ qnl (t) = q(t) in Q,
where
R R R
we wrote shortly qn = (un , zn ) for qτn ,hn ,εn = (uτn ,hn ,εn , zτn ,hn ,εn ).

9
Proof. We follow the steps for the existen e proof formulated in [14, 39℄. However,
we are more general than [14, 39℄ as we do not require [q0 ]h,ε to be stable.
Let us abbreviate
R (t)),
Gτ,h,ε (t) := Eε,h(t, qτ,h,ε R ; 0, t).
Dτ,h,ε (t) := VarDε,h (qτ,h,ε (3.33)

Step 1: A priori estimates. By (3.9) and (3.10), we an estimate the right-hand side
of (3.29) as
Z tkτ k−1 tkτ
∂Eε,h (t, qτ,h,ε )
Z
k−1


dt ≤ c1 Eε,h(t, qτ,h,ε ) + c0 dt
tk−1
τ
∂t tk−1
τ
Z tkτ
k−1
c1 ec1 (t−tτ )
Eε,h (tk−1 k−1


≤ τ , qτ,h,ε ) + c0 dt
tk−1
τ
k k−1
= ec1 (tτ −tτ )
− 1 Eε,h (tk−1 k−1
(3.34)



τ , qτ,h,ε ) + c0 .
Forgetting, for a moment, Dε,h in (3.29) and linking it with (3.34) yields
k k−1
Eε,h(tkτ , qτ,h,ε
k
) + c0 ≤ ec1 (tτ −tτ ) (Eε,h (tk−1 τ
k−1
, qτ,h,ε ) + c0 ) from whi h, by indu tion for
k = 1, 2, ..., kτ we get
k
Eε,h (tkτ , qτ,h,ε
k
) ≤ ec1 tτ Eε,h (0, qτ,h,ε
0
(3.35)


) + c0 − c0 .
By (3.32), we on lude that Eε,h(tkτ , qτ,h,ε
k
) is upper bounded independently of k , h,
τ , and ε. By (3.9b) we an bound Gτ,h,ε (t) from below and, by (3.35) with (3.10)
after some still some al ulations from above:
−c0 ≤ Gτ,h,ε (t) ≤ a∗ ec1 t − c0 with a∗ := c0 + sup Eε,h(0, qτ,h,ε
0
), (3.36)
τ,h,ε

where the sup is onsidered for (τ, h, ε) small enough. Note that a∗ < +∞ due to
(3.32) with the assumption E(0, q0 ) < +∞.
Using (3.35) again for (3.34) but summed for k = 1, ..., kτ , we obtain
Z T R )
∂Eε,h (t, qτ,h,ε kτ Z tkτ
X k−1
∂Eε,h (t, qτ,h,ε )
dt = dt
0 ∂t tk−1
τ k=1
∂t
kτ

X k k−1
0
ec1 tτ − ec1 tτ


≤ Eε,h(0, qτ,h,ε ) + c0

k=1
0
) + c0 ec1 T − 1 . (3.37)


= Eε,h(0, qτ,h,ε
Coming ba k to (3.29) and ombining it with the lower bound (3.9b) for Eε,h(T, qτ,h,ε
kτ
)
and with (3.37), we now an estimate the total variation of Dτ,h,ε as
kτ
X
k−1 k


Var(Dτ,h,ε; 0, T ) = Dε,h qτ,h,ε , qτ,h,ε
k=1
0 0
) + c0 ec1 T −1



≤ Eε,h (0, qτ,h,ε ) + c0 + Eε,h (0, qτ,h,ε
0
) + c0 ec1 T ≤ a∗ ec1 T (3.38)


= Eε,h(0, qτ,h,ε

10
with a∗ from (3.36). We an now estimate also the total variation of Gτ,h,ε simply
by (3.9b) and (3.10) as
Z T R )
∂Eε,h (t, qτ,h,ε Xkτ
k−1
Var(Gτ,h,ε ; 0, T ) = dt + Eε,h (tkτ , qτ,h,ε
k
) − Eε,h (tkτ , qτ,h,ε )
0 ∂t k=1
Z T kτ
X
Eε,h (tkτ , qτ,h,ε
k
) − Eε,h (tk−1 k−1


≤ c1 Gτ,h,ε (t) + c0 dt + τ , qτ,h,ε )
0 k=1
kτ Z tkτ k−1
X ∂Eε,h (t, qτ,h,ε )
+ dt =: T1 + T2 + T3 . (3.39)
k=1 tk−1
τ
∂t

The term T1 is bounded sin e we have already proved |Gτ,h,ε(t)| a-priori bounded,
and also T3 ≤ T1 , see (3.37), so it remains to bound T2 . By (3.29) and (3.30), we
an estimate
 Z T 
∂Eε,h (t, qτ,h,ε)
T2 ≤ Var(Gτ,h,ε ; 0, T ) + max T1 , dt
0 ∂t
 
0 1 1
+ max 0, Eε,h(0, qτ,h,ε) − Eε,h(tτ , qτ,h,ε) (3.40)

and, again by (3.9b) and (3.10),

T Z T
∂Eε,h (t, qτ,h,ε )
Z


dt ≤ c1 Eε,h(t, qτ,h,ε ) + c0 dt
0 ∂t 0
k τ
X τ Z tk kτ Z tkτ
k
X
k
c1 ec1 (tτ −t) Eε,h (tkτ , qτ,h,ε
k



= c1 Eε,h (t, qτ,h,ε )+c0 dt ≤ )+c0 dt
k=1 tk−1
τ k=1 tk−1
τ

kτ kτ
c1 (tkτ −tk−1 k−1
X
)
X k
k−1
Eε,h(tkτ , qτ,h,ε
k
ec1 (tτ −tτ )
−1 ec1 tτ a∗ eτ



= e τ
−1 ) + c0 ≤
k=1 k=1
kτ
X k k−1
ec1 tτ − ec1 tτ (3.41)

τ
a∗ e = ec1 T − 1 a∗ eτ ,


=
k=1

where we also used, by (3.35)(3.36), the estimate

k k−1
Eε,h (tkτ , qτ,h,ε
k
) + c0 ≤ ec1 tτ a∗ ≤ ec1 tτ a∗ eτ ,

and eventually the last term in (3.40) an be estimated simply be ause, by (3.9b)
and (3.32), Eε,h(t1τ , qτ,h,ε
1
) ≥ −c0 and Eε,h (0, qτ,h,ε
0
) ≤ a∗ − c0 , hen e this last term is
bounded from above by a∗ from (3.36).

Step 2: Sele tion of subsequen es. Sin e the s alar fun tions Gτ,h,ε and Dτ,h,ε from
(3.33) are uniformly bounded in BV([0, T ]) by (3.38) and (3.39) together with the
obvious bounds on |Gτ,h,ε(0)| = |Eε (0, [q0 ]h,ε )| ≤ max(|c0 |, a∗ ) with a∗ from (3.36) and
|Dτ,h,ε(0)| = 0, we may apply Helly's sele tion prin iple both in the lassi al form

11
and, relying on the assumptions (3.2), (3.6), (3.7), also in the form of Lemma 3.1 to
nd a subsequen e {(τn , hn , εn )}n∈N su h that for all t ∈ [0, T ] we have the following
onvergen e:

Gτn ,hn ,εn (t) → G(t), Dτn ,hn ,εn (t) → D(t), and zτn ,hn ,εn (t) → z(t) in Z, (3.42)

for suitable limit fun tions D, G and z satisfying also (3.15b). This shows that the
onvergen e at the point (iii) holds. We further set
∂
Pn (t) := Eε (t, qn (t)) (3.43)
∂t n
to denote the power of the external for es. Choosing another subsequen e (not
relabeled), if ne essary, we also obtain

Pn w*
→ p in L∞ ([0, T ]), (3.44)

sin e losed balls in L∞ ([0, T ]) are sequentially weakly* ompa t. For xed t, let

P(t) := lim sup Pn (t). (3.45)

n→∞

Using Fatou's lemma we on lude P ∈ L∞ (0, T ) and p(t) ≤ P(t) for a.a. t ∈ [0, T ].
Further, let us set
n ∂ o
A(t) := ũ ∈ U; E(t, ũ, z(t)) = P(t) . (3.46)
∂t
For any t xed, A(t) is nonempty: Indeed, we an hoose a subsequen e (ntj )j∈N
(depending on t!) su h that
∂
P(t) = lim Pntj (t) = lim Eε t (t, qntj (t)), (3.47)
j→∞ j→∞ ∂t nj

f. (3.45) and (3.43). Taking into a ount the energy bound Gτ,h,ε (t) obtained in
Step 1 and the ompa tness assumption (3.5), we an even assume that also qntj (t)
onverges to some q(t). By (3.42), q(t) = (u(t), z(t)) with z(t) just from (3.42). Let
τn
tj := max{θ ∈ [0, t]; θ = tk j , k = 0, ..., kτnj }. Then qntj (t) ∈ Sεnt ,hnt (tj ). Obviously,
j j
also tj → t. Hen e we an use (3.13) to obtain

∂ ∂
lim Eεnt ,hnt (tj , qntj (t)) = E(t, q(t)). (3.48)
j→∞ ∂t j j ∂t
Comparing it with (3.47) we get
∂
E(t, q(t)) = P(t). (3.49)
∂t
Thus u(t) forming the pair q(t) = (u(t), z(t)) lies in A(t) from (3.46). Ranging t
over [0, T ] thus yields a mapping u : [0, T ] → U with u(t) ∈ A(t) for all t ∈ [0, T ].

12
Step 3: Stability of the limit pro ess. The stability of the limit pro ess q is now
ensured by (3.16) as a dire t onsequen e of Lemma 3.3. For xed t ∈ (0, T ] onsider
qntj (t) and tj onverging for j → ∞ to q(t) and t in the position of qn and tn in the
ondition (3.16), respe tively, and then Lemma 3.3 just yields q(t) ∈ S(t). For t = 0,
stability of q(0) = q0 holds by assumption.

Step 4: Upper energy estimate. By (3.26) with r = 0 we have Gτn ,hn ,εn (t) +
Rt
Dτn ,hn ,εn (t) − Gτn ,hn,εn (0) ≤ 0 Pn (s) ds for any t = tkτn , k = 0, ..., kτn . For a gen-
eral t ∈ [0, T ], this inequality is fullled with an a ura y O(τn ); this is be ause
τn )| ≤ τn kPn kL∞ (0,T ) for t ∈ [tτn , tτn ) and be ause also
|Gτn ,hn ,εn (t) − Gτn ,hn ,εn (tk−1 k−1 k
Rt R tk−1
| 0 Pn (s) ds − 0 τn Pn (s) ds| ≤ τn kPn kL∞ (0,T ) while there is no additional error in
the pie e-wise onstant Dτn ,hn ,εn . By the onvergen e properties (3.42), (3.44) and
(3.45) with Fatou's lemma we get
Z t Z t
G(t) + D(t) − G(0) ≤ p(s)ds ≤ P(s)ds. (3.50)
0 0

Using further (3.8), (3.42), and the notation from Step 2, we have

E(t, q(t)) ≤ lim inf Eεnt ,hnt (tj , qntj (t)) = lim Gntj (t) = G(t). (3.51)
j→∞ j j j→∞

By (3.15b) with s = 0 and D(s) = D(0) = 0, we have VarD (q; 0, t) ≤ D(t). More-
over, by (3.32) we have G(0) = E(0, q(0)). Inserting this into (3.50) and using still
(3.49), we obtain

E(t, q(t)) + VarD (q; 0, t) − E(0, q(0)) ≤ G(t) + D(t) − G(0)

Z t Z t
∂
≤ P(s) ds = E(s, q(s)) ds, (3.52)
0 0 ∂s

whi h is the desired upper energy estimate.

Step 5: Lower energy estimate.

Rt ∂ The opposite estimate E(t, q(t)) + VarD (q; 0, t) −
E(0, q(0)) ≥ 0 ∂s E(s, q(s))ds is a onsequen e of the stability whi h is already
established in Step 3. We refer to [36, Prop. 5.7℄ or also [39, Prop. 2.4℄ for this
te hni al proof where (3.49) with P ∈ L∞ (0, T ) and (3.11) have been used. Thus,
we have proved that q : [0, T ] → Q is a solution.

Step 6: Improved Having energy equality, we on lude that in (3.50)

onvergen e.
all the inequalities must be equalities. In parti ular, this implies

p(t) = P(t), G(t) = E(t, q(t)) and Var(q; 0, t) = D(t). (3.53)

Together with the onvergen e properties established in Step 2, we obtain the as-
sertions (i)(iii). Finally, employing [14, Prop. A.2℄ together with p = P yields (iv).
2

13
Remark 3.9 (Two-sided energy estimate (3.26)(3.27).) In fa t, (3.26)(3.27) was
used only to prove the a-priori BV-bound for Gτ,h,ε in Step 1. This bound is not
really needed, sin e we may postpone the denition of G : [0, T ] → R from Step 2
to Step 4 and set G(t) = lim supn→∞ Gτn ,hn ,εn (t). Then (3.50) and (3.51) remain
true but the last equality in (3.51) whi h has to be repla ed by  ≤. Finally, Step 5
implies G(t) = E(t, q(t)) as before. However, the two-sided energy estimate (3.26)
(3.27) has its own relevan e as it an be used to he k implementation of numeri al
al ulations. Namely, evaluating the terms in (3.26)(3.27) at ea h time step and
he king a-posteriori the estimate (3.26)(3.27) may dete t, e.g., a failure of the
minimization pro edure, whi h we have to apply to solve numeri ally the global
optimization problem (3.22) at every urrent time step; see [31℄ for numeri al results
in a on rete example. Violation of (3.26) or (3.27) mean that qτ,h,ε
k
or qτ,h,ε
k−1
annot
be stable, respe tively.

Remark 3.10 (Numeri al integration.) Another approximation of Eε and Dε in-

volving, e.g., numeri al integration an quite easily be in orporated, too. For this,
Eε and Dε in the onditions in Se t. 3.1 as well as (3.16) should additionally depend
on h by still another way than only by adding δQh . As su h a generalization would
ompli ate, in parti ular, Se tion 4 and as it will not be used in Se tion 5, we have
omitted it ompletely.

4 Linear stru ture

We onsider now the ase that U and Z are subsets of some reexive separable
Bana h spa es U and Z , respe tively. This enables more detailed onsiderations.

4.1 Setting the data and their approximation

The weak topology, if restri ted on bounded onvex sets, will play the role of the
sequentially ompa t topology used in Se t. 3 for (3.3)(3.8), (3.13), and (3.16); yet
w
f. Remark 3.2 for (3.13). Here we will denote it by w-lim or  →  to distinguish
s
it from the norm topology whi h we will denote by s-lim or  → . In ase of
non-reexive spa es having preduals, we ould work with weak* topologies instead
of the weak ones. For an abstra t parameter h > 0, we onsider nite-dimensional
subspa es Uh ⊂ U and Zh ⊂ Z . The on rete onstru tions of Qh := Uh × Zh used
in numeri al analysis are reated by (here an abstra t) (quasi-)interpolation linear
bounded operators ΠU,h : U → U and ΠZ,h : Z → Z . We put Πh = ΠU,h × ΠZ,h :
Q → Q, and
Uh := ΠU,h U, Zh := ΠZ,h Z, Qh := Uh ×Zh = Πh Q. (4.1)
To guarantee the entral ondition (3.16), we assume the natural basi approxima-
tion property that Πh onverges pointwise to the identity, i.e.
∀q ∈ Q : s-lim Πh q = q. (4.2)
h→0

14
The quasi-interpolation operators need not be onformal with onstraints involved
impli itly in U and Z so that Qh need not be a subset of Q. As an analyti al tool
the Γ- onvergen e approa h allow also for su h situations ( f. [39℄) but, in order to
use the theory from Se tions 23 in a quantitative numeri al way, we will always
restri t ourselves on  onformal situations when

ΠU,h U ⊂ U and ΠU,h Z ⊂ Z; (4.3)

i.e. Qh = Πh Q ⊂ Q. Possible non onformities an be handled via the penalization

parameter ε.
For X another Bana h spa e, it is often useful to onsider a mapping Ξ : U ×Z → X
to des ribe possible equality onstraints of the form Ξ(u, z) = 0 that may impli itly
be involved in the denition of E . Moreover, we assume the for ing by f : [0, T ] →
U ∗ × Z ∗ to be given expli itly in E , whi h overs many appli ations (ex ept, e.g.,
hard-devi e loading of me hani al systems through Diri hlet boundary onditions).
Then, for E : U × Z → R we onsider

E(u, z) − hf (t), (u, z)i if u ∈ U, z ∈ Z, Ξ(u, z) = 0,


E(t, u, z) := (4.4)
+∞ otherwise.

The approximate energy deals with possible in ompatibility of the nite-dimensional

dis retization with the equality onstraints by a penalization of them ( f. [49℄):
( 1 α
E(u, z) − hf (t), (u, z)i + Ξ(u, z) X if u ∈ U, z ∈ Z,
Eε (t, u, z) := ε (4.5)
+∞ otherwise.

To satisfy (3.4), we assume a super-linear growth of E to dominate the linear be-

havior of hf (t), ·i:

E(q)
lim = +∞. (4.6)
q∈Q kqk
kqk→∞

Obviously, (4.4) and (4.5) yield simply ∂

∂t
E(t, q) = ∂
E (t, q)
∂t ε
∂
= h ∂t f (t), qi and (3.9a)
requires

f ∈ C 1 ([0, T ]; Q∗ ). (4.7)

The oer ivity (4.6) with (4.7) ensure also (3.9b), (3.11) and (3.13).
A quite anoni al way to indu e the dissipation distan es in simpler ases is through
a degree-1 homogeneous dissipation potentials. For this, we onsider K ⊂ Z a
losed onvex one with the vertex at 0, R : Z → R a ontinuous onvex degree-1
homogeneous fun tional, i.e. R(az) = aR(z) for any z ∈ Z and a ≥ 0. Then we
onsider the spe ial ase of D dened by

R(z2 − z1 ) if z2 − z1 ∈ K,

D(z1 , z2 ) := (4.8)
+∞ otherwise.

15
Note that D(z1 , z1 ) = 0 and the triangle inequality (2.1) holds. As R is onvex
and ontinuous and K onvex losed, D : Z × Z → R ∪ {+∞} is weakly lower
semi ontinuous.
If K 6= Z , then it might be numeri ally suitable to avoid the unilateral onstraints
involved by exa t penalization by hoosing the approximate potential Dε in the form
kz − ẑk
Dε (z1 , z2 ) := Rε (z2 − z1 ) where Rε (z) := R(z) + inf . (4.9)
ẑ∈K ε
As K is a one, Rε is again a homogeneous degree-1 fun tional for any ε > 0 and
(3.2) thus holds. As R is onvex and ontinuous and K is onvex, Rε is onvex
and ontinuous, and the weak lower-semi ontinuity (3.3) of Rε holds, too. Note
that always Rε ≤ R + δK . Unfortunately, smoothening of R + δK e.g. by Yosida's
approximation, whi h would be sometimes numeri ally desirable, does not fulll
(3.2) and expe tedly nontrivial modi ations of the theory in Se t. 3 would then be
needed.
The stability (3.31) of the initial ondition q0 is, in general, di ult to verify and
expli it onstru tions an be done in very spe ial ases only. Anyhow, there is one
universal way how to design a gentle start, namely taking q0 = (u0 , z0 ) minimizing
E(0, ·), i.e. here a solution to the problem

minimize E(u, z) − hf (0), (u, z)i,


(4.10)
subje t to Ξ(u, z) = 0, u ∈ U, z ∈ Z.

Su h a gentle start is, in fa t, pra ti ally the only option applied in engineering
simulations.
The other assumptions from Se t. 3 deserve a more detailed proof.

Proposition 4.1 (Veri ation of (3.5)(3.8).) Let E be weakly lower semi on-
tinuous, Ξ: Q→X be weakly ontinuous, and let K be onvex and losed, R be
onvex and also positive on K \ {0}, i.e.

∀z ∈ K : z 6= 0 ⇒ R(z) > 0. (4.11)

Then (3.5)(3.8) with  → referring to the weak topology hold.

Proof. In view of (4.4), the ondition Eε (θ, qh,ε ) ≤ a < +∞ in (3.5) implies E(qh,ε ) ≤
C +hf (θ), qh,εi, and by (4.6) a sequen e of {qh,ε }h,ε>0 must be bounded hen e it has a
subsequen e whi h onverges weakly (re all that we assume reexivity of Q), whi h
proves (3.5).
As to (3.6), for z2 − z1 ∈ K we have

kz̃h,ε − zh,ε − ẑk

lim inf Dε (zh,ε , z̃h,ε ) = lim inf R(z̃h,ε − zh,ε ) + inf
(h,ε)→(0,0) (h,ε)→(0,0) ẑ∈K ε
≥ lim inf R(z̃h,ε − zh,ε ) ≥ R(z̃ − z) = D(z, z̃) (4.12)
(h,ε)→(0,0)

16
be ause R is weakly lower semi ontinuous. If z2 −z1 6∈ K , then inf ẑ∈K kz̃ −z − ẑk > 0
be ause K is losed. Using also (4.11), we then have
kzh,ε − z̃h,ε − ẑk
lim inf Dε (zh,ε , z̃h,ε ) ≥ lim inf = +∞ = D(z, z̃). (4.13)
(h,ε)→(0,0) (h,ε)→(0,0) ẑ∈K ε

To prove (3.7), take z ∈ Z and a sequentially weakly ompa t K in Z and a sequen e

(zn )n∈N in K with limn→∞ min(D(zn , z), D(z, zn )) = 0. For a subsequen e we have
w
znj → z̃ , and the mentioned weak lower semi ontinuity of D implies either D(z, z̃) =
0 or D(z̃, z) = 0. Thus we an on lude z̃ = z and the whole sequen e must weakly
onverge, whi h proves (3.7).
As to (3.8), let us distinguish whether Ξ(q) = 0 or Ξ(q) 6= 0. The former ase
ensures the last equality in the following estimate:
1
lim inf Eε (θ, qh,ε ) = lim inf E(qh,ε ) − hf (θ), qh,ε i + kΞ(qh,ε )kαX
(h,ε,θ)→(0,0,t) (h,ε)→(0,0) ε
≥ lim inf E(qh,ε ) − hf (θ), qh,εi ≥ E(q) − hf (t), qi = E(t, q), (4.14)
(h,ε,θ)→(0,0,t)

where the last inequality is by the weak lower semi ontinuity of E . This proves
that (3.8) holds with respe t to the weak topology if Ξ(q) = 0. In the ase
w
Ξ(q) 6= 0, qh,ε → q and the weak ontinuity of Ξ ensures lim inf kΞ(qh,ε )kX ≥
kw- lim Ξ(qh,ε )kX = kΞ(q)kX > 0. Then, be ause of the oer ivity (4.6) of E , we
have
  1 α
lim inf Eε (θ, qh,ε ) ≥ inf E−f (θ) (q̃) + lim Ξ(qh,ε ) X = +∞ = E(t, q).
θ→t q̃∈Q (h,ε)→(0,0) ε
(h,ε)→(0,0) θ∈[0,T ]
2

In view of the above onsiderations, we have guaranteed the assumptions needed

in Theorem 3.8 ex ept (3.16) and (3.32). This onditions are still to be veried in
parti ular ases, some of them s rutinized in Se tions 4.24.4.

Remark 4.2 (BV-estimates.) Assuming oer ivity of R+δK on some Bana h spa e
Z1 ⊃ Z , i.e.

lim R(z) = +∞, (4.15)

z∈K, kzkZ1 →∞

together with the degree-1 homogeneity will make (4.11) more spe i , namely [R +
δK ](z) ≥ ckzkZ1 with some c > 0, hen e by (4.8) also D(q1 , q2 ) ≥ ckz1 − z2 kZ1 , and
by the denition of Var in (2.6) then also

VarD (q; 0, T ) ≥ cVark·kZ1 (z; 0, T ). (4.16)

In view of the denition (2.6) applied now with the norm k · kZ1 , the last expression
is just the standard total variation and the estimate (3.38) yields boundedness of
zτ,h,ε and thus also the limit z in the bounded-variation spa e BV(0, T ; Z1).

17
4.2 The ase K=Z
Let us onsider an additional norm | · |, whi h may indu e a weaker topology than
the anoni al norm making Q a Bana h spa e.

Proposition 4.3 K=Z .) Let (4.6) and

(Veri ation of (3.16) and (3.32) for
(4.7) hold, and let E : Q → R in (4.4) be weakly lower semi ontinuous
α ≥ 1, let
and norm ontinuous, both Ξ : Q → X and R : Z → R be weakly ontinuous, and
K = Z (hen e Rε ≡ R), and Ξ be also Lips hitz ontinuous with respe t to | · |, i.e.
∃ℓΞ ∈ R ∀q1 , q2 ∈ Q : Ξ(q1 ) − Ξ(q2 ) X
≤ ℓΞ q1 − q2 (4.17)
and let the operator Πh satises the onvergen e-rate estimate

∃γ > 0, C ∈ R ∀q ∈ Q : q − Πh q ≤ Chγ q . (4.18)

Then (3.16) and (3.32) with q0 ∈ S(0) are satised, the last two onditions relying
on the onvergen e riterion
1
s
(4.19)


H(ε) = o ε αγ and with q̃h,ε → q̃.

Proof. Let us prove (3.16). For any q̃ ∈ Q, with Ξ(q̃) = 0, by (4.17) and (4.18), we
have
Ξ(Πh q̃) X
= Ξ(Πh q̃) − Ξ(q̃) X
≤ ℓΞ q̃ − Πh q̃ ≤ CℓΞ hγ q̃ . (4.20)
For (h, ε) → (0, 0) with h ≤ H(ε) with H from (4.19) we therefore have
1 α hαγ α
Ξ(Πh q̃) X ≤ C α ℓαΞ q̃ → 0. (4.21)
ε ε
We put q̃h,ε := Πh q̃ for (3.16); note that, in fa t, we do not need any expli it
dependen e on ε ex ept that we assume h ≤ H(ε). As E is strongly ontinuous
s w
and, by (4.2), q̃h,ε → q̃, and as R is weakly ontinuous and qh,ε → q is assumed in
(3.16), it holds
lim Eε (θ, q̃h,ε ) + D(qh,ε, q̃h,ε ) = lim E(q̃h,ε ) − hf (θ), q̃h,ε i + R(q̃h,ε −qh,ε )
h≤H(ε) h≤H(ε)
(ε,h)→(0,0) (ε,h)→(0,0)
1 α
+ Ξ(Πh q̃) X = E(q̃) − hf (t), q̃i + R(q̃−q) = E(t, q̃) + D(q, q̃)
ε
whenever Ξ(q̃) = 0. Combining this with (3.8), we obtain (3.16) for Ξ(q̃) = 0. If
Ξ(q̃) 6= 0, then due to the denition (4.4) the right-hand side in (3.16) is +∞ and
(3.16) is fullled trivially.
The stability of q0 onsidered in Theorem 3.8 implies E(0, q0) < +∞, and then the
assumption (3.32) is fullled if one hooses
(4.22)
 
q0 h,ε := Πh q0
s
in (3.32). Indeed, [q0 ]h,ε → q0 for h → 0 just by (4.2) and then also Eε (0, [q0 ]h,ε ) =
E(Πh q0 ) + ε kΞ(Πh q0 )kX − hf (0), Πh q0 i → E(q0 ) − hf (0), q0 i = E(0, q0 ) be ause
1 α

E is assumed norm ontinuous and be ause, sin e the nite energy of q0 implies
Ξ(q0 ) = 0, we an employ the estimate (4.21) for h ≤ H(ε). 2

18
4.3 The ase K $Z
Certain appli ations to unidire tional pro esses (like damage, delamination, debond-
ing, or hardening in plasti ity or in ferromagnets) require modelling with K $ Z .
This needs further ner investigations for whi h we onsider some topology σ on
U × Z whi h is ner than the weak one and oarser than the norm one; see the
parti ular examples in Se t. 5.

Proposition 4.4 (Un onditional onvergen e for K $ Z .) Let E : Q → R

be weakly lower semi ontinuous and σ- ontinuous. Assume both R : Z → R and
Ξ:Q→X be weakly ontinuous, and let (4.7), and that the following attainability
ondition, expressing ertain onsisten y of the dis retization with the onstraints
given by Ξ and K, hold:
w
∀q, q̃ ∈ Q, Ξ(q) = 0, q̃−q ∈ K, Ξ(q̃) = 0, ∀qh ∈ Qh , qh → q
σ
∃q̃h ∈ Qh : q̃h → q̃, Ξ(q̃h ) X ≤ Ξ(qh ) X , q̃h −qh ∈ K. (4.23)
Then (3.16) with Dε,h from (4.9) is satised, now with H ≡ 1, i.e. un onditionally.
Moreover, the quali ation (3.32) of the stable initial ondition q0 holds if
σ
∃q0h ∈ Qh : Ξ(q0h ) = 0 & q0h → q0 . (4.24)

Proof.The a-priori bound Eε,h(θ, qh,ε ) ≤ C assumed in (3.16) means

1 α
(4.25)
 
Ξ(qh,ε ) X ≤ C − E(qh,ε ) + f (θ), qh,ε ≤ C + sup f (θ)−E (q) < +∞
ε q∈Q
θ∈[0,T ]

due to (4.6) so that kΞ(qh,ε )kX = O(ε1/α ). In the limit therefore Ξ(q) = 0 be ause
Ξ is assumed weakly ontinuous. Thus we take qh,ε from (3.16) for qh in (4.23). As
(3.16) is trivially satised if Ξ(q̃) 6= 0 be ause the right-hand side in (3.16) is +∞,
we an onsider only Ξ(q̃) = 0. Then we an take q̃h from (4.23) for q̃h,ε in (3.16).
Note that q̃h,ε − qh,ε ∈ K in (4.23) ensures Dε,h (qh,ε , q̃h,ε ) = R(qh,ε − q̃h,ε ) due to
the denition (4.9) and by the assumed weak ontinuity of R and losedness and
onvexity of K , we have
lim Dε,h (qh,ε , q̃h,ε ) = lim R(q̃h,ε − qh,ε ) = R(q̃ − q) = D(q, q̃). (4.26)
(h,ε)→(0,0) (h,ε)→(0,0)

Then, using the σ - ontinuity and weak lower semi ontinuity of E the ontinuity of
f (see (4.7)), and kΞ(q̃h,ε )kX ≤ kΞ(qh,ε )kX (see (4.23)), we obtain
  
lim sup Eε,h (θ, q̃h,ε ) + Dε,h (qh,ε , q̃h,ε ) − Eε,h(θ, qh,ε ) = lim sup E(q̃h,ε )
(h,ε,θ)→(0,0,t) (h,ε,θ)→(0,0,t)
1 α 1 α

−hf (θ), q̃h,ε − qh,ε i + kΞ(q̃h,ε )kX + D(qh,ε , q̃h,ε ) − E(qh,ε ) − kΞ(q̃h,ε )kX
 ε ε 
≤ lim sup E(q̃h,ε ) − hf (θ), q̃h,ε − qh,ε i + D(qh,ε , q̃h,ε ) − E(qh,ε )
(h,ε,θ)→(0,0,t)
 
= lim E(q̃h,ε ) − hf (θ), q̃h,ε − qh,ε i + D(qh,ε , q̃h,ε ) − lim inf E(qh,ε )
(h,ε,θ)→(0,0,t) (h,ε)→(0,0)

≤ E(q̃) − hf (t), q̃−qi + D(q, q̃) − E(q) = E(t, q̃) + D(q, q̃) − E(t, q). (4.27)

19
Eventually, we are to prove (3.32) provided (4.24) and provided q0 ∈ S(0); the last
in lusion implies E(0, q0 ) < +∞ whi h here further implies Ξ(q0 ) = 0). Then, with
[q0 ]h,ε := q0h in (4.24), it holds

Eε,h (0, [q0 ]h,ε ) = E(q0h ) − hf (0), q0h i → E(q0 ) − hf (0), q0 i = E(0, q0 ) (4.28)

as required in (3.32) be ause E is assumed σ - ontinuous. Note that the last equality
in (4.28) relies on Ξ(q0 ) = 0 for whi h σ - ontinuity of Ξ is needed; in fa t, we
assumed even weak ontinuity of Ξ. 2

4.4 The ase K $Z and semiquadrati  E.

Some appli ations exhibits the main part of the stored energy E quadrati in terms
of the dissipating variable z in the sense
1
E(u, z) := hBz, zi + E0 (u, z) , B : Z → Z ∗ linear and bounded,
2
E0 : U×Z → R (s×w)- ontinuous. (4.29)

In smooth ases, this orresponds to problems governed by semilinear mappings

E ′ (q) = 00 B
0
+ E ′ (q). Su h problems are well tted for un onditional onvergen e
0
under some parti ular ir umstan es.
As to (3.32), we an guarantee it again through (4.24) now with σ the strong topology
to have the quadrati term in (4.29) ontinuous. The veri ation of (3.16) is now
more sophisti ated:

Proposition 4.5 (Semiquadrati  ase: un onditional onvergen e.) Let

(4.7) and (4.29) hold, R be Ξ be independent of u, ane and
ontinuous, let further
ontinuous, i.e. in the form Ξ(u, z) = Ξ0 (z) + ξ with ξ ∈ X , and Ξ0 ∈ L(Z, X)
ompatible with the dis retization operator ΠZ,h in the sense that ΠZ,h (Ker Ξ0 ) ⊂
Ker Ξ0 . Let also Z + K ⊂ Z , and the one K be ompatible with ΠZ,h in the sense
that ΠZ,h K ⊂ K . Then (3.16) with H ≡ 1 holds.

Proof. We will prove (3.16) by using Proposition 4.4 and for this we will verify
(4.23) with σ being the strong×weak topology on U × Z . The re overy element q̃h
in (4.23) an be hosen simply as

ũh := ΠU,h ũ, (4.30a)

z̃h := zh + ΠZ,h (z̃ − z). (4.30b)

It holds q̃h ∈ Qh ; indeed, ũh ∈ Uh just by the denitions (4.1) and (4.30a) while
z̃h ∈ Zh be ause z̃ −z ∈ K , assumed in (4.23), implies z̃h −zh = ΠZ,h (z̃ −z) ∈ ΠZ,h K
and further Z + K ⊂ Z implies Zh = ΠZ,h Z ⊃ ΠZ,h (Z + K) = Zh + ΠZ,h K and
eventually zh ∈ Zh is assumed in (4.23), hen e z̃h ∈ Zh indeed follows.

20
Also, the inequality kΞ(qh )kX ≤ kΞ(q̃h )kX in (4.23) follows from

Ξ(q̃h ) = Ξ0 z̃h + ξ = Ξ0 (zh + ΠZ,h (z̃ − z)) + ξ = Ξ(qh ) (4.31)

be ause Ξ0 (ΠZ,h (z − z̃)) = 0 holds. Indeed, Ξ(q̃) = 0 is also expli itly assumed in
w
(4.23) while Ξ(q) = 0 follows from qh → q assumed in (4.23) by the ontinuity
of Ξ, and therefore Ξ0 (z − z̃) = Ξ(q) − Ξ(q̃) = 0, hen e z − z̃ ∈ Ker Π0 , and by
the assumed ompatibility ΠZ,h (Ker Ξ0 ) ⊂ Ker Ξ0 also ΠZ,h (z − z̃) ∈ Ker Ξ0 , hen e
eventually Ξ0 (ΠZ,h (z − z̃)) = 0. Then also, by using also (4.2), it holds
 
(s×w)-lim q̃h = s-lim ũh , w-lim z̃h
h→0 h→0 h→0
 
= s-lim ΠU,h ũ , w-lim zh + s-lim ΠZ,h (z̃−z) = ũ, z + (z̃−z) = q̃.


h→0 h→0 h→0

Although for σ =s×w the energy E itself need not be σ - ontinuous like in Proposi-
tion 4.4, in the ase (4.29) it is however possible to pass to the limit in the dieren e
E(θ, q̃h ) − E(θ, qh ) by using (4.31) and the binomial formula:

Eε (θ, q̃h ) − Eε (θ, qh ) = E(θ, q̃h ) + 1ε kΞ(q̃h )kαX − E(θ, qh ) − 1ε kΞ(qh )kαX
= E(θ, q̃h ) − E(θ, qh )
1
= 2
hB z̃h , z̃h i − 21 hBzh , zh i + E0 (q̃h ) − E0 (qh ) − hf (θ), q̃h −qh i
1
= 2
hB(z̃h − zh ), z̃h + zh i + E0 (q̃h ) − E0 (qh ) − hf (θ), q̃h −qh i
1
→ 2
hB(z̃ − z), z̃ + zi + E0 (q̃) − E0 (q) − hf (t), q̃ − qi
1
= 2
hB z̃, z̃i − 21 hBz, zi + E0 (q̃) − E0 (q) − hf (t), q̃ − qi
= E(t, q̃) − E(t, q). (4.32)
s
For the limit passage it was important that z̃h −zh = ΠZ,h (z̃−z) → z̃−z be ause of
(4.2) so that

hB(z̃h −zh ), z̃h +zh i → hB(z̃ − z), z̃ + zi (4.33)

w
be ause z̃h +zh → z+z̃ . We have z̃h − zh = ΠZ,h (z̃ − z) ∈ ΠZ,h K ⊂ K . Then, in
view of the denition in (4.8) and the strong ontinuity of R, we have

lim Dε (qh , q̃h ) = lim

Rε (z̃h −zh ) = lim R(z̃h −zh )
(ε,h)→(0,0) (ε,h)→(0,0) h→0

(4.34)



= lim R ΠZ,h (z̃−z) = R lim ΠZ,h (z̃−z) = R(z̃−z) = D(q, q̃).
h→0 h→0

By (4.32) and (4.34), we an pass dire tly to the limit in (4.27). Thus (3.16) with
H ≡ 1 is proved in this ase, too. 2

Alternatively to the setting (4.29), we an onsider a variant with a fully quadrati

main part of E :

21
Proposition 4.6 (Semiquadrati ase II: un onditional onvergen e.) Let
1
E(q) := hBq, qi + E0 (q) , B : Q → Q∗ linear and bounded,
2
E0 : Q → R w- ontinuous. (4.35)
hold, R be ontinuous and U = U and and Ξ be ane and ontinuous, i.e. in the
form Ξ(q) = Ξ0 q + ξ with ξ ∈ X and Ξ0 ∈ L(Q, X) su h that Πh (Ker Ξ0 ) ⊂ Ker Ξ0 .
Let again Z + K ⊂ Z , ΠZ,h K ⊂ K , and f satisfy (4.7). Then (3.16) with H ≡ 1
holds.

Proof. Instead of (4.30), we take

q̃h := qh + Πh (q̃ − q). (4.36)
Then it su es to modify the proof of Proposition 4.5 quite straightforwardly, e.g. to
onsider q 's instead of z 's in (4.31) and (4.32). 2

Remark 4.7 (No penalization.) In ase of the un onditional onvergen e, one an

onsider a numeri al s heme with ε = 0, i.e. with the original E and D instead of
Eε,h and Dε,h . The orresponding in remental problem might then involve unilateral
onstraint; f. also Remark 5.3.

5 Parti ular examples in ontinuum me hani s

The doubly-nonlinear in lusion (2.2) is a framework for des ription of so- alled gen-
eralized standard materials with internal parameters as introdu ed by Halphen and
Nguen [21℄ in those ases where onvexity of stored and dissipated energies an be
expe ted and inertial ee ts an be negle ted. Here we have in mind various inelas-
ti rate-independent pro esses in su h materials having possibly a non onvex stored
energy. The following examples illustrate how the general theory applies in parti -
ular situations, f. Table 1 for a survey. As a by-produ t of the presented numeri al
theory, we obtain analyti al existen e/ onvergen e results whi h have not yet been
derived in literature. For the sake of explanatory lu idity, we onne ourselves to
rather onventional models from ontinuum me hani s although some less onven-
tional models (e.g. those involving a mi rostru ture des ribed by so- alled Young
measures, see [32, 52, 53, 54℄) allow for su h numeri al analysis, too. In Se t. 5.7
we present a ombination of me hani al and ferromagneti al ee ts, i.e. magne-
tostri tion with hystereti al ee ts, but the ombination with ferroele tri al ee ts,
i.e. piezoele tri ity with hysteresis (see [43℄), or even purely non-me hani al rate-
independent models developed in ferromagneti s (e.g. [52, 53, 58, 59℄) and ferro-
ele tri s (e.g. [25, 48, 56℄) ould be treated similarly. We negle t any temperature
dependen e or, in other words, if there is a possible dependen e of data on tem-
perature, we onsider su iently slow pro esses so that the released heat due to
dissipative pro esses an e iently be transferred away to allow for onsidering
isothermal pro esses.

22
 pro ess unidire tional onstraints quadrati proposition
/ se tion (i.e. K $ Z ) (i.e. X6={0}) energy E used
plasti ity with hardening
at small strains / 5.2
+ − + 4.5 or 4.6

phase transformation:
mixture approa h / 5.3
− − − 4.4 (σ=s)

phase transformation:
non-mixture approa h/5.4
− + − 4.4 (σ=s)

damage / 5.5 + − ± 4.5

debonding / 5.6 + − − 4.4 (σ=s×w∗ )

magnetostri tion / 5.7 − + − 4.3

Table 1. Organization and features of the examples presented in Se tion 5.

5.1 Sket h of ontinuum me hani s of deformable bodies

We assume a spe imen o upying in its referen e onguration a bounded domain

Ω ⊂ R3 . As usual, y : Ω → R3 denotes the deformation and u : Ω → R3 the
displa ement, related to ea h other by y(x) = x+u(x), x ∈ Ω. Hen e the deformation
gradient equals F = ∇y = I + ∇u with I ∈ R being the identity matrix and ∇
3×3

is the gradient operator. For simpli ity, we will treat only the soft-devi e loading
realized through tra tion (Neumann or Robin-type) boundary onditions. The state
of the material and possibly also of boundary onditions is assumed to depend on (a
set of) ertain parameters z that may evolve in time in a rate-independent manner.
Then naturally U and Z used before will be the spa es of u's and of z 's, respe tively.
The spe i energy stored in the inter-atomi links in the homogeneous (possibly
anisotropi ) ontinuum ϕ̂ = ϕ̂(F, z) is phenomenologi ally des ribed as a fun tion
of the deformation gradient F and the mentioned variable z ∈ Rm . Mostly the
ve tor z ∈ Z0 ⊂ Rm in not dire tly a essible for a ma ros opi al loading (for an
ex eption see Se t. 5.7) and will thus play the role of internal parameters. The frame-
indieren e, i.e. ϕ̂(F, z) = ϕ̂(RF, z) for any R ∈ SO(3) = the group of orientation-
preserving rotations, requires that ϕ̂(·, z) in fa t depends only on the (right) Cau hy-
Green stret h tensor

F ⊤ F = (I + ∇u)⊤ (I + ∇u) = I + (∇u)⊤ + ∇u + (∇u)⊤ ∇u. (5.1)

An importantR property of ϕ̂(·, z) is quasi onvexity, whi h means ϕ(A, z) ≤

inf u∈W 1,p (Ω;R3 ) Ω ϕ(A+∇u, z) dx for any A ∈ R3×3 . The following assertion modies
0
the elebrated result by A erbi and Fus o [1℄:

Lemma 5.1 Let ϕ : R3×3 × Rm → R be ontinuous, ϕ(·, z) quasi onvex, p, p1 ∈

(1, +∞) and, for some c2 ≥ c1 > 0,

∀A∈R3×3 ∀z∈Z0 : c1 |A|p +|z|p1 −1 ≤ ϕ(A, z) ≤ c2 1+|A|p +|z|p1 . (5.2)

23
 R
Then the fun tional (u, z) 7→ Ω ϕ(∇u, z) dx is (w×s)-lower semi ontinuous on
W 1,p (Ω; R3 ) × {z∈Lp1 (Ω; Rm ); z(·)∈Z0 a.e. on Ω}.

Sket h of the proof.By oer ivity, we do not need to distinguish between sequential
and topologi al lower semi ontinuity.
Let us take a sequen e {(un , zn )}n∈N (w×s)- onverging to (u, z). Then (∇un , zn )
(w×s)- onverges to (∇u, z) in Lp (Ω; R3×3 ) × Lp1 (Ω; Rm ). Also, sele ting a suitable
subsequen e, it generates (a set) of Lp ×Lp1 -Young measures of the form ν ⊗µz where
µz = {δz(x) }x∈Ω with δz(x) denoting here the Dira distribution on Rm supported at
z(x); f. [44, Corollary 3.4℄. This means, in terms of a mentioned subsequen e, that
Z Z Z
 
lim v(∇un , zn )dx = v(A, r) νx ⊗ δz(x) (dA×dr)dx
n→∞ Ω 3×3 m
ZΩ ZR ×R
= v(A, z(x)) νx (dA)dx (5.3)
Ω R3×3

for any v ontinuous of a growth less than p in the A-variable, while for ϕ ontinuous
satisfying (5.2) we have only
Z Z Z
lim inf ϕ(∇un , zn ) dx ≥ ϕ(A, z(x)) νx (dA)dx; (5.4)
n→∞ Ω Ω R3×3

f. [45, Theorem 3.2℄.

As νx is a gradient Lp -Young measure with R3×3 Aνx (dA) = ∇u(x) for a.a. x ∈ Ω,
R

and as ϕ(·, z(x)) is quasi onvex, for a.a. x ∈ Ω it holds

Z Z 
ϕ(A, z(x))νx (dA) ≥ ϕ Aνx (dA), z(x) = ϕ(∇u(x), z(x)). (5.5)
R3×3 R3×3

see [30, 45℄. Combining (5.4) and (5.5) yields

R
lim inf n→∞ Ω
ϕ(∇un , zn ) dx ≥
ϕ(∇u(x), z(x)) dx. As the Young measure is not involved in the last estimate
R
Ω
at all, this estimate holds, in fa t, for the whole original sequen e. 2

An example of a frame-indierent quasi onvex (in fa t even poly onvex, i.e. onvex
in terms of F and its determinant and ofa tors) energy ϕ̂(F, z) := ϕ̃(F ) satisfying
(5.2) is the Ogden-type material

p/2 p0
ϕ(F, z) = α1 tr F ⊤ F − I + α2 tr cof(F ⊤ F )−I (5.6)



+ φ0 det(F ) ;

here α1 , α2 > 0, p ≥ 3, p0 ≤ p/2, φ0 is a onvex fun tion of at most p/3 growth, and
nally tr(·) in (5.6) denotes the tra e of a matrix.
As F = I+∇u, we an express the spe i stored energy in terms of the displa ement
gradient as

(5.7)


ϕ = ϕ(∇u, z) = ϕ̂ I+∇u, z .

24
The Piola-Kir hho stress σ : R3×3 → R3×3 is given by σ = ϕ′∇u (∇u, z) = ϕ̂′F (I+
∇u, z) with ϕ′∇u and ϕ̂′F denoting the tensor-valued partial gradients.
If the displa ement gradient ∇u is small, one an negle t the quadrati term (5.1)
so that the Green-Lagrange strain tensor E from (5.6) turns into a so- alled small-
strain tensor e(u) := 21 ∇u + 21 (∇u)⊤ , i.e.
1 ∂ui 1 ∂uj
eij (u) = + , i, j = 1, ..., 3. (5.8)
2 ∂xj 2 ∂xi

For all examples below, we assume Ω ⊂ R3 to be a polyhedral domain. The dis retiza-
tion is made by a nested family of regular triangulations of Ω with the mesh param-
eter h > 0 and ΠU,h and ΠZ,h will always be onsidered as quasi-interpolation op-
erators related with standard onformal nite elements of polynomial type, namely
P0 (i.e. element-wise onstant fun tions) or P1 (i.e. element-wise ane ontinu-
ous fun Rtions). To be more expli it, we an onsider a mollier u 7→ ũh with
ũh (x) = Ω kh (x, ξ)u(ξ)dξ using a ontinuous kernelR kh : Ω×Ω → R+ supported on
an h-neighbourhood of the diagonal in Ω×Ω and Ω kh (x, ξ) dξ = 1 for all x ∈ Ω.
Then dene uh = ΠU,h u as a Lagrange pie ewise ane interpolation of ũh using
the nodal points in ase of P1-elements, or pie ewise onstant interpolation using
bary enters of the simplexes of the parti ular triangulation in ase of P0-elements.
Moreover, we will assume the nested triangulations onformal with the spe i dis-
joint partition of Γ where possibly dierent boundary onditions are pres ribed. As
to the initial ondition q0 , we will always assume its stability (3.31), e.g. ensured
through a gentle start (4.10) and thus not dis ussed in parti ular ases.

5.2 Plasti ity with hardening at small strains

The rst example on whi h we want to demonstrate our theory is a fully rate-
independent plasti ity with isotropi hardening. The ve tor of the internal param-
eters z := (π, η) ∈ L2 (Ω; R3×3
sym,0 ) × L (Ω) =: Z is therefore now omposed from the
2

plasti strain π and a hardening variable η ; here we used the notation

R3×3 A ∈ R3×3 ; A⊤ = A, tr(A) = 0 . (5.9)


sym,0 :=
For simpli ity, we onsider homogeneous Diri hlet boundary onditions on a part Γ0
of the boundary ∂Ω with nonvanishing surfa e measure, so that
U := U = u ∈ W 1,2 (Ω; R3 ); u = 0 a.e. on Γ0 }, (5.10)

2 3×3 2
(5.11)


Z := L (Ω; Rsym,0 ) × L (Ω) ∩ K
where K is the one of admissible evolution dire tions, see (5.14) below. The oin i-
den e that the z - omponent of states an be restri ted equally in the stored energy
and dissipation energy is important for (5.17) below. We postulate the stored energy
as
1
Z
E(u, z) ≡ E(u, π, η) := (e(u) − π)⊤ C(e(u) − π) + bη 2 dx (5.12)
2 Ω

25
where C = [Cijkl ] ∈ R3×3×3×3 is a positive-denite 4th-order tensor of elasti moduli
and b > 0 a hardening parameter. There are no onstraints of the type Ξ(u, π, η) = 0
so we onsider Eε ≡ E . In view of Remark 3.10, it also means that no numeri al-
integration error is expe ted. Considering the loading by a time-varying for e g
a ting on Γ1 := ∂Ω \ Γ0 , we postulate f as
Z
hf (t), (u, z)i := g(t, x)·u(x) dS. (5.13)
Γ1

The hardening is a unidire tional pro ess and is, in standardly a epted models,
ree ted by the one of admissible evolution dire tions in the form

K := {z = (π, η); η ≥ δP∗ (π) a.e. on Ω}. (5.14)

Here P ⊂ R3×3 is a onvex losed neighbourhood of the origin, δP is its indi ator
fun tion,Pand δP∗ the onjugate fun tional to δP with respe t to the duality pairing
3
σ : e = i,j=1 σij eij . Note that the physi al dimension of this pairing is Pa=J/m3 .
Hen e, δP∗ is onvex, homogeneous degree-1 and positive ex ept at the origin, and
thus K is indeed a one. The interior of P is alled elasti ity domain while its
boundary is alled the yield surfa e. More pre isely, it orresponds to the initial
elasti ity domain if η = 1 is onsidered as an initial ondition while the a tual
elasti ity domain may be inated during the loading pro ess just by the isotropi al
hardening. The ontinuous part of the degree-1 homogeneous dissipation potential
is
Z
R(z) := δP∗ (π) dx (5.15)
Ω

so that the overall dissipation distan e is, in view of (4.8),

δ (π2 −π1 ) dx if η2 −η1 ≥ δP∗ (π2 −π1 ) on Ω,

 R ∗
Ω P
D(z1 , z2 ) ≡ D(π1 , η1 , π2 , η2 ) :=
+∞ otherwise.

This leads naturally to Z1 := L1 (Ω; R3×3

sym,0 ) × L (Ω) in Remark 4.2. Beside the men-
1

tioned initial ondition η(0, ·) = 1, we must pres ribe π(0, ·) = π0 ∈ L2 (Ω; R3×3 sym,0 ).
The required stability (3.31) of q0 , a hieved e.g. through the gentle start (4.10) as
suggested in Se t. 5.1, yields z0 = (π0 , η0 ) ∈ K , i.e. here δK
∗
(π0 ) ≤ 1. The mentioned
initial ondition η0 = 1 is, in general, guaranteed by this way only if f (0) is small
enough. Moreover, it is well-known ( f. [22, 36℄) that this problem has a unique
energeti solution (u, z) ∈ W 1,∞ ([0, T ]; U × Z).
We assume a polyhedral domain Ω with also Γ0 and Γ1 having a polyhedral shape,
and assume Ω triangulated by a nested family of regular triangulations with the
mesh parameter h > 0 onformal with the partition Γ = Γ0 ∪ Γ1 , and ΠU,h and ΠZ,h
quasi-interpolation operators related with onformal P1-elements and P0-elements,
respe tively. It is also important that the P0-elements are onformal with the one
K from (5.14) used also for Z in (5.11) in the sense ΠZ,h K ⊂ K , as needed for
Propositions 4.5 and 4.6. As there is no Ξ in this problem, we have Eε = E but Rε

26
from (4.9) is to be onsidered (unless one thinks about R + δK in pla e of Re as
suggested in Remark 4.7), and also (4.24) with σ the norm topology works simply
for [q0 ]h,ε := Πh q0 .

Corollary 5.2 Let the data Ω, Γ0 , Γ1 , P , and q0 be qualied as above, and g ∈

1 4/3 3
C ([0, T ]; L (Γ1 ; R )) and [q0 ]h,ε be taken as above. Then the approximate solutions
qε,τ,h = (uε,τ,h, πε,τ,h , ηε,τ,h) with

uε,τ,h ∈ L∞ (0, T ; W 1,2(Ω; R3 )), (5.16a)

πε,τ,h ∈ L∞ (0, T ; L2 (Ω; R3×3 1
sym,0 )) ∩ BV([0, T ]; L (Ω; R
3×3
)), (5.16b)
ηε,τ,h ∈ L∞ (0, T ; L2 (Ω)) ∩ BV([0, T ]; L1 (Ω)), (5.16 )

based on the P0-elements for π and η and the P1-elements for u onverge for
(ε, τ, h) → (0, 0, 0) (even as the whole sequen e in the sense of Theorem 3.8 with
Remark 4.2) to the energeti solution of the problem given by E , R, K , f and q0
above.

Proof. The oer ivity (4.6) is ensured due to the Poin aré inequality through the
Diri hlet boundary onditions, ensuring

E(u, π, η) ≥ c kuk2W 1,2(Ω;R3 ) + kπk2L2 (Ω;R3×3 ) + kηk2L2 (Ω) (5.17)

provided also δP∗ (π) ≤ η ; note that su h oer ivity does not hold for general (π, η) ∈
Z , whi h is why for the denition (5.11) of Z the restri tion to K had to be used.
As P is assumed bounded, δP∗ is Lips hitz ontinuous, and hen e R is ontinuous.
Moreover, stability of q0 as well as (4.24) have already been dis ussed above.
Using the oer ivity of RE already proved, we an verify (3.9b) with Eε = E by
∂E
∂t
(t, q) = −h ∂f
∂t
, qi = − Γ1 ∂g
∂t
(t, x) · u(x) dS and the estimate

∂E ∂g N 2 G21 c 2
(t, q) ≤ N u W 1,2 (Ω;R3 )
≤ NG1 kqk ≤ + q
∂t ∂t L4/3 (Γ1 ;R3 ) 2c 2
N 2 G21

≤ + E(t, q) = c1 E(t, q) + c0
c
with c1 = 1 and c0 = N 2 G21 /c. Here c is from (5.17) and N is the norm of the
tra e operator u 7→ u|Γ1 in Lin(W 1,2 (Ω), L4 (Γ1 )) and G1 = kgkC 1([0,T ];L4/3 (Γ1 )) . Here
we used the estimate E(t, q) = E(q) − hf (t), qi = E(q) − Γ1 ∂g
R
∂t
(t, x) · u(x) dS ≥
ckqk − NG1 kqk ≥ 2 kqk − 2c N G1 .
2 c 2 1 2 2

Then we use the assertions from Se t. 3 through either Propositions 4.5 or 4.6. In
the former ase, the setting (4.29) takes now

e(u)⊤ Ce(u)
Z
− e(u)⊤ Cπ dx,


B(π, η) := Cπ , bη , E0 (u, π, η) :=
Ω 2

27
while for the latter ase the setting (4.35) works simply with B = E ′ and E0 = 0,
i.e.



B(u, π, η) := div C(e(u) − π) , C(π − e(u)) , bη , E0 := 0,

with the div term onsidered in the weak sense, of ourse. Note that Z + K ⊂ Z ,
holds, too. Eventually, due to the uniqueness result [22, 36℄ or [51, Se t.11.1.3℄, we
on lude that the whole sequen e onverges. 2

Remark 5.3 (Implementation without regularization by LQ-programme.) In

anisotropi media like single- rystals, the domain P is onsidered to be polyhe-
dral, f. e.g. [12℄, hen e δP∗ has a polyhedral epigraph and the in remental problem
(3.22) without any regularization ( f. Remark 4.7) represents a minimization prob-
lem of a sum of a quadrati and a polyhedral-graph fun tional whi h an be, after
a omputationally heap enhan ement, solved by e ient linear-quadrati solvers;
f. [52, Lemma 4℄ for this enhan ement.

Remark 5.4 The P0/P1-dis retization of this plasti ity problem has been already
used by Alberty and Carstensen [2℄ and thus Corollary 5.2 re overs some results
from [2℄. Note that our onvergen e result does not use higher-order regularity of
the solutions (u, z) ∈ W 1,∞ (0, T ; U×Z). Hen e we annot expe t onvergen e rates
as in [2℄ and thus our results are loser to [23℄ where the above onvergen e result
was established already by a more elaborate method.

5.3 Phase transformation: a mixture approa h

In engineering, modelling of inelasti response of the materials undergoing marten-

siti transformation is of high interest. Here we want to demonstrate our theory on
a simplied mixture-like model for martensiti transformation.
Pm
Taking Γ0 as in Se tion 5.2 and Z0 := {s ∈ Rm ; sl ≥ 0 & l=1 sl = 1} the Gibbs
simplex, we put

U := U = u ∈ W 1,p (Ω; R3 ); u = 0 a.e. on Γ0 }, (5.18)



Z := z ∈ Z := W α,2 (Ω; Rm ); z(x) ∈ Z0 for a.a. x ∈ Ω (5.19)



with α > 0 denoting (possibly a fra tional) order of derivatives of the ve tor of the
internal parameters z whi h now represents volume fra tions referring to m phases
(or phase variants). For simpli ity, we onsider the loading again through g as in
Se t. 5.2, i.e. f is again dened by (5.13). We postulate the stored energy as

κ
Z
E(u, z) := ϕ(∇u, z) dx + |z|2α (5.20)
Ω 2

28
with κ, α > 0 and | · |α denoting the usual seminorm in the Sobolev (or, for α
noninteger, Sobolev-Slobodetski) spa e, i.e.
 Z


 |∇α z|2 dx for α ∈ N,
2
|z|α = ΩZ Z
(5.21)
1 |∇[α] z(x) − ∇[α] z(ξ)|2
dxdξ for α > 0 noninteger


4 Ω Ω |x − ξ|3+2(α−[α])


with [α] the integer part of α. In prin iple, more physi ally justied kernels with a
support lo alized around the diagonal {x = ξ} with the same singular behaviour as
|x − ξ|−3−2(α−[α]) for |x − ξ| → 0 ould equally be used in (5.21).
The degree-1 homogeneous dissipation potential is now postulated as
Z
R(z) := ∗
δM (z) dx (5.22)
Ω

where δM∗
is determined, in analogy with δP∗ from Se t. 5.2, by a onvex om-
pa t neighbourhood M ⊂ Rm of the origin whi h pres ribes a tivation energies
for martensite/austenite phase-transformation or for re-orientation of parti ular
martensiti variants. In parti ular, the martensiti transformation is a reversible
pro ess, so that K = Z . Also, there is nor Ξ neither K 6= Z and thus both Eε ≡ E
and Dε ≡ D and the ε-regularization is irrelevant here.
For the dis retization, we onsider naturally P1-elements for u and either P0-
elements for z (if α < 1/2) or P1-element also for z if (α < 3/2). Again, taking
[q0 ]h,ε := Πh q0 guarantees (4.24) with σ being the norm topology.

Corollary 5.5 Let the data Ω, Γ0 , and Γ1 be qualied as in Se t. 5.2, let ϕ be

qualied as in Lemma 5.1 (note that p1 is irrelevant as Z0 is bounded here), and
further let
 2p
= 3−p for p < 3,
p# /(p# −1)
1
g ∈ C ([0, T ]; L 3
(Γ1 ; R )), where p #
< +∞ for p = 3, (5.23)
= +∞ for p > 3,


and q0 ∈ S(0) be approximated by [q0 ]h,ε := Πh q0 . Then the approximate solutions

qτ,h = (uτ,h , zτ,h ) with

uτ,h ∈ L∞ (0, T ; W 1,p(Ω; R3 )), (5.24a)

zτ,h ∈ L∞ (0, T ; W α,2(Ω; Rm )) ∩ BV([0, T ]; L1 (Ω; Rm )), (5.24b)

based on the P1-elements for u and the P0- or P1-elements for z onverge for
(τ, h) → (0, 0) (in terms of subsequen es in the sense of Theorem 3.8 with Re-
mark 4.2) to energeti solutions of the problem given by E , R, f and q0 above.

Proof. Coer ivity on Q = U × Z follows from the assumed oer ivity (5.2) of ϕ(·, z)
by Poin aré inequality ombined with the Diri hlet ondition on Γ0 and by the

29
regularizing κ-term in (5.20) ombined with the onstraint z(x) ∈ Z0 involved in Z
in (5.19).
The lower-semi ontinuity of the rst term in (5.20) needed for (3.8) follows by
Lemma 5.1 with p1 < +∞ arbitrary sin e Z0 is now bounded.
The ontinuity of R : L1 (Ω) → R follows from (in fa t is equivalent to) the assumed
boundedness of M ⊂ Rm .
The assumption in Proposition 4.4 are satised simply if σ : equals the strong
topology on W 1,p (Ω; R3 ) × W α,2 (Ω; Rm ). Here the onvexity of the Gibbs simplex
Z0 involved in Z is used, whi h makes both P0- and P1-elements ompatible with Z
in the sense ΠZ,h Z ⊂ Z , f. (4.3), whi h makes our results from Se tion 4.3 working.
2

Example 5.6 At small strains, a popular model takes a mixture of quadrati

energies in the form
m
X (e(u)−eℓ )⊤ Cℓ (e(u)−eℓ ) U ⊤ +Uℓ
ϕ(∇u, z) := zℓ + ψ(z) where eℓ := ℓ ,
ℓ=1
2 2

with the distortion matri es Uℓ of parti ular pure phases (or phase variants). The
setting here is related with the situation of martensiti transformation in a single-
rystal and z 's are volume fra tions of the so- alled austenite and of parti ular
variants of martensite, e.g. m = 4 or 7 for tetragonal or orthorhombi martensite,
respe tively. The fun tion ψ ree ts the dieren e between hemi al energies of
austenite and martensite and also between pure phases and mixtures. As ϕ(·, z)
is now onvex, it qualies for Lemma 5.1 with Z0 bounded. The philosophy of
mixtures of austenite/martensite phases in so- alled shape-memory alloys has been
proposed by Frémond [15℄; in rate-independent variant also presented in [16℄. For its
analysis and numeri al implementation see [10, 11, 17, 24℄. Gradients of mesos opi al
volume fra tions (i.e.. (5.20) with α = 1 has already been used in Frémond's model
[16, p.364℄ or [17, Formula (7.20)℄. Another way for obtaining physi ally relevant
mixture energies is the quasi onvexi ation under volume onstrains, also alled
ross-quasi onvexi ation, see [42℄.

Example 5.7 If the elasti -moduli tensors Cℓ = C are equal for all phases, the
spe i energy in Example 5.6) transforms to
m m
X (e(u)−etr (z))⊤ C(e(u)−etr (z)) X
ϕ(∇u, z) = + ψ̃(z) with etr (z) = zℓ eℓ ,
ℓ=1
2 ℓ=1

where etr (z) the is so- alled transformation strain. Note that, although (5.20) has
got now a quadrati form ex ept the lower-order term ψ̃(z), we annot use Proposi-
tion 4.5 or 4.6 be ause of the onstraint z(x) ∈ Z0 . Hen e, the quadrati stru ture
of the regularizing term κ|z|2α annot be exploited and a non-quadrati regularizing
term ould equally be onsidered through this se tion. For su h a model we refer
e.g. to [5, 6, 8, 19, 20, 28, 60℄.

30
5.4 Phase transformation: non-mixture approa h

The mixture approa h in Se t. 5.3 is rather designed for phenomenologi al models of

poly rystals but is too oarse for the des ription of ompli ated mi rostru tures o -
urring in shape-memory-alloy single- rystals. An attempt to build a mi ros opi al
model has been done in [34℄ (see also [35℄) by restri ting z to be valued in verti es of
the Gibbs' simplex, i.e. only pure phase(variant)s are allowed; then α < 1/2 should
be taken in (5.20) or, as onsidered in [34, 35℄, a BV-like term κ|∇z|. In this model,
z swit hes ϕ.
A dierent philosophy with presumably similar ee ts, pioneered by Falk [13℄, on-
siders the ve torial order parameter z related to the deformation gradient ∇u and
parti ular shapes are then swit hed rather by ∇u. Spinodal regions are then allowed
instead of mixtures. The spe i stored energy ϕ now depends only on ∇u but need
not be quasi onvex. For example, in [3, 4, 32, 50, 54℄, a multiwell potential ϕ̂ (re-
lated with ϕ by (5.7)) arises by the ombination of St.Venant-Kir hho materials
onsidered for ea h parti ular phase:
1 
fˆ(F ) := min (Uℓ−⊤ F ⊤ F Uℓ−1 − I)⊤ Cℓ (Uℓ−⊤ F ⊤ F Uℓ−1 − I) + cℓ , (5.25)
ℓ=1,..,m 2

where Uℓ are distortion matri es as in Example 5.6, Cℓ are elasti -moduli tensors,
cℓ are some onstants, and Uℓ−⊤ := (Uℓ⊤ )−1 . Now naturally p = 4.
We postulate the stored energy in terms of E and Ξ as
κ
Z
E(u, z) := ϕ(∇u) dx + |u|2α , (5.26)
Ω 2
Ξ(u, z) := z − L(∇u), (5.27)

with κ > 0, α > 1 and L : R3×3 → Z0 playing the role of a phase indi ator
with Z0 being again the Gibbs simplex. The seminorm | · |α dened in (5.21) used
for 1 < α < 2 with the Frobenius norm in the enumerator, now a ting on (3×3)-
matri es is frame-indierent, as observed by Arndt in [3℄. We onsider the same
loading as in Se ts. 5.2 and 5.3, i.e. f from (5.13), but now we put

U := U = u ∈ W α,2 (Ω; R3 ); u = 0 a.e. on Γ0 }, (5.28)



Z := z ∈ Z := L (Ω; R ); z(x) ∈ Z0 for a.a. x ∈ Ω ,

2 m
(5.29)


and then naturally X := Z . The dissipation potential R is again from (5.22). There
is no K involved, hen e Dε = D , but as Ξ from (5.27) o urs, the regularization Eε
is, in prin iple, to be onsidered.
Choosing α < 3/2 allows for the usage of P1-elements for u and P0-elements for z .
As now Q = Q and K = Z , so in parti ular their onformity (4.3) is automati .
The proof of the following assertion shows that they are onformal also with the
onstraints Ξ(q) = 0 so, in view of Remark 4.7, it would be possible to avoid the
ε-regularization at all. When taking [u0 ]h,ε = ΠU,h u0 , we have ∇[u0 ]h,ε element-wise
onstant and so is L(∇[u0 ]h,ε ) =: [z0 ]h,ε , and (4.24) is satised.

31
Corollary 5.8 Let ϕ : R3×3 → R be ontinuous (not ne essarily quasi onvex)
satisfying (5.2) here with m := 0 (so no z -dependen e), let g satisfy (5.23),
L : R3×3 → Z0 be ontinuous, and α ∈ (1, 3/2) and p < 6/(5−2α) in (5.2), and q0
and [q0 ]h,ε as spe ied above. Then the approximate solutions qε,τ,h = (uε,τ,h , zε,τ,h )
with

uε,τ,h ∈ L∞ (0, T ; W α,2(Ω; R3 )), (5.30a)

zε,τ,h ∈ L∞ (0, T ; L∞ (Ω; Rm )) ∩ BV([0, T ]; L1 (Ω; Rm )), (5.30b)

based on the P1-elements for u and the P0-elements for z onverge for (ε, τ, h) →
(0, 0, 0) (in terms of subsequen es in the sense of Theorem 3.8 with Remark 4.2) to
energeti solutions of the problem given by E , R, f and q0 above.

Proof. Weak lower semi ontinuity of E is due to onvexity of the regularizing term
κ|u|2α in (5.26) while ϕ is now treated by ompa tness of the embedding W α,2 (Ω) ⊂
W 1,p (Ω) (guaranteed if p < 6/(5−2α)) as a lower-order term. The limit passage
in the z -variable is trivial. This ompa tness also ensures the weak ontinuity of
Ξ : U × Z → X.
As K = Z , ondition (4.23) with σ being the strong topology holds, if we show, for
σ
given z̃ = L(∇ũ), the existen e of (ũh , z̃h ) → (ũ, z̃) su h that z̃h = L(∇ũh ). As far
as ũh , this an be done by a density argument of smooth fun tions in W α,2 (Ω; R3 ),
and then the usual Lagrange interpolation. By the embedding W α,2 (Ω) ⊂ W 1,p (Ω),
s s
∇ũh → ∇ũ in Lp (Ω; R3×3 ) and z̃h = L(∇ũh ) → L(∇ũ) = z̃ by ontinuity of the
Nemytski mapping indu ed by L.
Then we use the results from Se t. 3 via Proposition 4.4 with σ being the strong
topology on W α,2 (Ω; R3 ) × L2 (Ω; Rm ). 2

Remark 5.9 The inequalities α < 3/2 and p < 6/(5−2α) restri t us to p < 3,
whi h unfortunately ex ludes (5.25). Hen e we are tempted to take higher α whi h,
however, brings the ne essity to use higher-order elements (or to split the problem
to a system). Considering P2-elements for u would allow for α < 5/2 whi h, in turn,
would allow for arbitrarily high p. Sin e L is inevitable nonlinear, it is no longer
onformal with the onstraint Ξ(q) = 0 no matter what (polynomial) elements are
taken for z . This would drive us to a penalization te hnique based on Proposition 4.3.
However, here it is simpler to modify our analysis to allow for expressing the model
in terms of u only, f. the following Remark 5.10.

Remark 5.10 In fa t, a vis ous rate-dependent variant of the above model was
proposed in [50℄, for the rate-independent dissipation term f. [50, Formula (33)℄.
The regularizing term | · |α used for α < 1/2 and the P0/P1-dis retization was sug-
gested and implemented in [3℄ and omputational experiments on NiMnGa single
rystals reported in [4℄. In [46℄, the model was analyzed and implemented in the
1-dimensional ase with α = 2. Pure analysis then followed also in [47℄; in parti u-
lar for α ≥ 3, [47, Prop.3℄ investigated an invis id variant of this model a ounting,

32
 ontrary to our paper, also for inertial ee ts. In fa t, the model was formulated
only in termsR of u in [3, 46, 47, 50℄ but then the dissipation distan e took the form
D(u1, u2 ) = Ω |L(∇u1) − L(∇u2)| dx, having lost the stru ture based on the degree-
1 homogeneous potential R. Negle ting di ulties in numeri al evaluation of su h
D if α = 2 would be onsidered, by this way one gets rid of the ne essity to penalize
Ξ (whi h, in ase α < 1/2, is made possible due to Corollary 5.8 together with
Remark 4.7 in our ase too). Nevertheless, a fully rate-independent model, used
in fa t for al ulations in [4℄, has not been subje ted to any rigorous mathemati-
al/numeri al analysis, and therefore Corollary 5.8 brings indeed new results.

5.5 Damage at large strains

In engineering, other inelasti pro ess in the materials of a high interest is damage.
We onsider a fully rate-independent isotropi and nonlo al damage, and again
onsider the body Ω xed on a nonvanishing part Γ0 and loaded by a surfa e for e
g on Γ1 = ∂Ω \ Γ0 , so that U = U is again from (5.18). As we onsider isotropi
damage, the internal parameter z ∈ Z will be s alar valued with

Z := z ∈ Z := W α,2 (Ω); z(x) ≥ 0 for a.a. x ∈ Ω . (5.31)



We postulate the stored energy again by the formula (5.20); κ > 0 in (5.20) is now
a oe ient responsible for nonlo al ee ts in gradient-of-damage theories as, e.g.,
in [16℄, f. [38℄ for a dis ussion and more referen es. Note that we admitted, rather
formally, ϕ operating on the argument z nonrestri ted from above to allow for a
simple onstru tion of the re overy sequen e (4.30). The loading f is onsidered
again by (5.13).
Like isotropi hardening in Se t. 5.2, the pro ess of damaging is unidire tional in
the sense that, if in progress, it an only in rease but the material never an heal,
whi h leads us to dene the one of admissible evolution dire tions as

K := {z ∈ W α,2 (Ω); z ≥ 0 a.e. on Ω} ≡ Z. (5.32)

The degree-1 homogeneous dissipation potential is onsidered as

Z
R(z) := c1 z dx, (5.33)
Ω

where c1 is a phenomenologi al spe i energy (with physi al dimension J/m3 =Pa)

expressing the energy needed for a damage of a unit volume des ribed by a unit
jump of the damage parameter z . Considering the initial ondition for z0 = 0
and ϕ(A, ·) de reasing for z ∈ [0, 1] and with ϕ(A, z) = ϕ(A, 1) + (z − 1)2 for
z ∈ (1, +∞), we ee tively for e the values of z to range only the interval [0, 1] and
c1 refers to the spe i energy dissipated by damaging the original material (having
the stored-energy ϕ(·, 0)) to the fully damaged material (having the stored-energy
ϕ(·, 1) assumed to be still oer ive so we ex lude the ase when the material fully
disintegrates).

33
As no equality onstraints of the type Ξ(q) = 0 are involved, we have Eε = E but
the ε-regularization Dε from (4.9) is to be still onsidered unless one takes R + δK
instead of Rε , f. Remark 4.7. For the dis retization, as in Se t. 5.3, we onsider
P1-elements for u and either P0-elements for z (if α < 1/2) or P1-element also for z
if (α < 3/2). Again, both P0- and P1-elements are onformal with the onstraints
in Z = K from (5.31)(5.32) in the sense ΠZ,h Z ⊂ Z and ΠZ,h K ⊂ K , as required
in Proposition 4.5.

Corollary 5.11 Let the data Ω, Γ0 , and Γ1 be qualied as in Se t. 5.2, let ϕ be

qualied as in Lemma 5.1 with m := 1 and Z0 := {z ≥ 0} and p1 := 2, let g
satisfy (5.23), and let q0 ∈ S(0) and [q0 ]h,ε := Πh q0 . Then the approximate solutions
qε,τ,h = (uε,τ,h, zε,τ,h ) with

uε,τ,h ∈ L∞ (0, T ; W 1,p(Ω; R3 )), (5.34a)

zε,τ,h ∈ L∞ (0, T ; W α,2(Ω)) ∩ BV([0, T ]; L1 (Ω)), (5.34b)

based on the P1-elements for u and the P0- or P1-elements for z onverge for
(ε, τ, h) → (0, 0, 0) (in terms of subsequen es in the sense of Theorem 3.8 with
Remark 4.2) to energeti solutions of the problem given by E , R, K , f and q0
above.

Proof. Coer ivity on Q = U × Z follows from the assumed ondition |A|p ≤ ϕ(A, z)
by the Poin aré inequality ombined with the Diri hlet ondition on Γ0 and by the
regularizing κ-term in (5.20) ombined with the onstraint z(x) ≥ 0 involved in Z
in (5.31). Then we use Proposition 4.5 with the de omposition (4.29) using B = E1′
with E1 (z) := κ2 |z|2α and E0 (u, z) = Ω ϕ(∇u, z) dx. Note also that [q0 ]h,ε := Πh q0
R

satises (4.24). 2

Remark 5.12 The partial damage at large strains has been analyzed in [38℄ but
without any numeri al approximation and nonquadrati regularizing term pκ1 |∇z|p1
with p1 > 3 had to be used, ontrary to the quadrati term in (5.20) whi h is
usual in engineering literature but never was mathemati ally analyzed so far. Hen e
Corollary 5.11 represents a new extension in this eld.

Example 5.13 (Engineering  (1−d)-model.) Considering two materials having

linear response des ribed in small strains by elasti moduli tensors C1 and C2 , the
rst one undergoing a damage in a linear way leads to the potential ϕ in the form
e(u)⊤ C1 e(u) e(u)⊤ C2 e(u)
ϕ(∇u, z) = (1−z)+ + + ((z−1)+ )2
2 2
where (·)+ = max(0, ·). This potential satises all our assumptions with p = p1 = 2
in (5.2) if C1 is positive semidenite and C2 positive denite. Su h a model is alled
in engineering literature a 1−d model (here rather 1−z ) and an be used for two-
omponent materials as e.g. lled polymers or lled rubbers whi h do not undergo
a full damage.

34
5.6 Debonding at large strains

Other inelasti pro esses may o ur not in the materials themselves but on the
boundary. Here we want to onsider a possible debonding of an elasti support on a
part Γ2 of the boundary ∂Ω. The internal parameter z ∈ L∞ (Γ2 ) is therefore now a
s alar debonding parameter assumed to range [0, 1] and expressing volume fra tion
of the adhesive whi h xes elasti ally the body on Γ2 if not debonded. It is natural
also to onsider a unilateral Signorini onta t on Γ2 . Moreover, we again onsider
the body Ω xed on a nonvanishing part Γ0 of ∂Ω (disjoint with Γ2 ) and loaded by
a surfa e time-varying for e g on Γ1 = ∂Ω \ (Γ0 ∪ Γ2 ), so that

U := u ∈ W 1,p (Ω; R3 ); u = 0 a.e. on Γ0 , ν · u ≥ 0 a.e. on Γ2 }, (5.35)



Z := z ∈ Z := L∞ (Γ2 ); 0 ≤ z ≤ 1 a.e. on Γ2 (5.36)



with ν = ν(x) a normal to Γ2 . We postulate the stored energy as

Z Z
E(u, z) := ϕ(∇u) dx + (1 − z)ψ(u) dS, (5.37)
Ω Γ2

where ψ : R3 → R+ des ribes the elasti response of the adhesive xing the body
on Γ2 .
Considering naturally that debonding an only develop but never heal ba k leads
us to pose the one of admissible evolution dire tions as

K := {z ∈ L∞ (Γ2 ); z ≥ 0 a.e. on Γ2 }. (5.38)

Similarly like in (5.33), the degree-1 homogeneous dissipation potential is

Z
R(z) := c2 z dS (5.39)
Γ2

with c2 a phenomenologi al spe i energy (with physi al dimension J/m2) express-

ing the energy needed for a full debonding of a unit area of Γ2 .
Natural nite-element approximation is now P1-elements for u and P0-elements on
the boundary for z . We assume that the disjoint partition Γ = Γ0 ∪ Γ1 ∪ Γ2 is
polyhedral and that the nested triangulations are onformal with this partition. To
simplify te hni al details, let us assume that Γ2 is at; this ensures ΠU,h U ⊂ U ,
f. (4.3). Also the onstraints in (5.36) are onformal with P0-elements in the sense
ΠZ,h Z ⊂ Z . As there is no Ξ here, we have Eε ≡ E but Dε 6= D is still to be
onsidered.

Corollary 5.14 Let the disjoint partition Γ = Γ0 ∪ Γ1 ∪ Γ2 be polyhedral, Γ2 at,

and the nested triangulations be onformal with this partition, ϕ be qualied as in
Lemma 5.1 with n := 0 (i.e. with no z -dependen e in ϕ), g satisfy (5.23), and
3 #
ψ:R →R be ontinuous satisfying the growth ondition 0 ≤ ψ(u) ≤ C(1 + |u|p −ǫ )

35
with p# from (5.23) and ǫ > 0, and q0 ∈ S(0) is approximated by [q0 ]h,ε := Πh q0 .
Then the approximate solutions qε,τ,h = (uε,τ,h, zε,τ,h) with
uε,τ,h ∈ L∞ (0, T ; W 1,p(Ω; R3 )), (5.40a)
zε,τ,h ∈ L∞ (0, T ; L∞ (Γ2 )) ∩ BV([0, T ]; L1 (Γ2 )), (5.40b)
based on the P1-elements for u and the P0-elements for z onverge for (ε, τ, h) →
(0, 0, 0) (in terms of subsequen es in the sense of Theorem 3.8 with Remark 4.2) to
some energeti solutions of the problem given by E , R, K , f and q0 above.

Proof. The oer ivity of E follows as in Corollary 5.5; note that the term on Γ2 ,
being nonnegative, annot destroy it. The weak lower-semi ontinuity is again as in
Corollary 5.5, the term on Γ2 being even weakly ontinuous due to anity in z -
variable and due to the ompa tness of the tra e operator u 7→ u|Γ2 : W 1,p (Ω; R3 ) →
#
Lp −ǫ (Γ2 ; R3 ).
We will expli itly onstru t the re overy sequen e {q̃h }h>0 for (4.23). As to ũh we
use the onstru tion (4.30a); as Γ2 is at, ν is onstant on Γ2 , and ΠU,h U = Uh ∩ U ,
whi h ensures Uh ⊂ U be ause Uh := ΠU,h U . As for ΠZ,h , we have in mind the
standard Clément's quasi-interpolation by element-wise onstant averages, hen e
e.g. fun tions valued in [0,1℄ are again mapped to (element-wise onstant) fun tions
valued in [0,1℄. Then we put
 1 − z̃ 
z̃h := 1 − (1 − zh )ΠZ,h . (5.41)
1−z
If z(x) = 1, then also z̃(x) = 1 be ause always z ≤ z̃ ≤ 1 and the fra tion in (5.41)
an be dened arbitrarily in valued [0, 1]. The produ t of element-wise onstant
fun tions 1 − zh and ΠZ,h ( 1−z
1−z̃
) is again element-wise onstant, hen e zh ∈ Zh . As
0 ≤ ΠZ,h ( 1−z ) ≤ 1, we have also zh ≤ z̃h ≤ 1, hen e z̃h ∈ Zh and z̃h − zh ∈ K .
1−z̃

s
As ΠZ,h ( 1−z̃
1−z
) → 1−z̃
1−z
in any Lp (Γ2 ), p < +∞, and zh w*
→ z , from (5.41) we have
z̃h w*
→ 1 − (1 − z) 1−z̃
1−z
= z̃ in fa t in L∞ (Γ2 ) due to the a-priori bound of values in
[0,1℄.
Then, having (4.23) proved, we an verify (3.16) through Proposition 4.4 used with
the topology σ := s × w∗ on W 1,p (Ω; R3 ) × L∞ (Γ2 ). 2

Remark 5.15 As we do not have any gradient-type regularization like in Se t. 5.5,

we had to assume ψ(u, ·) ane to allow for a passage via weak onvergen e. It
however does not allow for any arti ial denition of ψ like we did for ψ in Se t. 5.5
for z > 1, whi h is why here we had to in lude the onstraint z(x) ∈ [0, 1] expli itly
into Z in (5.36) but this, in turn, destroyed any quadrati stru ture in z and hen e
we had to rely on Proposition 4.4 supported by the rather sophisti ated onstru tion
(5.41).

Remark 5.16 A debonding on a-priori pres ribed surfa es inside the body, alled
then rather a delamination,ould be treated similarly only by introdu ing a more

36
extensive notation, f. [31℄. Let us emphasize that Corollary 5.14 adapted to su h a
problem substantially improves results from [31℄, where onvergen e has only been
proved for a semidis retization in time while the onvergen e of the full time-spa e
dis retization has only silently been assumed under an additional onvergen e ri-
terion h/τ → 0.

5.7 Magnetostri tion at small strains

In this se tion, we illustrate our theory on a deformable ferromagnet o upying a

domain Ω ⊂ R3 and undergoing quasistati isothermal evolution at small strains.
Again, the non-dissipative omponent u : Ω → R3 will be the displa ement while the
dissipating variable z : Ω → R3 will now be the magnetization ve tor; thus m = 3
here. The stored energy is then onsidered in the form

κ µ0
Z   Z
2 2
E(u, z) := ϕ ∇u(x), z(x) + ∇z dx + ∇φ dx. (5.42)
Ω 2 2 R 3

The parti ular terms in (5.42) represent the me hani al stored energy intera ting
with an anisotropi magnetization energy, the ex hange energy (with κ > 0 a oef-
 ient having a quantum-me hani al origin), and the energy of the demagnetizing
eld φ ∈ W
1,2
(Ω) (with µ0 > 0 the va uum permeability) whi h is determined by
the magnetization z by the (weak solution to the) following 2nd-order linear ellipti
equation on the whole spa e R3 :

div(µ0 ∇φ − χΩ z) = 0 on R3 , (5.43)

where χΩ : R3 → {0, 1} denotes the hara teristi fun tion on Ω. The external
for ing might be both me hani al and magneti al. Let us onsider it again via a
surfa e for e g (as in Se tion 5.2) and by an external magneti eld hext :
Z Z
hf (t), (u, z)i := g(t, x) · u(x) dS + hext (t, x) · z(x) dx. (5.44)
Γ1 Ω

Contrary to the previous se tions, z is not any internal parameter be ause it an be

subje ted dire tly to outer loading by hext . For notational simpli ity, we onsider
again the Diri hlet ondition on Γ0 and then take U = U from (5.18) while Z is
naturally to be taken as W 1,2 (Ω; R3 ). The standard model involves also the so-
alled Heisenberg onstraint

z(x) = ms for a.a. x (5.45)

with ms > 0 a given saturation magnetization. In fa t, due to (5.45) we an redene

ϕ(A, z) for |z| > ms , if needed, suitably so that the oer ivity (5.2) holds. For the
dissipation potential R we onsider, for example,
Z
R(z) := d0 z + d1 e3 · z dx (5.46)
Ω

37
where d0 > 0 and d1 ≥ 0 and e3 = (0, 0, 1). The d0 -term has been onsidered in
[58℄ while for the d1 -term see [59℄ or also [52, 53℄. The former term orresponds
to a basi dissipation and ensures oer ivity of R while the latter term des ribes
dissipation during remagnetization in a uniaxial magnet with easy-magnetization
axis in the dire tion e3 ; then the anisotropi energy ϕ(A, ·) is assumed to have
minima along this axis and d0 + d1 is a so- alled oer ive for e whi h determines
the width of a parent hysteresis loop during y li magnetization pro esses. The
magnetization pro ess is fully reversible (be ause we do not onsider any sort of
unidire tional hardening like in [53℄) and therefore we put K = Z = W 1,2 (Ω; R3 ).
The initial magnetization z0 should satisfy the onstraint (5.45) and, together with
u0 , be stable with respe t to the loading hext (0, ·) and g(0, ·); we will not spe ify this
rather te hni al ondition.
We annot simply involve the onstraint (5.45) into Z be ause (4.3) annot onven-
tionally be a hieved be ause no polynomial nite elements are ompatible with the
Heisenberg onstraints (5.45). Hen e we implement it by Ξ and then take simply
Z := Z = W 1,2 (Ω; R3 ) and dene Ξ as

|z|2 − m2s
2
Ξ : U × Z → X := L (Ω) : (u, z) 7→ p . (5.47)
|z|2 + m2s

Note that the nonlinearity r 7→ (|r|2 − m2s )/ |r|2 + m2s involved in (5.47) has a
p

linear growth and is Lips hitz ontinuous, and so is Ξ : L2 (Ω; R3 × R3 ) → L2 (Ω).

Simultaneously, Ξ is weakly ontinuous on U × Z due to the ompa t embedding of
U × Z into L2 (Ω; R3 × R3 ).
Again we onsider a polyhedral domain Ω and its nested regular triangulations,
and in view of (5.42) take P1-elements both for u and z . Then, in prin iple, exa t
integration formulae an be exploited for (5.43) and for the last term in (5.42), too.
So no dis retization of ϕ would be needed, although pra ti al al ulations usually
exploit some numeri al approximation of this pro edure (and hen e of E itself, too).
As we did not onsider it in previous se tions, we omit it here too. Be ause of
the mentioned in ompatibility of the P1-elements (and in fa t with any polynomial
nite-elements), with the onstraint Ξ(u, z) = 0, i.e. |z| = ms , we must onsider the
penalization method. Using α = 2 in (4.5), it yields

2 2 2


|z| −m
Z 

κ µ0
Z
s
Eε (u, z) = 2
ϕ ∇u(x), z(x) + |∇z| +
dx+ |∇φ|2 dx. (5.48)
Ω 2 ε |z| 2 +m2
s 2 R 3

The onformity (4.3) is here automati be ause there are no other onstraints in-
volved, i.e. Q ≡ Q and K ≡ Z .

Corollary 5.17 Let the data Ω, Γ0 , and Γ1 be qualied as in Se t. 5.2, let ϕ be

qualied as in Lemma 5.1 with Z0 := Rm , m = 3, p1 = 2, let g satisfy (5.23), and
let further hext ∈ C 1 ([0, T ]; L6/5 (Ω; R3 )), q0 ∈ S(0) and [q0 ]h,ε := Πh q0 . Then the

38
approximate solutions qε,τ,h = (uε,τ,h, zε,τ,h ) with

uε,τ,h ∈ L∞ (0, T ; W 1,p(Ω; R3 )), (5.49a)

zε,τ,h ∈ L∞ (0, T ; W 1,2(Ω; R3 )) ∩ BV([0, T ]; L1 (Ω; R3 )), (5.49b)
based on the P1-elements and the penalization of the Heisenberg onstraint (5.45)
as in (5.48) onverge for (ε, τ, h) → (0, 0, 0) (in terms of subsequen es in the sense
of Theorem 3.8 with Remark 4.2) to energeti solutions of the problem given by E,
R, Ξ, f and q0 above under the onvergen e riterion h2 /ε → 0.

For the onvergen e riterion h ≤ H(ε) an take H , e.g., in the form

H(ε) = εa with any 0 < a < 1/2. (5.50)

Proof of Corollary 5.17. The weak lower semi ontinuity in the sense (3.8) of the
ϕ-term in (5.48) is by Lemma 5.1, while that of the terms |∇z|2 and |∇φ|2 is due to
the onvexity and linearity of (5.43). The penalty term in (5.48) has the 2nd-order-
polynomial growth and is therefore ontinuous be ause of the ompa t embedding
of W 1,2 (Ω) into L2 (Ω). The oer ivity of E on U × Z follows from (5.2) through
Poin aré's inequality.
For our P1-elements, the estimate (4.18) with γ = 1 is then known to hold with | · |
and k · k being respe tively the L2 - and the W 1,2 -norms. The Lips hitz ontinuity
(4.17) of Ξ from (5.47) holds for X := L2 (Ω), whi h just makes the penalty form in
(5.48) with α = 2. The hoi e [q0 ]h,ε := Πh q0 again satises (4.24). Our assertion
then follows from Theorem 3.8 through Proposition 4.3 where (4.19) just says that
√
h = o( ε), as laimed. 2

Remark 5.18 Referen es for magnetostri tion usually addresses large strains where
more ompli ations arise, f. [9, 26, 27, 55, 57℄. Mathemati al analysis at large
strains needs some additional regularization, e.g. like [55℄. A onventional form
of ϕ in (5.42) in term of small strains, as onsidered here, is ϕ(∇u, z) = ϕ0 (z) +
1
2
(e(u) − ez )⊤ C(e(u) − ez ) with ez a preferred strain tensor orresponding to the
magnetization z ; for the on rete form of ez we refer to [27, 57℄. No numeri al and
even purely theoreti al analysis of this rate-independent evolution problem seems
to be reported in literature hen e Corollary 5.17 represents a new result for this
on eptual algorithm.

A knowledgments. The rst author a knowledges the support from C18-subproje t of

the Resear h Center Matheon (Deuts he Fors hungsgemeinshaft). The se ond author

a knowledges the hospitality of Weierstraÿ Institute, Berlin, where the majority of this

resear h has been arried out, supported through the Alexander von Humboldt Foundation.
Partial support of this resear h from the grants A 1075402 (GA AV ƒR), and LC 06052 and
MSM 21620839 (M’MT ƒR) as well as from the European grants HPRN-CT-2002-00284

Smart systems and MRTN-CT-2004-505226 Multi-s ale modelling and hara terisation

for phase transformations in advan ed materials is a knowledged, too.

39
Referen es
[1℄ E. A erbi, N. Fus o, Semi ontinuity problems in the al ulus of variations. Ar h. Ration.
Me h. Anal. 86 (1984) 125-145.
[2℄ Alberty, J., Carstensen, C.: Numeri al analysis of time-dependent primal elastoplasti ity
with hardening. SIAM J. Numer. Anal. 37 (2000), 12711294.
[3℄ Arndt, M., Griebel, M., Roubí£ek, T.: Modelling and numeri al simulation of martensiti
transformation in shape memory alloys. Continuum Me h. Thermodyn. 15 (2003), 463-485.
[4℄ Arndt, M., Griebel, M., Novák, V., Roubí£ek, T., ’ittner, P.: Martensiti transformation in
NiMnGa single rystals: numeri al simulations and experiments. Int. J. Plasti ity, 22 (2006),
1943-1961.
[5℄ Auri hio, F., Mielke, A., Steaneli, U.: A rate-independent model for the isothermal quasi-
stati evolution of shape-memory materials. In preparation.
[6℄ Auri hio, F., Petrini, L.: Improvements and algorithmi al onsiderations on a re ent three-
dimensional model des ribing stress-indu ed solid phase transformations. Int. J. Numer.
Meth. in Engr. 55 (2002), 12551284.
[7℄ B. Bourdin, G. Fran fort, J.-J. Marigo: Numeri al experiments in revisited brittle fra ture.
J. Me h. Phys. Solids 48 (2000), 797826.
[8℄ Boyd, J.G., Lagoudas, D.C.: A thermodynami al onstitutive model for shape memory ma-
terials. Part I. The monolithi shape memory alloys. Int. J. Plasti ity 12 (1996), 805842.
[9℄ Brown, W.F.: Magnetoelasti intera tions. In: C.Truesdel, ed., Springer Tra ts in Natural
Philosophy 9, Springer, 1966.
[10℄ Colli, P., Frémond, M. and Visintin, A.: Thermo-me hani al evolution of shape memory
alloys. Quarterly Appl. Math. 48 (1990), 3147.
[11℄ Colli, P., Sprekels, J.: Global existen e for a three-dimensional model for the thermo-
me hani al evolution of shape memory alloys. Nonlinear Anal. 18 (1992), 873888.
[12℄ Conti, S., Ortiz, M.: Dislo ation mi rostru tures and efe tive behaviour of single rystals.
Ar hive Rat. Me h. Anal. 176 (2005), 103147.
[13℄ Falk, F.: Model free energy, me hani s and thermodynami s of shape memory alloys. A ta
Metallurgi a 28 (1980), 1773-1780.
[14℄ Fran fort, G., Mielke, A.: An existen e result for a rate-independent material model in the
ase of non onvex energies. J. reine u. angew. Math. 595 (2006), 5591.
[15℄ Frémond, M.: Matériaux à mémoire de forme. C.R. A ad. S i. Paris 304, Série II (1987),
239244.
[16℄ Frémond, M.: Non-Smooth Thermome hani s. Springer, Berlin, 2002.
[17℄ Fried, E., Gurtin, M.E.: Dynami solid-solid transitions with phase hara terized by an order
perameter. Physi a D 72 (1994), 287308.
[18℄ A. Gia omini, M. Ponsiglione: Dis ontinuous nite element approximation of quasistati
ra k growth in nonlinear elasti ity. Math. Models Methods Appl. S i. 16 (2006), 77118.
[19℄ Govindjee, S., Miehe, Ch.: A multi-variant martensiti phase transformation model: for-
mulation and numeri al implementation. Comp. Meth. in Appl. Me h. Engr. 191 (2001),
215238.
[20℄ Govindjee, S., Mielke, A., Hall, G.J.: Free-energy of mixing for n-variant martensiti phase
transformations using quasi- onvex analysis. J. Me h. Physi s Solids 50 (2002), 18971922.
[21℄ Halphen, B, Nguyen, Q.S.: Sur les matériaux standards généralisés. J. Mé anique 14 (1975),
3963.

40
[22℄ Han, W., Reddy, B.D: Plasti ity. Mathemati al theory and numeri al analysis. Springer, New
York, 1999.
[23℄ Han, W., Reddy, B.D: Convergen e of approximations to the primal problem in plasti ity
under onditions of minimal regularity. Numer. Math. 87 (2000), 283315.
[24℄ Homann, K.-H., Niezgódka, M., Zheng Songmu: Existen e and uniqueness of global solu-
tions to an extended model of the dynami al development in shape memory alloys. Nonlinear
Analysis, Th. Meth. Appl. 15 (1990), 977990.
[25℄ Huber, J.E., Fle k, N.A., Landis, C.M., M Meeking, R.M.: A onstitutive model for ferro-
ele tri poly rystals. J. Me h. Phys. Solids 49 (2001), 785811.
[26℄ James, R.D., Kinderlehrer, D.: Theory of magnetostri tion with appli ations to
Tbx Dy1−x Fe2 . Phil. Mag. 68 (1993), 237274.
[27℄ James, R.D., Wuttig, M.: Magnetostri tion of martensite. Phil. Mag. A 77 (1998), 1273.
[28℄ Jung, Y., Papadopoulos, P., Rit hie, R.O.: Constitutive modeling and numeri al simulation
of multivariant phase transformation in superelasti shape-memory alloys. Int. J. Numer.
Meth. Engineering 60 (2004), 429460.
[29℄ Kamlah, M.: Ferroele tri and ferroelasti piezo erami s-modelling of ele trome hani al hys-
teresis phenomenona. Cont. Me h. Thermodyn. 13 (2001), 219268.
[30℄ Kinderlehrer, D., Pedregal, P.: Gradient Young measures generated by sequen es in Sobolev
spa es. J. Geom. Anal. 4 (1994), 5990.
[31℄ Ko£vara, M., Mielke, A., Roubí£ek, T.: A rate-independent approa h to the delamination
problem. Mathemati s and Me hani s of Solids 11 (2006), 423-447.
[32℄ Kruºík, M., Mielke, A., Roubí£ek, T.: Modelling of mi rostru ture and its evolution in shape-
memory-alloy single- rystals, in parti ular in CuAlNi. Me ani a 40 (2005), 389-418..
[33℄ Levitas, V.I.: The postulate of realizibility: formulation and appli ations to postbifur ational
behavior and phase transitions in elastoplasti materials. Int. J. Eng. S i. 33 (1995), 921971.
[34℄ Mainik, A.: A rate-independent model for phase transformations in shape-memory alloys.
PhD-thesis, Fa hberei h Mathematik, Universität Stuttgart, 2004.
[35℄ Mainik, A., Mielke, A.: Existen e results for energeti models for rate-independent systems.
Cal . Var. 22 (2005), 7399.
[36℄ Mielke, A.: Evolution of rate-independent systems. In: Handbook of Dierential Equations:
Evolutionary Di. Eqs. (Eds. C.Dafermos, E.Feireisl), pp. 461559, Elsevier, Amsterdam,
2005.
[37℄ Mielke, A., Roubí£ek, T.: Rate-independent model of inelasti behaviour of shape-memory
alloys. Multis ale Modeling Simul. 1 (2003), 571-597.
[38℄ Mielke, A., Roubí£ek, T.: Rate-independent damage pro esses in nonlinear inelasti ity. M3AS
16 (2006), 177-209.
[39℄ Mielke, A., Roubí£ek, T., Steaneli, U.: Relaxation and Γ-limits for rate-independent evolu-
tion equations. (Preprint No.1156, WIAS, Berlin, 2006) Cal . Var. P.D.E., submitted.
[40℄ Mielke, A., Theil, F.: A mathemati al model for rate-independent phase transformations
with hysteresis. In: Models of ontinuum me hani s in analysis and engineering. (Eds.: H.-
D.Alber, R.Balean, R.Farwig), Shaker Ver., Aa hen, 1999, pp.117-129.
[41℄ Mielke, A., Theil, F.: On rate-independent hysteresis models. Nonlin. Di. Eq. Appl. 11
(2004), 151189.
[42℄ Mielke, A., Theil, F., Levitas, V.I.: A variational formulation of rate-independent phase
transformations using an extremum prin iple. Ar hive Rat. Me h. Anal. 162 (2002), 137
177.

41
 [43℄ Mielke, A., Timofte, A.: An energeti material model for time-dependent ferroele tri behav-
ior: existen e and uniqueness. Math. Methods in the Appl. S i. 29 (2006), 1393-1410.
[44℄ Müller, S.: Variational models for mi rostru ture and phase transitions. (Le t.Notes No.2,
Max-Plan k-Institut für Math., Leipzig, 1998). In: Cal ulus of variations and geometri
evolution problems. (Eds.: S.Hildebrandt et al.) Le t. Notes in Math. 1713 (1999), Springer,
Berlin, pp.85210.
[45℄ Pedregal, P.: Parametrized Measures and Variational Prin iples. Birkhäuser, Basel, 1997.
[46℄ Ple há£, P., Roubí£ek, T.: Vis o-elasto-plasti model for martensiti phase transformation in
shape-memory alloys. Math. Methods in the Appl. S i. 25 (2002), 1281-1298.
[47℄ Rajagopal, K.R., Roubí£ek, T.: On the ee t of dissipation in shape-memory alloys. Nonlinear
Anal., Real World Appl. 4 (2003), 581-597.
[48℄ Romanowski, H., S hröder, J.: Modelling of the nonlinear ferroele tri hysteresis within
a thermodynami ally onsistent framework. In: Trends in Appli ations of Mathemati s to
Me hani s (Eds. Y.Wang, K.Hutter), Shaker Ver., Aa hen, 2005, pp.419428.
[49℄ Roubí£ek, T.: A note on an intera tion between penalization and dis retization. In: Pro .
IFIP-IIASA Conf. Modelling and Inverse Problems of Control for Distributed Parameter Sys-
tems. (A.Kurzhanski, I.Lasie ka, eds.) Le t. Notes in Control and Inf. S i. 154, Springer,
1991, pp.145-150.
[50℄ Roubí£ek, T.: Dissipative evolution of mi rostru ture in shape memory alloys. In: Le tures on
Applied Mathemati s. (H-J.Bungartz, R.H.W.Hoppe, C.Zenger, eds.) Springer, Berlin, 2000,
pp.4563.
[51℄ Roubí£ek, T.: Nonlinear Partial Dierential Equations with Appli ations. Birkhäuser, Basel,
2005.
[52℄ Roubí£ek, T., Kruºík, M.: Mi rostru ture evolution model in mi romagneti s. Zeit. für
angew. Math. und Physik, 55 (2004), 159-182.
[53℄ Roubí£ek, T., Kruºík, M.: Mesos opi model for ferromagnets with isotropi hardening. Zeit.
für angew. Math. und Physik 56 (2005), 107-135.
[54℄ Roubí£ek, T., Kruºík, M.: Mesos opi model of mi rostru ture evolution in shape mem-
ory alloys, its numeri al analysis and omputer implementation. 3rd GAMM Seminar on
mi rostru tures. (Ed.C.Miehe), GAMM Mitteilungen 2006 29, J.Wiley, in print.
[55℄ Rybka, P., Luskin, M.: Existen e od energy minimizers for magnetostri tive materials. SIAM
J. Math. Anal. 36 (2005), 2004-2019.
[56℄ Shu, Y., Bhatta harya, K.: Domain patterns and ma ros opi behaviour of ferroele tri
materials. Phil. Mag. B 81 (2001), 20212054.
[57℄ Ti kle, R.: Ferromagneti shape memory materials. PhD-thesis, Univ. of Minnesota, Min-
neapolis, 2000.
[58℄ Visintin, A., Modied Landau-Lifshitz equation for ferromagnetism. Physi a B 233 (1997),
365369.
[59℄ Visintin, A., Maxwell's equations with ve tor hysteresis. Ar h. Rat. Me h. Anal. 175 (2005),
137.
[60℄ Vivet, A., Lex ellent, C.: Mi rome hani al modelling for tension- ompression pseudoelasti
behaviour of AuCd single rystals. Euro Phys.J. A.P.4(1998), 125-132.

42

View publication stats