A rich source of dualizable categories are rigid symmetric monoidal categories, introduced by Gaitsgory–Rozenblyum [Gaitsgory and Rozenblyum (2017), Chapter 1, Section 9]. We have already met them in Definition 1.4; we now give a more detailed treatment.

Definition 6.1 ([Gaitsgory and Rozenblyum (2017), Chapter 1, Definition 9.1.2]). Let \(C \in \CAlg (\PrL _{\st })\) be a presentably symmetric monoidal stable category, with unit \(\unit _C\) and multiplication \(\mu \colon C \otimes C \to C\). Then \(C\) is called rigid if

(1)

the unit \(\unit _C\) is compact in \(C\), or equivalently, the unit functor \(u\colon \Sp \to C\), \(\S \mapsto \unit _C\), is strongly continuous;

(2)

the right adjoint \(\mu ^R\colon C \to C \otimes C\) preserves colimits, and its canonical lax \(C \otimes C\)-linear structure is strict; equivalently, \(\mu \) admits a right adjoint in \(\Mod _{C \otimes C}(\PrL _{\st })\).

Remark 6.2. The equivalence of the two formulations in (2) is a general fact about adjunctions internal to module categories: a right adjoint of a \(C \otimes C\)-linear functor inherits a lax \(C \otimes C\)-linear structure, and it is a right adjoint in \(\Mod _{C \otimes C}(\PrL _{\st })\) if and only if it preserves colimits and this lax structure is strict. Definition 6.1 is the definition of Gaitsgory–Rozenblyum, and it agrees with Ramzi’s definition [Ramzi (2024b), Definitions 4.5 and 4.36] (Ramzi additionally requires \(C\) to be dualizable, which is automatic by Proposition 6.3). It is also equivalent to Definition 1.4, which asks that for every finite set \(I\) the multiplication \(\mu _I\colon C^{\otimes I} \to C\) admits a right adjoint in \(\Mod _{C^{\otimes I}}(\PrL _{\st })\). Indeed, for \(I = \emptyset \) the map \(\mu _{\emptyset }\) is the unit functor \(u\colon \Sp \to C\), and \(\Mod _{\Sp }(\PrL _{\st }) = \PrL _{\st }\), so this case is condition (1); the case \(|I| = 1\) is vacuous, and the case \(|I| = 2\) is condition (2). Conversely, if (2) holds, then by induction \(\mu _{I \sqcup \{*\}} = \mu \circ (\mu _I \otimes \id _C)\) is a composite of two functors admitting right adjoints in \(\Mod _{C^{\otimes I} \otimes C}(\PrL _{\st })\): for \(\mu _I \otimes \id _C\) this follows by tensoring the adjunction for \(\mu _I\) with \(C\), and for \(\mu \) by restricting the \(C \otimes C\)-linear adjunction along the algebra map \(\mu _I \otimes \id _C\colon C^{\otimes I} \otimes C \to C \otimes C\).

Proposition 6.3 ([Gaitsgory and Rozenblyum (2017), Chapter 1, Section 9.2.1]). Let \(C\) be rigid. Then \(C\) is self-dual in \(\PrL _{\st }\), with evaluation and coevaluation given by \[ \ev \colon C \otimes C \xrightarrow {\mu } C \xrightarrow {\Gamma := \hom _C(\unit _C,-)} \Sp , \qquad \coev \colon \Sp \xrightarrow {u} C \xrightarrow {\mu ^R} C \otimes C. \] In particular, \(C\) is dualizable.

Proof.

Note first that both functors preserve colimits: \(\Gamma \) preserves colimits since \(\unit _C\) is compact (and \(C\) is stable), and \(\mu ^R\) preserves colimits by assumption. Moreover, \(\Gamma \) is right adjoint to \(u\). Write \(\eta := \mu ^R(\unit _C) \in C \otimes C\), so that \(\coev \) is the colimit-preserving functor \(\Sp \to C \otimes C\) sending \(\S \) to \(\eta \). We verify the first triangle identity, i.e. that the composite \[ C \simeq \Sp \otimes C \xrightarrow {\coev \otimes C} C \otimes C \otimes C \xrightarrow {C \otimes \ev } C \otimes \Sp \simeq C \] is equivalent to the identity; the second one follows by symmetry. Since \(C \otimes \ev = (C \otimes \Gamma ) \circ (C \otimes \mu )\), the composite sends \(c \in C\) to \[ (C \otimes \Gamma )\big ((C \otimes \mu )(\eta \boxtimes c)\big ) \simeq (C \otimes \Gamma )\big (\eta \cdot (\unit _C \boxtimes c)\big ), \] where \(\cdot \) denotes the multiplication in \(C \otimes C\). Now we use the linearity of \(\mu ^R\): for \(a \in C \otimes C\) and \(x \in C\) we have \(\mu ^R(\mu (a) \otimes x) \simeq a \cdot \mu ^R(x)\). Taking \(a = \unit _C \boxtimes c\) and \(x = \unit _C\) gives \[ \eta \cdot (\unit _C \boxtimes c) = (\unit _C \boxtimes c) \cdot \mu ^R(\unit _C) \simeq \mu ^R(c). \] Hence the composite is \((C \otimes \Gamma ) \circ \mu ^R\colon C \to C\), naturally in \(c\) (as all functors involved preserve colimits, it suffices to identify the functors, which the above does naturally in \(c\)). Since \(C \otimes - \) is a functor of \((\infty ,2)\)-categories, \(C \otimes \Gamma \) is right adjoint to \(C \otimes u\colon C \to C \otimes C\), \(x \mapsto x \boxtimes \unit _C\). Hence \((C \otimes \Gamma ) \circ \mu ^R\) is right adjoint to \(\mu \circ (C \otimes u) \simeq \id _C\), and is therefore equivalent to the identity. □

The linearity of the right adjoint to multiplication is what makes the triangle identity work. Trace-class morphisms express the corresponding duality data for individual morphisms.

Next, we introduce trace-class morphisms, following Clausen–Scholze [Scholze (2026)].

Definition 6.4. Let \(C\) be a symmetric monoidal category. A morphism \(f\colon x \to y\) in \(C\) is called trace-class if there exist an object \(z \in C\) and maps \(\alpha \colon \unit _C \to y \otimes z\) and \(\beta \colon z \otimes x \to \unit _C\) such that \(f\) is homotopic to the composite \[ x \simeq \unit _C \otimes x \xrightarrow {\alpha \otimes x} y \otimes z \otimes x \xrightarrow {y \otimes \beta } y \otimes \unit _C \simeq y. \]

Lemma 6.5. Let \(C\) be a symmetric monoidal category whose homotopy category is idempotent complete; for instance, \(C\) stable and idempotent complete [Lurie (2017), Lemma 1.2.4.6]. Then an object \(x \in C\) is dualizable if and only if \(\id _x\) is trace-class.

Proof.

If \(x\) is dualizable with dual \(x^{\vee }\), then the coevaluation and evaluation are a witness for \(\id _x\). Conversely, let \((z,\alpha ,\beta )\) be a witness for \(\id _x\), so that \(\alpha \colon \unit \to x \otimes z\) and \(\beta \colon z \otimes x \to \unit \) satisfy the first triangle identity \((x \otimes \beta )(\alpha \otimes x) \simeq \id _x\), but not necessarily the second one. Consider \[ e := (\beta \otimes z)(z \otimes \alpha )\colon z \to z \otimes x \otimes z \to z. \] Inserting the first triangle identity in the middle of \(e \circ e\) shows that \(e \circ e \simeq e\), and similarly \(\beta (e \otimes x) \simeq \beta \) and \((x \otimes e)\alpha \simeq \alpha \). As the homotopy category is idempotent complete, \(e\) splits: \(e \simeq sr\) with \(r\colon z \to z'\), \(s\colon z' \to z\) and \(rs \simeq \id _{z'}\). Put \(\alpha ' := (x \otimes r)\alpha \) and \(\beta ' := \beta (s \otimes x)\). Then \[ (x \otimes \beta ')(\alpha ' \otimes x) \simeq (x \otimes \beta )(x \otimes e \otimes x)(\alpha \otimes x) \simeq (x \otimes \beta )(\alpha \otimes x) \simeq \id _x \] and \((\beta ' \otimes z')(z' \otimes \alpha ') \simeq r e s \simeq rsrs \simeq \id _{z'}\). So \((\alpha ',\beta ')\) exhibits \(z'\) as a dual of \(x\); since dualizability can be checked in the homotopy category, \(x\) is dualizable. □

Remark 6.6.

(1)

If \(C\) is closed, write \(x^{\vee } := \iHom (x,\unit _C)\) for the predual of \(x\). Then \(f\) is trace-class if and only if the map \(\unit _C \to \iHom (x,y)\) classifying \(f\) factors through the canonical map \(y \otimes x^{\vee } \to \iHom (x,y)\): given \((z,\alpha ,\beta )\) as above, \(\beta \) is adjoint to a map \(z \to x^{\vee }\), and composing \(\alpha \) with it gives the desired factorization; conversely, given a factorization \(\unit _C \to y \otimes x^{\vee }\), take \(z = x^{\vee }\) and \(\beta \) the evaluation map.

(2)

Trace-class morphisms form a two-sided ideal: if \(f\) is trace-class with witness \((z,\alpha ,\beta )\), then \(hfk\) is trace-class with witness \((z,(h \otimes z)\alpha ,\beta (z \otimes k))\).

(3)

If \(C\) is stable and idempotent complete (for instance presentable and stable), then the identity of \(x\) is trace-class if and only if \(x\) is dualizable, see Lemma 6.5.

(4)

A symmetric monoidal functor preserves trace-class morphisms, since it preserves the data \((z,\alpha ,\beta )\).

Lemma 6.7. Let \(C \in \CAlg (\PrL _{\st })\) be such that \(\unit _C\) is compact. Then every trace-class morphism in \(C\) is compact in the sense of Definition 3.3.

Proof.

Let \(f\colon x \to y\) be trace-class with witness \((z,\alpha ,\beta )\). We construct a factorization of \(f^*\colon \Map _C(y,\colim (-)) \to \Map _C(x,\colim (-))\) through \(\Map _{\Ind (C)}(y(x),-)\), naturally in \(W = \quotcolim _j w_j \in \Ind (C)\). Given \(g\colon y \to \colim _j w_j\), consider \[ (g \otimes z) \circ \alpha \colon \unit _C \to (\colim _j w_j) \otimes z \simeq \colim _j (w_j \otimes z). \] Since \(\unit _C\) is compact, this defines a point of \(\colim _j \Map _C(\unit _C,w_j \otimes z)\), and postcomposing with the natural map \[ \colim _j \Map _C(\unit _C,w_j \otimes z) \to \colim _j \Map _C(x,w_j), \qquad \gamma \mapsto (w_j \otimes \beta ) \circ (\gamma \otimes x), \] we obtain a point of \(\colim _j \Map _C(x,w_j) = \Map _{\Ind (C)}(y(x),W)\). All these maps are natural in \(W\), and the image of this point in \(\Map _C(x,\colim _j w_j)\) is \((\colim _j w_j \otimes \beta ) \circ ((g \otimes z)\alpha \otimes x) \simeq g \circ (y \otimes \beta ) \circ (\alpha \otimes x) \simeq gf\). □

The converse holds for rigid categories; this is the main structural result about rigid categories. It was formulated in Clausen’s lectures as the definition of rigidity for dualizable symmetric monoidal categories.

Theorem 6.8 (Ramzi [Ramzi (2024b), Corollaries 2.6, 4.55 and 4.60, Example 2.9]; see also [Efimov (2025), Propositions 1.6 and 1.7]). Let \(C \in \CAlg (\PrL _{\st })\). Then the following are equivalent:

(1)

\(C\) is rigid;

(2)

\(\unit _C\) is compact, and \(C\) is generated under colimits by objects of the form \(\colim (x_0 \to x_1 \to x_2 \to \cdots )\) in which each map \(x_n \to x_{n+1}\) is trace-class;

(3)

\(C\) is dualizable, and a morphism in \(C\) is compact if and only if it is trace-class.

We will not prove this theorem. Note however that some implications are formal given what we know: (1) \(\implies \) the first half of (3) is Proposition 6.3, and (3) \(\implies \) (2) follows from Lemma 3.5(2), since \(\id _{\unit _C}\) is trace-class and hence compact, and Remark 3.21, which shows that a dualizable category is generated by sequential colimits along compact morphisms. The hard part is that in a rigid category every compact morphism is trace-class [Ramzi (2024b), Corollary 4.55, Example 2.9 and Corollary 2.6], and that (2) implies (1) [Ramzi (2024b), Corollary 4.60]. Note that the dualizability hypothesis in (3) was not stated in the talk.

Corollary 6.9. Let \(F\colon C \to D\) be a colimit-preserving symmetric monoidal functor between rigid categories. Then \(F\) is strongly continuous. Consequently, the category \(\mathrm {Rig}\) of rigid symmetric monoidal categories and colimit-preserving symmetric monoidal functors is a full subcategory of \(\CAlg (\Pr ^{\dual })\).

Proof.

By Theorem 6.8, the compact morphisms in \(C\) are trace-class. Since \(F\) is symmetric monoidal, it preserves trace-class morphisms (Remark 6.6(4)), and trace-class morphisms in \(D\) are compact by Lemma 6.7. So \(F\) preserves compact morphisms, and hence is strongly continuous by Proposition 3.22. □

Remark 6.10. The argument shows more generally that a colimit-preserving symmetric monoidal functor \(F\colon C \to D\) with \(C\) rigid and \(D\) dualizable with compact unit is strongly continuous. In fact, the dualizability hypothesis on \(D\) can be removed [Efimov (2025), Proposition 1.9].

Example 6.11. Let \(C \in \CAlg (\PrL _{\st })\) be compactly generated. Then \(C\) is rigid if and only if \(\unit _C\) is compact and every compact object of \(C\) is dualizable, if and only if the compact objects of \(C\) are precisely the dualizable objects [Gaitsgory and Rozenblyum (2017), Chapter 1, Lemma 9.1.5 and Corollary 9.1.7]. Indeed, if \(\unit _C\) is compact, then dualizable objects are compact (as \(\Map (x,-) \simeq \Map (\unit _C,x^{\vee } \otimes -)\)), and conversely if all compact objects are dualizable, then \(C = \Ind (C^{\omega })\) is generated by objects whose identity is trace-class, so (2) of Theorem 6.8 holds. Examples include \(\D (R)\) for a commutative ring \(R\), more generally \(\D _{\mathrm {qc}}(X)\) for a qcqs scheme \(X\), and \(\Sp \) itself.

Example 6.12. Let \(X\) be a compact Hausdorff space. Then \(\Shv (X;\Sp )\) is rigid; this can be seen as a manifestation of Verdier duality, see [Efimov (2025), Example 1.4]. For instance, the unit (the constant sheaf) is compact since global sections \(\Gamma (X;-)\) commute with filtered colimits for compact \(X\) [Lurie (2009), Corollary 7.3.4.12]. In particular, \(\Shv (X;\Sp )\) is dualizable and self-dual. For every presentable stable category \(D\) we have \(\Shv (X;D) \simeq \Shv (X;\Sp ) \otimes D\) [Krause et al. (2024), Proposition 2.8.7], so \(\Shv (X;D)\) is dualizable whenever \(D\) is. We will study these categories in detail in Chapter 9.

Example 6.13. If \(X\) is merely locally compact Hausdorff and not compact, then \(\Shv (X;\Sp )\) is not rigid, since the unit is not compact. It is still dualizable, and even locally rigid in the sense of Ramzi [Ramzi (2024b), Definition 4.5 and Example 4.10] (see also [Efimov (2025), Definition 1.36 and Example 1.38] and [Krause et al. (2024), Example 4.2.9, Corollary 4.4.13]): it satisfies condition (2) of Definition 6.1. Indeed, for \(j\colon X \hookrightarrow X^+\) the one-point compactification, \(j_!\) identifies \(\Shv (X;\Sp )\) with a smashing ideal in the rigid category \(\Shv (X^+;\Sp )\).