An ε-free rank-6 decoupling estimate for the paraboloid surface
Pylyp Cherevan
TL;DR
The paper proves an $\varepsilon$-free rank-6 decoupling estimate for the paraboloid, showing $\|F\|_{L^{6}(Q_{\lambda})}$ is controlled by a negative-exponent combination of $\lambda$ and $D=\lambda^{1/12}$ times the $\ell^2$-sum of cap pieces. The approach fuses broad-rank-3 geometry with trilinear Kakeya–BCT insertions, a detailed kernel analysis via 12 integrations by parts and Schur/$TT^{*}$, a robust Kakeya block with a density threshold, an algebraic shell that removes mass near low-rank sets, a tube-packing viewpoint, and a narrow cascade accessed through double $7/8$-scale rescalings. The cumulative exponents are negative, ensuring the traditional $\lambda^{\varepsilon}$ and $D^{\varepsilon}$ losses are eliminated, yielding a robust, quantitative decoupling bound. The techniques advance anisotropic decoupling and geometric harmonic analysis, with potential applicability to broader hypersurface settings and related decoupling problems.
Abstract
For the paraboloid decomposition $F=\sum_Θ F_Θ$ with $Θ\subset{|ξ|\simλ}$ and radius $r=λ^{-2/3}$, we prove a log-free estimate $|F|{L^{6}(Qλ)}\lesssim λ^{Σ_λ} D^{Σ_{D}} \big(\sum_Θ|F_Θ|{L^{6}}^{2}\big)^{1/2}$ as $λ\to\infty$, where $D=λ^{1/12}$. Key components: (i) broad geometry of rank 3: bilipschitz behavior of normals gives $\max{i<j<k}|n_i\wedge n_j\wedge n_k|\gtrsim λ^{-5/4}$, which via a trilinear Kakeya-BCT insertion contributes $+5/36$ in $λ$; (ii) kernel estimate: twelve integrations (6 in $t$, 6 in $x^{\prime}$) and measure analysis (Schur and $TT^{}$) yield $|K|{L^2\to L^2}\lesssim λ^{-9/2} D^{-3}$; (iii) robust Kakeya: a density threshold $> c{} D$ brings a factor $D$ ($+1/12$ in $λ$, $+1$ in $D$); (iv) algebraic shell: excluding a neighborhood $N_β(P)$ contributes $-1/12$ in $λ$ and $-1$ in $D$; (v) tube packing: explanatory only; (vi) narrow cascade: a double $7/8$ rescaling exits the narrow regime and contributes $-5/64$ in $λ$ (zero in $D$). Summing exponents: $Σ_λ=5/36-9/2-5/64=-2557/576\approx -4.44<0$ and $Σ_{D}=-3+1-1=-3<0$, hence both $λ^{\varepsilon}$- and $D^{\varepsilon}$-losses are removed.