On the Cauchy Problems for Polymer Flooding with Gravitation

Transcription

1 On the Cauchy Problems for Polymer Flooding with Gravitation Wen Shen Mathematics Department, Penn State University. November 5, 2015 Abstract We study two systems of conservation laws for polymer flooding in secondary oil recovery, one with gravitation force and one without. For each model, we prove global existence of weak solutions for the Cauchy problems, under rather general assumptions on the flux functions. Approximate solutions are constructed through a front tracking algorithm, and compactness is achieved through the bound on suitably defined wave strengths. 1 Introduction In this paper we study two models for enhanced secondary oil recovery by polymer flooding in two phases without adsorption terms. In the first model we neglect the effect of the gravitation force. The model takes the form of a 2 2 system of conservation laws, { s t + f(s, c) x = 0, (1.1) (cs) t + (cf(s, c)) x = 0, associated with the initial data (1.2) { s(0, x) = s(x), c(0, x) = c(x). Here s R is the saturation of the water phase and c R is the fraction of the polymer. For any fixed c, the mapping s f(s, c) is typically s-shaped. When c remains constant, the two equations in (1.1) are the same, and the system reduces to a scalar Buckley-Levrett equation [5]. We refer to (1.1)-(1.2) as the non-gravitation model. In the second model, we consider the effect of the gravitation force. The model takes the same form as (1.1), but with a different flux function. We have { s t + F (s, c) x = 0, (1.3) (cs) t + (cf (s, c)) x = 0, where the flux function F (s, c) is (1.4) F (s, c) = f(s, c)(1 K g λ(s, c)). 1

2 Here f(s, c) is the same as the fractional flow in (1.1), K g 0 is a constant that reflects the effect of the gravitation force, and λ(s, c) denotes the mobility of the oil phase. We referred to (1.3)-(1.2) as the gravitation model. For a detailed derivation of the models, see e.g. [9, 18, 4]. The system (1.1) and its variations have been studied by many authors. In particular, solutions of Riemann problems are extensively studied. Johansen, Tveito and Winther [14, 15, 16] constructed global Riemann solver for a model for polymer flooding with adsorption, under the additional monotonicity assumption f c < 0, and conducted numerical simulations with front tracking. Isaacson and Temple [12] studied the Riemann problem of a polymer flooding model. A model for a multicomponent chromatography in two-phase environment is studied in [7]. The system demonstrates many interesting properties. It is observed in [27] that if the initial data c(x) is smooth, then c(t, x) remains smooth for all t > 0. Such special features can be viewed more conveniently in a Lagrangian coordinate. Following Wagner s construction [28], one can introduce a coordinate change (t, x) (ψ, φ), where (ψ, φ) denote the Lagrangian coordinates defined as { φ x = s, φ t = f(s, c), (1.5) ψ = x. Here φ can be viewed as the potential of the first equation in (1.1). The Eulerian system (1.1) now takes an interesting form in the Lagrangian coordinates, ( ) 1 ( ) s = 0, ψ f(s, c) φ f(s, c) (1.6) c ψ = 0. Note that we obtain a triangular system in (1.6), where the second equation is decoupled. Thus the value c remains constant in the time variable ψ. Therefore, if the initial data c(0, φ) is smooth, it will remain smooth for all ψ > 0. Thanks to the classical result of Wagner [28] on the equivalence between the Eulerian and Lagrangian equations, the weak solutions for (1.1) and (1.6) are equivalent for s s o > 0. For the triangular system (1.6), solutions can be constructed by solving the second equation for c, which is constant in time ψ but possibly discontinuous in the space φ, and plugging it back in the first equation. This procedure results in a scalar conservation law with discontinuous flux function. Scalar conservation laws with discontinuous coefficient have extensive interests and have been studied by many authors.for nonlinear hyperbolic systems, it is well known that some additional entropy condition needs to be imposed to obtain a unique weak solution. These include the Lax condition [20], the Liu condition [21] and the vanishing viscosity limit [2, 3], each defined for suitable cases. For non-hyperbolic systems, there have been several definitions. Keyfitz & Kranzer [17] proposed a generalized Lax condition for a model of elasticity. Based on this entropy condition, Isaacson & Temple [12] constructed approximation solutions using Glimm s Randon Choice scheme, removing the monotonicity assumption of f c < 0. Furthermore, they proposed a singular mapping to measure wave strength, and obtained a 2

3 global bound on the total wave strength. This estimate provides the necessary compactness for the convergence of the approximate solutions generated by the Glimm s scheme, establishing the existence of weak solutions in L 1 loc. Isaacson & Temple [25, 13] also showed that in general total variation of the conserved quantity blows up in finite time due to nonlinear resonance, therefore solutions are not in BV. A different entropy condition, referred to as the minimum-jump condition, suitable for scalar conservation laws with discontinuous flux functions and nonlinear resonance, was proposed by Gimse & Risebro [8, 9]. They proved its equivalence to the vanishing viscosity limit. We remark that, the minimum-jump condition and the generalized Lax condition give different solutions to Riemann problems in many cases. In this paper, we adopt the minimum-jump condition by Gimse and Risebro [8, 9]. Under the monotonicity assumption f c < 0, Temple s functional [12] can be applied to the solutions satisfying the minimum-jump condition, and existence of solution could be established for the Cauchy problem for (1.1). In this paper, we remove the assumption f c < 0, and prove a more general result. In particular, we allow the graphs of the mappings s f(s, c) for different c values to intersect with each other, therefore the Temple functionals in [12] can not be applied. Wave front tracking algorithm will be used to generate piecewise constant approximate solutions. In order to show convergence of the approximate solutions, the key estimate is the bound on the total wave strength, suitably defined for this general case. In [23], for a model of slow erosion of granular flow with rough geological data, a new wave strength functional is introduced, for the case where the graphs of the flux functions can intersect. This functional leads to the uniform bound on the total wave strength. The slow erosion model was originally derived and studied in [1, 24]. However the flux function in the slow erosion model is strictly convex and takes a simpler shape than the Buckley-Leverett flux in (1.1). In this paper, such nonlinear mapping is extended to the more complex flux functions. The main results in this paper are the uniform bounds on the total variations of the flux functions for approximate solutions generated by front tracking algorithm. The main tools of analysis are two new functionals to measure the wave strengths for the two models in (1.1) and (1.3). Using these new functionals, we show that the total wave strengths is non-increasing at any interactions. Solutions of the Riemann problems for (1.1) are constructed using implicitly the insight in (1.6). In order to obtain result also in the case s = 0, we work directly in the Eulerian system (1.1), solving implicit Riemann problems for a scalar conservation law with discontinuous flux. The second model (1.3) with gravitation force was studied by various authors. In [9], the Cauchy problem for a scalar conservation law for two-phase flow with gravitation and rough permeability function was studied. Under the further assumption that λ = λ(s), and f c < 0, the simplified model is studied in [18], where the author proved the existence and L 1 stability of the solution. The simplification leads to a key property: there exists an ŝ such that F (ŝ, c) = 0 for all c, and F c > 0 for s (0, ŝ), and F c < 0 for s (ŝ, 1). Therefore, (ŝ, 0) is the only intersection points of the graphs of the functions F (s, c 1 ) and F (s, c 2 ) for any two different values of c 1, c 2. The authors took advantage of this structure and designed a functional to measure the wave strength, such that the total wave strength is non-increasing in time, in a suitable Lagrangian coordinate. In this paper, we will remove these simplifications, 3

4 and study the more general model. The rest of the paper is organized as follows. In Section 2 we study the non-gravitation model. After stating the precise assumptions and the main results in Section 2.1, we analyze in some detail the solutions of Riemann problems and the structure of the admissible shocks in both families in Sections A new wave strength functional is introduced in Section 2.4, and interaction estimates are derived in Section 2.5. The front tracking algorithm is defined in Section 2.6, as well as the convergence. In Section 3 we study the gravitation model, following a similar structure. Finally, several concluding remarks are provided in Section 4. 2 The Non-Gravitation Model 2.1 Assumptions, Preliminaries and Main Results We denote the domain for the solutions as { } (2.1) D = (s, c) 0 s 1, 0 c 1. The fractional flow f(s, c) satisfies the following assumptions: For every c, we have f(0, c) = 0, f(1, c) = 1, f s (0, c) = 0, f s (1, c) = 0. For every c, there exists a unique value ŝ(c) (depending on c), such that f ss (ŝ(c), c) = 0. Furthermore, it holds: (2.2) { fss (s, c) > 0, for 0 s < ŝ(c), f ss (s, c) < 0, for ŝ(c) < s 1. Note that we do not have any monotonicity assumption on the mapping c f. This allows the graphs of s f with different c values to intersect with each other, possibly multiple times, on the interval 0 < s < 1. We define the function (2.3) g(s, c) = Straight computation gives the following relations: f(s, c). s (2.4) g(0, c) = f s (0, c), g s (0, c) = 1 2 f ss(0, c). The assumptions on f(s, c) lead to the following assumptions on the function g(s, c): For every c we have g(0, c) = 0, g(1, c) = 1, g s (0, c) > 0. 4

5 For every c, there exists a unique value s M (c) where the mapping s g attains its maximum value with g s (s M (c), c) = 0 and g ss (s M (c), c) < 0. Furthermore, the mapping s g is strictly increasing for 0 < s < s M, and strictly decreasing for s M < s < 1. For every c, there exists a unique value s E (c) such that 0 < s E < 1 and g(s E (c), c) = 1. Again, the graphs of s g with different c values may intersect with each other, and the intersection points are the same as those for s f. Remark 2.1. We remark that the assumptions on the strict convexity of f in (2.2) can be relaxed. Since g s = f s s f s 2, g ss = f ss s 2f s s 2 + 2f s 3, then at the maximum point s M we have f(s M (c), c) = sf s (s M (c), c), and g ss (s M (c), c) = 1 s f ss(s M (c), c). Therefore, we can relax the strict inequalities in (2.2), and only require strictly convexity at the following two cases: f ss (0, c) > 0 and f ss (s M (c), c) < 0. We now define a function Φ(s, c) which will be used later to measure the wave strength. For every c, we let (2.5) Φ(s, c) = s 0 g s (σ, c) dσ. Clearly, the mapping s Φ is non-decreasing for every c. Illustrative plots of the three functions f(s, c), g(s, c) and Φ(s, c) for a given c can be found in Figure 1. Figure 1: Typical plots for the functions f, g, Φ for a fixed c. 5

6 Definition 2.2. The functions (s, c) are called a weak solution of (1.1)-(1.2), if the equation is satisfied in a distributional sense, such that (2.6) (2.7) 0 0 (sφ t + f(s, c)φ x ) dx dt + (scφ t + cf(s, c)φ x ) dx dt + φ(x, 0) s(x) dx = 0, φ(x, 0) s(x) c(x) dx = 0, for every test function φ C 1 0. Our first main theorem provides the existence of weak solutions for the Cauchy problems for the non-gravitation model. Theorem 2.3. Assume that the initial data ( s, c) has bounded total variation. Then, there exists a weak solution (s(t, x), c(t, x)) for the Cauchy problem (1.1)-(1.2). The rest of this section is devoted to the proof of this Theorem. 2.2 Basic Analysis The system (1.1) can be written in the quasi-linear form [ ] [ ] [ s fs (s, c) f (2.8) + c (s, c) s c 0 f(s, c)/s c where the coefficient matrix has two eigenvalues t (2.9) λ 1 = f s (s, c), λ 2 = f(s, c)/s, and the corresponding two right-eigenvectors [ ] [ 1 (2.10) r 1 =, r 0 2 = ] x = 0, f c (s, c) f s (s, c) f(s, c)/s Note that the wave families are not labelled according their wave speeds. We refer to the first family as the s-family, and the second one as the c-family. For the s-family, the integral curves of the eigenvectors are horizontal lines, while for the c-family, they are shown in Figure 2. The directional derivative of λ 2 along the integral curves of the eigenvector r 2 is [ fs /s f/s λ 2 r 2 = 2 f c /s ] [ f c f s f/s ]. ] = 0. Thus the second family is linearly degenerate, and every jump in this family is a contact discontinuity. Note that for both families, the shock curves and the rarefaction curves coincide, therefore the system is a Temple class. However, it is known that the system is not hyperbolic. Along the curve where f s (s, c) = f(s, c)/s, we have λ 1 = λ 2, and r 1 is parallel to r 2, thus the system is parabolic degenerate. This curve goes through the point on the c-integral curve where it reaches a horizontal tangent. 6

7 Figure 2: Integral curves of the eigenvectors for the c-family for the non-gravitation model. There are two families of waves, referred to as the s-waves and the c-waves. Along an s-wave, the c-value remains constant. Consider the Riemann data (s L, c) and (s R, c) as the left and right states. Then, the solution of the Riemann problem consists of only s-waves, which are simply the solution of the Riemann problem for a scalar conservation law { s s t + f(s, c) x = 0, s(x, 0) = L, x < 0, s R, x > 0. Since the mapping s f is S-shaped, the solution of the Riemann problem may consist of multiple s-waves. Along a c-wave, both s and c values will vary. Let (s L, c L ) and (s R, c R ) denote the left and right state of a c-wave. The wave speed σ c must satisfy the Rankine-Hugoniot jump conditions σ c (s L s R ) = f(s L, c L ) f(s R, c R ), σ c (c L s L c R s R ) = c L f(s L, c L ) c R f(s R, c R ). Simple computation shows that the wave speed must satisfy (2.11) σ c = f(sr, c R ) s R = f(sl, c L ) s L. For simplicity of the notations, we denote the functions f L (s) = f(s, c L ), f R (s) = f(s, c R ). This shows that, two states (s L, c L ) and (s R, c R ) can be connected by a c-wave, if on the graphs of the functions f L and f R, the extension of secant line connecting the left and right states goes through the origin. Furthermore, the slope of the ray is exactly the speed of the c-wave. See the left plot in Figure 3. This fact was observed in various previous literature, e.g. [12, 27]. 7

8 f f L f R s g σ c 1 s Figure 3: Connecting the left and right states of a c-wave. Left plot: function f(s, c). Right plot: g(s, c). 2.3 Solutions of the Riemann Problems To solve a Riemann problem with any given Riemann data (s L, c L ) and (s R, c R ), the key step is to locate the path of the c-wave. Once this path is located, then one can find the remaining s-waves by solving a scalar conservation law. From the above discussion, we see that we need to solve the Riemann problem for a scalar conservation law with an implicitly discontinuous flux function, i.e., (2.12) s t + f(s, x) x = 0, where f(s, x) = { f L (s), x < σ c t, f R (s), x > σ c t, with Riemann data (2.13) s(x, 0) = { s L, x < 0, s R, x > 0. Note that σ c is the speed of the c-wave, which is unknown and will be solved together as we locate the c-wave path. This makes the Riemann problem implicit. We now denote the functions (s) = g(s, c L ) and (s) = g(s, c R ). Note that (2.10) implies (2.14) σ c = (s R ) = (s L ). We observe that, on the graphs of the functions and, the c-wave path is a horizontal line. See the right plot in Figure 3. The speed of the c-wave is precisely the g-value of this horizontal line. This motivates the study of the Riemann problem for the following scalar conservation law with a discontinuous flux function: { g (2.15) s t + g(s, x) x = 0, g(s, x) = L (s), x < 0, (s), x > 0, 8

9 with the Riemann data (2.13). Note that we do not have the wave speed σ c in the Riemann problem, therefore the problem is no longer implicit. Problem (2.15) is equivalent to problem (2.11) in the sense that the location of the c-wave is the same for both problems. For scalar conservation laws with discontinuous flux functions, the existence and uniqueness of solutions of Riemann problem is proved by Gimse & Risebro [8], for a class of flux functions under rather general assumptions, using the minimum-jump entropy condition. To facilitate the readability of this paper, we describe the corresponding simpler version for our model. We denote by I (s L, ) the set of ŝ values where the solution of the Riemann problem { s t + s (s) x = 0, s(x, 0) = L, x < 0, ŝ, x > 0, consists of only non-positive waves. The set I (s L, ) can be constructed manually. We have {s L }, 0 s L < s 1, I (s L, ) = {s L } [ s L, 1], s 1 s L < s M, [s M, 1], s M s L 1. Note that the set may include an isolate point {s L }. See Figure 4 for an illustration. This gives us the set that contains candidates for the left state of the c-wave path. We observe that, restricted to the set I (s L, ), the function is decreasing. gl gl L gl L L s L s 1 s s 1 s s L s M s L 1 s 1 s M 1 s Figure 4: The set I (s L, ) where the Riemann problem consists of only non-positive waves. Similarly, we denote by I + (s R, ) the set of ŝ values where the solution of the Riemann problem { s t + ŝ, x < 0, (s) x = 0, s(x, 0) = s R, x > 0, consists of only non-negative waves. Again, the set I + (s R, ) can be constructed by hand. We have { I + (s R, [0, s M ], 0 < s R < s M, ) = [0, s R ] {s R }, s M s R 1. 9

10 Note that the set could include an isolated point {s R }. See Figure 5 for an illustration. The set I + (s R, ) contains candidates for the right state of the c-wave path. We observe that, restricted to the set I + (s R, ), the function is increasing. gr gr R 1 1 R s M s s R s M s R 1 s Figure 5: The set I + (s R, ) where the Riemann problem consists of only non-negative waves. Given the Riemann data (s L, c L ) and (s R, c R ), the path of the c-wave must be a horizontal line that connects the two points, one on the graph of restricted to the set I (s L, ) and the other on the graph of restricted to the set I + (s R, ). Following the minimum-jump condition, this path is chosen such that the jump in the c-value along this path is minimized. Thanks to the monotone properties of the g-values on the sets I and I + for the function and, respectively, the path always exists and is unique. In the case where one or two isolated points are on the path, then the path with the most isolated points will be chosen. We call such c-waves and their paths as admissible. Thus, we have proved the following lemma. Lemma 2.4. There exists a unique global solution for the Riemann problem of (2.15). Discussions on admissible c-wave paths. Let (s L, c L ) and (s R, c R ) be the left and right state of a c-wave, respectively, with g(s L, c L ) = g(s R, c R ), and write (s) = g(s, c L ) and (s) = g(s, c R ). We also denote by G L M, GR M the maximum values of the functions gl,, respectively. We define the conditions: (2.16) There exists an ŝ such that (ŝ) = (ŝ) and ( ) (ŝ) < 0, ( ) (ŝ) > 0. Here is an immediate technical Lemma. Lemma 2.5. For any given values of c 1 c 2, there exists at most one intersection point ŝ for the graphs s g(s, c 1 ) and s g(s, c 2 ) that satisfies (2.17) g s (ŝ, c 1 ) g s (ŝ, c 2 ) 0. 10

11 Thus, for any (c L, c R ), there exists at most one ŝ that satisfies (2.16). We discuss two cases, depending on whether (2.16) is satisfied. Case 1. If (2.16) is not satisfied, then the c-wave path must lie below the smaller value of M and gr M. Say, if gl M < gr M, then the g-value of the c-wave path must be less than gl M, and the path must connect the parts of graphs where the slopes are of the same sign. See Figure 6. In the left plot, admissible c-wave paths are illustrated. In the middle plot, the path is admissible but its location is unstable with respect to perturbation. Adding a small perturbation such that (s L ) (s R ), the resulting c-wave path will be in the region in the left plot, depending on the perturbation. Such an unstable path will not be part of any Riemann solution except when this is the Riemann data at t = 0. If this occurs, we will add some small perturbation to avoid it. Finally, in the right plot the path is not admissible, and the actual c-wave path would be at the line where g = M. actual path unstable non-admissible Figure 6: The admissible paths of the c-wave, where (2.16) is not satisfied, and M < gr M. The case where M > gr M is totally similar, and is illustrated in Figure 7. actual path unstable non-admissible Figure 7: The admissible paths of the c-wave, when (2.16) is not satisfied, and M > gr M. We observe that, the stable admissible c-wave path can only connect the points on graphs of and where the slopes are of the same sign, i.e., ( ) (s L )( ) (s R ) 0. 11

12 Case 2. If (2.16) holds, then the g-value for any c-waves must be smaller than the g value at the intersection point, i.e., g (ŝ) = (ŝ). In Figure 8, the left plot indicates the admissible c-wave paths. Note that the intersection point is an admissible path. In the middle plot we show some admissible paths whose locations are unstable with respect to perturbation. The path in the right plot is not admissible, and the actual path will be at the intersection point. Note that in this case, the intersection point acts as an attractor for many c-wave paths. In fact, if (s L ) (ŝ) and (s R ) (ŝ), then the c-wave path will be at this intersection point. actual path unstable non-admissible Figure 8: The admissible paths of the c-wave, where (2.16) holds. Remark 2.6. We observe that, for an admissible c-wave paths (both stable and unstable), we have (i) If s L < s R, then the upper envelope of the graphs of the functions and lies above the c-wave path on the interval s L < s < s R. (ii) If s L = s R, then this intersection point can be an admissible path. (iii) If s L > s R, then the lower envelope of the graphs of the functions and lies below the c-wave path on the interval s L < s < s R. We remark its resemblance to the classical Oleinik (E) condition [22]. 2.4 Definitions of Wave Strengths The wave strength of an s-wave is defined through the function Φ in (2.5). Let (s L, ĉ) and (s R, ĉ) be the left and right states of an s-wave. Then its strength is (2.18) s = Φ(s L, ĉ) Φ(s R, ĉ). The strength of the c-waves needs to be defined carefully. Let (s L, c L ) and (s R, c R ) be the left and right states of a c-wave, and let (s) = g(s, c L ) and (s) = g(s, c R ) be the flux function at the left and right sides of the wave. We denote s L M and sr M the points where gl 12

13 and reach their max values, respectively, and denote their max values by M = gl (s L M ) and M = gr (s R M ). First, we locate the unique point (ŝ, ĝ) as follows. If the conditions in (2.16) are satisfied, then we let ŝ be the intersection point, and let ĝ = (s o ) = (s o ). Otherwise, ŝ will be the point where or attains its maximum values, whichever has the smaller maximum value. To be precise, we let { (s L M (2.19) (ŝ, ĝ) =, gl M ), if gl M gr M, (s R M, gr M ), if gl M gr M. See Figure 9 for an illustration. ĝ g ĝ g ĝ g ŝ s ŝ s ŝ s Figure 9: The location of (ŝ, ĝ) for various cases. By the properties of the function g(s, c), the existence and uniqueness of the values (ŝ, ĝ) is obvious. Now we define two non-negative quantities (see Figure 10): (2.20) M L = M ĝ, M R = M ĝ. The strength of the c-wave is defined as: 2M L + 2M R, if ( ) (s L ) 0 and ( ) (S R ) 0, 2M L + 4M R, if ( ) (s L ) 0 and ( ) (S R ) 0, (2.21) c = 4M L + 2M R, if ( ) (s L ) > 0 and ( ) (S R ) 0, (unstable), if ( ) (s L ) > 0 and ( ) (S R ) < 0. Note that the case ( ) (s L ) 0 and ( ) (S R ) 0 happens only at: (i) an intersection point satisfying (2.16); or (ii) one of the extreme point of and is in the path, i.e., either ( ) (s L ) = 0, or ( ) (s R ) = 0, or both. The unstable case will be avoided by a small perturbation in the initial data. Remark 2.7. If the conditions in (2.16) do not hold, then the definition in (2.21) is equivalent to the functional proposed by Temple [26]. 13

14 ĝ g M L M R ĝ g M R M L = 0 s ŝ ŝ = s L M Figure 10: The definition of M L and M R in two cases. s 2.5 Interaction Estimates and the Bound on the Total Wave Strength In this section we prove the following interaction estimates. Lemma 2.8. At any wave interaction for the non-gravitation model, the total wave strength measured in (2.18) and (2.21) is non-increasing. Thus, the total wave strength is nonincreasing in time. Proof. When two s-waves interact, they must have the same c values, say c = ĉ. This interaction is the same as that of the scalar conservation law s t + g(s, ĉ) x = 0. Clearly, the total wave strength is non-increasing. Since all c-waves are contact discontinuities, they will never interact with each other. It remains to consider the interactions between a c-wave and an s-wave. We study two cases. Case 1: when (2.16) does not hold. Here the wave strength is the same as Temple s function [26], therefore the interaction estimates are the same as those in [12], and the wave strength is non-increasing after interaction. The proof follows also a similar argument as the case below. We skip the details. Case 2: when (2.16) holds. Let (ŝ, ĝ) be the intersection point of the graphs of,. There are several types of interactions. If the paths of the incoming and outgoing c-waves are on the same side of the intersection point, then the total wave strength is clearly unchanged. It can happen that, after interaction, the c-wave path is moved to the intersection point, either from the left or from the right. See Figure 11, plots (1a) and (1b). Let L, M in and R denote the three states of the incoming waves. In plot (1a), the incoming c-wave path is on the left of and below the intersection point, and it interacts with an s-rarefaction which comes from the left, and L lies above the intersection point. After interaction, the 14

15 intermediate state of the out-going waves M out is located at the intersection point, and we have three waves coming out: an s-shock from L to M out, a c-wave at M out, and an s-rarefaction from M out to R. The total strength of the incoming waves are: W in = s in + c in = [ Φ L (s L ) Φ L (s M in ) ] + [ 4M L + 2M R]. The total strength of the 3 outgoing waves are: Since W out = s in,1 + c in + s in,2 = [ Φ L (s Mout ) Φ L (s L ) ] + [ 2M L + 2M R] + [ Φ R (s Mout ) Φ R (s R ) ]. Φ L (s Mout ) Φ L (s L ) 2M L and Φ R (s Mout ) Φ R (s R ) Φ L (s L ) Φ L (s M in ), we have W out W in 0. The case in plot (1b) is totally similar. An incoming c-wave at the intersection can also be moved to a different location after the interaction, as illustrated in plots (2a) and (2b) in Figure 11. In plot (2a) an incoming s-shock interacts with a c-wave which is at the intersection point, where the point L is on the left of and below the intersection point, and R coincide with M in. After the interaction, the c-wave path is moved to the left. We have W in = s in + c in = [ Φ L (s M in ) Φ L (s L ) ] + [ 2M L + 2M R], and Since W out = c out + s out = [ 4M L + 2M R] + [ Φ R (s R ) Φ R (s Mout ) ]. Φ L (s M in ) Φ L (s L ) = 2M L + [ Φ R (s R ) Φ R (s Mout ) ], we have W out = W in. Also, the case in plot (2b) is totally similar. Finally, the c-wave path can move from one side of the intersection point to the other after wave interact, as shown in plots (3a) and (3b) in Figure 11. In plot (3a), a c wave on the left interacts with a large s-shock on the right, and produce a left-going s-shock and a c-wave. We have W in = s in + c in = [ 4M L + 2M R] + [ Φ R (s R ) Φ R (s M in ) ], and Since W out = s out + c out = [ Φ L (s Mout ) Φ L (s L ) ] + [ 2M L + 4M R]. [ Φ R (s R ) Φ R (s M in ) ] 2M R = [ Φ L (s Mout ) Φ L (s L ) ] 2M L, we conclude W out = W in. Again, the case in plot (3b) is totally similar. This completes the proof for Lemma

16 M in (1a) L R M out (2a) L M in,r M out L M in M out R (3a) (1b) M out L R M in (2b) M in,l M out R (3b) L Min R M out Case 1 Case 2 Case 3 Figure 11: Interaction estimates for Case Front Tracking Approximations and the Convergence We generate piecewise constant approximate solutions using a front tracking algorithm. Let ε > 0, and we construct two finite sets G = {g i } and C = {c i }, such that c i < c i+1, g i < g i+1, g i+1 g i ε, c i+1 c i ε. Furthermore, we ensure that the set G includes the maximum values g(s M (c i ), c i ) for every i, as well as the g values at the intersection points of the graphs of the functions s g(s, c i ) and s g(s, c j ) that satisfies the conditions in (2.16), for every i j. For each c i C, the flux function s g(s, c i ) is approximated by a piecewise polygonal function following Dafermos [6], whom we denote by g ε (s ε, c ε ). We denote by Φ ε (s ε, c ε ) the discrete version of Φ. The initial data ( s, c) are approximated by piecewise constant functions ( s ε, c ε ), where the discrete flux g ε and c ε take only the values in the sets G and C, respectively. Furthermore, the following estimates hold, for some bounded constant M: s( ) s ε ( ) L 1 Mε, c( ) c ε ( ) L 1 Mε, TV {Φ ε ( s ε, c ε )} TV {Φ( s, c)}, TV { c ε } TV { c}. By the assumption on the total variation of the initial data, this discretization is well-defined. Let X = {x i } denote the set of the jumps in either s ε or c ε. At t = 0, we solve a Riemann problem at every x i. Rarefaction waves are represented by a series of small fronts whose size 16

17 is bounded by ε. All fronts travel with Rankine-Hugoniot speed. As time evolves, the fronts will interact and new Riemann problems will be solved. The process continues until the final time T is reached. The convergence, and consequently the existence of weak solutions, is a direct consequence of the bound on the total wave strength. Since the initial data ( s ε, c ε ) have bounded variation, the initial total wave strength is bounded. Thanks to the wave interaction estimates in Section 2.5, the total wave strength remains non-increasing in time. Thus, the total variation of the flux g(s ε (, t), c ε (, t)) is uniformly bounded for all t > 0. A straight application of the arguments in Temple [26] or Gimse & Risebro [9] leads to the convergence of a subsequence of (s ε, c ε ) to a weak solution (s(, t), c(, t)) in L 1 loc as ε 0+, proving the existence of solutions for the Cauchy problem for the non-gravitation model. 3 The model with gravity 3.1 Assumptions, Preliminaries and Main Result For the gravitation model (1.3), we assume that the oil mobility function λ(s, c) satisfies (3.1) λ(0, c) = 1, λ o (1, c) = 0, for all c, and for every (s, c) D, we have (3.2) λ(s, c) 0, λ s (s, c) 0, λ ss (s, c) 0. We defined the functions (3.3) G(s, c) = F (s, c) s and Ψ(s, c) = s 0 G ( s, c) s d s. If 0 < K g 1, then the flux F (s, c) > 0 for 0 < s 1, and the model is equivalent to the non-gravitation model. For the rest, we assume K g > 1, such that F (s, c) < 0 for small values of s. See Figure 12 for an illustration for the plots of the three functions s F (s, c), s G(s, c), and s Ψ(s, c) for a given c, if K g > 1. By the assumptions on f(s, c) and λ(s, c), it is easy to verify that the function G(s, c) has the following properties: G(0, c) = 0, G(1, c) = 1, and G s (0, c) < 0 for every c. For every c, there exist two values s m (c) and s M (c) (depending on c) with 0 < s m < s M < 1, such that the mapping s G(s, c) attains the minimum and maximum values G m (c) and G M (c), respectively. Also, we have G ss (s m (c), c) > 0 and G ss (s M (c), c) < 0. G s < 0 for s (0, s m ) (s M, 1), and G s > 0 for s (s m, s M ). For every c, there exists a unique value s 0 (c) (0, 1) such that G(s 0 (c), c) = 0, and another unique value s 1 (c) (0, 1) such that G(s 1 (c), c) = 1, with s 0 < s 1. 17

18 Figure 12: The function F (s, c), G(s, c) and Ψ(s, c) for a given c. See the middle plot in Figure 12 for an illustration of these notations. The definition of a weak solution of (1.1)-(1.2) is similar to that of the non-gravitation model. Definition 3.1. The functions (s, c) are called a weak solution of (1.3)-(1.2), if the equation is satisfied in a distributional sense, such that (3.4) (3.5) 0 0 (sφ t + F (s, c)φ x ) dx dt + (scφ t + cf (s, c)φ x ) dx dt + φ(x, 0) s(x) dx = 0, φ(x, 0) s(x) c(x) dx = 0, for every test function φ C 1 0. Here is the second main Theorem, for the gravitation model. Theorem 3.2. Assume that the initial data ( s, c) have bounded total variation. Then, the Cauchy problem (1.3)-(1.2) admits a weak solution (s(t, x), c(t, x)). We will prove this Theorem in the rest of this section. The proof follows the same setting as the one for the non-gravitation model, with different details. The solutions of the Riemann problems are slightly different. Furthermore, the functional to measure the strength for c- waves will be modified. Finally, the proof of interaction estimates is more complicated. Once we obtain these estimates, the convergence of the front tracking approximate solutions and the existence of solution of the Cauchy problem for the gravitation model follow in a same way. Below, we provide a brief proof, explaining only the new details for this model. 3.2 Basic Analysis and Solutions of Riemann problems Replacing the function f(s, c) with F (s, c), the eigenvalues and eigenvectors are computed in the same way as in Section 2.2. Due to the more complicated shape of the function s F, 18

19 Figure 13: Typical integral curves of the eigenvectors for the c-family, the gravitation model the integral curves of the eigenvectors for the c-family are different, see Figure 13 for a typical plot. The c-family is linearly degenerate everywhere. Furthermore, there are two curves in the domain with F s (s, c) = F (s, c)/s where the system is parabolic degenerate. These are the curves that passes through the horizontal tangent points in the c-integral curves, one in the region G < 0 and the other in G > 0. The s-waves are waves that connects two states with the same c-value. The jumps in the c-family travel with Rankine-Hugoniot speed (2.11) after replacing f with F. The c-wave path in the plots of F L (s), F R (s) will lie on a straight line through the origin, similar to the situation in Figure 3. By a same argument, the implicit Riemann problem using F (s, c) can be replaced by a non-implicit one using G(s, c) as the flux function. The c-wave paths are horizontal lines in the s G plane. The construction of the Riemann problem relies on the sets I (s L, G L ) and I + (s R, G R ), as in Section 2.3. For the gravitation model, the sets I (s L, G L ) and I + (s R, G R ) are illustrated in Figure 14 for all cases. We have [0, s m ], if 0 s L s m, [0, s L ] {s L }, if s m < s L s 0, (3.6) I (s L, G L ) = {s L }, if s 0 < s L < s 1, {s L } [ s L, 1], if s 1 s L < s M, [s M, 1], if s M s L 1. and {s R } [ s R, s M ], if 0 s R < s m, (3.7) I + (s R, G R ) = [s m, s M ], if s m s R s M, [s m, s R ] {s R }, if s M < s R 1. 19

20 G L 1 0 s m 0 0 s L s L s L s L s L 1 s M 1 G R 1 0 s R s R s M s m s M s m s R s R Figure 14: The sets I (s L, G L ) (above) and I + (s R, G R ) (below) for the gravitation model. Note that, G L (s) is decreasing for s I (s L, G L ) and G R (s) is increasing for s I + (s R, G R ). Therefore, the minimum jump path exists and is unique, and the Riemann problem has a unique solution. The following technical Lemma is immediate. Lemma 3.3. For any given c 1 c 2, the graphs of the mappings s g(s, c 1 ) and s g(s, c 2 ) can have at most two intersection points that satisfy the conditions in (2.16), one for g < 0 and the other for g > 0. A totally similar analysis as those for the non-gravitation model reveals that, the admissible and stable c-wave path for g 0 remains the same, while for g 0 it is symmetric to the part where g 1. See Figure 15 for details. The observation in Remark 2.6 on admissible c-shock path also holds for the gravitation model. The front tracking algorithm follows in the same way as in Section 3.3, after replacing g with G. Furthermore, the set G now would also include the minimum values G(s m (c i ), c i ). 3.3 Definitions of Wave Strengths and Interaction Estimates The strength of an s-wave is defined using the function Ψ in (3.3). Let (s L, ĉ) and (s R, ĉ) be the left and right state of an s-wave. Its strength is measured as (3.8) s = Ψ(s L, ĉ) Ψ(s R, ĉ). 20

21 G = 0 G L G R G R G L Figure 15: The path for admissible and stable c-wave path, in the region where g 0. We now modify the functional for measuring the strength of c-waves. Let (s L, c L ) and (s R, c R ) be the left and right states of a c-wave, where G(s L, c L ) = G(s R, c R ). Also we denote G L (s) = G(s, c L ) and G R (s) = G(s, c R ). In the region where G 0, the values (ŝ, Ĝ) and (M L, M R ) are defined in the same way as in Section 2.4 for the non-gravitation model, after replacing g with G. In the region where G 0, we locate the values (ŝ, Ĝ ) as follows. If (2.16) holds, then we let ŝ be the intersection point, and Ĝ = G L (ŝ ) = G R (ŝ ). Otherwise, ŝ will be the point where G L (s) or G R (s) attains its minimum values, whichever has the larger value, i.e., { (3.9) (ŝ, Ĝ ) (s L = m, G L m), if G L m G R m, (s R m, G R m), if G L m < G R m. The corresponding non-negative values M L, M R are defined as: (3.10) M L = Ĝ G L m, M R = Ĝ G R m. We are now ready to define the strength for the c-wave. If G L (s L ) = G R (s R ) 0, then the wave strength is: (3.11) 4M L + 2M R + 2M L + 2M R, if (G L ) (s L ) 0 and (G R ) (s R ) 0, 4M L + 2M R + 2M L + 4M R, if (G L ) (s L ) 0 and (G R ) (s R ) 0, c = 4M L + 2M R + 4M L + 2M R, if (G L ) (s L ) > 0 and (G R ) (s R ) 0, (unstable), if (G L ) (s L ) > 0 and (G R ) (s R ) < 0. If G L (s L ) = G R (s R ) 0, then the wave strength is: (3.12) 4M L + 2M R + 2M L + 2M R, if (G L ) (s L ) 0 and (G R ) (s R ) 0, 4M L + 2M R + 2M L + 4M R, if (G L ) (s L ) 0 and (G R ) (s R ) 0, c = 4M L + 2M R + 4M L + 2M R, if (G L ) (s L ) > 0 and (G R ) (s R ) 0, (unstable), if (G L ) (s L ) > 0 and (G R ) (s R ) < 0. 21

22 Note that, if G = 0 for the c-wave, then we must have (G L ) (s L ) > 0 and (G R ) (s R ) > 0, and c = 4M L + 2M R + 4M L + 2M R from both (3.11) and (3.12). We now study the interaction estimates. We only need to consider the interactions between an s-wave and a c-wave. There are three cases. Case 1. We consider that the three states of the two incoming waves lie in the region G 0. In this region, the definition in (3.11) is equivalent to (2.21), except that an additional constant term is added. Therefore, the interaction estimate remains the same as those for the equation without gravitation force. Thus, total wave strength is non-increasing at such interactions. Case 2. We consider that the three states of the two incoming waves lie in the region G 0. Similarly, we observe that the definition in (3.12) is also equivalent to (2.21), but symmetric about the s-axis. Again, the total wave strength is non-increasing at such interactions. Case 3. We consider the case where the three states of the two incoming waves lie in different regions of G 0 and G 0. Since the G value is constant along the c-wave, then the incoming s-wave must cross the line G = 0. We first claim that, two neighboring waves, with a c-wave on the left and an s-wave on the right while the s-wave crosses the line G = 0, are non-approaching. Indeed, if the c-wave path has G < 0, then s-wave must have positive wave speed. Similarly, if the c-wave path has G > 0, then the s-wave must also have positive wave speed. Since the c-wave is stationary, the two waves are non-approaching and can never interact. It remains to check the case for two neighboring waves, with a c-wave on the right and an s-wave on the left, while the s-wave crosses the line G = 0. It is easy to verify that these waves are approaching. Let (s L, c L ), (s m, c m ), (s R, c R ) denote the three states of the two incoming waves, where (s L, c L ) (s m, c m ) is an s-wave and (s m, c m ) (s R, c R ) is a c-wave. Let (s Lo, c Lo ) be the point where the s-wave crosses the line G = 0. Note that we have c L = c m = c L 0. We claim that the solution of the Riemann problem at the interaction is equivalent to the following three-step Riemann solver, illustrated in Figure 16 for a typical situation: (i) We solve the interaction with (s Lo, c Lo ), (s m, c m ), (s R, c R ) as the three states for the incoming waves. This interaction belongs to Case 1, where we proved that the total wave strength is non-increasing. We let (s Mo, c M 0 ) denote the out-going intermediate state. Note that c M 0 = c R. (ii) We solve the interaction with incoming states (s L, c L ), (s Lo, c Lo ), (s Mo, c M 0 ). This interaction belongs to Case 2, where we proved that the total wave strength is also nonincreasing. We let (s M, c M ) denote the right state of the outgoing v-front. Note that c M = c M 0 = c R. (iii) We treat the two out-going s-fronts from the previous two steps, and merge them by solving a Riemann problem for a scalar conservation with flux G R, with (s M, c M ), (s R, c R ) (and c M = c R ) as the left and right states, to generate the outgoing s-fronts. The total wave strength is non-increasing with this merging, which is the same as the interaction between two s-waves. 22

23 G L G R G L G R G L G R R m L M = R m L 0 M 0 + R m L M L 0 M 0 Figure 16: Wave interaction for Case 3, a typical example. Indeed, it suffices to show that the location of the outgoing c-wave is equivalent in both solvers. We first observe that, for any Riemann data (s L, c L ), (s R, c R ), since the set I (s L, G L ) in (3.6) contains only G values that lie on the same side of the line G = 0 as G L = G(s L, c L ), then the c-wave path must also lie on the same side of G = 0 as G L. Then, for this case, after interaction, the c-wave path is moved to the other side of G = 0, i.e., the same side where (s L, c L ) lies. Furthermore, we observe that, for a fixed left state (s L, c L ), the c-wave path is the same for the two cases: (i) (s R, c R ) strictly on the other side of G = 0, and (ii) (s R, c R ) lies on the line G = 0. Thus, we conclude that the outgoing c-wave is the same for the normal Riemann problem and our three-step Riemann solver. Finally, since the totally wave strength is non-increasing in each of the three steps, it remains non-increasing after the interaction. This estimate gives a uniform bound on the total wave strength in all time, and in term it gives the convergence of the front tracking approximate solutions and the existence of weak solution for the Cauchy problem for the gravitation model. This completes the proof of Theorem Concludinemarks We have several concluding remarks. (1). It would be interesting to extend the result to a multi-component model where c is a vector in R n 1, such that (1.1) (or (1.3)) is an n n system. In the Lagrangian coordinate the system still becomes decoupled, and (1.6) holds. We have n 1 decoupled equations, one for each component c i. The first equation with s is the only coupled equation that connects s with the c values. We denote these linearly degenerate families as the c i -families for i = 1, 2,, n 1. Thanks to this decoupling property, the c i -waves will never intersect with each other, not even among c i -families with different i-values. Much of the analysis could be carried out in a very similar way as in this paper, and the same results hold. (2). The functionals designed here to measure wave strength, in (2.21) and (3.11)-(3.12), are unfortunately very custom-made. It is rather hard to extend these nonlinear singular mappings to general flux functions. It is highly desirable, if possible, to device a more flexible definition that allows such extensions. 23

24 (3). The Remark 2.6 on the admissible c-shock path is valid for both non-gravitation and gravitation models. In fact, this can be viewed as a generalization of the Oleinik entropy condition [22] for scalar conservation laws with non-convex flux functions. Observe that if (s) (s) this coincides with the Oleinik (E) condition, which is equivalent to the vanishing viscosity limit. We conjecture that Remark 2.6, after some modification for general flux functions, holds for solutions constructed by the minimum-jump condition. Thus, for scalar conservation laws with discontinuous flux, we have three entropy conditions: (i) the minimum jump condition, (ii) the generalized Oleinik (E) condition, and (iii) the vanishing viscosity limit. We will show that these three conditions are equivalent in a forthcoming paper. (4). The uniqueness and continuous dependence in L 1 for these weak solutions is still open. Since the front tracking approximate solutions satisfy Kruzhkov entropy inequality, a suitable definition of the entropy weak solution could be defined for both models. A classic variable-doubling technique with mollifiers [19] could be employed to yield the uniqueness and the continuous dependence result. The detailed analysis will be carried out in a forthcoming paper. References [1] D. Amadori, W. Shen, The slow erosion limit in a model of granular flow, Arch. Rational Mech. Anal., 199 (2011), pp [2] S. Bianchini, A. Bressan, Vanishing viscosity solutions of nonlinear hyperbolic systems, Ann. of Math. 161 (2005), pp [3] A. Bressan, Hyperbolic Systems of Conservation Laws. The One Dimensional Cauchy Problem. Oxford University Press, [4] P.G. Bedrikovetsky, Mathematical Theory of Oil and Gas Recovery. Kluwer Academic Publishers, London, [5] S.E. Buckley, M. Leverett, Mechanism of fluid displacement in sands. Transactions of the AIME, 146 (1942), pp [6] C.M. Dafermos, Polygonal approximation of solution to the initial value problem for a conservation law, J. Math. Anal. Appl., 38 (1972), pp [7] O. Dahl, T. Johansen, A. Tveito, R. Winther, Multicomponent chromatography in a two phase environment. SIAM J. Appl. Math. 52 (1992), pp [8] T. Gimse, N.H. Risebro, Riemann problems with a discontinuous flux function. In Proc. Third Internat. Conf. on Hyperbolic Problems. Theory, Numerical Method and Applications. (B. Engquist, B. Gustafsson. eds.) Studentlitteratur/Chartwell-Bratt, Lund- Bromley, (1991), pp [9] T. Gimse, N.H. Risebro, Solution of the Cauchy problem for a conservation law with discontinuous flux function. SIAM J. Math. Anal. 23 (1992), pp

25 [10] J. Glimm, Solutions in the large for nonlinear hyperbolic systems of equations, Comm. Pure Appl. Math., 18 (1965), pp [11] H. Holden, N. H. Risebro, Front Tracking for Hyperbolic Conservation Laws, Applied Mathematical Sciences, 152, Springer, New-York, [12] E. Isaacson, B. Temple, Analysis of a singular hyperbolic system of conservation laws. J.Differential Equations, 65 (1986), pp [13] E. Isaacson, B. Temple, Nonlinear resonance in inhomogeneous systems of conservation laws. Mathematics of nonlinear science (Phoenix, AZ, 1989), Contemp. Math., 108, Amer. Math. Soc., Providence, RI, (1990), pp [14] T. Johansen, A. Tveito, R. Winther, A Riemann solver for a two-phase multicomponent process. SIAM J. Sci. Statist. Comput. 10 (1989), pp [15] T. Johansen, R. Winther, The solution of the Riemann problem for a hyperbolic system of conservation laws modelling polymer flooding. SIAM J. Math. Anal., 19 (1988), pp [16] T. Johansen, R. Winther, Mathematical and numerical analysis of a hyperbolic system modeling solvent flooding. SIAM J. Math. Anal., 20 (1989), pp [17] B. Keyfitz, H. Kranzer, Asystem of non-strictly hyperbolic conservation laws arising in elasticity theory, Arch. Rational Mech. Anal., 72 (1980), pp [18] C. Klingenberg, N. H. Risebro, Stability of a resonant system of conservation laws modeling polymer flow with gravitation, J. Differential Equations, 170 (2001), pp [19] S. Kruzhkov, First-order quasilinear equations with several space variables. Math. USSR Sb. 10 (1970), pp [20] P.D. Lax, Hyperbolic systems of conservation laws II. Comm. Pure Appl. Math. 10 (1957), pp [21] T.P. Liu, The entropy condition and the admissibility of shocks. Journal of Mathematical Analysis and Applications 53 (1976), pp [22] O. Oleinik, Uniqueness and stability of the generalized solution of the Cauchy problem for a quasilinear equation, Uspehi Mat. Nauk. 14 (1959) no. 2 (86), pp (Russian). English Translation in Amer. Math. Soc. Transl. Ser. 2, 33 (1964), pp [23] W. Shen, Slow Erosion with Rough Geological Layers. SIAM J. Math. Anal., 47 (2015), pp [24] W. Shen and T. Y. Zhang, Erosion profile by a global model for granular flow, Arch. Rational Mech. Anal. 204 (2012), pp

26 [25] B. Temple, Stability and decay in systems of conservation laws. In Proc. Nonlinear Hyperbolic Problems, St. Etienne, France, C. Carasso, P. A. Raviart, D. Serre, eds., Springer- Verlag, Berlin, New York, [26] B. Temple, Global solution of the Cauchy problem for a class of 2x2 non-strictly hyperbolic conservation laws, Adv. in Appl. Math., 3 (1982), pp [27] A. Tveito, R. Winther, Existence, uniqueness, and continuous dependence for a system of hyperbolic conservation laws modeling polymer flooding. SIAM J. Math. Anal. 22 (1991), no. 4, pp [28] D. Wagner, Equivalence of the Euler and Lagrangian equations of gas dynamics for weak solutions. J. Differential Equations, 68 (1987), pp