Numerical Analysis of a Quasistatic Problem of Sliding Frictional Contact with Wear

Jiuhua Chen1 , Weimin Han1 and Mircea Sofonea2

We consider numerical approximations of a quasistatic problem modeling the sliding frictional contact with wear between a viscoelastic body and a rigid moving foundation. The contact is modeled with the Coulomb's law of dry friction and the wear is described by a version of Archard's law. The variational formulation of the problem consists of a nonlinear evolutionary equation coupled with a time-dependent variational inequality with nonlinear di erential operators, which has a unique solution under certain assumptions on the given data. We derive error estimates for both spatially semi-discrete and fully discrete schemes to solve the problem. Under appropriate regularity assumptions on the exact solution, we establish optimal order error estimates.

Abstract.

1 Introduction
We consider a mathematical model for the process of bilateral frictional contact of a viscoelastic body with a rigid moving foundation, such that there is no lose of contact between the body and the foundation. The external time dependent volume forces and tractions are assumed to vary slowly; as a result the mechanical states evolve quasistatically. We assume a sliding frictional contact which involves wear of the contacting surface. The friction is modeled with Coulomb's law and the wear is modeled by a version of Archard's law. Situations of frictional contact between deformable bodies can be frequently found in industry and everyday life such as train wheels with the rails, a shoe with the oor, tectonic plates, the car's braking system, etc. For this reason, considerable progress has been made with the modeling and analysis of contact problems. An early attempt to study frictional contact problems within the framework of variational inequalities was made in 4]. An excellent reference on analysis and numerical approximations of contact problems involving elastic materials with or without friction is 12]. The mathematical, mechanical and numerical state of the art can be found in the proceedings 13]. Wear is one of the plagues which reduce the lifetime of modern machine elements. It arises when a hard rough surface slides against a softer surface, digs into it, and its asperities plough a series of grooves. When two surfaces come into contact, rearrangement of the surface asperities takes place. When they are in relative motion some of the peaks will break and therefore the harder surface removes the softer material. This phenomenon involves the wear of the contacting
1 2

Department of Mathematics, University of Iowa, Iowa City, IA 52242, USA Laboratoire de Theorie des Systemes, University of Perpignan, France

surfaces. Evolution of the wear leads to a new, interesting and nonstandard boundary condition which relates the normal stress to the time rate of displacement. General models of frictional contact with wear were derived from thermodynamic considerations in 17, 18] where consistency of such models is discussed. A dynamic thermoelastic contact problem with normal compliance and surface wear has been analysed in 2]. Recently, variational analysis including existence of weak solutions in the study of viscoelastic frictional contact problems with wear has been provided in 14, 15, 16]. The present paper represents a continuation of 16]. Its aim is to provide numerical analysis of the quasistatic problem of sliding frictional contact with wear studied in 16]. In a variational formulation, the problem consists of a nonlinear evolution equation coupled with a time-dependent variational inequality with nonlinear di erential operators. The literature is abundant on numerical treatment of variational inequality, see for instance the monographs 5, 6, 10, 12]. Of particular relevance to this paper are the works on numerical analysis of variational inequalities arising in plasticity, cf. 7, 8, 9]. The paper is organized as follows. In Section 2 we present the mechanical problem together with its variational formulation. We then list the assumptions on the data and recall the existence and uniqueness result obtained in 16], which states that under a smallness assumption on the given data, the mechanical problem has a unique weak solution. In Sections 3 and 4 we analyze spatially semi-discrete and fully discrete schemes, respectively. We use the nite element method to discretize the spatial domain and a backward Euler scheme to obtain the fully discrete problems. We also derive error estimates for both spatially semi-discrete and fully discrete schemes. Finally, under appropriate regularity assumptions on the exact solution, we obtain optimal order error estimates.

2 The problem of sliding frictional contact with wear

In this section we describe a model for the contact problem with wear, present its variational formulation, list the assumptions imposed on the problem data and recall the existence and uniqueness result obtained in 16]. The physical setting is as follows. We consider a viscoelastic body whose material particles occupy a bounded domain of IRd (d = 2; 3 in applications). For the domain , we assume that its boundary ? is Lipschitz continuous, and is partitioned into three disjoint measurable parts ?1, ?2 and ?3, with meas (?1 ) > 0. Displacement and surface traction conditions will be speci ed on ?1 and ?2, respectively. On ?3, the body is in frictional bilateral contact with a moving rigid foundation, which results in the wear of the contacting surface. We assume that there is only sliding contact which is always maintained. Let 0; T ] be the time interval of interest. As usual, we will use the notation u = (ui) : 2 0; T ] ! IRd for the displacement eld and

= ( ij ) : 0; T ] ! Sd for the stress eld. Here and throughout this paper, the indices i and j run between 1 and d, we adopt the summation convention over repeated indices, unless stated otherwise, and the index that follows a comma indicates a partial derivative with respect to the corresponding component of the independent variable. We use Sd to represent the space of second order symmetric tensors on IRd, or equivalently, the space of symmetric matrices of order d. We de ne the inner products and the corresponding norms on IRd and Sd by
u v = u i vi ;

ij

jvj = (v v)1=2 ; 8 u; v 2 IRd; )1=2 ; 8 ; 2 Sd: ij ; j j = (

Since the boundary ? is Lipschitz continuous, the unit outward normal vector on the boundary is de ned a.e. For every vector eld v, we use the notation v to denote the trace of v on ? and we denote by v and v the normal and the tangential components of v on the boundary given by v =v ; v =v?v : We also use the notation "(v) for the tensor eld de ned by 1 "(v ) = ("ij (v )); "ij (v ) = (vi;j + vj;i ): 2 For a stress eld , the application of its trace on the boundary to is the Cauchy stress vector . We de ne, similarly, the normal and tangential components of the stress on the boundary by the formulae =( ) ; = ? : Finally, in the sequel div will denote the divergence operator for tensor elds, i.e. div = ( (2.1)
ij;j ):

The material is assumed to be viscoelastic, its constitutive relation being _ )) + G("(u)): = A("(u In (2.1) and below, to simplify the notation, we do not explicitly indicate the dependence of various functions on x 2 ? and t 2 0; T ]. Moreover, throughout the paper, a dot above a variable represents its time derivative. We assume that the viscoelastic body is clamped on ?1, and therefore the displacement eld vanish there. We also assume that a body force of density f act in , and surface traction of density g are imposed on ?2 . The densities f and g may depend on the time variable and are assumed to vary slowly in time; the accelerations are therefore neglected in the equations of motion, leading to a quasistatic approach of the process. 3

We introduce the wear function w : ?3 0; T ] ! IR+ representing the accumulated wear of the surface. The evolution of the wear of the contacting surface is governed by a version of Archard's law of wear (cf. 17, 18]), namely _ ? v j: w _ = ?kw ju _ ? v is the relative Here kw > 0 is a wear coe cient, v is the velocity of the foundation, and u velocity, called the slip velocity, between the contact surface and the foundation. We see that the rate of wear is assumed to be proportional to the contact stress and the slip velocity. We also assume that on the time scale of the quasistatic process, the surface rearranges itself in such a way that the slip velocity is just v which, for the sake of simplicity, is assumed to be a positive constant. In other words, we will use the following version of Archard's law (2.2) w _ = ?kw v where v > 0: Let us now represent the e ect of the wear as the recession of ?3. Since the body is in bilateral contact with the foundation, we impose the condition (2.3) u = ?w on ?3: Thus the position of the contact evolves with the wear. We can now eliminate the unknown function w from our problem. Let = kw v and = 1= . Using (2.2) and (2.3) we have (2.4) = u _ : We model the frictional contact between the viscoelastic body and the foundation with Coulomb's law of dry friction. Since there is only sliding contact it follows that _ ; j j = j j; =? u 0 where > 0 is the coe cient of friction. Moreover, the wear increase in time, i.e. w _ 0 and therefore, from (2.2) it results that 0. Thus, the friction condition on ?3 implies _ ; (2.5) j j=? ; =? u 0: Using now (2.1) as constitutive law and (2.4), (2.5) as contact conditions, the classical formulation of the mechanical problem of sliding frictional contact with wear is the following: nd a displacement u : 0; T ] ! IRd and a stress tensor : 0; T ] ! Sd such that
8 > _ )) + G("(u)) in = A("(u (0; T ); > > > > > div + f = in (0; T ); > > > < u= on ?1 (0; T ); > = g on ?2 (0; T ); > > > > > _ ; = u _ ; j j=? ; =? u > > > : u(0) = u in :

(2.6)

0 on ?3 (0; T );

Here u0 represent the given initial displacement. To present the variational formulation of this problem we need to introduce some functional notation. For the displacement variable, we use the space

V = fv = (vi) 2 (H 1( ))d : v = 0 on ?1g

with the canonical inner product de ned by
Z

hu; viV = uivi dx + "ij (u)"ij (v) dx

and the associate norm For stress and strain elds, we use the space with the canonical inner product de ned by

kvkV = hv; viV :

ij

Q = f 2 (L2 ( ))d d :

= ji; 1 i; j

dg

h ; i=
and the corresponding norm de ned by

ij ij dx;

k k2 Q = h ; i:
We will also need the space

H (div ; ) = f 2 Q : div 2 (L2 ( ))dg:

Since meas (?1) > 0, Korn's inequality holds ( 11]): (2.7)

kvkV c k"(v)kQ 8 v 2 V;

where c is a positive constant depending on and ?1. Everywhere in this paper, the symbol c will represent a positive constant which may change its value from place to place, and may depend on the input data, but independent of discretization parameters h and k to be introduced later. From (2.7) it follows that v 7! k"(v)kQ is a norm over the space V , equivalent to the norm v 7! kv kV . For any normed space X , C ( 0; T ]; X ) denotes the space of continuous functions from 0; T ] to X , with the norm kf kC ( 0;T ];X ) = tmax kf (t)kX : 2 0;T ] 5

Similarly, C 1( 0; T ]; X ) denotes the space of continuously di erentiable functions from 0; T ] to X , with the norm kf kC 1( 0;T ];X ) = tmax kf (t)kX + tmax kf_(t)kX : 2 0;T ] 2 0;T ] When X is a Banach space, both C ( 0; T ]; X ) and C 1 ( 0; T ]; X ) are Banach spaces. In the study of the mechanical problem (2.6) we make the following assumptions. 8 (a) A : S d ! Sd : > > > > > (b) There exists LA > 0 such that > > > > > jA(x; "1 ) ? A(x; "2 )j LAj"1 ? "2 j 8 "1 ; "1 2 Sd ; a:e: x 2 : > < (2.8) > (c) There exists m > 0 such that > > > (A(x; "1 ) ? A(x; "2)) ("1 ? "2 ) m j"1 ? "2j2 8 "1 ; "1 2 Sd ; a:e: x 2 : > > > > > (d) For any " 2 Sd; x 7! A(x; ") is Lebesgue measurable on : > > : (e) The mapping x 7! A(x; 0) 2 Q: 8 (a) G : S d ! Sd : > > > > > (b) There exists an LG > 0 such that > < (2.9) > jG(x; "1) ? G(x; "2)j LG j"1 ? "2j 8 "1; "2 2 Sd; a:e: x 2 ; : > > > (c) For any " 2 Sd; x 7! G(x; ") is measurable: > > : (d) The mapping x 7! G(x; 0) 2 Q: 8 f 2 C ( 0; T ]; (L2( ))d ); > > > > 2 d > > < g 2 C ( 0; T ]; (L (?2 )) ); (2.10) > 2 L1(?3); (x) 0 a:e: on ?3; > > > 2 L1(?3); (x) > 0 a:e: on ?3 ; > > : u0 2 V: We de ne the functionals

L(t; v) = j (u; v) =

Z Z

f (t) v dx +

Z
?2

g (t) v ds;

for all u; v 2 V and t 2 0; T ], where ds denotes the surface element. The variational formulation of the mechanical problem (2.6) can be stated as follows. : 0; T ] ! Sd, such that for t 2 0; T ], _ (t))) + G("(u(t))) in ; (2.11) (t) = A("(u _ (t))i + j (u _ (t); v) ? j (u _ (t); u _ (t)) L(t; v ? u _ (t)) 8 v 2 V; (2.12) h (t); "(v ? u and the initial condition (2.13) u(0) = u0 : 6
Problem P. Find a displacement u :

?3

ju j ( jv j + v ) ds

0; T ] ! IRd and a stress eld

Well-posedness of this problem has been studied in 16] where the following result is proved.
Theorem 2.1. Let the assumptions (2:8){(2:10) hold. Then there is a constant

depends only on , ?1 , ?3 and A, such that if

3

> 0 which

k kL1(? ) (k kL1(? ) + 1) < 0; then the problem P has a unique solution u 2 C 1( 0; T ]; V ), 2 C ( 0; T ]; H (div ; )).
3

The proof of Theorem 2.1 is carried out in several steps. It is based on classical results for elliptic variational inequalities followed by xed point arguments. In the rest of the paper, we assume the conditions stated in Theorem 2.1 are satis ed so that the contact problem P has a unique solution.

3 Spatially semi-discrete approximation

In this section we consider an approximation of the problem P by discretizing only the spatial domain. Let V h V and Qh Q be nite-dimensional spaces which for example, can be constructed by the nite element method. We assume that these spaces satisfy
"(V h )

Qh :

This assumption is very natural and is valid when the polynomial degree for the space V h is at most one higher than that for the space Qh. Let PQh : Q ! Qh be the orthogonal projection de ned through the relation (3.1)

hPQh q; qhi = hq; qhi 8 q 2 Q;

q h 2 Qh :

Obviously, we have (3.2) kPQh qkQ kqkQ 8 q 2 Q: This property will be used on various occasions. We now discuss a spatially semi-discrete scheme.
such that (3.3) and for t 2 0; T ],
Problem Ph. Find the displacement eld uh : 0; T ] ! V h and the stress eld h : 0; T ] ! Qh
uh(0) = uh 0;

_ h(t))) + PQh G("(uh(t))) in ; (3.4) h(t) = PQh A("(u _ h(t))i + j (u _ h(t); vh) ? j (u _ h(t); u _ h(t)) L(t; vh ? u _ h(t)) 8 vh 2 V h: (3.5) h h(t); "(vh ? u 7

h Here, uh 0 2 V is an appropriate approximation of u0 .

Using the arguments in 16], it can be shown that under the conditions stated in Theorem 2.1, the problem Ph has a unique solution uh 2 C 1( 0; T ]; V h), h 2 C ( 0; T ]; Qh). Our main purpose here is to derive estimates for the errors u ? uh and ? h. To this end, let t 2 0; T ]. From (2.11) and (3.4), we have _ (t))) ? A("(u _ h(t))) + G("(u(t))) ? G("(uh(t)))]; (t) ? h(t) = (IQ ? PQh ) (t) + PQh A("(u where IQ : Q ! Q is the identity operator. Using the conditions (2.8), (2.9) and the property (3.2), we obtain _ (t) ? u _ h(t))kQ + k"(u(t) ? uh(t))kQ): (3.6) k (t) ? h(t)kQ k(IQ ? PQh ) (t)kQ + c (k"(u Since Zt _ (r) ? u _ h(r)) dr; "(u "(u(t) ? uh (t)) = "(u0 ? uh ) + 0 0 we have Zt _ (r) ? u _ h(r))kQ dr: (3.7) k"(u k"(u(t) ? uh(t))kQ k"(u0 ? uh ) k + 0 Q 0 Thus from (3.6), _ (t) ? u _ h(t))kQ + k"(u0 ? uh k (t) ? h(t)kQ k(IQ ? PQh ) (t)kQ + c (k"(u 0 )kQ ) Zt _ (r) ? u _ h(r))kQ dr: + c k"(u Recalling Korn's inequality (2.7), we conclude _ ?u _ hkC ( 0;T ];V ) + ku0 ? uh (3.8) k ? hkC ( 0;T ];Q) k(IQ ? PQh ) kC ( 0;T ];Q) + c (ku 0 kV ): _ ?u _ hkC ( 0;T ];V ). We take v = u _ h(t) in (2.12) and use (2.11) to obtain It remains to estimate ku _ (t))) + G("(u(t))); "(u _ h(t) ? u _ (t))i + j (u _ (t); u _ h(t)) ? j (u _ (t); u _ (t)) L(t; u _ h(t) ? u _ (t)): hA("(u Using (3.4) and (3.5), we get _ h(t))) + G("(uh(t))); "(vh ? u _ h(t))i + j (u _ h(t); vh) ? j (u _ h(t); u _ h(t)) L(t; vh ? u _ h(t)): hA("(u Adding these two inequalities and performing some elementary manipulations, we obtain _ (t))) ? A("(u _ h(t))); "(u _ (t) ? u _ h(t))i hA("(u _ (t))) ? A("(u _ h(t))); "(u _ (t) ? v h)i hA("(u _ (t) ? vh)i + hG("(u(t))) ? G("(uh(t))); "(u _ (t) ? u _ h(t))i ? hG("(u(t))) ? G("(uh(t))); "(u _ (t); vh) + D(u _ (t); u _ h(t); vh); + R(t; u 8
0

where _ (t); vh) = h (t); "(vh ? u _ (t))i + j (u _ (t); vh) ? j (u _ (t); u _ (t)) ? L(t; vh ? u _ (t)); (3.9) R(t; u h h h h h h h h _ (t); u _ (t); v ) = j (u _ (t); u _ (t)) ? j (u _ (t); v ) + j (u _ (t); v ) ? j (u _ (t); u _ (t)): (3.10) D(u Using the assumptions (2.8) and (2.9) on the functions A and G, we then have _ (t) ? u _ h(t))k2 k"(u Q or equivalently, _ (t) ? u _ h(t))kQ k"(u _ (t) ? vh)kQ + k"(u(t) ? uh(t))kQ) c (k"(u _ (t); vh)j1=2 + jD(u _ (t); u _ h(t); vh)j1=2): + c (jR(t; u _ (t) ? vhk ku
Z

h 2 _ (t) ? vh)k2 c (k"(u Q + k"(u(t) ? u (t))kQ) _ (t); vh)j + jD(u _ (t); u _ h(t); vh)j); + c (jR(t; u

Recalling again Korn's inequality (2.7) and the inequality (3.7), we obtain (3.11) _ (t) ? u _ h(t)k ku
V t h _ (r) ? u _ h(r)kV c V + ku0 ? u0 kV + ku 0 h 1 = 2 _ (t); v )j + jD(u _ (t); u _ h(t); vh)j1=2 ): + c (jR(t; u

dr

From the trace theorem V ,! L2 (?) (cf. 1]), we have (3.12)

kvkL (? ) kvkL (?) c kvkV 8 v 2 V:

2 3 2

Now we are ready to estimate the term _ (t); u _ h(t); vh) = D (u Since _ (t); u _ h(t); vh)j jD(u _ h(t) ? v hkL (? ) + ku _ h(t) ? vhkL (? )) k kL1(? ) ku _ (t) ? u _ h(t)kL (? ) (k kL1(? ) ku _ (t) ? u _ h(t)kL (? ) ku _ h(t) ? vhkL (? ) k kL1(? ) (k kL1(? ) + 1) ku _ (t) ? u _ h(t)kL (? ) k kL1(? ) (k kL1(? ) + 1) ku _ h(t) ? u _ (t)kL (? ) + ku _ (t) ? vhkL (? )); (ku
3 2 3 3 2 3 2 3 3 3 2 3 2 3 3 3 2 3 2 3 2 3

?3

_ h(t)j ? jvhj) + u _ h(t) ? vh) ds: (ju _ (t)j ? ju _ h(t)j) ( (ju

using the trace theorem (3.12), we have _ (t); u _ h(t); vh)j jD(u _ (t) ? u _ h(t)kV (ku _ h(t) ? u _ (t)kV + ku _ (t) ? vhkV ): c k kL1(? ) (k kL1(? ) + 1) ku
3 3

Therefore, it follows that (3.13) _ (t); u _ h(t); vh)j jD(u _ (t) ? u _ h(t)k2 _ (t) ? v hk2 c k kL1(? ) (k kL1(? ) + 1) ku V + c ku V:
3 3

Using (3.13) in (3.11), we obtain

=2 1=2 _ (t) ? u _ (t) ? u _ h(t)kV c k k1 _ h(t)kV ku L1(? ) (k kL1(? ) + 1) ku Zt h h _ (t) ? v kV + ku0 ? u0 kV + ku _ (r) ? u _ h(r)kV dr + jR(t; u _ (t); vh)j1=2 : + c ku 0
3 3

Then, if k kL1(?3 )(k kL1(?3 ) + 1) is su ciently small, we have (3.14)

_ (t) ? u _ h(t)kV ku Zt _ (t) ? vhkV + ku0 ? uh _ (r) ? u _ h(r)kV dr + jR(t; u _ (t); vh)j1=2 : c ku k + k u 0 V 0 _ (t) ? u _ h(t)kV max ku
h

Notice that v h is arbitrary in the above inequality. Using the Gronwall's inequality, we obtain (3.15) _ (t) ? vhkV + jR(t; u _ (t); vh)j1=2 : c ku0 ? uh inf ku 0 kV + c tmax 2 0;T ] vh 2V h The following estimate for ? (3.16)
t2 0;T ] t2 0;T ]

follows directly from (3.8) and (3.15).

Q

max k (t) ?

h (t)k

c ku0 ? uh 0 kV + k(IQ ? PQh ) kC ( 0;T ];Q)

_ (t) ? vhkV + jR(t; u _ (t); vh)j1=2 : + c tmax inf ku 2 0;T ] vh 2V h

Summarizing, we have shown the following result.

Assume the conditions stated in Theorem 2.1. Then if k kL1(?3 ) (k kL1(?3 ) + 1) is su ciently small, the estimates (3:15) and (3:16) hold.
Theorem 3.1. Let (u; ) and (uh ; h ) be the solutions of the problems P and Ph respectively.

In order to give more concrete results based on (3.15) and (3.16), we now brie y specify the nite dimensional spaces V h and Qh via the nite element method. Details can be found in 3]. For simplicity, we assume that is polygonal. Let T h be a regular nite element partition of the domain . Here we use linear elements for the space V h and piecewise constants for Qh . To perform our convergence analysis, we need the following density result from 19].
Lemma 3.2. Assume that X is a Banach space, X0

C ( 0; T ]; X0) is dense in C ( 0; T ]; X ).

X is a dense subspace of X . Then

10

Now we are ready to give convergence and error analysis of the spatially discrete solution for the problem P.
Theorem 3.3. Keep all the assumptions in Theorem 3.1. Assume the initial approximation
h uh 0 2 V is chosen such that

(3.17)
Then (3.18) Moreover, if

ku0 ? uh 0 kV ! 0; as h ! 0:
h (t)k + k (t) ? _ _ max ( k u ( t ) ? u V t2 0;T ] h (t)k V) ! 0

as h ! 0:

(3.19)

u 2 C 1 ( 0; T ]; V

\ (H 2( ))d \ (H 2(?3 ))d);

c h;

2 C ( 0; T ]; (H 1( ))d d );

and (3.20) ku0 ? uh 0 kV then we have the optimal order error estimate

(3.21)

t2 0;T ]

_ (t) ? u _ h(t)kV + k (t) ? max (ku

h (t)k

V)

c h:

Proof. Since V \ (H 2 ( ))d and (H 1 ( ))d d \ Q are dense in V and Q respectively, then by Lemma 3.2, C ( 0; T ]; V \ (H 2 ( ))d ) and C ( 0; T ]; (H 1( ))d d \ Q) are dense in C ( 0; T ]; V )

~ 2 C ( 0; T ]; V \ (H 2( ))d ) and and C ( 0; T ]; Q) respectively. Therefore, 8 " > 0, there exist u ~ 2 C ( 0; T ]; (H 1( ))d d \ Q) such that

(3.22)

_ ?u ~ kC ( 0;T ];V ) "; k ? ~ kC ( 0;T ];Q) ": ku Let hv 2 V h be the piecewise linear interpolant of v 2 V \ (H 2( ))d . Then (cf. 3]) (3.23) kv ? hv kV c h kvk(H ( ))d :
2

_ (t); vh)j in (3.15) de ned by (3.9). Recalling the assumption Now we estimate the term jR(t; u (2.10) and the trace theorem (3.12), we can show that _ (t); vh)j c ku _ (t) ? vhkV : jR(t; u By (3.22) and (3.23), we obtain from (3.15) that
h (t)k _ _ max k u ( t ) ? u V t2 0;T ] =2 _ c ku0 ? uh inf k u ( t ) ? v h k1 0 kV + c tmax V h h 2 0;T ] v 2V =2 1=2 + c h1=2 ku ~ k1 c ku0 ? uh 0 kV + c " C ( 0;T ];(H 2 ( =2 1=2 + c max ku ~ (t) ? hu ~ (t)k1 c ku0 ? uh V 0 kV + c " t2 0;T ]
))d )

11

_ h to u _. Using the assumption (3.17), we conclude from the above estimate the convergence of u Observe that the error bound of (3.16) di ers from that of (3.15) only in one extra term k(IQ ? PQh ) kC ( 0;T ];Q). Thus the convergence of h to follows from the estimate

k(IQ ? PQh ) kC ( 0;T ];Q)

k ? ~ kC ( 0;T ];Q) + tmax inf k ~ ? qhkQ 2 0;T ] qh 2Qh " + c h k ~ kC ( 0;T ];(H ( ))d d ) ;
1

by using (3.2), (3.22) and nite element interpolation error estimates ( 3]). We now prove the error estimate (3.21) under the solution regularity condition (3.19) and the assumption (3.20). We re-estimate the term _ (t); vh) = R(t; u Integrate by parts, _ (t); vh) = R(t; u
Z Z

_ (t)) ? f (t) (vh ? u _ (t)) dx ? (t) "(v h ? u

?3

_ (t)j) + vh ? u ju _ (t)j ( (jvhj ? ju _ (t)) ds:

Z

?2

_ (t)) ds g (t) (v h ? u

_ (t))) dx (?div (t) ? f (t)) (vh ? u

Z Z?
?3

+ + _ (t); vh) = R(t; u Therefore,

Z

(t)

_ (t)) ds ? (vh ? u

ju _ (t)j

_ (t)j) + vh ? u (jvhj ? ju _ (t) ds:

?2

_ (t)) ds g (t) (v h ? u

Using the equilibrium equation and the boundary conditions on ?1 ?2, we have
?3

_ (t)) + ju _ (t)j) + vh ? u (t) (vh ? u _ (t)j ( (jvhj ? ju _ (t)) ds:

_ (t); vh)j (k (t) k(L (? ))d + k kL1(? )(k kL1(? ) + 1)ku _ (t) ? v hkL (? ); jR(t; u _ (t)kL (? ) )ku
2 3 3 3 2 3 2 3

which with the aid of the result (3.12) yields _ (t); vh)j jR(t; u k kC ( 0;T ];(H (
1

))d d )

_ kC ( 0;T ];V ) ku _ (t) ? vhkL2(?3 ): + k kL1(?3 ) (k kL1(?3 ) + 1)ku

Hence from the estimate (3.15), it follows that (3.24)

t2 0;T ]

_ (t) ? u _ h(t)kV max ku

=2 _ _ k u ( t ) ? v hkV + ku ( t ) ? v h k1 c ku0 ? uh inf 2 (?3 ) : 0 kV + c tmax L h h 2 0;T ] v 2V

12

With the regularity condition (3.19), using the interpolation results, we have ku(t) ? hu(t)kV c h ku(t)k(H 2 ( ))d ; ku(t) ? hu(t)kL2(?3 ) c h2ku(t)k(H 2 (?3 ))d ; k(IQ ? PQh ) (t)kQ c h k (t)k(H 1 ( ))d d for all t 2 0; T ]. Then the error estimate (3.21) follows from (3.16), (3.20) and (3.24).

4 Fully discrete approximation

Now we consider a fully discrete approximation of the problem P . In addition to the nite dimensional spaces V h and Qh introduced in the last subsection, we need a partition of the time interval 0; T ] : 0 = t0 < t1 < < tN = T . We denote the step-size kn = tn ? tn?1 for n = 1; : : : ; N . We allow non-uniform partition of the time interval, and denote k = maxn kn the maximal step-size. For a continuous function w(t), we use the notation wn = w(tn). For a sequence fwngN n=0 , we denote wn = wn ? wn?1 for the di erence, and wn = wn =kn the corresponding divided di erence. In this section, no summation is implied over the repeated index n. Then a fully discrete approximation method based on a backward Euler scheme is the following.
N Problem Phk . Find the displacement eld uhk = fuhk n gn=0 h (4.1) uhk 0 = u0 ; and for n = 1; : : : ; N , hk = P h A("( uhk )) + P h G("(uhk )) in ; (4.2) Q Q n?1 n n h h h hk hk h hk h hk hk (4.3) h n ; "(v ? un )i + j ( un ; v ) ? j ( uhk n ; un ) L(tn ; v ? un ) 8 v 2 V :

hk gN n n=0

Qh ,

such that

V h and the stress eld

hk

Using the argument technique of 16], it can be shown that the fully discrete problem Phk has a unique solution if k kL1(?3 )(k kL1(?3 ) + 1) is su ciently small. We now derive some error estimates for the numerical solution. We will use the notations _n = u _ (tn) and n = (tn ). We use (2.11) and (2.12) at t = tn and take v = uhk un = u(tn ), u n to obtain _ n)) + G("(un)); "( uhk _ n)i + j (u _ n; uhk _ n; u _ n) L(tn; uhk _ n): (4.4) hA("(u n ?u n ) ? j (u n ?u Substituting (4.2) into (4.3), we get hk h hk hk h hk hk (4.5) hA("( uhk n )) + G("(un?1 )); "(v ? un )i + j ( un ; v ) ? j ( un ; un ) L(t; v h ? uhk n ): 13

Adding (4.4) and (4.5) with rearrangement of the terms, we obtain _ n)) ? A("( uhk _ n ? uhk hA("(u n )); "(u n )i hk _ n ? uhk ?hG("(un)) ? G("(un?1 )); "(u n )i hk hk _ n)) + G("(un?1)) ? G("(un)); "(vh ? u _ n)i + hA("( un )) ? A("(u h hk h _ n ; v ) + D (u _ n; un ; v ); + R(tn; u h _ n; vh) and D(u _ n; uhk where the quantities R(tn; u n ; v ) are de ned in (3.9) and (3.10). Using the conditions (2.8) and (2.9), we obtain 2 _ n ? uhk k"(u n )kQ _ n ? uhk c k"(un ? uhk n?1 )kQk"(u n )kQ hk _ n ? uhk _ n ? vh)kQ + c (k"(u n )kQ + k"(un ? un?1 )kQ) k"(u h _ n; vh)j + jD(u _ n; uhk + c (jR(tn; u n ; v )j): Then we have _ n ? uhk k"(u n )kQ h 1=2 : _ n ? v h)kQ + jR(tn; u _ n; vh)j1=2 + jD(u _ n; uhk c k"(un ? uhk n?1 )kQ + k"(u n ; v )j Korn's inequality (2.7) can be applied here to yield _ n ? uhk (4.6) ku n kV h 1=2 : _ n ? vhkV + jR(tn; u _ n; vh)j1=2 + jD(u _ n; uhk c kun ? uhk n ; v )j n?1 kV + ku From the estimate (3.13) and the result (3.12), h 1=2 _ n; uhk jD(u n ; v )j =2 1=2 _ ? uhk k + c ku _ n ? v hkV : c k k1 n n V L1(?3 ) (k kL1(?3 ) + 1) ku Using this estimate in (4.6), we obtain =2 1=2 _ ? uhk k _ n ? uhk ku c k k1 n n kV n V L1(?3 ) (k kL1(?3 ) + 1) ku _ n ? vhkV + jR(tn; u _ n; vh)j1=2 : ku + c kun ? uhk inf n?1 kV + c v h 2V h If k kL1(?3 )(k kL1(?3 ) + 1) is su ciently small, we have (4.7) Write _n ? ku
uhk n kV

_ n ? vhkV + jR(tn; u _ n; vh)j1=2 : ku c kun ? uhk inf n?1 kV + c vh 2V h

n ?1 X j =1

h un ? uhk n?1 = u0 ? u0 + un ? un?1 +

_ j) + kj ( uj ? u

n ?1 X j =1

_ j ? uhk kj (u j ):

14

Then (4.8) Hence from (4.7), we get (4.9) _n? ku

h kun ? uhk n?1 kV ku0 ? u0 kV + kun ? un?1 kV

n ?1 X j =1

_ j ? uj kV + kj ku

n ?1 X j =1

_ j ? uhk kj ku j kV :
n ?1 X j =1

uhk n kV

c ku0 ? uh 0 kV + c kun ? un?1 kV + c

_ n ? vhkV ku _ n; vh)j1=2 + jR(tn; u +c

_ j ? uj kV kj ku

+ c vh inf 2V h

n ?1 X j =1

_ j ? uhk kj ku j kV :

To proceed further, we need the following result.

N Lemma 4.1. Assume fgn gN n=1 and fen gn=1 are two sequences of non-negative numbers, satisfying

en c g n + c
Then (4.10) Therefore, (4.11)
Proof. Denote

n ?1 X j =1

kj ej ; n = 1; : : : ; N:

en c gn +
1

n ?1 X j =1

kj gj ; n = 1; : : : ; N: c 1max g: n N n
n X j =1

max e n N n

En =

kj ej :

Then (4.12) Now

en c gn + c En?1; n = 1; : : : ; N: En ? En?1 = knen c kngn + c knEn?1;

which implies (4.13) En ? (1 + c kn) En?1 c kngn n = 1; : : : ; N: We introduce a sequence of numbers fzngN n=0 by setting z0 = 1 and

zn =
Using the inequalities

n Y

j =1

(1 + c kj ); 1 n N:

1 1 + c kj

ec kj ; j = 1; : : : ; N;
15

we have the following bounds: (4.14) 1

N Y i=j +1

(1 + cki) ec(T ?tj ) ; j = 1; : : : ; N:

Now with the above notations, the inequality (4.13) can be rewritten as En ? En?1 c kngn : zn zn?1 zn A simple induction argument gives us n g n n X Y X j (1 + cki)gj ; En c zn kj z = c kj j j =1 j =1 i=j +1 which can be combined with (4.12) and (4.14) to yield (4.10). The inequality (4.11) follows easily from (4.10). 2 Applying Lemma 4.1 to the inequality (4.9), we obtain the following estimate, (4.15)
1

_ ? max ku n N n

uhk n kV

ku0 ? uh 0 kV

N ?1 X n=1

_ nkV knk un ? u

_ n ? vhkV + jR(tn; u _ n; vh)j1=2 : + c 1max ku ? un?1kV + c 1max inf ku n N n n N vh 2V h Now we turn to bound
hk n? n n ? hk n.

Subtracting (4.2) from (2.11) at t = tn, we have

hk _ n)) ? A("( uhk A("(u n )) + G("(un )) ? G("(un?1 ))]; hk uhk n kV + ckun ? un?1 kV :

= (IQ ? PQh )
hk k n Q

n + PQh

which, by the assumptions (2.8), (2.9) and the property (3.2), yields

k n?
(4.16) 1max k n N
n?

_n ? k(IQ ? PQh ) nkQ + cku

Recalling (4.8), we then get the estimate

hk k n V

c ku0 ? uh 0 kV +

N ?1 X n=1

_ nkV + k(IQ ? PQh ) kC ( 0;T ];Q) knk un ? u

_ n ? vhkV + jR(tn; u _ n; vh)j1=2 : + c 1max ku ? un?1kV + c 1max inf ku n N n n N vh 2V h Summarizing, we have shown the following result.
hk N Theorem 4.2. Let (u; ) and f(uhk n ; n )gn=1 be the solutions of the problems

respectively. Assume the conditions stated in Theorem 2.1. Then if k kL1 (?3 )(k kL1(?3 ) + 1) is su ciently small, the estimates (4:15) and (4:16) hold.

P and Phk

16

The following result (cf. 19]) is needed for convergence analysis.

Lemma 4.3. Assume that X is a Banach space. Then C 1 ( 0; T ]; X ) is dense in C 1 ( 0; T ]; X ). Theorem 4.4. Keep all the assumptions in Theorem 4.2. Assume the initial approximation
h uh 0 2 V is chosen such that

(4.17)
Then (4.18) Moreover, if (4.19)
(
1

ku0 ? uh 0 kV ! 0; as h ! 0:
_ n ? uhk max (ku n kV + k n N
hk n ? n kV ) ! 0

as h; k ! 0:

u 2 C 1 ( 0; T ]; V

and (4.20) ku0 ? uh 0 kV then we have the optimal order error estimate

\ (H 2( ))d \ (H 2(?3 ))d) \ W 2;1(0; T ; V ); 2 C ( 0; T ]; (H 1( ))d d );

c h; c (h + k):

(4.21)

_ n ? uhk max (ku n kV + k n N

hk n ? n kV )

Proof. First we show the convergence result (4.18). Most terms on the right-hand sides of the

estimates (4.15) and (4.16) can be handled in much the same way as those similar terms on the right-hand side of the estimate (3.16), as was done in the proof of Theorem 3.3. Here it is enough for us to estimate the two terms
N ?1 X n=1

_ nkV and knk un ? u

max ku ? un?1kV : n N n

By Lemma 4.3, C 1( 0; T ]; V ) is dense in C 1( 0; T ]; V ). Hence 8 " > 0, there exists w 2 C 1( 0; T ]; V ) such that ku ? wkC 1( 0;T ];V ) ": Since

_ nkV un ? u

"+
we have
N ?1 X n=1

_n?w _ n)kV + k k (un ? w n ) ? (u Z t n 1 _ (t) ? w _ n)dtkV " + kk (w t

Z tn
n?1

_ nkV wn ? w

tn?1

kw(t)kV dt; n = 1; : : : ; N ? 1;
T" + k kwkL1(0;T ;V );
17

_ nkV knk un ? u

Thus we have the converngence (4.18).

N ?1 X n=1

where w(t) denotes the second-order derivative of w(t) with respect to the time variable t. It is easily seen that _ kC ( 0;T ];V ): max kun ? un?1kV k ku 1 n N

Under the regularity condition (4.19), we have _ nkV knk un ? u

k kukL1 (0;T ;V ) :

The estimate (4.21) can be proved in much the same way as the estimate (3.21) was proved in the previous section. 2

References
1] R.A. Adams, Sobolev Spaces , Academic Press, New York, 1975. 2] K.T. Andrews, A. Klarbring, M. Shillor and S. Wright, A dynamic thermoviscoelastic body in frictional contact with a rigid obstacle, to appear in Eur. J. Appl. Math. 3] P.G. Ciarlet, The Finite Element Method for Elliptic Problems , North Holland, Amsterdam, 1978. 4] G. Duvaut and J.L. Lions, Inequalities in Mechanics and Physics, Springer-Verlag, Berlin, 1976. 5] R. Glowinski, Numerical Methods for Nonlinear Variational Problems, Springer-Verlag, New York, 1984. 6] R. Glowinski, J.-L. Lions and R. Tremolieres, Numerical Analysis of Variational Inequalities , North-Holland, Amsterdam, 1981. 7] W. Han and B.D. Reddy, Computational plasticity: the variational basis and numerical analysis, Computational Mechanics Advances 2 (1995), 283{400. 8] W. Han and B.D. Reddy, Convergence of approximations to the primal problem in plasticity under conditions of minimal regularity, submitted. 9] W. Han and B.D. Reddy, Plasticity: Mathematical Theory and Numerical Analysis , Springer-Verlag, New York, 1999. 10] I. Hlavacek, J. Haslinger, J. Necas and J. Lov sek, Solution of Variational Inequalities in Mechanics, Springer-Verlag, New York, 1988. 18

11] N. Kikuchi and T.J. Oden, Contact Problems in Elasticity, SIAM, Philadelphia, 1988. 12] J. Necas and I. Hlavacek, Mathematical Theory of Elastic and Elastoplastic Bodies: An Introduction, Elsevier, Amsterdam, 1981. 13] M. Raous, M. Jean and J.J. Moreau, Contact Mechanics, Plenum Press, New York, 1995. 14] M. Rochdi, M. Shillor and M. Sofonea, Quasistatic viscoelastic contact with normal compliance and friction, to appear in J. of Elasticity. 15] M. Rochdi, M. Shillor and M. Sofonea, A quasistatic contact problem with directional friction and damped response, Applicable Analysis 68 (1998), 409-422. 16] M. Shillor and M. Sofonea, A quasistatic viscoelastic contact problem with friction, preprint. 17] N. Stromberg, Continuum Thermodynamics of Contact, Friction and Wear, Ph.D. Thesis, Linkoping University, Sweden, 1995. 18] N. Stromberg, L. Johansson and A. Klarbring, Derivation and analysis of a generalized standard model for contact friction and wear, Int. J. Solids Structures 33 (1996), 1817{ 1836. 19] E. Zeidler, Nonlinear Functional Analysis and its Applications, Volume II/A: Linear Monotone Operators , Springer-Verlag, New York, 1985.

19