Homotopy Kan Extensions

• Homotopy Kan extensions are categorical constructions that generalize ordinary Kan extensions by incorporating weak equivalences and model structures to compute derived functors.
• They are computed via homotopy colimits and explicit replacement functors, ensuring that universal properties hold under various localization regimes.
• Applications span sheaf theory, operads, and even cellular automata, demonstrating their ability to coherently extend local data to global structures.

Homotopy Kan extensions are Kan extensions interpreted in settings where weak equivalences, localizations, model structures, or higher-categorical coherence are part of the ambient formalism. In the literature, they appear as derived functors along localization functors, as homotopy left Kan extensions computed by homotopy colimits, as homotopy right Kan extensions along the Yoneda embedding for sheaves on generalized spaces, and as pointwise or 2-categorical Kan extensions whose stability properties are explicitly compared with homotopical closure phenomena (Govzmann et al., 2021, Gonzalez, 2011).

1. Universal property and localization

The starting point is the ordinary 2-categorical universal property. For a 2-category $\mathcal{K}$, a right Kan extension of a 1-cell $x\colon A\to X$ along $f\colon A\to B$ is a 1-cell $\mathrm{ran}_f x\colon B\to X$ equipped with a 2-cell

$\epsilon:\mathrm{ran}_f x\circ f \Rightarrow x$

such that, for every $y\colon B\to X$,

$$\mathcal{K}(A,X)(y f,x)\cong \mathcal{K}(B,X)(y,\mathrm{ran}_f x).$$

This is the formal template that is later transported to homotopical settings (Liberti et al., 23 Apr 2025).

In homotopical algebra, the relevant Kan extension is typically taken along a localization functor. If $L\colon \mathcal{C}\to \mathcal{C}[W^{-1}]$ is a localization at weak equivalences and $F\colon \mathcal{C}\to \mathcal{D}$ is a functor, then the Kan-derived functor is defined by

$L^K F:=\mathrm{Ran}_L F$

for right derived functors, or dually by left Kan extension for left derived functors. Its universal property is expressed by

$x\colon A\to X$0

Within this framework, the paper on comparison theorems distinguishes faint, weak, strong, and strict localizations, and uses those distinctions to compare different notions of derived functor (Govzmann et al., 2021).

A central point is that equivalent homotopy categories need not present the same universality data. The Kan homotopy category is a weak localization, and the Quillen homotopy category is a strict localization. Accordingly, Kan-derived functors coincide with faint-derived functors in the classical Kan homotopy category, while Kan-derived functors coincide with strong-derived functors in the Quillen homotopy category. The same paper states that homotopy Kan extension, as understood in much of the literature, is always a $x\colon A\to X$1-derived functor in this sense, and that its computation and universal property depend on the type of localization being used (Govzmann et al., 2021).

For model categories and suitable functors, explicit computation is available through replacement functors. The comparison paper records the formula

$x\colon A\to X$2

with $x\colon A\to X$3 a cofibrant replacement functor, and emphasizes that uniqueness is up to unique isomorphism for faint or weak localizations, and strict for strong or strict localizations. A recurrent misconception is therefore avoided: homotopy Kan extension is not merely “the ordinary Kan extension after passing to a homotopy category,” but a construction whose precise universal force depends on the localization regime (Govzmann et al., 2021).

2. Model-categorical realizations and homotopy colimits

In model categories, homotopy left Kan extensions are realized by homotopy colimits. The paper on realizable homotopy colimits defines a realizable homotopy colimit

$x\colon A\to X$4

as a relative left adjoint to the constant diagram functor

$x\colon A\to X$5

in the 2-category of relative categories $x\colon A\to X$6 (Gonzalez, 2011).

For a model category $x\colon A\to X$7, the same work proves that the Bousfield–Kan construction of homotopy colimit is the absolute left derived functor of the colimit. The corrected Bousfield–Kan formula is

$x\colon A\to X$8

where $x\colon A\to X$9 is a cofibrant replacement functor, $f\colon A\to B$0 is the nerve of the overcategory of $f\colon A\to B$1, and $f\colon A\to B$2 denotes the tensoring over simplicial sets. The point of the construction is precisely that ordinary colimits do not in general preserve weak equivalences, while the left derived functor does (Gonzalez, 2011).

This same formalism yields homotopy left Kan extensions. Given $f\colon A\to B$3 and a diagram $f\colon A\to B$4, the homotopy left Kan extension is given by

$f\colon A\to B$5

where $f\colon A\to B$6 is the forgetful functor. This formula makes the dependence on comma categories explicit and exhibits homotopy left Kan extension as a pointwise homotopy-colimit construction (Gonzalez, 2011).

Under exact coproduct hypotheses, realizable homotopy colimits satisfying a cofinality property are characterized by a simplicial-replacement formula. If

$f\colon A\to B$7

then

$f\colon A\to B$8

with $f\colon A\to B$9 a simple functor playing the role of geometric realization. This identifies a large class of homotopy left Kan extension procedures with geometric realization after simplicial replacement, thereby relating abstract universal properties to explicit computations (Gonzalez, 2011).

3. Homotopy right Kan extension along the Yoneda embedding

Homotopy right Kan extension also appears as a sheaf-theoretic extension process. Let $\mathrm{ran}_f x\colon B\to X$0 be a category, let $\mathrm{ran}_f x\colon B\to X$1 be its free cocompletion, and let $\mathrm{ran}_f x\colon B\to X$2 denote the Yoneda embedding. For a left proper, cellular or combinatorial simplicial model category $\mathrm{ran}_f x\colon B\to X$3 and a homotopy sheaf $\mathrm{ran}_f x\colon B\to X$4, the homotopy right Kan extension along $\mathrm{ran}_f x\colon B\to X$5 is defined by

$\mathrm{ran}_f x\colon B\to X$6

The same paper gives a model-categorical presentation

$\mathrm{ran}_f x\colon B\to X$7

for fibrant $\mathrm{ran}_f x\colon B\to X$8, with $\mathrm{ran}_f x\colon B\to X$9 denoting the enriched end and $\epsilon:\mathrm{ran}_f x\circ f \Rightarrow x$0 the canonical cofibrant replacement (Bunk, 2020).

The extension is compatible with a transferred Grothendieck pretopology. If $\epsilon:\mathrm{ran}_f x\circ f \Rightarrow x$1 is a site, then the induced pretopology $\epsilon:\mathrm{ran}_f x\circ f \Rightarrow x$2 on $\epsilon:\mathrm{ran}_f x\circ f \Rightarrow x$3 is defined by $\epsilon:\mathrm{ran}_f x\circ f \Rightarrow x$4-local epimorphisms: a morphism $\epsilon:\mathrm{ran}_f x\circ f \Rightarrow x$5 in $\epsilon:\mathrm{ran}_f x\circ f \Rightarrow x$6 is a $\epsilon:\mathrm{ran}_f x\circ f \Rightarrow x$7-local epimorphism if, for every $\epsilon:\mathrm{ran}_f x\circ f \Rightarrow x$8 and $\epsilon:\mathrm{ran}_f x\circ f \Rightarrow x$9, there exists a covering family $y\colon B\to X$0 such that each composite $y\colon B\to X$1 lifts through $y\colon B\to X$2. These local epimorphisms generate a Grothendieck topology on $y\colon B\to X$3 (Bunk, 2020).

The fundamental structural result is a Quillen adjunction

$y\colon B\to X$4

whose right adjoint $y\colon B\to X$5 is, on fibrant objects, naturally isomorphic to the homotopy right Kan extension along the Yoneda embedding. Thus homotopy sheaves on $y\colon B\to X$6 extend to homotopy sheaves on $y\colon B\to X$7 while preserving the sheaf condition (Bunk, 2020).

A further theorem identifies the essential image of this extension. Colimit-preserving sheaves on $y\colon B\to X$8 are precisely the Kan-extended images of sheaves on $y\colon B\to X$9, via an equivalence

$$\mathcal{K}(A,X)(y f,x)\cong \mathcal{K}(B,X)(y,\mathrm{ran}_f x).$$0

between the quasi-category of sheaves on the site and the full subcategory of $$\mathcal{K}(A,X)(y f,x)\cong \mathcal{K}(B,X)(y,\mathrm{ran}_f x).$$1-sheaves on $$\mathcal{K}(A,X)(y f,x)\cong \mathcal{K}(B,X)(y,\mathrm{ran}_f x).$$2 whose inverse image along $$\mathcal{K}(A,X)(y f,x)\cong \mathcal{K}(B,X)(y,\mathrm{ran}_f x).$$3 preserves colimits. The paper applies this to diffeological vector bundles, bundle gerbes with connection, and smooth bordism-type field theories, all of which are presented as cases where descent is preserved under homotopy right Kan extension (Bunk, 2020).

4. Cofinality, change of indexing, and homotopy invariance

The behavior of homotopy Kan extensions under change of indexing is governed by homotopy cofinality. For a functor $$\mathcal{K}(A,X)(y f,x)\cong \mathcal{K}(B,X)(y,\mathrm{ran}_f x).$$4 between small categories, homotopy cofinality means that for every object $$\mathcal{K}(A,X)(y f,x)\cong \mathcal{K}(B,X)(y,\mathrm{ran}_f x).$$5, the nerve of the comma category $$\mathcal{K}(A,X)(y f,x)\cong \mathcal{K}(B,X)(y,\mathrm{ran}_f x).$$6 is contractible. In that case, for every diagram $$\mathcal{K}(A,X)(y f,x)\cong \mathcal{K}(B,X)(y,\mathrm{ran}_f x).$$7, the induced map

$$\mathcal{K}(A,X)(y f,x)\cong \mathcal{K}(B,X)(y,\mathrm{ran}_f x).$$8

is a weak equivalence of pointed simplicial sets (Husainov, 2023).

This invariance result has direct consequences for non-Abelian homology. For a group diagram $$\mathcal{K}(A,X)(y f,x)\cong \mathcal{K}(B,X)(y,\mathrm{ran}_f x).$$9, the paper defines

$L\colon \mathcal{C}\to \mathcal{C}[W^{-1}]$0

and proves that homotopy cofinality yields isomorphisms

$L\colon \mathcal{C}\to \mathcal{C}[W^{-1}]$1

The same work further shows that if $L\colon \mathcal{C}\to \mathcal{C}[W^{-1}]$2 is a virtual discrete cofibration, then left Kan extension preserves non-Abelian homology: $L\colon \mathcal{C}\to \mathcal{C}[W^{-1}]$3 Here

$L\colon \mathcal{C}\to \mathcal{C}[W^{-1}]$4

is the objectwise formula for left Kan extension, and in the group-diagram case an explicit free-product formula is given in terms of final objects of $L\colon \mathcal{C}\to \mathcal{C}[W^{-1}]$5 (Husainov, 2023).

A parallel theory exists for 2-functors. For a small 2-category $L\colon \mathcal{C}\to \mathcal{C}[W^{-1}]$6 and a 2-functor $L\colon \mathcal{C}\to \mathcal{C}[W^{-1}]$7, the homotopy colimit is the simplicial 2-category

$L\colon \mathcal{C}\to \mathcal{C}[W^{-1}]$8

Its classifying space is homotopy equivalent to that of the Grothendieck construction: $L\colon \mathcal{C}\to \mathcal{C}[W^{-1}]$9 The same paper proves 2-categorical analogues of homotopy invariance, Quillen’s Theorems A and B, and the Homotopy Cofinality Theorem, including the statement that if a 2-functor $F\colon \mathcal{C}\to \mathcal{D}$0 is homotopy left cofinal, then pullback of diagrams along $F\colon \mathcal{C}\to \mathcal{D}$1 preserves homotopy colimits up to weak equivalence (Cegarra et al., 2015).

These cofinality results show that homotopy Kan extensions are not only derived versions of ordinary Kan extensions, but also change-of-indexing constructions whose stability depends on the contractibility of comma or slice-type categories. This suggests that the shape of the indexing category is part of the data of the extension, not merely a bookkeeping device.

5. Pointwise and 2-categorical interpretations

In higher categorical settings, homotopy Kan extension phenomena often appear through pointwise constructions and formal exactness conditions. In the 2-category $F\colon \mathcal{C}\to \mathcal{D}$2 of Grothendieck topoi, geometric morphisms, and 2-cells, pointwise Kan extensions are studied באמצעות a proarrow equipment

$F\colon \mathcal{C}\to \mathcal{D}$3

sending a topos to its underlying category and a geometric morphism to its direct image functor. A Kan extension in $F\colon \mathcal{C}\to \mathcal{D}$4 is pointwise when it is preserved by this equipment, and the paper states that the equipment-based notion agrees with the comma-object notion of pointwise Kan extension (Liberti et al., 23 Apr 2025).

A particularly important case concerns fully faithful morphisms. If $F\colon \mathcal{C}\to \mathcal{D}$5 is a geometric embedding and the right Kan extension $F\colon \mathcal{C}\to \mathcal{D}$6 exists and is pointwise, then

$F\colon \mathcal{C}\to \mathcal{D}$7

The paper interprets this as the coincidence of weak Kan injectivity and strict Kan injectivity along fully faithful morphisms. It then defines, for a 1-cell $F\colon \mathcal{C}\to \mathcal{D}$8, weak right Kan injectivity and right Kan injectivity, and forms the corresponding classes $F\colon \mathcal{C}\to \mathcal{D}$9 and $L^K F:=\mathrm{Ran}_L F$0 for a family $L^K F:=\mathrm{Ran}_L F$1 of morphisms. These classes axiomatize topos-theoretic notions and support a formal theory with closure under adjoint or coadjoint retracts, bicategorical limits, and slicing; the paper explicitly describes these closure properties as reminiscent of homotopical closure properties (Liberti et al., 23 Apr 2025).

A complementary conceptual interpretation is that pointwise left Kan extensions are partial colimits. The paper “Kan extensions are partial colimits” formulates this in terms of partial evaluations in the bar construction for the pseudomonads of diagrams and small presheaves on $L^K F:=\mathrm{Ran}_L F$2. For a pseudomonad $L^K F:=\mathrm{Ran}_L F$3 and a $L^K F:=\mathrm{Ran}_L F$4-pseudoalgebra $L^K F:=\mathrm{Ran}_L F$5, a partial evaluation from $L^K F:=\mathrm{Ran}_L F$6 to $L^K F:=\mathrm{Ran}_L F$7 is an object $L^K F:=\mathrm{Ran}_L F$8 such that

$L^K F:=\mathrm{Ran}_L F$9

Its main theorems state that pointwise left Kan extensions correspond precisely to such partial evaluations, first for the diagram pseudomonad along split opfibrations and then for the small-presheaf pseudomonad via image presheaves (Perrone et al., 2021).

The same paper identifies an “image” morphism of pseudomonads from diagrams to small presheaves and uses it to generalize confinal functors and absolute colimits. It also proposes a probabilistic analogy: if coends are like integrals, then Kan extensions are like conditional expectations. This is not a homotopy-theoretic theorem, but it provides a precise categorical interpretation of why pointwise Kan extension calculations often behave like partial aggregation procedures rather than total colimit formation (Perrone et al., 2021).

6. Algebraic and applied contexts

Homotopy Kan extensions also interact with algebraic structure. The paper on algebraic Kan extensions defines an algebraic left Kan extension as a left Kan extension that interacts well with the algebraic structure present in the situation, and states that such extensions appear in various subjects such as the homotopy theory of operads and in the study of conformal field theories. The governing condition is Guitart-exactness, which guarantees the algebraicness of left Kan extending along the relevant morphisms (Weber, 2015).

The setting is 2-monadic. Given a colax or pseudo algebra morphism

$x\colon A\to X$00

and a lax morphism

$x\colon A\to X$01

one asks whether the underlying pointwise left Kan extension of $x\colon A\to X$02 along $x\colon A\to X$03 exists in the base 2-category and lifts uniquely to the algebraic level. The cited paper shows that this lift exists when the target algebra is algebraically cocomplete relative to $x\colon A\to X$04 and $x\colon A\to X$05 is exact. A principal application is to internal algebra classifiers and polynomial monads, where exactness of the induced morphism between classifier categories yields left adjoints to forgetful functors between categories of internal algebras (Weber, 2015).

A distinct but structurally revealing application appears in the theory of cellular automata. For a local rule $x\colon A\to X$06, the paper “Cellular Automata and Kan Extensions” shows that two natural extensions to partial configurations are given by Kan extensions. The coarse transition function satisfies

$x\colon A\to X$07

while the fine transition function satisfies

$x\colon A\to X$08

where $x\colon A\to X$09 is a projection from shifted local configurations, $x\colon A\to X$10 is the inclusion of global configurations into partial configurations, and $x\colon A\to X$11 is the global transition function (Fernandez et al., 2021).

These examples indicate that Kan extension technology is not restricted to abstract homotopy theory. A plausible implication is that homotopy Kan extensions belong to a larger categorical pattern in which local data are extended to global data under explicitly controlled universal properties. In homotopical settings the control is expressed through weak equivalences, cofibrant or fibrant replacement, and cofinality; in algebraic or combinatorial settings it is expressed through exactness, pointwiseness, or order-theoretic extremality.