Remark 3: loss, Littlewood–Paley input and Neumann estimates

PDF p. 7: Remark 3

Assume the Axiom of Choice and the hypotheses of Global Strichartz estimate with derivative loss. Its derivative loss is the extra $1\mathrm{/}p$ derivatives required of the initial datum compared with the target regularity. The article explains that the loss in its estimate (2.5) comes from the near-obstacle part of the argument, where local smoothing and Sobolev embedding are used. The homogeneous smoothing estimate referred to here is Homogeneous half-derivative local smoothing. This describes the method used for that estimate, not a claim that every estimate must lose derivatives.

Blair, Smith and Sogge obtain scale-invariant Strichartz estimates on non-trapping exterior domains for a restricted range of exponents. Here scale-invariant means that the regularity and time-space exponents obey the scaling balance of the free Schrödinger equation. In three dimensions their section 5 uses a Littlewood–Paley argument, decomposing data into spectral frequency pieces, to obtain the energy estimate recorded in Three-dimensional L4 time Linfinity space estimate. They use this estimate to prove well-posedness for the energy-critical quintic equation; their introduction attributes the earlier small-energy well-posedness result to Ivanovici and Planchon.

For the boundary-condition comparison, use the classical convention: if $u$ and its first spatial derivatives extend continuously from $\mathrm{\Omega }$ to its boundary on every bounded truncation, define ${\mathrm{\partial }}_{\nu }u={\sum }_{j=1}^n{\nu }_j{\mathrm{\partial }}_{j}u$ there, with $\nu$ the outward unit normal of $\mathrm{\Omega }$. For complex $u$ this is applied to real and imaginary parts. Homogeneous Neumann means ${\mathrm{\partial }}_{\nu }u=0$; homogeneous Dirichlet means $u=0$ on $\mathrm{\partial }\mathrm{\Omega }$. This is the local exterior-boundary adaptation of Classical normal derivative and the outward-normal convention of Bounded C1 domains and their outward normals, justified below using Exterior and truncated obstacle domains. This classical comparison does not assert a normal trace for arbitrary energy data; the cited evolution results retain their own operator realizations.

Burq, Gérard and Tzvetkov say the Neumann case can be handled by similar ideas. Their Remark 2.3 concerns the high-frequency resolvent, while Remark 2.9 obtains a corresponding local smoothing estimate with constants depending on the time interval. This qualification does not assert the time-uniform Dirichlet bound of Homogeneous half-derivative local smoothing for Neumann evolution.

Choose $R$ with $\mathrm{\Theta }\subset B_R(0)$ as in Exterior and truncated obstacle domains. The bounded truncation ${\mathrm{\Omega }}_R=B_R(0)\setminus \mathrm{\Theta }$ has smooth boundary: near $\mathrm{\partial }\mathrm{\Theta }$ it has the given one-sided smooth boundary charts, and near $\mathrm{\partial }{B}_R(0)$ it agrees with the ball, since these two boundary pieces are disjoint. The outward-normal and classical-normal-derivative definitions of Bounded C1 domains and their outward normals and Classical normal derivative therefore apply to real and imaginary parts of the restriction of $u$ to ${\mathrm{\Omega }}_R$. At each point of $\mathrm{\partial }\mathrm{\Theta }$, ${\mathrm{\Omega }}_R$ and $\mathrm{\Omega }$ agree in a neighborhood. Their outward normal and the continuous interior gradient are consequently the same there. Any other admissible radius gives that same local formula, so the classical derivative on the obstacle boundary is independent of the truncation. No boundary condition is imposed on the artificial sphere in this definition.

The original article's Remark 3 is the attribution for this comparison. The cited independent papers support the descriptions of their methods and results. This item does not claim a new proof of those results or an independent reading of Ivanovici and Planchon's proof.