---
title: Semicommutative Hardy Spaces
url: https://www.emergentmind.com/topics/semicommutative-hardy-spaces
type: topic
---

# Semicommutative Hardy Spaces

Searching arXiv for the cited papers and closely related semicommutative Hardy-space work.
Semicommutative Hardy spaces are Hardy-type function spaces in which the underlying geometry remains classical while the function values lie in a noncommutative algebra, typically a von Neumann algebra with trace. The basic ambient object is a semicommutative algebra of the form \(L_\infty(X)\,\bar\otimes\,M\) or \(L_\infty(\Omega)\,\bar\otimes\,\mathcal M\), where the \(L_\infty\)-factor is commutative and the \(M\)- or \(\mathcal M\)-factor may be genuinely noncommutative. In the recent literature, the term encompasses several distinct but related theories: projection-defined Hardy spaces driven by interpolation and singular integrals, row/column and conditioned martingale Hardy spaces attached to filtrations, and operator-valued Hardy–BMO spaces on spaces of homogeneous type defined through Lusin area integrals [2604.23864], [1301.2071], [2311.15775]. This suggests that “semicommutative Hardy space” is best understood as a family of operator-valued Hardy theories rather than a single canonical construction.

## 1. Ambient framework and basic meaning

In the semicommutative setting of singular-integral and interpolation theory, one starts with a semifinite von Neumann algebra \(M\) equipped with a normal semifinite faithful trace \(\tau\), a measure space \(X\), and the tensor-product algebra
\[
N:=L_\infty(X)\,\bar\otimes\, M,
\]
equipped with the tensor-product trace \(\sigma\). Elements of \(N\) are essentially bounded \(M\)-valued functions on \(X\), and \(L_p(N)\) is the corresponding noncommutative \(L_p\)-space. The framework is called semicommutative because the \(L_\infty(X)\)-factor is commutative while the \(M\)-factor may be genuinely noncommutative [2604.23864].

In the martingale setting, the same structural idea appears with
\[
\mathcal N = L_\infty(\Omega)\,\bar\otimes\,\mathcal M,
\]
equipped with a filtration \((\mathcal N_t)\) coming from a classical filtration on \(\Omega\) or from mixed classical/operator-valued filtrations. Here semicommutative Hardy spaces arise as special cases of Hardy spaces over finite von Neumann algebras with trace-preserving conditional expectations [1301.2071].

On spaces of homogeneous type, the ambient semicommutative algebra is
\[
\mathcal N=L_\infty(\mathbb X)\,\overline\otimes\,\mathcal M,
\]
where \((\mathbb X,d,\mu)\) is a space of homogeneous type in the sense of Coifman–Weiss and \(\mathcal M\) is a von Neumann algebra equipped with a normal semifinite faithful trace \(\tau\). In that theory, geometry is classical, algebraic size and duality are noncommutative, and the Hardy and BMO spaces come in column, row, and mixture forms [2311.15775].

A recurrent clarification is that semicommutativity does not mean one is working with the rotation or noncommutative torus in Rieffel’s sense. For the torus model in the projection-based theory, the algebra is
\[
N = L_\infty(\mathbb T^d)\,\bar\otimes\, M,
\]
so the spatial variable is the classical torus and only the values are operator-valued [2604.23864].

## 2. Principal models of semicommutative Hardy spaces

The recent literature supports at least three principal models.

| Framework | Ambient algebra | Defining mechanism |
|---|---|---|
| Projection-based Hardy spaces | \(L_\infty(X)\bar\otimes M\) | Weak closure of \(E(M)\cap P(L_2(M))\) |
| Martingale Hardy spaces | \(L_\infty(\Omega)\bar\otimes\mathcal M\) | Row/column/conditioned square functions over filtrations |
| Operator-valued Hardy spaces on homogeneous type | \(L_\infty(\mathbb X)\overline\otimes\mathcal M\) | Lusin area functions from a Calderón reproducing family |

In the projection-based theory, if \(P\) is a self-adjoint projection on \(L_2(M)\) and \(E(M)\) is an exact interpolation space for \((L_1(M),L_\infty(M))\), the Hardy space is defined by
\[
H_E(P)
\]
as the \(\sigma(E(M),E^\times(M))\)-weak closure in \(E(M)\) of
\[
E(M)\cap P(L_2(M)).
\]
For \(E=L_p\), one writes
\[
H_p(P):=H_{L_p}(P), \qquad 1\le p\le \infty,
\]
and by definition
\[
H_2(P)=P(L_2(M)).
\]
These are abstract Hardy spaces associated with \(P\). The construction is explicitly not based on square functions, maximal functions, row/column structures, conditioned martingale norms, or subdiagonal algebras as the primary model [2604.23864].

In the martingale theory for continuous filtrations, the discrete column and row Hardy norms are
\[
\|x\|_{H_p^c} = \Big\|\Big(\sum_n |d_n(x)|^2\Big)^{1/2}\Big\|_p, \qquad \|x\|_{H_p^r} = \Big\|\Big(\sum_n |d_n(x^*)|^2\Big)^{1/2}\Big\|_p,
\]
with conditioned norms
\[
\|x\|_{h_p^c} = \Big\|\Big(\sum_n \mathcal E_{n-1}|d_n(x)|^2\Big)^{1/2}\Big\|_p, \qquad \|x\|_{h_p^r}=\|x^*\|_{h_p^c},
\]
and diagonal norm
\[
\|x\|_{h_p^d} = \Big(\sum_n \|d_n(x)\|_p^p\Big)^{1/p}.
\]
For continuous filtrations, these norms are replaced by partition limits and ultrafilter constructions based on the brackets
\[
[x,x]_\sigma=\sum_{t\in\sigma}|d_t^\sigma(x)|^2,
\qquad
\langle x,x\rangle_\sigma = \sum_{t\in\sigma}\mathcal E_{t^-(\sigma)}|d_t^\sigma(x)|^2.
\]
The corresponding spaces are denoted \(\mathcal H_p^c,\mathcal H_p^r,h_p^c,h_p^r,h_p^d\) [1301.2071].

In the operator-valued Hardy–BMO theory on spaces of homogeneous type, Hardy spaces are defined through Lusin area functions. For an \(\mathcal M\)-valued simple function \(f\), with cones
\[
\Gamma_x=\{(y,t)\in \mathbb X\times \mathbb R_+:\ d(x,y)<t\},
\]
the column and row area functions are
\[
S^c(f)(x) := \left( \int_{\Gamma_x} |D_t(f)(y)|^2 \frac{d\mu(y)\,dt}{V_t(x)\,t} \right)^{1/2},
\]
\[
S^r(f)(x) := \left( \int_{\Gamma_x} |(D_t(f)(y))^*|^2 \frac{d\mu(y)\,dt}{V_t(x)\,t} \right)^{1/2}.
\]
The norms are
\[
\|f\|_{H_p^c(\mathbb X,\mathcal M)} := \|S^c(f)\|_{L_p(\mathcal N)}, \qquad \|f\|_{H_p^r(\mathbb X,\mathcal M)} := \|S^r(f)\|_{L_p(\mathcal N)},
\]
and the mixture space is defined by
\[
H_p^{cr}=H_p^c+H_p^r \quad (1\le p<2), \qquad H_p^{cr}=H_p^c\cap H_p^r \quad (2\le p<\infty).
\]
At \(p=2\),
\[
H_2^c(\mathbb X,\mathcal M) = H_2^r(\mathbb X,\mathcal M) = H_2^{cr}(\mathbb X,\mathcal M) = L_2(\mathcal N)
\]
[2311.15775].

## 3. Spectral and analytic Hardy spaces on the torus

The torus provides the main concrete model for projection-defined semicommutative Hardy spaces. Let
\[
N=L_\infty(\mathbb T^d)\bar\otimes M.
\]
With Fourier basis \((\chi_n)_{n\in\mathbb Z^d}\), where
\[
\chi_n(z)=z^n=z_1^{n_1}\cdots z_d^{n_d},
\]
the \(M\)-valued Fourier coefficients are defined by
\[
\sigma(f(x\otimes \chi_n))=\tau(\widehat f(n)x),
\qquad
x\in (L_1\cap L_\infty)(M),\ n\in\mathbb Z^d.
\]
Given \(I\subset \mathbb Z^d\), the spectral projection is
\[
P(f)=\sum_{n\in I} \chi_n\otimes \widehat f(n) \quad\text{in }L_2(N),
\]
and the associated Hardy-type subspace is
\[
H(P):=\{f\in (L_1+L_\infty)(N):\ \forall n\notin I,\ \widehat f(n)=0\}.
\]
For exact interpolation spaces \(E(N)\) with order-continuous norm, one has
\[
H_E(P)=E(N)\cap H(P), \qquad H_{E^\times}(P)=E^\times(N)\cap H(P),
\]
and \((L_1\cap L_\infty)(N)\cap H(P)\) is dense in the appropriate norm or \(w^*\)-topology. Thus the abstract Hardy space associated with a spectral projection is exactly the Fourier-support-defined subspace one expects [2604.23864].

For \(1\le p\le \infty\), the analytic Hardy space on the torus is defined as the subspace \(H_p(N)\subset L_p(N)\) consisting of all \(f\) such that
\[
\int_{\mathbb T^d} f(z) z^n\,dz=0 \qquad \text{for every } n\in \mathbb N_+^d.
\]
Equivalently, Fourier coefficients vanish on the “negative” spectrum indexed by \(\mathbb N_+^d\) in the convention of the paper. In dimension \(d=1\), this is the standard operator-valued Hardy space of analytic \(M\)-valued functions on the circle [2604.23864].

For \(d=1\) and \(I=\mathbb N_+\), the spectral projection \(P\) is the Riesz projection, whose kernel is
\[
k(z,w)=\frac{w}{w-z},\qquad z,w\in \mathbb T,\ z\neq w,
\]
or equivalently
\[
k(p(x),p(y))=\frac12+\frac{i}{2}\cot\!\big((x-y)/2\big).
\]
This kernel is singular on \(\mathbb T\), so the general semicommutative singular-integral theory applies to analytic operator-valued Hardy spaces on the circle [2604.23864].

A distinct but related commutative model appears in the theory of symmetric norms. For a continuous rotationally symmetric norm \(\alpha\), one defines
\[
H^\alpha(\mathbb T)=\overline{\operatorname{span}\{1,z,z^2,\dots\}}^{\,\alpha},
\]
and the Fourier characterization is
\[
H^\alpha(\mathbb T)
=
\left\{
F\in L^\alpha(\mathbb T):\ \widehat F(n)=0,\ \forall n<0
\right\}.
\]
Also,
\[
H^\alpha(\mathbb T)=L^\alpha(\mathbb T)\cap H^1(\mathbb T).
\]
The paper does not define a noncommutative Hardy space \(H^\alpha(\mathcal A)\) for a subdiagonal algebra, but it presents this theory as a blueprint for such an extension [1407.7920].

## 4. Duality, interpolation, and \(K\)-closedness

A major theme of semicommutative Hardy-space theory is that Hardy subspaces behave stably under interpolation and admit BMO-type duals.

In the projection-based theory, the organizing tool is the Peetre \(K\)-functional
\[
K_t(u,E_0,E_1):= \inf\Big\{\|u_0\|_{E_0}+t\|u_1\|_{E_1}:\ u=u_0+u_1,\ u_0\in E_0,\ u_1\in E_1\Big\},
\]
together with the notions of \(K\)-closedness and quasi-complementation. If \(P\) is a Jones-type projection on \(L_2(M)\), then for every \(1\le p_0,p_1\le \infty\),
\[
(H_{p_0}(P),H_{p_1}(P))
\]
is quasi-complemented in
\[
(L_{p_0}(M),L_{p_1}(M)),
\]
with constant depending only on \(p_0,p_1\). Consequently the Hardy couple is \(K\)-closed in the ambient \(L_p\)-couple. If \(\Phi\) is a \(K\)-parameter such that
\[
E(M)=K_\Phi(L_{p_0}(M),L_{p_1}(M))
\]
has order-continuous norm, then
\[
H_E(P)=K_\Phi(H_{p_0}(P),H_{p_1}(P))
\]
with equivalent norms; in particular,
\[
H_p(P)= (H_1(P),H_\infty(P))_{\theta,p,K},
\qquad
\theta=1-\frac1p,\quad 1<p<\infty.
\]
For noncommutative \(L_p\)-spaces, Holmstedt’s formula
\[
K_t(x,L_1(M),L_\infty(M)) = \int_0^t \mu_x(s)\,ds
\]
underlies the real interpolation results [2604.23864].

For the torus Hardy spaces in dimension \(d=1\), the recovered Pisier–Xu result states that
\[
(H_1(N),H_\infty(N))
\]
is quasi-complemented in
\[
(L_1(N),L_\infty(N))
\]
with a universal constant, and for every \(1<p<\infty\),
\[
(H_1(N),H_\infty(N))_{\theta,p,K}=H_p(N)
\]
with equivalent norms, where \(\theta=1-\frac1p\), and the constants depend only on \(p\). The same framework yields complex interpolation:
\[
[H_{p_0}(P),H_{p_1}(P)]_\theta = H_{p_\theta}(P), \qquad \frac1{p_\theta}=\frac{1-\theta}{p_0}+\frac{\theta}{p_1}
\]
[2604.23864].

In the continuous-filtration martingale theory, one has the Hardy/Lebesgue equivalence
\[
L_p(\mathcal M)=\mathcal H_p \qquad (1<p<\infty),
\]
where
\[
\mathcal H_p=
\begin{cases}
\mathcal H_p^c+\mathcal H_p^r, & 1<p<2,\\
\mathcal H_p^c\cap \mathcal H_p^r, & 2\le p<\infty.
\end{cases}
\]
More explicitly,
\[
\|x\|_p \simeq \|x\|_{\mathcal H_p^c+\mathcal H_p^r}, \qquad 1<p<2,
\]
and
\[
\|x\|_p \simeq \max\{\|x\|_{\mathcal H_p^c},\|x\|_{\mathcal H_p^r}\}, \qquad 2\le p<\infty.
\]
The paper also proves
\[
(\mathcal H_p^c)^*=\mathcal H_{p'}^c \qquad (1<p<\infty),
\]
\[
(\mathcal H_1^c)^*=BMO^c,
\qquad
(\mathcal H_1)^*=BMO:=BMO^c\cap BMO^r,
\]
and interpolation identities
\[
\mathcal H_p^c=[BMO^c,\mathcal H_1^c]_{1/p},
\qquad
L_p(\mathcal M)=[BMO,\mathcal H_1]_{1/p}
\]
[1301.2071].

In the operator-valued Hardy theory on spaces of homogeneous type, the main structural theorem states for \(\dagger\in\{c,r,cr\}\):
\[
(H_1^\dagger(\mathbb X,\mathcal M))^* \simeq \mathrm{BMO}^\dagger(\mathbb X,\mathcal M),
\]
\[
(H_p^\dagger(\mathbb X,\mathcal M))^*\simeq H_{p'}^\dagger(\mathbb X,\mathcal M), \qquad 1<p<\infty,
\]
\[
H_p^\dagger(\mathbb X,\mathcal M)=L_pMO^\dagger(\mathbb X,\mathcal M), \qquad 2<p<\infty,
\]
and
\[
H_p^{cr}(\mathbb X,\mathcal M)=L_p(\mathcal N), \qquad 1<p<\infty.
\]
Interpolation takes the form
\[
[\mathrm{BMO}^\dagger(\mathbb X,\mathcal M),\, H_q^\dagger(\mathbb X,\mathcal M)]_{q/p} = H_p^\dagger(\mathbb X,\mathcal M),
\]
as well as the real interpolation identity
\[
[\mathrm{BMO}^c(\mathbb X,\mathcal M),H_q^c(\mathbb X,\mathcal M)]_{q/p,p} = H_p^c(\mathbb X,\mathcal M)
\]
[2311.15775].

## 5. Calderón–Zygmund structure and geometric extensions

A decisive recent development is the transplantation of Bourgain’s method for \(K\)-closedness to the semicommutative setting. In that framework, a bounded operator \(T\) on \(L_2(M)\) is of Calderón–Zygmund type if for every \(s>0\) and \(y\in (L_1\cap L_2)(M)\), there exist
\[
y=a+b,\qquad a,b\in L_2(M),
\]
and a projection \(p\in M\) such that
\[
\|a\|_2^2 \le C^2 s \|y\|_1,
\]
\[
\tau(1-p)\le C^2 s^{-1}\|y\|_1,
\]
\[
\|pT(b)p\|_1\le C\|y\|_1.
\]
This is the semicommutative substitute for the classical good/bad decomposition plus localization by a “good” projection. Such operators are weak type \((1,1)\), and if a projection \(P\) satisfies the technical assumptions and is of Calderón–Zygmund type, then
\[
(H_1(P),H_2(P))
\]
is quasi-complemented in
\[
(L_1(M),L_2(M)).
\]
This is the semicommutative analogue of Bourgain’s key \(K\)-closedness step [2604.23864].

The technical engine is a semicommutative Calderón–Zygmund decomposition on
\[
N=L_\infty(X)\bar\otimes M,
\]
for \(X\) Ahlfors \(d\)-regular and \(M\) semifinite. Starting from Cuculescu projections
\[
q_n := 1_{[0,s]}(q_{n-1}E_n(f)q_{n-1}),\qquad q_0=1,
\]
and setting
\[
e_n=q_{n-1}-q_n,\qquad q=\bigwedge_{n\ge1} q_n,\qquad e=\sum_{n\ge1} e_n=1-q,
\]
one decomposes
\[
f=a+b,\qquad b=b_d+b_o,
\]
with
\[
a:=qfq+\sum_{n\ge1}E_{n-1}(e_n f e_n),
\]
\[
b_d:=\sum_{n\ge1}\big(e_n f e_n - E_{n-1}(e_n f e_n)\big),
\]
\[
b_o:=f-(a+b_d)=f-qfq-\sum_{n\ge1} e_n f e_n.
\]
The presence of the diagonal part \(b_d\) and off-diagonal part \(b_o\) reflects genuinely noncommutative interactions. The associated projection
\[
p:=\Big(\bigvee_{n\ge1}\bigvee_{Q\in\mathcal Q_n} 1_{B_{3d_Q}(y_Q)}\otimes e_Q\Big)^\perp
\]
encodes the bad region. The key estimates are
\[
\|a\|_2^2 \lesssim s\|f\|_1,
\qquad
\sigma(p^\perp)\lesssim s^{-1}\|f\|_1,
\]
and for singular-kernel operators \(T\),
\[
\|p(T\otimes I)(b_d)p\|_1\lesssim \|f\|_1,
\qquad
\|p(T\otimes I)(b_o)p\|_1\lesssim \|f\|_1.
\]
Combining these yields
\[
\|pT(b)p\|_1\lesssim \|f\|_1,
\]
which is exactly what Bourgain’s method requires [2604.23864].

The same paper shows that if \(T=S\otimes I\) is a bounded operator on \(L_2(N)\) given by a scalar singular kernel \(k\) on an Ahlfors-regular space \(X\), satisfying the size estimate
\[
|k(x,y)|\lesssim V(x,y)^{-1}
\]
and an \(L_2\)-Hörmander regularity condition, then \(T\) is of Calderón–Zygmund type with constants independent of \(M\). Consequently, a projection \(P\) on \(L_2(N)\) that satisfies the technical assumptions and is given by a singular kernel on \(X\) is of Jones-type. Hence the associated semicommutative Hardy spaces are \(K\)-closed and form a real interpolation scale [2604.23864].

A parallel geometric extension is provided by the operator-valued Hardy–BMO theory on spaces of homogeneous type. There the standing assumption is a Calderón reproducing formula
\[
f=\int_0^\infty D_t^2(f)\,\frac{dt}{t},
\]
with kernels \(D_t(x,y)\) satisfying size, Hölder regularity in both variables, and cancellation:
\[
\int_{\mathbb X} D_t(x,y)\,d\mu(x)=\int_{\mathbb X} D_t(x,y)\,d\mu(y)=0.
\]
A key point is that the theory works under doubling only and does not assume reverse doubling. Under these assumptions it establishes \(H_1\)–BMO duality, atomic decomposition of \(H_1\), interpolation between Hardy spaces and BMO spaces, and \(L_p\)-boundedness of operator-valued Calderón–Zygmund operators [2311.15775].

## 6. Martingale, atomic, and generalized-norm perspectives

The continuous-filtration theory supplies the semicommutative martingale analogue of classical Hardy/BMO structure without pathwise stopping-time methods. The continuous bracket is defined through an ultrafilter:
\[
[x,x]_{\mathcal U} = \mathcal E_{\mathcal U}\big(([x,x]_\sigma)^\bullet\big),
\]
and
\[
\|x\|_{\mathcal H_p^c} := \|[x,x]_{\mathcal U}\|_{p/2}^{1/2}.
\]
A second candidate norm is
\[
\|x\|_{\widehat{\mathcal H}_p^c} := \lim_{\sigma,\mathcal U}\|[x,x]_\sigma\|_{p/2}^{1/2},
\]
and the paper proves
\[
\mathcal H_p^c=\widehat{\mathcal H}_p^c
\]
with equivalent norms. The same holds for conditioned Hardy spaces:
\[
\mathrm h_p^c = h_p^c \quad\text{with equivalent norms.}
\]
A structural fact important in semicommutative applications is that the norms are intrinsic up to equivalence, even though the bracket elements themselves are introduced through ultrafilters [1301.2071].

The same martingale theory establishes continuous Davis decompositions. For \(1<p<2\),
\[
\mathcal H_p^c = h_p^d + h_p^c
\]
with equivalent norms, and for \(2\le p<\infty\),
\[
\mathcal H_p^c = J_p^d \cap h_p^c.
\]
It also proves the conditioned/global Burkholder–Rosenthal theorem
\[
L_p(\mathcal M)=h_p \qquad (1<p<\infty),
\]
where
\[
h_p=
\begin{cases}
h_p^d+h_p^c+h_p^r, & 1<p<2,\\
J_p^d\cap h_p^c\cap h_p^r, & 2\le p<\infty.
\end{cases}
\]
When specialized to semicommutative algebras \(L_\infty(\Omega)\bar\otimes\mathcal M\), these become continuous operator-valued semicommutative Hardy, BMO, conditioned, and diagonal decompositions [1301.2071].

The operator-valued Hardy theory on spaces of homogeneous type adds an atomic endpoint description. An \(\mathcal M\)-atom \(a\in L_1(\mathcal M;L_2(\mathbb X))\) is supported in a ball \(B\), satisfies
\[
\|a\|_{L_1(\mathcal M;L_2(\mathbb X))}\le \mu(B)^{-1/2},
\]
and has cancellation
\[
\int_{\mathbb X} a(x)\,d\mu(x)=0.
\]
The atomic Hardy space \(H_1^{c,\mathrm{at}}\) consists of all
\[
f=\sum_{j\in\mathbb N}\lambda_j a_j
\]
with \(\sum_j|\lambda_j|<\infty\), and the theorem
\[
H_1^\dagger(\mathbb X,\mathcal M)=H_1^{\dagger,\mathrm{at}}(\mathbb X,\mathcal M)
\]
holds for \(\dagger\in\{c,r,cr\}\). This gives a semicommutative \(H_1\) atomic theory in rough geometric settings [2311.15775].

Finally, the symmetric-norm framework indicates one route beyond \(L_p\)-based Hardy spaces. For \(\alpha\in\mathcal R_c\), the spaces \(L^\alpha(\mathbb T)\) and \(H^\alpha(\mathbb T)\) generalize \(L^p(\mathbb T)\) and \(H^p(\mathbb T)\). In the von Neumann algebra setting, one defines \(L^\alpha(\mathcal M,\tau)\) by completing \(\mathcal M\) in the induced symmetric gauge norm. The corrected reflexivity criterion is
\[
L^\alpha(\mathcal M,\tau)\text{ is reflexive}
\iff
\alpha \text{ and } \alpha' \text{ are both strongly continuous.}
\]
Moreover, if \(\alpha\) is continuous and \(\alpha'\) is strongly continuous, then
\[
L^\alpha(\mathcal M,\tau)^\# = L^{\alpha'}(\mathcal M,\tau).
\]
The paper does not build semicommutative Hardy spaces in this generalized-norm setting, but it explicitly presents duality and module-map representation theorems as foundational tools for such an extension [1407.7920].

## 7. Scope, distinctions, and recurring misconceptions

A central distinction is that semicommutative Hardy spaces are not exhausted by one definition. The projection-based spaces \(H_E(P)\), the martingale spaces \(\mathcal H_p^c,\mathcal H_p^r,h_p^c,h_p^r,h_p^d\), and the Lusin-area-function spaces \(H_p^c(\mathbb X,\mathcal M),H_p^r(\mathbb X,\mathcal M),H_p^{cr}(\mathbb X,\mathcal M)\) are all semicommutative, but they are built from different primitives: spectral projections and interpolation, filtrations and square functions, or Calderón reproducing families and area integrals [2604.23864], [1301.2071], [2311.15775].

Another recurring clarification concerns the role of row and column structures. In the projection-based Bourgain framework, the Hardy spaces are not introduced via row/column norms, conditioned Hardy spaces, square function characterizations, or maximal function characterizations. By contrast, in the martingale and homogeneous-type theories, row and column spaces are fundamental, and the row/column distinction disappears only when the value algebra is commutative [2604.23864], [1301.2071].

The torus terminology also requires care. The “semicommutative torus” in the interpolation and Bourgain theory means \(L_\infty(\mathbb T^d)\bar\otimes M\), namely the classical torus with values in a von Neumann algebra, not the rotation/noncommutative torus in Rieffel’s sense [2604.23864].

Taken together, the available theories show that semicommutative Hardy spaces form a broad operator-valued Hardy paradigm. In one direction, singular-kernel projections on \(L_\infty(X)\bar\otimes M\) produce \(K\)-closed Hardy scales and recover analytic torus Hardy spaces via the Riesz projection. In another, continuous filtrations produce row, column, conditioned, and diagonal Hardy/BMO theories for operator-valued martingales. In a third, Calderón reproducing formulas on spaces of homogeneous type yield a full Hardy–BMO theory with atomic decomposition, interpolation, and Calderón–Zygmund applications. A plausible implication is that semicommutativity functions less as a single definition than as a unifying structural principle: classical geometry or probability on the outside, noncommutative \(L_p\)- and duality structure on the inside.

Source: https://www.emergentmind.com/topics/semicommutative-hardy-spaces