Classification of anisotropic Triebel-Lizorkin spaces

Sarah Koppensteiner, Jordy Timo van Velthoven, Felix Voigtlaender

TL;DR

This paper establishes a complete classification for anisotropic Triebel-Lizorkin spaces, proving that two different expansive dilation matrices define the same function space if and only if the matrices are geometrically equivalent. The only exception to this rule is the specific case of $L^p$ spaces ($\alpha=0, q=2, p \in (1, \infty)$), where the underlying anisotropy effectively vanishes.

What Problem Does This Paper Address?

In harmonic analysis, we often study function spaces (like Sobolev or Hardy spaces) by looking at how functions behave under different scales. Usually, we scale everything uniformly (isotropic scaling). However, many physical processes and mathematical operators scale differently in different directions—think of a heat equation where one axis diffuses faster than another. This is called anisotropy, and it is modeled using an expansive matrix $A$ (where all eigenvalues have absolute value $> 1$).

The "Triebel-Lizorkin" family is a massive "super-family" of function spaces that includes Lebesgue, Sobolev, and Besov spaces as special cases. While we knew how to define these spaces for a given matrix $A$, a fundamental question remained: If I pick a different matrix $B$, do I get a different space?

Previous work had answered this for Hardy spaces (2003) and Besov spaces (2020). This paper fills the gap for the full scale of Triebel-Lizorkin spaces, providing a "rigidity" theorem that tells us exactly when the choice of matrix matters.

Methods

The authors employ a mix of Fourier analysis and geometric measure theory. Key techniques include:

  • Homogeneous Quasi-norms: They use a function $\rho_A$ that acts as a "distance" measure tailored to the matrix $A$. Two matrices are considered "equivalent" if their associated quasi-norms are equivalent.
  • Homogeneous Covers: The proof relies on analyzing how the Fourier transform of a function is "covered" by dilations of a specific set. They use a criterion (Equation 1.1 in the paper) that measures the overlap between the geometric structures induced by matrices $A$ and $B$.
  • Khintchine’s Inequality: To prove the "necessity" part of the theorem (that the spaces must be different if the matrices aren't equivalent), the authors use Khintchine’s inequality. This probabilistic tool helps show that if the geometries don't match, the norm estimates will fail unless the "square function" parameter $q$ is exactly 2.
  • Peetre-type Maximal Functions: For the "sufficiency" part, they use maximal functions to characterize the spaces, allowing them to compare the norms induced by $A$ and $B$ directly.

Datasets

Not discussed in the paper. As a work of pure mathematics, the "data" consists of abstract function spaces and operator definitions rather than empirical datasets.

Key Insights and Ideas

  • Parameter Rigidity: One of the most striking findings is that you cannot "trade" a change in the matrix for a change in the space parameters ($\alpha, p, q$). If two spaces are equal, their smoothness ($\alpha$), integrability ($p$), and sum-ability ($q$) must be identical.
  • The $L^p$ "Blind Spot": The authors identify that for $q=2$ and $\alpha=0$ (with $p \in (1, \infty)$), the Triebel-Lizorkin space becomes the standard $L^p$ space. Because $L^p$ is defined purely by the size of the function and not its directional derivatives or oscillations, it "doesn't care" about the matrix $A$. This is the only case where anisotropy is irrelevant.
  • Geometric Equivalence: The paper reinforces the idea that the "spectral properties" (the eigenvalues) of the dilation matrix are what truly define the "shape" of the function space. If two matrices stretch space in a way that is bounded relative to each other, they are essentially the same "ruler" for these spaces.

Key Results

  • Theorem 1.1 (Rigidity): If $\dot{\mathbf{F}}^{\alpha_1}_{p_1,q_1}(A) = \dot{\mathbf{F}}^{\alpha_2}_{p_2,q_2}(B)$, then $(p_1, q_1, \alpha_1) = (p_2, q_2, \alpha_2)$.
  • Theorem 1.2 (Classification): For any $\alpha \in \mathbb{R}$ and $p, q \in (0, \infty]$, the spaces $\dot{\mathbf{F}}^\alpha_{p,q}(A)$ and $\dot{\mathbf{F}}^\alpha_{p,q}(B)$ are equal if and only if $A$ and $B$ are equivalent (except for the $L^p$ case mentioned above).
  • Diagonalizability: The authors show that a matrix $B$ is equivalent to a simple scalar dilation (like $A=2I$) if and only if $B$ is diagonalizable over $\mathbb{C}$ and all its eigenvalues have the same absolute value.

Limitations and Future Work

  • Homogeneous vs. Inhomogeneous: This paper focuses on homogeneous spaces (which are defined modulo polynomials). While these are mathematically elegant, many applications in PDE use inhomogeneous spaces. The authors suggest their methods could be adapted, but it is not explicitly handled here.
  • Exponential Matrices: The authors mention that their results allow them to drop the common assumption that $A$ must be an "exponential" matrix ($A = \exp(C)$), simplifying the requirements for future researchers in the field.
  • Variable Anisotropy: The paper deals with a fixed matrix $A$. A more complex frontier is "variable anisotropy," where the scaling matrix $A(x)$ changes depending on the position $x$ in space. This remains a highly active and difficult area of research.

Summary updated

Abstract

This paper provides a classification theorem for expansive matrices $A \in \mathrm{GL}(d, \mathbb{R})$ generating the same anisotropic homogeneous Triebel-Lizorkin space $\dot{\mathbf{F}}^α_{p, q}(A)$ for $α\in \mathbb{R}$ and $p,q \in (0,\infty]$. It is shown that $\dot{\mathbf{F}}^α_{p, q}(A) = \dot{\mathbf{F}}^α_{p, q}(B)$ if and only if the homogeneous quasi-norms $ρ_A, ρ_B$ associated to the matrices $A, B$ are equivalent, except for the case $\dot{\mathbf{F}}^0_{p, 2} = L^p$ with $p \in (1,\infty)$. The obtained results complement and extend the classification of anisotropic Hardy spaces $H^p(A) = \dot{\mathbf{F}}^{0}_{p,2}(A)$, $p \in (0,1]$, in [Mem. Am. Math. Soc. 781, 122 p. (2003)].
BibTeX
@inproceedings{koppensteiner2023classification,
  author = {Sarah Koppensteiner and Jordy Timo van Velthoven and Felix Voigtlaender},
  title = {{Classification of anisotropic Triebel-Lizorkin spaces}},
  booktitle = {Mathematische},
  year = {2023},
  doi = {10.1007/s00208-023-02690-y},
}
Figure