We first show that semiorthogonal decompositions and gluings behave well with respect to dualizability. We then explain why a general extension behaves differently. The Tate-object classification at the end of this section is a more specialized application and may be omitted on a first reading.

Proposition 5.23 ([Efimov (2024), Proposition 2.76]). Let \(E = \lra {C,D}\) be a semiorthogonal decomposition of presentable stable categories in the sense of Definition 2.21. Then \(E\) is dualizable if and only if \(C\) and \(D\) are dualizable. In this case \(E^{\omega } = \lra {C^{\omega },D^{\omega }}\).

Proof.

The projections \(i_C^L\) and \(i_D^R\) preserve colimits, so \(C\) and \(D\) are retracts of \(E\) in \(\PrL _{\st }\). Thus dualizability of \(E\) implies that of \(C\) and \(D\).

Conversely, if \(C\) and \(D\) are dualizable, then \(E\) is dualizable by Proposition 4.25, since the strongly continuous inclusions \(i_C\) and \(i_D\) jointly generate \(E\).

The inclusions preserve compact objects. The projection \(i_C^L\) does too, since its right adjoint \(i_C\) preserves colimits. For compact \(e \in E\), the cofiber sequence \(i_Di_D^Re \to e \to i_Ci_C^Le\) therefore shows that \(i_Di_D^Re\) is compact. Since \(i_D\) is fully faithful and preserves filtered colimits, \(i_D^Re\) is compact in \(D\). This restricts the semiorthogonal decomposition to compact objects. □

Construction 5.24 (Gluing along a functor). Let \(F\colon D \to C\) be an accessible exact functor between presentable stable categories. Let \(E\) be the category of triples \((x,y,u)\) with \(x \in C\), \(y \in D\), and \(u\colon x \to F(y)\). The same comma category occurs in the construction of a recollement in [Topos, Construction 5.48]. This is a presentable stable category [Efimov (2024), Proposition 1.2]; colimits are computed on \(x\) and \(y\), using the natural maps \(\colim F(y_j) \to F(\colim y_j)\).

The fully faithful colimit-preserving functors \[ i_C(c) = (c,0,0), \qquad i_D(d) = (0,d,0) \] have right adjoints \(i_C^R(x,y,u) = \fib (u)\) and \(i_D^R(x,y,u) = y\). Moreover, \(i_C^L(x,y,u) = x\) is left adjoint to \(i_C\), and every object fits into a cofiber sequence \[ i_D(y) \to (x,y,u) \to i_C(x). \] We have \(\hom _E(i_D(d),i_C(c)) \simeq 0\). The functor \(i_D\) is always strongly continuous, since its right adjoint \((x,y,u) \mapsto y\) preserves colimits.

In the talk, as in [Efimov (2024), Definition 1.1], the inclusion of \(C\) was instead defined as \(c \mapsto (c[-1],0,0)\). Its right adjoint is then \((x,y,u) \mapsto \cofib (u)\), and its composite with \(i_D\) is \(F\) itself, while in our convention \(i_C^Ri_D \simeq \Omega F\). This shift plays no role in what follows.

Proposition 5.25 ([Efimov (2024), Proposition 2.77]). Let \(F\colon D \to C\) be an accessible exact functor between dualizable categories, and let \(E\) be the gluing of Construction 5.24. Then the following are equivalent:

(1)

\(F\) preserves colimits;

(2)

\(E\) is dualizable, and \(i_C\) and \(i_D\) are strongly continuous.

In this case, \(i_C\) and \(i_D\) exhibit a semiorthogonal decomposition \(E = \lra {C,D}\).

Proof.

(2) \(\implies \) (1): If \(i_C\) is strongly continuous, then \(i_C^R\) preserves colimits, and hence so does \(i_C^Ri_D \simeq \Omega F\). So \(F\) preserves colimits. This does not use the dualizability of \(E\).

(1) \(\implies \) (2): If \(F\) preserves colimits, then \(i_C^R(x,y,u) = \fib (u\colon x \to F(y))\) preserves colimits, so \(i_C\) is strongly continuous. The functor \(i_D\) is always strongly continuous. The functors \(i_C\) and \(i_D\) are fully faithful, we have \(\hom _E(i_D(d),i_C(c)) \simeq 0\), and every object of \(E\) is an extension of an object of \(i_C(C)\) by one of \(i_D(D)\). So we obtain a semiorthogonal decomposition \(E = \lra {C,D}\) in the sense of Definition 2.21, and \(E\) is dualizable by Proposition 5.23. □

The strong continuity condition in Proposition 5.25 is essential: a general strongly continuous localization sequence with dualizable outer terms need not have dualizable middle term (Example 5.30).

Semiorthogonal decompositions are inherited by categories of strongly continuous functors. For presentable stable categories \(T\) and \(E\), we write \(\Fun ^{\mathrm {LL}}(T,E) \subseteq \Fun ^{\colim }(T,E)\) for the full subcategory of strongly continuous functors. It is a stable subcategory: shifts of strongly continuous functors are strongly continuous, and the right adjoint of the cofiber of a map \(G \to G'\) of colimit-preserving functors is the fiber of the induced map \(G'^R \to G^R\) of right adjoints, which preserves colimits if \(G^R\) and \(G'^R\) do. In the following, we use the notion of a semiorthogonal decomposition of a (not necessarily presentable) stable category from Remark 2.24.

Proposition 5.26 ([Efimov (2024), Proposition 2.78]). Let \(E = \lra {i_1(C),i_2(D)}\) be a semiorthogonal decomposition of a presentable stable category, let \(\pi _1 := i_1^L\colon E \to C\) and \(\pi _2 := i_2^R\colon E \to D\) be the projections, and let \(T\) be a presentable stable category.

(1)

A colimit-preserving functor \(G\colon T \to E\) is strongly continuous if and only if \(\pi _1G\) and \(\pi _2G\) are strongly continuous. Postcomposition with \(i_1\) and \(i_2\) gives a semiorthogonal decomposition \[ \Fun ^{\mathrm {LL}}(T,E) = \lra {\Fun ^{\mathrm {LL}}(T,C),\Fun ^{\mathrm {LL}}(T,D)}. \]

(2)

A colimit-preserving functor \(G\colon E \to T\) is strongly continuous if and only if \(Gi_1\) and \(Gi_2\) are strongly continuous. Precomposition with \(\pi _1\) and \(\pi _2\) gives a semiorthogonal decomposition \[ \Fun ^{\mathrm {LL}}(E,T) = \lra {\Fun ^{\mathrm {LL}}(C,T),\Fun ^{\mathrm {LL}}(D,T)}. \]

Proof.

Recall from Proposition 2.23 that \(i_1\), \(i_2\), \(\pi _1\) and \(\pi _2\) are strongly continuous, that \(\pi _1i_1 \simeq \id \), \(\pi _2i_2 \simeq \id \) and \(\pi _1i_2 \simeq 0\), and that there is a cofiber sequence \(i_2\pi _2 \to \id _E \to i_1\pi _1\). In particular, composition with any of these four functors preserves strong continuity.

(1) If \(G\) is strongly continuous, then so are \(\pi _1G\) and \(\pi _2G\). Conversely, if \(\pi _1G\) and \(\pi _2G\) are strongly continuous, then so are \(i_1\pi _1G\) and \(i_2\pi _2G\), and hence so is \(G\), by the cofiber sequence \(i_2\pi _2G \to G \to i_1\pi _1G\) and the stability of \(\Fun ^{\mathrm {LL}}(T,E)\). Let \(j_1 := i_1 \circ -\) and \(j_2 := i_2 \circ -\). The pointwise adjunctions \(\pi _1 \dashv i_1\) and \(i_2 \dashv \pi _2\) show that \(\pi _1 \circ -\) is left adjoint to \(j_1\) and that \(\pi _2 \circ -\) is right adjoint to \(j_2\). Hence \(j_1\) and \(j_2\) are fully faithful, since \(\pi _1i_1 \simeq \id \) and \(\pi _2i_2 \simeq \id \), and \[ \Map (j_2(H'),j_1(H)) \simeq \Map (\pi _1i_2H',H) \simeq * \] for \(H \in \Fun ^{\mathrm {LL}}(T,C)\) and \(H' \in \Fun ^{\mathrm {LL}}(T,D)\). Finally, every \(G\) is an extension \(i_2\pi _2G \to G \to i_1\pi _1G\) of an object in the image of \(j_1\) by one in the image of \(j_2\).

(2) The criterion for strong continuity follows from Lemma 4.24, applied to the functors \(i_1\) and \(i_2\), whose essential images generate \(E\); the converse holds since \(i_1\) and \(i_2\) are strongly continuous. Let \(j_1' := - \circ \pi _1\) and \(j_2' := - \circ \pi _2\). For an adjunction \(L \dashv R\), precomposition with \(R\) is left adjoint to precomposition with \(L\). Hence \(j_1'\) has the left adjoint \(- \circ i_1\), and \(j_2'\) has the right adjoint \(- \circ i_2\). Again \(j_1'\) and \(j_2'\) are fully faithful, as \(\pi _1i_1 \simeq \id \) and \(\pi _2i_2 \simeq \id \). Moreover, \[ \Map (j_2'(H'),j_1'(H)) \simeq \Map (H',H\pi _1i_2) \simeq * \] for \(H \in \Fun ^{\mathrm {LL}}(C,T)\) and \(H' \in \Fun ^{\mathrm {LL}}(D,T)\). Finally, precomposing \(G\) with the cofiber sequence \(i_2\pi _2 \to \id _E \to i_1\pi _1\) gives a cofiber sequence \(j_2'(Gi_2) \to G \to j_1'(Gi_1)\). □

In the talk, the decomposition in (2) was written with the two factors in the opposite order. The order above, which agrees with [Efimov (2024), Proposition 2.78], is forced by the vanishing of maps from the image of \(j_2'\) to the image of \(j_1'\).

Warning 5.27. The converse of Proposition 3.19 fails: if \(K\) and \(E\) are dualizable, then \(D\) need not be dualizable, even if \(K\) and \(E\) are compactly generated. An explicit counterexample is given in Example 5.30 below. However, dualizable categories are closed under pullbacks along strongly continuous localizations, see Proposition 7.21(2).

Let us take a closer look at extensions of compactly generated categories. Consider a strongly continuous localization sequence \[ \Ind (A) \xrightarrow {F} D \xrightarrow {G} \Ind (B) \] in \(\PrL _{\st }\), with \(A,B \in \Cat _{\perf }\). Since \(\hom _D(Fa,G^Rb) \simeq \hom _{\Ind (B)}(GFa,b) \simeq 0\), the category \(D\) admits a semiorthogonal decomposition \[ D = \lra {G^R(\Ind (B)),F(\Ind (A))} \] in the ordinary sense of semiorthogonality and generation. It need not satisfy Definition 2.21: although \(G^R\) preserves colimits, its right adjoint need not do so. Thus Proposition 5.23 does not apply to an arbitrary extension. Since \(G\) is strongly continuous, \(G^R\) admits a right adjoint \(G^{RR}\), and the extension \(D\) is encoded by the accessible functor \[ G^{RR} \circ F\colon \Ind (A) \to \Ind (B). \] An accessible exact functor \(\Phi \colon \Ind (A) \to \Ind (B)\) is in turn determined by the functor \(\Psi \colon B \to \Pro (\Ind (A))\) sending \(b \in B\) to the pro-object corepresenting the accessible functor \(x \mapsto \hom _{\Ind (B)}(b,\Phi (x))\).

Following Hennion [Hennion (2017)], we denote by \(\mathrm {Tate}^{\mathrm {el}}(A) \subseteq \Pro (\Ind (A))\) the stable subcategory generated by \(\Ind (A)\) and \(\Pro (A)\) (the elementary Tate objects), and by \(\mathrm {Tate}(A) \subseteq \Pro (\Ind (A))\) its idempotent completion. By [Hennion (2017), Theorem 3], an object of \(\Pro (\Ind (A))\) lies in \(\mathrm {Tate}^{\mathrm {el}}(A)\) if and only if it is an extension \(u \to t \to v\) with \(u \in \Pro (A)\) and \(v \in \Ind (A)\).

Proposition 5.28 (Efimov [Efimov (2024), Proposition 6.3]). In the above situation, let \(\Psi \colon B \to \Pro (\Ind (A))\) be the functor corresponding to \(\Phi = G^{RR} \circ F\). Then the following are equivalent:

(1)

\(D\) is dualizable;

(2)

\(D\) is compactly generated;

(3)

The essential image of \(\Psi \) is contained in \(\mathrm {Tate}(A)\).

Proof.

We first show that \(D\) is compactly generated if and only if for every \(b \in B\) there is a compact object \(d \in D^{\omega }\) with \(G(d) \simeq b \oplus \Sigma b\). If \(D\) is compactly generated, this follows from Theorem 2.17(2) and Remark 2.9. Conversely, suppose such lifts \(d_b\) exist. Since \(F\) is strongly continuous, the objects \(F(a)\), \(a \in A\), are compact, and we claim that together with the \(d_b\) they generate \(D\). Indeed, if \(x \in D\) is right orthogonal to all of them, then \(F^R(x) = 0\), so \(x \simeq G^RG(x)\), and \(0 \simeq \hom _D(d_b, G^RG(x)) \simeq \hom _{\Ind (B)}(b \oplus \Sigma b, G(x))\) for all \(b\), so \(G(x) = 0\) and hence \(x = 0\).

(1) \(\Rightarrow \) (2): If \(D\) is dualizable, then the required lifts exist by [Efimov (2024), Proposition 2.10] (which reduces to the compactly generated case by embedding \(D\) strongly continuously into a compactly generated category). (2) \(\Rightarrow \) (1) is clear.

(2) \(\Leftrightarrow \) (3): Let \(d \in D\), and put \(z := G(d)\) and \(w := F^R(d)\), so that we have a fiber sequence \(F(w) \to d \to G^R(z)\). Applying \(\hom _D(-,F(w'))\) for \(w' \in \Ind (A)\) and using \(\hom _D(G^R(z),F(w')) \simeq \hom _{\Ind (B)}(z,\Phi (w'))\), we obtain a fiber sequence of functors \(\Ind (A) \to \Sp \) \[ \hom _{\Ind (B)}(z,\Phi (-)) \to \hom _D(d,F(-)) \to \hom _{\Ind (A)}(w,-). \] Every filtered colimit in \(D\) is an extension of a filtered colimit in \(F(\Ind (A))\) by one in \(G^R(\Ind (B))\), as \(F^R\) and \(G\) preserve colimits. Since \(\hom _D(d,G^R(-)) \simeq \hom _{\Ind (B)}(z,-)\), it follows that \(d\) is compact if and only if \(z\) is compact and \(\hom _D(d,F(-))\) preserves filtered colimits. Now identify exact accessible functors \(\Ind (A) \to \Sp \) with \(\Pro (\Ind (A))\catop \); under this identification, the functors preserving filtered colimits correspond to \(\Pro (A)\). The first term of the fiber sequence corresponds to \(\Psi (z)\) by definition, and the last one to \(w \in \Ind (A)\). So \(d\) is compact if and only if \(z \in B\) and there is a cofiber sequence \(u \to \Psi (z) \to \Sigma w\) in \(\Pro (\Ind (A))\) with \(u \in \Pro (A)\). Conversely, given \(z \in B\) and a cofiber sequence \(u \to \Psi (z) \to v\) with \(u \in \Pro (A)\) and \(v \in \Ind (A)\), the map \(\Psi (z) \to v\) corresponds to a map \(z \to \Phi (v)\), i.e. to a map \(G^R(z) \to F(v)\), and its fiber \(d\) is a compact object with \(G(d) \simeq z\) by the same computation. Using Hennion’s theorem, we conclude that \(z \in B\) lifts to \(D^{\omega }\) if and only if \(\Psi (z) \in \mathrm {Tate}^{\mathrm {el}}(A)\).

Combining this with the first paragraph, \(D\) is compactly generated if and only if \(\Psi (b) \oplus \Sigma \Psi (b) \in \mathrm {Tate}^{\mathrm {el}}(A)\) for all \(b \in B\). This implies \(\Psi (b) \in \mathrm {Tate}(A)\). Conversely, if \(\Psi (b) \in \mathrm {Tate}(A)\), then \(\Psi (b) \oplus \Sigma \Psi (b) \in \mathrm {Tate}^{\mathrm {el}}(A)\) by Thomason’s classification of dense subcategories, as in Remark 2.9. □

Remark 5.29. Applied to compactly generated \(D\), the proof shows that short exact sequences \(A \to D_0 \to B\) in \(\Cat _{\perf }\) correspond to exact functors \(B \to \mathrm {Tate}(A)\); this classification first appeared in [Calmès et al. (2020), Theorem A.5.1 and Corollary A.5.10], see also [Efimov (2024), Theorem 6.4].

Example 5.30 (Efimov). Let \(k\) be a field, and consider the category \[ D := \Big \{(V,W,\phi ) \;\Big |\; V,W \in \D (k),\ \phi \colon \bigoplus _{\N }V \to \bigoplus _{\N }W\Big \}. \] The functors \(W \mapsto (0,W,0)\) and \((V,W,\phi ) \mapsto V\) exhibit \(D\) as an extension of \(\D (k)\) by \(\D (k)\), both of which are compactly generated. Nevertheless, \(D\) is not dualizable.

Proof (following [Efimov (2024), Proposition 6.8]). Here \(\Phi (W) = \prod _{\N }\bigoplus _{\N }W\), since \(\hom (\bigoplus _{\N }V,\bigoplus _{\N }W) \simeq \hom (V,\prod _{\N }\bigoplus _{\N }W)\). By Proposition 5.28, it suffices to show that \(\Psi (k) \notin \mathrm {Tate}(\Perf (k))\), where \(\Psi (k)\) is the pro-object corepresenting \(W \mapsto \hom (k,\Phi (W)) = \prod _{\N }\bigoplus _{\N }W\). Using (AB6) in \(\D (k)\), we may rewrite this functor as \[ \prod _{\N }\bigoplus _{\N }W \simeq \colim _{f\colon \N \to \N }\ \prod _{n \in \N }\bigoplus _{i=0}^{f(n)-1}W \simeq \colim _{f\colon \N \to \N }\hom (X_f,W), \qquad X_f := \bigoplus _{n \in \N }k^{f(n)}, \] where \(\N ^{\N }\) is ordered pointwise, and for \(f \leq g\) the transition map is induced by the projection \(X_g \to X_f\). Hence \(\Psi (k) \simeq \text {``}\!\lim \nolimits _f\!\text {} X_f\), a pro-system in which all transition maps are split epimorphisms.

Let \(q\colon \D (k) \to \Calk (\Perf (k))\) be the projection to \(\Calk (\Perf (k)) := \bigcup _{\kappa }\Calk _{\kappa }(\Perf (k))\) (the transition functors are fully faithful by Remark 2.27); this is an idempotent complete stable category, and \(q(x) = 0\) if and only if \(x \in \Perf (k)\). Suppose that \(\Psi (k) \in \mathrm {Tate}(\Perf (k))\). Then \(\Psi (k) \oplus \Sigma \Psi (k) \in \mathrm {Tate}^{\mathrm {el}}(\Perf (k))\), so there is a cofiber sequence \(u \to \Psi (k) \oplus \Sigma \Psi (k) \to v\) with \(u \in \Pro (\Perf (k))\) and \(v \in \D (k)\). Applying \(\Pro (q)\) kills \(u\), so \(\Pro (q)(\Psi (k)) = \text {``}\!\lim \nolimits _f\!\text {} q(X_f)\) is a retract of the constant pro-object \(q(v)\), and hence is itself isomorphic to a constant pro-object \(c\).

We now use the following elementary observation: if a pro-system \((Y_f)_f\) indexed by a directed poset, with split epimorphisms as transition maps \(t_{gf}\colon Y_g \to Y_f\), is isomorphic to a constant pro-object \(c\), then there is \(f_0\) such that \(t_{gf}\) is an isomorphism for all \(f_0 \leq f \leq g\). Indeed, the isomorphism is given by maps \(p_f\colon Y_f \to c\) for \(f \geq f_0\) (compatible with the \(t_{gf}\)) and a compatible family \(s_f\colon c \to Y_f\), such that \(p_fs_f = \id _c\) for \(f \geq f_0\), and for every \(f\) there is \(g \geq f,f_0\) with \(s_fp_g = t_{gf}\). If \(\sigma \) is a section of \(t_{gf}\), then \(s_f(p_g\sigma ) = \id \). So for \(f \geq f_0\), the map \(s_f\) is both a split mono and a split epi, hence an isomorphism, and then \(t_{gf} = s_fs_g^{-1}\) is an isomorphism for \(f_0 \leq f \leq g\).

Applied to \(Y_f = q(X_f)\), this gives a contradiction: for every \(f\), the fiber of \(X_{f+1} \to X_f\) is \(\bigoplus _{\N }k \notin \Perf (k)\), so \(q(X_{f+1}) \to q(X_f)\) is not an isomorphism.