---
title: Integral Projective Tensor Product
url: https://www.emergentmind.com/topics/integral-projective-tensor-product
type: topic
---

# Integral Projective Tensor Product

The integral projective tensor product is not presented in the surveyed literature as a single universally fixed tensor norm. The stable background object is the completed projective tensor product \(X\widehat{\otimes}_{\pi}Y\) of Banach spaces, defined by
\[
\|u\|_\pi=\inf\Big\{\sum_{i=1}^n \|x_i\|\,\|y_i\|:u=\sum_{i=1}^n x_i\otimes y_i\Big\},
\]
while “integral” enters through several closely related mechanisms: integral bilinear forms and integral operators, Bochner-integral representations of projective norm-attaining tensors, and duality with injective-type constructions [2602.23796] [1305.0791].

## 1. Terminology and scope

In Banach-space theory, the phrase belongs to the orbit of projective tensor products, integral operators, and measure-theoretic representations. It does not refer to the algebro-geometric usage of “tensor product surface” in which a surface is parametrized from \(\mathbb P^1\times \mathbb P^1\) by bihomogeneous forms; that terminology is explicitly unrelated to the integral/projective tensor product of Banach space theory [1908.02086].

Within the \(C^*\)-algebraic and operator-space literature surveyed here, an important terminological point is that several papers study the Banach-space projective tensor product \(A\otimes_\gamma B\) or the operator-space projective tensor product \(A\widehat{\otimes}B\), but do not introduce a separate tensor norm called the integral tensor product. Instead, “integral” appears through integral bilinear forms and integral operators, which are used to describe canonical dual embeddings and comparison maps [1305.0791] [1809.01131].

A second terminological distinction concerns recent work on norm attainment. There, “integral projective norm-attaining tensor” denotes a tensor in \(X\widehat\otimes_\pi Y\) represented by a Bochner integral over \(B_X\times B_Y\) with total mass equal to the projective norm; this is a measure-theoretic enlargement of the classical class of norm-attaining tensors, not a replacement for the underlying projective tensor norm itself [2602.23796].

## 2. Classical projective tensor products and their duality

For Banach spaces \(X\) and \(Y\), the projective tensor product \(X\widehat{\otimes}_\pi Y\) is the completion of the algebraic tensor product under \(\|\cdot\|_\pi\). Its dual admits the standard identification
\[
(X\widehat\otimes_\pi Y)^*=\mathcal B(X\times Y)=\mathcal L(X,Y^*),
\]
and the pairing is given on elementary tensors by \(B(x\otimes y)=B(x,y)\) [2602.23796]. This duality is the basic reason projective tensor products linearize bounded bilinear forms.

The same pattern persists in higher tensor powers. For \(n\ge 1\), the \(n\)-fold projective tensor product \(\widehat\otimes_\pi^n X\) is defined recursively, and its dual is the Banach space of bounded \(n\)-linear forms:
\[
(\widehat\otimes_\pi^n X)^*=\mathcal L^n(X).
\]
In the case \(X=c_0\), the projective tensor tower is especially rigid: for \(m<n\), \(\widehat{\otimes}_\pi^n c_0\) is not isomorphic to any subspace of any quotient of \(\widehat{\otimes}_\pi^m c_0\), and
\[
(\widehat{\otimes}_\pi^n c_0)^* \cong \widehat{\otimes}_\varepsilon^n \ell_1
\]
under the hypotheses established in the paper [2012.13437].

An analogous projective formalism exists beyond Banach spaces. For Banach \(L^0\)-modules \(M\) and \(N\), the projective pointwise norm is
\[
|a|_{\pi} := \bigwedge \left\{ \sum_{i=1}^{n} |v_i|\,|w_i| \;\middle|\; a=\sum_{i=1}^{n} v_i\otimes w_i \right\},
\]
the completed projective tensor product is \(M\widehat\otimes_{\pi}N\), and its universal property is
\[
B(M,N;Q)\cong \operatorname{HOM}(M\widehat\otimes_{\pi}N;Q).
\]
Its dual is correspondingly identified as
\[
(M\widehat\otimes_{\pi}N)^*\cong B(M,N)
\]
[2308.03634]. This \(L^0\)-valued construction preserves the classical projective theme while changing both the scalar ring and the norm range.

## 3. Integral operators, bilinear forms, and projective tensor products

A central bridge between “integral” and “projective” is duality. For \(C^*\)-algebras \(A\) and \(B\), the canonical map
\[
\theta:A\otimes_\gamma B\to (A^*\otimes_\lambda B^*)^*
\]
is bounded above and below by
\[
\frac12 \|u\|_\gamma \le \|\theta(u)\| \le \|u\|_\gamma,
\]
and \((A^*\otimes_\lambda B^*)^*\) is identified in the paper with a space of integral operators [1305.0791]. The same paper defines a natural embedding
\[
\mu:A^{**}\otimes_\gamma B^{**}\to (A\otimes_\gamma B)^{**}
\]
with the same bicontinuity estimate,
\[
\frac12 \|u\|_\gamma \le \|\mu(u)\| \le \|u\|_\gamma.
\]
Under additional hypotheses it records the coincidence
\[
N(B^*,A^{**}) = PJ(B^*,A^{**}) = J(B^*,A^{**}),
\]
linking nuclear, Pietsch integral, and integral operators.

This operator-theoretic perspective complements the usual projective-tensor description by bilinear forms. For Banach algebras \(A\) and \(B\), Arens regularity of \(A\otimes^\gamma B\) is equivalent to the statement that every bounded bilinear form \(m:A\times B\to \mathbb C\) is biregular, while Arens regularity of the operator-space projective tensor product \(A\widehat{\otimes}B\) is equivalent to biregularity of every jointly completely bounded bilinear form [1401.1997]. In other words, projective tensor products are classified by the classes of bilinear forms they linearize, and integral-operator language enters through the dual embeddings that control these spaces.

The same dual mechanism appears in recent work on norm attainment. If a functional on \(X\widehat\otimes_\pi Y\) attains its norm at an integral norm-attaining tensor, then the corresponding bilinear form is norm-attaining on \(X\times Y\) [2602.23796]. This relation is one reason integral representations are useful: they convert extremal questions in the tensor product into norm-attainment questions for bilinear forms and operators.

## 4. Integral projective norm-attaining tensors

The most explicit recent integral formulation is the class \(INA_\pi(X\widehat\otimes_\pi Y)\) of integral projective norm-attaining tensors. A tensor \(u\in X\widehat\otimes_\pi Y\) belongs to this class if there exists a finite positive Borel measure \(\mu\) on \(B_X\times B_Y\) such that the canonical map
\[
\varphi:B_X\times B_Y\to X\widehat\otimes_\pi Y,\qquad \varphi(x,y)=x\otimes y,
\]
is \(\mu\)-Bochner integrable and
\[
u=\int_{B_X\times B_Y} x\otimes y\,d\mu(x,y), \qquad \|\mu\|=\|u\|_\pi.
\]
Classical norm-attaining tensors yield such representations by taking a discrete measure, so
\[
NA_\pi(X\widehat\otimes_\pi Y)\subseteq INA_\pi(X\widehat\otimes_\pi Y)\subseteq X\widehat\otimes_\pi Y
\]
[2602.23796].

This integral class admits a support-functional characterization parallel to the discrete theory. If
\[
u=\int x\otimes y\,d\mu(x,y),
\]
then \(u\in INA_\pi\) if and only if there exists \(B\in S_{\mathcal B(X,Y)}\) such that
\[
B(x,y)=1 \quad \text{for \(\mu\)-a.e. }(x,y)\in B_X\times B_Y.
\]
Positive measures suffice, and any witnessing measure is concentrated on \(S_X\times S_Y\). The class is also stable under absolute continuity: if \(\mu'\ll \mu\), then
\[
u'=\int x\otimes y\,d\mu'
\]
is again in \(INA_\pi\) [2602.23796].

A major structural theorem states that every integral norm-attaining tensor can be approximated in projective norm by finitely norm-attaining tensors:
\[
INA_\pi(X\widehat\otimes_\pi Y)\subseteq \overline{FNA_\pi(X\widehat\otimes_\pi Y)}^{\|\cdot\|_\pi}.
\]
Consequently, the Bishop–Phelps density problem is unchanged by passing from discrete to integral representations:
\[
\overline{NA_\pi(X\widehat\otimes_\pi Y)}^{\|\cdot\|_\pi}=X\widehat\otimes_\pi Y
\iff
\overline{INA_\pi(X\widehat\otimes_\pi Y)}^{\|\cdot\|_\pi}=X\widehat\otimes_\pi Y.
\]
The same paper proves that if an extreme point of \(B_{X\widehat\otimes_\pi Y}\) belongs to \(INA_\pi\) and is witnessed by a Radon measure, then it must be an elementary tensor \(x\otimes y\) [2602.23796].

Weak and weak\(^*\) variants are developed by replacing the norm-topology Borel structure on \(B_X\times B_Y\) with weak or weak\(^*\) product topologies. Most formal properties persist, but genuine Bochner integrability becomes delicate. Under separability together with either BAP or a Dunford–Pettis-type hypothesis, the tensor-valued map \(\varphi(x,y)=x\otimes y\) is Bochner integrable in the weak setting. The same work also extends known non-norm-attainment phenomena to the integral setting, showing for instance that \(L_1\widehat\otimes_\pi L_p\) and the real \(c_0\widehat\otimes_\pi L_p\) contain non-norm-attaining tensors for \(1<p<\infty\) [2602.23796].

## 5. Geometric and asymptotic structure

Projective tensor products have a rigid convex geometry. For bounded closed convex sets \(C\subseteq X\) and \(D\subseteq Y\), the set
\[
\overline{\operatorname{co}(C\otimes D)}\subseteq X\widehat\otimes_\pi Y
\]
is the natural tensorial convex hull, and when \(K(X,Y^*)\) separates points of \(X\widehat\otimes_\pi Y\), every nonzero preserved extreme point of this set is of the form \(x\otimes y\), with \(x\) and \(y\) preserved extreme in \(C\) and \(D\) respectively [2211.13559]. Applied to unit balls, this yields a factorization theorem for preserved extreme points of \(B_{X\widehat\otimes_\pi Y}\).

The same paper gives a partial converse for weak-strongly exposed points. Under the additional assumption that \(x\otimes y\) has a compact neighborhood system for the weak topology in \(\overline{\operatorname{co}(C\otimes D)}\), weak-strong exposure tensorizes:
\[
x\otimes y \text{ is weak-strongly exposed }
\iff
x \text{ and } y \text{ are weak-strongly exposed.}
\]
The extra hypothesis is essential. There exists a Banach space \(X\) isomorphic to \(\ell_2\) with a weak-strongly exposed point \(x_0\in B_X\) such that \(x_0\otimes x_0\) is not weak-strongly exposed in \(B_{X\widehat\otimes_\pi X}\) [2211.13559].

Asymptotically, iterated projective tensor products form a strict hierarchy. For \(m<n\),
\[
\widehat{\otimes}_\pi^n c_0
\]
is not isomorphic to any subspace of any quotient of
\[
\widehat{\otimes}_\pi^m c_0.
\]
This is stronger than mere non-isomorphism and is proved through asymptotic invariants such as the classes \(\mathcal A_{N,p}\) and Szlenk-type bounds [2012.13437]. A plausible implication is that projective tensoring increases asymptotic complexity in a systematic way rather than merely enlarging dimension or codimension.

These geometric results interact naturally with the integral perspective. The recent extremal theorem for Radon-integral norm-attaining tensors gives a measure-theoretic route to the same rank-one conclusion that older convex-geometric arguments derive from compact-operator separation [2602.23796] [2211.13559]. This suggests a convergence of two lines of thought: one through weak topology and operator ideals, the other through Bochner representations.

## 6. \(C^*\)-algebraic and operator-space variants

For \(C^*\)-algebras, the Banach-space projective tensor product is denoted \(A\otimes_\gamma B\). It is a Banach \(*\)-algebra, and its ideal theory is unusually explicit. If \(A_1\subseteq A\) and \(B_1\subseteq B\) are \(C^*\)-subalgebras, the identity map on \(A_1\otimes B_1\) extends to an isometric \(*\)-algebra map from \(A_1\otimes_\gamma B_1\) onto the closed \(*\)-subalgebra \(A_1\otimes_\gamma B_1\subseteq A\otimes_\gamma B\) [1809.01131]. When one factor is topologically simple, every closed ideal of \(A\otimes_\gamma B\) is of the form \(A\otimes_\gamma J\). More generally, if one factor has finitely many closed ideals, every closed ideal is a finite sum of product ideals. The same paper identifies maximal ideals by
\[
J = A\otimes_\gamma N + M\otimes_\gamma B
\]
for maximal ideals \(M\subseteq A\), \(N\subseteq B\), and proves the center formula
\[
Z(A\otimes_\gamma B)=Z(A)\otimes_\gamma Z(B)
\]
[1809.01131].

The operator-space projective tensor product uses a different norm. For operator spaces \(X\) and \(Y\),
\[
\|u\|_\wedge = \inf \{ \|\alpha\|\, \|x\|\, \|y\|\, \|\beta\| : u=\alpha(x\otimes y)\beta\},
\]
with \(x\in M_p(X)\), \(y\in M_q(Y)\), \(\alpha\in M_{1,pq}\), and \(\beta\in M_{pq,1}\) [1401.1997]. In the \(C^*\)-algebraic setting, the canonical embedding
\[
\mu_{\wedge}:A^{**}\widehat\otimes B^{**}\to (A\widehat\otimes B)^{**}
\]
is \(*\)-preserving, positive, completely bounded, and satisfies
\[
\frac12\|u\|_\wedge \le \|\mu_{\wedge}(u)\| \le \|u\|_\wedge
\]
[1305.0791].

Arens regularity sharply links the Banach and operator-space projective theories. For \(C^*\)-algebras \(A\) and \(B\), the Arens regularity of
\[
A\otimes^\gamma B,\quad A\widehat\otimes B,\quad A\otimes^h B,\quad A\otimes^s B
\]
is equivalent [1401.1997]. Thus, at the level of bidual multiplication, classical projective, operator-space projective, Haagerup, and Schur tensor norms exhibit the same regularity behavior in the \(C^*\)-algebraic category.

## 7. Generalized projective tensor products beyond Banach spaces

The Banach \(L^0\)-module setting extends projective tensor products to norms with values in the ring \(L^0(X)\). There the projective tensor product \(M\widehat\otimes_\pi N\) is defined by completion of the algebraic tensor product over \(L^0(X)\), and its dual satisfies
\[
(M\widehat\otimes_{\pi}N)^*\cong B(M,N).
\]
It also obeys a series representation formula
\[
|a|_{\pi} = \bigwedge \left\{ \sum_{n\in\mathbb N}|u_n|\,|w_n| \;\middle|\; a=\sum_{n}u_n\otimes w_n \right\},
\]
and a pullback compatibility theorem
\[
\varphi^{*}(M\widehat\otimes_{\pi}N) \cong (\varphi^{*}M)\widehat\otimes_{\pi}(\varphi^{*}N)
\]
for measurable maps \(\varphi\) under the stated hypotheses [2308.03634]. This is a projective tensor theory over a function ring rather than over \(\mathbb R\) or \(\mathbb C\).

An even more noncommutative extension is the projective tensor product of protoquantum spaces. For PQ-spaces \(E\) and \(F\), the proto-operator-projective norm on \(\mathcal F(E\otimes F)\) is
\[
\|U\|_{pop} := \inf \left\{ \sum_{k=1}^n \|a_k\|\,\|u_k\|\,\|v_k\|\,\|b_k\| \right\},
\]
where the infimum runs over decompositions
\[
U = \sum_{k=1}^n a_k\cdot (u_k\diamond v_k)\cdot b_k.
\]
The resulting completed tensor product \(E\widehat{\otimes}_{pop}F\) linearizes completely bounded bilinear maps and satisfies an adjoint associativity theorem
\[
CB(E\otimes_{pop}F,G)\cong CB(F,CB(E,G))
\]
[1706.00621]. In this framework, the standard operator-space projective formula is too rigid outside the operator-space category: the paper exhibits tensors \(V_n\) for which
\[
\|V_n\|_{pop}=n,\qquad \|V_n\|_{op}=n^2.
\]
This shows that a projective tensor product suitable for general matricially normed spaces cannot simply be the usual operator-space projective tensor product [1706.00621].

Taken together, these developments show that “integral projective tensor product” names not a single immutable object but a family of tightly connected ideas. The fixed core is the projective tensor product and its universal property; the integral aspect appears through integral operators, bilinear forms, and, most explicitly, through Bochner-integral representations of norm-attaining tensors. The interaction between these themes governs extremal geometry, non-attainment phenomena, Arens regularity, ideal structure, and the extension of projective tensor methods to operator spaces, \(C^*\)-algebras, \(L^0\)-modules, and protoquantum spaces.

Source: https://www.emergentmind.com/topics/integral-projective-tensor-product