Given a functor \(\phi \colon I \to J\), we would like to extend a functor \(F\colon I \to C\) to \(J\). Left and right Kan extensions formulate this as a universal problem, using adjoint objects for the restriction functor \(\phi ^*\colon \Fun (J,C) \to \Fun (I,C)\).

Definition 22.4.1. Let \(I\), \(J\) and \(C\) be \(\infty \)-categories and let \(\phi \colon I \to J\) be a functor.

(1)

Given a functor \(F\colon I \to C\), a left Kan extension of \(F\) along \(\phi \) is a left adjoint object to \(F\) under the restriction functor \(\phi ^*\colon \Fun (J,C) \to \Fun (I,C)\). More explicitly, it is a functor \(\phi _!(F)\colon J \to C\) equipped with a natural transformation \(\eta _F\colon F \to \phi ^*\phi _!(F)\) such that for every other functor \(G\colon J \to C\) the composite \[ \Nat (\phi _!(F),G) \xrightarrow {\phi ^*} \Nat (\phi ^*\phi _!(F), \phi ^*G) \xrightarrow {- \circ \eta _F} \Nat (F,\phi ^*G) \] is an equivalence. Diagrammatically:

Commutative triangle with vertices I, C, J and arrows labeled phi, alpha.
Commutative triangle with vertices I, C, J and arrows labeled phi, \eta F.
(2)

Dually, a right Kan extension of \(F\) along \(\phi \) is a right adjoint object to \(F\) under \(\phi ^*\), i.e. a functor \(\phi _*(F)\colon J \to C\) equipped with a transformation \(\epsilon _F\colon \phi ^*\phi _*(F) \to F\) inducing equivalences \[ \Nat (G,\phi _*(F)) \iso \Nat (\phi ^*(G),F) \] for all \(G\colon J \to C\).

If every functor \(F\colon I \to C\) admits a left Kan extension along \(\phi \), then by Corollary 22.1.5 these left Kan extensions assemble into a left adjoint \[ \phi _!\colon \Fun (I,C) \to \Fun (J,C) \] to the restriction functor \(\phi ^*\colon \Fun (J,C) \to \Fun (I,C)\), called the functor of left Kan extension along \(\phi \). Dually, if every functor \(F\) admits a right Kan extension along \(\phi \) then \(\phi ^*\) admits a right adjoint \[ \phi _*\colon \Fun (I,C) \to \Fun (J,C) \] called right Kan extension along \(\phi \).

Observation 22.4.2. Note that the left Kan extension of a functor \(F\colon I \to C\) along the functor \(p_I \colon I \to *\) is a colimit of \(F\). Dually, a right Kan extension of \(F\) along \(p_I\) is a limit of \(F\).

The Kan extensions used in this book are computed objectwise by colimits or limits over relative slice categories. Such Kan extensions are called pointwise Kan extensions. The following theorem gives the formulas and the hypotheses under which they apply.

Theorem 22.4.3 (Pointwise formula Kan extensions). Let \(F\colon I \to C\) and \(\phi \colon I \to J\) be functors.

(1)

Assume that for every object \(j\) of \(J\) the composite functor \[ I_{/j} \xrightarrow {s} I \xrightarrow {F} C \] admits a colimit in \(C\). Then these colimits assemble into a functor \[ \phi _!(F) \colon J \to C. \] Furthermore, the maps \[ F(i) \simeq \colim _{i' \in I_{/i}} F(i') \to \colim _{i' \in I_{/\phi (i)}} F(i') = \phi _!(F)(\phi (i)) = \phi ^*\phi _!(F)(i) \] assemble into a natural transformation \(\eta _F\colon F \to \phi ^*\phi _!(F)\) exhibiting \(\phi _!(F)\) as a left Kan extension of \(F\) along \(\phi \).

(2)

Dually, if for every \(j\) in \(J\) the composite \[ I_{j/} \xrightarrow {t} I \xrightarrow {F} C \] admits a limit in \(C\), then these limits assemble into a functor \[ \phi _*(F) \colon J \to C \] which is a right Kan extension of \(F\).

In (1), \(I_{/i}\) denotes the ordinary slice in \(I\), while \(I_{/\phi (i)}\) is the relative slice for \(\phi \). The functor \((i',f)\mapsto (i',\phi (f))\) induces the displayed map between colimits. The first isomorphism follows from Corollary 22.3.3, Lemma 22.2.4.

Proof. See Theorem 13.1.8, Theorem 13.1.9 of [Cisinski et al. (2026)]. □

Definition 22.4.4. If the condition in (1) is satisfied, we say that \(F\) admits a pointwise left Kan extension along \(\phi \), and if the condition in (2) is satisfied we say that \(F\) admits a pointwise right Kan extension along \(\phi \).

In general, a Kan extension need not restrict back to the original functor: the unit \(F \to \phi ^*(\phi _!F)\) or counit \(\phi ^*(\phi _*F) \to F\) need not be invertible. However, these maps are invertible for pointwise Kan extensions along a fully faithful functor \(\phi \). For the proof we first record a characterization of full faithfulness in terms of relative slices.

Lemma 22.4.5. Let \(F\colon C \to D\) be a functor. Then the following conditions are equivalent:

(1)

The functor \(F\) is fully faithful;

(2)

For every object \(y\) of \(C\) the commutative square

Commutative square with maps C /y to D /Fy labeled F, C to D labeled F, C /y to C labeled s, D /Fy to D labeled s.

is a pullback square;

(3)

For every object \(x\) of \(C\) the commutative square

Commutative square with maps C x/ to D Fx/ labeled F, C to D labeled F, C x/ to C labeled t, D Fx/ to D labeled t.

is a pullback square.

Proof. We prove the equivalence between (1) and (2); the equivalence with (3) is dual. Suppose that \(F\) is fully faithful. For every \(y\in C\), the induced functor \(C_{/y}\to C\times _D D_{/Fy}\) is essentially surjective: every object \((x,u\colon Fx\to Fy)\) of the target comes from a morphism \(f\colon x\to y\). For \(f\colon x\to y\) and \(g\colon x'\to y\), the hom anima from \(f\) to \(g\) and the hom anima between their images are, respectively, \[ \Hom _C(x,x')\times _{\Hom _C(x,y)}\{f\} \qquad \text {and}\qquad \Hom _C(x,x')\times _{\Hom _D(Fx,Fy)}\{Ff\}, \] by Lemma 22.3.2, Chapterexercise 1.4. The maps from \(\Hom _C(x,x')\) are postcomposition with \(g\) and its composite with \(F\), respectively. Full faithfulness of \(F\) identifies these pullbacks, so the induced functor is fully faithful and hence an equivalence by Axiom J.

Conversely, taking the fibers of the square in (2) over \(x\) and \(Fx\) gives an isomorphism \(\Hom _C(x,y)\iso \Hom _D(Fx,Fy)\) induced by \(F\). Thus \(F\) is fully faithful. □

Lemma 22.4.6. Assume that \(\phi \colon I \hookrightarrow J\) is a fully faithful functor, and let \(F\colon I \to C\) be a functor.

(1)

Assume that \(F\) admits a pointwise left Kan extension \(\phi _!(F)\) along \(\phi \). Then the unit map \(\eta _F\colon F \to \phi ^*\phi _!(F)\) is a natural isomorphism.

(2)

Dually, if \(F\) admits a pointwise right Kan extension \(\phi _*(F)\) along \(\phi \) then the counit \(\epsilon _F\colon \phi ^*\phi _*(F) \to F\) is a natural isomorphism.

Proof. We prove (1); the proof of (2) is dual. By Axiom K we may check that \(\eta _F\) is a pointwise isomorphism. For an object \(i\) of \(I\), the pointwise formula of Theorem 22.4.3 identifies the unit component \(\eta _F(i)\colon F(i)\to (\phi _!F)(\phi (i))\) with the map on colimits induced by the functor \[ I_{/i} \to I_{/\phi (i)}, \quad (i',f\colon i' \to i) \mapsto (i',\phi (f)\colon \phi (i') \to \phi (i)). \] But since \(\phi \) is fully faithful, this functor is an equivalence by Lemma 22.4.5, and it follows that \(\eta _F(i)\) is an isomorphism. □