A general wavelet-based profile decomposition in the critical embedding of function spaces (Q643855)

Let \((X,\|\cdot\|_X)\) and \((Y,\|\cdot\|_Y)\) be two generic homogeneous function spaces and \(X\subset Y\), which have the same scaling property in the sense that for any function \(f\) and \(h>0\), \(h^r\|f(h\cdot)\|_X=\|f\|_X \quad \text{and} \quad h^r\|f(h\cdot)\|_Y=\|f\|_Y\) for the same value of \(r\). Let the function \(\psi\) be the so called mother wavelet. Given a function \(f\), its wavelet decomposition means that \(f\) has the following form \(f=\sum_{\lambda\in\nabla}d_\lambda\psi_\lambda,\) where \(\nabla:=\mathbb{Z}\times\mathbb{Z}^d\) and \(\lambda=(j,k)\) concatenates the scale index \(j=j(\lambda)\) and space index \(k=k(\lambda)\). For \(d=1\), \(\psi_{j,k}=\psi_{\lambda}=2^{j/2}\psi(s^j\cdot-k)\), \(j\in\mathbb{Z}\), \(k\in\mathbb{Z}\). For \(d>1\), let \(E\) be a finite set, \(\{\psi^e\}_{e\in E}\) be generating functions, \(\psi_{\lambda}:=(\psi_{\lambda}^e)_{e\in E}^T\), \(d_\lambda=(d_\lambda^e)_{e\in E}\), and \(\psi_{j,k}=\psi_\lambda=2^{dj/2}\psi(2^j\cdot-k)\), \(j\in\mathbb{Z},\,k\in\mathbb{Z}^d\). Because of the property of being generic homogeneous, one can systematically normalize the wavelets in \(X\) which is equivalent to normalizing them in \(Y\) by \(\psi_{j,k}=\psi_\lambda=2^{rj}\psi(2^j\cdot-k)\). The wavelet bases \(\{\psi_\lambda\}\) are called unconditional if for such spaces \(X\), there exists a positive constant \(D\) such that for any finite subset \(E\subset\nabla\) and coefficients vectors \((c_\lambda)_{\lambda\in E}\) such that \(|c_\lambda|\leq|d_\lambda|\) for all \(\lambda\), one has \(\|\sum_{\lambda\in E}c_\lambda\psi\lambda \|_X\leq D\|\sum_{\lambda\in E}d_\lambda\psi\lambda\|_X.\) For each \(M>0\), if \(f\in X\) has the expansion in the wavelet basis, then the nonlinear projector is defined by \(Q_Mf:=\sum_{\lambda\in E_M}d_\lambda\psi_\lambda,\) where \(E_M=E_M(f)\) is the subset of \(\nabla\) of cardinality \(M\) that corresponds to the \(M\) largest values of \(|d_\lambda|\). Assumption (1): The nonlinear projection satisfies \(\lim_{M\to\infty}\max_{\|f\|_X\leq 1}\|f-Q_Mf\|_Y=0.\) Assumption (2): If for a sequence \(\{f_n\}_{n>0}\) of functions, which are uniformly bounded in \(X\) and may be written as \(f_n=\sum_{\lambda\in\nabla}c_{\lambda,n}\psi_\lambda,\) and for all \(\lambda\), the sequence \(c_{\lambda,n}\) converges towards a finite limit \(c_\lambda\) as \(n\to \infty\), then the series \(\sum_{\lambda\in\nabla}c_\lambda\psi_\lambda\) converges in \(X\) with \(\|\sum_{\lambda\in\nabla}c_\lambda\psi_\lambda\|_X\leq C\liminf_{n\to\infty}\|f_n\|_X,\) where \(C\) is a positive constant only depending on the space \(X\) and on the choice of the wavelet basis. In this paper, the authors prove the following: Assume that \(X\) and \(Y\) are two function spaces with the same scaling and continuous embedding \(X\subset Y\), and assume that there exists a wavelet basis \(\{\psi_\lambda\}_{\lambda\in \nabla}\) which is unconditional for both \(X\) and \(Y\) such that Assumptions (1) and (2) hold. Let \(\{u_n\}_{n>0}\) be a bounded sequence in \(X\). Then, up to sequence extraction, there exists a family of functions \(\{\phi^l\}_{l>0}\) in \(X\) and sequences of scale-space indices \(\{\lambda_l(n)\}_{n>0}\) for each \(l>0\) such that \(u_n=\sum_{l=1}^L\phi_{\lambda_l(n)}^l+r_{n,L},\) where \(\lim_{L\to\infty}(\limsup_{n\to\infty}\|r_n\|_Y)=0\) and the decomposition above is asymptotically orthogonal in the sense that for any \(k\neq l\), \(|j(\lambda_k(n))-j(\lambda_l(n))|\to\infty\) or \(|k(\lambda_k(n))-2^{j(\lambda_k(n))-j(\lambda_l(n))}k(\lambda_l(n))| \to\infty\) as \(n\to\infty.\) Moreover, when \(X\) is a Besov or Triebel-Lizorkin space, the decomposition in the above result is stable in the sense that for some \(\tau=\tau(X)\), one has that \(\|(\|\phi^l\|_X)_{l>0}\|_{\ell^\tau}\leq CK,\) where \(C\) is a positive constant that only depends on \(X\) and the choice of the wavelet basis and \(K:=\sup_{n\geq0}\|u_n\|_X\).