Trace theorems and boundary data

Let $\Omega$ be a bounded Lipschitz domain and $1\le p<\infty$. Restricting functions smooth on a neighborhood of the closure to the boundary satisfies a bound by their $W^{1,p}(\Omega)$ norm in the boundary $L^p$ norm, where the boundary measure is surface measure. Smooth approximation therefore extends restriction uniquely to a bounded linear trace map on the Sobolev space. In one dimension this is simply the pair of endpoint values of the absolutely continuous representative. In higher dimensions the trace is a boundary equivalence class, not necessarily a pointwise continuous boundary function.

For $1<p<\infty$, the sharper trace range is the fractional space $W^{1-1/p,p}(\partial\Omega)$, interpreted in Lipschitz charts; the map is onto that space and has a bounded linear lifting. For $p=2$ this is $H^{1/2}$, the natural Dirichlet data space for H¹ solutions. We use only the L¹ trace conclusion at $p=1$ here rather than copying the fractional formula to an endpoint without qualification. The boundary loses one-over-p derivatives, reflecting the reduction in integration dimension.

For the same domain and finite p, the kernel of the first-order trace is $W^{1,p}_0(\Omega)$. This identifies the closure definition with a genuine homogeneous Dirichlet condition. If a prescribed datum g lies in the trace range, choose a lifting w with $Tw=g$ and write the unknown as $u=w+v$ with $v\in W^{1,p}_0$. Different liftings change v but not the final boundary constraint. In variational formulations this turns a nonzero-boundary problem into one whose unknown and test functions have homogeneous boundary behavior.

Trace continuity also constrains approximation. A sequence converging in W¹,ᵖ must have traces converging in boundary Lᵖ, so a nonzero trace cannot be approximated in the full norm by compactly supported smooth functions with zero trace. Conversely, Lᵖ convergence inside the domain alone provides no boundary control. Narrow boundary layers can have vanishing volume mass while retaining a fixed endpoint value. Their derivative norms reveal the missing cost. This gives a concrete reason to include derivatives in the energy space rather than regarding the boundary condition as a harmless pointwise annotation.

An H¹ function has first derivatives in L², but an arbitrary L² vector field does not have pointwise boundary values. Consequently one cannot take the trace of the gradient, or a normal derivative, merely from H¹ membership. On a sufficiently smooth boundary, H² regularity supplies additional derivative traces. Another route uses the PDE: if a vector flux lies in the space of square-integrable fields with square-integrable divergence, its normal component has a boundary trace in a negative-order space, defined by a Green identity. This is a weak flux trace rather than ordinary evaluation.

The geometry assumption matters throughout: a Lipschitz boundary provides charts and surface measure compatible with the trace estimate. A cusp or crack may require a different formulation, and boundary pieces of zero surface measure cannot generally carry arbitrary trace constraints through the same Lᵖ argument. Even on smooth domains, a function represented by altered values on a dense volume-null set belongs to the same Sobolev class. Its trace is obtained from the stable operator and interior behavior, so those arbitrary changes cannot alter the prescribed boundary datum.