Proposition 6.17 (Clausen–Scholze; see [Clausen and Scholze (2026), Theorem 8.6] and [Efimov (2025), Propositions 1.27 and 1.28]).

(1)

For \(M \in C\), the following are equivalent:

(a)

\(M\) is nuclear;

(b)

for every compact \(P\), the map \(\hom _C(\unit _C,P^{\vee } \otimes M) \to \hom _C(P,M)\) is an equivalence of spectra (equivalently, since we may replace \(P\) by its shifts, the corresponding map of mapping spaces is an equivalence for every compact \(P\));

(d)

\(M \simeq \colim _{i \in I} M_i\) for a directed poset \(I\) and compact objects \(M_i\), such that for every \(i\) there is \(j \geq i\) with \(M_i \to M_j\) trace-class.

Moreover, these conditions are implied by

(c)

for every compact \(P\), the map \(P^{\vee } \otimes M \to \iHom (P,M)\) is an equivalence,

and if the tensor product of any two compact objects of \(C\) is compact, then (c) is equivalent to (a), (b) and (d).

(2)

The subcategory \(\Nuc (C) \subseteq C\) is closed under colimits. It is \(\omega _1\)-presentable, and its \(\omega _1\)-compact objects are precisely the basic nuclear objects.

Proof.

We first prove (1). (b) \(\Rightarrow \) (a): Taking \(\pi _0\) of the mapping spaces, every morphism \(P \to M\) lies in the image of \(\pi _0\lambda _M\) (with \(Q = \unit _C\)), i.e. is trace-class.

(a) \(\Rightarrow \) (d): Since \(C\) is compactly generated, \(M\) is a filtered colimit of compact objects, and by Theorem A.8 we may write \(M \simeq \colim _{i \in I} M_i\) with \(I\) a directed poset and \(M_i\) compact; denote by \(t_{ij}\) the transition maps and by \(\iota _i\colon M_i \to M\) the canonical maps. Fix \(i\). Since \(\iota _i\) is trace-class, it is \(\lambda _M(w)\) for some \(w \in \pi _0\Map _C(\unit _C, M \otimes M_i^{\vee })\). As \(\unit _C\) is compact and \(M \otimes M_i^{\vee } \simeq \colim _j M_j \otimes M_i^{\vee }\), the class \(w\) is the image of some \(w' \in \pi _0\Map _C(\unit _C, M_j \otimes M_i^{\vee })\) with \(j \geq i\). Then \(g := \lambda _{M_j}(w')\colon M_i \to M_j\) is trace-class, and \(\iota _j g \simeq \lambda _M(w) \simeq \iota _i\) by naturality of \(\lambda \). As \(M_i\) is compact, \(\pi _0\Map _C(M_i,M) \simeq \colim _k \pi _0\Map _C(M_i,M_k)\), so there is \(k \geq j\) with \(t_{jk}g \simeq t_{ik}\). Since trace-class morphisms form an ideal (Remark 6.6(2)), \(t_{ik}\) is trace-class.

(d) \(\Rightarrow \) (b): This is Lemma 6.16 with \(Q = \unit _C\).

(c) \(\Rightarrow \) (b): Apply \(\hom _C(\unit _C,-)\) and use \(\hom _C(\unit _C,\iHom (P,M)) \simeq \hom _C(P,M)\).

(d) \(\Rightarrow \) (c), if compact objects are closed under tensor products: For compact \(Q\), the map \(\hom _C(Q,P^{\vee } \otimes M) \to \hom _C(Q,\iHom (P,M)) \simeq \hom _C(Q \otimes P,M)\) is an equivalence by Lemma 6.16. Since \(C\) is compactly generated, the compact objects detect equivalences, so \(P^{\vee } \otimes M \to \iHom (P,M)\) is an equivalence.

We now prove (2). By (1), \(M\) is nuclear if and only if the maps in (b) are equivalences. Since \(\unit _C\) and \(P\) are compact, the functors \(\hom _C(\unit _C,P^{\vee } \otimes -)\) and \(\hom _C(P,-)\) preserve filtered colimits, and being exact they preserve all small colimits. Hence \(\Nuc (C)\) is closed under colimits in \(C\); in particular, colimits in \(\Nuc (C)\) are computed in \(C\).

Basic nuclear objects are nuclear by (d) (with \(I = \N \)). They are countable colimits of compact objects, hence \(\omega _1\)-compact in \(C\) (Corollary A.4), and hence also in \(\Nuc (C)\).

Next, every nuclear \(M\) is an \(\omega _1\)-filtered colimit of basic nuclear objects. Write \(M \simeq \colim _{i \in I} M_i\) as in (d), and choose for every \(i\) an element \(\sigma (i) \geq i\) such that \(M_i \to M_{\sigma (i)}\) is trace-class. Let \(\Sigma _{\sigma }\) be the set of countable directed subsets \(J \subseteq I\) with \(\sigma (J) \subseteq J\), ordered by inclusion. The proof of Proposition A.9 applies verbatim to \(\Sigma _{\sigma }\) instead of \(\Sigma \) (adding in the inductive construction also \(\sigma (u)\) for \(u \in U_k\)), and shows that \(\Sigma _{\sigma }\) is \(\omega _1\)-filtered and that \(M \simeq \colim _{J \in \Sigma _{\sigma }} M_J\) with \(M_J = \colim _{i \in J} M_i\). Each \(M_J\) is basic nuclear: enumerate \(J = \{q_0,q_1,\dots \}\), and choose inductively \(j_0 = q_0\) and \(j_{n+1} \in J\) with \(j_{n+1} \geq \sigma (j_n), q_{n+1}\). Then \(n \mapsto j_n\) is cofinal (Lemma A.7), so \(M_J \simeq \colim _n M_{j_n}\), and each \(M_{j_n} \to M_{j_{n+1}}\) factors through the trace-class map \(M_{j_n} \to M_{\sigma (j_n)}\), hence is trace-class.

The basic nuclear objects form an essentially small collection, since \(C^{\omega }\) is essentially small. Thus \(\Nuc (C)\) admits small colimits and is generated under \(\omega _1\)-filtered colimits by an essentially small collection of \(\omega _1\)-compact objects, so it is \(\omega _1\)-accessible by [Lurie (2009), Proposition 5.4.2.2], and hence \(\omega _1\)-presentable.

Finally, let \(X \in \Nuc (C)\) be \(\omega _1\)-compact. Writing \(X \simeq \colim _{J \in \Sigma _{\sigma }} X_J\) as above, the identity of \(X\) factors through some \(X_J\), so \(X\) is a retract of a basic nuclear object \(Y = \colim _n a_n\), where the \(a_n\) are compact and the transition maps \(a_n \to a_{n+1}\) are trace-class. Now we argue as in the proof of Proposition A.13, applied in \(C \simeq \Ind (C^{\omega })\): by Lemma A.11 and Lemma A.12, we find a strictly increasing \(f\colon \N \to \N \) with \(f(n) \geq n\) and a natural transformation \(\eta \colon a_{\bullet } \to a_{f(\bullet )}\) such that \(X \simeq \colim _n \Phi (n,n)\), where \(\Phi (k,n) = a_{f^k(n)}\). The transition map \(\Phi (n,n) \to \Phi (n+1,n+1)\) is the composite of the transition map \(a_{f^n(n)} \to a_{f^n(n+1)}\) of \(a_{\bullet }\), which is trace-class as \(f^n(n+1) > f^n(n)\), with \(\eta _{f^n(n+1)}\). So it is trace-class, and \(X\) is basic nuclear. □