Monoidal, Rigid, and Braided Language

Ordinary composition, a monoidal product, duality, braiding, and symmetry are different structures. Composition combines arrows sequentially when their types match. A monoidal product combines pairs of objects and arrows through a bifunctor. Rigidity adds evaluation and coevaluation maps satisfying snake identities. Braiding adds coherent natural exchange maps, and symmetry adds the requirement that a double exchange is trivial. These axioms organize a claim; by themselves they do not prove physical tensor factorization, locality, statistics, positivity, unitarity, or the existence of a QFT realizing them.

Required background. Categories, Functors, Natural Transformations, and Universal Properties supplies the functor laws, naturality equations, and typed commuting diagrams used here without rederivation.

Monoidal structure · Rigidity · Braiding · QFT translation · Check your understanding

Tensor is not ordinary composition

For $f:X\to X'$ and $g:Y\to Y'$, the expression $f\otimes g$ has type

$$f\otimes g:X\otimes Y\longrightarrow X'\otimes Y'.$$

No codomain-to-domain match between $f$ and $g$ is required. This differs from $g\circ f$, which is defined only when $X'=Y$ and then has type $X\to Y'$. The object $X\otimes Y$ is also not automatically a categorical product: a monoidal product need not come with projections or satisfy a product universal property.

A monoidal category consists of

$$(\mathcal C,\otimes,\mathbf 1,\alpha,\lambda,\rho),$$

where

• $\otimes:\mathcal C\times\mathcal C\to\mathcal C$ is a bifunctor;
• $\mathbf 1$ is a unit object;
• the associator is a natural isomorphism $\alpha_{X,Y,Z}:(X\otimes Y)\otimes Z\to X\otimes(Y\otimes Z)$;
• the left and right unitors are natural isomorphisms $\lambda_X:\mathbf 1\otimes X\to X$ and $\rho_X:X\otimes\mathbf 1\to X$.

Bifunctoriality means

$$\operatorname{id}_X\otimes\operatorname{id}_Y = \operatorname{id}_{X\otimes Y},$$

and, for composable arrows $f:X\to X'$, $f':X'\to X''$, $g:Y\to Y'$, and $g':Y'\to Y''$,

$$(f'\circ f)\otimes(g'\circ g) = (f'\otimes g')\circ(f\otimes g).$$

The second equation is the interchange law. It relates sequential and parallel composition without identifying them.

Coherence controls rebracketing

The associator and unitors must obey the pentagon and triangle identities. The pentagon says that the two canonical routes from $((W\otimes X)\otimes Y)\otimes Z$ to $W\otimes(X\otimes(Y\otimes Z))$ agree:

$$\begin{aligned} &\alpha_{W,X,Y\otimes Z} \circ\alpha_{W\otimes X,Y,Z}\\ &\quad= (\operatorname{id}_W\otimes\alpha_{X,Y,Z}) \circ\alpha_{W,X\otimes Y,Z} \circ(\alpha_{W,X,Y}\otimes\operatorname{id}_Z). \end{aligned}$$

The triangle says that removing a unit before or after reassociation gives the same result:

$$(\operatorname{id}_X\otimes\lambda_Y) \circ\alpha_{X,\mathbf 1,Y} = \rho_X\otimes\operatorname{id}_Y.$$

A monoidal category is strict when the associator and unitors are identity maps in the chosen presentation. Strictness is not rigidity. Mac Lane’s strictness theorem gives a monoidal equivalence to a strict monoidal category; it does not make differently parenthesized objects literally equal inside the original category. The coherence theorem does license suppressing brackets in canonical composites built from $\alpha$, $\lambda$, $\rho$, and their inverses, because all such canonical routes with the same endpoints agree. For the monoidal data, coherence, strictification, duals, rigidity, braiding, and symmetry used throughout this page, see Etingof, Gelaki, Nikshych, and Ostrik 2015, §§2.1–2.2, pp. 21–25; §2.4, pp. 30–31; §§2.8–2.10, pp. 36–42; and §§8.1–8.2, pp. 195–198, PDF.

A graded-vector-space test category

Fix an integer $N\ge2$, let $G=\mathbb Z/N\mathbb Z$, and choose $\zeta=e^{2\pi i/N}$. Let $\mathcal V_N^\zeta$ have finite-dimensional $G$-graded complex vector spaces

$$V=\bigoplus_{a\in G}V_a$$

as objects and degree-preserving linear maps as morphisms. Define

$$(V\otimes W)_r = \bigoplus_{a+b=r}V_a\otimes W_b,$$

where all degrees are computed in $G$. The unit is $\mathbb C_0$, concentrated in degree zero. The usual vector-space associator and unitors preserve the grading, so they give the required natural isomorphisms.

For degree-preserving maps $f:V\to V'$ and $g:W\to W'$,

$$(f\otimes g)(v_a\otimes w_b) = f(v_a)\otimes g(w_b).$$

Homogeneous pure tensors span, so the interchange law follows from

$$\begin{aligned} &((f'\circ f)\otimes(g'\circ g))(v_a\otimes w_b)\\ &\quad= f'(f(v_a))\otimes g'(g(w_b))\\ &\quad= ((f'\otimes g')\circ(f\otimes g))(v_a\otimes w_b). \end{aligned}$$

Likewise, both pentagon routes send

$$(((u_a\otimes v_b)\otimes w_c)\otimes z_d)$$

to $u_a\otimes(v_b\otimes(w_c\otimes z_d))$, and both triangle routes send $(v_a\otimes1)\otimes w_b$ to $v_a\otimes w_b$. In the usual construction the differently parenthesized vector spaces are canonically isomorphic, not literally the same object.

Rigidity adds duals

Dual terminology varies between sources, so the arrow types will be part of the convention here. A left dual of $X$ is an object ${}^{\vee}X$ with

$$\mathrm{ev}^{L}_X:{}^{\vee}X\otimes X\longrightarrow\mathbf 1, \qquad \mathrm{coev}^{L}_X:\mathbf 1\longrightarrow X\otimes{}^{\vee}X.$$

A right dual is an object $X^{\vee}$ with

$$\mathrm{ev}^{R}_X:X\otimes X^{\vee}\longrightarrow\mathbf 1, \qquad \mathrm{coev}^{R}_X:\mathbf 1\longrightarrow X^{\vee}\otimes X.$$

Suppressing only the coherent associators and unitors, the left-dual snake identities are

$$\begin{aligned} (\operatorname{id}_X\otimes\mathrm{ev}^{L}_X) \circ(\mathrm{coev}^{L}_X\otimes\operatorname{id}_X) &=\operatorname{id}_X,\\ (\mathrm{ev}^{L}_X\otimes\operatorname{id}_{{}^{\vee}X}) \circ(\operatorname{id}_{{}^{\vee}X}\otimes\mathrm{coev}^{L}_X) &=\operatorname{id}_{{}^{\vee}X}. \end{aligned}$$

The right-dual identities reverse the tensor orders:

$$\begin{aligned} (\mathrm{ev}^{R}_X\otimes\operatorname{id}_X) \circ(\operatorname{id}_X\otimes\mathrm{coev}^{R}_X) &=\operatorname{id}_X,\\ (\operatorname{id}_{X^{\vee}}\otimes\mathrm{ev}^{R}_X) \circ(\mathrm{coev}^{R}_X\otimes\operatorname{id}_{X^{\vee}}) &=\operatorname{id}_{X^{\vee}}. \end{aligned}$$

An object is dualizable when it has the required dual data. This page calls a monoidal category rigid when every object has both a left and a right dual; some sources separately say left-rigid or right-rigid.

Duals in the running example

For $V\in\mathcal V_N^\zeta$, define the graded dual by

$$(V^\vee)_r=(V_{-r})^*.$$

It realizes both left and right duals. Choose homogeneous bases $e_{a,i}\in V_a$ with dual basis elements $e^{a,i}\in(V^\vee)_{-a}$. The left-dual maps are

$$\mathrm{ev}^{L}_V(e^{a,i}\otimes e_{b,j}) = \delta_{ab}\delta_{ij}, \qquad \mathrm{coev}^{L}_V(1) = \sum_{a,i}e_{a,i}\otimes e^{a,i}.$$

The right-dual maps use the reversed tensors:

$$\mathrm{ev}^{R}_V(e_{b,j}\otimes e^{a,i}) = \delta_{ab}\delta_{ij}, \qquad \mathrm{coev}^{R}_V(1) = \sum_{a,i}e^{a,i}\otimes e_{a,i}.$$

For example, the first left snake sends

$$v \longmapsto \sum_{a,i}e_{a,i}\otimes e^{a,i}\otimes v \longmapsto \sum_{a,i}e_{a,i}e^{a,i}(v) =v.$$

The left coevaluation tensor is basis independent: it corresponds to $\operatorname{id}_V$ under the usual identification of $V\otimes V^*$ with finite-rank endomorphisms. The right coevaluation is its ordinary vector-space tensor flip—not the chosen $\zeta$-braiding—and is basis independent for the same reason.

Finite dimensionality is essential. If an infinite-dimensional algebraic vector space had a coevaluation

$$1\longmapsto\sum_{i=1}^{m}v_i\otimes\phi_i,$$

then the snake identity would express every $v$ as $\sum_i v_i\phi_i(v)$, forcing the entire space into the finite-dimensional span of the $v_i$. Therefore the category of all complex vector spaces is symmetric monoidal but not rigid.

Dualizable does not mean tensor-invertible. For instance, $\mathbb C^2$ is dualizable, but $\mathbb C^2\otimes(\mathbb C^2)^*$ is four-dimensional and therefore not isomorphic to the unit $\mathbb C$.

Braiding is coherent exchange

A braiding is a natural isomorphism

$$c_{X,Y}:X\otimes Y\longrightarrow Y\otimes X$$

satisfying two hexagon identities. Naturality means that for $f:X\to X'$ and $g:Y\to Y'$,

$$c_{X',Y'}\circ(f\otimes g) = (g\otimes f)\circ c_{X,Y}.$$

In coherence-suppressed notation, the two hexagons become

$$\begin{aligned} c_{X,Y\otimes Z} &= (\operatorname{id}_Y\otimes c_{X,Z}) \circ(c_{X,Y}\otimes\operatorname{id}_Z),\\ c_{X\otimes Y,Z} &= (c_{X,Z}\otimes\operatorname{id}_Y) \circ(\operatorname{id}_X\otimes c_{Y,Z}). \end{aligned}$$

In a non-strict presentation, associators occur along these routes. Coherence suppression is what makes the shorter equations well-typed; it is not a claim that the associators were absent.

A braided but nonsymmetric calculation

Equip $\mathcal V_N^\zeta$ with

$$c_{V,W}(v_a\otimes w_b) = \zeta^{ab}w_b\otimes v_a$$

on homogeneous tensors. The phase is well-defined on residue classes because $\zeta^N=1$. The exchange preserves total degree because $G$ is abelian, and degree-preserving maps make naturality immediate. The two hexagons reduce to

$$\zeta^{a(b+c)}=\zeta^{ab}\zeta^{ac}, \qquad \zeta^{(a+b)c}=\zeta^{ac}\zeta^{bc}.$$

Thus the exchange phase is a bicharacter; an arbitrary phase assignment would not necessarily satisfy the hexagons.

A braided monoidal category is symmetric only if

$$c_{Y,X}\circ c_{X,Y} = \operatorname{id}_{X\otimes Y}$$

for all $X,Y$. In the running example, the double braid is

$$(c_{W,V}\circ c_{V,W})(v_a\otimes w_b) = \zeta^{2ab}v_a\otimes w_b.$$

For $N=2$, this is the symmetric super sign rule $c(v_a\otimes w_b)=(-1)^{ab}w_b\otimes v_a$. For $N=3$ and $a=b=1$, the double braid is multiplication by $\zeta^2\ne1$, so the category is braided but not symmetric. In a merely braided category, $c_{Y,X}$ need not equal $c_{X,Y}^{-1}$. The reverse-crossing, double-braid, symmetry, and graphical-coherence distinctions are reviewed in Selinger 2011, §3.1, pp. 9–11; §§3.3–3.5, pp. 14–18; and §4.1, pp. 18–19.

Let $L_a$ denote $\mathbb C$ concentrated in degree $a$. Then

$$L_a\otimes L_b\cong L_{a+b}, \qquad L_a^\vee\cong L_{-a}.$$

For $N=3$, $L_1\otimes L_1\cong L_2$ and $L_1\otimes L_2\cong L_0$, while the double braid on $L_1\otimes L_1$ is $\zeta^2\operatorname{id}$. This is a controlled model for the algebraic words “fusion,” “conjugate,” and “exchange,” not evidence that a QFT with these sectors exists.

An ordinary functor need not preserve tensor structure

For a functor $F:\mathcal C\to\mathcal D$ to be strong monoidal in the convention used here, one supplies coherent natural isomorphisms

$$J_{X,Y}:F(X)\otimes_{\mathcal D}F(Y) \xrightarrow{\sim} F(X\otimes_{\mathcal C}Y), \qquad J_0:\mathbf 1_{\mathcal D}\xrightarrow{\sim}F(\mathbf 1_{\mathcal C}).$$

Some sources reverse both comparison arrows; invertibility makes the strong versions equivalent, but the two conventions must not be mixed. A braided strong monoidal functor additionally obeys

$$J_{Y,X}\circ c^{\mathcal D}_{F(X),F(Y)} = F(c^{\mathcal C}_{X,Y})\circ J_{X,Y}.$$

Forgetting the grading gives a strong monoidal functor

$$U:\mathcal V_N^\zeta\longrightarrow\mathbf{FinVect}_{\mathbb C},$$

using the canonical tensor identifications. For nontrivial exchange phases it is not braided relative to the ordinary flip in $\mathbf{FinVect}_{\mathbb C}$: the source route contributes $\zeta^{ab}$ while the target flip does not. Monoidality and braidedness are therefore additional structure and compatibility, not consequences of being a functor.

A controlled QFT translation

Let $\mathfrak N$ be a von Neumann algebra. There is a category whose objects are unital $*$-endomorphisms $\alpha:\mathfrak N\to\mathfrak N$. An arrow $t:\alpha\to\beta$ is an intertwiner $t\in\mathfrak N$ satisfying

$$t\,\alpha(n)=\beta(n)\,t \qquad (n\in\mathfrak N).$$

Arrow composition is multiplication in $\mathfrak N$, and the identity arrow at every endomorphism is $1_{\mathfrak N}$. Choose the object-tensor convention

$$\alpha_1\otimes\alpha_2 := \alpha_1\circ\alpha_2, \qquad \mathbf 1:=\operatorname{id}_{\mathfrak N}.$$

For $t_i:\alpha_i\to\beta_i$, define

$$t_1\otimes t_2 := t_1\,\alpha_1(t_2) = \beta_1(t_2)\,t_1.$$

The equality uses the intertwiner equation for $t_1$. The claimed type can be checked directly:

$$\begin{aligned} &(t_1\alpha_1(t_2))\,\alpha_1\alpha_2(n)\\ &\quad= t_1\alpha_1(t_2\alpha_2(n))\\ &\quad= t_1\alpha_1(\beta_2(n)t_2)\\ &\quad= \beta_1\beta_2(n)\,t_1\alpha_1(t_2). \end{aligned}$$

Hence $t_1\otimes t_2$ is an intertwiner $\alpha_1\alpha_2\to\beta_1\beta_2$. To verify bifunctoriality, let $s_i:\beta_i\to\gamma_i$. Identity arrows obey

$$1_{\alpha_1}\otimes1_{\alpha_2} = 1_{\mathfrak N}\,\alpha_1(1_{\mathfrak N}) = 1_{\mathfrak N} = 1_{\alpha_1\alpha_2},$$

and the interchange law follows from

$$\begin{aligned} (s_1t_1)\otimes(s_2t_2) &=s_1t_1\alpha_1(s_2t_2)\\ &=s_1\beta_1(s_2)t_1\alpha_1(t_2)\\ &=(s_1\otimes s_2)(t_1\otimes t_2). \end{aligned}$$

Tensoring arrows is strictly associative as well:

$$(t_1\otimes t_2)\otimes t_3 = t_1\alpha_1(t_2\alpha_2(t_3)) = t_1\otimes(t_2\otimes t_3).$$

Together with literal associativity and unitality of endomorphism composition, this gives a strict monoidal category. Its tensor product is composition of endomorphisms, not a Hilbert-space tensor factorization. The endomorphism tensor category and the tensoring formula for intertwiners are given in Bischoff, Longo, Kawahigashi, and Rehren 2015, §2, p. 6, and §3, p. 14.

Nothing in this construction automatically supplies duals or a braiding. In the DHR setting, a suitable subcategory of localized, transportable endomorphisms acquires a braiding from locality and charge transporters under the framework’s operator-algebraic hypotheses; conjugates and finite statistical dimension require further results. Those are physical and analytic theorems, not consequences of the words “monoidal category.” The additional hypotheses leading to the DHR subcategory and its braiding are summarized in Bischoff, Longo, Kawahigashi, and Rehren 2015, §§5.1.2–5.1.3, pp. 75–77; the primary braid-statistics and exchange-algebra analysis is Fredenhagen, Rehren, and Schroer 1989, §2, pp. 203–206, and p. 221.

The canonical specialist continuation is Endomorphisms, Intertwiners, and Tensor Products. The fuller localization, transport, and sector-theoretic derivation belongs at that canonical continuation. This primer only supplies the categorical types needed to read such a derivation and does not by itself satisfy every specialist prerequisite.

Separating the structures

The following failures prevent common overclaims:

False implication	Counterexample or missing input
strict $\Rightarrow$ rigid	The discrete monoidal category $(\mathbb N,+,0)$ is strict, but no $n>0$ has a dual.
symmetric $\Rightarrow$ rigid	Vector spaces are symmetric monoidal, but infinite-dimensional ones are not algebraically dualizable.
rigid $\Rightarrow$ braided	The discrete monoidal category on a nonabelian group is rigid, but $gh\to hg$ need not exist.
dualizable $\Rightarrow$ tensor-invertible	$\mathbb C^2$ is dualizable, but $\mathbb C^2\otimes(\mathbb C^2)^*\not\cong\mathbb C$.
braided $\Rightarrow$ symmetric	$\mathcal V_3^\zeta$ has double braid $\zeta^{2ab}$, which is nontrivial for $a=b=1$.
monoidal $\Rightarrow$ physical factorization	A monoidal product needs an independent physical interpretation and theorem.

In the nonabelian-group row, “discrete” means that the objects are group elements and the only morphisms are identities. Tensoring objects uses group multiplication. Every $g$ has dual $g^{-1}$, but a braiding would require an arrow $gh\to hg$ for every pair; such an arrow exists only when the two objects are equal.

Common pitfalls

Confusing tensor with sequential composition. The types of $f\otimes g$ and $g\circ f$ are different, and the former does not require the latter’s matching condition. The interchange law relates the operations without collapsing them.

Turning coherence into literal equality. Associators and unitors remain maps in a non-strict category. Coherence makes canonical rebracketing unambiguous; strictification replaces the category by a monoidally equivalent presentation.

Equating rigidity with strictness or invertibility. Rigidity is controlled by evaluation, coevaluation, and the snake identities. Neither strict associativity nor the existence of a dual makes evaluation an isomorphism.

Calling an arbitrary swap a braiding. A braiding must be natural and obey both hexagons. Symmetry is stronger still: the correctly typed condition is $c_{Y,X}\circ c_{X,Y}=\operatorname{id}_{X\otimes Y}$.

Assuming a functor preserves tensor data. A strong monoidal functor needs comparison isomorphisms and coherence. Preserving a braiding is a further compatibility condition.

Inferring physics from categorical structure. A braided rigid category is not automatically unitary, semisimple, finite, modular, or realized by a QFT. Locality and exchange statistics enter only through framework-specific hypotheses and theorems.

Check your understanding

Retrieval. For $f:X\to X'$ and $g:Y\to Y'$, type $f\otimes g$ and explain why $g\circ f$ may be undefined.

Answer check. The tensor arrow is $f\otimes g:X\otimes Y\to X'\otimes Y'$. The composite $g\circ f$ requires $X'=Y$, a condition not needed for the tensor arrow.

Proof checkpoint. If an infinite-dimensional algebraic vector space had a coevaluation represented by a finite tensor $\sum_{i=1}^m v_i\otimes\phi_i$, what would the snake identity imply?

Answer check. It would give $v=\sum_i v_i\phi_i(v)$ for every $v$, so the whole space would lie in the finite span of the $v_i$, a contradiction.

Counterexample check. In $\mathcal V_3^\zeta$, compute the double braid on $v_1\otimes w_1$ and decide whether the braiding is symmetric.

Answer check. It is $\zeta^2v_1\otimes w_1\ne v_1\otimes w_1$. The category is braided but not symmetric.

Calculation checkpoint. Verify the first hexagon phase on $v_a\otimes w_b\otimes z_c$.

Answer check. Exchanging $v_a$ past $w_b\otimes z_c$ in one step gives $\zeta^{a(b+c)}$; two exchanges give $\zeta^{ab}\zeta^{ac}$. They agree because the phase is a bicharacter.

Transfer. In the endomorphism category, type $t_1\alpha_1(t_2)$ and list what is still missing before it can represent a physical exchange law.

Answer check. It lies in $\operatorname{Hom}(\alpha_1\alpha_2,\beta_1\beta_2)$. A physical exchange requires an appropriate localized and transportable sector subcategory, locality and the relevant operator-algebraic hypotheses, plus a construction of the braiding; none follows from endomorphism composition alone.

What has been established

A monoidal category coherently combines objects and arrows in parallel; rigidity supplies typed dual data; braiding supplies coherent exchange; and symmetry makes the double exchange trivial. The graded-vector-space example checks all of those laws and displays a genuinely nonsymmetric braiding. The endomorphism example then shows how a strict tensor product can arise in QFT language while leaving locality, conjugates, and exchange as separate theorems.

For the chapter-wide comparison of monoidal, homological, and local-to-global requirements, continue eventually to Derived, Higher, and Factorization Frameworks: a Boundary Map.

References

• Marcel Bischoff, Roberto Longo, Yasuyuki Kawahigashi, and Karl-Henning Rehren, Tensor Categories and Endomorphisms of von Neumann Algebras (with Applications to Quantum Field Theory), SpringerBriefs in Mathematical Physics 3 (2015), arXiv:1407.4793v3, §2, p. 6; §3, p. 14; and §§5.1.2–5.1.3, pp. 75–77. The strict endomorphism tensor category, tensoring intertwiners, and the additional hypotheses used to obtain the DHR subcategory and braiding.

• Pavel Etingof, Shlomo Gelaki, Dmitri Nikshych, and Victor Ostrik, Tensor Categories, Mathematical Surveys and Monographs 205, American Mathematical Society (2015), author-final manuscript, PDF, §§2.1–2.2, pp. 21–25; §2.4, pp. 30–31; §§2.8–2.10, pp. 36–42; and §§8.1–8.2, pp. 195–198. Monoidal data and coherence, strong monoidal functors, strictification, left and right duals, rigidity, braiding, and symmetry. Consulted together with the official corrections dated 14 February 2026, PDF, especially the corrected compatible-uniqueness wording for duals on p. 2.

• Klaus Fredenhagen, Karl-Henning Rehren, and Bert Schroer, “Superselection Sectors with Braid Group Statistics and Exchange Algebras I: General Theory,” Communications in Mathematical Physics 125 (1989), 201–226, DOI:10.1007/BF01217906, §2, pp. 203–206, and p. 221. The primary braid-statistics and exchange-algebra analysis, built on the DHR localized-endomorphism and intertwiner framework, and the roles of Einstein causality and positivity.

• Peter Selinger, “A Survey of Graphical Languages for Monoidal Categories,” in New Structures for Physics, Lecture Notes in Physics 813, Springer (2011), arXiv:0908.3347v1, §3.1, pp. 9–11; §§3.3–3.5, pp. 14–18; and §4.1, pp. 18–19 (arXiv pagination). Pentagon and triangle coherence, the distinction between a reverse crossing and a double braid, symmetry, dual maps, and the scope of graphical coherence.