Failure of weak-type endpoint restriction estimates for quadratic manifolds.

It is well-known that the Fourier extension operator for the paraboloid in $$\mathbb {R}^d$$ R d cannot be weak-type bounded at the restriction endpoint $$q = 2d/(d-1)$$ q = 2 d / ( d - 1 ) , since such an estimate would imply bounds for the Kakeya maximal function which contradict the existence of Besicovitch sets. We generalize this approach to prove that the Fourier extension operator for an n -dimensional quadratic manifold $$\mathcal {M}$$ M cannot be weak-type bounded at the restriction endpoint. The key step in this proof is constructing a set $$K \subset \mathbb {R}^d$$ K ⊂ R d containing a translate of every plane normal to $$\mathcal {M}$$ M which can be covered by $$\lesssim \delta ^{-d}\left( \frac{\log \log (1/\delta )}{\log (1/\delta )}\right) ^{n/(d-n)}$$ ≲ δ - d log log ( 1 / δ ) log ( 1 / δ ) n / ( d - n ) many $$\delta $$ δ -balls. Such a set rules out endpoint bounds for the associated Kakeya maximal function and hence weak-type endpoint estimates for the restriction operator.