Kolmogorov-type inequalities for derivatives (Q3839263)

By a Kolmogorov type inequality the authors mean an inequality of the following form \[ \| x^{(k)}(\cdot)\| _{L_q(T)} \leq K\| x(\cdot)\| _{L_p(T)}^\alpha \| x^{(n)}(\cdot)\| _{L_r(T)}^\beta, \] (where \(0\leq k\leq n\) are integers, \(1\leq p,q,r\leq\infty\), \(\alpha,\beta\geq 0\), \(T={\mathbb R}\) of \({\mathbb R}_+\)) which must hold for all functions \(x(\cdot)\in L_p(T)\) with \((n-1)\)th derivative locally absolutely continuous on \(T\) and \(x^{(n)}(\cdot)\in L_r(T)\). The finding of the optimal constant \(K\) in the above inequality is equivalent to finding the precise solution to the following optimization problem: \[ \| x^{(k)}(\cdot)\| _{L_q(T)}\to\max \quad \text{subject to} \quad \| x(\cdot)\| _{L_p(T)}^\alpha\leq\gamma_1, \| x^{(n)}(\cdot)\| _{L_r(T)}^\beta,\leq\gamma_2, \] where \(\gamma_1,\gamma_2\) are fixed positive constants. Thus, the discussion of the original inequality problem can be carried out with the technique and results of optimization theory using, for instance, Lagrange principle of duality methods. In this paper, the authors consider the even more general optimization problem \[ \| D_0x(\cdot)\| _{X_0(T)}\to\max \quad \text{subject to} \quad \| D_jx(\cdot)\| _{X_j(T)}\leq\gamma_j \quad (j=1,\cdots,N), \tag{1} \] where \(T\) is manifold (a line, a half-line, a closed interval, a sphere, the space \({\mathbb R}^d\), etc.), \(X_j(T)\) is a function space over \(T\) and the \(D_j\) are certain differential operators (\(j=0,1,\dots,N\)). The following special cases of (1) are discussed in detail. (i) \(T={\mathbb R}^d\), \(X_j(T)=L_2(T)\), and \(D_j=D^{\alpha_j}\) is the partial differential operator of fractional order \(\alpha_j\in{\mathbb R}_+^d\) in Weyl's sense \((j=0,1,\dots,N)\) [generalization of the Hardy-Littlewood-Pólya inequality]; (ii) \(T={\mathbb R}^d\), \(X_0(T)=C^b(T)\), \(X_j(T)=L_2(T)\) \((j=1,\dots,N)\), and \(D_j=D^{\alpha_j}\) is the partial differential operator of fractional order \(\alpha_j\in{\mathbb R}_+^d\) \((j=0,1,\dots,N)\) [generalization of Talkov's inequality]; (iii) \(T={\mathbb R_+}\), \(X_0(T)=C^b(T)\), \(X_j(T)=L_2(T)\) \((j=1,2)\) and \(D_0,D_1=\text{id}\), \(D_2=d/dx\) [generalization of Gabushin's inequality]; (iv) \(T={\mathbb R_+}\), \(X_j(T)=L_2(T)\) \((j=0,1,2)\) and \(D_0=d^k/dx^k\), \(D_1=\text{id}\), \(D_2=d^n/dx^n\) \((0\leq k\leq n)\) [Lyubich-Kuptsov inequality]; (v) \(T=S^d=\{x\in{\mathbb R}^{d+1}:\| x\| =1\}\), \(X_j(T)=L_2(T)\), and \(D_j=(-\Delta)^{\alpha_j/2}\) is the Laplace-Beltrami operator of degre \(\alpha_j/2\) on the sphere \(\alpha_j\in{\mathbb R}_+^d\) \((j=0,1,\dots,N)\) [Hardy-Littlewood-Pólya problem on the manifold \(S^d\)]; (vi) \(T={\mathbb R_+}\), \(X_0(T)=C^b(T)\), \(X_j(T)=L_2(T)\) \((j=1,2)\) and \(D_0=D^\alpha\) (where \(\alpha\in\mathbb R\)) and \(D^\alpha\) is the fractional derivative in the Riemann-Liouville sense, \(D_1=\text{id}\), \(D_2=d^n/dx^n\) [further generalization of Gabushin's inequality]. The results and methods of this paper summarize and unify many classical and known contributions of the literature. This work can be considered as a corner-stone paper; it is likely that it will have a great influence on the theory of inequalities involving derivatives.