Endpoint estimates of commutators of singular integrals vs. Conditions on the symbol (Q2291552)

Given a singular integral operator \(T \in B(L^2)\) and a function \(b \in\mathrm{BMO}\), one can consider the commutator \[ T_{b}:f \mapsto T(bf)-bT(f). \] A classical result in harmonic analysis is that \(T \in B(L^{p}(\mathbb{R}^{d}))\), for all Calderón-Zygmund operators \(T\), all \(\mathrm{BMO}\) functions \(b\), and all \(p \in (1,\infty)\). This has many applications to PDE. For instance, interior estimates for elliptic problems involving a non divergence form operator \(L\) can be deduced by representing the derivatives of the solution using commutators \(T_{b}\) of the (Green) solution operator and the coefficients (see e.g. [\textit{D.-C. Chang} and \textit{S.-Y. Li}, Ann. Sc. Norm. Super. Pisa, Cl. Sci., IV. Ser. 28, No. 2, 341--356 (1999; Zbl 0952.42008)]). At the end points \(p=1,\infty\), a problem is that \(T_{b}\) does not behave as well as \(T\) does. For instance, Theorem 3.1 in [\textit{E. Harboure} et al., Ill. J. Math. 41, No. 4, 676--700 (1997; Zbl 0892.42009)] shows that, when \(T\) is the Hilbert transform, \(T_{b}\) maps \(L^{\infty}\) to \(\mathrm{BMO}\) (or \(H^{1}\) to \(L^1\)) if and only if \(b\) is constant. Appropriate endpoint results thus need to involve other spaces than the usual Hardy and \(\mathrm{BMO}\) spaces for the space of admissible functions \(b\) to be non-trivial. The paper [\textit{Y. Sun} and \textit{W. Su}, J. Funct. Anal. 234, No. 2, 235--260 (2006; Zbl 1156.42306)] successfully identifies such spaces, and demonstrates their relevance in PDE. Commutators \(T_{b}\) turn out to be bounded on \(\mathrm{BMO} \cap L^{q}\) (and \(H^{1}+L^q\)) for all \(q \in (1,\infty)\) and all \(b \in\mathrm{LMO} \cap L^{\infty}\). The space \(\mathrm{LMO}\) is a subspace of \(\mathrm{BMO}\) defined via the seminorm \[ [f]_{\mathrm{LMO}}:= \underset{x \in \mathbb{R}^{d}, r<\frac{1}{4}}{\sup} \frac{1+|\ln(r)|}{|B(x,r)|} \int _{B(x,r)} \left|f(y) - \left(\frac{1}{|B(x,r)|} \int _{B(x,r)}f(z)dz\right)\right|dy+\] \[+\underset{x \in \mathbb{R}^{d}, r\geq \frac{1}{4}}{\sup} \frac{1}{|B(x,r)|} \int _{B(x,r)} \left|f(y) - \left(\frac{1}{|B(x,r)|} \int _{B(x,r)}f(z)dz\right)\right|dy. \] Note that this space arises as a space of bounded multiplication operators on \(h^1\) (see [\textit{E. Nakai} and \textit{K. Yabuta}, J. Math. Soc. Japan 37, 207--218 (1985; Zbl 0546.42019)]). More generally, Nakai and Yabuta considered spaces \(\mathrm{BMO}_{\omega}\) and \(\mathrm{LMO}_{\omega}\) where the mean oscillations \[ \frac{1}{|B(x,r)|} \int _{B(x,r)} \left|f(y) - \left(\frac{1}{|B(x,r)|} \int _{B(x,r)}f(z)dz\right)\right|dy\] are replaced by weighted versions \[ \frac{1}{\omega(x,r)|B(x,r)|} \int _{B(x,r)} \left|f(y) - \left(\frac{1}{|B(x,r)|} \int _{B(x,r)}f(z)dz\right)\right|dy. \] The paper under review proves a very natural generalisation of the aforementioned \(\mathrm{BMO} \cap L^{q}\) boundedness result: \(T_{b}\) is bounded on \(\mathrm{BMO}_{\omega} \cap L^{q}\) spaces whenever \(b \in \mathrm{LMO}_{\Omega}\) for an appropriate \(\Omega\), given explicitly in (1.6). This generalisation is sharp in the sense that, if \(T\) is a Riesz transform and \(T_{b} \in B(\mathrm{BMO}_{\omega} \cap L^{q})\), then \(b \in\mathrm{LMO}_{\Omega}\). The main result (Theorem 1.9) is proven under mild regularity assumptions on \(\omega\): \( \omega(x,r) \leq C \omega(x,s) \quad \forall s>r\), and \(\omega(x,r) \leq C \omega(y,r) \quad \forall y \in B(x,r)\), as well as \(\omega(\cdot,1)^{-1} \in L^{\infty}\), and the slightly more technical assumptions: \[ \int \limits _{r} ^{\infty} \frac{\omega(x,s)}{s^{2}}ds \leq C\frac{\omega(x,r)}{r} \quad \forall r>0 \quad \forall x \in \mathbb{R}^{d},\] on \(\omega\) and \(\Omega\). Example 2.4. shows that these conditions are satisfied by many natural choices of \(\omega\).