Electronic Journal of Differential Equations, Vol. 2020 (2020), No. 07, pp. 1–15. ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu PASSING TO THE LIMIT ON SMALL PARAMETERS FOR GENERALIZED VISCOUS CAHN-HILLIARD TYPE EQUATIONS WITH NONLINEAR SOURCE BUI LE TRONG THANH, NGUYEN NGOC QUOC THUONG Communicated by Jesus Ildefonso Diaz Abstract. We study the well-posedness of the generalized viscous Cahn- Hilliard equation with nonlinear source term. Then, we analyze the singu- lar limits when the relaxed terms vanish. In the sense of Young measures, we obtain the measure-valued solution of a forward-backward parabolic type equation. 1. Introduction We study the forward-backward parabolic problem ut = ∆ϕ(u) + f(u) in Ω× (0, T ) =: QT u = 0 on ∂Ω× (0, T ) u = u0 in Ω× {0} , (1.1) by considering the limit of solutions of the generalized viscous Cahn-Hilliard prob- lems ut = ∆[ϕ(u)− ε∆u+ δut] + f(u) in Ω× (0, T ) =: QT u = ∆u = 0 on ∂Ω× (0, T ) u = u0 in Ω× {0} (1.2) where Ω is a smooth bounded subset of RN (N ≤ 3), ε > 0, and δ > 0. We use the the following assumptions: (H1) ϕ ∈ C2(R), ϕ(0) = 0, f ∈ C1(R), f(0) = 0; (H2) ϕ′(s) ≥ −C0, C0 ≥ 0; (H3) ϕ(s)s ≥ C1Φ(s)− C2 ≥ −C3 where C1, C2, C3 ≥ 0 and Φ(s) = ∫ s 0 ϕ(r)dr; (1.3) (H4) There exist C1 > 0, C2 ∈ R such that f2(s) ≤ C1Φ(s) + C2; 2010 Mathematics Subject Classification. 35B25, 35K55, 35R25, 28A33, 35D99. Key words and phrases. Generalized Cahn-Hilliard equation; singular limits; forward-backward parabolic equations; Young measure. c©2020 Texas State University. Submitted July 7, 2019. Published January 13, 2020. 1 2 B. L. T. THANH, N. N. Q. THUONG EJDE-2020/07 (H5) There exist a δ > 0 such that for all u ∈ L2(Ω), ‖f(u)‖‖u‖ ≤ δ ∫ Ω Φ(u)dx+ Cδ; (H6) There exist C1, C2 > 0 such that ‖f ′(u)‖2 ≤ C1 ∫ Ω Φ(u)dx+ C2, where ‖ · ‖ denotes the usual norm in L2(Ω). Note that when ε = 0 and δ = 0, problem (1.2) becomes (1.1). Letting ε > 0 and δ = 0 in (1.2) leads to the generalized Cahn-Hilliard equation ut = ∆[ϕ(u)− ε∆u] + f(u). (1.4) Depending on the choice of f , we get the corresponding equation which was widely investigated in the literature. For example, in the case f(s) = −cs with c > 0, equation (1.4) is known as the Cahn-Hilliard-Oono equation which is an application in the phase separation process (see [14]). If f(s) = αs(1 − s) and α > 0, then (1.4) has an application in biology, in particular, in models wound healing and tumor growth (see [10]). The well-posedness of equation (1.4) was studied in [12] with the Dirichlet boundary condition, and in [6] with the Neumann boundary condition. In these articles, they also gave the asymptotic behavior of solution in terms of finite-dimensional attractors. The case ε = 0 and δ > 0 leads to the equation ut = ∆[ϕ(u) + δut] + f(u). (1.5) This equation arises as a model for populations with the tendency to form groups which was studied by Padron (see [15]). The model of aggregating population with a migration rate determined by ϕ, and total birth and mortality rates characterized by f . He showed that the aggregating mechanism induced by ϕ allows the survival of a species in danger of extinction. For more information on the application of equation (1.5), we refer to [15] and the references therein. Now taking into account ε = 0 in equation (1.4) or δ = 0 in equation (1.5), we obtain the forward-backward parabolic type equation ut = ∆ϕ(u) + f(u). (1.6) This equation has a variety of applications in biology such as aggregating popula- tions (see [8, 11, 7] and references therein). In aggregation of population models, the nonlinearity ϕ may be increasing or decreasing therefore, the standard initial boundary value problems for (1.6) are in general ill-posed. That is the reason for studying the regularized problem of equation (1.6) by adding some regular terms. It is worth to mention that in the case of vanishing source term, the forward-backward parabolic equation ut = ∆ϕ(u) (1.7) has no weak solution if the general initial data is considered. Often a higher order term is added to the right-hand side to regularize the equation. There are mainly two classes of additional terms which can be found in the mathematical literature, which, e.g. in case of equation (1.7) reduce to: (i) ε∆[ψ(u)]t, with ψ′ > 0, leading to third order pseudo-parabolic equations (ε > 0 being a small parameter; see for example [1, 13, 17]); EJDE-2020/07 CAHN-HILLIARD EQUATION 3 (ii) −ε∆2u, leading to fourth-order Cahn-Hilliard type equations (see for ex- ample [16, 18] and references therein). It is remarkable, taking advantage of the cubic-like growth of ϕ at infinity, which gives rise to better estimates of the family {uε} of solutions of the regularized problem, they proved the existence of solutions in the sense of Young measures. Moreover, it is worthy to mention that the Cahn-Hilliard equation with a loga- rithmic nonlinear term has been investigated by many authors (see [5, 4, 3] and references therein). In all these references, the logarithmic potential is approx- imated by regular ones. Then, when passing to the limit in the approximated problems, it is difficult to prove that the limit of the order parameter remains in (−1, 1). We refer readers to the survey [2] for more applications and other aspects of the Cahn-Hiliard equation. In light of the above considerations, we first prove the existence and uniqueness of solution of problem (1.2) by Galerkin approximation and compactness method which are the same approaches as in [12, 6]. Secondly, we give the rigorous analysis of the convergence of a family {uε,δ} of solutions of (1.2) as δ → 0 to obtain the existence of solution of (1.1). Finally, we investigate the convergence of a family {uε} of solutions of (1.1). Because of lacking of the compactness, we prove the appearance of measure-valued solution of (1.1). Actually, taking of the advantage of the growth of nonlinearities ϕ, f , we only have the L2 uniform bounded estimate on solutions. Our approach is almost the same as in [18, 16]. Remark 1.1. If we choose ϕ(u) = u3 − u and f(u) = αu(1− u) with α > 0, then ϕ, f will satisfy the assumptions (H1)–(H6). The choice of ϕ, f is widely used in the literature. This article is organized as follows: we introduce our problem and some assump- tions in Section 1, we study the well-posedness of problem (1.2) in Section 2 and give the rigorous convergence of the solutions of (1.2) as δ vanishing in Section 3. Finally, we investigate the existence of measure-valued solution of (1.1) in Section 4. 2. Well-posedness of problem (1.2) In what follows, the symbols c, c′, c′′, ci (i ≥ 0) will denote positive constants and may vary from line to line. Q : R+ → R+ will be a positive increasing monotone function and may also vary from line to line or even in a same line. 2.1. Mathematical formulation and results. In this section, we study the well- posedness of problem (1.2). By setting A := −∆, the first equation of problem (1.2) is written in the form ut +A ( ϕ(u) + εAu+ δut ) − f(u) = 0. (2.1) Operator A : D(A) → L2(Ω) is a strictly positive self-adjoint linear with compact inverse on L2(Ω), and domain D(A) = H2(Ω) ∩H1 0 (Ω). In this article, we denote( ·, · ) as an usual scalar product in L2(Ω) and set ‖ · ‖−1 = ‖A−1/2 · ‖. In general, we introduce the family of Hilbert spaces H2s = D(As), ∀s ∈ R with scalar products ((u, v))2s := (Asu,Asv), ∀u, v ∈ Hs. 4 B. L. T. THANH, N. N. Q. THUONG EJDE-2020/07 Definition 2.1. For any interval (0, T ), a function uε,δ(x, t) = u(x, t) is called a solution of problem (1.2) if (u, ut) ∈ L∞(0, T ;D(A)× L2(Ω)) and ut +A (ϕ(u) + εAu+ δut)− f(u) = 0 in D(A−1), a.e. t ∈ (0, T ), and u(0) = u0(x) ∈ D(A), for a.e. x ∈ Ω. Theorem 2.2. Let assumptions (H1)–(H6) hold and u0 ∈ H2(Ω) ∩H1 0 (Ω). Then problem (1.2) admits a unique global solution as in definition 2.1. Moreover, let u1, u2 be two solutions of (1.2) with initial data u0,1 and u0,2 respectively, then there exists a constant c ≥ 0 such that ‖u1(t)− u2(t)‖2−1 + δ‖u1(t)− u2(t)‖2 ≤ ectQ(‖u0,1‖H2 , ‖u0,2‖H2)(‖u0,1 − u0,2‖2−1 + δ‖u0,1 − u0,2‖2) for any t ≥ 0. Proof of Theorem 2.2. We first prove the uniqueness of solution of (1.2). Let u1, u2 be two solutions of (1.2) with initial data u0,1 and u0,2 respectively. We set u = u1 − u2 and u0 = u0,1 − u0,2 and then u satisfies ut +A ( ϕ(u1)− ϕ(u2) + εAu+ δut ) − f(u1) + f(u2) = 0, u = ∆u = 0 on ∂Ω, u(x, 0) = u0(x). (2.2) We multiply (2.2) by A−1u and we have 1 2 d dt ( ‖A−1/2u‖2 + δ‖u‖2 ) + ε‖A1/2u‖2 + ( ϕ(u1)− ϕ(u2), u ) − ( f(u1)− f(u2), A−1u ) = 0. (2.3) Note that from (H2), (ϕ(u1)− ϕ(u2), u) ≥ −C0‖u‖2. Furthermore,∣∣(f(u1)− f(u2), A−1u )∣∣ ≤ ∫ Ω |A−1u||u| ∫ 1 0 |f ′(su1 + (1− s)u2)| ds dx ≤ Q(‖u0,1‖H2 , ‖u0,2‖H2)‖A−1u‖∞‖u‖ ≤ Q(‖u0,1‖H2 , ‖u0,2‖H2)‖u‖2 thank to the continuous embedding H2 ⊂ C(Ω̄). Therefore, d dt ( ‖A−1/2u‖2 + δ‖u‖2 ) + ε‖A1/2u‖2 ≤ Q(‖u0,1‖H2 , ‖u0,2‖H2)‖u‖2. (2.4) From ‖u‖2 = ( A−1/2u,A1/2u ) and Young’s inequality, we have d dt ( ‖A−1/2u‖2 + δ‖u‖2 ) + ε‖A1/2u‖2 ≤ Q(‖u0,1‖H2 , ‖u0,2‖H2)(‖A−1/2u‖2 + δ‖u‖2). (2.5) By Gronwall’s lemma we obtain ‖u1(t)− u2(t)‖2−1 + δ‖u1(t)− u2(t)‖2 ≤ ectQ(‖u0,1‖H2 , ‖u0,2‖H2)(‖u0,1 − u0,2‖2−1 + δ‖u0,1 − u0,2‖2) EJDE-2020/07 CAHN-HILLIARD EQUATION 5 for any t ≥ 0, c ≥ 0. We have the uniqueness and the continuous dependence with respect to initial data. � The existence result relies on a standard approximation - a priori estimates and passage to the limit procedure. The Faedo-Galerkin scheme is as follows. Since A−1 is compact and self-adjoint operator on L2(Ω), there exists an orthonormal basis of L2(Ω) consisting of eigenvectors {ei} of A and the corresponding eigenvalues λi with Dirichlet boundary condition; that is, Aei = λiei, for i = 1, 2, . . . and 0 < λ1 < λ2 ≤ · · · ≤ λk → ∞. It is easy to check that {λ−1/2 i ei} is an orthonormal basis of H1 0 (Ω). For any integer number n ≥ 1, let Vn := span{e1, . . . , en}. We state the approximating problem as follows. Problem Pn: Find tn > 0 and ui ∈ C2([0, tn]) for i = 1, · · · , n such that un := N∑ i=1 ui(t)ei(x), belongs to C2([0, tn], D(A)) and satisfies 〈unt , v〉+ 〈ϕ(un) + εAun + δunt , Av〉 − 〈f(un), v〉 = 0, ∀v ∈ Vn, (2.6) un(0) = un0 , (2.7) where un0 ∈ Vn such that un0 → u0 in D(A). Problem Pn consist of a n-dimensional system of nonlinear ordinary differential equations. By the Cauchy-Lipschitz Theorem, there exists a unique local in time solution un in the maximal interval [0, T ∗). We now derive some a priori estimates that will permit us to prove the existence result by passage the limit as n→∞. The procedure is standard and so we only give a priori estimates in the next subsection. 2.2. A priori estimates. We multiply the first equation of (1.2) (2.1) by A−1u, and integrate over Ω and by parts to obtain 1 2 d dt ( ‖u‖2−1 + δ‖u‖2 ) + ( ϕ(u), u ) + ε‖A1/2u‖2 − ( f(u), A−1u ) = 0. (2.8) Thank to assumptions (H3) and (H5), we have 1 2 d dt ( ‖u‖2−1 + δ‖u‖2 ) + ε‖A1/2u‖2 + c1 ∫ Ω Φ(u)dx ≤ ‖f(u)‖‖u‖+ c2, d dt ( ‖u‖2−1 + δ‖u‖2 ) + c ( ε‖u‖2H1 + ∫ Ω Φ(u)dx ) ≤ c′, c > 0. (2.9) Multiplying (2.1) by u and integrating over Ω, we obtain 1 2 d dt ( ‖u‖2 + δ‖A1/2u‖2 ) + ε‖Au‖2 + (ϕ′(u)∇u,∇u)− (f(u), u) = 0. (2.10) Thank to (H2) and (H5), we have 1 2 d dt ( ‖u‖2 + δ‖A1/2u‖2 ) + ε‖Au‖2 ≤ c0‖A1/2u‖+ ‖f(u)‖‖u‖. (2.11) 6 B. L. T. THANH, N. N. Q. THUONG EJDE-2020/07 Therefore, d dt ( ‖u‖2 + δ‖A1/2u‖2 ) + cε‖u‖2H2 ≤ c1‖A1/2u‖+ c2 ∫ Ω Φ(u)dx+ c3, (2.12) with c > 0. Finally, we take sum of (2.9) and (2.12) times δ1 > 0, where δ1 is small enough, to obtain d dt ( ‖u‖2−1 + δ‖u‖2 + δ1‖u‖2 + δ1δ‖A1/2u‖2 ) + c ( ε‖u‖2H2 + ∫ Ω Φ(u)dx ) ≤ c′, c > 0. (2.13) Note that from (2.13) and Gronwall’s lemma, we have ‖u(t)‖2−1 + δ‖u(t)‖2 + δ1‖u(t)‖2 + δ1δ‖A1/2u(t)‖2 ≤ e−ctQ(‖uo‖H1 0 ) + c′, (2.14) for t > 0 with c > 0. Multiplying equation (2.1) by Au and integrating over Ω, we have (ut + δAut, Au) + ε‖A3/2u‖2 + (Aϕ(u)− f(u), Au) = 0. (2.15) By Holder’s inequality, 1 2 d dt ( ‖A1/2u‖2 + δ‖Au‖2 ) + ε‖A3/2‖2 ≤ c ( ‖Aϕ(u)‖2 + ‖f(u)‖2 ) . (2.16) Since H2(Ω) ⊂ C(Ω̄) with continuous embedding as N ≤ 3 and ϕ, f ∈ C2(R), ‖Aϕ(u)‖2 + ‖f(u)‖2 ≤ Q(‖u‖H2). Thus, d dt ( ‖A1/2u‖2 + δ‖Au‖2 ) ≤ Q(‖A1/2u‖2 + δ‖Au‖2). (2.17) Let y be the solution to the ordinary differential equation y′ = Q(y), y(0) = ‖A1/2u0‖2 + δ‖Au0‖2. Then by the comparison principle, there exists a time T ∗ = T ∗(‖u0‖H2) > 0 such that ‖A1/2u(t)‖2 + δ‖Au(t)‖2 ≤ y(t), t ≤ T ∗. In summary we have δ‖u(t)‖H2 ≤ Q(‖u0‖H2 ), ∀t ≤ T ∗. (2.18) Multiplying (2.1) by A−1ut and have ε 2 d dt ‖A1/2u‖2 + ‖ut‖2−1 + δ‖ut‖2 + (ϕ(u), ut)− ( f(u), A−1ut ) = 0, (2.19) which, by (2.18) and Holder’s inequality, for t ≤ T ∗ yields ε 2 d dt ‖A1/2u‖2 + ‖ut‖2−1 + δ‖ut‖2 ≤ c ( ‖ϕ(u)‖2H1 + ‖f(u)‖2 ) ≤ Q(‖u‖H2) ≤ Q(‖u0‖H2). (2.20) Therefore, ε‖A1/2u(t)‖2 + ∫ T∗ 0 ‖ut‖2−1 + δ‖ut‖2ds ≤ Q(‖u0‖H2), t ≤ T ∗. (2.21) Differentiating (2.1) with respect to time and setting v = ut we have A−1vt + δvt + εAv + ϕ′(u)v −A−1(f ′(u)v) = 0. (2.22) EJDE-2020/07 CAHN-HILLIARD EQUATION 7 Multiplying (2.22) by tv, for t ≤ T ∗ we obtain d dt ( t‖v‖2−1 + δt‖v‖2 ) + εt‖A1/2v‖2 ≤ Q(‖u0‖H2)(t‖v‖2) + ‖v‖2−1 + δ‖v‖2. (2.23) Noting that ‖v‖2 ≤ c‖v‖−1‖A1/2v‖, we have d dt ( t‖v‖2−1 + δt‖v‖2 ) ≤ Q(‖u0‖H2)(t‖v‖2 + δt‖v‖2) + ‖v‖2−1 + δ‖v‖2. (2.24) Integrate (2.24) over (0, t) with t ≤ T ∗; then thanks to (2.21) we have t‖v‖2−1 + δt‖v‖2 ≤ c ∫ t 0 (s‖v‖2 + δs‖v‖2)ds+ ∫ t 0 ‖v‖2−1 + δ‖v‖2ds (2.25) ≤ c ∫ t 0 (s‖v‖2 + δs‖v‖2)ds+Q(‖u0‖H2). (2.26) By Gronwall’s inequality, ‖v‖2−1 + δ‖v‖2 ≤ 1 t Q(‖u0‖H2), 0 < t ≤ T ∗. (2.27) We now multiply (2.22) by v to have d dt ( ‖v‖2−1 + δ‖v‖2 ) + 2ε‖A1/2v‖+ 2 (ϕ′(u)v, v) ≤ 2| ( A−1(f ′(u)v), v ) |. (2.28) Thank to (H2), Young’s inequality and ‖f ′(u)‖ ≤ c1 ∫ Φ(u)dx+ c2, we have d dt ( ‖v‖2−1 + δ‖v‖2 ) + 2ε‖A1/2v‖ ≤ c0‖v‖2 + c‖f ′(u)‖‖v‖−1‖A1/2v‖ ≤ ( c1 ∫ Φ(u)dx+ c2 ) ( ‖v‖2−1 + δ‖v‖2 ) . (2.29) From (2.9), we have ∫ t 0 ∫ Ω Φ(u) dx ds ≤ c‖u0‖2 + c′t+ c′′. (2.30) Using Gronwall’s inequality and (2.29) we have ‖v‖2−1 + δ‖v‖2 ≤ ( ‖v(T ∗)‖2−1 + δ‖v(T ∗)‖2 ) e ∫ t 0 dt ∫ Ω c1Φ(u)dx+c2 . (2.31) Thanks to (2.27), we have ‖v‖2−1 + δ‖v‖2 ≤ ectQ(‖u0‖H2), t ≥ T ∗. (2.32) Again rewritten our equation (2.1) in the following form εAu+ ϕ(u)−A−1f(u) = −A−1v − δv := h. (2.33) It is clearly that from (2.32) ‖h‖ ≤ ectQ(‖u0‖H2), ∀t ≥ T ∗. Multiplying (2.33) by u and integrating over Ω, we have ε‖A1/2u‖2 + (ϕ(u), u) ≤ ‖h‖‖u‖+ ‖f(u)‖‖u‖. (2.34) This implies ε‖A1/2u‖2 + c ∫ Ω Φ(u)dx ≤ ectQ(‖u0‖H2 ) + c′′. (2.35) Multiplying (2.33) by Au and integrating over Ω, we have ε‖Au‖2 + (ϕ′(u)∇u,∇u) ≤ ‖h‖‖Au‖+ ‖f(u)‖‖u‖. (2.36) 8 B. L. T. THANH, N. N. Q. THUONG EJDE-2020/07 This implies ε‖Au‖2 ≤ ‖h‖2 + c0‖A1/2u‖2 + c1 ∫ Ω Φ(u)dx+ c2. (2.37) Adding (2.29) and δ2 times (2.37), where δ2 is small enough, yields ε‖u(t)‖2H2 ≤ ectQ(‖u0‖H2) + c′, ∀t ≥ T ∗, c ≥ 0. (2.38) Combining this with (2.18), we obtain ε‖u(t)‖2H2 ≤ ectQ(‖u0‖H2 ) + c′, ∀t ≥ 0, c ≥ 0. (2.39) From (2.13), we have ε ∫ 1 0 ‖u(t)‖H2dt ≤ c‖u0‖2H1 + c′, (2.40) so that there exists a T ∈ (0, 1) such that ‖u(T )‖H2 ≤ c‖u0‖2H1 + c′. (2.41) If we start from time t = T instead of t = 0, inequality (2.41) holds for T = 1, that is, ‖u(1)‖H2 ≤ c‖u0‖2H1 + c′, c ≥ 0. (2.42) Again from (2.13) and Gronwall’s lemma, we can prove that for any t ≥ 0,∫ t+1 t ‖u(s)‖H2ds ≤ e−ctQ(‖u0‖H1) + c′, c ≥ 0. Hence for every t ≥ 1, there exists a t1 ∈ [t− 1, t] such that ‖u(t1)‖H2 ≤ e−ctQ(‖u0‖2H1) + c′, (2.43) which implies, for t2 ∈ [0, 1], that t = t1 + t2. Thanks to (2.39) and (2.43), ‖u(t)‖2H2 = ‖u(t1 + t2)‖2H2 ≤ ect2Q(‖u(t1)‖H2) + c′ ≤ c1e−c2t2Q(‖u(t1)‖H2) + c3 ≤ c1e−c2t2Q ( e−c ′tQ′(‖u0‖2H2) + c′′ ) + c3 ≤ e−ctQ′(‖u0‖2H1) + c′, that is, ‖u(t)‖H2 ≤ ectQ(‖u0‖H2) + c′, c > 0, t ≥ 0. (2.44) 3. Convergence of solutions of problem (1.2) as δ → 0 In this section, we study the well-posedness of problem (1.2) in the sense of convergence of a family of solutions {uε,δ} of (1.2) as δ → 0. Definition 3.1. Let T > 0, u0 ∈ H1 0 (Ω)∩H2(Ω), ε > 0, by a solution uε of (1.2), we mean a function uε ∈ L∞(0, T ;H1 0 (Ω))∩L2(0, T ;H2(Ω)), uεt ∈ L∞(0, T ;H−1(Ω)), ϕ(uε), f(uε) ∈ L2(QT ) such that∫ T 0 〈uεt, η〉ds+ ∫ T 0 〈ϕ(uε) + εAuε, Aη〉ds− ∫ T 0 〈f(uε), η〉ds = 0, (3.1) for any test function η ∈ C1(0, T ;H1 0 (Ω) ∩ H2(Ω)), and uε(x, 0) = u0(x) for a.e. x ∈ Ω. EJDE-2020/07 CAHN-HILLIARD EQUATION 9 Theorem 3.2. Let (H1)–(H6) hold, and u0 ∈ H2(Ω)∩H1 0 (Ω). Then problem (1.1) admits a unique solution as in Definition 3.1. Proof. We focus mainly on proving the existence result. Thanks to estimates (2.13), (2.14), (2.21), (2.32) and (2.44), there exists a positive constant C independent of δ such that ‖uε,δ‖L∞(0,T :H1 0 (Ω)) ≤ C, (3.2) ‖uε,δ‖L2(0,T :H2(Ω)) ≤ C, (3.3) ‖uε,δt‖L∞(0,T :H−1(Ω)) ≤ C, (3.4) √ δ‖uε,δt‖L∞(0,T :L2(Ω)) ≤ C, (3.5) By standard argument of compactness and Aubin-Lions Lemma, there exist a function uε ∈ L∞(0, T ;H1 0 (Ω)) ∩ L2(0, T ;H2(Ω)), uεt ∈ L∞(0, T ;H−1(Ω)) and a subsequence of {uε,δ} (still denote {uε,δ}) such that • uε,δ converges weakly-star to uε in L∞(0, T ;H1 0 (Ω)), • uε,δ converges weakly to uε in L2(0, T ;H2(Ω)), • uε,δ converges strongly to uε in L2(QT ) and a.e. in QT , • uε,δt converges weakly-star to uεt in L∞(0, T ;H−1(Ω)). Now we are ready to take limit δ → 0 in the weak formulation of solution uε,δ of problem (1.2). For any η ∈ C1(0, T ;H1 0 ∩H2), T > 0, we have∫ T 0 〈uε,δt, η〉ds+ ∫ T 0 〈ϕ(uε,δ)+εAuε,δ+δuε,δt, Aη〉ds− ∫ T 0 〈f(uε,δ), η〉ds = 0. (3.6) Taking δ → 0 in (3.6), using above convergences of {uε,δ} and noting that∣∣ ∫ T 0 ∫ Ω δuε,δtη dx ds ∣∣ = ∣∣ ∫ T 0 ∫ Ω δ1/2δ1/2uεδtη dx ds ∣∣ ≤ δ1/2‖δ1/2uε,δt‖L2(QT )‖η‖L2(QT ) ≤ Cδ1/2 → 0 as δ → 0, which yields the weak formulation (3.1) as in Definition 3.1. It is also easy to prove that uε(x, 0) = u0(x) a.e. in Ω. Concerning the uniqueness, we first observe that for a.e. t ∈ (0, T ), 〈uεt, η〉+ 〈ϕ(uε) + εAuε, Aη〉 − 〈f(uε), η〉 = 0 (3.7) for any η ∈ H1 0 (Ω) ∩H2(Ω). Let u1, u2 be two solutions of (1.1) with initial data u0,1 and u0,2 respectively. We set u = u1−u2 and u0 = u0,1 = u0,2 and then u1, u2 satisfy (3.7). We choose η = A−1u and subtract equations of u1 and u2 to obtain 1 2 d dt ‖A−1/2u‖2 + ε‖A1/2u‖2 + (ϕ(u1)− ϕ(u2), u)− ( f(u1)− f(u2), A−1u ) = 0. (3.8) Note that from (H2) (ϕ(u1)− ϕ(u2), u) ≥ −C0‖u‖2. Furthermore,∣∣ (f(u1)− f(u2), A−1u ) ∣∣ ≤ ∫ Ω |A−1u||u| ∫ 1 0 |f ′(su1 + (1− s)u2)| ds dx ≤ Q(‖u0,1‖H2 , ‖u0,2‖H2)‖A−1u‖∞‖u‖ 10 B. L. T. THANH, N. N. Q. THUONG EJDE-2020/07 ≤ Q(‖u0,1‖H2 , ‖u0,2‖H2)‖u‖2 by the continuous embedding H2 ⊂ C(Ω̄). And thus, we have d dt ‖A−1/2u‖2 + ε‖A1/2u‖2 ≤ Q(‖u0,1‖H2 , ‖u0,2‖H2)‖u‖2. (3.9) From ‖u‖2 = ( A−1/2u,A1/2u ) and Young’s inequality we have d dt ( ‖A−1/2u‖2 ) + ε‖A1/2u‖2 ≤ Q(‖u0,1‖H2 , ‖u0,2‖H2)‖A−1/2u‖2. (3.10) By Gronwall’s lemma we obtain ‖u1(t)− u2(t)‖2−1 ≤ ectQ(‖u0,1‖H2 , ‖u0,2‖H2)‖u0,1 − u0,2‖2−1 for any t ≥ 0, some c ≥ 0. We complete the proof of Theorem 3.2. � 4. Existence of measure-valued solution of problem (1.1) Concerning the well-posedness of problem (1.2), we refer to Section 3 or to Alain Miranville [12]. Set vε(x, t) = ϕ(uε(x, t)) − ε∆uε(x, t), then ut = ∆v + f(u). We state the equivalence of problem (1.2), finding (uε, vε) of uεt = ∆vε + f(uε) in Ω× (0, T ) =: QT uε = vε = 0 on ∂Ω× (0, T ) u = u0 in Ω× {0} (4.1) Theorem 4.1 (Well-posedness of (4.1)). Let (H1)–(H6) hold and u0 ∈ H2(Ω) ∩ H1 0 (Ω). Then problem (4.1) admits a unique global solution uε(·, t) ∈ H2(Ω) ∩ H1 0 (Ω), vε ∈ L2((0, T ), H1 0 (Ω)∩H2(Ω)), uεt ∈ L2(QT ) for all t ≥ 0 in strong sense. Proposition 4.2 (A priori estimates). Let (H1)–(H6) hold and ∫ Ω Φ(u0)dx < ∞. Then the family of solutions {uε, vε}ε>0 which are guaranteed by Theorem 4.1 sat- isfy the following inequalities ‖uε(·, t)‖L2(Ω) ≤ C, ∀t ∈ [0, T ], (4.2)∫ Ω ε|∇uε(x, t)|2dx ≤ C, ∀t ∈ [0, T ], (4.3) ‖vε‖L2(0,T ;H1 0 (Ω)) ≤ C (4.4) where C is a positive constant independent of ε. Next we prove the existence of Young measure solutions of problem (1.1) in the sense of the following definition. For the definition and properties of Young measure, we refer to [19]. Definition 4.3. Let N ≤ 3, u0 ∈ H1 0 (Ω). By a Young measure solution of problem (1.1) in QT we mean a triplet (u, v, τ) such that: (i) u ∈ L2(QT ), ut ∈ L2((0, T ), H−1(Ω)); (ii) v ∈ L2((0, T );H1 0 (Ω)), τ ∈ Y(QT ;P(R)); (iii) for almost every (x, t) ∈ QT it holds u(x, t) = 〈τ(x,t), id 〉R = ∫ R ξ dτ(x,t)(ξ) , (4.5) where id(ξ) := ξ (ξ ∈ R) and τ(x,t) ∈ P(R) denotes the disintegration of τ ; EJDE-2020/07 CAHN-HILLIARD EQUATION 11 (iv) for any ζ ∈ C1([0, T );C1 c (Ω)) and t ∈ (0, T )∫ t 0 ∫ Ω [ u ζs −∇v · ∇ζ + f∗ζ ] (x, s) dx ds = ∫ Ω u(x, t)ζ(x, t) dx− ∫ Ω u0(x) ζ(x, 0) dx , (4.6) where v(x, t) and f∗ satisfy v(x, t) = ϕ∗(x, t) := 〈τ(x,t), ϕ〉R = ∫ R ϕ(ξ) dτ(x,t)(ξ), (4.7) f∗(x, t) := 〈τ(x,t), f〉R = ∫ R f(ξ) dτ(x,t)(ξ) (4.8) for almost every (x, t) ∈ QT . A Young measure solution of problem (1.1) in Q∞, which exists in QT for any T ∈ (0,∞), is said to be global. Proposition 4.4. Let assumptions in Theorem 4.1 hold. Then there exist functions (u, v) ∈ L2(QT )×L2(0, T ;H1 0 (Ω)) and subsequences {uεk}, {vεk} of {uε}, {vε} (still denote {uε}, {vε} for convenience) such that: (i) uε converges weakly to u in L2(QT ); (ii) vε converges weakly to v in L2(0, T ;H1 0 (Ω)); (iii) The sequence of Young measure {τk} associated with the sequence {uεk}. There exists a Young measure τ such that τk → τ narrowly in the sense of Definition 5.5; (iv) We have u(x, t) = 〈τ(x,t), id 〉R = ∫ R ξ dτ(x,t)(ξ) . (4.9) Moreover for any φ ∈ C1(R) there exists a function φ∗ such that φ∗(x, t) := 〈τ(x,t), φ〉R = ∫ R φ(ξ) dτ(x,t)(ξ) . (4.10) Our main result is as follows. Theorem 4.5. Let assumptions in Theorem 4.1 hold. Then problem (1.1) admits a global Young measure solution as in Definition 4.3. Proof of Proposition 4.2. Multiplying the two-sides of the first equation of (1.1) by vε and integrating over Ω, yields∫ Ω uεt[ϕ(uε)− ε∆uε]dx = ∫ Ω ∆vεvε + f(uε)vεdx, d dt (∫ Ω Φ(uε) + ε 2 |∇uε|2dx ) + ∫ Ω |∇vε|2dx = ∫ Ω f(uε)vεdx. Using Poincare’ inequality for vε and Hölder’s inequality for the integral in the right-hand side, we obtain∫ Ω Φ(uε) + ε 2 |∇uε|2dx+ ∫ t 0 ∫ Ω |∇vε|2dx = ∫ t 0 ∫ Ω f(uε)vεdx+ C(‖u0‖H1 0 (Ω)),∫ Ω Φ(uε) + ε 2 |∇uε|2dx+ ∫ t 0 ∫ Ω |∇vε|2dx ≤ C1 ∫ t 0 ∫ Ω f2(uε)dx+ C2. 12 B. L. T. THANH, N. N. Q. THUONG EJDE-2020/07 Now using assumption (H4),∫ Ω Φ(uε) + ε 2 |∇uε|2dx+ ∫ t 0 ∫ Ω |∇vε|2dx ≤ C ∫ t 0 ∫ Ω Φ(uε)dx+M. By Gronwall’s inequality, we obtain∫ Ω Φ(uε)dx ≤ C(u0, T,Ω). Then the assumption on the growth of Φ implies inequality (4.2), ‖uε‖L2(Ω) ≤ C. From this, we can easily obtain the remaining estimates in Proposition 4.2 � Proof of Proposition 4.4. The statements in this Proposition follow directly from estimates (4.2), (4.4) and the Fundamental Theorem of Young measure. � Proof of Theorem 4.5. Firstly, we prove the statement (4.7). Indeed, for any η ∈ C∞c (QT ),∣∣ ∫ QT [ϕ(uεk)− v]η dx dt ∣∣ = ∣∣ ∫ QT [ϕ(uεk)− vεk + vεk − v]η dx dt ∣∣ ≤ ∣∣ ∫ QT [ϕ(uεk)− vεk ]η dx dt ∣∣+ ∣∣ ∫ QT [vεk − v]η dx dt ∣∣ = ∣∣ ∫ QT εk∆uεkη dx dt ∣∣+ ∣∣ ∫ QT [vεk − v]η dx dt ∣∣ = ∣∣ ∫ QT εk∇uεk∇η dx dt ∣∣+ ∣∣ ∫ QT [vεk − v]η dx dt ∣∣ → 0 as k →∞,∣∣ ∫ QT [vεk − v]η dx dt ∣∣→ 0 by Proposition 4.4 (ii), ∣∣ ∫ QT εk∇uεk∇η dx dt ∣∣ = √ εk‖ √ εk∇uεk‖L2(QT )‖∇η‖L2(QT ) → 0 by the uniform bounded estimate (4.3) of Proposition 4.2 Secondly, by the well-posedness Theorem 4.1 for problem (1.2), with εk instead of ε, and with the initial datum u0εk := u0. For any function ζ ∈ C1([0, T );C1 c (Ω)) and t ∈ (0, T ) we have∫ t 0 ∫ Ω [ (uεk)s ζ −∆vζ ] (x, s) dx ds = ∫ Ω f(uεk(x, t))ζ(x, t) dx. Integration by parts yields∫ t 0 ∫ Ω [ uεk ζs −∇v · ∇ζ + f(uεk)ζ ] (x, s) dx ds = ∫ Ω [uεkζ](x, t)dx− ∫ Ω u0(x)ζ(x, 0). By Proposition 4.4 and the above statement of weak convergence of {vεk}, we send k →∞ to get the weak formulation (4.6),∫ t 0 ∫ Ω [ u ζs −∇v · ∇ζ + f∗ζ ] (x, s) dx ds EJDE-2020/07 CAHN-HILLIARD EQUATION 13 = ∫ Ω u(x, t)ζ(x, t) dx− ∫ Ω u0(x) ζ(x, 0) dx , where f∗ is barycenter of the nonlinearity f which is defined as in (4.8). This completes the proof. � 5. Appendix Concerning Young measures on Q× R (e.g., see [9, 19] and references therein). Definition 5.1. By a Young measure on Q × R we mean any positive Radon measure τ such that τ(E × R) = |E| (5.1) for any Lebesgue measurable set E ⊆ Q. The set of Young measures on Q×R will be denoted by Y(Q;R). If f : Q→ R is Lebesgue measurable, the Young measure associated to f is the measure τ ∈ Y(Q;R) such that τ(E × F ) = |E ∩ f−1(F )| (5.2) for any Lebesgue measurable set E ⊆ Q and any Borel set F ⊆ R. Remark 5.2. In view of (5.2), if τ is the Young measure associated to a Lebesgue measurable function f : Q → R, for any τ -integrable function ψ : Q × R → R̄ we have ∫ Q×R ψ dτ = ∫∫ Q ψ(x, t, f(x, t)) dx dt . (5.3) Proposition 5.3. Let τ ∈ Y(Q;R). Then for almost every (x, t) ∈ Q there exists a measure τ(x,t) ∈ P(R), such that for any function ψ : Q × R → R bounded and continuous: (i) the map (x, t)→ 〈τ(x,t), ψ(x, t, ·)〉R = ∫ R ψ(x, t, ξ) dτ(x,t)(ξ) is Lebesgue measurable; (ii) it holds 〈τ, ψ〉Q×R := ∫ Q×R ψ dτ = ∫∫ Q 〈τ(x,t), ψ(x, t, ·)〉R dx dt = ∫∫ Q dx dt ∫ R ψ(x, t, ξ) dτ(x,t)(ξ) . (5.4) Therefore, every τ ∈ Y(Q × R) can be identified with the associated family {τ(x,t) : (x, t) ∈ Q}, which is called the disintegration of τ . Remark 5.4. If τ is the Young measure associated to a Lebesgue measurable function f : Q→ R, equalities (5.3)-(5.4) imply ψ(x, t, f(x, t)) = 〈τ(x,t), ψ(x, t, ·)〉R = ∫ R ψ(x, t, ξ) dτ(x,t)(ξ) (5.5) for almost every (x, t) ∈ Q; where ψ ∈ BC(Q×R) and {τ(x,t)} is the disintegration of τ . In this case τ(x,t) = δf(x,t) for almost every (x, t) ∈ Q , where δP denotes the Dirac mass concentrated in P ∈ R. 14 B. L. T. THANH, N. N. Q. THUONG EJDE-2020/07 Definition 5.5. Let {τn} ⊆ Y(Q;R), τ ∈ Y(Q;R) (n ∈ N). We say that τn → τ narrowly in Q× R, if ∫ Q×R ψ dτn → ∫ Q×R ψ dτ (5.6) for any function ψ : Q × R → R bounded and measurable, such that ψ(x, t, ·) is continuous for almost every (x, t) ∈ Q. Theorem 5.6. Let {fn} be a bounded sequence in L1(Q), and {τn} the sequence of associated Young measures. Then: (i) there exist subsequences {fk} ≡ {fnk } ⊆ {fn}, {τk} ≡ {τnk} ⊆ {τn} and a Young measure τ on Q× R such that τk → τ narrowly in Q× R; (ii) for any ρ ∈ C(R) such that the sequence {ρ ◦ fn} ⊆ L1(Q) is uniformly integrable, it holds ρ ◦ fk ≡ ρ ◦ fnk ⇀ ρ∗ in L1(Q) , (5.7) where ρ∗(x, t) := 〈τ(x,t), ρ〉R = ∫ R ρ(ξ) dτ(x,t)(ξ) a.e. (x, t) ∈ Q (5.8) and {τ(x,t)} is the disintegration of τ . Acknowledgement. This research received funding from a research grant of Viet- nam National University, HCM city, project number B2019-18-01. References [1] G. I. Barenblatt, M. Bertsch, R. Dal Passo, M. Ughi; A degenerate pseudo-parabolic regu- larization of a nonlinear forward-backward heat equation arising in the theory of heat and mass exchange in stably stratified turbulent shear flow, SIAM J. Math. Anal., 24 (1993), 1414-1439. [2] L. Cherfils, A. Miranville, S. Zelik; The Cahn-Hilliard Equation with Logarithmic Potentials, Milan J. Math., Vol. 79 (2011), 561-596. [3] A. Debussche, L. Dettori; On the Cahn-Hilliard equation with degenerate mobility, SIAM J. Math. Anal., 27 (1996), 404-423. [4] C. M. Elliott, H. Garcke; On the Cahn-Hilliard equation with a logarithmic free energy, Nonlinear Anal., 24 (1995), 1491-1514. [5] C. M. Elliott, S. Luckhaus; A generalized diffusion equation for phase separation of a multi- component mixture with interfacial energy, SFB 256 Preprint No. 195, University of Bonn, (1991). [6] H. Fakih; Asymptotic behavior of a generalized Cahn-Hilliard equation with a mass source, Applicable Analysis., 96 (2017), 324-348. [7] P. Grindrod; Models of individual aggregation in single and multispecies communities, J. Math. Biol., 26 (1988), 651–660. [8] M. E. Gurtin, R. C. MacCamy; On the diffusion of biological populations, Mathematical biosciences, 33 (1977), 35–49. [9] M. Giaquinta, G. Modica, J. Souček; Cartesian Currents in the Calculus of Variations (Springer, 1998). [10] E. Khain, L.M. Sander; A generalized Cahn-Hilliard equation for biological applications, Phys. Rev. E, 77 (2008), p. 051129. [11] M. Lizana, V. Padron; A spatially discrete model for aggregating populations, J. Math. Biol., 38 (1999), 79–102. [12] A. Miranville; Asymptotic behaviour of a generalized Cahn-Hilliard equation with a profiler- ation term, Applicable Analysis., 92 (2013), 1308-1321. [13] A. Novick-Cohen, R. L. Pego; Stable patterns in a viscous diffusion equation, Trans. Amer. Math. Soc. 324 (1991), 331-351. EJDE-2020/07 CAHN-HILLIARD EQUATION 15 [14] Y. Oono, S. Puri; Computationally efficient modeling of ordering of quenched phases, Phys. Rev. Lett. 58 (1987), 836-839. [15] V. Padrón; Effect of aggregation on population revovery modeled by a forward-backward pseudoparabolic equation, Trans. Amer. Math. Soc., no. 7, 356 (2004), 2739-2756. [16] P. I. Plotnikov; Passage to the limit over a small parameter in the Cahn-Hilliard equations, Siberian Math. J., 38 (1997), 550-566. [17] P. I. Plotnikov; Passing to the limit with respect to viscosity in an equation with variable parabolicity direction, Diff. Equ., 30 (1994), 614-622. [18] M. Slemrod; Dynamics of measure-valued solutions to a backward-forward heat equation, J. Dynam. Differential Equations, 3 (1991), 1-28. [19] M. Valadier; A Course on Young Measures, Rend. Ist. Mat. Univ. Trieste, 26 (1994), suppl., 349-394 (1995). Bui Le Trong Thanh Faculty of Mathematics and Computer Science, University of Science, 227 Nguyen Van Cu, D. 5, Ho Chi Minh City, Vietnam. Vietnam National University, Ho Chi Minh City, Vietnam Email address: bltthanh@hcmus.edu.vn Nguyen Ngoc Quoc Thuong Faculty of Mathematics and Statistics, Quy Nhon University, 170 An Duong Vuong Street, Quy Nhon City, Vietnam Email address: nguyenngocquocthuong@qnu.edu.vn 1. Introduction 2. Well-posedness of problem (??) 2.1. Mathematical formulation and results 2.2. A priori estimates 3. Convergence of solutions of problem (??) as 0 4. Existence of measure-valued solution of problem (??) 5. Appendix Acknowledgement References