Failure of zero extension in parabolic Sobolev spaces

Technical Summary: Failure of Zero Extension in Parabolic Sobolev Spaces

Problem Statement
The paper addresses a fundamental question regarding the functional analytic properties of solutions to parabolic partial differential equations (PDEs). Specifically, it investigates whether the spatial zero extension of a function $u$ vanishing on the lateral boundary of a domain $\Omega$ preserves its membership in the parabolic Sobolev space $\mathring{H}^1_p((0, T) \times \Omega)$.

In the context of elliptic equations, it is a standard result that the zero extension of a function $u \in \mathring{W}^1_p(\Omega)$ belongs to $W^1_p(\mathbb{R}^d)$. This property is crucial for reducing boundary value problems to whole-space problems, thereby enabling the use of global tools such as Fourier analysis and Sobolev embeddings. The authors ask whether an analogous property holds for parabolic spaces, where the time derivative is treated in a distributional sense (specifically, $u_t = D_i g_i + f$). The central question (Q) is: Does the zero extension of a function $u(t, x)$ with zero lateral trace remain in the same parabolic Sobolev class when extended by zero to the whole space?

Methodology
The authors employ a constructive approach to disprove the extension property for $p \in [1, \infty)$. Their methodology involves:

1. Construction of a Counterexample: They explicitly construct a function $u \in \mathring{H}^1_p((0, T) \times \mathbb{R}^d_+)$ defined on the half-space. The function is designed to exhibit a self-similar boundary layer concentrated at the initial-boundary corner $(t, x) = (0, 0)$.

   • For $d=1$, the function is defined as $u(t, x) = \partial_x v(t, x)$, where $v(t, x) = t^{-a} e^{-x^2/t}$.
   • The exponent $a$ is carefully chosen to ensure that $u$ and its spatial derivatives belong to $L^p$, and that the time derivative $u_t$ admits a divergence structure $u_t = \partial_x g$ with $g \in L^p$ on the half-line.

2. Analysis of Singularities: The authors analyze the scaling behavior of the constructed function as $t \to 0$. They demonstrate that while the singularity is "harmless" on the half-line due to the divergence structure (absorbing the boundary flux), the spatial zero extension creates a non-integrable boundary-supported flux defect.

   • Upon extending $u$ by zero to $\mathbb{R}$, the distributional time derivative $\bar{u}_t$ formally acquires a term proportional to $g(t, 0)\delta_0$.
   • For $p < \infty$, the value $g(t, 0)$ behaves like $t^{-a-1}$ (or similar singular powers depending on $p$), which is not integrable near $t=0$.

3. Proof by Contradiction: The authors assume that the zero extension $\bar{u}$ belongs to $H^1_p((0, T) \times \mathbb{R}^d)$. This implies the existence of functions $G_i, F \in L^p$ such that $\bar{u}_t = D_i G_i + F$. By testing this identity against specific test functions $\phi(t, x) = \eta_m(t)\varphi(x)$ that concentrate near the corner $(0,0)$, they derive a contradiction. The left-hand side of the integral identity diverges due to the temporal singularity, while the right-hand side remains bounded by the assumed $L^p$ norms of $G$ and $F$.

4. Extension to General Domains and Mixed Norms: The result is generalized to arbitrary Lipschitz domains via boundary flattening arguments and to mixed-norm spaces $H^1_{p,q}$ by adjusting the scaling parameters.

Key Contributions and Results

• Theorem 1.1: For any $p \in [1, \infty)$, there exists a function $u \in \mathring{H}^1_p((0, T) \times \mathbb{R}^d_+)$ such that its spatial zero extension $\bar{u}$ does not belong to $H^1_p((0, T) \times \mathbb{R}^d)$. Specifically, while $\bar{u}$ and $D\bar{u}$ are in $L^p$, the time derivative $\bar{u}_t$ cannot be represented as $D_i G_i + F$ with $G_i, F \in L^p$.
• Mechanism of Failure: The failure is attributed to a "self-similar boundary layer" concentrated at the initial-boundary corner. The zero extension creates a flux defect supported on the boundary that is too singular to be captured by the $L^p$ integrability of the time derivative's divergence form.
• The Case $p = \infty$: The authors prove that the extension property does hold for $p = \infty$. In this case, the singularity at the corner is bounded, and the zero extension remains in $H^1_\infty$.
• Comparison of Function Spaces: The paper provides a detailed comparison of three Sobolev-type spaces used in parabolic theory:
  • $H^1_p$: The natural space for divergence-form equations with bounded coefficients. Zero extension fails.
  • $V_p$ (or $L^p(0, T; W^1_p) \cap L^\infty(0, T; L^p)$): Zero extension holds, but elements in this space cannot always be realized as weak solutions to divergence-form parabolic equations.
  • $H^{1/2, 1}_p$: A space involving a half-order time derivative, proposed as a suitable framework for equations with unbounded lower-order terms, where zero extension holds.

Significance and Claims
The paper claims to provide the first explicit construction and rigorous proof of the failure of spatial zero extension in standard parabolic Sobolev spaces $\mathring{H}^1_p$ for $p < \infty$. While previous literature (e.g., [19]) had hinted at this phenomenon, no detailed justification or counterexample was available.

The significance of this result lies in its implications for the solvability theory of parabolic equations:

1. Limitations of Global Tools: The failure of zero extension means that standard techniques relying on reducing boundary value problems to whole-space problems (such as global Fourier analysis or standard Sobolev embeddings on $\mathbb{R}^d$) cannot be directly applied to functions in $\mathring{H}^1_p$ with zero lateral boundary conditions.
2. Suitability of Solution Spaces: The results highlight that $\mathring{H}^1_p$ is the "optimal" space for solvability with bounded lower-order terms, but its structural properties differ fundamentally from the elliptic case. Conversely, spaces like $V_p$, while allowing zero extension, are shown to be too large to serve as solution spaces for divergence-form equations, as they contain functions that do not satisfy the necessary distributional time derivative structure.
3. Corner Singularities: The work underscores the critical role of the initial-boundary corner in parabolic regularity, showing that singularities concentrated there can destroy the global integrability of the time derivative upon extension, even when the boundary is flat.

The authors conclude that the choice of function space for parabolic problems must be made with care, balancing the need for zero extension properties against the requirement that the space accurately represents weak solutions to the specific class of equations being studied.