Extension operators and domain geometry

For a bounded Lipschitz open set $\Omega\subset\mathbb R^n$ and $1\le p<\infty$, there is a bounded linear operator $E:W^{1,p}(\Omega)\to W^{1,p}(\mathbb R^n)$ satisfying $Eu=u$ almost everywhere inside. Lipschitz means the boundary can locally be written as a graph of a Lipschitz function, with the domain on one side; corners are allowed, while arbitrary cracks and cusps need not satisfy this description. The operator norm depends on the domain geometry and exponent, but not on the particular function being extended.

The conclusion is stronger than extending each individual function by an unspecified formula. Linearity and uniform boundedness let an entire bounded family share a common whole-space estimate. A cutoff can arrange that the extension is supported in a fixed bounded neighborhood of a bounded domain, at the cost of another controlled norm factor. This is useful in compactness arguments, where both translation control and spatial tightness are needed. Lipschitz regularity is a sufficient hypothesis, not a necessary characterization of every possible extension domain.

On an interval adjoining a flat boundary, reflect a function evenly across the boundary. Its values agree on both sides, so the first weak derivative is the original derivative on one side and its reflected negative on the other. Integrating the pth powers over equal halves doubles both the function and derivative contributions. This simple construction is the local model for extension across a Lipschitz graph: flatten the graph with a bi-Lipschitz coordinate map, reflect, then transform back. A partition of unity patches finitely many boundary charts and an interior region.

For first-order Sobolev functions, the chain rule under bi-Lipschitz coordinate changes controls derivatives through essentially bounded Jacobians, and change of variables controls integrals. These estimates explain the geometric constants. A naive even reflection is generally insufficient for higher-order extension: a smooth function can acquire a jump in its first derivative across the reflected interface, creating a singular second derivative. Higher-order theorems use constructions that match enough derivative information. One should not promote a successful first-order reflection calculation into a proof for every order without checking the interface terms.

If a function has nonzero boundary values, extending it by zero creates a jump. The derivative of that continuation contains a boundary distribution, so it need not belong to $W^{1,p}$ on the whole space even when the original function is smooth inside. In contrast, for $u\in W^{1,p}_0(\Omega)$ with finite $p$, zero extension belongs to $W^{1,p}(\mathbb R^n)$ on any open set. Prove this first for compactly supported smooth approximants, where the identities are immediate, and then pass to the norm limit. This argument does not need a Lipschitz boundary.

An extension theorem often lets us transfer embeddings and approximation estimates from the whole space to a domain, but the transferred constant inherits the extension norm. Families of domains with increasingly narrow necks or poor geometric parameters can therefore have deteriorating constants. On an arbitrary open set, an interior estimate may remain true locally while a uniform global extension estimate fails. The correct question is which geometric hypothesis the desired conclusion uses, rather than whether a drawing happens to look sufficiently smooth.