← All writingMathematics blog

Kan's Ex⁡∞\operatorname{Ex}^\infty functor

Elementary facts and proofs about Ex ∞, and their consequences for the Kan-Quillen model structure on simplicial sets.

Updated

We denote by sSetKQ\sset_{\KQ} the category of simplicial sets equipped with the Kan-Quillen model structure. It has various special properties over the other frequently-used model categories in simplicial homotopy theory and ∞\infty-category theory, and many of these come from the existence and properties of Kan's Ex⁡∞\Ex^\infty functor. This blog post discusses and proves some of these results, often in a more pedestrian manner than some of the standard references.

Remark 1 (conventions). For the purposes of this post, we use the following conventions:

  1. A Kan fibration is a morphism of simplicial sets with the RLP against horn inclusions Λkn↪Δn\Lambda^n_k\mono\D^n for n≥1n\geq1 and 0≤k≤n0\leq k\leq n. A Kan complex is a simplicial set XX such that X→Δ0X\to\D^0 is a Kan fibration.

  2. An anodyne extension is a morphism of simplicial sets with the LLP against all Kan fibrations; equivalently, by the small object argument, it is an element of the smallest saturated class of monomorphisms containing the horn inclusions Λkn↪Δn\Lambda^n_k\mono\D^n for n≥1n\geq1 and 0≤k≤n0\leq k\leq n.

  3. A weak homotopy equivalence is a morphism of simplicial sets X→YX\to Y such that for each Kan complex KK, the induced map

    [Y,K]Δ1→[X,K]Δ1[Y,K]_{\D^1}\to[X,K]_{\D^1}

    between Δ1\D^1-homotopy classes of maps is bijective.

  4. We will assume that in the model category sSetKQ\sset_\KQ:

    • Cofibrations are monomorphisms.
    • All anodyne maps are trivial cofibrations.
    • The weak equivalences are weak homotopy equivalences.
    • The fibrant objects are Kan complexes, and the fibrations between Kan complexes are precisely Kan fibrations.

    But we will not assume that all trivial cofibrations are anodyne - this will be a consequence of our Ex⁡∞\Ex^\infty results. In particular, we won't at the beginning assume that a Kan fibration between arbitrary objects is a fibration of sSetKQ\sset_\KQ.

The subdivision functor Sd⁡\Sd

Defining Sd⁡\Sd and Ex⁡\Ex

We begin by defining a functor which acts as barycentric subdivision of simplicial sets.

  • If PP is a poset, we define its subdivision s(P)s(P) to be the poset of all finite, nonempty, totally ordered subsets of PP, ordered by inclusion.
  • If f ⁣:P→P′f\colon P\to P' is an order-preserving map, there is an induced order-preserving map f∗ ⁣:s(P)→s(P′)f_*\colon s(P)\to s(P') sending U↦f(U)U\mapsto f(U). This defines a functor from posets to posets.
  • For each PP, there is a canonical map s(P)→Ps(P)\to P sending U↦max⁡(U)U\mapsto \max(U), and this defines a natural transformation s→Ids\to \Id because taking maxima commutes with order-preserving maps.

Based on this, there is a functor Δ→sSet\DD\to\sset given by [n]↦N(s([n]))[n]\mapsto N(s([n])), and we define the functor

Sd⁡ ⁣:sSet→sSet\Sd\colon \sset\to\sset

to be its extension by colimits. In particular, it is characterized by being colimit-preserving and satisfying Sd⁡(Δn)=N(s([n]))\Sd(\D^n)= N(s([n])), and it comes equipped with a right adjoint written

Ex⁡ ⁣:sSet→sSet\Ex\colon\sset\to\sset

such that Ex⁡(X)n=Hom⁡sSet(Sd⁡(Δn),X)\Ex(X)_n=\Hom_\sset(\Sd(\D^n),X). The natural maps Sd⁡(Δn)→Δn\Sd(\D^n)\to\D^n in nn provide a map X→Ex⁡(X)X\to\Ex(X) natural in XX, and thus by adjunction a map Sd⁡(A)→A\Sd(A)\to A natural in AA. We observe that the map s[n]→[n]s[n]\to[n] is an epimorphism split by k↦[k]⊂[n]k\mapsto [k]\subset[n], so Sd⁡(Δn)→Δn\Sd(\D^n)\to\D^n is a split epimorphism, and thus X→Ex⁡(X)X\to\Ex(X) is a monomorphism.

Mathematical diagram
Diagram 1. A visual representation of Sd⁡(Δ2)\Sd(\D^2). The natural map Sd⁡(Δ2)→Δ2\Sd(\D^2)\to\D^2 collapses the entire diagram onto the lower left-hand copy of Δ2\D^2, where the colored arrows are mapped onto the same-colored arrow.
Diagram TeX source
\[\begin{tikzcd}[column sep=small,row sep=scriptsize]
	&&& {\{2\}} &&& \\
	&& {\{0,2\}} && {\{1,2\}} \\
	&&& {\{0,1,2\}} \\
	{\{0\}} &&& {\{0,1\}} &&& {\{1\}}
	\arrow[from=1-4, to=2-3]
	\arrow[from=1-4, to=2-5]
	\arrow[from=1-4, to=3-4]
	\arrow[from=2-3, to=3-4]
	\arrow[from=2-5, to=3-4]
	\arrow[color={rgb,255:red,51;green,82;blue,255}, from=4-1, to=2-3]
	\arrow[color={rgb,255:red,51;green,82;blue,255}, from=4-1, to=3-4]
	\arrow[color={rgb,255:red,88;green,227;blue,79}, from=4-1, to=4-4]
	\arrow[color={rgb,255:red,255;green,68;blue,51}, from=4-4, to=3-4]
	\arrow[color={rgb,255:red,255;green,68;blue,51}, from=4-7, to=2-5]
	\arrow[color={rgb,255:red,255;green,68;blue,51}, from=4-7, to=3-4]
	\arrow[from=4-7, to=4-4]
\end{tikzcd}\]

Definition 2 (Sd and Ex). We define Sd⁡0=Ex⁡0=IdsSet\Sd^0=\Ex^0=\Id_\sset, and for each k≥1k\geq1 inductively

Sd⁡k≔Sd⁡∘Sd⁡k−1Ex⁡k≔Ex⁡∘Ex⁡k−1,\Sd^k\coloneqq \Sd\circ\Sd^{k-1}\qquad \Ex^k\coloneqq\Ex\circ\Ex^{k-1},

equipped with sequences of natural maps

⋯→Sd⁡2→Sd⁡1→IdId→Ex⁡→Ex⁡2→⋯ .\cdots\to\Sd^2\to\Sd^1\to\Id\qquad \Id\to\Ex\to\Ex^2\to\cdots.

In particular Sd⁡k⊣Ex⁡k\Sd^k\dashv\Ex^k. Finally, we define the functor

Ex⁡∞≔lim→⁡(Id→Ex⁡→Ex⁡2→⋯),\Ex^\infty\coloneqq\varinjlim\big(\Id\to\Ex\to\Ex^2\to\cdots\big),

equipped with natural monomorphisms Ex⁡k↪Ex⁡∞\Ex^k\mono\Ex^\infty for k≥0k\geq0.

We remark that although Ex⁡∞\Ex^\infty isn't a right adjoint to anything, the fact that Ex⁡k\Ex^k is implies that Ex⁡∞\Ex^\infty preserves finite limits, since filtered colimits in sSet\sset commute with finite limits.

Observation 3. In the poset s([n])s([n]), there is a chain

{0}⊂{0,1}⊂⋯⊂{0,…,n}\{0\}\subset \{0,1\}\subset \cdots\subset \{0,\dots,n\}

of length nn (among others), which is clearly maximal. Therefore Sd⁡(Δn)\Sd(\D^n) is nn-skeletal, and it follows that Ex⁡(X)k\Ex(X)_k depends only on the skeleton Sk⁡k(X)\Sk_k(X), i.e. the inclusion Sk⁡kX→X\Sk_k X\to X induces an isomorphism Sk⁡kEx⁡(Sk⁡kX)→Sk⁡kEx⁡(X)\Sk_k\Ex(\Sk_k X)\to\Sk_k\Ex(X).

We will now ignore the higher Sd⁡\Sd's and Ex⁡\Ex's and focus on each of the k=1k=1 cases individually.

Basic observations about Sd⁡\Sd

Let's begin with some easy observations which follow directly from the definition:

  • Sd⁡(Δ0)=Δ0\Sd(\D^0)=\D^0, meaning that the inclusion X→Ex⁡(X)X\to\Ex(X) is a bijection on objects.

  • Sd⁡(Δ1)=Δ1∪{1}Δ1=Λ22\Sd(\D^1)=\D^1\cup_{\{1\}}\D^1=\Lambda^2_2, so a morphism in Ex⁡(X)\Ex(X) is a cospan in XX. The map Sd⁡(Δ1)→Δ1\Sd(\D^1)\to\D^1 is the map Λ22↪Δ2→s1Δ1\Lambda^2_2\mono\D^2\xrightarrow{s^1}\D^1, so the inclusion X→Ex⁡(X)X\to\Ex(X) sends a morphism f ⁣:x→yf\colon x\to y to x→fy←1yx\xrightarrow{f} y\xleftarrow{1}y.

  • Let n≥1n\geq1. Then the inclusion s([n]∖{i})↪s([n])s([n]\setminus\{i\})\mono s([n]) views a totally ordered subset of [n]∖{i}[n]\setminus\{i\} as one of [n][n]. As subsets of s([n])s([n]) we have

    s([n]∖{i,j})=s([n]∖{i})∩s([n]∖{j}),s\big( [n]\setminus\{i,j\} \big) = s\big( [n]\setminus\{i\} \big) \cap s\big( [n]\setminus\{j\} \big),

    because a subset of [n][n] excludes {i,j}\{i,j\} iff it excludes ii and jj. Taking nerves shows that Sd⁡(Δ[n]∖{i,j})=Sd⁡(Δ[n]∖{i})∩Sd⁡(Δ[n]∖{j})\Sd(\D^{[n]\setminus\{i,j\}})=\Sd(\D^{[n]\setminus\{i\}})\cap\Sd(\D^{[n]\setminus\{j\}}).

Example 4. Apply the observation above to the defining coequalizer

∐i,jΔ[n]∖{i,j}⇉∐iΔ[n]∖i→∂Δn\coprod_{i,j}\D^{[n]\setminus\{i,j\}}\rightrightarrows \coprod_i\D^{[n]\setminus i}\to\partial\D^n

corresponding to the decomposition ∂Δn=⋃iΔ[n]∖{i}\partial\D^n=\bigcup_i\D^{[n]\setminus\{i\}}. Then we deduce that Sd⁡(∂Δn)=⋃iSd⁡(Δ[n]∖{i})\Sd(\partial\D^n)=\bigcup_{i}\Sd(\D^{[n]\setminus\{i\}}) and that the union is taken as subobjects of Sd⁡(Δn)\Sd(\D^n). In particular Sd⁡(∂Δn)→Sd⁡(Δn)\Sd(\partial\D^n)\to\Sd(\D^n) is a monomorphism for all n≥0n\geq0. Since Sd⁡\Sd preserves colimits, the morphisms A→BA\to B of simplicial sets for which Sd⁡(A)→Sd⁡(B)\Sd(A)\to\Sd(B) is a monomorphism form a saturated class, and therefore Sd⁡\Sd preserves all monomorphisms. In particular we have

Sd⁡(Λkn)=⋃i≠kSd⁡(Δ[n]∖{i})\Sd(\Lambda^n_k)=\bigcup_{i\neq k}\Sd(\D^{[n]\setminus\{i\}})

as well inside of Sd⁡(Δn)\Sd(\D^n).

Remark 5. Even after applying Sd⁡\Sd once, the resulting subdivisions of Λkn\Lambda^n_k and ∂Δn\partial\D^n already become nerves of posets. Indeed, write ∂s[n]⊂s[n]\partial s[n]\subset s[n] for the subset s[n]∖{[n]}s[n]\setminus\{[n]\}. Then we claim that

Sd⁡(∂Δn)=⋃i=0nN(s([n]∖{i}))=N(∂s[n])\Sd(\partial\D^n)=\bigcup_{i=0}^n N(s([n]\setminus\{i\}))=N(\partial s[n])

as subcomplexes of Sd⁡(Δn)\Sd(\D^n). Indeed, the ⊂\subset inclusion is clear because s([n]∖{i})⊂∂s[n]s([n]\setminus\{i\})\subset \partial s[n]; conversely, a non-degenerate kk-simplex on the RHS is a chain E0⊊⋯⊊EkE_0\subsetneq\cdots\subsetneq E_k of subsets of [n][n] where [n]≠Ek[n]\neq E_k. Hence Ek⊂[n]∖{i}E_k\subset [n]\setminus\{i\} for some ii, meaning that the chain is contained in N(s([n]∖{i}))N(s([n]\setminus\{i\})). Likewise, define Λks[n]⊂s[n]\Lambda_k s[n]\subset s[n] to consist of the subsets E⊂[n]E\subset [n] such that EE doesn't contain [n]∖{k}[n]\setminus \{k\} (i.e. so that [n]∖E≠∅,{k}[n]\setminus E\neq \varnothing,\{k\}). Then the same argument shows that Sd⁡(Λkn)=N(Λks[n])\Sd(\Lambda^n_k)=N(\Lambda_k s[n]).

We can generalize some of the observations above. For each n≥0n\geq0 and poset PP, the canonical map

Fun⁡([n],P)→Fun⁡(s[n],s(P))≅Hom⁡(Sd⁡Δn,Ns(P))\Fun([n],P)\to\Fun(s[n],s(P))\cong \Hom(\Sd\D^n,Ns(P))

gives a map N(P)→Ex⁡(Ns(P))N(P)\to\Ex(Ns(P)) natural in PP, and hence a map Sd⁡(N(P))→N(s(P))\Sd(N(P))\to N(s(P)) natural in PP.

Proposition 6. For each poset PP, the natural map Sd⁡(N(P))→N(s(P))\Sd(N(P))\to N(s(P)) is an isomorphism of simplicial sets.

Proof. By definition it is an isomorphism when P=[n]P=[n] for each nn. Since every finite nonempty totally ordered set EE is isomorphic to [n][n], it is also an isomorphism if P=EP=E.

For arbitrary PP, note that N(P)=⋃E∈s(P)N(E)N(P)=\bigcup_{E\in s(P)}N(E), because an nn-simplex (a0≤⋯≤an)(a_0\leq \cdots\leq a_n) is contained in the nerve of the chain {a0,…,an}\{a_0,\dots,a_n\}. Applying Sd⁡\Sd to the corresponding coequalizer gives the top row of the diagram:

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{\coprod_{E,E'} \Sd N(E\cap E')} & {\coprod_{E}\Sd N(E)} & {\Sd N(P)} \\
	{\coprod_{E,E'}N(s(E\cap E'))} & {\coprod_E N(sE)} & {N(sP)}
	\arrow[shift left, from=1-1, to=1-2]
	\arrow[shift right, from=1-1, to=1-2]
	\arrow["\cong"', from=1-1, to=2-1]
	\arrow[from=1-2, to=1-3]
	\arrow["\cong"', from=1-2, to=2-2]
	\arrow[from=1-3, to=2-3]
	\arrow[shift left, from=2-1, to=2-2]
	\arrow[shift right, from=2-1, to=2-2]
	\arrow[from=2-2, to=2-3]
\end{tikzcd}

In particular it will complete the proof to show that the bottom row is a coequalizer, or in other words that N(sP)=⋃EN(sE)N(sP)=\bigcup_{E} N(sE) with intersections N(s(E∩E′))N(s(E\cap E')). The former is clear: if E0⊂⋯⊂ErE_0\subset\cdots\subset E_r is an rr-simplex in N(sP)N(sP), then it is also an rr-simplex in N(sEr)N(sE_r).

For the latter, since NN preserves intersections, it suffices to see that s(E∩E′)=sE∩sE′s(E\cap E')=sE\cap sE'. Indeed a nonempty totally ordered subset U⊂PU\subset P is contained in E∩E′E\cap E' if and only if it is contained in EE and in E′E'. □

Key properties of Sd⁡\Sd

Let us prove our first few nontrivial results. We will use the following standard lemma, which comes from basic properties of joins.

Lemma 7. Let A↪BA\mono B be an anodyne extension and U↪VU\mono V be a monomorphism. Then:

  1. The induced map B∗U∪A∗V↪B∗VB * U \cup A * V \mono B * V is an anodyne extension.
  2. The induced map U∗Δ0→V∗Δ0U*\D^0\to V*\D^0 is (right) anodyne.

Proof. For (1), we may by the colimit-preserving properties of joins reduce to showing that Δn∗U∪Λkn∗V→Δn∗V\D^n*U\cup\Lambda^n_k* V\to \D^n*V is anodyne for n≥1n\geq1. Hence the same reasoning allows us to further reduce to the case

Δn∗∂Δm∪Λkn∗Δm→Δn∗Δm\D^n*\partial\D^m\cup\Lambda^n_k*\D^m\to\D^n*\D^m

for n≥1n\geq1 and m≥0m\geq0. This is precisely the inclusion Λkn+m+1↪Δn+m+1\Lambda^{n+m+1}_{k}\mono\D^{n+m+1}.

For (2), we similarly reduce to ∂Δm∗Δ0→Δm∗Δ0\partial\D^m *\D^0\to\D^m *\D^0 for m≥0m\geq0, which is just the inclusion Λm+1m+1↪Δm+1\Lambda^{m+1}_{m+1}\mono\D^{m+1}. □

Example 8. If PP is a poset with a greatest element ee, then any map Δ0→N(P)\D^0\to N(P) is anodyne. Indeed, we may factorize it as

Δ0→Δ0∗{e}→N(P∖{e})∗{e}=N(P).\D^0\to\D^0 *\{e\}\to N(P\setminus \{e\}) * \{e\} = N(P).

The first map is the horn inclusion {0}→Δ1\{0\}\to\D^1 and the second is obtained from Lemma 7(2) applied to the monomorphism Δ0↪N(P∖{e})\D^0\mono N(P\setminus\{e\}). This works as long as Δ0→N(P)\D^0\to N(P) is not the inclusion of ee itself; in that case, the same reasoning works with ∅∗{e}→N(P∖{e})∗{e}=N(P)\varnothing * \{e\}\to N(P\setminus\{e\})*\{e\}= N(P).

Proposition 9. For each simplicial set AA, the canonical map Sd⁡(A)→A\Sd(A)\to A is a weak homotopy equivalence. In particular Sd⁡\Sd preserves weak homotopy equivalences, and

Sd⁡ ⁣:sSetKQ⇄sSetKQ:Ex⁡\Sd\colon \sset_{\KQ}\rightleftarrows \sset_{\KQ} : \Ex

is a Quillen equivalence.

Proof. Because Sd⁡\Sd and IdsSet\Id_\sset are colimit-preserving functors which send monomorphisms to cofibrations, we can use the saturated-by-monomorphisms trick to reduce to showing that Sd⁡(Δn)→Δn\Sd(\D^n)\to\D^n is a weak equivalence for each n≥0n\geq0. Choose any point of the source, which yields a commutative diagram:

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	& {\D^0} & \\
	{\Sd(\D^n)} && {\D^n}
	\arrow[from=1-2, to=2-1]
	\arrow[from=1-2, to=2-3]
	\arrow[from=2-1, to=2-3]
\end{tikzcd}

Since s[n]s[n] and [n][n] have terminal objects ([n][n] and nn respectively), the vertical maps are anodyne extensions by Example 8, yielding a horizontal weak equivalence.

Since Sd⁡(A)→A\Sd(A)\to A is natural in AA, it immediately follows from 2-out-of-3 that Sd⁡\Sd preserves weak equivalences. Since it already preserves monomorphisms, Sd⁡\Sd is left Quillen for the Kan-Quillen model structure. Then the natural map Sd⁡→IdsSet\Sd\to\Id_\sset descends to a morphism of derived functors LSd⁡→IdH\mathbf{L}\Sd\to\Id_{\mathcal{H}}, and the first statement shows that its components are all isomorphisms. In particular LSd⁡\mathbf{L}\Sd is isomorphic to an equivalence of categories. □

At this point we know that Sd⁡\Sd preserves trivial cofibrations in sSetKQ\sset_\KQ, but we don't yet know that these coincide with anodyne extensions. The next theorem proves the anodyne extension case explicitly. We recall the notion of a generalized horn: If S⊂[n]S\subset[n] is a nonempty subset, then we set

ΛSn≔⋃i∉SΔ[n]∖{i}⊂Δn.\Lambda^n_S\coloneqq\bigcup_{i\notin S}\D^{[n]\setminus\{i\}}\subset \D^n.

We will use this notion in the proof of the following result.

Theorem 10. The functor Sd⁡ ⁣:sSet→sSet\Sd\colon\sset\to\sset preserves anodyne extensions.

Proof. Since Sd⁡\Sd preserves colimits, it suffices to show that the inclusion Sd⁡(Λkn)→Sd⁡(Δn)\Sd(\Lambda^n_k)\to\Sd(\D^n) is anodyne for n≥1n\geq1. We will more generally prove by induction on n≥1n\geq1 that Sd⁡(ΛSn)→Sd⁡(Δn)\Sd(\Lambda^n_S)\to\Sd(\D^n) is anodyne for any ∅⊊S⊊[n]\varnothing\subsetneq S\subsetneq[n].

If n=1n=1, the map is either of the endpoint inclusions Δ0→N(s[1])\D^0\to N(s[1]), so this case follows from Example 8. Assuming n≥2n\geq2, our first goal is to find an anodyne map Δ0→Sd⁡ΛSn\D^0\to\Sd\Lambda^n_S. If ∣S∣=n\abs{S}=n, then ΛSn≅Δn−1\Lambda^n_S\cong\D^{n-1}, so we can already do this by Example 8. Otherwise, pick k∉Sk\notin S and consider ΛS∪{k}n⊂ΛSn\Lambda^n_{S\cup\{k\}}\subset\Lambda^n_S. There is a pushout diagram:

Mathematical diagram
Diagram TeX source
\begin{tikzcd}[sep=2.25em]
	{\Lambda^{n-1}_{S'}} & {\Lambda^{[n]\setminus\{k\}}_{S}} & {\Lambda^n_{S\cup\{k\}}} \\
	{\D^{n-1}} & {\D^{[n]\setminus k}} & {\Lambda^n_{S}}
	\arrow[equals, from=1-1, to=1-2]
	\arrow[hook, from=1-1, to=2-1]
	\arrow[hook, from=1-2, to=1-3]
	\arrow[hook, from=1-2, to=2-2]
	\arrow[hook, from=1-3, to=2-3]
	\arrow[equals, from=2-1, to=2-2]
	\arrow[hook, from=2-2, to=2-3]
	\arrow["\lrcorner"{anchor=center, pos=0.125, rotate=180}, draw=none, from=2-3, to=1-2]
\end{tikzcd}

This is because Δ[n]∖{i}∩Δ[n]∖{k}=Δ([n]∖{k})∖{i}\D^{[n]\setminus\{i\}}\cap\D^{[n]\setminus \{k\}}=\D^{([n]\setminus\{k\})\setminus\{i\}}, so S′⊂[n−1]S'\subset[n-1] is the preimage of SS under dk ⁣:[n−1]→[n]d^k\colon[n-1]\to[n]. Now apply Sd⁡\Sd: by induction, Sd⁡ΛS∪{k}n↪Sd⁡ΛSn\Sd\Lambda^n_{S\cup\{k\}}\mono\Sd\Lambda^n_S is anodyne, and we can repeat this process with S∪{k}S\cup\{k\} until we obtain an anodyne extension Sd⁡ΛTn→Sd⁡ΛSn\Sd\Lambda^n_T\to\Sd\Lambda^n_S with ∣T∣=n\abs{T}=n. The case we started with then provides the anodyne extension Δ0→Sd⁡ΛSn\D^0\to\Sd\Lambda^n_S.

Here's how we can finish the inductive step. From Remark 5, we know that

Sd⁡(∂Δn)∗Δ0=N(∂s[n])∗Δ0=N(s[n])=Sd⁡(Δn)\Sd(\partial\D^n)*\D^0=N(\partial s[n])*\D^0=N(s[n])=\Sd(\D^n)

since [n][n] is the terminal object of s([n])s([n]). Let v ⁣:Δ0→Sd⁡(Δn)v\colon\D^0\to\Sd(\D^n) denote this vertex [n][n]; we factor the inclusion Sd⁡(ΛSn)→Sd⁡(Δn)\Sd(\Lambda^n_S)\to\Sd(\D^n) as

Sd⁡ΛSn→Sd⁡(ΛSn)∪Δ0∗{v}↪Sd⁡(ΛSn)∗{v}↪Sd⁡(∂Δn)∗{v}=Sd⁡Δn\Sd\Lambda^n_S\to \Sd(\Lambda^n_S)\cup\D^0 *\{v\}\mono\Sd(\Lambda^n_S)*\{v\}\mono\Sd(\partial\D^n)*\{v\}=\Sd\D^n

noting that the last map is anodyne by Lemma 7. The first map is a pushout of Δ0→Δ0∗{v}\D^0\to\D^0*\{v\} which is a horn inclusion, hence anodyne. Lemma 7(1) implies that the second map is anodyne, applied to Δ0→Sd⁡(ΛSn)\D^0\to\Sd(\Lambda^n_S) and ∅→{v}\varnothing\to\{v\}. □

The extension functor

Here is a simple consequence of our results in the previous section:

Corollary 11. The functors Ex⁡k\Ex^k for 0≤k≤∞0\leq k\leq \infty preserve Kan fibrations and trivial fibrations.

Proof. For all finite kk, it suffices to prove the case that k=1k=1. By adjunction, this is equivalent to the question of whether Sd⁡\Sd preserves monomorphisms and anodyne extensions, which we know is true by Example 4 and Theorem 10.

For k=∞k=\infty, we may therefore show that the classes of Kan fibrations and trivial fibrations are both stable under filtered colimits. This follows from the more general statement that if A↪BA\mono B is an inclusion of finite simplicial sets, then RLP⁡(A↪B)\operatorname{RLP}(A\mono B) is closed under filtered colimits.

Indeed, suppose that {Xi→Yi:i∈I}\{X_i\to Y_i:i\in I\} is a morphism of filtered diagrams such that each Xi→YiX_i\to Y_i has the RLP against A↪BA\mono B. In other words

Hom⁡sSet(B,Xi)→Pi≔Hom⁡sSet(A,Xi)×Hom⁡sSet(A,Yi)Hom⁡(B,Yi)\Hom_\sset(B,X_i)\to P_i\coloneqq\Hom_\sset(A,X_i)\times_{\Hom_\sset(A,Y_i)}\Hom(B,Y_i)
(1)

is surjective for each ii. Finiteness means that Hom⁡sSet(A,−)\Hom_\sset(A,-) and Hom⁡sSet(B,−)\Hom_\sset(B,-) preserve filtered colimits, and since finite limits commute with filtered colimits we deduce that

lim→⁡i∈IPi=Hom⁡sSet(A,X)×Hom⁡sSet(A,Y)Hom⁡(B,Y)\varinjlim_{i\in I} P_i = \Hom_\sset(A,X)\times_{\Hom_\sset(A,Y)}\Hom(B,Y)

where X=lim→⁡iXiX=\varinjlim_i X_i and Y=lim→⁡iYiY=\varinjlim_i Y_i. Taking lim→⁡i\varinjlim_i of Equation (1) gives the result, as surjections are stable under filtered colimits. □

This is the first of many nice properties enjoyed by Ex⁡∞\Ex^\infty. We will discuss the other major ones next. We will need to use a slightly nontrivial result about bisimplicial sets, which we prove now.

The diagonal lemma

Definition 12. A bisimplicial set is a functor

Δop×Δop→Set,\DD^\op\times\DD^\op\to\SetCat,

and we define bisSet\mathbf{bisSet} to be the category of bisimplicial sets. Moreover:

  • If XX is a bisimplicial set, we write X∙,kX_{\bullet,k} and Xk,∙X_{k,\bullet} for the simplicial sets [n]↦Xn,k[n]\mapsto X_{n,k} and [n]↦Xk,n[n]\mapsto X_{k,n}, respectively.
  • If XX is a bisimplicial set, we write diag⁡X\diag{X} for the simplicial set [n]↦Xn,n[n]\mapsto X_{n,n}.
  • If XX and YY are simplicial sets, then we write X⊠YX\boxtimes Y for the bisimplicial set given by ([m],[n])↦Xm×Yn([m],[n])\mapsto X_m\times Y_n.

Observe that in a bisimplicial set, we have face and degeneracy maps si ⁣:X∙,k−1→X∙,ks_i\colon X_{\bullet,k-1}\to X_{\bullet,k} and di ⁣:Xk+1,∙→Xk,∙d_i\colon X_{k+1,\bullet}\to X_{k,\bullet} which are maps of simplicial sets, rather than of sets. Bisimplicial sets come up somewhat frequently in the theory of simplicial sets, often in proofs whose statements involve only ordinary simplicial sets. There are good model structures on bisSet\mathbf{bisSet} which interact well with simplicial sets, and such techniques are one way to prove the following theorem. Here we present an elementary version of the proof.

Theorem 13 (diagonal lemma). Let f ⁣:X→Yf\colon X\to Y be a morphism of bisimplicial sets. If fk ⁣:X∙,k→Y∙,kf_k\colon X_{\bullet,k}\to Y_{\bullet,k} is a weak homotopy equivalence for each k≥0k\geq0, then the induced map diag⁡f ⁣:diag⁡X→diag⁡Y\diag{f}\colon\diag{X}\to\diag{Y} is a weak homotopy equivalence.

Of course, the automorphism flipping the factors in Δ×Δ\DD\times\DD induces an automorphism of bisSet\mathbf{bisSet}, and under this we see that the above lemma works if Xk,∙→Yk,∙X_{k,\bullet}\to Y_{k,\bullet} is a weak equivalence for all k≥0k\geq0.

Remark 14. This proof is the most technical and least geometric part of this post. I recommend skipping to Ex ∞ as a fibrant replacement to see how one applies this result first.

The proof will take a couple steps, each of which is elementary individually; we phrase them in terms of ordinary simplicial sets.

Lemma 15. Let f ⁣:X→Yf\colon X\to Y be a morphism of simplicial sets and II a finite set. We further assume that:

  • We have decompositions X=⋃i∈IXiX=\bigcup_{i\in I}X_i and Y=⋃i∈IYiY=\bigcup_{i\in I}Y_i such that f(Xi)⊂Yif(X_i)\subset Y_i.

  • For each nonempty I′⊂II'\subset I, the induced map

    ⋂i∈I′Xi→f⋂i∈I′Yi\bigcap_{i\in I'}X_i\xrightarrow{f}\bigcap_{i\in I'}Y_i

    is a weak homotopy equivalence.

Then ff is a weak homotopy equivalence.

Proof. We prove this simultaneously for all maps ff by induction on ∣I∣≥1\abs{I}\geq1. If ∣I∣=1\abs{I}=1 there is nothing to prove, so assume ∣I∣>1\abs{I}>1. Choose j∈Ij\in I, and consider the commutative diagram

Mathematical diagram
Diagram TeX source
\begin{tikzcd}[sep=large]
	{\bigcup_{i\neq j}X_i} & {\bigcup_{i\neq j}X_i\cap X_j} & {X_j} \\
	{\bigcup_{i\neq j}Y_i} & {\bigcup_{i\neq j}Y_i\cap Y_j} & {Y_j}
	\arrow[from=1-1, to=2-1]
	\arrow[hook', from=1-2, to=1-1]
	\arrow[hook, from=1-2, to=1-3]
	\arrow[from=1-2, to=2-2]
	\arrow[from=1-3, to=2-3]
	\arrow[hook', from=2-2, to=2-1]
	\arrow[hook, from=2-2, to=2-3]
\end{tikzcd}

The right-hand map is a weak equivalence by assumption, and the left-hand map is a weak equivalence by the inductive hypothesis. Moreover, the inductive hypothesis also applies to the middle map because the finite intersections of {Xi∩Xj:i∈I∖{j}}\{X_i\cap X_j:i\in I\setminus\{j\}\} are all finite intersections of {Xi:i∈I}\{X_i:i\in I\}. Finally, since all horizontal maps are cofibrations, the induced map

X=⋃i∈IXi→Y=⋃i∈IYiX=\bigcup_{i\in I}X_i\to Y=\bigcup_{i\in I}Y_i

between pushouts of the two spans is a weak equivalence. □

The next two results concern degeneracies.

Lemma 16. Let XX be a simplicial set and n≥1n\geq1. For 0≤i1<⋯<ik<n0\leq i_1<\cdots<i_k<n, the following subsets of XnX_n are equal:

  1. The intersection si1(Xn−1)∩⋯∩sik(Xn−1)s_{i_1}(X_{n-1})\cap\cdots \cap s_{i_k}(X_{n-1}).
  2. The image of si1⋯ik,∗ ⁣:Xn−k→Xns^{i_1\cdots i_k,*}\colon X_{n-k}\to X_{n}, where si1⋯ik ⁣:[n]→[n−k]s^{i_1\cdots i_k}\colon [n]\to[n-k] is the unique surjection identifying iji_j and ij+1i_j+1 for j=1,…,kj=1,\dots,k.
  3. The image siksik−1⋯si2si1(Xn−k)⊂Xns_{i_k}s_{i_{k-1}}\cdots s_{i_2} s_{i_1}(X_{n-k})\subset X_n.

Proof. Clearly (2)=(3)(2)=(3) because si1⋯ik=si1∘⋯∘siks^{i_1\cdots i_k}=s^{i_1}\circ\cdots\circ s^{i_k} in Δ\DD. Now (under the Eilenberg-Zilber lemma) identify XnX_n with pairs (α,u)(\alpha,u) of non-degenerate simplices α∈Xm\alpha\in X_m and surjections u ⁣:[n]↠[m]u\colon[n]\epi[m]. Then (α,u)(\alpha,u) lies in sij(Xn−1)s_{i_j}(X_{n-1}) iff u(ij)=u(ij+1)u(i_j)=u({i_j+1}), thus it lies in (1)(1) iff u(ij)=u(ij+1)u(i_j)=u(i_j+1) for j=1,…,kj=1,\dots,k. This is equivalent to uu factoring through si1⋯iks^{i_1\cdots i_k}, which is in turn equivalent to (2)(2). □

Lemma 17. Let AA be a simplicial set. For each n≥1n\geq1, there is a canonical pushout diagram

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{A^\degn_n\times(\D^n)_k\cup A_n\times(\partial\D^n)_k} & {(\Sk_{n-1}A)_k} \\
	{A_n\times(\D^n)_k} & {(\Sk_{n}A)_k}
	\arrow[""{name=0, anchor=center, inner sep=0}, from=1-1, to=1-2]
	\arrow[from=1-1, to=2-1]
	\arrow[from=1-2, to=2-2]
	\arrow[from=2-1, to=2-2]
	\arrow["\lrcorner"{anchor=center, pos=0.125, rotate=180}, draw=none, from=2-2, to=0]
\end{tikzcd}

for all k≥0k\geq0.

Proof. The lower map sends (α,u)↦u∗(α)(\alpha,u)\mapsto u^*(\alpha). We have a diagram

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{A_n^\nd\times(\partial\D^n)_k} & {A^\degn_n\times(\D^n)_k\cup A_n\times(\partial\D^n)_k} & {(\Sk_{n-1}A)_k} \\
	{A^\nd_n\times(\D^n)_k} & {A_n\times(\D^n)_k} & {(\Sk_{n}A)_k}
	\arrow[hook, from=1-1, to=1-2]
	\arrow[hook, from=1-1, to=2-1]
	\arrow[""{name=0, anchor=center, inner sep=0}, from=1-2, to=1-3]
	\arrow[hook, from=1-2, to=2-2]
	\arrow[hook, from=1-3, to=2-3]
	\arrow[hook, from=2-1, to=2-2]
	\arrow[from=2-2, to=2-3]
	\arrow["\lrcorner"{anchor=center, pos=0.125, rotate=180}, draw=none, from=2-3, to=0]
\end{tikzcd}

where the outer rectangle is the evaluation at kk of the skeletal filtration pushout associated to AA. The left-hand square is cartesian because Andeg∩Annd=∅A_n^\degn\cap A^\nd_n=\varnothing, and it is hence cocartesian because the corresponding union is evidently all of An×(Δn)kA_n\times(\D^n)_k. Thus the pasting lemma shows that the right-hand square is cocartesian. □

We can now finally prove the diagonal lemma.

Proof (of Theorem 13). For a bisimplicial set XX, we write LnXL_nX for the simplicial set (LnX)k=(Xk,∙)ndeg(L_nX)_k=(X_{k,\bullet})_n^{\degn}. It is indeed a simplicial set because we have

LnX=⋃i=0n−1im⁡(si ⁣:X∙,n−1→X∙,n)⊂X∙,n.L_nX=\bigcup_{i=0}^{n-1}\operatorname{im}\big(s_i\colon X_{\bullet,n-1}\to X_{\bullet,n}\big)\subset X_{\bullet,n}.

Our first observation is that LnX→LnYL_nX\to L_nY is a weak homotopy equivalence for all n≥0n\geq0. To prove this we are going to apply Lemma 15, which says that it is enough to show that

si1(X∙,n−1)∩⋯∩sik(X∙,n−1)→si1(Y∙,n−1)∩⋯∩sik(Y∙,n−1)s_{i_1}(X_{\bullet,n-1})\cap\cdots\cap s_{i_k}(X_{\bullet,n-1})\to s_{i_1}(Y_{\bullet,n-1})\cap\cdots\cap s_{i_k}(Y_{\bullet,n-1})

is a weak homotopy equivalence for every sequence i1<⋯<ik<ni_1<\cdots<i_k<n. By Lemma 16, this map is isomorphic to the top row of

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{X_{\bullet,n-k}} & {Y_{\bullet,n-k}} \\
	{X_{\bullet,n}} & {Y_{\bullet,n}}
	\arrow[from=1-1, to=1-2]
	\arrow["{s^{i_1\dots i_k,*}}"', from=1-1, to=2-1]
	\arrow["{s^{i_1\dots i_k,*}}", from=1-2, to=2-2]
	\arrow[from=2-1, to=2-2]
\end{tikzcd}

as the vertical maps are induced by split surjections. We have assumed that this is a weak homotopy equivalence.

Next, write Sk⁡n(2)X\Sk^{(2)}_nX for the bisimplicial set with

(Sk⁡n(2)X)k,∙≔Sk⁡n(Xk,∙),(\Sk^{(2)}_nX)_{k,\bullet}\coloneqq \Sk_n(X_{k,\bullet}),

namely it is the nn-skeleton with respect to the second direction. For n≥1n\geq1, there is a skeletal pushout diagram:

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{L_nX\times\D^n\cup X_{\bullet,n}\times\partial\D^n} & {\diag\Sk_{n-1}^{(2)}X} \\
	{X_{\bullet,n}\times\D^n} & {\diag\Sk_{n}^{(2)}X}
	\arrow[""{name=0, anchor=center, inner sep=0}, from=1-1, to=1-2]
	\arrow[hook, from=1-1, to=2-1]
	\arrow[hook, from=1-2, to=2-2]
	\arrow[from=2-1, to=2-2]
	\arrow["\lrcorner"{anchor=center, pos=0.125, rotate=180}, draw=none, from=2-2, to=0]
\end{tikzcd}

Indeed, evaluating at kk provides the diagram

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{(X_{k,\bullet})^{\degn}_n\times(\D^n)_k\cup X_{k,n}\times(\partial\D^n)_k} & {\Sk_{n-1}(X_{k,\bullet})_k} \\
	{X_{k,n}\times(\D^n)_k} & {\Sk_{n}(X_{k,\bullet})_k}
	\arrow[""{name=0, anchor=center, inner sep=0}, from=1-1, to=1-2]
	\arrow[hook, from=1-1, to=2-1]
	\arrow[hook, from=1-2, to=2-2]
	\arrow[from=2-1, to=2-2]
	\arrow["\lrcorner"{anchor=center, pos=0.125, rotate=180}, draw=none, from=2-2, to=0]
\end{tikzcd}

which is exactly the pushout of Lemma 17 applied to the simplicial set Xk,∙X_{k,\bullet}.

To conclude, write PnX≔LnX×Δn∪X∙,n×∂ΔnP_nX\coloneqq L_nX\times\D^n\cup X_{\bullet,n}\times\partial\D^n. The diagram

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{L_nX\times\D^n} & {L_nX\times\partial\D^n} & {X_{\bullet,n}\times\partial\D^n} \\
	{L_nY\times\D^n} & {L_nY\times\partial\D^n} & {Y_{\bullet,n}\times\partial\D^n}
	\arrow[from=1-1, to=2-1]
	\arrow[hook', from=1-2, to=1-1]
	\arrow[hook, from=1-2, to=1-3]
	\arrow[from=1-2, to=2-2]
	\arrow[from=1-3, to=2-3]
	\arrow[hook', from=2-2, to=2-1]
	\arrow[hook, from=2-2, to=2-3]
\end{tikzcd}

has rows all cofibrations and vertical maps weak equivalences (by the first observation). Thus the induced map PnX→PnYP_nX\to P_nY on pushouts is a weak equivalence. We use this to prove by induction that

diag⁡Sk⁡n(2)X→diag⁡Sk⁡n(2)Y\diag\Sk_{n}^{(2)}X \to \diag\Sk_{n}^{(2)}Y

is a weak equivalence for all n≥0n\geq0. When n=0n=0, note that

(diag⁡Sk⁡0(2)X)k=Sk⁡0(Xk,∙)k=Xk,0,(\diag\Sk_{0}^{(2)}X)_k=\Sk_0(X_{k,\bullet})_k=X_{k,0},

so the map is just X∙,0→Y∙,0X_{\bullet,0}\to Y_{\bullet,0} which is a weak equivalence. For n≥1n\geq1, the commutative diagram

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{X_{\bullet,n}\times\D^n} & {P_nX} & {\diag\Sk_{n-1}^{(2)}X} \\
	{Y_{\bullet,n}\times\D^n} & {P_nY} & {\diag\Sk_{n-1}^{(2)}Y}
	\arrow[from=1-1, to=2-1]
	\arrow[hook', from=1-2, to=1-1]
	\arrow[from=1-2, to=1-3]
	\arrow[from=1-2, to=2-2]
	\arrow[from=1-3, to=2-3]
	\arrow[hook', from=2-2, to=2-1]
	\arrow[from=2-2, to=2-3]
\end{tikzcd}

has vertical maps weak equivalences, and the first horizontal map in each row a cofibration. Hence the induced map on pushouts

diag⁡Sk⁡n(2)X→diag⁡Sk⁡n(2)Y\diag\Sk_{n}^{(2)}X \to \diag\Sk_{n}^{(2)}Y

is a weak equivalence, completing the inductive step.

Lastly, we have assembled a morphism of direct systems

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{\diag\Sk_{0}^{(2)}X} & {\diag\Sk_{1}^{(2)}X} & {\diag\Sk_{2}^{(2)}X} & \cdots \\
	{\diag\Sk_{0}^{(2)}Y} & {\diag\Sk_{1}^{(2)}Y} & {\diag\Sk_{2}^{(2)}Y} & \cdots
	\arrow[hook, from=1-1, to=1-2]
	\arrow[from=1-1, to=2-1]
	\arrow[hook, from=1-2, to=1-3]
	\arrow[from=1-2, to=2-2]
	\arrow[hook, from=1-3, to=1-4]
	\arrow[from=1-3, to=2-3]
	\arrow[hook, from=2-1, to=2-2]
	\arrow[hook, from=2-2, to=2-3]
	\arrow[hook, from=2-3, to=2-4]
\end{tikzcd}

in which all vertical maps are weak equivalences and horizontal maps are cofibrations. Thus lim→⁡ndiag⁡Sk⁡n(2)X→lim→⁡ndiag⁡Sk⁡n(2)Y\varinjlim_n\diag\Sk_{n}^{(2)}X\to\varinjlim_n\diag\Sk_{n}^{(2)}Y is a weak equivalence. Since diag⁡\diag preserves monomorphisms and colimits are computed degreewise, clearly both sides are exactly diag⁡X\diag X and diag⁡Y\diag Y, so we're done. □

Ex⁡∞\Ex^\infty as a fibrant replacement

In this section we get to the two main results of this post.

Theorem 18. Let XX be a simplicial set. Then Ex⁡∞(X)\Ex^\infty(X) is a Kan complex.

Proof. Let Λkn→Ex⁡∞(X)\Lambda^n_k\to\Ex^\infty(X) be a horn. Since the source is a finite simplicial set, the map factors through Ex⁡r(X)\Ex^r(X) for some r≥1r\geq1. Thus it suffices to find a commutative diagram of the form:

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{\Lambda^n_k} & {\Ex^r(X)} \\
	{\D^n} & {\Ex^{r+1}(X)}
	\arrow[from=1-1, to=1-2]
	\arrow[hook, from=1-1, to=2-1]
	\arrow[hook, from=1-2, to=2-2]
	\arrow[dashed, from=2-1, to=2-2]
\end{tikzcd}

By adjunction, this is in correspondence with commutative diagrams of the form:

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{\Sd^2\Lambda^n_k} & {\Sd\Lambda^n_k} \\
	{\Sd^2\D^n} & {\Ex^{r-1}(X)}
	\arrow[from=1-1, to=1-2]
	\arrow[hook, from=1-1, to=2-1]
	\arrow[from=1-2, to=2-2]
	\arrow[dashed, from=2-1, to=2-2]
\end{tikzcd}

Therefore it will suffice to show that there is a diagonal map Sd⁡2Δn→Sd⁡Λkn\Sd^2\D^n\to\Sd\Lambda^n_k making the left-hand triangle commute. In other words, under Proposition 6, we may find a commutative triangle:

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{s(\Lambda_ks[n])} & \\
	{s(s[n])} & {\Lambda_ks[n]}
	\arrow[hook, from=1-1, to=2-1]
	\arrow[curve={height=-12pt}, from=1-1, to=2-2]
	\arrow["\psi", from=2-1, to=2-2]
\end{tikzcd}

We can define this explicitly as follows. First define an auxiliary function h ⁣:s[n]→[n]h\colon s[n]\to[n] by h(S)=max⁡Sh(S)=\max{S} if S∈Λks[n]S\in\Lambda_k s[n], and h(S)=kh(S)=k otherwise. Then we define

ψ(U)≔{h(S):S∈U}⊂[n].\psi(U)\coloneqq\{h(S):S\in U\}\subset[n].

The following checks are in order:

  1. First view ψ\psi as a map to s([n])s([n]). Then it is obviously order-preserving (since s(s[n])s(s[n]) is ordered by inclusion).
  2. If we have U⊂Λks[n]U\subset \Lambda_k s[n] (i.e. U∈s(Λks[n])U\in s(\Lambda_k s[n])), then by definition of hh, U↦{max⁡S:S∈U}U\mapsto\{\max{S}:S\in U\}. This says precisely that the triangle above commutes.
  3. It remains to see that ψ(U)∈Λks[n]\psi(U)\in \Lambda_k s[n] for all UU. If U⊂Λks[n]U\subset \Lambda_k s[n] then we already know this from (2); otherwise, we must have k∈ψ(U)k\in\psi(U), so we just have to show that ψ(U)≠[n]\psi(U)\neq[n].
  4. Let SS be the largest element of UU in Λks[n]\Lambda_k s[n] (if no element lies in Λks[n]\Lambda_k s[n] then ψ(U)={k}∈Λks[n]\psi(U)=\{k\}\in\Lambda_k s[n]). Then every T∈U∩Λks[n]T\in U\cap \Lambda_k s[n] satisfies T⊂ST\subset S, meaning h(T)=max⁡T∈Sh(T)=\max T\in S. Hence ψ(U)⊂S∪{k}\psi(U)\subset S\cup\{k\}, and since SS excludes some j≠kj\neq k, ψ(U)\psi(U) also excludes jj.

We have found the requisite diagram, proving the horn-lifting property for Ex⁡∞(X)\Ex^\infty(X). □

In the next result, we will use some basic properties of Δ1\D^1-homotopies and Δ1\D^1-homotopy equivalences (which we recall is stronger than being a weak homotopy equivalence). Most importantly:

  1. Let h ⁣:Δ1×X→Yh\colon\D^1\times X\to Y be a Δ1\D^1-homotopy between h0h_0 and h1h_1. Then for all simplicial sets KK we have an induced map

    Hom‾⁡(Y,K)→Hom‾⁡(Δ1×X,K)≅Hom‾⁡(Δ1,Hom‾⁡(X,K)),\iHom(Y,K)\to\iHom(\D^1\times X,K)\cong\iHom(\D^1,\iHom(X,K)),

    which corresponds (by adjunction) to a homotopy of the form

    Δ1×Hom‾⁡(Y,K)→Hom‾⁡(X,K)\D^1\times\iHom(Y,K)\to\iHom(X,K)

    between h0∗h_0^* and h1∗h_1^*. In particular, if a map f ⁣:X→Yf\colon X\to Y is a Δ1\D^1-homotopy equivalence, then so too is Hom‾⁡(Y,K)→Hom‾⁡(X,K)\iHom(Y,K)\to\iHom(X,K) for all KK.

  2. Let CC be a small category with a terminal object e∈Ce\in C. For each object c∈Cc\in C, the terminal map c→ec\to e commutes with all other maps, meaning that it gives a natural transformation Id→e\Id\to e (where e ⁣:C→Ce\colon C\to C is the constant functor). In other words this is a homotopy Δ1×N(C)→N(C)\D^1\times N(C)\to N(C), and it exhibits N(C)→Δ0N(C)\to\D^0 as a Δ1\D^1-homotopy equivalence with inverse e ⁣:Δ0→N(C)e\colon\D^0\to N(C). 1

Theorem 19. Let XX be a simplicial set. For 0≤k≤∞0\leq k\leq\infty, the canonical map

X→Ex⁡k(X)X\to\Ex^k(X)

is a weak homotopy equivalence. In particular, X→Ex⁡∞(X)X\to\Ex^\infty(X) is a functorial fibrant replacement in sSetKQ\sset_\KQ.

Proof. It suffices to prove the k=1k=1 case; indeed this will imply that X→Ex⁡(X)X\to\Ex(X) is a Kan-Quillen trivial cofibration, and so the composite

X→Ex⁡(X)→Ex⁡2(X)→⋯→Ex⁡k(X)X\to\Ex(X)\to\Ex^2(X)\to\cdots\to\Ex^k(X)

(where possibly k=∞k=\infty) is a transfinite composition of trivial cofibrations, hence a trivial cofibration.

We define two bisimplicial sets as follows:

E(X)m,n=Hom⁡(Δm×Sd⁡Δn,X)D(X)m,n=Hom⁡(Δm×Δn,X)E(X)_{m,n}=\Hom(\D^m\times\Sd\D^n,X)\qquad D(X)_{m,n}=\Hom(\D^m\times\D^n,X)

By definition, they fit into a canonical commutative diagram

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{\D^0\boxtimes X} & {D(X)} \\
	{\D^0\boxtimes \Ex{X}} & {E(X)}
	\arrow[from=1-1, to=1-2]
	\arrow[from=1-1, to=2-1]
	\arrow[from=1-2, to=2-2]
	\arrow[from=2-1, to=2-2]
\end{tikzcd}

of bisimplicial sets, where on (m,n)(m,n) the top map is Hom⁡(Δn,X)→Hom⁡(Δm×Δn,X)\Hom(\D^n,X)\to\Hom(\D^m\times\D^n,X) induced by projection, and the other map is

Hom⁡(Δn,Ex⁡X)=Hom⁡(Sd⁡Δn,X)→Hom⁡(Δm×Sd⁡Δn,X)\Hom(\D^n,\Ex{X})=\Hom(\Sd\D^n,X)\to\Hom(\D^m\times\Sd\D^n,X)

also induced by projection. If we apply diag⁡(−)\diag(-) to the square, the left-hand map becomes X→Ex⁡(X)X\to\Ex(X), so we want to show that the other three maps become weak equivalences. The trick will be to apply the diagonal lemma (Theorem 13) to each of these three arrows, but changing the parity for one.

  1. Top arrow: Note that D(X)m,∙=Hom‾⁡(Δm,X)D(X)_{m,\bullet}=\iHom(\D^m,X), so we use that

    X→Hom‾⁡(Δm,X)X\to\iHom(\D^m,X)

    is a weak homotopy equivalence because Δm→Δ0\D^m\to\D^0 is a Δ1\D^1-homotopy equivalence, as explained above. Then the diagonal lemma ensures that the induced diagonal map is a weak equivalence.

  2. Bottom arrow: Now E(X)m,∙=Ex⁡(Hom‾⁡(Δm,X))E(X)_{m,\bullet}=\Ex(\iHom(\D^m,X)), and the map is

    Ex⁡(X)→Ex⁡(Hom‾⁡(Δm,X)),\Ex(X)\to\Ex\big( \iHom(\D^m,X) \big),

    so for this case it suffices to see that Ex⁡\Ex preserves Δ1\D^1-homotopies. Indeed given Δ1×K→L\D^1\times K\to L we have the composite

    Δ1×Ex⁡(K)↪Ex⁡(Δ1)×Ex⁡(K)→Ex⁡(L)\D^1\times\Ex(K)\mono \Ex(\D^1)\times\Ex(K)\to\Ex(L)

    as Ex⁡\Ex preserves products, giving the required homotopy.

  3. Right arrow: This time we consider D(X)∙,n→E(X)∙,nD(X)_{\bullet,n}\to E(X)_{\bullet,n} and apply the opposite of the diagonal lemma. This map is the lower arrow in the triangle:

    Mathematical diagram
    Diagram TeX source
    \begin{tikzcd}
      & X & \\
      {\iHom(\D^n,X)} && {\iHom(\Sd\D^n,X)}
      \arrow[from=1-2, to=2-1]
      \arrow[from=1-2, to=2-3]
      \arrow[from=2-1, to=2-3]
    \end{tikzcd}

    Both legs of the triangle are Hom‾⁡(−,X)\iHom(-,X) applied to the Δ1\D^1-homotopy equivalences Δn→Δ0\D^n\to\D^0 and Sd⁡Δn→Δ0\Sd\D^n\to\D^0 (because [n][n] and s[n]s[n] have terminal objects), hence are Δ1\D^1-homotopy equivalences. In particular the bottom map is a weak equivalence.

As explained, this implies that X→Ex⁡(X)X\to\Ex(X) is a weak homotopy equivalence, completing the proof. □

Consequences for the Kan-Quillen model structure

Here are some of the immediate corollaries of the existence of Ex⁡∞\Ex^\infty as fibrant replacement. We begin with a simple but very useful one.

Example 20. If XX is a simplicial set, recall that we define the set π0(X)\pi_0(X) as the coequalizer of d0,d1 ⁣:X1⇉X0d_0,d_1\colon X_1\rightrightarrows X_0. In other words, it is the quotient of X0X_0 by the equivalence relation generated by the existence of an edge between vertices. For instance, by definition π0Hom‾⁡(A,X)=[A,X]Δ1\pi_0\iHom(A,X)=[A,X]_{\D^1} for any AA and XX, so if XX is a Kan complex then this also equals Hom⁡H(A,X)\Hom_{\mathcal{H}}(A,X). Evidently π0\pi_0 preserves colimits.

A fact which isn't immediate from the definitions is that π0\pi_0 sends arbitrary weak homotopy equivalences to bijections. To see this, first note that for a weak homotopy equivalence X→YX\to Y of Kan complexes, it is clear because

π0(X)=Hom⁡H(Δ0,X)≅Hom⁡H(Δ0,Y)=π0(Y).\pi_0(X)=\Hom_{\mathcal{H}}(\D^0,X)\cong \Hom_{\mathcal{H}}(\D^0,Y)=\pi_0(Y).

So for the general case we may prove that X→Ex⁡∞(X)X\to\Ex^\infty(X) induces a bijection on π0\pi_0. Since π0\pi_0 preserves colimits, it suffices to show that π0(X)→π0(Ex⁡(X))\pi_0(X)\to\pi_0(\Ex(X)) is a bijection, which is obvious: an edge between x0,x1x_0,x_1 in XX provides one in Ex⁡(X)\Ex(X), and an edge in Ex⁡(X)\Ex(X) between x0,x1x_0,x_1 provides a zig-zag x0→y←x1x_0\to y\leftarrow x_1 between x0x_0 and x1x_1 in XX.

Example 21. A fun immediate application of the previous example is that the following conditions are equivalent for a map f ⁣:X→Yf\colon X\to Y:

  1. ff is a Δ1\D^1-homotopy equivalence.
  2. f∗ ⁣:[Y,A]Δ1→[X,A]Δ1f^*\colon[Y,A]_{\D^1}\to[X,A]_{\D^1} is a bijection for all AA.
  3. f∗ ⁣:Hom‾⁡(Y,A)→Hom‾⁡(X,A)f^*\colon\iHom(Y,A)\to\iHom(X,A) is a weak homotopy equivalence for all AA.

Indeed, the equivalence of the first two is clear, for instance by the Yoneda lemma applied to the quotient of sSet\sset by the congruence of Δ1\D^1-homotopy. We already showed that (1) implies that the map of (3) is a Δ1\D^1-homotopy equivalence, hence a weak equivalence. Conversely applying π0\pi_0 and the previous example to (3) yields (2).

Corollary 22. The functors Sd⁡i,Ex⁡k ⁣:sSetKQ→sSetKQ\Sd^i,\Ex^k\colon\sset_\KQ\to\sset_\KQ for 0≤i<∞0\leq i<\infty and 0≤k≤∞0\leq k\leq\infty preserve and reflect homotopy (co)limits.

Proof. This is a completely general fact about a functor F ⁣:M→MF\colon M\to M from a model category to itself, for which there is a natural transformation F→IdF\to\Id (or Id→F\Id\to F) that is a pointwise weak equivalence. (The functors listed satisfy these conclusions by Proposition 9 and Theorem 19).

E.g. let's do the case of Id→F\Id\to F and homotopy limits, as the other cases are identical. Assuming the injective model structure on Fun⁡(I,M)\Fun(I,M) exists, let D ⁣:I→MD\colon I\to M be a diagram and XI→DX_I\to D a cone. Because of our natural transformation, we have a commutative diagram

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	X & {\mathbf{R}\varprojlim D} \\
	FX & {\mathbf{R}\varprojlim FD}
	\arrow[from=1-1, to=1-2]
	\arrow["\simeq", from=1-1, to=2-1]
	\arrow["\simeq", from=1-2, to=2-2]
	\arrow[from=2-1, to=2-2]
\end{tikzcd}

in Ho⁡(M)\operatorname{Ho}(M). The left-hand map is an isomorphism by assumption, and the morphism D→FDD\to FD in Fun⁡(I,M)\Fun(I,M) is a weak equivalence (because they are pointwise in the injective model structure), hence induces an isomorphism in Ho⁡(M)\operatorname{Ho}(M). □

Corollary 23. If Y′→YY'\to Y is a Kan fibration, then a commutative square of simplicial sets of the form

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{X'} & {Y'} \\
	X & Y
	\arrow[from=1-1, to=1-2]
	\arrow[from=1-1, to=2-1]
	\arrow[from=1-2, to=2-2]
	\arrow[from=2-1, to=2-2]
\end{tikzcd}

is homotopy cartesian in sSetKQ\sset_\KQ if and only if the induced map

X′→X×YY′X'\to X\times_Y Y'

is a weak homotopy equivalence.

Proof. By Corollary 22, the diagram is homotopy cartesian iff the same is true after applying Ex⁡∞\Ex^\infty. Moreover since Ex⁡∞\Ex^\infty preserves finite limits, the new induced map

Ex⁡∞(X′)→Ex⁡∞(X)×Ex⁡∞(Y)Ex⁡∞(Y′)\Ex^\infty(X')\to\Ex^\infty(X)\times_{\Ex^\infty(Y)}\Ex^\infty(Y')

is obtained from applying Ex⁡∞\Ex^\infty to the old one; the pointwise weak equivalence Id→Ex⁡∞\Id\to\Ex^\infty also means that Ex⁡∞\Ex^\infty preserves and reflects weak equivalences. Therefore by Theorem 18 and Corollary 11, we may assume that X′,Y′,XX',Y',X, and YY are all Kan complexes. In this case we know that the claim holds in any model category. □

Example 24. The most common use case of this result is the assertion that the pullback of a weak homotopy equivalence by a Kan fibration is a weak homotopy equivalence. Indeed, Corollary 23 shows that the corresponding cartesian square is homotopy cartesian, so the result follows.

Theorem 25. The fibrations in the Kan-Quillen model structure are precisely Kan fibrations, and the trivial cofibrations are precisely the anodyne extensions.

Proof. The two statements are equivalent, and we claim that it is enough to show that a Kan fibration which is a weak equivalence is a trivial fibration. Indeed if A→BA\to B is a trivial cofibration, factor it as A→A′→BA\to A' \to B where A→A′A\to A' is anodyne and A′→BA'\to B is a Kan fibration. Then A′→BA'\to B is a trivial fibration, so A→BA\to B is a retract of A→A′A\to A' by the retract lemma, hence anodyne.

Let p ⁣:X→Yp\colon X\to Y be a Kan fibration which is a weak homotopy equivalence, and suppose that

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{\partial\D^n} & X \\
	{\D^n} & Y
	\arrow["a", from=1-1, to=1-2]
	\arrow[hook, from=1-1, to=2-1]
	\arrow["p", from=1-2, to=2-2]
	\arrow["\sigma", from=2-1, to=2-2]
\end{tikzcd}

is a lifting problem. Let F={σ(0)}×YXF=\{\sigma(0)\}\times_{Y}X be the corresponding fiber, and note that Corollary 23 says that the defining square for FF is homotopy cartesian. In particular F→Δ0F\to\D^0 is both a weak equivalence and a Kan fibration between Kan complexes, hence a trivial fibration. We now move the lifting problem into the contractible fiber.

Let h ⁣:Δ1×Δn→Δnh\colon \D^1\times\D^n\to\D^n be the standard nullhomotopy with h0=0h_0=0 and h1=1Δnh_1=1_{\D^n}. First we perform two lifts as in the diagrams below:

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{\{1\}\times\partial\D^n} && X & {\partial\D^n} & F \\
	{\D^1\times\partial\D^n} && Y & {\D^n} & {\D^0}
	\arrow["a", from=1-1, to=1-3]
	\arrow[hook, from=1-1, to=2-1]
	\arrow["p", from=1-3, to=2-3]
	\arrow["{\wt{h}_0}", from=1-4, to=1-5]
	\arrow[from=1-4, to=2-4]
	\arrow[from=1-5, to=2-5]
	\arrow["{\wt{h}}"{description}, dashed, from=2-1, to=1-3]
	\arrow["{\sigma h\rvert_{\partial\D^n}}"', from=2-1, to=2-3]
	\arrow["c"{description}, dashed, from=2-4, to=1-5]
	\arrow[from=2-4, to=2-5]
\end{tikzcd}

On the left, commutativity is because we chose h1=Idh_1=\Id, and the lift exists because 1×∂Δn→Δ1×∂Δn1\times\partial\D^n\to\D^1\times\partial\D^n is anodyne. In particular ph~0=σh0=σ(0)p\wt{h}_0=\sigma h_0=\sigma(0), so h~0\wt{h}_0 factors through the fiber FF, and the right-hand diagram makes sense. Now we may form:

Mathematical diagram
Diagram TeX source
\begin{tikzcd}
	{\D^1\times\partial\D^n\cup\{0\}\times\D^n} && X \\
	{\D^1\times\D^n} && Y
	\arrow["{\wt{h}\cup c}", from=1-1, to=1-3]
	\arrow[hook, from=1-1, to=2-1]
	\arrow["p", from=1-3, to=2-3]
	\arrow["g"{description}, dashed, from=2-1, to=1-3]
	\arrow["{\sigma h}"', from=2-1, to=2-3]
\end{tikzcd}

Here h~\wt{h} and cc agree on {0}×∂Δn\{0\}\times\partial\D^n by construction, and σh\sigma h restricts to ph~p\wt{h} on Δ1×∂Δn\D^1\times\partial\D^n by construction. Likewise it restricts to pc=σ(0)pc=\sigma(0) on 0×Δn0\times\D^n. Since the left-hand map above is anodyne, a lift gg exists. In particular the map g1g_1 satisfies

g1∣∂Δn=h~1=apg1=σh1=σ,g_1\rvert_{\partial\D^n}=\wt{h}_1=a\qquad p g_1=\sigma h_1=\sigma,

providing a solution to the initial lifting problem. □

References

Footnotes

  1. This also holds for an ∞\infty-category C\scr{C}. Indeed, ee is a terminal object iff C//e→C\scr{C}//e\to\scr{C} is a trivial fibration, so we can use a section to obtain C→C//e↪Hom‾⁡(Δ1,C)\scr{C}\to\scr{C}//e\mono\iHom(\D^1,\scr{C}). This corresponds to a map Δ1×C→C\D^1\times\scr{C}\to\scr{C} starting at the identity and ending at the constant map at ee. ↩︎

Back to the blog →