Published as : P.G. LeFloch, Propagating phase boundaries: formulation of the problem and existence via the Glimm scheme, Arch. Rational Mech. Anal. 123 (1993), 153–197 PROPAGATING PHASE BOUNDARIES: FORMULATION OF THE PROBLEM AND EXISTENCE 7 0 VIA GLIMM METHOD 0 2 n Philippe G. LeFLOCH a J Address from 1990 to 1992 : 3 Courant Institute of Mathematical Sciences 2 New York University ] 251 Mercer Street, New York, NY 10012 P A . h Current address : t a Laboratoire Jacques-Louis Lions m Centre National de la Recherche Scientifique [ Universit´e de Paris 6 1 4 place Jussieu, 75252 Paris, France. v E-mail: [email protected] 6 4 6 Abstract. In this paper we consider the hyperbolic-elliptic system of two conservation laws 1 that describes the dynamics of an elastic material governed by a non-monotone strain-stress 0 function. Following Abeyaratne and Knowles, we propose a notion of admissible weak solution 7 for this system in the class of functions with bounded variation. The formulation includes an 0 / entropy inequality, a kinetic relation (imposed along any subsonic phase boundary), and an h nucleationcriterion(for theappearanceof new phaseboundaries). We prove the L1–continuous t a dependence of the solution to the Riemann problem. Our main result yields the existence and m the stability of propagating phase boundaries. The proof is based on Glimm’s scheme and in v: particular on techniques going back to Glimm-Lax. In order to deal with the kinetic relation, i we prove a result of pointwise convergence for the phase boundary. X r a 0. INTRODUCTION This paper deals with the following system of two conservation laws which describes the motion of an elastic material (0.1) ∂ w ∂ v =0, ∂ v ∂ σ(w) =0. t x t x − − Here w > 1 and v represent the deformationgradient and the velocity of the material, respec- − tively. The stress function σ :] 1, [ IR is assumed to be monotonically increasing except − ∞→ in an interval ]w ,w [. Such a form of the stress function is typical in the modeling of solid M m materials which admit different phases. A van der Waals gas also is described by a very similar system. System(0.1)isofmixedtype,i.e. hyperbolicinthephase1region = w < w and 1 M H { } in the phase 3 region = w > w , but elliptic in the intermediate region of phase 2 states. 3 m H { } 1 2 The phase 2 states are known to be both mathematically and physically unstable (James [23]). We will consider here exclusively solutions which take their values in the phase 1 or phase 3 regions only. Solutions to (0.1) in general are discontinuous and so must be understood in the sense of distributions; see Lax [25], [26] for background on weak solutions. Such discontinuous solutions are in general non-unique and those having a physical meaning must be selected through an admissibility (or entropy) criterion. We refer to Dafermos [7] for a review of entropy conditions in the setting of hyperbolic problems. As was pointed out by James [23], the mixed system (0.1) possesses a high degree of non-uniqueness, that a number of authors have attempted to resolve by means of suitable generalizations of entropy criteria from the theory of hyperbolic conservation laws. First of all, Shearer [38] considered the Lax entropy criterion [25], [26]. The viscosity and viscosity-capillarity approches have been analyzed by Slemrod [41], [42]; cf. also Hagan-Slemrod [18], Pego [35] and Shearer [39], [40]. Next, Hattori [19], [20] has investigated the application to (0.1) of the entropy rate admissibility criterion proposed by Dafermos [6]. Hsiao [22] has considered the Liu entropy criterion [32] which allows one to treat equations of state losing genuine nonlinearity in hyperbolic regions. Another approach to resolve the non- uniquenesscan be found in a work by Keyfitz[24]. Additional materialon system (0.1) is found in [12], [13] and [36]. All theabove worksconsidertheRiemannproblemonly, i.e. a Cauchyproblemfor(0.1)with initial condition which consists of two constant states. This problem can be solved explicitly (in a possibly non-unique way) by using simple waves (i.e. shock waves, rarefaction waves or contact discontinuities). Adding an “admissibility criterion” allows one to reduce the class of (admissible) solutions and in most situations to select a unique solution. However, it must be emphasizedthatthesolutionoftheRiemannproblem(whenitisunique)dependsonthechosen admissibility criterion. It turns out that there is no preferred criterion for the selection of the “physically meaningful” solutions of (0.1). A different approach was recently investigated by Abeyratne and Knowles in [2]. The main suggestion of these authors is that system (0.1) is not physically complete enough to describe the evolution of a phase boundary in an elastic material. It must be completed with a kinetic relation imposed along any subsonic phase boundary: this kinetic relation actually yields the rate of entropy dissipation across the phase discontinuity. Moreover, Abeyaratne and Knowles add an initiation criterion which controls the possible appearance of a new phase. We refer to [1] and the references therein for the motivation of introducing a kinetic relation and an initiation criterion which are actually classical in the context of quasi-static problems. Cf. also Gurtin [17] and Truskinovsky [44] for related ideas. Abeyaratne and Knowles proved in [2] that the Riemann problem for (0.1) always admits a unique admissible solution, i.e. a weak solution satisfying the kinetic relation and the initiation criterion, as well as the entropy inequality which reads v2 (0.2) ∂ W(w)+ ∂ σ(w)v 0, t x 2 − ≤ (cid:0) (cid:1) (cid:0) (cid:1) where W :] 1, [ IR is the internal energy function defined by − ∞→ w (0.3) W(w) = σ(y)dy for w ] 1, [. ∈ − ∞ Z0 Next they showed in [3] that the solution of the Riemann problem found by Slemrod through theviscosity-capillarityapproximationcorrespondstoaspecialchoiceofkineticrelationintheir 3 approach. It is not difficult to check also that the solution found by Shearer using Lax entropy inequalitiescoincideswiththemaximallydissipativekineticrelationinvestigatedin[4]. (Ithank Michael Shearer for pointing that out to me.) Thepresentpaperis devotedtocontinuingtheanalysis ofsystem(0.1)throughtheapproach of Abeyaratneand Knowles. As in [2], we will restrict ourselves to the case of a piecewise linear stress-function. This assumption simplifies the calculationsbut it is not a real restrictionto the results of this paper. Our purpose is first (Sections 1 and 2) to give a slightly different presentation of the ideas of [2], whichaswe thinkclarifiestheconceptsofkineticrelationandinitiationcriterionintroduced by Abeyaratne and Knowles. Section 1 presents the mathematical formulation of a well-posed (at least for Riemann data) problem associated with system (0.1). As is usual for hyperbolic problems, we consider bounded solutions of bounded variation (BV). Our formulation follows [2] with however two main modifications. The kinetic relation is introduced from a completely dynamical point of view and not as a generalization of the quasi-static point of view as was done in [2]. That leads us to a larger range of admissible values for the — as we call it below — entropy dissipation function in the kinetic relation. Furthermore, the initiation criterion at some point x is formulated here in two different ways, depending on whether x is in an interior point of the space interval [a,b] where we set the problem, or x is a point of its boundary. For definiteness, we allow spontaneous initiation of a new phase only at the extremities of the bar [a,b], which is consistent with the classical static theory. Then Section 2 describes briefly the solution of the Riemann problem. We explain how to take into account the two changes above in the construction of [2]. The main result of this section establishes the L1 continuous dependence of the Riemann solution with respect to its initial states. It must be emphasized that the two observations above are essential for the continuous dependence property to hold, especially our condition that a new phase may occur spontaneously only at the extremities of the bar. The same results are also obtained for the Riemann problem in a half-space. Note that, although uniqueness of the admissible solution holds for the Riemann problem, nothing is known for the general Cauchy problem. As a matter of fact, theissue of uniquenessfor conservationlaws is understoodin a few numberof situations only. (See, for hyperbolic problems, LeFloch-Xin [31] and the references therein.) The second part of the paper (Sections 3 and 4) focuses on the solutions of the Cauchy problem for system (0.1), which are BV perturbations of a single propagating phase boundary separating a phase 1 state and a phase 3 state. We prove the existence of admissible weak solutions of this form, when the initial data on both sides of the phase discontinuity has small total variation. We treat the case of any non-characteristic phase boundary as well as the case of a characteristic phase boundary provided that no strong wave arises from perturbating the states on both sides of the phase boundary. The random-choice scheme due to Glimm [15] is used to construct approximate solutions to the problem. Its stability in the BV norm is proved from an essentially linear estimate of wave interactions between two Riemann solutions. Such linear interaction terms were used in a different situation by Chern [5] and Schochet [37]. Note that the strength of the phase discontinuity is not (and can not be) assumed to be small in any sense. The stability of the scheme in the total variation norm is sufficient to extract a subsequence converging to a weak solution of the problem. This convergence result holds almost everywhere with respectto the Lebesgue measure. This is sufficient to show that the scheme convergesto a weak solution of the problem. But, proving that this solution is admissible requires a result of pointwise convergence of the phase boundary. In Section 4, we establish this property by using the technique of analysis due to Glimm-Lax [16]. We next prove that it is sufficient, at least for non-stationary phase boundaries, for the passage to the limit in the kinetic relation. 4 An extensionof the resultsin this paper to arbitrarylarge initial data would require a better understanding of the phenomena of initiation of new phases. Manyideasin thispaperarerelatedtothosein thedevelopingtheoryof nonlinearhyperbolic systems in non-conservative form for which we refer the reader to Dal Maso–LeFloch–Murat[8] and LeFloch–Liu [30]; see also [27] to [29]. Acknowledgements. This work was done while the author was assistant professor at CIMS (Courant Instructor), and on leave from a CNRS research position at the Ecole Polytechnique (France). This work was in part supported by a NSF grant DMS-88-06731 and by the CNRS (CentreNational de la RechercheScientifique). It is a greatpleasure for me to thankBob Kohn who encouraged me to study this problem and accepted the job of reading the first version of this paper. The hospitality of Peter Lax at Courant Institutewas highly appreciated. I am also graceful to Rohan Abeyaratne and Michael Shearer for discussions. 1. MATHEMATICAL FORMULATION OF THE PROBLEM This section describes the formulation of the Cauchy problem associated with the mixed system(0.1). Theformulationincludesthesystemofconservationlaws(mass,momentum)(0.1) togetherwith the (Clausius-Duhem)entropyinequality associatedwith theentropy W(w)+v2. 2 Itismadecompletebyaddingtothesebothakineticrelationalonganysubsonicphaseboundary and an initiation criterion for the occurrence of possible new phase boundaries in the solution. We specify below the assumptions on the kinetic relation and the initiation criterion which will be essential to the results of Section 2. This section also introduces notation which will be of constant use throughout this paper. We write system (0.1) in the form v σ(w) (1.1) ∂ u+∂ f(u)=0, u= , f(u)= − . t x w v (cid:18) (cid:19) (cid:18) − (cid:19) For simplicity, we shall assume that the stress-function σ:] 1, [ IR is a piecewise linear − ∞→ function of the following form k w for 1 w w , 1 M − ≤ ≤ (1.2) σ(w) = k w +(k w k w )(w w )/(w w ) for w w w ,  3 m 1 M 3 m m M m M m − − − ≤ ≤  k w for w w. 3 m ≤  The constants k,k ,w and w in (1.2) are assumed to satisfy the properties 1 3 m M (1.3) 0 <k < k and 0 <w <w . 3 1 M m We shall use the notation σ =k w and σ = k w . M 1 M m 3 m The phase 1 region = 1 <w w and the phase 3 region = w w correspond 1 M 3 m H {− ≤ } H { ≥ } toobservableandstablestates. Inourformulationbelow,thesolutioncannot entertheunstable phase2region w < w <w andsomustjumpfrom to orconversely. Adiscontinuity M m 1 3 { } H H between two states in different phases is called a phase boundary. 5 System (1.1) is linear hyperbolic in and , and the corresponding characteristic speeds 1 3 H H are c = k in and c = k in . 1 1 1 3 3 3 ± ± H ± ± H In view of (1.3), the waves inpphase 1 travel faster than thosepin phase 3, i.e. c > c . We may 1 3 also use the notation (1.4) c(w) =c if w w , c if w w . 1 M 3 m ≤ ≥ Note that c(w) is not defined if w belongs to ]w ,w [. With some abuse of notation, and M m 1 H will sometimes also denote (v,w)/ 1 < w w , v IR and (v,w)/w w, v IR , 3 M m H { − ≤ ∈ } { ≤ ∈ } respectively. We also set = . 1 3 H H ∪H Since (1.1) is linear hyperbolic in and , possible discontinuities in the initial data for 1 3 H H (1.1) are simply advected along the characteristic lines of slopes either c or c . (This is 1 3 ± ± true at least up to the time of appearance of a new phase.) The elementary waves in each of the regions and are contact discontinuities. Hence, the special choice (1.2) for the 1 3 H H constitutive law is very convenient. It makes quite simple the analysis in the hyperbolic regions and allows us to focus on the phase boundaries between and . We will see that the 1 3 H H description of the appearance and the evolution of the phase boundaries is far from trivial. In the theory of hyperbolic conservation laws, it is standard to consider solutions u =(v,w) to (1.1) in the functional space L∞(IR IR, ) (recall that = ) which satisfy loc +× H H H1 ∪H3 (1.5a) system (1.1) in the sense of distributions, (1.5b) the entropy inequality (0.2), (0.3) in the sense of distributions and (1.5c) an initial condition u at t =0 in the L1 sense. 0 loc Here u is a given function in L∞(IR, ) and, for future reference, we rewrite the entropy 0 loc H inequality in the form (1.6a) ∂ U(u)+∂ F(u) 0 t x ≤ with v2 w (1.6b) U(u) =W(w)+ , F(u)= σ(w)v and W(w) = σ(y)dy. 2 − Z0 Werecallthatsolutionsinthesense(1.5)areunique(atleastforRiemanndata)inthestandard situation of a (genuinely nonlinear or linearly degenerate) increasing strain-stress function σ. This is no longer true in the case of the mixed system underconsiderationhere: see for instance James [23]. We also point out that the entropy function U is not a convex function. To complete the formulation (1.5), we follow Abeyaratne and Knowles in [2]. Let us give first some motivation for their suggestion. Suppose uǫ =(vǫ,wǫ) is the solution of a regularized version of system (1.1) obtained by adding high-order terms, dependingon a (small) parameter ǫ, in the right-hand side of the equations (e.g. use the viscosity-capillarity terms as was done by Slemrod [41]). As was pointed out by Lax for generalsystems of conservationlaws, the limit 6 u =limuǫ –if itexists(andiftheconvergenceholdsin asuitabletopology)–must beasolution to (1.1) in the sense (1.5); in particular the entropy inequality (1.6) must hold. Since (1.5) is incomplete, it seems natural to “keep more information” about the limiting function u from its regularization uǫ. Specifically Abeyaratne and Knowles’ suggestion is equivalent to replacing (1.6) with the stronger requirement that (1.7) ∂ U(u)+∂ F(u)=µ, t x where µ is a given non-positive measure that clearly must satisfy certain restrictions. Note that in principle µ could be determined by the formula µ= weak-star lim(∂ U(uǫ)+∂ F(uǫ)) t x ǫ→0 (at least when uǫ has uniformly bounded total variation in (t,x)). This formula may not give a veryexplicitexpressionforµ. Hopefully,itturnsoutthat(1.7)isneeded(toachieveuniqueness) only for one kind of discontinuity: the subsonic phase boundaries. Moreover, in that case, we can allow a large range of measures µ. Here, we call subsonic (respectively supersonic) those phase boundaries that travel with speed less (resp. greater) than the contact discontinuities in phase . 3 H The precise formulation of condition (1.7) given below requires that u is a bounded function ofboundedvariation. Whenuhasboundedvariation,wecallentropydissipationthevalueofthe measure ∂ U(u)+∂ F(u) along a curve of (contact or phase) discontinuity of u. According to t x [2], the kinetic relation yields this entropy dissipation along any subsonic phase boundary, as an explicit function, say φ(V), of the speed V of propagationof this discontinuity. In applications, theactualkineticrelation,thatis thefunctionφ, must bedeterminedfromthepropertiesofthe specific material under consideration. This kind of constitutive model is already in extensive use in the quasi-static setting for problems of phase transition in solids. We refer the reader to [1] as well as Truskinovsky [44] and the references cited there. The speed V can also be interpreted as an internal variable and the kinetic relation indeed determines the evolution of this internal parameter. Remark 1.1. 1) That subsonic and supersonic phase boundariesmust be treatedin a different way is clear, for instance when solving Riemann problems. A wave structure with a supersonic phase boundary contains two waves, while one with a subsonic boundary is composed of three waves. This latter case suffers, without a kinetic relation, from a strong lack of uniqueness. Cf. James [23] and Section 2. 2) The approach considered here has some similarity to the theory of nonlinear hyperbolic sys- tems in non-conservative form; cf. Dal Maso-LeFloch-Murat [8] and LeFloch-Liu [30]. Namely, asis thecasefor systems(1.1),the weaksolutionsto thesesystemsarenotuniquelydetermined by the partial differential equations and an entropy inequality, but an additional constitutive relationmust beaddedto ensureuniqueness. ThisfactwasfirstpointedoutbyLeFloch; cf.[27] to [29]. 3) Conservation laws with measure source-term like (1.7) have been useful in various contexts, cf. Di Perna [9], Di Perna-Majda [11], Hou-LeFloch [21]. Let us introduce some notations and recall some facts about functions of bounded variation, that can be found in Volpert [45] and Federer [14]. Let Ω be an open subset of IRm. A func- tion u: Ω IRp belongs to the space BV(Ω,IRp) (respectively BV (Ω,IRp)) if u L1(Ω,IRp) loc → ∈ ∂u (resp. L1 (Ω,IRp)) and the distributional derivatives for 1 j m are bounded (resp. loc ∂y ≤ ≤ j 7 locally bounded) Borel measures on Ω. In what follows, we will always consider functions in L∞(Ω,IRp) BV(Ω,IRp) or L∞(Ω,IRp) BV (Ω,IRp), often called for short BV functions or ∩ loc ∩ loc BV functions. For each BV function u, we have the following decomposition loc loc Ω =C(u) S(u) E(u), ∪ ∪ where C(u) is the set of all points of approximate continuity for u, S(u) is the set of all points of approximate jump for u and E(u) is the set of exceptional points with the property H (E(u)) =0. m−1 Here H is the (m 1)-dimensional Hausdorff measure on IRm. For each point y in S(u), m−1 there exists a unit nor−mal ν IRm and approximate left and right limits for u that we denote ∈ by u (y). The set S(u) consists of the union of a countable number of rectifiable curves. ± We denote the norm of u by u = u + Du(Ω), where Du is the measure BV(Ω,IRp) L1(Ω,IRp) k k k k | | ∂u ∂u ∂u ( , , ..., ). When u = u(t,x) L∞(IR IR, ) BV (IR IR, ), we use the ∂y ∂y ∂y ∈ loc + × H ∩ loc + × H 1 2 m notation: ν (t,x) t ν(t,x) =(ν (t,x), ν (t,x)) and V(t,x) = t x −ν (t,x) x valid for all (t,x) S(u). The ratio V(t,x) represents the speed of propagation of the disconti- ∈ nuity in u at the point (t,x). Note that system(1.1) has the propertyof propagationwith finite velocity (in regions and ). So ν (t,x) will never vanish, and for definiteness we always 1 3 x H H choose ν (t,x) > 0. In the following, we shall always have: u(t) BV for all times t. x ∈ Let φ: ] c ,c [ IR be a function, called below entropy dissipation function, satisfying the 3 3 − → following properties: (1.8a) φ belongs to 2(] c ,0[ ]0,c ]) and φ(0 ) and φ′(0 ) exist, 3 3 C − ∪ ± ± (1.8b) lim φ =ψ(c ) and φ′′′(c ) exists, 3 3 V→c− − 3 (1.8c) lim φ = , V→−c+ −∞ 3 (1.8d) φ is increasing on ] c ,c ] 3 3 − and ψ(V) φ(V) 0 for V ] c ,0], 3 ≤ ≤ ∈ − (1.8e)   0 φ(V) ψ(V) for V [0,c ]. 3 ≤ ≤ ∈ In (1.8b) and (1.8e), the minimal and maximal entropy dissipation functions ψ: ] c ,0] IR 3 − − → and ψ: [0,c ] IR are defined by 3 + → (k k ) k V2 1 3 1 (1.9a) ψ(V) = − w (w − w ) for V ] c ,0] 2 M m− k V2 M ∈ − 3 3 − 8 and (k k ) k V2 1 3 3 (1.9b) ψ(V) = − w (w − w ) for V [0,c ]. 2 m M − k V2 m ∈ 3 1 − Remark 1.2. 1) Inequalities (1.8e) give the range of values taken by the entropy dissipation rate (u) (see below) when varying the left and right values at a discontinuity satisfying the E Rankine-Hugoniot relations and the entropy condition. 2) In [2], instead of (1.8e), Abeyaratne and Knowles assume the (more restrictive) condition: ψ(0) φ(V) 0 for V ] c ,0], 3 ≤ ≤ ∈ − (1.8e)′   0 φ(V) ψ(0) for V [0,c ]. 3 ≤ ≤ ∈   3) Assumptions (1.8) made in this paper are indeed satisfied in the examples considered by [3] and [4]. For instance, they are fulfilled by the maximally dissipative function φ defined by: max φ (V) =ψ(V) for V ] c ,0], ψ(V) for V [0,c ]. max 3 3 ∈ − ∈ We next define the entropy dissipation rate (u) associated with any function E u L∞(IR IR, ) BV (IR IR, ) ∈ loc +× H ∩ loc +× H by the following formula ν x (1.10) (u) = (U(u ) U(u )) (F(u ) F(u )) + − + − E − − − ν − t which defines (u)(t,x) at H –almost every point (t,x), where ν (t,x) = 0 (i.e. V(t,x) = 0). 1 t E 6 6 (u) is the product of 1 by the jump of the measure ∂ U(u) +∂ F(u) along the curve of E −νt t x approximate jump of u. Formula (1.10) makes sense only if ν (t,x) = 0. However, it is a simple t 6 observationthat if u is assumed to be a weak solution to system (1.1), then the above jump (i.e. the entropy dissipation) vanishes at the points where ν vanishes. This fact allows us to define t (u)(t,x) H –almost everywhere, as shown by the following lemma. 1 E Lemma 1.1. If u L∞(IR IR, ) BV (IR IR, ) is a weak solution to (1.1), then one ∈ loc +× H ∩ loc +× H has w+ 1 (1.10)′ (u) = σ(y) (σ(w )+σ(w )) dy, + − E − − 2 Zw− (cid:26) (cid:27) at H –almost every (t,x) such that ν (t,x) =0. 1 t 6 From now on, we use (1.10)′ to define (u)(t,x). E 9 Proof. At a point of approximate discontinuity (t,x) of the solution u, the following Rankine- Hugoniot relations hold: ν (w w ) ν (v v ) = 0, t + − x + − − − −   ν (v v ) ν (σ(w ) σ(w )) =0. t + − x + − − − −  These relations used in (1.10) yield: w+ 1 ν (u) = σ(y)dy+ (v2 v2) x(σ(w )v σ(w )v ) −E 2 +− − − ν + +− − − Zw− t w+ 1 ν ν x x = σ(y)dy+ (v +v ) (σ(w ) σ(w )) (σ(w )v σ(w )v ). + − + − + + − − 2 ν − − ν − Zw− t t We thus get w+ ν x (u) = σ(y)dy+ v σ(w )+v σ(w ) v σ(w ) v σ(w ) + + − + + − − − −E 2ν − − Zw− t (cid:8) 2σ(w )v +2σ(w )v , + + − − − (cid:9) so that w+ 1 ν x (u) = σ(y)dy (v v )(σ(w )+σ(w )), + − + − −E − 2 ν − Zw− t which, in view of the Rankine-Hugoniot relations above, gives the desired result (1.10)′. ⊔⊓ Let usdenoteby B (u) theset of all pointsof approximate discontinuityin a weaksolution sub u that correspond to a subsonic phase boundary. This means: B (u)= (t,x) S(u) either : u (t,x) , u (t,x) and V c , sub − 1 + 3 3 ∈ | ∈ H ∈ H | | ≤ (cid:26) or : u (t,x) , u (t,x) and V c . − 3 + 1 3 ∈ H ∈ H | | ≤ (cid:27) Inviewof(1.6),theBorelmeasure∂ U(u)+∂ F(u)isgloballynon-positive. Thekinetic relation t x now specifies the value itself (and not only the sign) of this measure along any subsonic phase boundary. In other words, for H –almost all (t,x) (u), one must have 1 sub ∈ B φ(V(t,x)) if u (t,x) , − 1 − ∈ H (1.11) (u)(t,x) =  E  φ( V(t,x)) if u (t,x) . − 3 − ∈ H   Remark 1.3. Asamatteroffact,thetravelingwavesobtainedthroughtheviscosity-capillarity regularizationtosystem(1.1)convergetoweaksolutionsof(1.1)thatsatisfythekineticrelation (1.1)withaspecificchoiceoffunctionφ. Thisfunctioncanbedeterminedexplicitlyanddepends only on the viscosity and capillarity coefficients introduced in the regularization (cf. [3]). 10 Finally, we have to formulate the initiation criterion, which together with the above kinetic relation will allow us to rule out all non-physical solutions to our problem. Let ]a,b[ be a space interval in which we are going to set the problem, with a < b and possibly a = and/or −∞ b =+ . The initiation criterion will reflect the following facts: ∞ no new phase occurs from any point x in ]a,b[ (1.12) ( except if no solution exists without creation of a new phase, a new phase state may occur at the boundary point x= a, (1.13) even if a solution with no new phase exists;    a criterion is required to make the choice   and a new phase state may occur at x =b, (1.14) even if a solution with no new phase exists;    a criterion is required to make the choice.   From the mathematical point of view, condition (1.12) is essential: it ensures that sponta- neous initiation of a new phase inside ]a,b[ cannot occur from two nearby initial states in the same phase (cf. Section 2). This does not exclude the possibility (and it really happens) that an initial discontinuity with large jump gives rise to, for instance, a phase 1 state although the states on both sides of the initial discontinuity are in phase 3. However, by condition (1.12), a single constant state is always a (trivial) admissible solution. (This property was not satisfied in the construction of [2].) This is also essential to get the L1 continuous dependence property for Riemann solutions, proved below in Section 2. Conditions (1.13) and (1.14) follow the quasi-static theory [1]. They allow “spontaneous nucleation” of a new phase only at the end points of [a,b]. Note that, more generally, we could as well allow nucleation at some arbitrary given points of [a,b]. Our actual restriction is that the points of spontaneous nucleation are known a priori and follow a selection criterion of the formspecifiedbelow. However,whilethisformulationisfullysatisfactoryfromthemathematical pointof view, it doesnotreproducewhatisreallyobservedin practicalexperimentswith elastic bars. Namely, in experiments, when pulling out an elastic bar uniformly in phase 1, initiation of phase 3 regionsin the bar occurssuccessively and (apparently)randomly at various places in the bar. Physicists assert that initiation occurs at microscopic inhomogeneities of the material. A complete treatement of the initiation mecanism is beyond the scope of this paper and would probably require a statistical description. (As a matter of fact, this might be included in the random choice scheme, studied below, quite easily.) It remains to provide an analytic version of the conditions (1.12) to (1.14). For convenience, we use here an averaged strain in our formulation. (In [1] and [2], the stress and the entropy dissipation rate, respectively, are used instead.) Given any function u = (v,w) in L∞(IR loc + × IR, ) BV (IR IR, ), we set loc + H ∩ × H c(w )w +c(w )w +v v − − + + + − (1.15) h = − u c(w )+c(w ) − +