Poincaré and Friedrichs inequalities

For a bounded open set $\Omega$ and $1\le p<\infty$, the zero-boundary space $W^{1,p}_0(\Omega)$ satisfies $\|u\|_{L^p}\le C\|\nabla u\|_{L^p}$. One can prove this by zero-extending compactly supported smooth functions into a containing box, applying the fundamental theorem along a coordinate direction, and passing to the closure. No smooth boundary is needed for this zero-closure formulation. The bounded containing box supplies a length scale; without a size restriction a uniform inequality on the whole space fails by dilation.

For a bounded connected Lipschitz domain, subtracting the average gives another version: $\|u-u_\Omega\|_{L^p}\le C\|\nabla u\|_{L^p}$. Connectedness matters because a function with zero gradient is only constant on each connected component. One overall average cannot remove several independent componentwise constants. The Lipschitz hypothesis is a convenient sufficient setting for a uniform domain inequality; arbitrary connected open sets with very thin connections need not enjoy the same control. Boundary normalization and mean normalization are different subspaces, even though both remove the constant ambiguity.

Rescale a domain by a factor L. The function Lᵖ norm gains a factor $L^{n/p}$, while its gradient norm gains $L^{n/p-1}$. Thus the Poincaré constant has the units and scaling of length. A numerical constant from a unit interval cannot be reused unchanged on a long interval. On $(0,L)$ with two zero endpoints and $p=2$, the sharp constant is $L/\pi$. Expanding in Dirichlet sine modes proves the inequality, and the first sine mode attains equality. This special exact constant depends on the geometry and boundary condition.

For the Dirichlet energy $a(u,u)=\int|\nabla u|^2$, Poincaré gives coercivity on $H^1_0$: gradient energy controls the full H¹ norm up to a domain constant. This makes the energy appropriate for uniqueness and existence arguments in weak elliptic equations. Without a normalization, constants have zero energy and uniqueness may fail. In a pure Neumann problem, one commonly works modulo constants or in the zero-mean subspace and also checks the forcing compatibility condition. An inequality cannot repair an incompatible equation merely by changing the norm.

A Friedrichs-type variant controls a function by its gradient when its trace vanishes on a boundary subset of positive surface measure, on a bounded connected Lipschitz domain. The constant depends on the domain and the constrained subset. If that subset shrinks, one cannot assume the same constant stays valid. A related estimate includes a boundary norm instead of imposing zero data. Terminology varies across references, so specify the actual formula and subspace rather than relying on whether a source calls the estimate Poincaré, Friedrichs, or Poincaré–Steklov.

A direct contradiction proof for the zero-mean inequality illustrates the roles of the assumptions. If no constant existed, normalize a sequence to have Lᵖ norm one while its gradients tend to zero and its means remain zero. Rellich compactness on a bounded Lipschitz domain gives an Lᵖ-convergent subsequence. The limit has zero weak gradient, hence is constant by connectedness; its mean forces that constant to zero, contradicting unit norm. This proof uses a compactness theorem, not merely the completeness of the Sobolev space. A disconnected domain invalidates the constant-identification step.