Extremal maps in best constants vector theory. I: Duality and compactness (Q652439)

Continuity (Theorem 1.1), duality (Theorem 1.2) and compactness (Theorem 1.3) of sharp potential type Riemannian \(L^2\)-Sobolev inequalities are proved by means of a local geometric Sobolev inequality of the same kind and suitable De Giorgi-Nash-Moser estimates. Let \((M,g)\) be a smooth Riemannian manifold of dimension \(n\geq 3\) and \(\beta: M\to\mathbb R\) be a positive continuous function. Then there are constants \(A\), \(B\) such that \[ \Biggl(\int_M|u|^{2^*} dv_g\Biggr)^{2/2^*}\leq A \int_M |\nabla_g u|^2\,dv_g+ B\int_M \beta(x)\,u^2 dv_g \] for any \(u\in C^\infty(M)\), where \(2^*={2n\over n-2}\). The infimum of \(A\)'s satisfy this inequality for some \(B\) is the first sharp Riemannian \(L^2\)-Sobolev best constant \(A_0(n,\beta, g)\). Similarly, the second sharp Riemannian \(L^2\)-Sobolev best constant \(B_0(n,\beta,g)\) is defined to be the infimum of \(B\)'s which satisfy the first sharp \(L^2\)-Sobolev inequality \[ \Biggl(\int_M |u|^{2^*} dv_g\Biggr)^{2/2^*}\leq A_0(n,\beta,g) \int_M |\nabla_g u|^2\,dv_g+ B\int_M \beta(x)\,u^2dv_g \] (cf. [\textit{E. Hebey} and \textit{M. Vaugon}, Ann. Inst. Henri Poincaré, Anal. Non Linéaire 13, No. 1, 57--93 (1996; Zbl 0849.53035)]). An extremal function is a function which satisfies the second sharp \(L^2\)-Sobolev inequality \[ \Biggl(\int_M |u|^{2^*} dv_g\Biggr)^{2/2^*}\leq A_0(n,\beta, g)\int_M |\nabla_g u|^2 dv_g+ B_0(n,\beta, g)\int_M\beta(x)\,u^2 dv_g \] with equality. Then \(A_0(n,\beta,g)= A_0(n)\) depends only on \(n\) [\textit{T. Aubin}, J. Differ. Geom. 11, 573--598 (1976; Zbl 0371.46011)], and the following are known results. 1. Continuity: The map \(g\in{\mathcal M}\to B_0(n,1,g)\) is continuous if \(n\geq 4\), where \({\mathcal M}\) is the space of Riemannian metrics on \(M\) endowed with the \(C^2\) topology [\textit{E. R. Barbosa} and \textit{M. Montenegro}, Commun. Pure Appl. Anal. 8, No. 6, 1759--1777 (2009; Zbl 1201.32012)]. 2. Duality: If \(n\geq 4\), \[ B_0(n,\beta,g)\beta(x_0)= {(n-2)\over 4(n-1)} A_0(n) S_g(x_0),\text{ for some }x_0\in M. \] Here \(S_g\) is the scalar curvature of the metric \(g\) [\textit{E. Hebey} and \textit{M. Vaugon}, Math. Z. 237, No. 4, 737--767 (2001; Zbl 0992.58016)]). 3. Compactness: The set of extremal functions with unit \(L^2\)-norm is compact if \[ B_0(n,\beta,g) \beta(x)> {(n-2)\over 4(n-1)} A_0(n) S_g(x), \] for all \(x\) [Hebey and Vaugon, 2001, loc. cit.]. In this paper, these results are extended to sharp potential type Riemannian \(L^2\)-Sobolev inequalities. Let \(C^\infty_k(M)\) be the space of smooth \(k\)-maps (maps to \(\mathbb R^k\)) on \(M\), \(F:\mathbb R^k\to\mathbb R\) and \(G: M\times\mathbb R^k\to\mathbb R\) be positive continuous maps with \(F\) homogeneous of degree \(2^*\) and \(G\) homogeneous degree 2 in the second variable. The space of such functions considered as a subspace of \(C^0_{\text{loc}}(\mathbb R^k)\) or \(C^0_{\text{loc}}(M\times \mathbb R^k)\) is denoted by \({\mathcal F}_k\) and \({\mathcal G}_k\), respectively. Then there are constants \({\mathcal A}\), \({\mathcal B}\in\mathbb R\) such that \[ \Biggl(\int_M F(U) dv_g\Biggr)^{2/2^*}\leq{\mathcal A} \int_M |\nabla_g U|^2 dv_g+{\mathcal B} \int_M G(x,U)\,dv_g, \] where \(U= (u_1,\dots, u_k)\). The first Riemannian \(L^2\)-Sobolev best constant \({\mathcal A}_0(n,F,G,g)\), etc., are defined similar to \(A_0(n,\beta,g)\), etc., and it is shown that the first sharp potential type Riemannian \(L^2\)-Sobolev inequality \[ \Biggl(\int_M F(U) dv_g\Biggr)^{2/2^a st}\leq{\mathcal A}_0(n, F,G,g) \int_M |\nabla U|^2 dv_g+{\mathcal B}\int_M G(x,U)\,dv_g \] is valid if \(n\geq 3\) (Proposition 2.2). It is also shown \[ {\mathcal A}_0(n,F,G,g)={\mathcal A}_0(n,F)= M^{2/2^*}_F A_0(n),\quad M_F= \max_{|t|=1} F(t). \] If \(n\geq 4\), the estimate \[ {\mathcal B}_0(n,F,G,g) G(x, t_0)\geq {n-2\over 4(n-1)}{\mathcal A}_0(n,F) S_g(x),\quad t_0\in\mathbb S^{k-1} \] is derived applying solutions of the Yamabe problem (Proposition 2.3, cf. [\textit{T. Aubin}, J. Math. pur. appl., IX. Sér. 55, 269--296 (1976; Zbl 0336.53033); \textit{R. Schoen}, J. Differ. Geom. 20, 479--495 (1984; Zbl 0576.53028)]). An extremal map is not differentiable, in general. But if \(F(t)\) and \(G(x,t)\) are of class \(C^1\) in \(t\), then up to a scaling, \(U= (u_1,\dots, u_n)\) satisfies the potential type elliptic system \[ -\Delta_g u_i+ {\lambda\over 2} {\partial G(x,U)\over\partial t_i}= {1\over 2^*} {\partial F(U)\over\partial t_i},\quad i= 1,\dots, k, \] for some \(\lambda >0\). In \S2.2, De Giorgi-Nash-Moser estimates for weak solutions of potential type elliptic systems (Proposition 2.4) and for minimizers of non-differentiable energy functions on \(H^{1,2}_k(M)\) are established (Proposition 2.5). After showing that positive homogeneous functions can be approximated by \(C^1\) functions of the same kind (Proposition 2.6), the local geometric Sobolev inequality of \textit{O. Druet} [Proc. Am. Math. Soc. 130, No. 8, 2351--2361 (2002; Zbl 1067.53026)] is extended to the vector context (Proposition 2.7) in \S2.3. Some technical details such as the vector version of the \textit{H. Brezis} and \textit{E. Lieb}' s lemma [Proc. Am. Math. Soc. 88, 486--490 (1983; Zbl 0526.46037)] used in these proofs are proved in the Appendix. The authors say that the assumption \(n\geq 5\) is only used in the proof of Proposition 2.7. By using these results, assuming \(M\) is compact, \(n\geq 5\) and \(k\geq 1\), the Theorems 1.1--1.3 are proved in \S3, \S4 and \S5. Theorem 1.1. The map \((F,G,g)\in{\mathcal F}_k\times{\mathcal G}_k\times{\mathcal M}^n\to{\mathcal B}_0(n,F,G,g)\) is continuous. Theorem 1.2. At least one of the following assertion holds: 1. The second sharp potential type Riemannian \(L^2\)-Sobolev inequality admits an extremal map. 2. \({\mathcal B}_0(n,F,G,g)G(x_0,t_0)= {n-2\over 4(n-1)}{\mathcal A}_0(n,F) S_g(x_0)\) for some \(x_0\in M\) and some maximal points \(t_0\) of \(F\) on \(\mathbb S^{k-1}\). Theorem 1.3. Let \(((F_\alpha, G_\alpha, g_\alpha))\) be a sequence converging to \((F,G,g)\) in \({\mathcal F}_k\times{\mathcal G}_k\times{\mathcal M}\) and \((U_\alpha)\) be a sequence of extremal maps. If \[ {\mathcal B}_0(n,F,G,g) G(x,t_0)> {n-2\over 4(n-1)}{\mathcal A}_0(n,F) S_g(x_0) \] for all \(x\in M\) and all maximum points \(t_0\) of \(F\) on \(\mathbb S^{k-1}\), then \((U_\alpha)\) is \(L^{2^*}_k\)-compact. Moreover, if \(((F_\alpha, G_\alpha, g_\alpha))\) converges in \(C^1_{\text{loc}}(\mathbb R^k)\times C^0(M, C^1_{\text{loc}}(\mathbb R^k))\times{\mathcal M}\), then \((U_\alpha)\) is \(C^0_k\)-compact. As for Theorem 1.2, there exist examples satisfying both of 1 and 2 (\S6. Example 2). \S6 also gives examples of nonexistence of extremal maps (Example 3), of non-compactness of the set of extremal maps (Example 4) and of losts of continuity of the second best constant (Example 5). The authors say that to extend results in this paper to low-dimensional manifolds, especially to 3-dimensional manifolds is an important problem. Another interesting problem is the study of \(L^p\)-Sobolev inequalities \((p\neq 2)\). They announce a paper on this problem as a sequel to this paper.