Vector-valued local approximation spaces (Q666641): Difference between revisions

The authors consider several spaces, namely Besov-type spaces \(B^{sM}_{pq}\), local approximation spaces \(A^{sN}_{pq}\) and Sobolev-type spaces \(W^{k,q}\), of vector-valued functions \(f: \mathbb R^n\to Y\) where \(Y\) is a Banach space. The aim of the paper is to analyze the embeddings between those spaces depending on the geometric properties of \(Y\). They, for each \(p,q\in [1,\infty]\), \(s\in \mathbb R\) and \(M>s/2\), define the (homogeneous) Besov norm by \[ [f]_{B^{sM}_{pq}} = \Bigg(\int_0^\infty t^{(M-s/2)q} \|T_t^{(M)}f\|_{L^p(\mathbb R^n, Y)}^q\frac{dt}{t}\Bigg)^{1/q}, \] where \(T_t=e^{\Delta t}\) is the heat semigroup and \(T^{(M)}_tf=\partial_t^M T_t f=\Delta ^M T_t f\). Also, for each \(s>0\) and \(N\in \mathbb N\), the (homogeneous) local approximation norm is given by \[ [f]_{A^{sN}_{pq}} = \Bigg(\int_0^\infty t^{-sq} \Big\| \inf_{\deg(P)\le N} \Big(\frac{1}{|B(x,t)|}\int_{B(x,t)} \|f(y)-P(y)\|^p_Y dy\Big)^{1/p}\Big\|_{L^p(\mathbb R^n)}^q \frac{dt}{t}\Bigg)^{1/q}, \] where the infimum is taken over of polynomials \(P\) such that \(\deg(P)\le N\). The corresponding spaces \(B^{sM}_{pq}\) and \(A^{sN}_{pq}\) consist of functions in \(L^p(\mathbb R^n, Y)\) and the norm is defined adding \(\|f\|_{L^p}\) to the above quantities. The authors also consider the non-homogeneous Sobolev norm given, for each \(k\in \mathbb N\) and \(q\ge 1\), by \[ \|f\|_{W^{k,q}}=\sum_{0\le |\alpha|\le k} \|\partial_x^\alpha f\|_{L^q(\mathbb R^n, Y)}. \] One of their results establishes that, for \(M>s/2\) and \(N>s-1\), the norms \(\|f\|_{A^{sN}_{pq}}\) and \(\|f\|_{B^{sM}_{pq}}\) are equivalent. The authors get a nice connection between the embedding of the Sobolev spaces into Besov or local approximation spaces in the vector-valued setting and some geometric properties of the space \(Y\). Namely, they show that, for \(k\), \(n\ge 1\), \(N\ge k\) integers, \(Y\) admits an equivalent norm with modulus of convexity of power type \(q\) if and only if \(\|f\|_{A^{kN}_{q q}}\le C \|f\|_{W^{k,q}}\) for all \(f\in W^{k,q}(\mathbb R^n, Y)\). The above result uses the fact that the embedding \(W^{k,q}(\mathbb R^n, Y)\subset B^{kM}_{qq}(\mathbb R^n,Y)\) for \(q\ge 2\) is equivalent to the fact that \(Y\) has martingale cotype \(q\). The proofs for the results rely also on the equivalence proved by \textit{G. Pisier} [Isr. J. Math. 20, 326--350 (1975; Zbl 0344.46030)] and uses also the result by \textit{T. Martínez} et al. [Adv. Math. 203, No. 2, 430--475 (2006; Zbl 1111.46008)] relating the embedding between \(L^q(\mathbb R^n, Y)\) and \(B^0_{qq}(\mathbb R^n, Y)\) with martingale cotype \(q\).