Pseudodifferential operators with rough symbols (Q2380392)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Pseudodifferential operators with rough symbols
scientific article

    Statements

    Pseudodifferential operators with rough symbols (English)
    0 references
    0 references
    26 March 2010
    0 references
    The classical classes of pseudodifferential operators are defined by \[ T_\sigma f(x)= \int \sigma(x,\xi)e^{2\pi i\xi\cdot x}\widehat{f}(\xi)\, d\xi \] for \(x,\xi\in\mathbb R^n\), with symbols in \(S^m_{\rho,\delta} \), where \(\sigma \in S^m_{\rho,\delta}\) if \[ |D_x^\beta D_\xi^\alpha \sigma(x,\xi)|\leq C_{\alpha,\beta}\langle\xi\rangle^{m-\rho |\alpha|+\delta|\beta|}, \] and \(\langle\xi\rangle= \sqrt{1+|\xi|^2}\). It is known that if \(\sigma\in S^0_{\rho,\rho}\), where \(\rho<1\), then \(T_\sigma: L^2\mapsto L^2\), and there is a counterexample at \(\rho=1\). Symbols with less smoothness in \(x\) have been considered by \textit{R.\,Coifman} and \textit{Y.\,Meyer} [``Au dela des opérateurs pseudo-différentiels'', Asterisque 57, Soc.\ Math.\ de France (1978; Zbl 0483.35082)]. They considered a modulus of continuity \(\omega:\mathbb R_+\mapsto\mathbb R_+\), where \(u\) satisfies that modulus of continuity if \(|u(x+y)-u(x)|\leq \omega(|y|)\), and said that a symbol \(\sigma\in C^\omega S^0_{1,0}\) if \[ \sup_{x, y, \xi}\langle\xi\rangle^{|\alpha|} |D_{\xi}^\alpha\sigma(x+y,\xi)- D_\xi^\alpha\sigma(x,\xi)|\leq C_\alpha\omega(|y|), \] and proved that, if \(\sum_{j\geq 0}\omega(2^{-j})^2<\infty\), then \(T_\sigma: L^p\mapsto L^p\), \(1<p<\infty\). The author investigates averaged smoothness conditions on the symbol which will give rise to bounded operators on \(L^p\). The motivation is recent work of \textit{I.\,Rodnianski} and \textit{T.\,Tao} [Commun.\ Math.\ Phys.\ 251, No.\,2, 377--426 (2004; Zbl 1106.35073)] and the author [\textit{A.\,Stefanov}, Adv.\ Math.\ 210, No.\,1, 246--303 (2007; Zbl 1121.35023)] in which parametrices were constructed for first order perturbations of the wave and Schrödinger equations. The parametrices did not possess pointwise smoothness conditions on the derivatives. Finally, the author is looking for scale invariant conditions on the symbol: in addition to getting an estimate \[ \|T_\sigma\|_{L^p\mapsto L^p}\leq C\|\sigma\|_Y, \] one wants the estimate to be such that \(\|\sigma(\lambda\cdot, \lambda^{-1}\cdot)\|_Y=\|\sigma\|_Y\), which is true for operators in the Hörmander class \(S^0_{1, 0}\) et al. The smoothness conditions are rather technical and are expressed in terms of the natural decomposition of the Besov spaces. One has a partition of unity on the Fourier transform side with a function whose Fourier transform \(\widehat{\varphi}\) is supported in \(1\leq|\xi|\leq 2\), \[ \sum_{k\in\mathbb Z} \widehat{\varphi} (2^{-k}\xi)=1, \quad \xi\neq 0, \] and so \(f=\sum_{k\in\mathbb Z}P_k f\) is the Littlewood-Paley decomposition of \(f\), where \(\widehat{P_k f}(\xi)= \varphi(2^{-k}\xi) \widehat{f}(\xi)\). He writes the homogeneous Besov space norm as \[ \|f\|_{\dot{B}^s_{p,q}}= \left(\sum_{l\in\mathbb Z} 2^{lsq}\|P_l f\|_{L^p}^q \right)^{1/q}, \] and the Triebel-Lizorkin spaces are defined by \[ \|f\|_{\dot{F}^s_{p,q}}= \left\|\bigg(\sum_{l\in\mathbb Z} 2^{lsq} |P_lf|^q\bigg)^{1/q} \right\|_{L^p}, \] and \(W^{p,s}\) are the usual Sobolev spaces of functions with \(s\) derivatives in \(L^p\) when \(s\) is an integer and are defined by the Littlewood-Paley decomposition when \(s\) is non-integer. One can also modify the definition of the Besov spaces on the unit ball and define the inhomogeneous Besov spaces \(B^s_{p,q}\) and Triebel-Lizorkin spaces \(F^s_{p,q}\). The statement of the theorem also requires the Littlewood-Paley decomposition in both the \(x\) and \(\xi\) variables, \[ P_j^x f(x,\xi)= 2^{jn} \int\widehat{\varphi} (2^j (x - y)) f(y, \xi) \, dy, \] \[ P_j^{\xi} f(x, \xi)= 2^{jn} \int \widehat{\varphi} (2^j (\xi - \eta)) f(x, \eta) \, d\eta, \] Theorem 1. Let \(\sigma:\mathbb R^n \times\mathbb R^n\to\mathbb C\), and let \(T_\sigma\) be the corresponding pseudodifferential operator. Then \[ \|T_\sigma\|_{L^2\mapsto L^2}\leq C \left(\sum_l 2^{\ln/2} \sup_x\|P_l^{\xi} \sigma(x, \cdot )\|_{L^2(\mathbb R^n)} \right) , \] and the result is sharp in the sense that for every \(p > 2\), there is a symbol \(\sigma(x, \xi)\) that satisfies \(|D_{\xi}^{\alpha} \sigma(x, \xi)|\leq C_{\alpha}|\xi|^{-|\alpha|}\), and \(\sup_x\|\sigma(x,\cdot)\|_{W^{p, n/p}}< \infty\), but the operator \(T_{\sigma}\) is unbounded on \(L^2\). The estimate is scale invariant and the counterexample shows that one needs essentially \(n/2\) distributional derivatives for the result. Theorem 2 gives \(L^p\) bounds and weak type one bounds under like conditions. Theorem 2. Let \(\sigma:\mathbb R^n \times\mathbb R^n\to\mathbb C\), and let \(T_\sigma\) be the corresponding pseudodifferential operator. Then, if \(2 \leq p < \infty\), \[ \|T_{\sigma}\|_{L^p\;\mapsto L^p}\;\leq C \left(\;\sum_l 2^{\ln/2} \sup_x\|P_l^{\xi} \sigma(x, \cdot )\|_{L^2(\mathbb R^n)}\right) , \] while, if \(1 < p < 2\), \[ \|T_{\sigma}\|_{L^p \mapsto L^p}+\|T_{\sigma}\|_{L^1 \mapsto L^{1, \infty}} \leq C \left( \sum_l 2^{\ln} \sup_x\|P_l^{\xi} \sigma(x, \cdot )\|_{L^1(\mathbb R^n)} \right) . \] He explains how the results are essentially necessary, and gives a version of the results for radial symbols and symbols homogeneous of degree zero (although the operators with symbols homogeneous of degree zero only map \(B^0_{p,1}\) into \(L^p\), \(2 \leq p < \infty\), and there are no weak type one results for this case). Several applications are given; I cite only one in some detail. For classical symbols, the composition of two symbols of class \(S^m_{\rho. \delta}\) is well understood, and there is a symbolic calculus. One cannot invoke the same symbolic calculus for rough symbols. The author shows that the operator whose symbol is the product of two symbols satisfying the sufficient conditions for \(L^p\) boundedness is bounded from \(L^2\) into \(L^2\) and its bound is the product of the bounds for the individual operators as maps of \(L^2\) into \(L^2\) from Theorem 1, while it is bounded from \(L^p\) into \(L^p\), \(1 < p < \infty\), and is bounded by the product of the operator norm bounds as a map of \(L^p\) into \(L^p\), \(1 < p < 2\), from Theorem 2. He also gives results for the maximal directional Hilbert transform and for the Carleson operator.
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    pseudodifferential operators
    0 references
    rough symbols
    0 references
    boundedness
    0 references
    scale invariant
    0 references
    Littlewood-Paley decomposition
    0 references
    0 references
    0 references