Commuting Tensor Product

• Commuting tensor product is a collection of tensor constructions defined by reordering, symmetry, and commutation conditions across diverse mathematical contexts.
• It appears in contexts such as matrix theory, Hilbert modules, quantum games, and category theory, providing a framework for analyzing operator models and tensor factorizations.
• Applications span from designing swap operators in finite matrices and ensuring double commutativity in analytic modules to forming Banach algebra structures and constructing commuting projections in numerical analysis.

In the literature considered here, commuting tensor product does not denote a single construction. The expression is used for several distinct, technically precise phenomena: a permutation or swap operator that reorders tensor factors in Kronecker products; a tensor-factorization theorem rigidified by double commutativity in Hilbert modules; a comparison between the commuting operator and tensor product models in nonlocal games; symmetric monoidal tensor products representing commuting multimorphisms in category theory; Banach-algebra tensor products adapted to commuting pairs of matrices; and tensorized commuting projections in discrete de Rham complexes. What these usages share is not a common formal definition, but the structural role of a tensor product constrained by an additional commutation, symmetry, or interchange law (Rakotonirina, 2011, Chattopadhyay et al., 2013, Gambino et al., 18 Nov 2025).

1. Matrix and Kronecker-product meanings

In finite-dimensional matrix theory, the basic object is the tensor commutation matrix or more generally a tensor permutation matrix. For vector spaces $E_1,\dots,E_k$ and a permutation $\sigma$, the associated tensor permutation operator is the linear map

$$U_\sigma(x_1\otimes \cdots \otimes x_k)=x_{\sigma(1)}\otimes \cdots \otimes x_{\sigma(k)}.$$

After choosing tensor-product bases, its matrix is a permutation matrix independent of the chosen bases. In the two-factor case this becomes the tensor commutation matrix $U_{n\otimes p}$, characterized by

$U_{n\otimes p}(a\otimes b)=b\otimes a,$

and, for square-compatible matrices,

$$U_{n\otimes p}(A\otimes B)=(B\otimes A)U_{n\otimes p}.$$

For rectangular factors the general identity is

$$U_\sigma\,(A_1\otimes \cdots \otimes A_k)\,V_\sigma^T = A_{\sigma(1)}\otimes \cdots \otimes A_{\sigma(k)}.$$

Thus the relevant sense of “commuting” is not literal equality $A\otimes B=B\otimes A$, but reordering by canonical permutation matrices (Rakotonirina, 2011).

This viewpoint is sharpened by explicit decompositions of the swap operator in orthogonal operator bases. For two qubits,

$$S_{2\otimes 2} = \frac{1}{2}I_2\otimes I_2 +\frac{1}{2}\sum_{i=1}^{3}\sigma_i\otimes\sigma_i,$$

and more generally, if $n=2^p$, the swap on $\sigma$0 admits the Pauli-string expansion

$\sigma$1

The paper also recalls the $\sigma$2-dimensional Gell-Mann analogue

$\sigma$3

An important negative result accompanies these formulas: the particular $\sigma$4 generalized Pauli candidates tested there do not reproduce $\sigma$5 in the same way; the successful construction is restricted to tensor products of ordinary Pauli matrices and hence naturally to dimensions $\sigma$6 (Rakotonirina et al., 2013).

2. Double commutativity and tensor factorization in Hilbert modules

In the Hilbert-module literature, “commuting tensor product” refers to a structural decomposition of quotient modules over product domains. Let

$\sigma$7

be a product reproducing kernel Hilbert module over $\sigma$8, with coordinate multipliers $\sigma$9, and let $$U_\sigma(x_1\otimes \cdots \otimes x_k)=x_{\sigma(1)}\otimes \cdots \otimes x_{\sigma(k)}.$$0 be a quotient module with compressed multipliers

$$U_\sigma(x_1\otimes \cdots \otimes x_k)=x_{\sigma(1)}\otimes \cdots \otimes x_{\sigma(k)}.$$1

The decisive condition is double commutativity: $$U_\sigma(x_1\otimes \cdots \otimes x_k)=x_{\sigma(1)}\otimes \cdots \otimes x_{\sigma(k)}.$$2 For analytic Hilbert modules, this condition is equivalent to the existence of a tensor decomposition

$$U_\sigma(x_1\otimes \cdots \otimes x_k)=x_{\sigma(1)}\otimes \cdots \otimes x_{\sigma(k)}.$$3

with each $$U_\sigma(x_1\otimes \cdots \otimes x_k)=x_{\sigma(1)}\otimes \cdots \otimes x_{\sigma(k)}.$$4 a one-variable quotient module. In this setting, being doubly commuting is therefore neither merely necessary nor merely sufficient: it is exactly the criterion for tensor-product splitting (Chattopadhyay et al., 2013).

Passing to orthogonal complements yields the classification of co-doubly commuting submodules. A submodule $$U_\sigma(x_1\otimes \cdots \otimes x_k)=x_{\sigma(1)}\otimes \cdots \otimes x_{\sigma(k)}.$$5 is co-doubly commuting precisely when

$$U_\sigma(x_1\otimes \cdots \otimes x_k)=x_{\sigma(1)}\otimes \cdots \otimes x_{\sigma(k)}.$$6

equivalently

$$U_\sigma(x_1\otimes \cdots \otimes x_k)=x_{\sigma(1)}\otimes \cdots \otimes x_{\sigma(k)}.$$7

The tensor decomposition also makes the mixed commutators explicit: $$U_\sigma(x_1\otimes \cdots \otimes x_k)=x_{\sigma(1)}\otimes \cdots \otimes x_{\sigma(k)}.$$8 From this, the paper derives rigidity results: for $$U_\sigma(x_1\otimes \cdots \otimes x_k)=x_{\sigma(1)}\otimes \cdots \otimes x_{\sigma(k)}.$$9, a co-doubly commuting submodule is essentially doubly commuting if and only if it has finite codimension, equivalently the quotient is essentially normal. The scope is not universal, however. The standard hypothesis is essential: if one factor is not standard, one can construct a doubly commuting quotient module that is an orthogonal sum of two tensor products rather than a single tensor product (Chattopadhyay et al., 2013).

3. Commuting-operator models versus tensor-product models

In quantum information, the phrase enters through the distinction between two models of multipartite nonlocal strategies. In the tensor product model, the shared state lies in

$U_{n\otimes p}$0

and each player measures only on its own tensor factor. In the commuting operator model, all observables act on a single Hilbert space, with the only locality condition being pairwise commutation across different players. The general inequality

$U_{n\otimes p}$1

holds, and the paper explicitly notes that equality can fail in general. The special case of perfect $U_{n\otimes p}$2XOR games is different: a 3XOR game has a perfect commuting operator strategy if and only if it has a perfect tensor product strategy using a 3-qubit GHZ state, so in that exact value-$U_{n\otimes p}$3 regime

$U_{n\otimes p}$4

The proof encodes the game into a right-angled Coxeter group, identifies perfect commuting strategies with the group-theoretic condition $U_{n\otimes p}$5, and then shows that for 3XOR this obstruction is already captured by a tractable abelian quotient, yielding a polynomial-time decision procedure (Watts et al., 2020).

The same paper is careful about scope. The equivalence is only for 3XOR games, only for perfect strategies, and does not establish equality of the full value functions for arbitrary games or approximate strategies. That restriction matters when set against negative approximation results for commutation. In normalized Hilbert–Schmidt norm,

$U_{n\otimes p}$6

is not Hilbert–Schmidt stable. Concretely, there exist contractions $U_{n\otimes p}$7 such that

$U_{n\otimes p}$8

but for every exactly commuting pair $U_{n\otimes p}$9 with

$U_{n\otimes p}(a\otimes b)=b\otimes a,$0

one still has

$U_{n\otimes p}(a\otimes b)=b\otimes a,$1

A plausible implication is that “commuting tensor product” phenomena that depend on exact commutation cannot, in general, be replaced by asymptotic Hilbert–Schmidt commutation without loss of structure (Ioana, 2021).

4. Categorical and operadic commuting tensor products

In category-theoretic settings, the phrase is closer to a genuine tensor product equipped with a universal commutation property. A basic symmetric example is the tensor product of correspondence functors. For correspondence functors $U_{n\otimes p}(a\otimes b)=b\otimes a,$2,

$U_{n\otimes p}(a\otimes b)=b\otimes a,$3

This makes the category $U_{n\otimes p}(a\otimes b)=b\otimes a,$4 symmetric monoidal: $U_{n\otimes p}(a\otimes b)=b\otimes a,$5 For a finite lattice $U_{n\otimes p}(a\otimes b)=b\otimes a,$6, the associated functor $U_{n\otimes p}(a\otimes b)=b\otimes a,$7 is a commutative algebra correspondence functor, with multiplication

$U_{n\otimes p}(a\otimes b)=b\otimes a,$8

and one has

$U_{n\otimes p}(a\otimes b)=b\otimes a,$9

In this context the commuting feature is ordinary symmetry of the monoidal product together with commutative algebra objects inside that symmetric monoidal category (Bouc et al., 2019).

A much more general treatment is given in the double-categorical framework of commuting tensor products of monads. There the ambient datum is a symmetric normal oplax monoidal closed double category $$U_{n\otimes p}(A\otimes B)=(B\otimes A)U_{n\otimes p}.$$0. One forms $$U_{n\otimes p}(A\otimes B)=(B\otimes A)U_{n\otimes p}.$$1, whose objects are horizontal monads, vertical arrows are monad morphisms, and horizontal arrows are bimodules. A commuting monad multimorphism is defined by an interchange hexagon built from the oplax monoidal structure; the commuting tensor product is then the representing object for such commuting multimorphisms. Under the stated hypotheses, the category $$U_{n\otimes p}(A\otimes B)=(B\otimes A)U_{n\otimes p}.$$2 becomes symmetric monoidal closed under this commuting tensor product, and $$U_{n\otimes p}(A\otimes B)=(B\otimes A)U_{n\otimes p}.$$3 becomes a symmetric oplax monoidal double category. This single construction recovers the tensor product of enriched categories, the Boardman–Vogt tensor product of operads and symmetric multicategories, and extensions of these tensor products to bimodules, profunctors, and multiprofunctors (Gambino et al., 18 Nov 2025).

5. Banach-algebra tensor products for commuting pairs

A different usage appears in multicentric functional calculus. Here the objective is not to define a commuting tensor product of operator algebras in the usual $$U_{n\otimes p}(A\otimes B)=(B\otimes A)U_{n\otimes p}.$$4-algebraic sense, but to construct a Banach algebra adapted to a commuting pair of matrices. Starting from one-variable multicentric algebras $$U_{n\otimes p}(A\otimes B)=(B\otimes A)U_{n\otimes p}.$$5 and $$U_{n\otimes p}(A\otimes B)=(B\otimes A)U_{n\otimes p}.$$6, the paper builds the two-variable space

$$U_{n\otimes p}(A\otimes B)=(B\otimes A)U_{n\otimes p}.$$7

with a polyproduct induced from the one-variable multiplications through

$$U_{n\otimes p}(A\otimes B)=(B\otimes A)U_{n\otimes p}.$$8

and proves the Banach-space identification

$$U_{n\otimes p}(A\otimes B)=(B\otimes A)U_{n\otimes p}.$$9

Scalar functions $$U_\sigma\,(A_1\otimes \cdots \otimes A_k)\,V_\sigma^T = A_{\sigma(1)}\otimes \cdots \otimes A_{\sigma(k)}.$$0 are represented as Gelfand transforms of matrix-valued functions $$U_\sigma\,(A_1\otimes \cdots \otimes A_k)\,V_\sigma^T = A_{\sigma(1)}\otimes \cdots \otimes A_{\sigma(k)}.$$1, with

$$U_\sigma\,(A_1\otimes \cdots \otimes A_k)\,V_\sigma^T = A_{\sigma(1)}\otimes \cdots \otimes A_{\sigma(k)}.$$2

and the construction is intended for commuting pairs $$U_\sigma\,(A_1\otimes \cdots \otimes A_k)\,V_\sigma^T = A_{\sigma(1)}\otimes \cdots \otimes A_{\sigma(k)}.$$3 after choosing $$U_\sigma\,(A_1\otimes \cdots \otimes A_k)\,V_\sigma^T = A_{\sigma(1)}\otimes \cdots \otimes A_{\sigma(k)}.$$4 so that $$U_\sigma\,(A_1\otimes \cdots \otimes A_k)\,V_\sigma^T = A_{\sigma(1)}\otimes \cdots \otimes A_{\sigma(k)}.$$5 and $$U_\sigma\,(A_1\otimes \cdots \otimes A_k)\,V_\sigma^T = A_{\sigma(1)}\otimes \cdots \otimes A_{\sigma(k)}.$$6 are diagonalizable. The paper explicitly notes that this is not the standard operator-algebraic commuting tensor product; it is a tensor-product-style Banach algebra tailored to commuting pairs of matrices (Andrei, 2021).

This analytic use is illuminated by contrast with tensor-algebra products that are not commuting. Shao’s general product of tensors extends ordinary matrix multiplication and satisfies the associative law

$$U_\sigma\,(A_1\otimes \cdots \otimes A_k)\,V_\sigma^T = A_{\sigma(1)}\otimes \cdots \otimes A_{\sigma(k)}.$$7

but is generally noncommutative. Even scalar placement is asymmetric: $$U_\sigma\,(A_1\otimes \cdots \otimes A_k)\,V_\sigma^T = A_{\sigma(1)}\otimes \cdots \otimes A_{\sigma(k)}.$$8 The paper therefore provides a useful counterpoint: tensor-product language alone does not imply any commuting behavior, and in that literature the central property is associativity rather than commutativity (Shao, 2012).

6. Tensor-product commuting projections in discrete de Rham theory

In numerical analysis and finite element exterior calculus, the phrase appears in the construction of commuting projection operators on tensor-product patch spaces. For 2D multipatch domains

$$U_\sigma\,(A_1\otimes \cdots \otimes A_k)\,V_\sigma^T = A_{\sigma(1)}\otimes \cdots \otimes A_{\sigma(k)}.$$9

with $A\otimes B=B\otimes A$0, each patch carries a local tensor-product de Rham sequence. On the reference patch,

$A\otimes B=B\otimes A$1

The scalar projector $A\otimes B=B\otimes A$2 is built from a tensor-product basis and dual basis, and the higher projectors are defined by tensorized antiderivative formulas such as

$A\otimes B=B\otimes A$3

and

$A\otimes B=B\otimes A$4

On a single patch these operators are projections, are $A\otimes B=B\otimes A$5-stable, and satisfy

$A\otimes B=B\otimes A$6

The tensor-product structure is used crucially through directional invariance preserved by $A\otimes B=B\otimes A$7 (Pinto et al., 2023).

The multipatch problem is subtler because adjacent patches may have non-matching interfaces. The paper first applies these tensor-product commuting projectors patchwise on broken spaces, then adds localized edge and vertex corrections so that the final global projectors

$A\otimes B=B\otimes A$8

become conforming while retaining commutation: $A\otimes B=B\otimes A$9 The result is local and stable in every $$S_{2\otimes 2} = \frac{1}{2}I_2\otimes I_2 +\frac{1}{2}\sum_{i=1}^{3}\sigma_i\otimes\sigma_i,$$0, $$S_{2\otimes 2} = \frac{1}{2}I_2\otimes I_2 +\frac{1}{2}\sum_{i=1}^{3}\sigma_i\otimes\sigma_i,$$1, under the assumptions that neighboring patches have nested resolutions and interior vertices are shared by exactly four patches. A plausible implication is that here “commuting tensor product” refers not to a tensor product of global spaces, but to a tensorized local mechanism whose cochain commutation survives global gluing (Pinto et al., 2023).

7. Scope, ambiguity, and recurrent distinctions

Across these sources, several recurrent distinctions govern the meaning of the term.

First, reordering is not equality. In matrix theory, tensor factors “commute” only through permutation matrices such as $$S_{2\otimes 2} = \frac{1}{2}I_2\otimes I_2 +\frac{1}{2}\sum_{i=1}^{3}\sigma_i\otimes\sigma_i,$$2 or $$S_{2\otimes 2} = \frac{1}{2}I_2\otimes I_2 +\frac{1}{2}\sum_{i=1}^{3}\sigma_i\otimes\sigma_i,$$3; the tensor product itself remains noncommutative as an ordered expression (Rakotonirina, 2011, Rakotonirina et al., 2013).

Second, commutation may be an operator-theoretic rigidity condition rather than a symmetry of a monoidal product. In analytic Hilbert modules, double commutativity of the compressed coordinate multipliers is equivalent to tensor-product factorization, but only under the paper’s standard or analytic hypotheses; outside that regime, doubly commuting quotients need not be single tensor products (Chattopadhyay et al., 2013).

Third, commuting and tensor-product models can coincide or separate depending on context. For perfect 3XOR games, perfect commuting operator strategies collapse to perfect tensor-product GHZ strategies, but the same paper explicitly refuses any general commuting$$S_{2\otimes 2} = \frac{1}{2}I_2\otimes I_2 +\frac{1}{2}\sum_{i=1}^{3}\sigma_i\otimes\sigma_i,$$4tensor-product conclusion beyond that exact setting (Watts et al., 2020). Conversely, approximate Hilbert–Schmidt commutation need not be perturbable to exact commuting relations, so asymptotic commutation is not an adequate replacement for exact commuting models in general (Ioana, 2021).

Fourth, some literatures use the term via a universal property. In the most abstract setting surveyed here, a commuting tensor product is the object representing commuting multimorphisms, and its extension to bimodules requires an oplax interchange law in a double category (Gambino et al., 18 Nov 2025). In less abstract but still categorical settings, the same broad phenomenon appears as a symmetric monoidal product with commutative algebra objects, as in correspondence functors (Bouc et al., 2019).

These distinctions suggest that commuting tensor product is best treated as a family of domain-specific notions organized by a common theme: a tensorial construction constrained by interchange, symmetry, exact commutation, or compatibility with differential structure, rather than by a single cross-disciplinary definition.