---
title: Local Multi-Norm Hardy Space
url: https://www.emergentmind.com/topics/local-multi-norm-hardy-space
type: topic
---

# Local Multi-Norm Hardy Space

Searching arXiv for the primary paper and closely related Hardy-space frameworks to ground the article in recent literature.
arXiv search query: 2507.09399
arXiv search query: "Littlewood-Paley square functions and the local Hardy space for Multi-Norm Structures on R^d"
The local multi-norm Hardy space \(h^1_{\mathbb E}(\mathbb R^d)\) is a local Hardy space determined by a multi-norm structure on \(\mathbb R^d\), where several anisotropic dilation families interact through a standard matrix \(\mathbb E\). In the formulation developed in "Littlewood-Paley square functions and the local Hardy space for Multi-Norm Structures on \(\mathbb R^d\)" [2507.09399], the space is defined through Littlewood–Paley square functions attached to a frequency decomposition organized by admissible scales, cones, and marked partitions, and it admits equivalent descriptions by square functions, Plancherel–Pólya variants, local multi-norm Riesz transforms, and atoms. The resulting theory recovers the classical local Hardy space in the single-parameter isotropic case, while occupying an intermediate position between one-parameter and product Hardy theories.

## 1. Matrix geometry and multi-norm structures

The ambient space is \(\mathbb R^d\) equipped with anisotropic one-parameter dilations
\[
\delta_t(x)=(t^{\lambda_1}x_1,\dots,t^{\lambda_d}x_d), \qquad \lambda_h>0,
\]
and with a block decomposition
\[
\mathbb R^d=\mathbb R^{d_1}\times\cdots\times \mathbb R^{d_n},
\]
where each factor is dilation-invariant. If \(E_i\subset\{1,\dots,d\}\) is the coordinate index block for \(\mathbb R^{d_i}\), then the homogeneous dimension and homogeneous norm on the \(i\)-th factor are
\[
q_i=\sum_{h\in E_i}\lambda_h,
\qquad
|x_i|=\sum_{h\in E_i}|x_h|^{1/\lambda_h}.
\]
A standard matrix \(\mathbb E=\{e(i,j)\}\in \mathbb R^{n\times n}\) satisfies three conditions: \(e(j,j)=1\); \(e(j,\ell)\le e(j,k)e(k,\ell)\) for all \(j,k,\ell\); and \(1<e(j,k)e(k,j)\) for \(j\neq k\). These conditions encode the coupled anisotropies that distinguish the multi-norm setting from a product of unrelated one-parameter structures [2507.09399].

Given \(\mathbb E\), one defines \(n\) families of dilations on physical and frequency space,
\[
\delta_t^i(x)=\big(t^{1/e(i,1)}x_1,\dots,t^{1/e(i,n)}x_n\big),
\qquad
\widehat\delta_t^i(\xi)=\big(t^{e(1,i)}\xi_1,\dots,t^{e(n,i)}\xi_n\big),
\]
together with the multi-norm gauges
\[
N_i(x)=\max_{1\le j\le n}|x_j|^{e(i,j)},
\qquad
\widehat N_i(\xi)=\max_{1\le j\le n}|\xi_j|^{1/e(j,i)}.
\]
The kernel class \(S_0(\mathbb E)\) and multiplier class \(M_\infty(\mathbb E)\) are controlled by these gauges. If \(\mathcal K\in S_0(\mathbb E)\), then on \(\mathbb R^d\setminus\{0\}\) it is represented by a smooth function \(K\) satisfying
\[
|\partial^\alpha K(x)|
\le
C_{\alpha,N}
\Big[\prod_{i=1}^n N_i(x)^{-q_i-\langle\alpha\rangle_i}\Big](1+\|x\|)^{-N},
\]
where
\[
\langle\alpha\rangle_i=\sum_{j=1}^n e(j,i)\langle\alpha^j\rangle,
\qquad
\langle\alpha^j\rangle=\sum_{h\in E_j}\lambda_h\alpha_h.
\]
Dually, if \(m\in M_\infty(\mathbb E)\), then
\[
|\partial^\alpha m(\xi)|
\le
C_\alpha\prod_{i=1}^n (1+\widehat N_i(\xi))^{-\langle\alpha^i\rangle}.
\]
Theorem 3.9 gives equivalent descriptions of \(M_\infty(\mathbb E)\) and \(S_0(\mathbb E)\) via dyadic decompositions in frequency space and via kernel expansions with prescribed cancellations aligned with \(\mathbb E\) and the dotted variables. This geometry is the structural substrate on which the local Hardy space is built.

## 2. Littlewood–Paley decomposition adapted to \(\mathbb E\)

The frequency decomposition is organized by the logarithmic map
\[
\pi(\xi)=\big(\log_2^+(|\xi_1|),\dots,\log_2^+(|\xi_n|)\big),
\]
tubes \(T_L\subset \mathbb R_+^n\) around integer cubes \(Q_L\), and dyadic blocks \(B_L=\pi^{-1}(T_L)\). The local Littlewood–Paley decomposition has the form
\[
\widehat{\mathbb R}^d = B(1)\cup \bigcup_{L\in\mathcal L} B_L,
\]
where \(\mathcal L\subset \mathbb Z^n\) is the set of admissible scales. The decomposition is refined by marked partitions
\[
S=\{(A_1,k_1),\dots,(A_s,k_s)\},
\]
which encode dominance relations of components in the gauges \(\widehat N_i\). Each marked partition determines a cone \(\Gamma_S\subset \mathbb R_+^n\) and a face \(F_S\); the principal cone is \(\Gamma(\mathbb E)\) [2507.09399].

For \(L\in \mathcal L_S\), Proposition 3.8.5 gives a quasi-product description of \(B_L\). In qualitative form, each block behaves like a product of an annulus in the dotted variable \(k_r\) and a ball in the remaining \(A_r\)-variables, at scales prescribed by \(L\). This point is decisive: the frequency tiles are neither simple products of annuli nor single-parameter annuli, but are instead adapted to the dominance pattern imposed by \(\mathbb E\).

Two kinds of Littlewood–Paley families are constructed. The tensor type starts from one-parameter families \(\{\phi_{i,\ell}\}\) in each \(\mathbb R^{d_i}\), forms \(\Psi_L\) as a product over \(i\), and imposes cancellations in dotted variables. The convolution type starts from families \(\{\rho_{i,\ell},\sigma_{i,\ell}\}\) adapted to \(\widehat\delta^i\) and defines
\[
\Sigma_L^{(-L)}
=
\Big(\ast_{i\in D_L}\sigma_{i,\ell_i}^{(-\ell_i)_i}\Big)
\ast
\Big(\ast_{i\notin D_L}\rho_{i,\ell_i}^{(-\ell_i)_i}\Big).
\]
Both constructions satisfy reproducing formulas; for instance, in the tensor case,
\[
\delta=\sum_{L\in\mathcal L}\Psi_L^{(-L)},
\qquad
\delta=\sum_{L\in\mathcal L}\Psi_L^{(-L)}\ast \widetilde\Psi_L^{(-L)}.
\]

For any Littlewood–Paley family \(\Xi=\{\Xi_L\}\), the associated square function is
\[
S_\Xi f(x)=\Big(\sum_{L\in\mathcal L}|f\ast \Xi_L^{(-L)}(x)|^2\Big)^{1/2}.
\]
Its local character is intrinsic: \(\mathcal L\) indexes scales near \(\Gamma(\mathbb E)\), the blocks \(B_L\) are large away from the origin, the sums start at \(\ell_i\ge 0\), and the unit ball \(B(1)\) is retained. The main structural result is that these square functions are \(L^1\)-equivalent. Theorem 5.1 proves the tensor-type equivalence, Theorems 6.2 and 6.4 compare convolution-type and tensor-type square functions, Theorem 6.11 identifies them with the full-parameter local convolution square function, and Theorem 7.1 proves equivalence with Plancherel–Pólya blockwise supremum variants.

## 3. Definition and elementary properties of \(h^1_{\mathbb E}(\mathbb R^d)\)

The local multi-norm Hardy space is defined by square functions:
\[
h^1_{\mathbb E}(\mathbb R^d)
:=
\{f\in \mathcal S'(\mathbb R^d): S_\Xi f\in L^1(\mathbb R^d)\},
\qquad
\|f\|_{h^1_{\mathbb E}}:=\|S_\Xi f\|_{L^1}.
\]
Because of the \(L^1\)-equivalence theorems, this definition is independent of the particular choice of \(\Xi\). Low frequency \(B(1)\) is handled by the inhomogeneous family, while the remaining blocks are supported away from the origin in frequency but near \(\Gamma(\mathbb E)\). The space is therefore local in the same sense as Goldberg’s theory: it measures small-scale oscillation while retaining an inhomogeneous low-frequency component [2507.09399].

Several basic properties are immediate from the square-function construction. Theorem 8.1 shows
\[
h^1_{\mathbb E}\subset L^1(\mathbb R^d),
\qquad
\|f\|_{L^1}\le C\|f\|_{h^1_{\mathbb E}}.
\]
Lemma 8.2 states that \(h^1_{\mathbb E}\cap L^2(\mathbb R^d)\) is dense in \(h^1_{\mathbb E}\). Proposition 8.4 proves invariance under multiplication by compactly supported smooth cutoffs, and Corollary 8.5 yields a partitioning argument. These facts confirm that the construction is genuinely local rather than merely inhomogeneous in a formal sense.

A common misidentification is to regard \(h^1_{\mathbb E}\) as either a standard anisotropic Hardy space attached to one dilation or a pure product Hardy space. The paper explicitly places multi-norm structures between one-parameter and product theories: they capture compositions of Calderón–Zygmund operators adapted to distinct dilation families via the \(\mathbb E\)-geometry, and the marked partitions reflect anisotropic dominance across components, not independence. Unlike pure product theory, the multi-norm blocks \(B_L\) are coarser than products of annuli but finer than any single one-parameter decomposition.

## 4. Equivalent characterizations and atomic decomposition

The square-function definition is only one of several equivalent descriptions. Theorem 7.1 gives a Plancherel–Pólya characterization using dyadic rectangles in \(\mathbb R^d\). Theorem 8.9 gives a Riesz-transform characterization: \(f\in h^1_{\mathbb E}\) if and only if all convolutions
\[
f\ast R_J\in L^1(\mathbb R^d),
\qquad
R_J=\mathcal K^\flat_{1,j_1}\ast\cdots\ast \mathcal K^\flat_{n,j_n},
\]
where \(\mathcal K^\flat_{i,j}\) are the local Calderón–Zygmund kernels characterizing \(h^1_{\widehat\delta^i}\). Moreover,
\[
\|S_\Xi f\|_1 \simeq \sum_J \|f\ast R_J\|_1.
\]
This characterization links the abstract square-function norm to a concrete family of local singular integrals [2507.09399].

The atomic theory is indexed by marked partitions \(S\in \mathcal S_{\mathbb E}^*=\mathcal S_{\mathbb E}\cup\{\emptyset\}\) and a parameter \(\tau>0\). There are three atom types.

| Type | Support and size | Cancellation |
|---|---|---|
| \(S=\emptyset\) | supported in a translate of \(2^\tau B(1)\), \(\|a\|_{L^2}\le 1\) | none |
| \(s=1\) | supported in \(2^\tau\circ \mathbf R\), \(\|a\|_{L^2}\le |\mathbf R|^{-1/2}\) | \(\int_{\mathbb R^{d_k}} a\,dx_k=0\) |
| \(s\ge 2\) | special product atom on \(\Omega\subset 2^\tau\circ Q\), \(\|a\|_{L^2}\le |\Omega|^{-1/2}\) | cancellation in each dotted variable \(k_r\) |

For \(s=1\), the supporting set is a dyadic rectangle at admissible scale \(L\in\mathcal L_S\), and dyadic rectangles coincide with dyadic cubes for the one-parameter dilations \(\delta^k\). For \(s\ge 2\), the atoms are “special Chang–Fefferman atoms”: cancellations are in dotted coordinates, not whole blocks. The cancellation order is minimal—integral zero in the dotted variables—consistent with local \(h^1\) frameworks, while higher cancellations enter the reproducing formulas rather than the atom definition.

Theorem 9.3 establishes the atomic decomposition: for any \(\tau>0\), there exists \(C_\tau\) such that every \(f\in h^1_{\mathbb E}(\mathbb R^d)\) can be written
\[
f=\sum_{j=0}^\infty \lambda_j a_j,
\]
where each \(a_j\) is an \((S_j,\tau)\)-atom for some \(S_j\in \mathcal S_{\mathbb E}^*\), and
\[
\sum_j |\lambda_j|\le C_\tau \|f\|_{h^1_{\mathbb E}}.
\]
The proof uses the multi-norm Calderón reproducing formulas, Plancherel–Pólya square functions, and a multi-parameter Journé covering lemma. The converse inclusion is also proved: Theorem 11.7 and Proposition 11.9 show that every \((S,\tau)\)-atom belongs to \(h^1_{\mathbb E}\) with a uniform norm bound.

## 5. Singular integrals, multipliers, and model cases

The space \(h^1_{\mathbb E}\) is stable under the singular integral operators naturally attached to the multi-norm geometry. Theorem 8.6 states that if \(\mathcal K\in S_0(\mathbb E)\) is a multi-norm singular kernel, then
\[
Tf=f\ast \mathcal K
\]
is bounded on \(h^1_{\mathbb E}(\mathbb R^d)\), with operator norm controlled by finitely many kernel constants from Definition 3.1. This makes \(h^1_{\mathbb E}\) the correct endpoint space for the local multi-norm Calderón–Zygmund class [2507.09399].

A principal source of examples comes from compositions of operators adapted to distinct dilation families. Theorem 3.10, recalled from earlier work in Section 3.5, states that if \(\mathcal K_i\) are local smooth Calderón–Zygmund kernels adapted to \(\widehat\delta^i\), \(i=1,\dots,n\), then
\[
m=\prod_{i=1}^n \widehat{\mathcal K_i}\in M_\infty(\mathbb E),
\qquad
\mathcal K_1\ast\cdots\ast \mathcal K_n\in S_0(\mathbb E).
\]
In particular, compositions of \(n\) Mihlin–Hörmander multipliers adapted to \(\widehat\delta^i\) define operators bounded on \(h^1_{\mathbb E}\) via Theorem 8.6.

The theory includes several limiting and illustrative cases. When \(n=1\) and \(\delta_t\) acts isotropically, \(\lambda_h\equiv 1\), the multi-norm structure reduces to the standard local Hardy space \(h^1(\mathbb R^d)\) in Goldberg’s sense; the square functions become classical inhomogeneous Littlewood–Paley square functions, and the atoms become local Goldberg atoms. At the other end, the two-dilation case already exhibits the essential geometry. If
\[
\mathbb E=
\begin{bmatrix}
1 & a\\
b & 1
\end{bmatrix},
\qquad ab>1,
\]
then
\[
\widehat\delta^1_r(\xi_1,\xi_2)=(r\xi_1,r^b\xi_2),
\qquad
\widehat N_1(\xi)=\max\{|\xi_1|,|\xi_2|^{1/b}\},
\]
and
\[
\widehat\delta^2_r(\xi_1,\xi_2)=(r^a\xi_1,r\xi_2),
\qquad
\widehat N_2(\xi)=\max\{|\xi_1|^{1/a},|\xi_2|\}.
\]
The marked partitions are \(S_p=\{\dot 1\}\{\dot 2\}\), \(S_1=\{\dot 1,2\}\), and \(S_2=\{1,\dot 2\}\), with regions
\[
\widehat E_{S_p}: |\xi_1|^{1/a}\le |\xi_2|\le |\xi_1|^b,
\]
\[
\widehat E_1: |\xi_1|\ge |\xi_2|^a,
\qquad
\widehat E_2: |\xi_2|\ge |\xi_1|^b.
\]
The frequency space is partitioned into the unit square plus rectangles aligned with these regions, and the corresponding square functions impose cancellation in the dotted variables according to the marked partition.

## 6. Technical features, surrounding literature, and unsettled directions

The general \(n\)-dilation case is described as considerably harder than the recent \(2\)-dilation work, and the analysis proceeds through a systematic organization of frequency tiles by marked partitions and cones. Lemma 3.7 gives a tube cover of \(\mathbb R_+^n\) with controlled overlaps, Lemma 3.8 places admissible scales within bounded distance of the relevant cone, and the proofs of \(L^1\)-equivalence use high-order cancellation of the auxiliary families, “borrowing” decay across coordinates to dominate mixed interactions, and vector-valued Fefferman–Stein inequalities. The Plancherel–Pólya comparison is obtained by dyadic rectangles, while the atomic theorem relies on a multi-parameter Journé covering lemma and the Carbery–Seeger proposition from the “journe” section [2507.09399].

The broader Hardy-space landscape shows several parallel local theories, but they encode locality in different ways. Anisotropic mixed-norm Hardy spaces \(H_{\vec a}^{\vec p}(\mathbb R^n)\) are defined by anisotropic grand maximal functions and admit atomic and Campanato dual descriptions [1804.05558]. Local Musielak–Orlicz Hardy spaces \(h_\phi(\mathbb R^n)\) and weighted local Orlicz–Hardy spaces \(h^\Phi_\omega(\mathbb R^n)\) are defined by local grand maximal functions, admit atomic decompositions, and have local \(\mathrm{BMO}\)-type duals [1108.2797], [1107.3266]. Local Hardy spaces associated with ball quasi-Banach spaces and non-negative self-adjoint operators use local Lusin area functions together with low-frequency control, and mixed-norm local Hardy spaces appear იქ in the operator-adapted setting [2503.03301], while multilinear pseudo-differential operators are studied on local Hardy spaces associated with ball quasi-Banach function spaces, including local mixed-norm Hardy spaces \(h^{\vec p}\) [2503.23592]. These comparisons suggest that the local multi-norm Hardy space belongs to a wider family of local endpoint constructions, but its distinctive feature is the \(\mathbb E\)-driven interaction of several anisotropies through marked partitions and dotted-variable cancellation rather than through a single anisotropic metric, a Musielak–Orlicz modular, or an operator semigroup.

One notable omission is duality. The paper develops a characterization via local multi-norm Riesz transforms, but it does not state explicitly the identification of \((h^1_{\mathbb E})^*\) with a local \(\mathrm{BMO}_{\mathbb E}\)-type space. The techniques and references suggest such a duality is plausible; however, it is not claimed as a theorem. In that sense, the theory is structurally complete at the level of square functions, atoms, and singular integrals, while the dual endpoint picture remains an evident direction for further development.

Source: https://www.emergentmind.com/topics/local-multi-norm-hardy-space