betaFeedback

WorksMathematics

Enriched nerves, interleaving distances and the multiplicative magnitude-path spectral sequence

Abstract

By regarding the nerves of categories enriched over a monoidal poset as persistence simplicial sets, we investigate an upper bound of the interleaving distance in the homotopy category between the nerves in the sense of Lanary and Scoccola. In particular, it holds that the distance between the nerves for rr-homotopy equivalent generalized metric spaces is less than or equal to the half of rr. Moreover, a product on the cochain complex associated with a persistence simplicial set is introduced and related to the cup product on the magnitude cochain complex in the sense of Hepworth. As a consequence, the product gives the multiplicative structure to the cohomological magnitude-path spectral sequence (MPSS) whose first term is isomorphic to the magnitude cohomology ring. Furthermore, we discuss at which term the MPSS for a connected finite digraph collapses by using the interleaving of the enriched nerve of the digraph.

Introduction

Magnitude homology for graphs is introduced by Hepworth and Willerton [18]. Leinster and Shulman [28] generalize the magnitude homology to a large class of categories enriched over a monoidal category. Subsequently, Otter [32] has given a framework to relate magnitude homology with persistence homology which appears in topological data analysis. The key idea is to introduce a simplicial set indexed by the poset [0,∞][0,\infty] for a given [0,∞]op[0,\infty]^{\mathrm{op}}-category CC, namely a generalized metric space. The persistence simplicial set is called the enriched nerve of CC. The blurred magnitude homology due to Otter, which is a persistence version of magnitude homology, is further studied in [9, 13, 31]. Very recently, Ivanov and Xu [21] have given a lot of fascinating results on ν\nu-Vietoris-Rips simplicial sets by which both the blurred magnitude chain complex and the Vietoris-Rips complex are induced. To compare such persistence objects homotopically, we may utilize the notion of the interleavings up to homotopy which is originally introduced for persistence spaces by Blumberg and Lesnick [6] and is investigated in a more general setting by Lanari and Scoccola [25].

It is worthwhile mentioning that there exists a spectral sequence, that is called the magnitude-path spectral sequence [18, 1, 11] (MPSS for short), converging to the reachability homology of a directed graph (digraph) GG with the shortest path metric; see [16]. The first page of the MPSS is indeed the magnitude homology of GG and the second one contains the path homology of GG in the sense of Grigor’yan, Lin, Muranov and Yau [14] on the horizontal line. This intimate relationship of the two homologies has been clarified by Asao [1]. Moreover, the entire first page is the bigraded path homology due to Hepworth [17] of the given digraph.

The aim of this article is to consider the enriched nerves in the category of persistence simplicial sets endowed with the projective model structure. Furthermore, we discuss a product on the cochain complex associated with the enriched nerve which gives a multiplicative structure to the cohomological MPSS.

More precisely, by regarding enriched nerves as persistence objects, we give an upper bound of the interleaving distance dIHCd_{\mathrm{IHC}} in the homotopy category between such nerves introduced by Lanari and Scoccola [27]. In particular, Theorem 3.9 asserts that the distance dIHCd_{\mathrm{IHC}} between the enriched nerves for 2r2r-homotopy equivalent generalized metric spaces is less than or equal to rr. As a corollary, we see that the half of the diameter of a generalized metric space MM is an upper bound of the distance dIHCd_{\mathrm{IHC}} between the enriched nerves for MM and a point; see Corollary 3.11. As seen in Remark 3.4, the colimit of the enriched nerve of every connected digraph is homotopy equivalent to that of a point. However, the distance dIHCd_{\mathrm{IHC}} indeed detects evolution of the nerves by the persistence parameter; see Corollaries 3.13 and 3.14 as toy computational examples of dIHCd_{\mathrm{IHC}}.

In other words, the distance dIHCd_{\mathrm{IHC}} gives a lower bound for the rr-homotopy equivalence of two given generalized metric spaces XX and YY. Thus, for example, if the bottleneck distance of the persistent homology of the enriched nerves for XX and YY is greater than or equal to ss, then there is no 2r2r-homotopy equivalence between XX and YY for r<sr < s; see Remark 3.12. While the interleaving distance of the homology groups of the enriched nerves of metric spaces is considered in [31 Section 6] and [21 Section 4], our objects in the study are persistence simplicial sets themselves.

In the meddle of this article, we introduce a product on the dual to the chain complex associated with a persistence simplicial set, which is compatible with persistence structure on the dual; see Theorem 4.1. In fact, the product is obtained by restricting the cup product on the dual to the reachability chain complex discussed in [16]. As a consequence, the product induces that on the magnitude cochain complex in the sense of Hepworth [15]; see Proposition 4.5. We observe that the magnitude cohomology algebra is a complete isometry invariant for extended quasi-metric spaces (X,d)(X,d) each of which inf⁡{d(a,b)∣a,b∈X, a≠b}\inf\{d(a,b)\mid a,b\in X,\ a\ne b\} is positive; see [15 Theorem 3.1 (Recovery theorem)]. Moreover, the product in Theorem 4.1 allows us to give a multiplicative structure to the cohomological MPSS converging to the reachability cohomology algebra of a given metric space; see Theorem 4.6.

The MPSS is constructed by a persistence simplicial chain complex whose domain is the poset of non-negative integers. Such persistence objects are incorporated into the notion of extended tame persistence objects; see [8, 22]. In the latter half of this article, we consider at which term the MPSS for a connected finite digraph collapses by using the interleaving distance dIHCetd_{\mathrm{IHC}}^{\mathrm{et}} in the homotopy category of extended tame chain complexes between the enriched nerve complexes of the digraph and a point; see Theorem 5.1. Thus, the distance dIHCetd_{\mathrm{IHC}}^{\mathrm{et}} from the acyclic chain complex to the enriched nerve complex of a digraph GG gives a sufficient condition for GG to be rr-contractible in one of the Brown categories of N\mathbb{N}-metric spaces in the sense of Livernet, Roff and Whitehouse; see [29 Section 6] and also [17 Section 7].

The rest of this manuscript is organized as follows. In Section 2, following Otter [32], we introduce the enriched nerve for a [0,∞]op[0,\infty]^{\mathrm{op}}-category. Section 3 deals with interleaving distances of enriched nerves and proves Theorem 3.9. In Section 4, we address the multiplicative MPSS and prove Theorem 4.6. In § 5, after recalling the model structure of extended tame persistence objects which takes values in a model category, Theorem 5.1 is proved. In Appendix A, we compare the distances dIHCd_{\mathrm{IHC}} and dIHCetd_{\mathrm{IHC}}^{\mathrm{et}} for persistence objects whose domains are the poset of non-negative integers; see Proposition A.1 and Corollary A.2.

The nerve of the category enriched over a monoidal category

In what follows, we may call a functor from a poset V\mathcal{V} to a category a persistence object, for example, a persistence simplicial set for a functor V→sSet\mathcal{V} \to\mathsf{sSet}.

We recall the nerve of an enriched category introduced by Leinster, Shulman [27] and Otter [32]. Let V=(V,⊗,1V)\mathcal{V} = (\mathcal{V}, \otimes, 1_{\mathcal{V}}) be a monoidal category and C\mathcal{C} a small V\mathcal{V}-category; that is, C\mathcal{C} has a set of objects and an object C(x,y)\mathcal{C}(x,y) in V\mathcal{V} is assigned for elements xx and yy in C\mathcal{C} together with a composition C(x,y)⊗C(y,z)→C(x,y)\mathcal{C}(x,y) \otimes\mathcal{C}(y,z) \to\mathcal{C}(x,y) and identity 1V→C(x,x)1_{\mathcal{V}} \to\mathcal{C}(x,x) which satisfy associativity and unitality conditions.1

By definition, a morphism f:C→C′f : \mathcal{C} \to\mathcal{C}' of V\mathcal{V}-categories is given by an object map x↦f(x)∈C′x \mapsto f(x) \in\mathcal{C}' for x∈Cx \in\mathcal{C} together with morphisms fx,y:C(x,y)→C′(f(x),f(y))f_{x,y} : \mathcal{C}(x,y) \to\mathcal{C}'(f(x),f(y)) in V\mathcal{V} for x,y∈Cx,y \in\mathcal{C} such that the diagrams

1V→C(x,x)↓↓fx,xC′(f(x),f(x))=C′(f(x),f(x))\begin{CD} 1_{\mathcal{V}} @>>> \mathcal{C}(x,x) \\ @VVV @VV{f_{x,x}}V \\ \mathcal{C}'(f(x),f(x)) @= \mathcal{C}'(f(x),f(x)) \end{CD}

and

C(y,z)⊗C(x,y)→C(x,z)fy,z⊗fx,y↓↓fx,zC′(f(y),f(z))⊗C′(f(x),f(y))→C′(f(x),f(z))\begin{CD} \mathcal{C}(y,z) \otimes\mathcal{C}(x,y) @>>> \mathcal{C}(x,z) \\ @V{f_{y,z}\otimes f_{x,y}}VV @VV{f_{x,z}}V \\ \mathcal{C}'(f(y),f(z)) \otimes\mathcal{C}'(f(x),f(y)) @>>> \mathcal{C}'(f(x),f(z)) \end{CD}

commute for x,y,z∈Cx,y,z \in\mathcal{C}. We have the category V-Cat\mathcal{V}\text{-}\mathsf{Cat} of V\mathcal{V}-categories.

Assume further that V\mathcal{V} is semicartesian; that is, the unit 1V1_{\mathcal{V}} is the terminal object. Then, a functor N(C):Vop→sSetN(\mathcal{C}) : \mathcal{V}^{\mathrm{op}} \to\mathsf{sSet} which is called the enriched nerve of C\mathcal{C} is defined by

N(C)(ε)n:=∐x0,x1,…,xn∈Ob⁡CV(ε,C(x0,x1)⊗C(x1,x2)⊗⋯⊗C(xn−1,xn)).N(\mathcal{C})(\varepsilon)_n := \coprod_{x_0,x_1,\ldots,x_n \in\operatorname{Ob}\mathcal{C}} \mathcal{V}\left(\varepsilon,\mathcal{C}(x_0,x_1) \otimes\mathcal{C}(x_1,x_2) \otimes\cdots\otimes\mathcal{C}(x_{n-1},x_n)\right).

where the face maps di:N(C)(ε)n→N(C)(ε)n−1d_i : N(\mathcal{C})(\varepsilon)_n \to N(\mathcal{C})(\varepsilon)_{n-1} and the degeneracy maps sj:N(C)(ε)n→N(C)(ε)n+1s_j : N(\mathcal{C})(\varepsilon)_n \to N(\mathcal{C})(\varepsilon)_{n+1} are defined using composition morphisms and identity-assigning morphisms, respectively. Observe that d0d_0 and dnd_n are defined by using the unique map C(x,y)→1V\mathcal{C}(x,y) \to1_{\mathcal{V}}.

In the case where (V,≤,⊗,1V)(\mathcal{V}, \leq, \otimes, 1_{\mathcal{V}}) is a monoidal poset whose opposite category Vop\mathcal{V}^{\mathrm{op}} is semicartesian and C\mathcal{C} is a Vop\mathcal{V}^{\mathrm{op}}-category, we see that

N(C)(ε)n={(x0,x1,…,xn)∣xi∈Ob⁡C, ℓ(x0,x1,…,xn)≤ε in V}.N(\mathcal{C})(\varepsilon)_n = \{(x_0,x_1,\ldots,x_n) \mid x_i \in\operatorname{Ob}\mathcal{C},\ \ell(x_0,x_1,\ldots,x_n) \leq\varepsilon\ \text{in }\mathcal{V}\}.

where ℓ(x0,x1,…,xn)\ell(x_0,x_1,\ldots,x_n) denotes the element C(x0,x1)⊗C(x1,x2)⊗⋯⊗C(xn−1,xn)\mathcal{C}(x_0,x_1) \otimes\mathcal{C}(x_1,x_2) \otimes\cdots\otimes\mathcal{C}(x_{n-1},x_n) in V\mathcal{V}. We call the element ℓ(x0,x1,…,xn)\ell(x_0,x_1,\ldots,x_n) the length of the tuple (x0,…,xn)(x_0,\ldots,x_n). Thus, the nerve construction above gives a functor N:Vop-Cat→sSetVN : \mathcal{V}^{\mathrm{op}}\text{-}\mathsf{Cat} \to\mathsf{sSet}^{\mathcal{V}} from categories of Vop\mathcal{V}^{\mathrm{op}}-categories to the category of persistence simplicial sets. We may write cN(C)\mathrm{c}N(\mathcal{C}) for the simplicial set colim⁡ℓN(C)(ℓ)∙\operatorname{colim}_{\ell}N(\mathcal{C})(\ell)_\bullet obtained by the colimit construction.

Remark 2.1. Let F:Vop→MopF : \mathcal{V}^{\mathrm{op}} \to\mathcal{M}^{\mathrm{op}} be a lax monoidal functor between monoidal category. For a Vop\mathcal{V}^{\mathrm{op}}-category C\mathcal{C}, one has a Mop\mathcal{M}^{\mathrm{op}}-category FCF\mathcal{C} whose composition FC(a,b)⊗FC(b,c)→FC(a,c)F\mathcal{C}(a,b) \otimes F\mathcal{C}(b,c) \to F\mathcal{C}(a,c) is defined by the composite

Fx⊗Fy→αx,yF(x⊗y)→F(x⊗y→z)FzF x \otimes F y \xrightarrow{\alpha_{x,y}} F(x \otimes y) \xrightarrow{F(x\otimes y\to z)} F z

with the structure map αx,y\alpha_{x,y} of FF and the composition x⊗y→zx\otimes y\to z, where x=C(a,b)x=C(a,b), y=C(b,c)y=C(b,c) and z=C(a,c)z=C(a,c). For example, the ceiling function ⌈ ⌉:[0,∞]→N:=N∪{∞}\lceil\ \rceil:[0,\infty]\to\mathbb{N}:=\mathbb{N}\cup\{\infty\} gives rise to a lax monoidal functor [0,∞)op→Nop[0,\infty)^{\mathrm{op}}\to\mathbb{N}^{\mathrm{op}} between the monoidal posets. Observe that ⌈a⌉+⌈b⌉≥⌈a+b⌉\lceil a\rceil+\lceil b\rceil\geq\lceil a+b\rceil.

Example 2.2. i) (An enriched nerve of a metric space) Let (X,d)(X,d) be a generalized metric space (an extended quasi-pseudometric space)†^\dagger; that is, the distance function d:X×X→[0,∞]d:X\times X\to[0,\infty] satisfies the conditions:

  1. d(x,x)=0d(x,x)=0 and

  2. d(x,y)≤d(x,z)+d(z,y)d(x,y)\leq d(x,z)+d(z,y).

Then, the space XX is identified with a [0,∞]op[0,\infty]^{\mathrm{op}}-category defined by X(x,y)=d(x,y)X(x,y)=d(x,y) for x,y∈Xx,y\in X; see [26]. Applying the enriched nerve construction mentioned above, we have a persistence simplicial set N(X):[0,∞]→sSetN(X):[0,\infty]\to\mathsf{sSet} of XX with the form

N(X)(l)n:={(x0,x1,…,xn)∣xi∈X, ∑i=0n−1d(xi,xi+1)≤l}N(X)(l)_n:=\left\{(x_0,x_1,\ldots,x_n)\mid x_i\in X,\ \sum_{i=0}^{n-1}d(x_i,x_{i+1})\leq l\right\}

whose face maps did_i and degeneracy maps sjs_j are defined by

di((x0,x1,…,xn))=(x0,…,xi−1,xi+1,…,xn)d_i((x_0,x_1,\ldots,x_n))=(x_0,\ldots,x_{i-1},x_{i+1},\ldots,x_n)

for i=0,…,ni=0,\ldots,n and

sj((x0,x1,…,xn))=(x0,…,xj,xj,…,xn)s_j((x_0,x_1,\ldots,x_n))=(x_0,\ldots,x_j,x_j,\ldots,x_n)

for j=0,…,nj=0,\ldots,n, respectively. By definition, the length ℓ(x0,…,xn)\ell(x_0,\ldots,x_n) has the form

ℓ((x0,…,xn))=∑i=0n−1d(xi,xi+1).\ell((x_0,\ldots,x_n))=\sum_{i=0}^{n-1}d(x_i,x_{i+1}).

We see that the construction gives rise to a functor N:[0,∞]op-Cat→sSet([0,∞],≤)N:[0,\infty]^{\mathrm{op}}\text{-Cat}\to\mathsf{sSet}^{([0,\infty],\leq)}. Moreover, the nerve yields a persistence chain complex KN(X)KN(X) whose homology is isomorphic to the blurred magnitude homology in the sense of Otter; see [32 Definition 9.3 and Theorem 9.4]. The colimit colim⁡lKN(X)(l)\operatorname{colim}_lKN(X)(l) is nothing but the reachability complex RN(X)RN(X) which induces the magnitude-path spectral sequence; see § 4.3.

ii) Let XX be a compact Riemannian manifold, more general, a compact Riemannian diffeological space defined in [24]. We have an Nop\mathbb{N}^{\mathrm{op}}-category of the type ⌈ ⌉X\lceil\ \rceil X and the a persistence simplicial set N(⌈ ⌉X)N(\lceil\ \rceil X); see Remark 2.1.

Interleavings between enriched nerves

Throughout this section, let V\mathcal{V} be the monoidal poset ([0,∞),≤,+,0)([0,\infty),\leq,+,0). Observe that Vop\mathcal{V}^{\mathrm{op}} is semicartesian. Thus, we have the functor N:Vop-Cat→sSetVN:\mathcal{V}^{\mathrm{op}}\text{-Cat}\to\mathsf{sSet}^{\mathcal{V}}. For a generalized metric space XX, the enriched nerve N(X)N(X) defined in Example 2.2 is a persistence simplicial set. The functor N:[0,∞]op-Cat→sSet([0,∞],≤)N:[0,\infty]^{\mathrm{op}}\text{-Cat}\to\mathsf{sSet}^{([0,\infty],\leq)} described in Example 2.2 i) and the inclusion functor ι:V→[0,∞]\iota:\mathcal{V}\to[0,\infty] give the functor ι∗N:[0,∞]op-Cat→sSetV\iota^*N:[0,\infty]^{\mathrm{op}}\text{-Cat}\to\mathsf{sSet}^{\mathcal{V}} defined by (ι∗N)(C)(l)=N(C)(ι(l))(\iota^*N)(C)(l)=N(C)(\iota(l)). The functor will be also denoted by NN. In what follows, we denote by W\mathcal{W} the monoidal poset V\mathcal{V} or [0,∞][0,\infty].

Some of results in this section are generalized to the case of a more general monoidal poset. While it is crucial to address for example the monoidal poset (Rn,≤,+,0)(\mathbb{R}^n,\leq,+,0) with the product order ≤\leq in the study of multiparameter persistence theory, we do not pursue such generalizations in this manuscript except for the argument in Remark 3.10.

Interleavings and rr-homotopies

In this section, we consider the interleaving distance between persistence simplicial sets N(C)N(\mathcal{C}) associated with Wop\mathcal{W}^{\mathrm{op}}-categories via the functor N:Wop-Cat→sSetVN:\mathcal{W}^{\mathrm{op}}\text{-Cat}\to\mathrm{sSet}^{\mathcal{V}}. We begin by reviewing the interleaving distance introduced in [5] and results in [4] related to the distance, which we use extensively in this article. By regarding the poset (V,≤)(\mathcal{V},\leq) as a category, we have the functor category C(V,≤)\mathcal{C}^{(\mathcal{V},\leq)} for a category C\mathcal{C}. For a real number ε≥0\varepsilon\geq0, define a functor Tε:(V,≤)→(V,≤)T_\varepsilon:(\mathcal{V},\leq)\to(\mathcal{V},\leq) by Tε(a)=a+εT_\varepsilon(a)=a+\varepsilon. Moreover, the ε\varepsilon-shift functor

( )ε:C(V,≤)→C(V,≤)(\ )^\varepsilon:\mathcal{C}^{(\mathcal{V},\leq)}\to\mathcal{C}^{(\mathcal{V},\leq)}

is defined by ( )ε(F)=Fε:=FTε(\ )^\varepsilon(F)=F^\varepsilon:=FT_\varepsilon. We have a natural transformation ηε:id⁡C(V,≤)⇒( )ε\eta^\varepsilon:\operatorname{id}_{\mathcal{C}^{(\mathcal{V},\leq)}}\Rightarrow(\ )^\varepsilon defined by the structure maps ηFε(i):=F(i≤i+ε):F(i)→F(i+ε)=Fε(i)\eta^\varepsilon_F(i):=F(i\leq i+\varepsilon):F(i)\to F(i+\varepsilon)=F^\varepsilon(i).

Definition 3.1. ([5 Definition 4.2], [4 Definition 3.1]) Objects FF and GG in C(V,≤)\mathcal{C}^{(\mathcal{V},\leq)} are ε\varepsilon-interleaved if there exist natural transformations φ:F→GTε\varphi:F\to G T_\varepsilon and ψ:G→FTε\psi:G\to F T_\varepsilon such that the diagram

\begin{CD} F @>{\eta^\varepsilon_F}>> F^\varepsilon@>{\eta^{\varepsilon}_{F^\varepsilon}}>> F^{\varepsilon+\varepsilon}\\ @V{\psi}VV @V{\psi^\varepsilon}VV @VV{\varphi^\varepsilon}V\\ G @>{\eta^\varepsilon_G}>> G^\varepsilon@>{\eta^{\varepsilon}_{G^\varepsilon}}>> G^{\varepsilon+\varepsilon} \tag*{(1)} \end{CD}

is commutative. Such a tuple (F,G,φ,ψ)(F,G,\varphi,\psi) is called an ε\varepsilon-interleaving.

For objects FF and GG in C(V,≤)\mathcal{C}^{(\mathcal{V},\leq)}, the interleaving distance between FF and GG is define by

dI(F,G):=inf⁡({ε≥0∣F,G are ε-interleaved}∪{∞}).d_I(F,G):=\inf(\{\varepsilon\geq0\mid F,G\text{ are }\varepsilon\text{-interleaved}\}\cup\{\infty\}).

We observe that FF and GG are ε+δ\varepsilon+\delta-interleaved for each δ≥0\delta\geq0 if the objects are ε\varepsilon-interleaved; see [4 Lemma 3.4].

In what follows, we use terminology in [12, 19] for model categories. Let M\mathcal{M} be a cofibrantly generated model category and M(V,≤)\mathcal{M}^{(\mathcal{V},\leq)} the model category endowed with the projective model structure whose weak equivalences and fibrations are pointwise weak equivalences and pointwise fibrations of M\mathcal{M}, respectively; see [19 Theorem 11.6.1].

We denote by Top and sSet the cofibrantly generated model categories of topological spaces and simplicial sets described for example in [19 11.1.5 Example], respectively.

We say that objects FF and GG in M(V,≤)\mathcal{M}^{(\mathcal{V},\leq)} are ε\varepsilon-interleaved in the homotopy category if they are ε\varepsilon-interleaved in Ho⁡(M(V,≤))\operatorname{Ho}(\mathcal{M}^{(\mathcal{V},\leq)}). The shift functor ( )ε:M(V,≤)→M(V,≤)(\ )^\varepsilon:\mathcal{M}^{(\mathcal{V},\leq)}\to\mathcal{M}^{(\mathcal{V},\leq)} preserves weak equivalences. Thus, we can consider the commutative diagram (3.1) for FF and GG in Ho⁡(M(V,≤))\operatorname{Ho}(\mathcal{M}^{(\mathcal{V},\leq)}). For objects FF and GG in Ho⁡(M(R,≤))\operatorname{Ho}(\mathcal{M}^{(\mathbb{R},\leq)}), Lanari and Scoccola [25] introduce the interleaving distance in the homotopy category define by

dIHC(F,G):=inf⁡({ε≥0∣F,G are ε-interleaved in the homotopy category}∪{∞}).d_{IHC}(F,G):=\inf(\{\varepsilon\geq0\mid F,G\text{ are }\varepsilon\text{-interleaved in the homotopy category}\}\cup\{\infty\}).

By the same definition as above, we define the interleaving distance in the homotopy category for the objects FF and GG in Ho⁡(M(V,≤))\operatorname{Ho}(\mathcal{M}(\mathcal{V},\leq)).

Remark 3.2. An interleaving in Ho⁡(M(V,≤))\operatorname{Ho}(\mathcal{M}(\mathcal{V},\leq)) mentioned above is considered with a homotopy commutative diagram of the form (3.1) in M(V,≤)\mathcal{M}(\mathcal{V},\leq) using fibrant-cofibrant replacements RQFRQF and RQGRQG of FF and GG, respectively. In fact, by definition, the Hom-set Hom⁡Ho⁡(M(V,≤))(F,G)\operatorname{Hom}_{\operatorname{Ho}(\mathcal{M}(\mathcal{V},\leq))}(F,G) between FF and GG is given by the set of homotopy classes of morphisms from RQFRQF to RQGRQG:

Hom⁡Ho⁡(M(V,≤))(F,G)=Hom⁡M(V,≤)(RQF,RQG)/≃.\operatorname{Hom}_{\operatorname{Ho}(\mathcal{M}(\mathcal{V},\leq))}(F,G)=\operatorname{Hom}_{\mathcal{M}(\mathcal{V},\leq)}(RQF,RQG)/\simeq.

We require that the homotopy which gives the homotopy commutative diagram is a morphism in M(V,≤)\mathcal{M}(\mathcal{V},\leq).

In what follows, let K\mathbb{K} be a commutative ring with unit. For a functor P:V→sSet⁡P:\mathcal{V}\to\operatorname{sSet} to the category of simplicial sets, we have a functor KP:V→Ch⁡K\mathbb{K}P:\mathcal{V}\to\operatorname{Ch}_{\mathbb{K}} to the category of chain complexes defined by (KP(l),dn=∑i=0n(−1)idi)(\mathbb{K}P(l),d_n=\sum_{i=0}^{n}(-1)^id_i) for l≥0l\geq0, where KP(l)n\mathbb{K}P(l)_n is the free K\mathbb{K}-module generated by the set P(l)nP(l)_n and did_i are the face maps in P(l)P(l). Let HnH_n denote the homology functor sSet⁡V→Ab⁡V\operatorname{sSet}^{\mathcal{V}}\to\operatorname{Ab}^{\mathcal{V}} defined by Hn(P):=Hn(ZP,d∗)H_n(P):=H_n(\mathbb{Z}P,d_*), where Ab⁡\operatorname{Ab} is the category of abelian groups. It follows from [4 Proposition 3.6] that

dI(Hn(X∙),Hn(Y∙))≤dIHC(X∙,Y∙)(2)d_I(H_n(X_\bullet),H_n(Y_\bullet))\leq d_{IHC}(X_\bullet,Y_\bullet) \tag*{(2)}

for persistence simplicial sets X∙X_\bullet and Y∙Y_\bullet and for each n≥0n\geq0.

In the rest of this section, we consider an upper bound of the interleaving distance in the homotopy category between enriched nerves.

Definition 3.3. ((cf. [20, §2.2], [21, Definition 5.2], [29, Definition 2.2.2], [16, Definition 2.2]) Let rr be an object in V\mathcal{V} and f,g:C→C′f,g:C\to C' morphisms of Wop\mathcal{W}^{\mathrm{op}}-categories, namely 1-Lipschitz maps in W\mathcal{W}; see the beginning of Section 2. We say there is an elementary rr-homotopy from ff to gg, and write f⇝rgf\rightsquigarrow_r g, if C′(f(x),g(x))≤rC'(f(x),g(x))\leq r in V\mathcal{V} for every x∈Cx\in C.

Moreover, maps ff and gg are rr-homotopic, denoted f∼rgf\sim_r g, if there is a zigzag sequence of elementary rr-homotopies; that is, there exists a tuple f1,…,fkf_1,\ldots,f_k of morphisms of Vop\mathcal{V}^{\mathrm{op}}-categories such that f=f1f=f_1 and fk=gf_k=g and, each ii, either fi⇝rfi+1f_i\rightsquigarrow_r f_{i+1} or fi+1⇝rfif_{i+1}\rightsquigarrow_r f_i. By definition, a morphism f:C→C′f:C\to C' is an rr-homotopy equivalence if there exists a morphism g:C′→Cg:C'\to C such that g∘f∼rid⁡Cg\circ f\sim_r\operatorname{id}_C and f∘g∼rid⁡C′f\circ g\sim_r\operatorname{id}_{C'}.

Remark 3.4. (1) Let f:C→C′f:C\to C' be a morphism of Wop\mathcal{W}^{\mathrm{op}}-categories. Then, the nerve functor NN mentioned in Section 2 gives a morphism N(f)r:N(C)→N(C′)rN(f)_r:N(C)\to N(C')^r defined by the composite ηr(N(C′))∘N(f):N(C)→N(C′)→N(C′)r\eta^r(N(C'))\circ N(f):N(C)\to N(C')\to N(C')^r for an object rr in Vop\mathcal{V}^{\mathrm{op}}; see the sentence before Definition 3.1 for ηr\eta^r.

(2) Morphisms f,g:C→C′f,g:C\to C' with f⇝rgf\rightsquigarrow_r g give rise to the simplicial homotopy {hj:N(C)(l)n→N(C′)r(l)n+1}0≤j≤n\{h_j:N(C)(l)_n\to N(C')^r(l)_{n+1}\}_{0\leq j\leq n} from N(f)rN(f)_r to N(g)rN(g)_r defined by

hj(x0,…,xn)=(g(x0),…,g(xj),f(xj),…,f(xn)).h_j(x_0,\ldots,x_n)=(g(x_0),\ldots,g(x_j),f(x_j),\ldots,f(x_n)).

As a consequence, we see that cN(C):=colim⁡lN(C)(l)cN(C):=\operatorname{colim}_{l}N(C)(l) is homotopic to cN(pt)cN(pt) if there exists r∈Vr\in\mathcal{V} such that C(x,y)≤rC(x,y)\leq r in V\mathcal{V} for xx and yy in CC.

(3) Let f,g:K→Lf,g:K\to L be simplicial maps and {{hj:Kn→Ln+1}0≤j≤n}\{\{h_j:K_n\to L_{n+1}\}_{0\leq j\leq n}\} a simplicial homotopy from ff to gg. Then, we have a simplicial map H:K×Δ[1]→LH:K\times\Delta[1]\to L with H∘ε0=fH\circ\varepsilon_0=f and H∘ε1=gH\circ\varepsilon_1=g. In fact, the homotopy HH is defined by

Hn(xn,cj)={gn(xn)for j=0,fn(xn)for j=n+1,djhj−1for 1≤j≤n.(3)H_n(x_n,c_j)= \begin{cases} g_n(x_n) & \text{for } j=0,\\ f_n(x_n) & \text{for } j=n+1,\\ d_jh_{j-1} & \text{for } 1\leq j\leq n. \end{cases} \tag*{(3)}

Here, we regard Δ[1]n=Δ([n],[1])\Delta[1]_n=\Delta([n],[1]) as the set {c0,…,cn+1}\{c_0,\ldots,c_{n+1}\} with ci−1(0)={0,…,i−1}c_i^{-1}(0)=\{0,\ldots,i-1\}; see, for example, [34, Proposition 10.5.2].

Proposition 3.5. (cf. [21 Proposition 3.3]) With the same notation as in Remark 3.4, for morphisms f,g:C→C′f,g:C\to C' of Wop\mathcal{W}^{\mathrm{op}}-categories with f⇝rgf\rightsquigarrow_r g, the maps ∣N(f)r∣|N(f)_r| and ∣N(g)r∣|N(g)_r| are homotopic in the model category Top⁡(V,≤)\operatorname{Top}(\mathcal{V},\leq) endowed with the projective model structure.

To prove Proposition 3.5, we need the following two lemmas whose proofs are postponed to the end of this section.

Lemma 3.6. The object N(C)(∗)εN(\mathcal{C})(\ast)^\varepsilon is cofibrant in sSet⁡V\operatorname{sSet}^{\mathcal{V}} for each ε≥0\varepsilon\geq0.

Lemma 3.7. Let XX be a cofibrant object in Top⁡(V,≤)\operatorname{Top}(\mathcal{V},\leq). Suppose that H:X×I→YH:X\times I\to Y is a morphism with ε0H=f\varepsilon_0H=f and ε1H=g\varepsilon_1H=g, where II is the constant persistence space defined by I(l)=[0,1]I(l)=[0,1] for l∈Vl\in\mathcal{V} and (εi)(l)(x)=(x,i)(\varepsilon_i)(l)(x)=(x,i) for x∈X(l)x\in X(l). Then, ff is right homotopic to gg and hence f≃gf\simeq g in Top⁡(V,≤)\operatorname{Top}(\mathcal{V},\leq).

Proof of Proposition 3.5. By [19 Theorem 11.6.5] and Lemma 3.6, we see that the persistence space ∣N(C)(∗)ε∣|N(\mathcal{C})(\ast)^\varepsilon| is cofibrant. Observe that ∣N(C)(∗)ε∣=∣N(C)(∗)∙∣ε|N(\mathcal{C})(\ast)^\varepsilon|=|N(\mathcal{C})(\ast)_\bullet|^\varepsilon and ∣N(C)(∗)∙∣ε|N(\mathcal{C})(\ast)_\bullet|^\varepsilon is fibrant; see Remark 3.2.

The maps N(f)N(f) and N(g)N(g) are regarded as maps cN(C)→cN(C′)cN(\mathcal{C})\to cN(\mathcal{C}'). Then, Remark 3.4 allows us to obtain a commutative diagram

cN(C)×Δ[1]→HcN(C′)↑↑N(C)(l)∙×Δ[1]→Hres,lN(C′)(l+r)∙↑↑N(C)(l′)∙×Δ[1]→Hres,l′N(C′)(l′+r)∙(4)\begin{CD} cN(\mathcal{C})\times\Delta[1] @>H>> cN(\mathcal{C}')\\ @AAA @AAA\\ N(\mathcal{C})(l)_\bullet\times\Delta[1] @>H_{\mathrm{res},l}>> N(\mathcal{C}')(l+r)_\bullet\\ @AAA @AAA\\ N(\mathcal{C})(l')_\bullet\times\Delta[1] @>H_{\mathrm{res},l'}>> N(\mathcal{C}')(l'+r)_\bullet \tag*{(4)} \end{CD}

where Hres,kH_{\mathrm{res},k} is the restriction to N(C)(k)∙N(\mathcal{C})(k)_\bullet of HH and the vertical arrows denote the inclusions induced by the structure maps of the persistence simplicial set N(C)(∗)∙N(\mathcal{C})(\ast)_\bullet. Thus, we see that the realizations ∣Hres∣={∣Hres,l∣}l∈V|H_{\mathrm{res}}|=\{|H_{\mathrm{res},l}|\}_{l\in\mathcal{V}} of the homotopies Hres,lH_{\mathrm{res},l} gives rise to a homotopy from ∣N(f)r∣|N(f)_r| to ∣N(g)r∣|N(g)_r| in Top⁡(V,≤)\operatorname{Top}(\mathcal{V},\leq). In fact, the commutativity of the lower square in (4) enables us to deduce that the map ∣Hres∣:∣N(C)(∗)∙∣×I→∣N(C)(∗)∙∣|H_{\mathrm{res}}|:|N(\mathcal{C})(\ast)_\bullet|\times I\to|N(\mathcal{C})(\ast)_\bullet| is in Top⁡(V,≤)\operatorname{Top}(\mathcal{V},\leq). Lemma 3.7 gives the homotopy we require. □\square

Remark 3.8. A cofibration in M(V,≤)\mathcal{M}(\mathcal{V},\leq) endowed with the projective model structure is complicated. Indeed, it is a retract of a transfinite composition; see [19 Theorem 11.6.1] for more details. Thus, in order to prove Proposition 3.5, we use the notion of right homotopy instead of that of left homotopy.

Theorem 3.9. Let f:C→C′f : C \to C' be a 2r2r-homotopy equivalence in Vop\mathcal{V}^{\mathrm{op}}-categories. Then, the persistence simplicial sets N(C)N(C) and N(C′)N(C') in sSet⁡(V≤)\operatorname{sSet}(\mathcal{V}^{\leq}) are rr-interleaved in the homotopy category. As a consequence, one has

dIHC(N(C),N(C′))≤r.d_{\mathrm{IHC}}(N(C),N(C')) \leq r.

Proof. There exists a morphism g:C′→Cg : C' \to C such that f∘g∼2rid⁡f \circ g \sim_{2r} \operatorname{id} and g∘f∼2rid⁡g \circ f \sim_{2r} \operatorname{id}. By applying Proposition 3.5 repeatedly, we see that ∣(N(g)r)r∣∘∣N(f)r∣≃∣ηN(C)r+r∣\lvert(N(g)_r)^r\rvert\circ\lvert N(f)_r\rvert\simeq\lvert\eta^{r+r}_{N(C)}\rvert and ∣(N(f)r)r∣∘∣N(g)r∣≃∣ηN(C′)r+r∣\lvert(N(f)_r)^r\rvert\circ\lvert N(g)_r\rvert\simeq\lvert\eta^{r+r}_{N(C')}\rvert. In fact, it holds that

(N(g)r)r∘N(f)r=N(g∘f)r+randηN(C)r+r=N(id⁡)r+r.(N(g)_r)^r \circ N(f)_r = N(g\circ f)_{r+r} \quad\text{and}\quad\eta^{r+r}_{N(C)}=N(\operatorname{id})_{r+r}.

This yields that dIHC(S∗(∣N(C)∣),S∗(∣N(C′)∣))≤dIHC(∣N(C)∣,∣N(C′)∣)≤rd_{\mathrm{IHC}}(S_*(\lvert N(C)\rvert),S_*(\lvert N(C')\rvert)) \leq d_{\mathrm{IHC}}(\lvert N(C)\rvert,\lvert N(C')\rvert) \leq r, where S∗( )S_*(\ ) denotes the singular simplex functor. Since S∗(∣F∣)≅FS_*(\lvert F\rvert)\cong F in Ho⁡(sSet⁡(V≤))\operatorname{Ho}(\operatorname{sSet}(\mathcal{V}^{\leq})) for a cofibrant persistence simplicial set FF, we have the result; see [19 Theorems 8.5.23 and 11.6.5]. □

Remark 3.10. Let V\mathcal{V} be a monoidal poset. The notion of an rr-homotopy in Definition 3.3 is generalized to that for objects in Vop\mathcal{V}^{\mathrm{op}}-category. Moreover, the proof of Theorem 3.9 allows us to conclude that the assertion of the first half of the theorem holds replacing the term “a 2r2r-homotopy equivalence” with “an (r+r)(r+r)-homotopy equivalence” for W=V\mathcal{W}=\mathcal{V}.

Corollary 3.11. Let x0x_0 be an object in CC. Suppose that the supremum rCr_C of the set {C(x,x0)∣x∈ob⁡C}\{C(x,x_0)\mid x\in\operatorname{ob}C\}1 exists. Then, the objects N(C)∙N(C)_\bullet and N({x0})∙N(\{x_0\})_\bullet in sSet⁡(V≤)\operatorname{sSet}(\mathcal{V}^{\leq}) are rC2\frac{r_C}{2}-interleaved in the homotopy category.

Proof. The inclusion from the trivial category {x0}\{x_0\} to CC is an rCr_C-homotopy equivalence. Then, the result follows from Theorem 3.9. □

Remark 3.12. For a finite generalized metric space AA, the kkth homology group Hk(ZN(A)(l);K)H_k(\mathbb{Z}N(A)(l);K) with coefficients in a field KK is of finite dimension for each kk and ll. Let BAB_A denote the barcode of the persistence module Hk(ZN(A)(∗);K)H_k(\mathbb{Z}N(A)(*);K). Then, by [4 Proposition 3.6 and Theorem 4.16] and the inequality (2), we see that, for a finite generalized metric spaces XX and YY, the bottleneck distance dB(BX,BY)d_B(B_X,B_Y) of the persistent homology gives a lower bound of the interleaving distance in the homotopy category dIHC(N(X),N(Y))d_{\mathrm{IHC}}(N(X),N(Y)).

If the bottleneck distance is greater than or equal to ss, then XX and YY is not 2r2r-homotopy equivalent for any rr less than ss. This follows from Theorem 3.9. We observe that Hk(ZN(X)(∗);Z)H_k(\mathbb{Z}N(X)(*);\mathbb{Z}) is isomorphic to the blurred magnitude homology; see Example 2.2 i).

We refer the reader to [23 Theorem 3.3] for the equivalence of the distance dIHCd_{\mathrm{IHC}} of chain complexes with other interleaving distances up to homotopy; see [6, 25] for interleavings up to homotopy.

Toy examples as corollaries of Theorem 3.9

In this section, we give computational toy examples for dIHCd_{\mathrm{IHC}}.

Corollary 3.13. Let KnK_n be the complete graph endowed with the shortest path distance, where n≥2n\geq2. Then, one has dIHC(N(Kn)∙,N(pt)∙)=12d_{\mathrm{IHC}}(N(K_n)_\bullet,N(pt)_\bullet)=\frac{1}{2}.

Proof. Corollary 3.11 enables us to conclude that dIHC(N(Hm,n)∙,N(x1)∙)≤12d_{\mathrm{IHC}}(N(H_{m,n})_{\bullet},N(x_1)_{\bullet}) \le\frac{1}{2}. For each integer k≥0k \ge0, the structure map N(Kn)(k)∙→N(Kn)(k+ε)∙N(K_n)(k)_{\bullet} \to N(K_n)(k+\varepsilon)_{\bullet} is identity for ε<1\varepsilon< 1. The same condition holds for N(pt)∙N(pt)_{\bullet}. Therefore, if we have dIHC(N(Kn)∙,N(pt)∙)<12d_{\mathrm{IHC}}(N(K_n)_{\bullet},N(pt)_{\bullet}) < \frac{1}{2}, then dIHC(N(Kn)∙,N(pt)∙)=0d_{\mathrm{IHC}}(N(K_n)_{\bullet},N(pt)_{\bullet})=0. Thus, there exists a weak equivalence between N(Kn)∙N(K_n)_{\bullet} and N(pt)∙N(pt)_{\bullet}. In particular, the simplicial sets N(Kn)(0)∙=∐nΔ[0]N(K_n)(0)_{\bullet}=\coprod_{n}\Delta[0] and N(pt)(0)∙=Δ[0]N(pt)(0)_{\bullet}=\Delta[0] are weak homotopy equivalent. This is a contradiction. □

Corollary 3.14. Let (Tm,n,d)(T_{m,n},d) be the finite metric space consisting three points x0x_0, x1x_1 and x2x_2 with d(x0,x1)=md(x_0,x_1)=m, d(x1,x2)=nd(x_1,x_2)=n and d(x0,x1)=m+nd(x_0,x_1)=m+n, where m≤nm\le n. Then, dIHC(N(Tm,n)∙,N(x1)∙)=n2d_{\mathrm{IHC}}(N(T_{m,n})_{\bullet},N(x_1)_{\bullet})=\frac{n}{2}.

Proof. We write TT for the metric space Tm,nT_{m,n}. Corollary 3.11 enables us to deduce that dIHC(N(T)∙,N(x1)∙)≤n2d_{\mathrm{IHC}}(N(T)_{\bullet},N(x_1)_{\bullet})\le\frac{n}{2}. For the case where m=nm=n, the result follows from the same argument as in Corollary 3.13.

Suppose that m<nm<n. Assume further that dIHC(N(T)∙,N(x1)∙)<n2d_{\mathrm{IHC}}(N(T)_{\bullet},N(x_1)_{\bullet})<\frac{n}{2}. It follows from the inequality (3.2) that

dI(H0(N(T)),H0(N(x1)))≤dIHC(N(T)∙,N(x1)∙)<12.d_I(H_0(N(T)),H_0(N(x_1)))\le d_{\mathrm{IHC}}(N(T)_{\bullet},N(x_1)_{\bullet})<\frac{1}{2}.

Thus, there exists a positive real number ε\varepsilon such that H0(N(T))H_0(N(T)) and H0(N(x1))H_0(N(x_1)) are (n2−ε)(\frac{n}{2}-\varepsilon)-interleaved and m2<n2−ε\frac{m}{2}<\frac{n}{2}-\varepsilon. We have a commutative diagram

H0(N(T)(0))→θH0(N(T)(m))→≅⋅⟶H0(N(T)(n))↘φ↗ψH0(N(x1)(0))=H0(N(x1)(n2−ε))⟶H0(N(x1)(m)).\begin{aligned} H_0(N(T)(0))&\xrightarrow{\theta}H_0(N(T)(m))\xrightarrow{\cong} \cdot\longrightarrow H_0(N(T)(n))\\ &\searrow\varphi\qquad\qquad\qquad\quad\nearrow\psi\\ H_0(N(x_1)(0))&=H_0(N(x_1)(\frac{n}{2}-\varepsilon))\longrightarrow H_0(N(x_1)(m)). \end{aligned}

where θ\theta denotes the structure map H0(N(H)(0<m)H_0(N(H)(0<m). We have H0(N(T)(0))≅Z{x0−x1,x1,x2}H_0(N(T)(0))\cong\mathbb{Z}\{x_0-x_1,x_1,x_2\}. Since ZN(T)0(m)\mathbb{Z}N(T)_0(m) is generated by x0,x1,x2x_0,x_1,x_2 and the component ZN(T)1(m)\mathbb{Z}N(T)_1(m) is generated by (x0,x1)(x_0,x_1) and (x1,x0)(x_1,x_0), it follows that H0(N(T)(m))≅Z{x1,x2}H_0(N(T)(m))\cong\mathbb{Z}\{x_1,x_2\} and the map θ\theta is an epimorphism. The diagram above gives a commutative triangle

Z⊕3⟶Z⊕2\mathbb{Z}^{\oplus3}\longrightarrow\mathbb{Z}^{\oplus2}

in which the horizontal arrow is an epimorphism. Then, the map ψ\psi is an epimorphism, which is a contradiction. □

Proofs of Lemmas 3.6 and 3.7

Before proving Lemma 3.6, we recall a result due to Lanari and Scoccola [25] which is a characterization of a cofibrant persistence simplicial set in sSet⁡(L,≤)\operatorname{sSet}(L,\le) endowed with the projective model structure, where (L,≤)(L,\le) is an arbitrary poset.

Definition 3.15. ([25 Definition 4.1]) A LL-filtered simplicial set (X,β)(X,\beta) is a simplicial set XX equipped with the set of functions β={βn:Xn→L}n≥0\beta=\{\beta_n:X_n\to L\}_{n\ge0}, satisfying

  • βn−1(di(σ))≤βn(σ)\beta_{n-1}(d_i(\sigma))\le\beta_n(\sigma) for every n≥1n\ge1, σ∈Xn\sigma\in X_n, and face map di:Xn→Xn−1d_i:X_n\to X_{n-1};

  • βn+1(sj(σ))≤βn(σ)\beta_{n+1}(s_j(\sigma))\le\beta_n(\sigma) for every n≥0n\ge0, σ∈Xn\sigma\in X_n, and degeneracy map sj:Xn→Xn+1s_j:X_n\to X_{n+1}.

Definition 3.16. ([25 Definition 4.2]) Given a LL-filtered simplicial set (X,β)(X,\beta), define a persistence simplicial set (X‾,β‾)∈sSetL(\overline{X},\overline{\beta}) \in\mathrm{sSet}^{L} such that for r∈Lr \in L we have

(X‾,β‾)(r)n={σ∈Xn∣βn(σ)≤r}.(\overline{X},\overline{\beta})(r)_{n}=\{\sigma\in X_{n}\mid\beta_{n}(\sigma)\leq r\}.

with faces and degeneracies given by restricting the ones of XX.

The following is the result of the characterization mentioned above.

Proposition 3.17. ([25 Proposition 4.5]) A persistence simplicial set XX is isomorphic to one associated with a LL-filtered simplicial set if and only if XX is cofibrant in sSetL\mathrm{sSet}^{L}.

Proof of Lemma 3.6. It follows from Proposition 3.17 that N(C)(∗)∙N(\mathcal{C})(*)_{\bullet} is cofibrant in sSetV\mathrm{sSet}^{\mathcal{V}}. In fact, we have a V\mathcal{V}-filtered simplicial set (cN(C),β)(cN(\mathcal{C}),\beta), where βn(x0,…,xn)=ℓ(x0,…,xn)\beta_{n}(x_{0},\ldots,x_{n})=\ell(x_{0},\ldots,x_{n}) with the length ℓ\ell; see the paragraph before Remark 2.1 for the notations. Then, it follows that (cN(C),β)=N(C)(∗)∙(cN(\mathcal{C}),\beta)=N(\mathcal{C})(*)_{\bullet}.

Define functions βn′:Xn:=cN(C)n→V\beta'_{n}:X_{n}:=cN(\mathcal{C})_{n}\to\mathcal{V} by βn′(σ)=max⁡{0,βn(σ)−ε}\beta'_{n}(\sigma)=\max\{0,\beta_{n}(\sigma)-\varepsilon\}. It is readily seen that the set of functions β′={βn′}\beta'=\{\beta'_{n}\} satisfy the two conditions in Definition 3.15. By Proposition 3.17, in order to prove the result, it suffices to show that the third equality in

(X‾,β‾)(r)n=(X‾,β‾)(r+ε)n={σ∈Xn∣βn(σ)≤r+ε}={σ∈Xn∣βn′(σ)≤r}=(X‾,β′)(r)n=:En\begin{aligned} (\overline{X},\overline{\beta})(r)_{n}=(\overline{X},\overline{\beta})(r+\varepsilon)_{n}=\{\sigma\in X_{n}\mid\beta_{n}(\sigma)\leq r+\varepsilon\} \\ &=\{\sigma\in X_{n}\mid\beta'_{n}(\sigma)\leq r\}=(\overline{X},\beta')(r)_{n}=:E_{n} \end{aligned}

holds. The first, second and fourth equalities follow from the definitions.

For σ∈En\sigma\in E_{n}, suppose that βn′(σ)=0\beta'_{n}(\sigma)=0. Then βn(σ)−ε≤0\beta_{n}(\sigma)-\varepsilon\leq0. Since βn(σ)≤ε\beta_{n}(\sigma)\leq\varepsilon, it follows that σ∈(X‾,β‾)(ε)n⊆(X‾,β‾)(r+ε)n\sigma\in(\overline{X},\overline{\beta})(\varepsilon)_{n}\subseteq(\overline{X},\overline{\beta})(r+\varepsilon)_{n}. If βn′(σ)>0\beta'_{n}(\sigma)>0, then βn(σ)−ε=βn′(σ)≤r\beta_{n}(\sigma)-\varepsilon=\beta'_{n}(\sigma)\leq r. We see that σ\sigma is in (X‾,β‾)(r)n(\overline{X},\overline{\beta})(r)_{n}. Let σ\sigma be a simplex in (X‾,β‾)(r)n(\overline{X},\overline{\beta})(r)_{n}. Suppose that βn(σ)−ε≥0\beta_{n}(\sigma)-\varepsilon\geq0. We have βn′(σ)=βn(σ)−ε≤r\beta'_{n}(\sigma)=\beta_{n}(\sigma)-\varepsilon\leq r. If βn(σ)−ε<0\beta_{n}(\sigma)-\varepsilon<0, then βn′(σ)=0≤r\beta'_{n}(\sigma)=0\leq r. This implies that σ∈En\sigma\in E_{n}.

Proof of Lemma 3.7. By considering the adjoint ad(H)ad(H) to HH, we have a commutative diagram

X→ad(H)YI(f,g)↓↓πY×Y=Y×Y(5)\begin{CD} X @>{ad(H)}>> Y^{I}\\ @V{(f,g)}VV @VV{\pi}V\\ Y\times Y @= Y\times Y \tag*{(5)} \end{CD}

where π\pi is pointwise evaluation fibration. Observe that π\pi is a fibration in Top⁡(V,≤)\operatorname{Top}(\mathcal{V},\leq). Let c:Y→YIc:Y\to Y^{I} be the map assigns the constant map at yy to a point y∈Yy\in Y. Then, we have a commutative diagram

Y→ΔY×Yl↓↑πZ→pYIK↑↑ad(H)X=X\begin{CD} Y @>{\Delta}>> Y\times Y\\ @V{l}VV @AA{\pi}A\\ Z @>{p}>> Y^{I}\\ @A{K}AA @AA{ad(H)}A\\ X @= X \end{CD}

in which p∘ip\circ i is a decomposition of gg and KK is a lift of ad(H)ad(H). We observe that pp is a weak equivalence and XX is cofibrant.

Since the diagonal map Δ\Delta factors through ZZ, it follows that ZZ is a path object for YY. Then, the commutativity of the diagram (3.5) allows us to deduce that the morphism KK is a right homotopy between ff and gg. Each object in Top⁡(V,≤)\operatorname{Top}(\mathcal{V},\leq) is fibrant. Then, it follows from [19 Proposition 7.4.8] that f≃gf \simeq g in Top⁡(V,≤)\operatorname{Top}(\mathcal{V},\leq).

A product on a cochain complex associated with an enriched nerve

As in § 3, let V\mathcal{V} denote the monoidal poset ([0,∞),≤,+,0)([0,\infty),\leq,+,0). Let P ⁣:V→sSet⁡P\colon\mathcal{V}\to\operatorname{sSet} be a persistence simplicial set and KPKP the persistence chain complex associated with PP. Assume that the structure map P(ε≤ε′)P(\varepsilon\leq\varepsilon') is injective for ε≤ε′\varepsilon\leq\varepsilon'. We may identify the element σ\sigma in P(ε)P(\varepsilon) with P(ε≤ε′)σP(\varepsilon\leq\varepsilon')\sigma in P(ε′)P(\varepsilon').

Let cPcP and P(<ε)P(<\varepsilon) denote the colimits colim⁡lP(l)\operatorname{colim}_{l}P(l) and colim⁡l<εP(l)\operatorname{colim}_{l<\varepsilon}P(l), respectively. We observe that P(ε)=colim⁡l≤εP(l)P(\varepsilon)=\operatorname{colim}_{l\leq\varepsilon}P(l). It follows that the structure map P(ε′<ε)P(\varepsilon'<\varepsilon) induces a chain map ι<ε ⁣:KP(<ε)→KP(ε)\iota_{<\varepsilon}\colon KP(<\varepsilon)\to KP(\varepsilon). The chain complex RP:=KcP≅colim⁡lKP(l)RP:=KcP\cong\operatorname{colim}_{l}KP(l) is called the reachability (chain) complex.

Magnitude homology and cohomology

We recall magnitude cohomology introduced by Hepworth; see [15 Section 2, Definition 5.3 and Example 5.6]. Let XX be a generalized metric space. The (unnormalized) magnitude chain complex MCε,∗(X)MC_{\varepsilon,*}(X) of XX is defined by

MCε,n(X):=K{(x0,…,xn)∣∑i=0nd(xi,xi−1)=ε}MC_{\varepsilon,n}(X):=\mathbb{K}\{(x_0,\ldots,x_n)\mid\sum_{i=0}^{n}d(x_i,x_{i-1})=\varepsilon\}

with the differential dn:=∑i=0n(−1)idid_n:=\sum_{i=0}^{n}(-1)^id_i where σ=(x0,…,xn)\sigma=(x_0,\ldots,x_n),

di(σ)={(x0,…,xi−1,xi+1,…,xn)if (x0,…,xi−1,xi+1,…,xn) is in MCε,n−1(X)0otherwise.d_i(\sigma)= \begin{cases} (x_0,\ldots,x_{i-1},x_{i+1},\ldots,x_n) & \text{if }(x_0,\ldots,x_{i-1},x_{i+1},\ldots,x_n)\text{ is in }MC_{\varepsilon,n-1}(X)\\ 0 & \text{otherwise.} \end{cases}

The magnitude homology MHε,∗(X)MH_{\varepsilon,*}(X) is the homology of the magnitude chain complex. The magnitude cochain complex of XX is the dual to the magnitude chain complex: MCε∗(X)=Hom⁡(MCε,∗(X),K)MC^*_{\varepsilon}(X)=\operatorname{Hom}(MC_{\varepsilon,*}(X),\mathbb{K}). The magnitude cohomology MHε∗(X)MH^*_{\varepsilon}(X) is the cohomology of the magnitude cochain complex.

Give φ∈MCεm(X)\varphi\in MC^m_{\varepsilon}(X) and ψ∈MCδn(X)\psi\in MC^n_{\delta}(X), the product φ⋅ψ∈MCε+δm+n(X)\varphi\cdot\psi\in MC^{m+n}_{\varepsilon+\delta}(X) is defined on (x0,…,xm+n)∈MCε+δ,m+n(X)(x_0,\ldots,x_{m+n})\in MC_{\varepsilon+\delta,m+n}(X) by

φ⋅ψ(x0,…,xm+n)=φ(x0,…,xm)⋅φ(xm,…,xm+n)\varphi\cdot\psi(x_0,\ldots,x_{m+n})=\varphi(x_0,\ldots,x_m)\cdot\varphi(x_m,\ldots,x_{m+n})

if (x0,…,xm)∈MCε,∗(X)(x_0,\ldots,x_m)\in MC_{\varepsilon,*}(X) and (xm,…,xm+n)∈MCδ,∗(X)(x_m,\ldots,x_{m+n})\in MC_{\delta,*}(X), and by

φ⋅ψ(x0,…,xm+n)=0\varphi\cdot\psi(x_0,\ldots,x_{m+n})=0

otherwise. The product gives rise to that on the magnitude cohomology MH∗∗(X)MH^*_{*}(X). Moreover, we see that the product makes MH∗∗(X)MH^*_{*}(X) into a unital, associative bigraded ring with unit 1=[u]1=[u], where u∈MC00(X)u\in MC^0_0(X) is defined by u(x0)=1u(x_0)=1. Observe that the product is not graded commutative in general; see [15 Proposition 2.3].

We refer the reader to [15 Sections 6 and 7] for interesting computations of magnitude cohomology rings of finite graphs.

The cup product on a cochain complex

We consider a commutative diagram

Fε→F<ε→F<ε/Fεj↓i↓↓≅RP∗→pKP(ε)∗←(ι<ε)∗Ker⁡((ι<ε)∗)→u∗MCε∗(X)(6)\begin{CD} F^{\varepsilon} @>>> F^{<\varepsilon} @>>> F^{<\varepsilon}/F^{\varepsilon}\\ @V{j}VV @V{i}VV @VV{\cong}V\\ RP^* @>{p}>> KP(\varepsilon)^* @<<<{(\iota_{<\varepsilon})^*} \operatorname{Ker}((\iota_{<\varepsilon})^*) @>{u^*}>> MC^*_{\varepsilon}(X) \tag*{(6)} \end{CD}

where FεF^{\varepsilon} and F<εF^{<\varepsilon} are kernels of pp and p′p', respectively. We have an exact sequence 0→KP(<ε)→ι<εKP(ε)→KP(ε)/KP(<ε)→00 \to KP(<\varepsilon) \xrightarrow{\iota_{<\varepsilon}} KP(\varepsilon) \to KP(\varepsilon)/KP(<\varepsilon) \to0 of chain complexes.

Let MCε,∗(X)MC_{\varepsilon,*}(X) be the unnormalized magnitude chain complex for a generalized metric space XX; see [15 Section 4]. We observe that there exists an isomorphism u:MCε,∗(X)→KP(ε)/KP(<ε)u:MC_{\varepsilon,*}(X)\to KP(\varepsilon)/KP(<\varepsilon) of chain complexes, where P=N(X):[0,∞)→sSetP=N(X):[0,\infty)\to\mathrm{sSet} is the persistence simplicial set described in the beginning of Section 3. In fact, the inclusion MCε,∗(X)→KP(ε)MC_{\varepsilon,*}(X)\to KP(\varepsilon) gives rise to the isomorphism. Since KP(ε)/KP(<ε)KP(\varepsilon)/KP(<\varepsilon) is a free KK-module, it follows that the dual sequence

0←KP(<ε)∗←(ι<ε)∗KP(ε)∗←(KP(ε)/KP(<ε))∗←00 \leftarrow KP(<\varepsilon)^{*} \xleftarrow{(\iota_{<\varepsilon})^{*}} KP(\varepsilon)^{*} \leftarrow(KP(\varepsilon)/KP(<\varepsilon))^{*} \leftarrow0

is exact. Then, the natural map uu is an isomorphism. We have a copersistence cochain complex F:V→coChKF:\mathcal{V}\to\mathrm{coCh}_{K} defined by F(ε)=F<εF(\varepsilon)=F^{<\varepsilon} with the natural inclusion F(ε≤ε′):F(ε′)→F(ε)F(\varepsilon\leq\varepsilon'):F(\varepsilon')\to F(\varepsilon); that is, F:Vop→coChKF:\mathcal{V}^{\mathrm{op}}\to\mathrm{coCh}_{K} is a functor.

The usual cup product ∪\cup on the cochain complex RP∗:=Hom⁡K(RP,K)RP^{*}:=\operatorname{Hom}_{K}(RP,K) is defined by

(φ∪ψ)(σ)=((dp+1⋯dp+p′)∗φ)(σ)⋅K((d0⋯dp−1)∗ψ)(σ),(7)(\varphi\cup\psi)(\sigma)=((d_{p+1}\cdots d_{p+p'})^{*}\varphi)(\sigma)\cdot_{K}((d_{0}\cdots d_{p-1})^{*}\psi)(\sigma), \tag*{(7)}

for φ∈(RP)p\varphi\in(RP)^{p} and ψ∈(RP)p′\psi\in(RP)^{p'}, where ⋅K\cdot_{K} denotes the multiplication on KK. In what follows, we may write D+D_{+} and D−D_{-} for dp+1⋯dp+p′d_{p+1}\cdots d_{p+p'} and d0⋯dp−1d_{0}\cdots d_{p-1} in (7), respectively.

Consider the condition (A) below for a persistence simplicial set P:V→sSetP:\mathcal{V}\to\mathrm{sSet} whose structure maps are injective.

(A): D+(σ)∈Pm(<ε)D_{+}(\sigma)\in P_{m}(<\varepsilon) or D−(σ)∈Pn(<δ)D_{-}(\sigma)\in P_{n}(<\delta) for σ∈Pm+n(<ε+δ)\sigma\in P_{m+n}(<\varepsilon+\delta) and integers m,nm,n.

Theorem 4.1. Let PP be a persistence simplicial set for which the condition (A) holds. Then, the restriction of the cup product on RP∗RP^{*} to F(ε)⊗F(δ)F(\varepsilon)\otimes F(\delta) factors through F(ε+δ)F(\varepsilon+\delta). Thus, one has a commutative diagram

F(ε)⊗F(δ)→∪F(ε+δ)i⊗j↓↓kRP∗⊗RP∗→∪RP∗\begin{CD} F(\varepsilon)\otimes F(\delta) @>\cup>> F(\varepsilon+\delta)\\ @V{i\otimes j}VV @VV{k}V\\ RP^{*}\otimes RP^{*} @>\cup>> RP^{*} \end{CD}

where ii, jj and kk are natural inclusions. As a consequence, the product on F(∗)F(*) is associative. Moreover, the cup product compatible with the structure maps of the copersistence cochain complex; that is, the diagram

F(ε′)⊗F(δ′)→∪F(ε′+δ′)F(ε≤ε′)⊗F(δ≤δ′)↓↓F(ε+δ≤ε′+δ′)F(ε)⊗F(δ)→∪F(ε+δ)\begin{CD} F(\varepsilon')\otimes F(\delta') @>\cup>> F(\varepsilon'+\delta')\\ @V{F(\varepsilon\leq\varepsilon')\otimes F(\delta\leq\delta')}VV @VV{F(\varepsilon+\delta\leq\varepsilon'+\delta')}V\\ F(\varepsilon)\otimes F(\delta) @>\cup>> F(\varepsilon+\delta) \end{CD}

is commutative for ε≤ε′\varepsilon\leq\varepsilon' and δ≤δ′\delta\leq\delta'.

Proof. By definition, we see that F(ε)F(\varepsilon) is the kernel of the natural map RP∗→KP(<ε)RP^{*}\to KP(<\varepsilon). This gives the commutativity of the first square. The cup product on RP∗RP^{*} is associative and then so is the restriction. The direct computation yields the commutativity of the second square. □\square

Remark 4.2. Suppose that the unit 11 in RP0RP^{0} is in F(0)0F(0)^{0}. Then F(∗)F(*) is unital in the sense that 1∪ψ=ψ=ψ∪11\cup\psi=\psi=\psi\cup1 for ψ∈F(ε)\psi\in F(\varepsilon).

Proposition 4.3. Let CC be a WopW^{\mathrm{op}}-category, where WW is the monoidal poset V\mathcal{V} or ([0,∞],≤,+,0)([0,\infty],\leq,+,0). Then, the persistence simplicial set N(C):V→sSetN(C):\mathcal{V}\to\mathsf{sSet} described in Section 3 satisfies the condition (A).

Proof. For an element (x0,…,xm,xm+1,…,xm+n)(x_0,\ldots,x_m,x_{m+1},\ldots,x_{m+n}) in N(X)n+m(<ε+δ)N(X)_{n+m}(<\varepsilon+\delta), we see that ℓ(D−(xm,…,xn+m))<δ\ell(D_-(x_m,\ldots,x_{n+m}))<\delta if ℓ(D+(x0,…,xm+n))≥ε\ell(D_+(x_0,\ldots,x_{m+n}))\geq\varepsilon. □

Remark 4.4. Let VR⁡(X):[0,∞)→sSet\operatorname{VR}(X):[0,\infty)\to\mathsf{sSet} be the Vietoris-Rips simplicial set of a metric space (X,d)(X,d) defined by

VR⁡(X)(ε)n:={(x0,…,xn)∣xi∈X, d(xi,xj)≤ε for i,j}.\operatorname{VR}(X)(\varepsilon)_n:=\{(x_0,\ldots,x_n)\mid x_i\in X,\ d(x_i,x_j)\leq\varepsilon\text{ for }i,j\}.

In general, the condition (A) for the Vietoris-Rips simplicial set does not hold. In fact, for the complete graph KnK_n with n≥3n\geq3, consider the simplex σ=(x0,x1,x2)\sigma=(x_0,x_1,x_2) consisting of distinct points. While D+(σ)=(x0,x1)D_+(\sigma)=(x_0,x_1) and D−(σ)=(x1,x2)D_-(\sigma)=(x_1,x_2) are not in P1(<1)P_1(<1), we have σ∈P2(<2)\sigma\in P_2(<2).

Proposition 4.5. Let XX be a generalized metric space. One has a commutative diagram

F<ε/Fε⊗F<δ/Fδ⟶F<ε+δ/Fε+δα⊗α≅↓α⊗α≅α≅↓≅αKer⁡(ι<ε)∗⊗Ker⁡(ι<δ)∗⟶Ker⁡(ι<ε+δ)∗u∗⊗u∗≅↓u∗⊗u∗≅u∗≅↓≅u∗MCε∗(X)⊗MCδ∗(X)→⋅HMCε+δ∗(X),\begin{aligned} F^{<\varepsilon}/F^\varepsilon\otimes F^{<\delta}/F^\delta&\longrightarrow F^{<\varepsilon+\delta}/F^{\varepsilon+\delta}\\ \alpha\otimes\alpha\mathrel{\cong}\downarrow\phantom{\alpha\otimes\alpha\mathrel{\cong}}&\phantom{\alpha\mathrel{\cong}}\downarrow\mathrel{\cong}\alpha\\ \operatorname{Ker}(\iota_{<\varepsilon})^*\otimes\operatorname{Ker}(\iota_{<\delta})^*&\longrightarrow\operatorname{Ker}(\iota_{<\varepsilon+\delta})^*\\ u_*\otimes u_*\mathrel{\cong}\downarrow\phantom{u_*\otimes u_*\mathrel{\cong}}&\phantom{u_*\mathrel{\cong}}\downarrow\mathrel{\cong}u_*\\ MC_\varepsilon^*(X)\otimes MC_\delta^*(X)&\xrightarrow{\cdot_H}MC_{\varepsilon+\delta}^*(X), \end{aligned}

where ⋅H\cdot_H denotes the product on the (unnormalized) magnitude cochain complex MC∗∗(X)MC_*^*(X) defined by Hepworth; see § 4.1.

Proof. The result follows from the direct computation. □

The multiplicative magnitude-path spectral sequence (MPSS)

Let P:[0,∞)→sSetP:[0,\infty)\to\mathsf{sSet} be a persistence simplicial set for which the condition (A) holds. We restrict the domain of the functor PP to N\mathbb{N}. The same notation PP is used for the restricted functor. We recall the reachability complex RP:=colim⁡l∈NKP(l)≅K(colim⁡l∈NP)RP:=\operatorname{colim}_{l\in\mathbb{N}}KP(l)\cong K(\operatorname{colim}_{l\in\mathbb{N}}P).

With the same notation as in the diagram (6), we have the decreasing filtration

F={Fl(RP∗)}l∈NF=\{F^l(RP^*)\}_{l\in\mathbb{N}}

defined by

(Fl(RP∗))n:=(F<l)n=(Fl−1)n.(F^l(RP^*))^n:=(F^{<l})^n=(F^{l-1})^n.

The usual procedure to make a spectral sequence, we have a right half-plane spectral sequence {Er∗,∗,dr}\{E_r^{*,*},d_r\}. In fact, for a filtered complex (A,F,d)(A,F,d), the ErE_r-term of the spectral sequence {Er∗,∗,dr}\{E_r^{*,*},d_r\} is defined by

Erp,q:=Zrp,q/(Zr−1p+1,q−1+Br−1p,q),E_r^{p,q}:=Z_r^{p,q}/(Z_{r-1}^{p+1,q-1}+B_{r-1}^{p,q}),

where Zrp,qZ_r^{p,q} and Brp,qB_r^{p,q} are defined by

Zrp,q:=FpAp+q∩d−1(Fp−rAp+q+1)Z_r^{p,q}:=F^pA^{p+q}\cap d^{-1}(F^{p-r}A^{p+q+1})

and

Brp,q:=FpAp+q∩d(Fp−rAp+q−1),B_r^{p,q}:=F^pA^{p+q}\cap d(F^{p-r}A^{p+q-1}),

respectively. The differential drd_r on the ErE_r-term is induced by dd; see [30 The proof of Theorem 2.6]. Thanks to Theorem 4.1, the spectral sequence is endowed with a product structure.

The following theorem is a corollary of the consideration above.

Theorem 4.6. (The multiplicative magnitude-path spectral sequence) Let XX be a generalized metric space. Then, the filtration F:={FlKN(X)}l∈NF:=\{F^lKN(X)\}_{l\in\mathbb{N}} gives rise to a multiplicative spectral sequence {Er∗,∗,dr}r≥1\{E_r^{*,*},d_r\}_{r\geq1} with Erp,q=0E_r^{p,q}=0 for p+q<0p+q<0, E1p,q≅MHpp+q(X)E_1^{p,q}\cong MH_p^{p+q}(X) as a K-module and

E1∗,∗≅⨁p,qMHpp+q(X)E_1^{*,*}\cong\bigoplus_{p,q}MH_p^{p+q}(X)

as a bigraded algebra. Moreover, the spectral sequence converges to H∗(RN(X))H^*(RN(X)) as an algebra provided the filtration FF is bounded.

Proof. It follows from Proposition 4.3 and Theorem 4.1 that

∪(FsN(X)m⊗FtN(X)n)⊂Fs+tN(X)m+n\cup(F^sN(X)^m\otimes F^tN(X)^n)\subset F^{s+t}N(X)^{m+n}

for the cup product. Then, we see that the filtration FF gives a bigraded algebra. By [30 Theorem 2.14], we have the multiplicative spectral sequence associated by FF. Thus, Proposition 4.5 enables us to deduce that the E1E_1-term is isomorphic to the magnitude cohomology ring ⨁l,nMHln(X)\bigoplus_{l,n}MH_l^n(X). We have the result. □

Remark 4.7. The pp-axis E1∗,0E_1^{*,0} in the spectral sequence in Theorem 4.6 is nothing but the diagonal part of the magnitude cohomology ring ⨁n,nMHnn(X)\bigoplus_{n,n}MH_n^n(X).

Remark 4.8. i) If the generalized metric space XX is finite, then the filtration FF in Theorem 4.6 is bounded.

ii) Suppose that the filtration FF is degreewise bounded; that is, for each cohomology degree nn, there exists a filtration degree l=l(n)l=l(n) such that FlKPn=RPnF^lKP^n=RP^n. Then, for (p,q)(p,q) a pair of nonnegative integers, there exists an integer NN such that Erp,q≅E∞p,qE_r^{p,q}\cong E_\infty^{p,q} as a K-module for r≥Nr\geq N; see [30 The proof of Theorem 2.6].

iii) We may apply the construction of the spectral sequence in the proof of Theorem 4.6 to the enriched nerve of the Nop\mathbb{N}^{\mathrm{op}}-category ⌈ ⌉X\lceil\ \rceil X; see Example 2.2 ii). The E1E_1-term is isomorphic to the magnitude cohomology algebra of ⌈ ⌉X\lceil\ \rceil X.

iv) Let XX be a compact Riemannian manifold, more general, a Riemannian diffeological space which is compact for the D-topology; see [24 Theorem 3.21]. Then the map ℓ:Xn→[0,∞]\ell:X^n\to[0,\infty] is bounded for each nn; see [24 Theorem 3.21] for a diffeological space. This implies that the filtration FF which construct the spectral sequence in Theorem 4.6 and ii) above is bounded degreewise. Then, these spectral sequences converge degreewise.

Remark 4.9. The cohomology ring H∗(RN(X))H^*(RN(X)) is commutative. In fact, the cochain complex RN(X)RN(X) is obtained by the simplicila set colim⁡l∈NN(X)\operatorname{colim}_{l\in\mathbb{N}}N(X). However, the magnitude cohomology ring is not necessarily commutative; see [15 Proposition 2.3]. It is an interesting problem to consider which term in the multiplicative MPSS is (bigraded) commutative.

We end this section with a comment on a relationship between the MPSS and an rr-homotopy equivalence mentioned in § 3. The topic is already discussed in [2 Section 4], [29 Section 3] and [17 Section 2].

Let XX and X′X' be generalized metric spaces and f,g:X→X′f,g:X\to X' rr-homotopic maps. Assume that rr is a non-negative integer. The simplicial homotopy h:N(X)→N(X′)h:N(X)\to N(X') from N(f)N(f) to N(g)N(g) in Remark 3.4 gives rise to a chain homotopy h~:RN(X)∗→RN(X′)∗+1\widetilde{h}:RN(X)_*\to RN(X')_{*+1} from RN(f)RN(f) to RN(g)RN(g); see, for example, [3 Chapter 3, 4.2 Proposition]. By the construction, we see that $\widetilde{h}(KN(X)(p))\subset KN(X')(p+ r)r) for each p∈Np \in\mathbb{N}. This enables us to conclude that h~∗(Fp+r(RN(X′)∗))⊂Fp(RN(X)∗−1)\widetilde{h}^{*}(F^{p+r}(RN(X')^{*})) \subset F^{p}(RN(X)^{*-1}). Thus, by [7 Proposition XV, 3.1], we have

Proposition 4.10. ([2 Theorem 4.37] and [29 Proposition 3.3.3]) Suppose that f,g:X→X′f,g:X\to X' are rr-homotopic maps of generalized metric spaces. Then, Es(f)=Es(g)E_s(f)=E_s(g) for all s>rs>r. As a consequence, for an rr-homotopy equivalence f:X→X′f:X\to X', the morphism Er(f):Er∗,∗(X)→Er∗,∗(X′)E_r(f):E_r^{*,*}(X)\to E_r^{*,*}(X') is a quasi-isomorphism of differential bigraded algebras.

Interleavings and the MPSS

We describe a relationship between interleavings of enriched nerves and the notion of an rr-homotopy in Theorem 3.9. Proposition 4.10 relates the rr-homotopy to the MPSS. Therefore, one may expect a relationship of the MPSS and the interleaving between persistence chain complexes. However, in general, an interleaving is not a morphism of persistence objects and, therefore, does not naturally induce a morphism of spectral sequences. The following proposition gives a relationships of the MPSS and interleaving with an acyclic persistence chain complex.

Theorem 5.1. Let XX be a generalized metric space whose distance takes values in non-negative integers and {Er∗,∗(X),dr}\{E_r^{*,*}(X),d_r\} the MPSS for XX. Suppose that

dIHCet(KN(X),KN(pt))=r2,d_{\mathrm{IHC}}^{\mathrm{et}}(\mathrm{KN}(X),\mathrm{KN}(\mathrm{pt}))=\frac{r}{2},

where dIHCetd_{\mathrm{IHC}}^{\mathrm{et}} denotes the interleaving distance in the homotopy category of the category et(ChK[0,∞))\mathrm{et}(\mathrm{Ch}_{\mathbb{K}}^{[0,\infty)}) of extended tame persistence chain complexes endowed with the model structure described in [22]; see Theorem 5.3 below for the detail. Then, one has Es∗,∗(X)=KE_s^{*,*}(X)=\mathbb{K} for s>rs>r; see also Remark 5.6. In particular, the reachability complex RN(X)=colim⁡lKN(X)(l)RN(X)=\operatorname{colim}_l\mathrm{KN}(X)(l) is acyclic.

Before proving Theorem 5.1, we need to explain terminology used in the theorem. The model structure mentioned in Theorem 5.1 is given modifying that of the category of tame persistence objects defined in [8]. The advantage of using the model structure is that we have comparatively tractable cofibrations in the model category. On the other hand, it is difficult to say that a cofibration in the projective model structure of a functor category has an explicit form; see, for example, [19 Theorem 11.6.1].

We here consider the persistence chain complex KN(X)\mathrm{KN}(X) in the category et(ChK[0,∞))\mathrm{et}(\mathrm{Ch}_{\mathbb{K}}^{[0,\infty)}) of extended tame persistence chain complexes.

Definition 5.2. ([8 Definition 2.1], [22 Definition 2.4]) Let C\mathcal{C} be a category. A sequence τ0<τ1<⋯<τn<⋯\tau_0<\tau_1<\cdots<\tau_n<\cdots in [0,∞)[0,\infty), which is divergent or finite, discretises a functor X:[0,∞)→CX:[0,\infty)\to\mathcal{C} if X(s≤t):X(s)→X(t)X(s\leq t):X(s)\to X(t) may fail to be an isomorphism only if there is a τi∈N\tau_i\in\mathbb{N} such that s<τi≤ts<\tau_i\leq t. A functor X:[0,∞)→CX:[0,\infty)\to\mathcal{C} is called extended tame if there is a sequence that discretises it. Let et(C[0,∞))\mathrm{et}(\mathcal{C}^{[0,\infty)}) denote the full subcategory of the functor category C[0,∞)\mathcal{C}^{[0,\infty)} whose objects are extended tame functors. An object XX in et(C[0,∞))\mathrm{et}(\mathcal{C}^{[0,\infty)}) is called tame if the sequence which discretises XX is finite; see [8].

Let C\mathcal{C} be a category. We recall a factorization of a morphism g:X→Yg:X\to Y in et(C[0,∞))\mathrm{et}(\mathcal{C}^{[0,\infty)}). Let τ0<τ1<⋯<τn<⋯\tau_0<\tau_1<\cdots<\tau_n<\cdots be a sequence discretising both XX and YY; see [8 Section 2.2]. By induction on {0,1,2,…,n,…}\{0,1,2,\ldots,n,\ldots\}, we define morphisms gˉ(τa):X(τa)→Q(τa)\bar{g}(\tau_a): X(\tau_a) \to Q(\tau_a) and g^(τa):Q(τa)→Y(τa)\hat{g}(\tau_a): Q(\tau_a) \to Y(\tau_a) in C\mathcal{C} as follows: (gˉ(0):X(0)→Q(0)):=(1:X(0)→X(0))(\bar{g}(0): X(0) \to Q(0)) := (1: X(0) \to X(0)) and (g^(0):Q(0)→Y(0)):=(g(0):X(0)→Y(0))(\hat{g}(0): Q(0) \to Y(0)) := (g(0): X(0) \to Y(0)). For a>0a > 0, the object Q(τa)Q(\tau_a) is defined by

colim⁡(Y(τa−1)←g(τa−1)X(τa−1)→X(τa−1<τa)X(τa))\operatorname{colim}\left( Y(\tau_{a-1}) \xleftarrow{g(\tau_{a-1})} X(\tau_{a-1}) \xrightarrow{X(\tau_{a-1}<\tau_a)} X(\tau_a) \right)

with gˉ(τa):X(τa)→Q(τa)\bar{g}(\tau_a): X(\tau_a) \to Q(\tau_a) and g^(τa):Q(τa)→Y(τa)\hat{g}(\tau_a): Q(\tau_a) \to Y(\tau_a) which fit in the commutative diagram

X(τa−1)→X(τa−1<τa)X(τa)g(τa−1)↓↓g(τa)Y(τa−1)→Q(τa)→g^(τa)Y(τa)(8)\begin{CD} X(\tau_{a-1}) @>{X(\tau_{a-1}<\tau_a)}>> X(\tau_a) \\ @V{g(\tau_{a-1})}VV @VV{g(\tau_a)}V \\ Y(\tau_{a-1}) @>>> Q(\tau_a) @>{\hat{g}(\tau_a)}>> Y(\tau_a) \tag*{(8)} \end{CD}

consisting of the pushout square. For a>0a > 0, define Q(τa−1<τa):Q(τa−1)→Q(τa)Q(\tau_{a-1}<\tau_a): Q(\tau_{a-1}) \to Q(\tau_a) to be the composite of the morphisms g^(τa−1):Q(τa−1)→Y(τa−1)\hat{g}(\tau_{a-1}): Q(\tau_{a-1}) \to Y(\tau_{a-1}) and Y(τa−1)→Q(τa)Y(\tau_{a-1}) \to Q(\tau_a) which is the bottom horizontal arrow in (8). Thus, we have a factorization g=g^∘gˉg = \hat{g} \circ\bar{g} of gg.

Let M\mathcal{M} be a model category. We recall a model structure on the full subcategory of the functor category M[0,∞)\mathcal{M}^{[0,\infty)} consisting of extended tame persistence objects. The following results for tame objects are originally proved by Chachólski, Giunti and Landi [8]. The extended tame versions are also proved by the same argument as in the proof of [8 Theorem 2.2]; see [22 Comments after Proposition 2.6].

Theorem 5.3. ([8, Theorem 2.2], [22, Theorem 2.5]) Let M\mathcal{M} be a model category. With the notation and terminology mentioned above, the following choices of weak equivalences, fibrations and cofibrations form a model structure on et⁡(M[0,∞))\operatorname{et}(\mathcal{M}^{[0,\infty)}). A morphism g:X→Yg: X \to Y in et⁡(M[0,∞))\operatorname{et}(\mathcal{M}^{[0,\infty)}) is a

  • weak equivalence if g(t):X(t)→Y(t)g(t): X(t) \to Y(t) is a weak equivalence for all tt,

  • fibration if g(t):X(t)→Y(t)g(t): X(t) \to Y(t) is a fibration for all tt,

  • cofibration if g^(t):Q(t)→Y(t)\hat{g}(t): Q(t) \to Y(t) is a cofibration for all tt.

It is worthwhile mentioning that the model category in Theorem 5.3 is not necessarily cofibrantly generated. We have a characterization of a cofibrant object in et⁡(M[0,∞))\operatorname{et}(\mathcal{M}^{[0,\infty)}).

Proposition 5.4. ([8, Proposition 2.3], [22, Proposition 2.6]) An object XX in et⁡(M[0,∞))\operatorname{et}(\mathcal{M}^{[0,\infty)}) is cofibrant if and only if X(0)X(0) is cofibrant and, for any s<ts < t in [0,∞)[0,\infty), the transition morphism X(s<t):X(s)→X(t)X(s < t): X(s) \to X(t) is a cofibration in M\mathcal{M}.

We apply the construction in Theorem 5.3 to the category Ch⁡K\operatorname{Ch}_{K} of non-negatively graded chain complexes whose model structure is described in [12 7.2 Theorem]. Observe that by definition, a morphism i:M→Ni: M \to N in Ch⁡K\operatorname{Ch}_{K} is a cofibration if ik:Mk→Nki_k: M_k \to N_k is a monomorphism and the cokernel of iki_k is a projective K-module for k≥0k \ge0.

Let II be the chain complex over K generated by 0~\tilde{0}, 1~\tilde{1} and ι\iota, with d(ι)=0~−1~d(\iota) = \tilde{0} - \tilde{1} and deg⁡(ι)=1\deg(\iota) = 1. We regard II as the constant persistence chain complex.

Let MM be a cofibrant object in et⁡(ChK[0,∞))\operatorname{et}(\mathrm{Ch}_{K}^{[0,\infty)}). For the folding map ∇:M⊕2→M\nabla:M^{\oplus2}\to M, we have a commutative diagram

M⊕2→gM⊗I→πM∇↓∇~↓∥M=M=M(9)\begin{CD} M^{\oplus2} @>{g}>> M\otimes I @>{\pi}>> M \\ @V{\nabla}VV @V{\widetilde{\nabla}}VV @| \\ M @= M @= M \tag*{(9)} \end{CD}

in et⁡(ChK[0,∞))\operatorname{et}(\mathrm{Ch}_{K}^{[0,\infty)}). Here M⊗IM\otimes I is the persistence chain complex defined by the tensor product pointwise. Moreover, the map gg is induced by the natural inclusion and the map π\pi is defined by π(x⊗ı~)=x\pi(x\otimes\widetilde{\imath})=x for i=0,1i=0,1 and π(x⊗ι)=0\pi(x\otimes\iota)=0. It is readily seen that π\pi is a trivial fibration.

Lemma 5.5. Let MM be a cofibrant object in et⁡(ChK[0,∞))\operatorname{et}(\mathrm{Ch}_{K}^{[0,\infty)}). The map gg in (9) is a cofibration in the sense in Theorem 5.3. As a consequence, the persistence chain complex M⊗IM\otimes I is a cylinder object of MM in et⁡(ChK[0,∞))\operatorname{et}(\mathrm{Ch}_{K}^{[0,\infty)}).

Proof. It suffices to show that gg in the diagram (9) is a cofibration in et⁡(ChK[0,∞))\operatorname{et}(\mathrm{Ch}_{K}^{[0,\infty)}). We consider the commutative diagram

M(k)⊕2→η⊕2(k<k+1)M(k+1)⊕2g(k)↓↓g~(k+1)M(k)⊗I→M(k+1)⊕2⊕M(k)⊗K{ι}→g~(k+1)M(k+1)⊗I\begin{CD} M(k)^{\oplus2} @>{\eta^{\oplus2}(k<k+1)}>> M(k+1)^{\oplus2} \\ @V{g(k)}VV @VV{\widetilde{g}(k+1)}V \\ M(k)\otimes I @>>> M(k+1)^{\oplus2}\oplus M(k)\otimes K\{\iota\} @>{\widetilde{g}(k+1)}>> M(k+1)\otimes I \end{CD}

in which the inner square is the pushout. It follows that g~(k+1)\widetilde{g}(k+1) is a monomorphism in each degree of the chain complex and the cokernel of g~(k+1)\widetilde{g}(k+1) is a projective module in each degree. Then, the morphism g~(k+1)\widetilde{g}(k+1) is a cofibration in ChK\mathrm{Ch}_{K}; see the comment that follow Proposition 5.4. Thus, by definition, we see that gg is a cofibration in et⁡(ChK[0,∞))\operatorname{et}(\mathrm{Ch}_{K}^{[0,\infty)}). □

Proof of Theorem 5.1. Let K\mathbb{K} be the constant persistence chain complex K\mathbb{K} with K(l)0=K\mathbb{K}(l)_0=\mathbb{K} and K(l)i=0\mathbb{K}(l)_i=0 for i≥1i\ge1. Since KN(pt)≅K\mathrm{KN}(pt)\cong\mathbb{K} in the homotopy category Ho(et⁡(ChK[0,∞)))\mathrm{Ho}(\operatorname{et}(\mathrm{Ch}_{K}^{[0,\infty)})), it follows form the assumption that dIHCet(KN(X),K)=r/2d_{\mathrm{IHC}}^{\mathrm{et}}(\mathrm{KN}(X),\mathbb{K})=r/2.

We observe that the non-negative integers discretises the persistence chain complex KN(X)\mathrm{KN}(X). Therefore, we see that rr is an integer. In fact, if rr is not an integer, then KN(X)\mathrm{KN}(X) and K\mathbb{K} are (⌊r⌋/2)(\lfloor r\rfloor/2)-interleaved in the homotopy category, which is a contradiction.

Proposition 5.4 yields that KN(X)\mathrm{KN}(X) and K\mathbb{K} are cofibrant. Moreover, the both of the objects are fibrant. Thus, we have an r/2r/2-interleaving (KN(X),K,φ,ψ)(\mathrm{KN}(X),\mathbb{K},\varphi,\psi) in the homotopy category for which ψr/2∘φ≃ηKN(X)r\psi^{r/2}\circ\varphi\simeq\eta_{\mathrm{KN}(X)}^{r} and φr/2∘ψ≃ηKr=id⁡K\varphi^{r/2}\circ\psi\simeq\eta_{\mathbb{K}}^{r}=\operatorname{id}_{\mathbb{K}} via homotopies H1:KN(X)⊗I→KN(X)rH_1:\mathrm{KN}(X)\otimes I\to\mathrm{KN}(X)^r and H2:K⊗I→KrH_2:\mathbb{K}\otimes I\to\mathbb{K}^r with the cylinder objects described in Lemma 5.5, respectively; see Definition 3.1.

Since K\mathbb{K} is concentrated in the zero degree of each chain complex, it follows that φ:KN(X)→Kr/2\varphi:\mathrm{KN}(X)\to\mathbb{K}^{r/2} factors through K\mathbb{K}; that is, φ=ηKr/2∘φ′\varphi=\eta_{\mathbb{K}}^{r/2}\circ\varphi' for some morphism φ′:KN(X)→K\varphi':\mathrm{KN}(X)\to\mathbb{K} of persistence chain complexes. Moreover, the image of ψ\psi is in KN(X)(∗)0\mathrm{KN}(X)(*)_0 and then in the image of the structure map η:KN(X)(0)0→KN(X)(∗)0\eta:\mathrm{KN}(X)(0)_0\to\mathrm{KN}(X)(*)_0. Thus, there exists a morphism ψ′:K→KN(X)\psi':\mathbb{K}\to\mathrm{KN}(X) of persistence chain complexes such that ψ=ηKN(X)r/2∘ψ′\psi=\eta_{\mathrm{KN}(X)}^{r/2}\circ\psi'. The homotopy H1H_1 mentioned above is in et⁡(ChK[0,∞))\operatorname{et}(\mathop{\mathrm{Ch}}\nolimits_{\mathbb{K}}^{[0,\infty)}). Therefore, the homotopy induces a chain map RN(X)⊗I→RN(X)RN(X) \otimes I \to RN(X) and then a chain homotopy h1:RN(X)∗→RN(X)∗+1h_1: RN(X)_* \to RN(X)_{*+1} defined by h1(x)=(−1)deg⁡xH1(x⊗ι)h_1(x)=(-1)^{\deg x}H_1(x\otimes\iota). We have h1:ψ′φ′≃id⁡RN(X)h_1: \psi'\varphi'\simeq\operatorname{id}_{RN(X)}. The same argument above yields a chain homotopy h2h_2 from K=colim⁡K\mathbb{K}=\operatorname{colim}\mathbb{K} to itself which gives φ′ψ′≃id⁡K\varphi'\psi'\simeq\operatorname{id}_{\mathbb{K}}. The homotopies hih_i increases the filtration degree of RN(X)RN(X) and K\mathbb{K} by ss. Thus, the argument before Proposition 4.10 yields that Es∗,∗=KE_s^{*,*}=\mathbb{K} for s>rs>r. □\square

Remark 5.6. Let XX be a generalized metric space which satisfies the same condition as in Theorem 5.1. Consider the homological MPSS {Er∗,∗(X),dr}\{E_r^{*,*}(X),d^r\} for XX. Then, the argument before Proposition 4.10 and the proof of Theorem 5.1 yield that Es∗,∗=KE_s^{*,*}=\mathbb{K} for s>rs>r.

Remark 5.7. i) Let GG be a connected finite digraph with the shortest path metric. Then cN(G)cN(G) is contractible and then RN(G)RN(G) is acyclic; see Remark 3.4 (2). Therefore, the distance dIHCet(KN(X),KN(pt))d_{\mathrm{IHC}}^{\mathrm{et}}(KN(X),KN(pt)) is finite.

ii) The distance in the homotopy category described in Theorem 5.1 is computed by using the bottleneck distance between the persistence homology H∗(KN(X))H_*(KN(X)) and the constant persistence module K\mathbb{K} if K\mathbb{K} is a field; see Remark A.4 for more details.

The interleaving distance dIHCetd_{\mathrm{IHC}}^{\mathrm{et}} described in Theorem 5.1 is related to dIHCd_{\mathrm{IHC}} considered in Section 3; see Appendix A. The relationship enables us to give a computational example of the interleaving distance in the homotopy category between enriched nerves of a direct cycle and the point. We recall the definition of the directed cycle.

Definition 5.8. ([16 Definition 8.1]) For m≥1m\geq1, the directed cycle ZmZ_m of length mm is the digraph with mm cyclicially ordered vertices, and with a single directed edge adjacent vertices, consistently oriented.

In [16 Section 8], the homological MPSS is explored in detail the directed cycle ZmZ_m. By using the result of the computation, we have

Proposition 5.9. For the directed cycle ZmZ_m of length m≥3m\geq3,

dIHCet(KN(Zm),KN(pt))=dIHC(N(Zm)∙,N(x1)∙)=m−12.d_{\mathrm{IHC}}^{\mathrm{et}}(KN(Z_m),KN(pt))=d_{\mathrm{IHC}}(N(Z_m)_\bullet,N(x_1)_\bullet)=\frac{m-1}{2}.

Proof. By virtue of Corollary 3.11, we have dIHC(N(Zm)∙,N(x1)∙)≤m−12d_{\mathrm{IHC}}(N(Z_m)_\bullet,N(x_1)_\bullet)\leq\frac{m-1}{2}. It follows from Corollary A.2 that dIHCet(KN(Zm),KN(pt))≤dIHC(N(Zm)∙,N(x1)∙)d_{\mathrm{IHC}}^{\mathrm{et}}(KN(Z_m),KN(pt))\leq d_{\mathrm{IHC}}(N(Z_m)_\bullet,N(x_1)_\bullet). The result [16 Theorem 8.2] enables us to deduce that for {Er∗,∗(Zm),dr}\{E_r^{*,*}(Z_m),d^r\} the homology MPSS of ZmZ_m, the EmE_m-term is trivial but Em−1E_{m-1}-term is non-trivial. Then, Theorem 5.1 and Remark 5.6 imply that m−12≤dIHCet(KN(Zm),KN(pt))\frac{m-1}{2}\leq d_{\mathrm{IHC}}^{\mathrm{et}}(KN(Z_m),KN(pt)). We have the result. □\square

Acknowledgements. The author thanks Takahito Naito, Shun Wakatsuki and Toshihiro Yamaguchi for valuable discussions on interleavings of enriched nerves. The author was partially supported by JSPS KAKENHI Grant Number JP26K00604.

Appendix A. dIHCd_{\mathrm{IHC}} versus dIHCetd_{\mathrm{IHC}}^{\mathrm{et}}

We compare the interleaving distances dIHCd_{\mathrm{IHC}} and dIHCetd_{\mathrm{IHC}}^{\mathrm{et}} between enriched nerves. Let M\mathcal{M} be a cofibrantly generated model category and i:et⁡(M[0,∞))→M[0,∞)i:\operatorname{et}(\mathcal{M}^{[0,\infty)})\to\mathcal{M}^{[0,\infty)} the inclusion functor. We regard M[0,∞)\mathcal{M}^{[0,\infty)} and et⁡(M[0,∞))\operatorname{et}(\mathcal{M}^{[0,\infty)}) as the model categories endowed with the projective model structure and the model structure described in Theorem 5.3, respectively.

Let (⌊ ⌋)∗:MN∪{0}→et⁡(M[0,∞))(\lfloor\ \rfloor)_* : \mathcal{M}^{\mathbb{N}\cup\{0\}} \to\operatorname{et}(\mathcal{M}^{[0,\infty)}) be the functor induced by the floor function ⌊ ⌋:[0,∞)→N∪{0}\lfloor\ \rfloor: [0,\infty) \to\mathbb{N}\cup\{0\}. Observe that the image of the functor (⌊ ⌋)∗(\lfloor\ \rfloor)_* is in et⁡(M[0,∞))\operatorname{et}(\mathcal{M}^{[0,\infty)}). Thus, we have (⌊ ⌋)∗=i∘(⌊ ⌋)∗′(\lfloor\ \rfloor)_* = i \circ(\lfloor\ \rfloor)_*^{\prime}, where (⌊ ⌋)∗′:MN∪{0}→et⁡(M[0,∞))(\lfloor\ \rfloor)_*^{\prime} : \mathcal{M}^{\mathbb{N}\cup\{0\}} \to\operatorname{et}(\mathcal{M}^{[0,\infty)}) is the corestriction.

Proposition A.1. For objects FF and GG in the image of the functor (⌊ ⌋)∗(\lfloor\ \rfloor)_*, one has

dIHCet(F,G)=dIHC(iF,iG),d_{\mathrm{IHC}}^{\mathrm{et}}(F,G)=d_{\mathrm{IHC}}(iF,iG),

where dIHCetd_{\mathrm{IHC}}^{\mathrm{et}} denotes the interleaving distance in the homotopy category of et⁡(M[0,∞))\operatorname{et}(\mathcal{M}^{[0,\infty)}).

Proof. Let res:M[0,∞)→et⁡(M[0,∞))res : \mathcal{M}^{[0,\infty)} \to\operatorname{et}(\mathcal{M}^{[0,\infty)}) be the composite of (⌊ ⌋)∗(\lfloor\ \rfloor)_* and functor induced by the inclusion ι:N∪{0}→[0,∞)\iota: \mathbb{N}\cup\{0\} \to[0,\infty) of posets. We observe that res iF=Fres\ iF = F for the object in the image of the functor (⌊ ⌋)∗(\lfloor\ \rfloor)_*. The functors ii and resres preserve weak equivalences. Then, we have the right derived functors Ri:Ho⁡(et⁡(M[0,∞)))→Ho⁡(M[0,∞))Ri : \operatorname{Ho}(\operatorname{et}(\mathcal{M}^{[0,\infty)})) \to\operatorname{Ho}(\mathcal{M}^{[0,\infty)}) and Rres:Ho⁡(M[0,∞))→Ho⁡(et⁡(M[0,∞)))Rres : \operatorname{Ho}(\mathcal{M}^{[0,\infty)}) \to\operatorname{Ho}(\operatorname{et}(\mathcal{M}^{[0,\infty)})) between homotopy categories. Since each object in M[0,∞)\mathcal{M}^{[0,\infty)} and et⁡(M[0,∞))\operatorname{et}(\mathcal{M}^{[0,\infty)}) is fibrant, it follows from [12 Proposition 9.3] that RiF≅iFRiF \cong iF and Rres G≅GRres\ G \cong G for F∈et⁡(M[0,∞))F \in\operatorname{et}(\mathcal{M}^{[0,\infty)}) and G∈M[0,∞)G \in\mathcal{M}^{[0,\infty)}. Thus, we have

dIHCet(F,G)≥dIHC(RiF,RiG)=dIHC(iF,iG)≥dIHCet(Rres iF,Rres iG)=dIHCet(res iF,res iG)=dIHCet(F,G).\begin{aligned} d_{\mathrm{IHC}}^{\mathrm{et}}(F,G) \ge d_{\mathrm{IHC}}(RiF,RiG) = d_{\mathrm{IHC}}(iF,iG) \\ &\ge d_{\mathrm{IHC}}^{\mathrm{et}}(Rres\ iF,Rres\ iG) = d_{\mathrm{IHC}}^{\mathrm{et}}(res\ iF,res\ iG) \\ &= d_{\mathrm{IHC}}^{\mathrm{et}}(F,G). \end{aligned}

This completes the proof □

Corollary A.2. For generalized metric spaces XX and YY whose distances take values in non-negative integers, one has

dIHCet(N(X),N(Y))=dIHC(N(X),N(Y)),d_{\mathrm{IHC}}^{\mathrm{et}}(N(X),N(Y)) = d_{\mathrm{IHC}}(N(X),N(Y)),
dIHCet(KN(X),KN(Y))=dIHC(KN(X),KN(Y)),d_{\mathrm{IHC}}^{\mathrm{et}}(\mathrm{KN}(X),\mathrm{KN}(Y)) = d_{\mathrm{IHC}}(\mathrm{KN}(X),\mathrm{KN}(Y)),
dIHCet(KN(X),KN(Y))≤dIHCet(N(X),N(Y)) andd_{\mathrm{IHC}}^{\mathrm{et}}(\mathrm{KN}(X),\mathrm{KN}(Y)) \le d_{\mathrm{IHC}}^{\mathrm{et}}(N(X),N(Y)) \text{ and}
dIHC(KN(X),KN(Y))≤dIHC(N(X),N(Y)).d_{\mathrm{IHC}}(\mathrm{KN}(X),\mathrm{KN}(Y)) \le d_{\mathrm{IHC}}(N(X),N(Y)).

Proof. With the same notation as in the proof of Proposition A.1, we see that N(X)=res iN(X)N(X) = res\ iN(X) in et⁡(sSet[0,∞))\operatorname{et}(\mathrm{sSet}^{[0,\infty)}) and KN(X)=res iKN(X)\mathrm{KN}(X) = res\ i\mathrm{KN}(X) in et⁡(ChK[0,∞))\operatorname{et}(\mathrm{Ch}_{K}^{[0,\infty)}). The equalities follow from Proposition A.1. By virtue of [4 Proposition 3.6], we have the inequalities. □

We further relate the distance dIHC(KN(X),KN(Y))d_{\mathrm{IHC}}(\mathrm{KN}(X),\mathrm{KN}(Y)) to the cohomology interleaving distance dCohId_{\mathrm{CohI}} defined in [23 Definition 3.1]. Write dIHC(F,G)Cd_{\mathrm{IHC}}(F,G)_{\mathcal{C}} for the interleaving distance in the homotopy category between objects FF and GG in the category C=ChK[0,∞)\mathcal{C} = \mathrm{Ch}_{K}^{[0,\infty)} or ChKR\mathrm{Ch}_{K}^{\mathbb{R}}. Moreover, we denote by dI(H,K)Dd_{\mathrm{I}}(H,K)_{\mathcal{D}} the interleaving distance between objects HH and KK in D=ModK[0,∞)\mathcal{D} = \mathrm{Mod}_{K}^{[0,\infty)} or D=ModKR\mathcal{D} = \mathrm{Mod}_{K}^{\mathbb{R}}, the category of persistence modules over KK. In particular, for XX and YY in Corollary A.2, the distance dIHC(KN(X),KN(X))ChK[0,∞)d_{\mathrm{IHC} }(\mathrm{KN}(X),\mathrm{KN}(X))_{\mathrm{Ch}_{K}^{[0,\infty)}} is nothing but dIHC(KN(X),KN(Y))d_{\mathrm{IHC}}(\mathrm{KN}(X),\mathrm{KN}(Y)) with the notation in § 3.1 and Corollary A.2.

Proposition A.3. Let K\mathbb{K} be a field. For objects FF and GG in ChK[0,∞)\mathrm{Ch}_{\mathbb{K}}^{[0,\infty)}, one has

dIHC(F,G)ChK[0,∞)=dChoI(F,G):=sup⁡{dI(Hn(F),Hn(G))ModK[0,∞)∣n∈Z}.d_{\mathrm{IHC}}(F,G)_{\mathrm{Ch}_{\mathbb{K}}^{[0,\infty)}}=d_{\mathrm{ChoI}}(F,G):=\sup\{d_{\mathrm{I}}(H_n(F),H_n(G))_{\mathrm{Mod}_{\mathbb{K}}^{[0,\infty)}}\mid n\in\mathbb{Z}\}.

where HnH_n is the nnth homology functor from ChK\mathrm{Ch}_{\mathbb{K}} to ModK\mathrm{Mod}_{\mathbb{K}} the category of K\mathbb{K}-modules.

Proof. Consider the maps [0,∞)→jR→k[0,∞)[0,\infty)\xrightarrow{j}\mathbb{R}\xrightarrow{k}[0,\infty) defined by the inclusion and k(x)=0k(x)=0 for x≤0x\leq0 and k(x)=xk(x)=x for x≥0x\geq0. Then, the maps induce functors

D[0,∞)←j∗DR←k∗D[0,∞)\mathcal{D}^{[0,\infty)}\xleftarrow{j^*}\mathcal{D}^{\mathbb{R}}\xleftarrow{k^*}\mathcal{D}^{[0,\infty)}

with j∗k∗=id⁡j^*k^*=\operatorname{id} for D=ChK\mathcal{D}=\mathrm{Ch}_{\mathbb{K}} and ModK\mathrm{Mod}_{\mathbb{K}}. We have a sequence of equalities

dIHC(F,G)ChK[0,∞)=dIHC(k∗F,k∗G)ChKRd_{\mathrm{IHC}}(F,G)_{\mathrm{Ch}_{\mathbb{K}}^{[0,\infty)}}=d_{\mathrm{IHC}}(k^*F,k^*G)_{\mathrm{Ch}_{\mathbb{K}}^{\mathbb{R}}}

(by the same argument as in the proof of Proposition A.1 with the right derived functors of k∗k^* and j∗j^*)

=sup⁡{dI(Hn(k∗F),Hn(k∗G))ModKR∣n∈Z}=\sup\{d_{\mathrm{I}}(H_n(k^*F),H_n(k^*G))_{\mathrm{Mod}_{\mathbb{K}}^{\mathbb{R}}}\mid n\in\mathbb{Z}\}

(by [23 Theorem 3.3])

=sup⁡{dI(k∗Hn(F),k∗Hn(G))ModKR∣n∈Z}=\sup\{d_{\mathrm{I}}(k^*H_n(F),k^*H_n(G))_{\mathrm{Mod}_{\mathbb{K}}^{\mathbb{R}}\mid n\in\mathbb{Z}\}}

(by the commutativity of k∗k^* and HlH_l)

=sup⁡{dI(Hn(F),Hn(G))ModK[0,∞)∣n∈Z}=\sup\{d_{\mathrm{I}}(H_n(F),H_n(G))_{\mathrm{Mod}_{\mathbb{K}}^{[0,\infty)}\mid n\in\mathbb{Z}\}}

(by j∗k∗=id⁡j^*k^*=\operatorname{id})

=dChoI(F,G).=d_{\mathrm{ChoI}}(F,G).

We need the assumption that K\mathbb{K} is a field to apply [23 Theorem 3.3] in the argument above. The fact that the polynomial algebra K[t]\mathbb{K}[t] is a graded principal ideal domain is the key to proving the theorem; see [23 Lemmas 3.9 and 3.10] for the details. □

Remark A.4. Suppose that the underlying ring K\mathbb{K} is a field. Then, we have

dIHCet(KN(X),KN(Y))=dIHC(KN(X),KN(Y))=dChoI(KN(X),KN(Y))d_{\mathrm{IHC}}^{\mathrm{et}}(\mathbb{K}N(X),\mathbb{K}N(Y))=d_{\mathrm{IHC}}(\mathbb{K}N(X),\mathbb{K}N(Y))=d_{\mathrm{ChoI}}(\mathbb{K}N(X),\mathbb{K}N(Y))

for generalized metric spaces XX and YY whose distances take values in non-negative integers. This follows from Corollary A.2 and Proposition A.3.

The interleaving distance of persistence K\mathbb{K}-module HH and KK, which are of finite dimension pointwise, is determined by the bottleneck distance between HH and KK via the interval module decompositions of the modules; see, for example, [4 Theorem 4.16]. Thus, in particular, the interleaving distance dIHCet(KN(X),KN(pt))d_{\mathrm{IHC}}^{\mathrm{et}}(\mathbb{K}N(X),\mathbb{K}N(pt)) in Theorem 5.1 may be computed with the bottleneck distance between Hn(N(X);K)H_n(N(X);\mathbb{K}) and the constant persistence module K\mathbb{K} for each nn.

References

Department of Mathematical Sciences, Faculty of Science, Shinshu University, Matsumoto, Nagano 390-8621, Japan

Email address: [email protected]

References

  1. [1]Y. Asao, Magnitude homology and path homology, Bull. Lond. Math. Soc. 55 (2023), 375–398.arxiv.org/abs/2201.08047
  2. [2]Y. Asao, Magnitude and magnitude homology of filtered set enriched categories, preprint 2019. https://arxiv.org/abs/2303.05677v2.
  3. [3]M. Barr, Acyclic models, CRM Monograph Series. 17. Providence, RI: American Mathematical Society (AMS), 2002.
  4. [4]P. Bubenik and J.A. Scott, Categorification of persistent homology, Discrete Comput. Geom. 51 (2014), 600–627.
  5. [5]F. Chazal, D. Cohen-Steiner, M. Glisse, L.J. Guibas and S.Y. Oudot, Proximity of persistent modules and their diagrams, In Proceedings of the twenty-fifth annual symposium on Computational geometry (2009), 237–246.DOI
  6. [6]A.J. Blumberg and M. Lesnick, University of the homotopy interleaving distance, Trans. Amer. Math. Soc. 376 (2023), 8269–8307.arxiv.org/abs/1705.01690
  7. [7]H. Cartan and S. Eilenberg, Homological Algebra, Princeton University Press, Princeton, NJ, 1956.
  8. [8]W. Chachólski, B. Giunti and C. Landi, Invariants for tame parametrised chain complexes, Homology, Homotopy and Applications, 23 (2021), 183–213.
  9. [9]S. Cho, Quantales, persistence, and magnitude homology, preprint 2019. https://arxiv.org/abs/1910.02905.
  10. [11]S. Di, S.O. Ivanov, L. Mukoseev and M. Zhang, On the path homology of Cayley digraphs and covering digraphs. J. Algebra, (2024), 156–199.
  11. [12]W.G. Dwyer and J. Spaliński, Homotopy theories and model categories. In Handbook of algebraic topology, pages 73–126. North-Holland, Amsterdam, 1995.DOI
  12. [13]D, Govc and R. Hepworth, Persistent magnitude, Journal of Pure and Applied Algebra, 225 (2021) 106517.DOI
  13. [14]A.A. Grigor’yan, Y. Lin, Y.V. Muranov and S.-T. Yau, Homologies of path complexes and digraphs, 2012. https://arxiv.org/abs/1207.2834.
  14. [15]R. Hepworth, Magnitude cohomology, Math. Z. 301 (2022), 3617–3640.arxiv.org/abs/1807.06832
  15. [16]R. Hepworth and E. Roff, The reachability homology of a directed graph. Int. Math. Res. Not. IMRN: Paper No. rnae280, 18, 2025.
  16. [17]R. Hepworth and E. Roff, Bigraded path homology and the magnitude-path spectral sequence, preprint, 2026. arXiv:2404.06689v3.
  17. [18]R. Hepworth and S. Willerton, Categorifying the magnitude of a graph, Homology Homotopy Appl., 19 (2017), 31–60.DOI
  18. [19]P. Hirschhorn, Model Categories and Their Localizations, AMS Math. Survey and Monographs 99 (2002).
  19. [21]S.O. Ivanov and X. Xu, On ℓp-Vietoris-Rips complexes, J. Appl. Comput. Topol. 10 (2026), 46 p.
  20. [22]K. Kuribayashi, T. Naito, K. Sekizuka, S. Wakatsuki and T. Yamaguchi, A distance between maps via interleavings of relative Sullivan algebras, preprint, 2026. https://arxiv.org/abs/2604.06800.
  21. [23]K. Kuribayashi, T. Naito, S. Wakatsuki and T. Yamaguchi, The equalities of interleaving distances and cohomology interleavings of spaces over BS¹, J. Comput. Appl. Math. 10 (2026), 46 pages. https://doi.org/10.1007/s41468-026-00252-0.
  22. [24]K. Kuribayashi, K. Sakai and Y. Shiobara, Towards Riemannian diffeology, Proc. R. Soc. Edinb., Sect. A, Math. (2025) https://doi.org/10.1017/prm.2025.10114.
  23. [25]E. Lanari and L. Scoccola, Rectification of interleavings and a persistent Whitehead theorem Algebraic & Geometric Topology 23 (2023), 803–832.
  24. [26]F.W. Lawvere, Metric spaces, generalized logic, and closed categories, Rend. Sem. Mat. fis. Milano 43 (1974), 135–166. Reprinted as Repr. Theory Appl. Categ. 1 (2002), 1–37.
  25. [27]T. Leinster and M. Shulman, Online discussion at the n-Category Café. https://golem.ph.utexas.edu/ncatlab.org/nlab/show/Category/2016/09/magnitude_homology.html.
  26. [28]T. Leinster and M. Shulman, Magnitude homology of enriched categories and metric spaces, Algebr. Geom. Topol., 21 (2021), 2175–2221.
  27. [29]M. Livernet, E. Roff and S. Whitehouse, Homotopy theories via the magnitude-path spectral sequence, preprint, 2026, https://arxiv.org/abs/2606.09747.
  28. [30]J. McCleary, User’s guide to spectral sequences, Second Edition, Cambridge Studies in Advanced Mathematics 58. Cambridge: Cambridge University Press, (2001).DOI
  29. [31]L. Melodia, Magnitude homology is the associated graded of the length filtration, preprint 2026. https://arxiv.org/abs/2608.21479
  30. [32]N. Otter, Magnitude meets persistence: homology theories for filtered simplicial sets, Homology Homotopy Appl. 24 (2022), 365–387.DOI
  31. [33]E. Riehl, Categorical Homotopy Theory, New Mathematical Monographs 24, Cambridge: Cambridge University Press, (2014).DOI

Paper details

vlnc:W_3md7n9yp
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents