Sobolev Trace

Overview

A Sobolev trace supplies boundary values for functions whose regularity is measured by integrals rather than pointwise smoothness. It is what makes a Dirichlet condition meaningful for an energy solution of a PDE.

An $L^p$ equivalence class records values only up to sets of volume zero, while a smooth boundary itself has volume zero. Arbitrarily assigning values on that boundary therefore cannot define a boundary operator on equivalence classes. The trace is instead determined by interior Sobolev regularity and approximation.

Prerequisites

Know $W^{1,p}(\Omega)$ and weak derivatives. For $1\le p<\infty$, $W^{1,p}_0(\Omega)$ denotes the closure of $C_c^\infty(\Omega)$ in the Sobolev norm. That definition makes sense on any open set. Its identification with functions having zero boundary trace requires a theorem and suitable domain geometry.

Definition and conventions

For this article, let $\Omega\subseteq\mathbb R^n$ be a bounded domain with smooth boundary and $1\le p<\infty$. The first-order trace is the unique bounded linear map

agreeing with ordinary boundary restriction for smooth functions on $\overline\Omega$. There is a constant $C=C(\Omega,p)$ such that

If smooth $u_m\to u$ in $W^{1,p}$, then their boundary restrictions converge to $Tu$ in boundary $L^p$. The estimate makes that limit independent of the approximating sequence. Surface measure $dS$ is different from volume measure $dx$. On an interval, boundary data are the two endpoint values, measured with counting measure.

Intuition

Derivative control couples nearby interior layers, preventing them from carrying unrelated boundary values at small Sobolev cost. The trace estimate formalizes that coupling. It does not generally assert a pointwise boundary limit at every boundary point. The output is itself an equivalence class on the boundary.

Examples and non-examples

On $(0,1)$, a $W^{1,p}$ function has an absolutely continuous representative, and its trace is $(u(0),u(1))$. For $u(x)=x(1-x)$, both entries are zero, so $u\in W^{1,p}_0(0,1)$. For $v(x)=1$, the trace is $(1,1)$, and $v$ is not in that zero-boundary space.

Endpoint evaluation is not bounded in the $L^2(0,1)$ norm alone. Take $w_m(x)=\max(1-mx,0)$. Then $w_m(0)=1$ while

Its derivative has squared $L^2$ norm $m$, so the sequence does not tend to zero in $H^1$. This calculation pinpoints why the derivative term is essential to trace continuity.

Important properties

For a proof sketch, begin with a smooth compactly supported function on the closed upper half-space. Integrating along the inward normal gives, for $p>1$,

Integrate in $x'$, apply Hölder's inequality, and bound by the Sobolev norm. For $p=1$, use the fundamental theorem of calculus and the integral of $|\partial_nu|$. Density extends the boundary restriction operator. Boundary charts and a partition of unity transfer the estimate to a smooth bounded domain.

Under these hypotheses, $\ker T=W^{1,p}_0(\Omega)$. One inclusion follows by continuity from compactly supported smooth approximation. For the converse, zero trace permits cutoff and smoothing constructions that remove a boundary layer with vanishing Sobolev error.

In the Hilbert case, the natural trace range is $H^{1/2}(\partial\Omega)$, and $T:H^1(\Omega)\to H^{1/2}(\partial\Omega)$ is onto. Thus the $L^2$ target above is a convenient weaker description, not a claim that all $L^2$ boundary data arise from $H^1$ functions. For higher-order Sobolev spaces, zero function trace alone is insufficient to characterize $W^{k,p}_0$; derivative traces must also be considered.

Connections and further reading

Zero trace defines homogeneous Dirichlet data in weak elliptic problems. Nonzero data can be handled by choosing a Sobolev extension and subtracting it before solving a zero-boundary problem. Hunter, §3.9 develops the first-order construction; 程代展、赵寅, §3.9 describes the Hilbert-space trace range.