Introduction
In this paper three key ingredients come together. Chronologically, the first is the foundational work by R. Franzosa on connection matrices in the setting of dynamical systems, v. [9 , 11 ] . This provides an algebraic theory for Morse representations with information about connecting orbits and bifurcations in flows. The second is the work by J. Robbins & D. Salamon on connection matrix theory for discrete dynamical systems, v. [19 ] . In this work they utilize the novel observation that the attractors in a dynamical system display a lattice structure. The third is the introduction of spectral systems for filtered chain complexes by B. Matschke, which allow filterings over any poset, v. [16 ] . In this paper we introduce spectral systems in an abstract setting beyond filtered chain complexes. This abstract setting is applied to the theory of Morse representations and extends the approach via spectral sequences by P. Bartłomiejczyk, v. [1 , 2 ] ; cf. [6 ] .
The decomposition of a dynamical system ϕ \phi ϕ into a hierarchical structure of invariant sets is a cornerstone of the global analysis of dynamical systems.1 Central to this decomposition is the concept of Morse representation , defined as a finite partially ordered set consisting of pairwise disjoint, compact, non-empty invariant sets — Morse sets — outside of which ϕ \phi ϕ has gradient-like behavior, v. Sect. 2 and [5 , 11 , 15 , 13 ] . Formally, a Morse representation is equivalent to a finite sublattice of attractors A \mathcal{A} A of ϕ \phi ϕ . To each Morse set M M M , i.e. the intersection of an attractor A ∈ A A \in\mathcal{A} A ∈ A and the dual repeller A ′ ∗ A'^* A ′ ∗ of an attractor A ′ ∈ A A' \in\mathcal{A} A ′ ∈ A , with A ′ ⊆ A A' \subseteq A A ′ ⊆ A , we assign the Conley index H C ( M ) \mathrm{HC}(M) HC ( M ) , an algebraic-topological invariant defined as the relative homology H ( U , U ′ ) H(U,U') H ( U , U ′ ) , where U , U ′ U,U' U , U ′ , with U ′ ⊆ U U' \subseteq U U ′ ⊆ U , are trapping regions for A A A and A ′ A' A ′ respectively. The interplay between the Conley indices is captured by a Cartan–Eilenberg system E ( M ) \mathrm{E}(M) E ( M ) (or E-system ), v. [21 , 12 ] . This system provides an exhaustive algebraic framework consisting of R R R -modules H C ( M ) = H ( U , U ′ ) \mathrm{HC}(M)=H(U,U') HC ( M ) = H ( U , U ′ ) for every pair ( A , A ′ ) ∈ A × A (A,A')\in\mathcal{A}\times\mathcal{A} ( A , A ′ ) ∈ A × A , with A ′ ⊆ A A'\subseteq A A ′ ⊆ A . Within this system, every nested pair of attractors ( A , A ′ ) (A,A') ( A , A ′ ) yields an exact triangle of the form:
⋯ ⟶ H ( U ′ ) → i H ( U ) → j H ( U , U ′ ) → k H ( U ′ ) ⟶ ⋯ (1) \cdots\longrightarrow H(U') \xrightarrow{i} H(U) \xrightarrow{j} H(U,U') \xrightarrow{k} H(U') \longrightarrow\cdots \tag*{(1)} ⋯ ⟶ H ( U ′ ) i H ( U ) j H ( U , U ′ ) k H ( U ′ ) ⟶ ⋯ (1)
An E-system serves as a maximal system of invariants , encoding the totality of algebraic (homological) data regarding Morse sets and the connecting orbits between them through a network of commutative squares and exact triangles, v. Sect. 3. In [21 ] it is shown that E -systems for a Morse representation are equivalent to Franzosa’s homology braids.
An equivalent formulation of this information is via a spectral system 2 S ( M ) \mathrm{S}(M) S ( M ) (or S-system ), which represents the data as a collection of R R R -modules and differentials. If ( Q , ≤ ) (Q,\leq) ( Q , ≤ ) is a poset and Q n Q_n Q n is the poset of ordered n n n -tuples α 1 ≤ ⋯ ≤ α n \alpha_1\leq\cdots\leq\alpha_n α 1 ≤ ⋯ ≤ α n in Q Q Q , ordered componentwise and regarded as a thin category, then a spectral system S \mathrm{S} S is defined by a functor S : Q 4 → Mod R \mathrm{S}:Q_4\to\operatorname{Mod}_R S : Q 4 → Mod R , ( α , β , γ , δ ) ↦ S α β γ δ (\alpha,\beta,\gamma,\delta)\mapsto\mathrm{S}^{\gamma\delta}_{\alpha\beta} ( α , β , γ , δ ) ↦ S α β γ δ , together with differentials d : S η ζ γ δ → S α β γ δ d:\mathrm{S}^{\gamma\delta}_{\eta\zeta}\to\mathrm{S}^{\gamma\delta}_{\alpha\beta} d : S η ζ γ δ → S α β γ δ , ( α , β , γ , δ , ϵ , ζ ) ∈ Q 6 (\alpha,\beta,\gamma,\delta,\epsilon,\zeta)\in Q_6 ( α , β , γ , δ , ϵ , ζ ) ∈ Q 6 , which satisfy a list of axioms, v. Defn. 4.1 . In our setting Q = O ( M ) Q=\mathrm{O}(M) Q = O ( M ) , the lattice of down-sets in ( M , ≤ ) (M,\leq) ( M , ≤ ) , and
S α β γ δ = ker : E β γ → k E α β im : E γ δ → k E β γ , ∀ ( α , β , γ , δ ) ∈ O ( M ) 4 , \mathrm{S}^{\gamma\delta}_{\alpha\beta}=
\frac{\ker:E^\gamma_\beta\xrightarrow{k}E^\beta_\alpha}{\operatorname{im}:E^\delta_\gamma\xrightarrow{k}E^\gamma_\beta},
\qquad\forall(\alpha,\beta,\gamma,\delta)\in\mathrm{O}(M)_4, S α β γ δ = im : E γ δ k E β γ ker : E β γ k E α β , ∀ ( α , β , γ , δ ) ∈ O ( M ) 4 ,
where E α β = H ( F β U , F α U ) E^\beta_\alpha=H(F_\beta U,F_\alpha U) E α β = H ( F β U , F α U ) are the E -terms for E ( M ) \mathrm{E}(M) E ( M ) — Conley indices — and the sets F α U F_\alpha U F α U define a lattice U U U of trapping regions isomorphic to A A A , v. Sect. 2. The differentials k : E β γ → E α β k:E^\gamma_\beta\to E^\beta_\alpha k : E β γ → E α β are the connecting homomorphisms of E ( M ) \mathrm{E}(M) E ( M ) . Spectral systems indexed by a finite distributive lattice, such as O ( M ) \mathrm{O}(M) O ( M ) , will be called lattice-ordered .
The spectral net approach
In the tradition of Franzosa, chain-generated E -systems, v. Sect. 3, can be encoded by an M -graded differential module ( C , d ) (C,d) ( C , d ) , also referred to as a connection matrix (or spectral matrix in the context of Cartan–Eilenberg systems). We emphasize that a connection matrix is not a reduction of the data but an alternative, highly structured encoding of the system. Knowing ( C , d ) (C,d) ( C , d ) determines all E E E -terms, S S S -terms, and differentials. Connection matrices are not unique in general. The existence of connection matrices is proved by Franzosa [11 ] , but not over arbitrary rings. For the extension to arbitrary rings, v. [23 ] .
Because computing the full connection matrix requires global, chain-level information, which is often not available, we introduce the concept of a spectral net . Analogous to spectral sequences in homological algebra, spectral nets provide a framework to deduce structural information about E E E -systems and S S S -systems from partial, localized data by exploiting the underlying spectral structure.
Spectral systems offer several key advantages over traditional connection matrices. First, they operate over an arbitrary ring R R R , providing broad algebraic generality. Second, their differentials are uniquely determined by the E E E -system, which makes them robust under perturbations of ϕ \phi ϕ , v. [10 , 8 ] , and non-trivial differentials carry direct information about the underlying connecting orbit structure, v. Theorems A and B below. In contrast, connection matrix entries are generally non-unique and lack this straightforward geometric interpretation. Finally, spectral systems make information more accessible: while computing a connection matrix requires complete, global knowledge of the entire system up front, spectral systems can be analyzed using localized or partial data via spectral nets. To exploit this accessibility, spectral nets enable an iterative computational approach. Rather than relying on a complete system description, spectral nets incrementally recover information about the spectral system — higher S S S -terms and higher differentials — step-by-step from limited initial data: the E E E -terms E α β E_{\alpha}^{\beta} E α β with β ∖ α \beta\setminus\alpha β ∖ α a singleton, and the first-order differentials between them.
An example
We consider an E E E -system over the lattice of down-sets O ( P ) \mathcal{O}(P) O ( P ) of the poset P = N 5 P = N_{5} P = N 5 . Figure 1.1 gives the Hasse diagrams of both N 5 N_{5} N 5 and O ( N 5 ) \mathcal{O}(N_{5}) O ( N 5 ) . The
Figure 1.1. The poset P = N 5 = { 1 , 2 , 3 , 4 , 5 } P = N_{5} = \{1,2,3,4,5\} P = N 5 = { 1 , 2 , 3 , 4 , 5 } with covering relations 1 < 2 < 4 < 5 1 < 2 < 4 < 5 1 < 2 < 4 < 5 and 1 < 3 < 5 1 < 3 < 5 1 < 3 < 5 [left] and its lattice of down-sets O ( P ) \mathcal{O}(P) O ( P ) [right].
lattice O ( N 5 ) \mathcal{O}(N_{5}) O ( N 5 ) has 3 maximal chains with 3 associated linear extensions of N 5 N_{5} N 5 :
K 1 : 0 ⊆ α ⊆ β ⊆ ϵ ⊆ ζ ⊆ 1 , ⟹ 1 < 2 < 4 < 3 < 5 ; K 2 : 0 ⊆ α ⊆ β ⊆ δ ⊆ ζ ⊆ 1 , ⟹ 1 < 2 < 3 < 4 < 5 ; K 3 : 0 ⊆ α ⊆ γ ⊆ δ ⊆ ζ ⊆ 1 , ⟹ 1 < 3 < 2 < 4 < 5. \begin{aligned}
K_{1}:&\quad0 \subseteq\alpha\subseteq\beta\subseteq\epsilon\subseteq\zeta\subseteq1,\quad\Longrightarrow\quad1<2<4<3<5;\\
K_{2}:&\quad0 \subseteq\alpha\subseteq\beta\subseteq\delta\subseteq\zeta\subseteq1,\quad\Longrightarrow\quad1<2<3<4<5;\\
K_{3}:&\quad0 \subseteq\alpha\subseteq\gamma\subseteq\delta\subseteq\zeta\subseteq1,\quad\Longrightarrow\quad1<3<2<4<5.
\end{aligned} K 1 : K 2 : K 3 : 0 ⊆ α ⊆ β ⊆ ϵ ⊆ ζ ⊆ 1 , ⟹ 1 < 2 < 4 < 3 < 5 ; 0 ⊆ α ⊆ β ⊆ δ ⊆ ζ ⊆ 1 , ⟹ 1 < 2 < 3 < 4 < 5 ; 0 ⊆ α ⊆ γ ⊆ δ ⊆ ζ ⊆ 1 , ⟹ 1 < 3 < 2 < 4 < 5.
From the structure of the Cartan–Eilenberg system we define page-1 of the spectral net by considering E E E -terms corresponding to consecutive down-sets in the maximal chains. We assume that the E E E -system is excisive , in particular E α β ≅ E α ′ β ′ E_{\alpha}^{\beta} \cong E_{\alpha'}^{\beta'} E α β ≅ E α ′ β ′ for all β \ α = β ′ \ α ′ \beta\backslash\alpha=\beta'\backslash\alpha' β \ α = β ′ \ α ′ .
0 ⟶ E ζ 1 → d 1 ζ ε E ε ζ → d ε β E β ε → d β α E α β → d β α 0 E 0 α ⟶ 0 ↗ d 1 ζ ε ↙ d δ β α ↘ d β α 0 E δ ζ → d δ ε E β δ ↘ d δ γ E γ δ → d δ γ α E α γ → d γ α 0 E 0 α \begin{aligned}
0 &\longrightarrow E^1_{\zeta} \xrightarrow{d_{1\zeta\varepsilon}} E^\zeta_{\varepsilon} \xrightarrow{d_{\varepsilon\beta}} E^\varepsilon_{\beta} \xrightarrow{d_{\beta\alpha}} E^\beta_{\alpha} \xrightarrow{d_{\beta\alpha0}} E^\alpha_{0} \longrightarrow0 \\
&\qquad\qquad\quad\nearrow d_{1\zeta\varepsilon}\qquad\qquad\qquad\swarrow d_{\delta\beta\alpha}\qquad\qquad\qquad\searrow d_{\beta\alpha0}\\
&\qquad\qquad\qquad E^\zeta_{\delta} \xrightarrow{d_{\delta\varepsilon}} E^\delta_{\beta} \qquad\qquad\qquad\qquad\qquad\qquad\qquad\\
&\qquad\qquad\qquad\qquad\qquad\qquad\qquad\searrow d_{\delta\gamma}\qquad\qquad\qquad\qquad\qquad\qquad\\
&\qquad\qquad\qquad\qquad\qquad\qquad\qquad E^\delta_{\gamma} \xrightarrow{d_{\delta\gamma\alpha}} E^\gamma_{\alpha} \xrightarrow{d_{\gamma\alpha0}} E^\alpha_{0}
\end{aligned} 0 ⟶ E ζ 1 d 1 ζ ε E ε ζ d εβ E β ε d β α E α β d β α 0 E 0 α ⟶ 0 ↗ d 1 ζ ε ↙ d δ β α ↘ d β α 0 E δ ζ d δ ε E β δ ↘ d δ γ E γ δ d δ γ α E α γ d γ α 0 E 0 α
If we display each maximal chain we obtain the following diagram:
0 ⟶ E ζ 1 → d 1 ζ δ E δ ζ → d δ ε E β δ → d δ β α E α β → d β α 0 E 0 α ⟶ 0 0 ⟶ E ζ 1 → d 1 ζ δ E δ ζ → d δ β E β δ → d δ β α E α β → d β α 0 E 0 α ⟶ 0 0 ⟶ E ζ 1 → d 1 ζ δ E δ ζ → d δ γ E γ δ → d δ γ α E α γ → d γ α 0 E 0 α ⟶ 0 \begin{aligned}
0 &\longrightarrow E^1_{\zeta} \xrightarrow{d_{1\zeta\delta}} E^\zeta_{\delta} \xrightarrow{d_{\delta\varepsilon}} E^\delta_{\beta} \xrightarrow{d_{\delta\beta\alpha}} E^\beta_{\alpha} \xrightarrow{d_{\beta\alpha0}} E^\alpha_{0} \longrightarrow0\\
0 &\longrightarrow E^1_{\zeta} \xrightarrow{d_{1\zeta\delta}} E^\zeta_{\delta} \xrightarrow{d_{\delta\beta}} E^\delta_{\beta} \xrightarrow{d_{\delta\beta\alpha}} E^\beta_{\alpha} \xrightarrow{d_{\beta\alpha0}} E^\alpha_{0} \longrightarrow0\\
0 &\longrightarrow E^1_{\zeta} \xrightarrow{d_{1\zeta\delta}} E^\zeta_{\delta} \xrightarrow{d_{\delta\gamma}} E^\delta_{\gamma} \xrightarrow{d_{\delta\gamma\alpha}} E^\gamma_{\alpha} \xrightarrow{d_{\gamma\alpha0}} E^\alpha_{0} \longrightarrow0
\end{aligned} 0 0 0 ⟶ E ζ 1 d 1 ζ δ E δ ζ d δ ε E β δ d δ β α E α β d β α 0 E 0 α ⟶ 0 ⟶ E ζ 1 d 1 ζ δ E δ ζ d δ β E β δ d δ β α E α β d β α 0 E 0 α ⟶ 0 ⟶ E ζ 1 d 1 ζ δ E δ ζ d δ γ E γ δ d δ γ α E α γ d γ α 0 E 0 α ⟶ 0
where the dotted arrows indicate the isomorphisms between E E E -terms due to the excisive property. The latter diagram makes it easier to display the higher pages of the spectral net. By the excisive property we choose R R R -modules G p C G_pC G p C : G 1 C ≅ E 0 α G_1C \cong E^\alpha_0 G 1 C ≅ E 0 α , G 2 C ≅ E α β ≅ E β δ G_2C \cong E^\beta_\alpha\cong E^\delta_\beta G 2 C ≅ E α β ≅ E β δ , G 3 C ≅ E ε ζ ≅ E γ δ ≅ E α γ G_3C \cong E^\zeta_\varepsilon\cong E^\delta_\gamma\cong E^\gamma_\alpha G 3 C ≅ E ε ζ ≅ E γ δ ≅ E α γ , G 4 C ≅ E δ ζ ≅ E β ε G_4C \cong E^\zeta_\delta\cong E^\varepsilon_\beta G 4 C ≅ E δ ζ ≅ E β ε , and G 5 C ≅ E ζ 1 G_5C \cong E^1_\zeta G 5 C ≅ E ζ 1 , whose identifications are indicated with dotted arrows in the diagram above. If we identify the isomorphic E E E -terms with G 1 C , … , G 5 C G_1C,\ldots,G_5C G 1 C , … , G 5 C , then we obtain the following diagram:
0 ⟶ G 5 C → d 45 1 G 4 C → 0 G 3 C → d 24 1 G 2 C → d 12 1 G 1 C ⟶ 0 ↗ d 35 1 ↘ d 13 1 G 3 C G 4 C → 0 G 3 C G 2 C → 0 G 3 C (2) \begin{aligned}
0 &\longrightarrow G_5C \xrightarrow{d^1_{45}} G_4C \xrightarrow{0} G_3C \xrightarrow{d^1_{24}} G_2C \xrightarrow{d^1_{12}} G_1C \longrightarrow0\\
&\qquad\qquad\qquad\qquad\qquad\qquad\qquad\quad\nearrow d^1_{35}\qquad\qquad\qquad\searrow d^1_{13}\\
&\qquad\qquad\qquad\qquad\qquad\qquad\qquad G_3C \qquad\qquad\qquad\qquad\qquad\\
&\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\\
&\qquad\qquad\qquad G_4C \xrightarrow{0} G_3C \qquad\qquad\qquad G_2C \xrightarrow{0} G_3C \tag*{(2)}
\end{aligned} 0 ⟶ G 5 C d 45 1 G 4 C 0 G 3 C d 24 1 G 2 C d 12 1 G 1 C ⟶ 0 ↗ d 35 1 ↘ d 13 1 G 3 C G 4 C 0 G 3 C G 2 C 0 G 3 C (2)
Repeated application of homology yields the “higher” S S S -terms and associated differentials. The following diagram displays page-2:
0 ⟶ S ε ζ 11 → 0 S β ε ζ 1 → d 23 2 = 0 S α β ε ζ → 0 S 0 α β ε ⟶ 0 ↘ d 45 2 ↘ d 24 2 ↘ d 14 2 0 ⟶ S δ ζ 11 → 0 S β δ δ 1 → d 34 2 S α β δ ζ → 0 S 0 α β δ ⟶ 0 ↘ d 35 2 ↘ d 34 2 = 0 ↘ d 13 2 0 ⟶ S δ ζ 11 → 0 S γ δ δ 1 → d 34 2 = 0 S α γ δ γ → 0 S 0 α γ δ ⟶ 0 \begin{aligned}
0 &\longrightarrow S^{11}_{\varepsilon\zeta} \xrightarrow{0} S^{\zeta1}_{\beta\varepsilon} \xrightarrow{d^2_{23}=0} S^{\varepsilon\zeta}_{\alpha\beta} \xrightarrow{0} S^{\beta\varepsilon}_{0\alpha} \longrightarrow0\\
&\qquad\qquad\qquad\searrow d^2_{45}\qquad\qquad\qquad\searrow d^2_{24}\qquad\qquad\qquad\searrow d^2_{14}\\
0 &\longrightarrow S^{11}_{\delta\zeta} \xrightarrow{0} S^{\delta1}_{\beta\delta} \xrightarrow{d^2_{34}} S^{\delta\zeta}_{\alpha\beta} \xrightarrow{0} S^{\beta\delta}_{0\alpha} \longrightarrow0\\
&\qquad\qquad\qquad\searrow d^2_{35}\qquad\qquad\qquad\searrow d^2_{34}=0\qquad\qquad\qquad\searrow d^2_{13}\\
0 &\longrightarrow S^{11}_{\delta\zeta} \xrightarrow{0} S^{\delta1}_{\gamma\delta} \xrightarrow{d^2_{34}=0} S^{\delta\gamma}_{\alpha\gamma} \xrightarrow{0} S^{\gamma\delta}_{0\alpha} \longrightarrow0
\end{aligned} 0 0 0 ⟶ S εζ 11 0 S β ε ζ 1 d 23 2 = 0 S α β εζ 0 S 0 α β ε ⟶ 0 ↘ d 45 2 ↘ d 24 2 ↘ d 14 2 ⟶ S δ ζ 11 0 S β δ δ 1 d 34 2 S α β δ ζ 0 S 0 α β δ ⟶ 0 ↘ d 35 2 ↘ d 34 2 = 0 ↘ d 13 2 ⟶ S δ ζ 11 0 S γ δ δ 1 d 34 2 = 0 S α γ δ γ 0 S 0 α γ δ ⟶ 0
Spectral nets are based on partial information of the spectral system. At page-1 we start with the modules G p C G_pC G p C and differentials defined between consecutive (in the linear extensions of P P P ) G G G -terms given by the diagram in (2) . Each path of solid arrows from G 5 C G_5C G 5 C to G 1 C G_1C G 1 C in (2) represents a spectral sequence associated to a maximal chain in O ( P ) O(P) O ( P ) . The “bridges” (equal signs between G G G -terms) allow for additional paths and therefore new relations on the differentials, v. Thm. 6.1. In Figure 6.2 and Figure 6.3 additional relations are displayed. In Example 8.3 we give another example of the complexity of relations between differentials in a spectral net applied to a problem in [11 Ex. 6.2] .
Figure 6.2. The paths in red and blue correspond to the perspective pairs (i) and (ii) respectively. It indicates how the differentials are composed to give zero.
Figure 6.3. The paths in beige correspond to the perspective pair in (iii). The green boxes indicate the Morse set corresponding to the convex set { 2 , 4 } \{2,4\} { 2 , 4 } .
1.3 Main results. In Section 2 we start with the dynamical context of this paper. A semi-flow ϕ \phi ϕ on a compact topological space ( X , T ) (X,\mathcal{T}) ( X , T ) is given by a continuous function ϕ : R + × X → X \phi:\mathbb{R}^{+}\times X\to X ϕ : R + × X → X , with R + = [ 0 , ∞ ) \mathbb{R}^{+}=[0,\infty) R + = [ 0 , ∞ ) , that satisfies: (i) ϕ 0 ( x ) = x \phi^0(x)=x ϕ 0 ( x ) = x for all x ∈ X x\in X x ∈ X , (ii) ϕ t + s = ϕ t ∘ ϕ s \phi^{t+s}=\phi^t\circ\phi^s ϕ t + s = ϕ t ∘ ϕ s for all s , t ∈ R + s,t\in\mathbb{R}^{+} s , t ∈ R + (semi-group property), and (iii) ϕ t \phi^t ϕ t is proper for all t ∈ R + t\in\mathbb{R}^{+} t ∈ R + , where ϕ t : = ϕ ( t , ⋅ ) \phi^t:=\phi(t,\cdot) ϕ t := ϕ ( t , ⋅ ) , v. [13 ] . We summarize the duality between Morse representations and finite sublattices of attractors for ϕ \phi ϕ . In Section 3 we define the local invariants for Morse sets, the Conley index , and we introduce an algebraic structure that combines the “local” Conley indices into a “global” structure via a Cartan–Eilenberg system, v. [21 ] . In [9 , 11 ] the global structure is given by the equivalent structure called a homology braid,3 v. [21 ] . In Section 4 we introduce the notion of a spectral system motivated by Matschke’s notion of spectral systems for poset-filtered chain complexes, v. [16 ] . We give a new axiomatic definition of spectral system and we prove that spectral systems are equivalent to Cartan–Eilenberg systems, v. Thm. 4.7 . In Section 5 we give a formal definition of the fundamental spectral net for an excisive S S S -system. The first page consists of the E E E -terms E α β E^\beta_\alpha E α β , with β ∖ α = { p } \beta\setminus\alpha=\{p\} β ∖ α = { p } for all p ∈ P p\in P p ∈ P , and the higher pages are inductively defined by successively applying homology. In Section 1.2 we already explained the basics via an example. Other spectral nets such as subnets and nets of E E E -terms of convex sets can be defined to obtain different information about the spectral system and its differentials. In Section 6 we study properties of the differentials in an S S S -system and in particular the higher differentials. Section 6.1 starts with the notion of perspective pairs of down-sets, which provide additional relations on the first-order differentials of a spectral net besides the usual relations coming from the spectral sequences, v. Thm. 6.1. In Section 6.2 we show that first-order differentials allow derived differentials that carry information about extensions, v. Thm. 6.3 , and in Section 6.3 various properties of higher differentials are discussed, v. Thm. 6.8 and the subsequent results. In Section 7 we apply the properties of differentials to Morse representations, which yields direct information about the structure of connecting orbits. For flow-orders, v. Defn. 2.2 and [11 ] , we obtain the following result for connecting orbits, where C ( M ′ , M ) C(M',M) C ( M ′ , M ) denotes the set of connecting orbits from M ′ M' M ′ to M M M , and 0 = ∅ 0=\varnothing 0 = ∅ and 1 = M 1=M 1 = M are the bottom and top of O ( M ) O(M) O ( M ) :
Theorem A [v. Thm. 7.2] Let ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) be a flow-order. Suppose the differential
d : S β γ δ 1 → S α δ β γ \mathrm{d}:S^{\delta_1}_{\beta\gamma}\to S^{\beta\gamma}_{\alpha\delta} d : S β γ δ 1 → S α δ β γ is non-trivial , with β ∖ α = { M } \beta\setminus\alpha=\{M\} β ∖ α = { M } and δ ∖ γ = { M ′ } \delta\setminus\gamma=\{M'\} δ ∖ γ = { M ′ } . Then, there exists a directed path { M k } k = 0 n = δ ∖ α \{M_k\}_{k=0}^n=\delta\setminus\alpha { M k } k = 0 n = δ ∖ α from M ′ = M n M'=M_n M ′ = M n to M = M 0 M=M_0 M = M 0 (a path of covering relations, v. Cor. 6.12 ),
such that for all 1 ≤ k ≤ n 1 \le k \le n 1 ≤ k ≤ n either C ( M k , M k − 1 ) ≠ ∅ C(M_k,M_{k-1}) \ne\varnothing C ( M k , M k − 1 ) = ∅ , or C ( M k − 1 , M k ) ≠ ∅ C(M_{k-1},M_k) \ne\varnothing C ( M k − 1 , M k ) = ∅ , depending on whether M k > M k − 1 M_k > M_{k-1} M k > M k − 1 , or M k < M k − 1 M_k < M_{k-1} M k < M k − 1 respectively.
Other results are described in § 7 , such as Theorem 7.3 , which extend results in [11 ] and [2 ] . In contrast, the interpretation of connection matrices is less direct as their higher-order entries depend on extension data.
Theorem B [v. Thm. 7.5] Let ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) be a Morse representation and suppose the differential
d : S γ δ ξ → S α β γ δ d:S^{\xi}_{\gamma\delta} \to S^{\gamma\delta}_{\alpha\beta} d : S γ δ ξ → S α β γ δ is non-trivial, with γ ∖ β = { M } \gamma\setminus\beta= \{M\} γ ∖ β = { M } and ϵ ∖ δ = { M ′ } \epsilon\setminus\delta= \{M'\} ϵ ∖ δ = { M ′ } . Then,
(i) there exists a Morse set M ′ ′ ∈ δ ∖ β M'' \in\delta\setminus\beta M ′′ ∈ δ ∖ β , M ′ ′ < M ′ M'' < M' M ′′ < M ′ , such that C ( M ′ , M ′ ′ ) ≠ ∅ C(M',M'') \ne\varnothing C ( M ′ , M ′′ ) = ∅ ;
(ii) there exists a Morse set M ′ ′ ′ ∈ ϵ ∖ γ M''' \in\epsilon\setminus\gamma M ′′′ ∈ ϵ ∖ γ , M < M ′ ′ ′ M < M''' M < M ′′′ , such that C ( M ′ ′ ′ , M ) ≠ ∅ C(M''',M) \ne\varnothing C ( M ′′′ , M ) = ∅ .
Theorem B is a key result: in contrast to most previous results, it extracts information about connecting orbits from (higher) differentials alone, and it applies to arbitrary Morse representations, not only to flow-orders. In § 8 we illustrate the results on spectral nets and the interpretation of higher differentials by some basic examples, including the examples in [11 ] .
Morse representations
Let ϕ \phi ϕ be a semi-flow on a compact topological space X X X , v. Sect. 1.3. A trapping region U ⊆ X U \subseteq X U ⊆ X is a forward invariant attracting neighborhood, i.e. ϕ t ( U ) ⊆ U \phi^t(U) \subseteq U ϕ t ( U ) ⊆ U for all t ≥ 0 t \ge0 t ≥ 0 and ϕ τ ( cl U ) ⊆ int U \phi^\tau(\operatorname{cl} U) \subseteq\operatorname{int} U ϕ τ ( cl U ) ⊆ int U for some τ > 0 \tau> 0 τ > 0 . Every trapping region U U U defines an attractor A = ω ( U ) A = \omega(U) A = ω ( U ) , the ω \omega ω -limit set of U U U , which is non-empty and invariant for ϕ t \phi^t ϕ t if U U U is non-empty, v. [13 ] . The trapping regions ANbhd + ( ϕ ) \operatorname{ANbhd}_{+}(\phi) ANbhd + ( ϕ ) (with ∪ \cup ∪ and ∩ \cap ∩ ) and the attractors Att ( ϕ ) \operatorname{Att}(\phi) Att ( ϕ ) (with A ∨ A ′ = A ∪ A ′ A \vee A' = A \cup A' A ∨ A ′ = A ∪ A ′ and A ∧ A ′ = ω ( A ∩ A ′ ) A \wedge A' = \omega(A \cap A') A ∧ A ′ = ω ( A ∩ A ′ ) ) form bounded distributive lattices, v. [15 ] . All sublattices of ANbhd + ( ϕ ) \operatorname{ANbhd}_{+}(\phi) ANbhd + ( ϕ ) and Att ( ϕ ) \operatorname{Att}(\phi) Att ( ϕ ) are assumed to contain the bottom and the top element, i.e. ∅ \varnothing ∅ and X X X , resp. ∅ \varnothing ∅ and ω ( X ) \omega(X) ω ( X ) . The relationship between attractors and trapping regions is captured by the surjective lattice homomorphism4
ω ϕ : ANbhd + ( ϕ ) ↠ Att ( ϕ ) , U ↦ ω ϕ ( U ) = A . (3) \omega_{\phi}:\operatorname{ANbhd}_{+}(\phi)\twoheadrightarrow\operatorname{Att}(\phi),\quad U\mapsto\omega_{\phi}(U)=A. \tag*{(3)} ω ϕ : ANbhd + ( ϕ ) ↠ Att ( ϕ ) , U ↦ ω ϕ ( U ) = A . (3)
Every attractor has a well-defined dual repeller A ∗ = α ( U c ) A^*=\alpha(U^c) A ∗ = α ( U c ) , the α \alpha α -limit set of U c = X ∖ U U^c=X\setminus U U c = X ∖ U , where U U U is a trapping region with ω ( U ) = A \omega(U)=A ω ( U ) = A ; the set U c U^c U c is called a repelling region , v. [14 , 15 , 13 ] . The lattice of repellers is denoted by Rep ( ϕ ) \operatorname{Rep}(\phi) Rep ( ϕ ) . A pair of attractors A , A ′ ∈ Att ( ϕ ) A,A' \in\operatorname{Att}(\phi) A , A ′ ∈ Att ( ϕ ) defines a Morse set M = A − A ′ : = A ∩ A ′ ∗ M=A-A':=A\cap A'^* M = A − A ′ := A ∩ A ′ ∗ ,5 and a pair of trapping regions U , U ′ ∈ ANbhd + ( ϕ ) U,U' \in\operatorname{ANbhd}_{+}(\phi) U , U ′ ∈ ANbhd + ( ϕ ) defines a Morse tile T = U \ U ′ : = U ∩ U ′ c T=U\backslash U':=U\cap U'^c T = U \ U ′ := U ∩ U ′ c . This defines a homomorphism Inv : MTile ( ϕ ) ↠ Morse ( ϕ ) \operatorname{Inv}:\operatorname{MTile}(\phi)\twoheadrightarrow\operatorname{Morse}(\phi) Inv : MTile ( ϕ ) ↠ Morse ( ϕ ) , via Inv ( T ) = ω ( U ) − ω ( U ′ ) \operatorname{Inv}(T)=\omega(U)-\omega(U') Inv ( T ) = ω ( U ) − ω ( U ′ ) , with T = U \ U ′ T=U\backslash U' T = U \ U ′ , which is the maximal invariant set in T T T (so that it does not depend on the choice of U U U and U ′ U' U ′ ), and where MTile ( ϕ ) \operatorname{MTile}(\phi) MTile ( ϕ ) and Morse ( ϕ ) \operatorname{Morse}(\phi) Morse ( ϕ ) denote the semilattices of Morse tiles and Morse sets respectively, v. [15 ] .
For every finite sublattice U ⊆ ANbhd + ( ϕ ) U \subseteq\operatorname{ANbhd}_{+}(\phi) U ⊆ ANbhd + ( ϕ ) of trapping regions the homomorphism in (3) yields a finite sublattice A ⊆ Att ( ϕ ) A \subseteq\operatorname{Att}(\phi) A ⊆ Att ( ϕ ) of attractors via U ∋ U ↦ ω ( U ) = A ∈ A U \ni U \mapsto\omega(U)=A \in A U ∋ U ↦ ω ( U ) = A ∈ A , so that the diagram
ANbhd + ( ϕ ) → ω Att ( ϕ ) i ↑ ↑ i U → ω A \begin{CD}
\operatorname{ANbhd}_{+}(\phi) @>\omega>> \operatorname{Att}(\phi) \\
@A i AA @AA i A \\
U @>\omega>> A
\end{CD} ANbhd + ( ϕ ) i ⏐ ↑ U ω ω Att ( ϕ ) ⏐ ↑ i A
commutes. By Birkhoff’s representation theorem the surjective homomorphism ω : U → A \omega: U \to A ω : U → A dualizes to an order-embedding J ( ω ) : J ( A ) ↪ J ( U ) J(\omega): J(A) \hookrightarrow J(U) J ( ω ) : J ( A ) ↪ J ( U ) , where J J J is the contravariant functor that assigns to a finite, distributive lattice its poset of join-irreducible elements.6 Define J ( A ) ∋ A ↦ M = A − A ∗ J(A) \ni A \mapsto M = A - A^{*} J ( A ) ∋ A ↦ M = A − A ∗ , a Morse set, and J ( U ) ∋ U ↦ T = U \ U ∗ J(U) \ni U \mapsto T = U\backslash U^{*} J ( U ) ∋ U ↦ T = U \ U ∗ , a Morse tile. We obtain the order-embedding7
ι : M ( A ) ↪ T ( U ) , Inv ( ι ( M ) ) = M , (4) \iota: \mathop{\mathrm{M}}(A) \hookrightarrow\mathop{\mathrm{T}}(U), \quad\operatorname{Inv}(\iota(M)) = M, \tag*{(4)} ι : M ( A ) ↪ T ( U ) , Inv ( ι ( M )) = M , (4)
where M ( A ) = { M = A − A ∗ ∣ A ∈ J ( A ) } \mathop{\mathrm{M}}(A) = \{M = A - A^{*} \mid A \in J(A)\} M ( A ) = { M = A − A ∗ ∣ A ∈ J ( A )} , with M ⪯ M ′ M \preceq M' M ⪯ M ′ if and only if A ⊆ A ′ A \subseteq A' A ⊆ A ′ , A , A ′ ∈ J ( A ) A,A' \in J(A) A , A ′ ∈ J ( A ) (where M ′ = A ′ − A ′ ∗ M' = A' - A'^{*} M ′ = A ′ − A ′ ∗ ), T ( U ) = { T = U \ U ∗ ∣ U ∈ J ( U ) } \mathop{\mathrm{T}}(U) = \{T = U\backslash U^{*} \mid U \in J(U)\} T ( U ) = { T = U \ U ∗ ∣ U ∈ J ( U )} , with T ⪯ T ′ T \preceq T' T ⪯ T ′ if and only if U ⊆ U ′ U \subseteq U' U ⊆ U ′ , U , U ′ ∈ J ( U ) U,U' \in J(U) U , U ′ ∈ J ( U ) (where T ′ = U ′ \ U ′ ∗ T' = U'\backslash U'^{*} T ′ = U ′ \ U ′ ∗ ), and Inv ( ι ( M ) ) \operatorname{Inv}(\iota(M)) Inv ( ι ( M )) is the maximal invariant set in T = ι ( M ) T = \iota(M) T = ι ( M ) . The poset M ( A ) \mathop{\mathrm{M}}(A) M ( A ) is called a Morse representation , the poset T ( U ) \mathop{\mathrm{T}}(U) T ( U ) is called a Morse tessellation and the embedding in (4) is called a (tessellated ) Morse decomposition , v. [15 ] for details.
Notationally, the poset T ( U ) \mathop{\mathrm{T}}(U) T ( U ) is denoted by P P P and the isomorphism O ( P ) ≅ U \mathop{\mathrm{O}}(P) \cong U O ( P ) ≅ U , where O ( P ) \mathop{\mathrm{O}}(P) O ( P ) denotes the lattice of down-sets in P P P , v. [7 ] , is expressed as the lattice filtering
O ( P ) → ≅ U , α ↦ F α U ∈ U , \mathop{\mathrm{O}}(P) \xrightarrow{\cong} U,\quad\alpha\mapsto F_{\alpha}U \in U, O ( P ) ≅ U , α ↦ F α U ∈ U ,
v. [21 ] . The elements of T ( U ) \mathop{\mathrm{T}}(U) T ( U ) are denoted by G p U G_pU G p U , p ∈ P p \in P p ∈ P , i.e. G p U = F ↓ p U \ F ↓ p ∖ { p } U G_pU = F_{\downarrow p}U\backslash F_{\downarrow p\setminus\{p\}}U G p U = F ↓ p U \ F ↓ p ∖ { p } U , where ↓ p \downarrow p ↓ p is the down-set generated by p p p . For any convex set β \ α ∈ Co ( P ) \beta\backslash\alpha\in\operatorname{Co}(P) β \ α ∈ Co ( P ) , where Co ( P ) \operatorname{Co}(P) Co ( P ) denotes the meet-semilattice of convex sets in P P P , G β \ α U = F β U \ F α U G_{\beta\backslash\alpha}U = F_{\beta}U\backslash F_{\alpha}U G β \ α U = F β U \ F α U . This does not depend on the choice of α \alpha α and β \beta β , since
G β \ α U = ⋃ p ∈ β \ α G p U , G_{\beta\backslash\alpha}U = \bigcup_{p\in\beta\backslash\alpha} G_pU, G β \ α U = p ∈ β \ α ⋃ G p U ,
which defines the semilattice embedding Co ( P ) ↪ MTile ( ϕ ) \operatorname{Co}(P) \hookrightarrow\operatorname{MTile}(\phi) Co ( P ) ↪ MTile ( ϕ ) . The following theorem provides an equivalent characterization of a Morse representation.
Theorem 2.1 ([15 ] , Thm. 8.5) Let ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) be a finite poset of non-empty, pairwise disjoint, compact, invariant sets8 in X X X . Then, ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) is a Morse representation subordinate to a finite sublattice A ( M ) ⊆ Att ( ϕ ) A(M) \subseteq\operatorname{Att}(\phi) A ( M ) ⊆ Att ( ϕ ) if and only if for every x ∈ X x \in X x ∈ X there exists an M ∈ M M \in M M ∈ M such that ω ( x ) ⊆ M \omega(x) \subseteq M ω ( x ) ⊆ M , and for each complete orbit γ x \gamma_x γ x 9 with x ∉ ⋃ M ∈ M M x \notin\bigcup_{M\in M} M x ∈ / ⋃ M ∈ M M there exist M ≺ M ′ M \prec M' M ≺ M ′ such that ω ( x ) ⊆ M \omega(x) \subseteq M ω ( x ) ⊆ M and α o ( γ x ) ⊆ M ′ \alpha_o(\gamma_x) \subseteq M' α o ( γ x ) ⊆ M ′ . The associated lattice A ( M ) A(M) A ( M ) is the image of the injective lattice homomorphism ν : O ( M ) → Invset ( ϕ ) \nu: \mathop{\mathrm{O}}(M) \to\operatorname{Invset}(\phi) ν : O ( M ) → Invset ( ϕ ) given by
α ↦ ν ( α ) = ⋃ M ∈ α W u ( M ) ∈ Att ( ϕ ) , (5) \alpha\mapsto\nu(\alpha) = \bigcup_{M\in\alpha} W^{u}(M) \in\operatorname{Att}(\phi), \tag*{(5)} α ↦ ν ( α ) = M ∈ α ⋃ W u ( M ) ∈ Att ( ϕ ) , (5) where W u ( M ) = { x ∈ X ∣ there exists a backward orbit γ x with α o ( γ x ) ⊆ M } W^{u}(M) = \{x \in X \mid\text{there exists a backward orbit }\gamma_x\text{ with }\alpha_o(\gamma_x) \subseteq M\} W u ( M ) = { x ∈ X ∣ there exists a backward orbit γ x with α o ( γ x ) ⊆ M } is the unstable set of M M M . Moreover, M = M ( A ( M ) ) M = M(A(M)) M = M ( A ( M )) .
Let F : P o s e t → S e t F:\mathbf{Poset}\to\mathbf{Set} F : Poset → Set be the forgetful functor keeping the underlying set of a poset, and let F ( M ) ⊆ Set ( X ) F(M) \subseteq\operatorname{Set}(X) F ( M ) ⊆ Set ( X ) , where Set ( X ) \operatorname{Set}(X) Set ( X ) denotes the set of all subsets of X X X , be the collection of Morse sets in a Morse representation ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) .
Definition 2.2 A finite sublattice A ⊆ Att ( ϕ ) A \subseteq\operatorname{Att}(\phi) A ⊆ Att ( ϕ ) is flow-maximal if for every finite sublattice A ′ ⊆ Att ( ϕ ) A' \subseteq\operatorname{Att}(\phi) A ′ ⊆ Att ( ϕ ) such that F ( M ( A ′ ) ) = F ( M ( A ) ) F(\mathcal{M}(A')) = F(\mathcal{M}(A)) F ( M ( A ′ )) = F ( M ( A )) , it holds that A ′ ⊆ A A' \subseteq A A ′ ⊆ A .
The associated Morse representation ( M ( A ) , ≤ ) (\mathcal{M}(A), \leq) ( M ( A ) , ≤ ) is called a flow-order for ϕ \phi ϕ .10
A flow-order for ϕ \phi ϕ is obtained from a Morse representation ( M , ≤ ) (\mathcal{M}, \leq) ( M , ≤ ) by removing order relations (in finitely many steps) until the order is minimal with respect to the characterization in Theorem 2.1; since ν ( α ) \nu(\alpha) ν ( α ) in (5) does not depend on the order, a smaller order, which has more down-sets, yields a larger sublattice A ( M ) A(\mathcal{M}) A ( M ) . This implies that a flow-order ( M ( A ) , ≤ ) (\mathcal{M}(A), \leq) ( M ( A ) , ≤ ) is characterized by the property: M < M ′ M < M' M < M ′ if and only if there exists a chain { M i } i = 0 n ⊆ M ( A ) \{M_i\}_{i=0}^{n} \subseteq\mathcal{M}(A) { M i } i = 0 n ⊆ M ( A ) , with M 0 = M M_0 = M M 0 = M and M n = M ′ M_n = M' M n = M ′ , such that the sets of connecting orbits C ( M i , M i − 1 ) C(M_i,M_{i-1}) C ( M i , M i − 1 ) are non-empty for all 1 ≤ i ≤ n 1 \leq i \leq n 1 ≤ i ≤ n ; here C ( M ′ , M ) C(M',M) C ( M ′ , M ) denotes the set of complete orbits γ x \gamma_x γ x with α o ( γ x ) ⊆ M ′ \alpha_o(\gamma_x) \subseteq M' α o ( γ x ) ⊆ M ′ and ω ( x ) ⊆ M \omega(x) \subseteq M ω ( x ) ⊆ M .
Algebraic invariants
In order to organize the local data in a Morse representation, we start with a global structure reminiscent of homology braids as introduced in [9 , 11 ] , v. [21 ] .
Cartan–Eilenberg systems
3.1 Cartan–Eilenberg systems. Consider an arbitrary poset Q Q Q and let Q 1 Q_1 Q 1 be the associated posetal category. By Q 2 Q_2 Q 2 we denote the arrow category of Q 1 Q_1 Q 1 , which consists of ordered pairs ( α , β ) (\alpha,\beta) ( α , β ) , α ≤ β \alpha\leq\beta α ≤ β , ordered componentwise, i.e. ( α , β ) ≤ ( α ′ , β ′ ) (\alpha,\beta) \leq(\alpha',\beta') ( α , β ) ≤ ( α ′ , β ′ ) if and only if α ≤ α ′ \alpha\leq\alpha' α ≤ α ′ and β ≤ β ′ \beta\leq\beta' β ≤ β ′ . The (functor) categories Q n Q_n Q n consist of concatenations of n − 1 n-1 n − 1 arrows and are given by n n n -tuples α 1 ≤ ⋯ ≤ α n \alpha_1 \leq\cdots\leq\alpha_n α 1 ≤ ⋯ ≤ α n , again ordered componentwise, which may be regarded as a path of length n − 1 n-1 n − 1 in the directed graph induced by the poset Q Q Q . Let Mod R \operatorname{Mod}_R Mod R be the category of R R R -modules over a ring R R R .
Definition 3.1 (cf. [21 ] ) A Cartan–Eilenberg system , or E-system , E \mathbf{E} E over Q Q Q consists of:
E α β ∈ Mod R E^\beta_\alpha\in\operatorname{Mod}_R E α β ∈ Mod R , for all pairs ( α , β ) ∈ Q 2 (\alpha,\beta) \in Q_2 ( α , β ) ∈ Q 2 ;
homomorphisms ℓ : E α β → E α ′ β ′ \ell:E^\beta_\alpha\to E^{\beta'}_{\alpha'} ℓ : E α β → E α ′ β ′ for all pairs ( α , β ) ≤ ( α ′ , β ′ ) (\alpha,\beta) \leq(\alpha',\beta') ( α , β ) ≤ ( α ′ , β ′ ) , such that ℓ ′ ∘ ℓ = ℓ ′ ′ \ell' \circ\ell= \ell'' ℓ ′ ∘ ℓ = ℓ ′′ , where ℓ ′ : E α ′ β ′ → E α ′ ′ β ′ ′ \ell':E^{\beta'}_{\alpha'} \to E^{\beta''}_{\alpha''} ℓ ′ : E α ′ β ′ → E α ′′ β ′′ and ℓ ′ ′ : E α β → E α ′ ′ β ′ ′ \ell'':E^\beta_\alpha\to E^{\beta''}_{\alpha''} ℓ ′′ : E α β → E α ′′ β ′′ , for all triples ( α , β ) ≤ ( α ′ , β ′ ) ≤ ( α ′ ′ , β ′ ′ ) (\alpha,\beta) \leq(\alpha',\beta') \leq(\alpha'',\beta'') ( α , β ) ≤ ( α ′ , β ′ ) ≤ ( α ′′ , β ′′ ) , and such that ( α , β ) = ( α ′ , β ′ ) (\alpha,\beta) = (\alpha',\beta') ( α , β ) = ( α ′ , β ′ ) yields the identity map ℓ = id \ell= \operatorname{id} ℓ = id . Conditions (i) and (ii) define a functor Q 2 → Mod R Q_2 \to\operatorname{Mod}_R Q 2 → Mod R ;
connecting homomorphisms k : E β γ → E α β k:E^\gamma_\beta\to E^\beta_\alpha k : E β γ → E α β and k ′ : E β ′ γ ′ → E α ′ β ′ k':E^{\gamma'}_{\beta'} \to E^{\beta'}_{\alpha'} k ′ : E β ′ γ ′ → E α ′ β ′ such that the diagrams11
\begin{aligned}
E^\beta_\alpha&\xrightarrow{i} E^\gamma_\alpha\\
&\swarrow^{k} \quad\searrow^{j} \\
& E^\gamma_\beta
\end{aligned}
\qquad
\begin{array}{ccc}
E^\gamma_\beta&\xrightarrow{k}& E^\beta_\alpha\\
\downarrow^{\ell} && \downarrow^{\ell}\\
E^{\gamma'}_{\beta'}&\xrightarrow{k'}& E^{\beta'}_{\alpha'}
\end{array} \tag*{(6)}\end{aligned}
are exact triangles and commutative squares respectively for all pairs of ordered triples ( α , β , γ ) ≤ ( α ′ , β ′ , γ ′ ) (\alpha,\beta,\gamma) \leq(\alpha',\beta',\gamma') ( α , β , γ ) ≤ ( α ′ , β ′ , γ ′ ) .
By (ii) ℓ : E α β → ℓ E α β \ell:E^\beta_\alpha\xrightarrow{\ell} E^\beta_\alpha ℓ : E α β ℓ E α β is the identity homomorphism and the exactness of the left diagram in (6) shows that E α α = 0 E^\alpha_\alpha= 0 E α α = 0 for all ( α , α ) ∈ Q 2 (\alpha,\alpha) \in Q_2 ( α , α ) ∈ Q 2 , cf. [12 ] . In many situations Q Q Q is a modular12 lattice, in which case one can define a notion of excision. A Cartan–Eilenberg system is excisive if the homomorphisms ℓ : E α ∧ β α ∨ β → E α β \ell:E^{\alpha\vee\beta}_{\alpha\wedge\beta}\to E^\beta_\alpha ℓ : E α ∧ β α ∨ β → E α β are isomorphisms for all α , β ∈ Q \alpha,\beta\in Q α , β ∈ Q . In this paper we employ Cartan–Eilenberg systems over a finite, distributive lattice Q = L Q=L Q = L , represented as L = O ( P ) L=O(P) L = O ( P ) for some finite poset P P P via Birkhoff’s representation theorem, v. [7 ] . Since the elements in O ( P ) O(P) O ( P ) are down-sets, the excisive property is equivalent to: ι : E α β → E α ′ β ′ \iota:E^\beta_\alpha\to E^{\beta'}_{\alpha'} ι : E α β → E α ′ β ′ is an isomorphism for all ( α , β ) ≤ ( α ′ , β ′ ) (\alpha,\beta)\leq(\alpha',\beta') ( α , β ) ≤ ( α ′ , β ′ ) with β ∖ α = β ′ ∖ α ′ \beta\setminus\alpha=\beta'\setminus\alpha' β ∖ α = β ′ ∖ α ′ . In particular, E α β ≅ E α ′ β ′ E^\beta_\alpha\cong E^{\beta'}_{\alpha'} E α β ≅ E α ′ β ′ for all β ∖ α = β ′ ∖ α ′ \beta\setminus\alpha=\beta'\setminus\alpha' β ∖ α = β ′ ∖ α ′ , v. [21 ] .
A morphism between Cartan–Eilenberg systems E \mathbf{E} E and E ′ \mathbf{E}' E ′ over Q Q Q is given by homomorphisms h α β : E α β → E α ′ β h^\beta_\alpha:E^\beta_\alpha\to E'^\beta_\alpha h α β : E α β → E α ′ β such that the diagram
\begin{CD}
E^\beta_\alpha@>{i}>> E^\gamma_\alpha@>{j}>> E^\gamma_\beta@>{k}>> E^\beta_\alpha\\
@V{h^\beta_\alpha}VV @V{h^\gamma_\alpha}VV @V{h^\gamma_\beta}VV @V{h^\beta_\alpha}VV\\
E'^\beta_\alpha@>{i'}>> E'^\gamma_\alpha@>{j'}>> E'^\gamma_\beta@>{k'}>> E'^\beta_\alpha \tag*{(7)}
\end{CD}
commutes for any α ≤ β ≤ γ \alpha\leq\beta\leq\gamma α ≤ β ≤ γ . We write h : E → E ′ h:\mathbf{E}\to\mathbf{E}' h : E → E ′ , and h h h is an isomorphism if all h α β h^\beta_\alpha h α β are isomorphisms.
Remark 3.2 In some settings it is necessary to consider Cartan–Eilenberg systems over different posets, which requires extending the notion of morphism between Cartan–Eilenberg systems. Let E \mathbf{E} E and E ′ \mathbf{E}' E ′ be Cartan–Eilenberg systems over Q Q Q and Q ′ Q' Q ′ respectively. Then, a morphism between E \mathbf{E} E and E ′ \mathbf{E}' E ′ is given by an order-preserving map ψ : Q → Q ′ \psi:Q\to Q' ψ : Q → Q ′ and homomorphisms h α β : E α β → E ψ ( α ) ′ ψ ( β ) h^\beta_\alpha:E^\beta_\alpha\to E'^{\psi(\beta)}_{\psi(\alpha)} h α β : E α β → E ψ ( α ) ′ ψ ( β ) such that the diagram in (7) is satisfied for all α ≤ β ≤ γ \alpha\leq\beta\leq\gamma α ≤ β ≤ γ , with the bottom row replaced by R R R -modules E ψ ( α ) ′ ψ ( β ) E'^{\psi(\beta)}_{\psi(\alpha)} E ψ ( α ) ′ ψ ( β ) , E ψ ( α ) ′ ψ ( γ ) E'^{\psi(\gamma)}_{\psi(\alpha)} E ψ ( α ) ′ ψ ( γ ) and E ψ ( β ) ′ ψ ( γ ) E'^{\psi(\gamma)}_{\psi(\beta)} E ψ ( β ) ′ ψ ( γ ) .
In [16 Lem. 4.8] it is shown that the information in a Cartan–Eilenberg system can be captured by only considering exact triangles and commutative squares of the form:13
E 0 α → i E 0 β ↘ k ↙ j E α β E α β → k E 0 α ℓ ↓ ↓ i E α ′ β ′ → k ′ E 0 α ′ (8) \begin{aligned}
&E^\alpha_0 \xrightarrow{i} E^\beta_0 \\
&\quad\searrow k \qquad\swarrow j \\
&\qquad E^\beta_\alpha
\end{aligned}
\qquad\qquad
\begin{aligned}
&E^\beta_\alpha\xrightarrow{k} E^\alpha_0\\
&\ell\downarrow\qquad\downarrow i\\
&E^{\beta'}_{\alpha'} \xrightarrow{k'} E^{\alpha'}_0 \tag*{(8)}
\end{aligned} E 0 α i E 0 β ↘ k ↙ j E α β E α β k E 0 α ℓ ↓ ↓ i E α ′ β ′ k ′ E 0 α ′ (8)
This reduced system is called an exact couple system D \mathbf{D} D and is reminiscent of Boardman’s unrolled exact couple, cf. [3 , 12 ] . An exact couple system fully reconstructs a Cartan–Eilenberg system.
An invariant for Morse representations
A Morse set M M M is determined by a pair of attractors. If M = A − A ′ M=A-A' M = A − A ′ , then A − A ′ = A − ( A ′ ∧ A ) A-A'=A-(A'\wedge A) A − A ′ = A − ( A ′ ∧ A ) , which justifies considering only nested pairs ( A , A ′ ) (A,A') ( A , A ′ ) , A ′ ⊆ A A'\subseteq A A ′ ⊆ A . A pair ( U , U ′ ) (U,U') ( U , U ′ ) is called an index pair for M M M if (i) U , U ′ U,U' U , U ′ are trapping regions with U ′ ⊆ U U'\subseteq U U ′ ⊆ U , and (ii) M = A − A ′ M=A-A' M = A − A ′ , with A = ω ( U ) A=\omega(U) A = ω ( U ) and A ′ = ω ( U ′ ) A'=\omega(U') A ′ = ω ( U ′ ) . It holds that M ⊆ int ( U ∖ U ′ ) M\subseteq\operatorname{int}(U\setminus U') M ⊆ int ( U ∖ U ′ ) . The term index pair can be used without mention of the Morse set M M M . A nested pair U , U ′ U,U' U , U ′ of trapping regions with U ′ ⊊ U U'\subsetneq U U ′ ⊊ U does not guarantee that M M M is non-empty. Its relative homology yields a criterion to guarantee non-empty Morse sets M M M . We define the Conley index of M M M by
H C ( M ) : = H ( U , U ′ ) , (9) HC(M):=H(U,U'), \tag*{(9)} H C ( M ) := H ( U , U ′ ) , (9)
where H H H denotes (co)homology with coefficients in R R R , e.g. singular homology or Čech cohomology. We deliberately ignore the grading and regard H ( U , U ′ ) H(U,U') H ( U , U ′ ) as an R R R -module. The index does not depend on the choice of index pair, i.e. if ( V , V ′ ) (V,V') ( V , V ′ ) is an index pair with M = ω ( V ) − ω ( V ′ ) M = \omega(V) - \omega(V') M = ω ( V ) − ω ( V ′ ) , then H ( V , V ′ ) ≅ H ( U , U ′ ) H(V,V') \cong H(U,U') H ( V , V ′ ) ≅ H ( U , U ′ ) via natural isomorphisms, which for nested index pairs are induced by inclusion, v. [20 ] . The Conley index contains information about Morse sets. If H C ( M ) ≠ 0 HC(M) \ne0 H C ( M ) = 0 , then M ≠ ∅ M \ne\varnothing M = ∅ , v. [5 , 20 , 24 ] . The Conley index is invariant under continuation, v. [5 , 20 ] .
In this paper, however, we concentrate on Morse representations for a fixed dynamical system ϕ \phi ϕ . For a Morse representation ( M , ≤ ) (M,\leq) ( M , ≤ ) for ϕ \phi ϕ there exists a finite sublattice U ⊆ ANbhd + ( ϕ ) U \subseteq\operatorname{ANbhd}_{+}(\phi) U ⊆ ANbhd + ( ϕ ) of trapping regions such that M ≅ T ( U ) M \cong T(U) M ≅ T ( U ) .14 In other words, since M = M ( A ) M = M(A) M = M ( A ) , with A ⊆ Att ( ϕ ) A \subseteq\operatorname{Att}(\phi) A ⊆ Att ( ϕ ) , U U U is the image of a lift A ↪ ANbhd + ( ϕ ) A \hookrightarrow\operatorname{ANbhd}_{+}(\phi) A ↪ ANbhd + ( ϕ ) , v. [14 ] . A lattice U U U as indicated has the property that tiles in T ( U ) T(U) T ( U ) correspond to (non-empty) Morse sets in M M M . As before, we represent U U U by the lattice isomorphism O ( M ) → ≅ U O(M) \xrightarrow{\cong} U O ( M ) ≅ U given by α ↦ F α U ∈ U \alpha\mapsto F_{\alpha}^{U} \in U α ↦ F α U ∈ U . The same applies to O ( M ) → ≅ A O(M) \xrightarrow{\cong} A O ( M ) ≅ A given by α ↦ F α A ∈ A \alpha\mapsto F_{\alpha}^{A} \in A α ↦ F α A ∈ A .
Let C ( X ) C(X) C ( X ) be an R R R -module of topological chains and let d : C ( X ) → C ( X ) d:C(X)\to C(X) d : C ( X ) → C ( X ) be the differential making ( C ( X ) , d ) (C(X),d) ( C ( X ) , d ) a differential module, for example the singular chain complex of X X X with coefficients in R R R . The lattice filtering α ↦ F α U \alpha\mapsto F_{\alpha}^{U} α ↦ F α U yields a filtering of C ( X ) C(X) C ( X ) , i.e. α ↦ F α C ( X ) \alpha\mapsto F_{\alpha}C(X) α ↦ F α C ( X ) with F α C ( X ) : = C ( F α U ) F_{\alpha}C(X):=C(F_{\alpha}U) F α C ( X ) := C ( F α U ) , where C ( F α U ) C(F_{\alpha}U) C ( F α U ) are the topological chains in C ( X ) C(X) C ( X ) restricted to F α U F_{\alpha}U F α U . Observe that F α ∩ β C ( X ) = F α C ( X ) ∩ F β C ( X ) F_{\alpha\cap\beta}C(X)=F_{\alpha}C(X)\cap F_{\beta}C(X) F α ∩ β C ( X ) = F α C ( X ) ∩ F β C ( X ) . The same relation with respect to union does not hold in general. The differential satisfies d F α C ( X ) ⊆ F α C ( X ) dF_{\alpha}C(X)\subseteq F_{\alpha}C(X) d F α C ( X ) ⊆ F α C ( X ) for all α ∈ O ( M ) \alpha\in O(M) α ∈ O ( M ) , making α ↦ F α C ( X ) \alpha\mapsto F_{\alpha}C(X) α ↦ F α C ( X ) a meet semilattice filtered module. Define C α β = C ( F β U ) / C ( F α U ) C_{\alpha}^{\beta}=C(F_{\beta}U)/C(F_{\alpha}U) C α β = C ( F β U ) / C ( F α U ) and C α 0 = C ( F α U ) C_{\alpha}^{0}=C(F_{\alpha}U) C α 0 = C ( F α U ) , α ⊆ β \alpha\subseteq\beta α ⊆ β . We consider the following short exact sequences of differential modules. Given a triple α ⊆ β ⊆ γ \alpha\subseteq\beta\subseteq\gamma α ⊆ β ⊆ γ in O ( M ) O(M) O ( M ) , we have the short exact sequence:
0 ⟶ C α β → i C α γ → j C β γ ⟶ 0 , (10) 0 \longrightarrow C_{\alpha}^{\beta} \xrightarrow{i} C_{\alpha}^{\gamma} \xrightarrow{j} C_{\beta}^{\gamma} \longrightarrow0, \tag*{(10)} 0 ⟶ C α β i C α γ j C β γ ⟶ 0 , (10)
and for pairs α ⊆ β \alpha\subseteq\beta α ⊆ β and α ′ ⊆ β ′ \alpha'\subseteq\beta' α ′ ⊆ β ′ in O ( M ) O(M) O ( M ) , with α ⊆ α ′ \alpha\subseteq\alpha' α ⊆ α ′ , β ⊆ β ′ \beta\subseteq\beta' β ⊆ β ′ , we define the compositions:
C α β ⊆ C α ′ β ′ → π C β ′ α ′ . (11) C_{\alpha}^{\beta} \subseteq C_{\alpha'}^{\beta'} \xrightarrow{\pi} C_{\beta'}^{\alpha'}. \tag*{(11)} C α β ⊆ C α ′ β ′ π C β ′ α ′ . (11)
The homology associated to the filtering of ( C ( X ) , d ) (C(X),d) ( C ( X ) , d ) is denoted by H ( F α U ) = H ( C α 0 ) H(F_{\alpha}U)=H(C_{\alpha}^{0}) H ( F α U ) = H ( C α 0 ) and the relative homology is defined by H ( F β U , F α U ) = H ( C α β ) H(F_{\beta}U,F_{\alpha}U)=H(C_{\alpha}^{\beta}) H ( F β U , F α U ) = H ( C α β ) . The maps in (10) and (11) yield the exact triangles:
H ( F β U , F α U ) → i H ( F γ U , F α U ) ↙ k ↘ j H ( F γ U , F β U ) \begin{aligned}
H(F_{\beta}U,F_{\alpha}U) &\xrightarrow{i} H(F_{\gamma}U,F_{\alpha}U) \\
&\swarrow k \qquad\searrow j \\
& H(F_{\gamma}U,F_{\beta}U)
\end{aligned} H ( F β U , F α U ) i H ( F γ U , F α U ) ↙ k ↘ j H ( F γ U , F β U )
and commutative squares:
H ( F γ U , F β U ) → k H ( F β U , F α U ) ↓ ℓ ↓ ℓ H ( F γ ′ U , F β ′ U ) → k ′ H ( F β ′ U , F α ′ U ) \begin{array}{ccc}
H(F_{\gamma}U,F_{\beta}U) & \xrightarrow{k} & H(F_{\beta}U,F_{\alpha}U) \\
\downarrow\ell& & \downarrow\ell\\
H(F_{\gamma'}U,F_{\beta'}U) & \xrightarrow{k'} & H(F_{\beta'}U,F_{\alpha'}U)
\end{array} H ( F γ U , F β U ) ↓ ℓ H ( F γ ′ U , F β ′ U ) k k ′ H ( F β U , F α U ) ↓ ℓ H ( F β ′ U , F α ′ U )
where k k k and k ′ k' k ′ are the connecting homomorphisms, for all triples ( α , β , γ ) ≤ ( α ′ , β ′ , γ ′ ) (\alpha,\beta,\gamma)\leq(\alpha',\beta',\gamma') ( α , β , γ ) ≤ ( α ′ , β ′ , γ ′ ) in O ( M ) O(M) O ( M ) . Via the above defined homology modules, the lattice U U U defines a Cartan–Eilenberg system, which is denoted by E ( T ( U ) ) E(T(U)) E ( T ( U )) . The E E E -terms are given by
E α β : = H C ( M ) = H ( F β U , F α U ) , ∀ α ⊆ β , with M = ω ( F β U ) − ω ( F α U ) , (12) E_{\alpha}^{\beta}:=HC(M)=H(F_{\beta}U,F_{\alpha}U), \qquad\forall\alpha\subseteq\beta,\ \text{with } M=\omega(F_{\beta}U)-\omega(F_{\alpha}U), \tag*{(12)} E α β := H C ( M ) = H ( F β U , F α U ) , ∀ α ⊆ β , with M = ω ( F β U ) − ω ( F α U ) , (12)
It is important to point out that E ( T ( U ) ) E(\mathrm{T}(U)) E ( T ( U )) for a Morse representation is excisive even though ( C ( X ) , d ) (C(X),d) ( C ( X ) , d ) is not necessarily lattice-filtered, i.e. F α ∪ β C ( X ) = F α C ( X ) + F β C ( X ) F_{\alpha\cup\beta}C(X)=F_\alpha C(X)+F_\beta C(X) F α ∪ β C ( X ) = F α C ( X ) + F β C ( X ) need not hold. Indeed, for ( α , β ) ≤ ( α ′ , β ′ ) (\alpha,\beta)\leq(\alpha',\beta') ( α , β ) ≤ ( α ′ , β ′ ) with β ∖ α = β ′ ∖ α ′ \beta\setminus\alpha=\beta'\setminus\alpha' β ∖ α = β ′ ∖ α ′ we have that
ω ( F β U ) − ω ( F α U ) = ω ( F β ′ U ) − ω ( F α ′ U ) , \omega(F_\beta U)-\omega(F_\alpha U)=\omega(F_{\beta'}U)-\omega(F_{\alpha'}U), ω ( F β U ) − ω ( F α U ) = ω ( F β ′ U ) − ω ( F α ′ U ) ,
v. [15 ] , and therefore, by the invariance of the Conley index, the homomorphism ℓ : H ( F β U , F α U ) → H ( F β ′ U , F α ′ U ) \ell:H(F_\beta U,F_\alpha U)\to H(F_{\beta'}U,F_{\alpha'}U) ℓ : H ( F β U , F α U ) → H ( F β ′ U , F α ′ U ) , induced by the inclusion of index pairs, is an isomorphism, v. [20 ] , which proves the excisive property of E ( T ( U ) ) E(\mathrm{T}(U)) E ( T ( U )) . Since the E E E -terms are defined as homology, we say that E ( T ( U ) ) E(\mathrm{T}(U)) E ( T ( U )) is chain-generated . More generally, an E E E -system over O ( P ) \mathcal{O}(P) O ( P ) is called chain-generated if it is isomorphic to the E E E -system defined for instance as above by a differential module ( C , d ) (C,d) ( C , d ) with a filtering α ↦ F α C \alpha\mapsto F_\alpha C α ↦ F α C by submodules satisfying d F α C ⊆ F α C dF_\alpha C\subseteq F_\alpha C d F α C ⊆ F α C and F α ∩ β C = F α C ∩ F β C F_{\alpha\cap\beta}C=F_\alpha C\cap F_\beta C F α ∩ β C = F α C ∩ F β C , v. [23 ] .
Theorem 3.3 Let T ( U ) ≅ T ( U ′ ) \mathrm{T}(U)\cong\mathrm{T}(U') T ( U ) ≅ T ( U ′ ) be two Morse tessellations for the Morse representation M ( A ) \mathrm{M}(A) M ( A ) . Then, E ( T ( U ) ) ≅ E ( T ( U ′ ) ) E(\mathrm{T}(U))\cong E(\mathrm{T}(U')) E ( T ( U )) ≅ E ( T ( U ′ )) as Cartan–Eilenberg systems.
Proof. Since ω ( F α U ) = F α A = ω ( F α U ′ ) \omega(F_\alpha U)=F_\alpha A=\omega(F_\alpha U') ω ( F α U ) = F α A = ω ( F α U ′ ) and ω \omega ω is a lattice homomorphism, F α V : = F α U ∩ F α U ′ F_\alpha V:=F_\alpha U\cap F_\alpha U' F α V := F α U ∩ F α U ′ is a trapping region with ω ( F α V ) = F α A \omega(F_\alpha V)=F_\alpha A ω ( F α V ) = F α A for all α ∈ O ( M ) \alpha\in\mathcal{O}(M) α ∈ O ( M ) . The map α ↦ F α V \alpha\mapsto F_\alpha V α ↦ F α V is monotone, which is all that is used in (3.5) and (3.6). Replacing F α U F_\alpha U F α U by F α V F_\alpha V F α V therefore defines a Cartan–Eilenberg system E ( V ) E(V) E ( V ) with E E E -terms H ( F β V , F α V ) H(F_\beta V,F_\alpha V) H ( F β V , F α V ) . For α ⊆ β \alpha\subseteq\beta α ⊆ β the inclusion ( F β V , F α V ) ⊆ ( F β U , F α U ) (F_\beta V,F_\alpha V)\subseteq(F_\beta U,F_\alpha U) ( F β V , F α V ) ⊆ ( F β U , F α U ) induces chain maps between the short exact sequences in (3.5), which commute with the maps in (3.6). By naturality of the connecting homomorphisms the induced homomorphisms h α β : H ( F β V , F α V ) → H ( F β U , F α U ) h_\alpha^\beta:H(F_\beta V,F_\alpha V)\to H(F_\beta U,F_\alpha U) h α β : H ( F β V , F α V ) → H ( F β U , F α U ) define a morphism h : E ( V ) → E ( T ( U ) ) h:E(V)\to E(\mathrm{T}(U)) h : E ( V ) → E ( T ( U )) . Both pairs are index pairs for the Morse set F β A − F α A F_\beta A-F_\alpha A F β A − F α A . Hence each h α β h_\alpha^\beta h α β is an isomorphism, v. [20 ] , and h h h is an isomorphism. In the same way E ( V ) ≅ E ( T ( U ′ ) ) E(V)\cong E(\mathrm{T}(U')) E ( V ) ≅ E ( T ( U ′ )) . □ \square □
By Theorem 3.3 the isomorphism class of the Cartan–Eilenberg system E ( T ( U ) ) E(\mathrm{T}(U)) E ( T ( U )) is an invariant for ( M , ≤ ) (M,\leq) ( M , ≤ ) , which we denote by E ( M ) E(M) E ( M ) . In this case we write
E α β = HC ( M ) , ∀ α ⊆ β , with M = Inv ( G β ∖ α U ) , (13) E_\alpha^\beta=\operatorname{HC}(M),\qquad\forall\alpha\subseteq\beta,\quad\text{with }M=\operatorname{Inv}(G_{\beta\setminus\alpha}U), \tag*{(13)} E α β = HC ( M ) , ∀ α ⊆ β , with M = Inv ( G β ∖ α U ) , (13)
i.e. up to isomorphism the E E E -terms of E ( M ) E(M) E ( M ) are the Conley indices of the Morse sets Inv ( G β ∖ α U ) = ω ( F β U ) − ω ( F α U ) \operatorname{Inv}(G_{\beta\setminus\alpha}U)=\omega(F_\beta U)-\omega(F_\alpha U) Inv ( G β ∖ α U ) = ω ( F β U ) − ω ( F α U ) .
Remark 3.4 The term chain-generated is reminiscent of Franzosa’s approach of assigning invariants to Morse representations, v. [11 ] . The approach in [11 ] is more restrictive since the complexes C β α C_\beta^\alpha C β α are required to be excisive, i.e. to depend only on β ∖ α \beta\setminus\alpha β ∖ α (equivalently, F α ∪ β C = F α C + F β C F_{\alpha\cup\beta}C=F_\alpha C+F_\beta C F α ∪ β C = F α C + F β C ). This property is not needed in the theory of Morse representations presented in this paper.
Spectral systems
Cartan–Eilenberg systems over a poset Q Q Q , v. Sect. 3.1, may be regarded as page-1 of a spectral system as introduced in [16 ] . In this section we give an axiomatic definition of spectral systems that are not a priori generated by a Cartan–Eilenberg system.
Definition 4.1 A spectral system S S S , or S S S -system, over a poset Q Q Q consists of:
(i) S S S -terms S α β γ δ ∈ Mod R S_{\alpha\beta}^{\gamma\delta}\in\operatorname{Mod}_{R} S α β γ δ ∈ Mod R , for all quadruples ( α , β , γ , δ ) ∈ Q 4 (\alpha,\beta,\gamma,\delta)\in Q_4 ( α , β , γ , δ ) ∈ Q 4 ;
(ii) homomorphisms ℓ : S α β γ δ → S α ′ β ′ γ ′ δ ′ \ell:S_{\alpha\beta}^{\gamma\delta}\to S_{\alpha'\beta'}^{\gamma'\delta'} ℓ : S α β γ δ → S α ′ β ′ γ ′ δ ′ for all ( α , β , γ , δ ) ≤ ( α ′ , β ′ , γ ′ , δ ′ ) (\alpha,\beta,\gamma,\delta)\leq(\alpha',\beta',\gamma',\delta') ( α , β , γ , δ ) ≤ ( α ′ , β ′ , γ ′ , δ ′ ) in Q 4 Q_4 Q 4 , such that ℓ ′ ∘ ℓ = ℓ ′ ′ \ell'\circ\ell=\ell'' ℓ ′ ∘ ℓ = ℓ ′′ , where ℓ ′ : S α ′ β ′ γ ′ δ ′ → S α ′ ′ β ′ ′ γ ′ ′ δ ′ ′ \ell':S_{\alpha'\beta'}^{\gamma'\delta'}\to S_{\alpha''\beta''}^{\gamma''\delta''} ℓ ′ : S α ′ β ′ γ ′ δ ′ → S α ′′ β ′′ γ ′′ δ ′′ and ℓ ′ ′ : S α β γ δ → S α ′ ′ β ′ ′ γ ′ ′ δ ′ ′ \ell'':S_{\alpha\beta}^{\gamma\delta}\to S_{\alpha''\beta''}^{\gamma''\delta''} ℓ ′′ : S α β γ δ → S α ′′ β ′′ γ ′′ δ ′′ for all triples of quadruples ( α , β , γ , δ ) ≤ ( α ′ , β ′ , γ ′ , δ ′ ) ≤ ( α ′ ′ , β ′ ′ , γ ′ ′ , δ ′ ′ ) (\alpha,\beta,\gamma,\delta)\leq(\alpha',\beta',\gamma',\delta')\leq(\alpha'',\beta'',\gamma'',\delta'') ( α , β , γ , δ ) ≤ ( α ′ , β ′ , γ ′ , δ ′ ) ≤ ( α ′′ , β ′′ , γ ′′ , δ ′′ ) in Q 4 Q_4 Q 4 .
Identical quadruples ( α , β , γ , δ ) = ( α ′ , β ′ , γ ′ , δ ′ ) (\alpha,\beta,\gamma,\delta)=(\alpha',\beta',\gamma',\delta') ( α , β , γ , δ ) = ( α ′ , β ′ , γ ′ , δ ′ ) yield the identity map ℓ = i d \ell=\mathrm{id} ℓ = id . Conditions (i) and (ii) define a functor Q 4 → Mod R \mathbf{Q}_4\to\operatorname{Mod}_R Q 4 → Mod R ;
(iii) differentials d : S γ δ ϵ ζ → S α β γ δ \mathrm{d}:S_{\gamma\delta}^{\epsilon\zeta}\to S_{\alpha\beta}^{\gamma\delta} d : S γ δ ϵ ζ → S α β γ δ for all sextuples ( α , β , γ , δ , ϵ , ζ ) ∈ Q 6 (\alpha,\beta,\gamma,\delta,\epsilon,\zeta)\in\mathbf{Q}_6 ( α , β , γ , δ , ϵ , ζ ) ∈ Q 6 , and such that
S β δ ϵ ζ ≅ ker : S γ δ ϵ ζ → d S α β γ δ im : S ϵ ζ α β → d S γ δ ϵ ζ = H ( S γ δ ϵ ζ ) , (homology), (14) S_{\beta\delta}^{\epsilon\zeta}\cong\frac{\ker:S_{\gamma\delta}^{\epsilon\zeta}\xrightarrow{\mathrm{d}}S_{\alpha\beta}^{\gamma\delta}}{\operatorname{im}:S_{\epsilon\zeta}^{\alpha\beta}\xrightarrow{\mathrm{d}}S_{\gamma\delta}^{\epsilon\zeta}}=H(S_{\gamma\delta}^{\epsilon\zeta}),\quad\text{(homology),} \tag*{(14)} S β δ ϵ ζ ≅ im : S ϵ ζ α β d S γ δ ϵ ζ ker : S γ δ ϵ ζ d S α β γ δ = H ( S γ δ ϵ ζ ) , (homology), (14)
for all octuples ( α , β , γ , δ , ϵ , ζ , η , θ ) ∈ Q 8 (\alpha,\beta,\gamma,\delta,\epsilon,\zeta,\eta,\theta)\in\mathbf{Q}_8 ( α , β , γ , δ , ϵ , ζ , η , θ ) ∈ Q 8 , where S ϵ ζ η θ → d S γ δ ϵ ζ → d S α β γ δ S_{\epsilon\zeta}^{\eta\theta}\xrightarrow{\mathrm{d}}S_{\gamma\delta}^{\epsilon\zeta}\xrightarrow{\mathrm{d}}S_{\alpha\beta}^{\gamma\delta} S ϵ ζ η θ d S γ δ ϵ ζ d S α β γ δ satisfies d 2 = 0 \mathrm{d}^2=0 d 2 = 0 . The isomorphism in (14) is induced by the homomorphisms ℓ \ell ℓ in (ii), i.e. ℓ : S β δ ϵ ζ → S γ δ ϵ ζ \ell:S_{\beta\delta}^{\epsilon\zeta}\to S_{\gamma\delta}^{\epsilon\zeta} ℓ : S β δ ϵ ζ → S γ δ ϵ ζ is injective with image ker d \ker\mathrm{d} ker d , and ℓ ′ : S γ δ ϵ ζ → S γ δ η θ \ell':S_{\gamma\delta}^{\epsilon\zeta}\to S_{\gamma\delta}^{\eta\theta} ℓ ′ : S γ δ ϵ ζ → S γ δ η θ is surjective with kernel im d \operatorname{im}\mathrm{d} im d ;
(iv) commutative diagrams
S γ δ ϵ ζ → d S α β γ δ ℓ ↓ ↓ ℓ S γ ′ δ ′ ϵ ′ ζ ′ → d S α ′ β ′ γ ′ δ ′ (15) \begin{CD}
S_{\gamma\delta}^{\epsilon\zeta} @>\mathrm{d}>> S_{\alpha\beta}^{\gamma\delta}\\
@V\ell VV @VV\ell V\\
S_{\gamma'\delta'}^{\epsilon'\zeta'} @>\mathrm{d}>> S_{\alpha'\beta'}^{\gamma'\delta'} \tag*{(15)}
\end{CD} S γ δ ϵ ζ ℓ ↓ ⏐ S γ ′ δ ′ ϵ ′ ζ ′ d d S α β γ δ ↓ ⏐ ℓ S α ′ β ′ γ ′ δ ′ (15)
for all sextuples ( α , β , γ , δ , ϵ , ζ ) ⪯ ( α ′ , β ′ , γ ′ , δ ′ , ϵ ′ , ζ ′ ) (\alpha,\beta,\gamma,\delta,\epsilon,\zeta)\preceq(\alpha',\beta',\gamma',\delta',\epsilon',\zeta') ( α , β , γ , δ , ϵ , ζ ) ⪯ ( α ′ , β ′ , γ ′ , δ ′ , ϵ ′ , ζ ′ ) in Q 6 \mathbf{Q}_6 Q 6 ;
(v) short exact sequences
0 ⟶ S α β γ ϵ → i S α β δ ϵ → j S α γ δ ϵ ⟶ 0 , (16) 0\longrightarrow S_{\alpha\beta}^{\gamma\epsilon}\xrightarrow{i}S_{\alpha\beta}^{\delta\epsilon}\xrightarrow{j}S_{\alpha\gamma}^{\delta\epsilon}\longrightarrow0, \tag*{(16)} 0 ⟶ S α β γ ϵ i S α β δ ϵ j S α γ δ ϵ ⟶ 0 , (16)
for all quintuples ( α , β , γ , δ , ϵ ) ∈ Q 5 (\alpha,\beta,\gamma,\delta,\epsilon)\in\mathbf{Q}_5 ( α , β , γ , δ , ϵ ) ∈ Q 5 .15
Remark 4.2 The isomorphism in (14) is induced by the homomorphisms ℓ \ell ℓ as follows. Let ( α , β , γ , δ , ϵ , ζ , η , θ ) ∈ Q 8 (\alpha,\beta,\gamma,\delta,\epsilon,\zeta,\eta,\theta)\in\mathbf{Q}_8 ( α , β , γ , δ , ϵ , ζ , η , θ ) ∈ Q 8 . The two conditions in (iii) concern the diagram
S β δ ϵ ζ ℓ ↓ S ϵ ζ η θ → d , im d = ker ℓ ′ S γ δ ϵ ζ → d , ker d = im ℓ ′ S α β γ δ ℓ ′ ↓ S γ δ η θ \begin{CD}
&&S_{\beta\delta}^{\epsilon\zeta}\\
&&@V\ell VV\\
S_{\epsilon\zeta}^{\eta\theta}@>\mathrm{d},\ \operatorname{im}\mathrm{d}=\ker\ell'>>S_{\gamma\delta}^{\epsilon\zeta}@>\mathrm{d},\ \ker\mathrm{d}=\operatorname{im}\ell'>>S_{\alpha\beta}^{\gamma\delta}\\
&&@V\ell'VV\\
&&S_{\gamma\delta}^{\eta\theta}
\end{CD} S β δ ϵ ζ ℓ ↓ ⏐ S ϵ ζ η θ ℓ ′ ↓ ⏐ S γ δ η θ d , im d = k e r ℓ ′ S γ δ ϵ ζ d , k e r d = im ℓ ′ S α β γ δ
in which ℓ \ell ℓ and ℓ ′ \ell' ℓ ′ are homomorphisms as in (ii): ℓ \ell ℓ is injective with image ker d \ker\mathrm{d} ker d , the kernel of the right-hand differential, and ℓ ′ \ell' ℓ ′ is surjective with kernel im d \operatorname{im}\mathrm{d} im d , the image of the left-hand differential. Since im d ⊆ ker d \operatorname{im}\mathrm{d}\subseteq\ker\mathrm{d} im d ⊆ ker d , we have im d = ker d ∩ ker ℓ ′ \operatorname{im}\mathrm{d}=\ker\mathrm{d}\cap\ker\ell' im d = ker d ∩ ker ℓ ′ , and therefore ℓ ′ \ell' ℓ ′ induces an isomorphism
H ( S γ δ ϵ ζ ) = ker d im d = ℓ ( S β δ ϵ ζ ) ℓ ( S β δ ϵ ζ ) ∩ ker ℓ ′ → ℓ ′ ≅ ℓ ′ ℓ ( S β δ ϵ ζ ) ⊆ S γ δ η θ . H(S_{\gamma\delta}^{\epsilon\zeta})=\frac{\ker\mathrm{d}}{\operatorname{im}\mathrm{d}}=\frac{\ell(S_{\beta\delta}^{\epsilon\zeta})}{\ell(S_{\beta\delta}^{\epsilon\zeta})\cap\ker\ell'}\xrightarrow[\ell']{\cong}\ell'\ell(S_{\beta\delta}^{\epsilon\zeta})\subseteq S_{\gamma\delta}^{\eta\theta}. H ( S γ δ ϵ ζ ) = im d ker d = ℓ ( S β δ ϵ ζ ) ∩ ker ℓ ′ ℓ ( S β δ ϵ ζ ) ≅ ℓ ′ ℓ ′ ℓ ( S β δ ϵ ζ ) ⊆ S γ δ η θ .
To identify the image ℓ ′ ℓ ( S β δ ϵ ζ ) \ell'\ell(S_{\beta\delta}^{\epsilon\zeta}) ℓ ′ ℓ ( S β δ ϵ ζ ) , consider the square
S β δ ϵ ζ → ℓ ′ ′ S β δ η θ ℓ ↓ ↓ ℓ ′ ′ ′ S γ δ ϵ ζ → ℓ ′ S γ δ η θ \begin{CD}
S_{\beta\delta}^{\epsilon\zeta}@>\ell''>>S_{\beta\delta}^{\eta\theta}\\
@V\ell VV @VV\ell'''V\\
S_{\gamma\delta}^{\epsilon\zeta}@>\ell'>>S_{\gamma\delta}^{\eta\theta}
\end{CD} S β δ ϵ ζ ℓ ↓ ⏐ S γ δ ϵ ζ ℓ ′′ ℓ ′ S β δ η θ ↓ ⏐ ℓ ′′′ S γ δ η θ
of homomorphisms as in (ii). It commutes, since both compositions are the homomorphism S β δ ξ ζ → S γ δ η θ S_{\beta\delta}^{\xi\zeta} \to S_{\gamma\delta}^{\eta\theta} S β δ ξ ζ → S γ δ η θ of (ii). The map ℓ ′ \ell' ℓ ′ is surjective by the second condition in (iii) for the octuple ( β , β , β , δ , ϵ , ζ , η , θ ) (\beta,\beta,\beta,\delta,\epsilon,\zeta,\eta,\theta) ( β , β , β , δ , ϵ , ζ , η , θ ) , and ℓ ′ ′ ′ \ell''' ℓ ′′′ is injective by the first condition in (iii) for the octuple ( α , β , γ , δ , ϵ , η , θ , θ ) (\alpha,\beta,\gamma,\delta,\epsilon,\eta,\theta,\theta) ( α , β , γ , δ , ϵ , η , θ , θ ) . Hence ℓ ′ ℓ ( S β δ ξ ζ ) = ℓ ′ ′ ℓ ′ ′ ′ ( S β δ ξ ζ ) = ℓ ′ ′ ( S β δ η θ ) \ell'\ell(S_{\beta\delta}^{\xi\zeta})=\ell''\ell'''(S_{\beta\delta}^{\xi\zeta})=\ell''(S_{\beta\delta}^{\eta\theta}) ℓ ′ ℓ ( S β δ ξ ζ ) = ℓ ′′ ℓ ′′′ ( S β δ ξ ζ ) = ℓ ′′ ( S β δ η θ ) , and ℓ ′ ′ ′ \ell''' ℓ ′′′ maps S β δ η θ S_{\beta\delta}^{\eta\theta} S β δ η θ isomorphically onto this image.
Combining the two isomorphisms, the isomorphism in (14) is
S β δ ξ η → ≅ H ( S γ δ ξ ζ ) , x ↦ [ ℓ ( y ) ] , where y ∈ S β δ ξ ζ with ℓ ′ ′ ( y ) = x , S_{\beta\delta}^{\xi\eta}\xrightarrow{\ \cong\ }H(S_{\gamma\delta}^{\xi\zeta}),\qquad x\mapsto[\ell(y)],\quad\text{where }y\in S_{\beta\delta}^{\xi\zeta}\text{ with }\ell''(y)=x, S β δ ξ η ≅ H ( S γ δ ξ ζ ) , x ↦ [ ℓ ( y )] , where y ∈ S β δ ξ ζ with ℓ ′′ ( y ) = x ,
with inverse [ ℓ ( y ) ] ↦ ℓ ′ ( y ) [\ell(y)]\mapsto\ell'(y) [ ℓ ( y )] ↦ ℓ ′ ( y ) . The class [ ℓ ( y ) ] [\ell(y)] [ ℓ ( y )] does not depend on the choice of y y y : if ℓ ′ ′ ( y ) = ℓ ′ ′ ( y ′ ) \ell''(y)=\ell''(y') ℓ ′′ ( y ) = ℓ ′′ ( y ′ ) , then ℓ ′ ′ ( y − y ′ ) = ℓ ′ ′ ′ ℓ ′ ′ ( y − y ′ ) = 0 \ell''(y-y')=\ell''' \ell''(y-y')=0 ℓ ′′ ( y − y ′ ) = ℓ ′′′ ℓ ′′ ( y − y ′ ) = 0 , so that ℓ ( y − y ′ ) ∈ ker ℓ ′ = im d \ell(y-y')\in\ker\ell'=\operatorname{im}d ℓ ( y − y ′ ) ∈ ker ℓ ′ = im d .
Conversely, the two conditions in (iii) are equivalent to the exactness of the sequences (19) in Lemma 4.4 . For the spectral system S ( E ) \mathbf{S}(\mathbf{E}) S ( E ) of a Cartan–Eilenberg system E \mathbf{E} E , v. Sect. 4.2, both conditions hold: with S α β γ δ ≅ im ( E α γ → E β δ ) S_{\alpha\beta}^{\gamma\delta}\cong\operatorname{im}(E_{\alpha}^{\gamma}\to E_{\beta}^{\delta}) S α β γ δ ≅ im ( E α γ → E β δ ) , ℓ \ell ℓ is an inclusion of submodules of E δ ϵ E_{\delta}^{\epsilon} E δ ϵ and ℓ ′ \ell' ℓ ′ is surjective, v. [16 Lem. 4.11] .
Remark 4.3 Properties (ii) and (v) yield the short exact sequence 0 ⟶ S α α α α → id S α α α α → id S α α α α ⟶ 0 0\longrightarrow S_{\alpha\alpha}^{\alpha\alpha}\xrightarrow{\operatorname{id}}S_{\alpha\alpha}^{\alpha\alpha}\xrightarrow{\operatorname{id}}S_{\alpha\alpha}^{\alpha\alpha}\longrightarrow0 0 ⟶ S α α α α id S α α α α id S α α α α ⟶ 0 , and therefore S α α α α = 0 S_{\alpha\alpha}^{\alpha\alpha}=0 S α α α α = 0 for all α \alpha α . By (14) this implies that S α β β γ = 0 S_{\alpha\beta}^{\beta\gamma}=0 S α β β γ = 0 for all ( α , β , γ ) ∈ Q 3 (\alpha,\beta,\gamma)\in\mathbf{Q}^{3} ( α , β , γ ) ∈ Q 3 .
The E E E -terms for spectral systems
From Definition 4.1(iii) it follows that
ker : S γ δ ξ ϵ → d S α β γ δ ≅ S β δ ξ ϵ , and coker : S γ δ η θ → d S γ δ ξ ϵ ≅ S γ δ ξ η , (17) \ker:S_{\gamma\delta}^{\xi\epsilon}\xrightarrow{d}S_{\alpha\beta}^{\gamma\delta}\cong S_{\beta\delta}^{\xi\epsilon},\qquad\text{and}\qquad\operatorname{coker}:S_{\gamma\delta}^{\eta\theta}\xrightarrow{d}S_{\gamma\delta}^{\xi\epsilon}\cong S_{\gamma\delta}^{\xi\eta}, \tag*{(17)} ker : S γ δ ξ ϵ d S α β γ δ ≅ S β δ ξ ϵ , and coker : S γ δ η θ d S γ δ ξ ϵ ≅ S γ δ ξ η , (17)
for all octuples ( α , β , γ , δ , ϵ , ζ , η , θ ) ∈ Q 8 (\alpha,\beta,\gamma,\delta,\epsilon,\zeta,\eta,\theta)\in\mathbf{Q}^{8} ( α , β , γ , δ , ϵ , ζ , η , θ ) ∈ Q 8 , v. [16 Lem. 4.11] . In particular, for a triple ( α , β , γ ) ∈ Q 3 (\alpha,\beta,\gamma)\in\mathbf{Q}^{3} ( α , β , γ ) ∈ Q 3 , the differential d : S β β γ γ → S α α β β d:S_{\beta\beta}^{\gamma\gamma}\to S_{\alpha\alpha}^{\beta\beta} d : S β β γ γ → S α α β β is defined via the sextuple ( α , α , β , β , γ , γ ) (\alpha,\alpha,\beta,\beta,\gamma,\gamma) ( α , α , β , β , γ , γ ) . Then, the short exact sequences in (16) yield the short exact sequence
0 ⟶ S α α β γ → i S α α γ γ → j S α β γ γ ⟶ 0 , (18) 0\longrightarrow S_{\alpha\alpha}^{\beta\gamma}\xrightarrow{i}S_{\alpha\alpha}^{\gamma\gamma}\xrightarrow{j}S_{\alpha\beta}^{\gamma\gamma}\longrightarrow0, \tag*{(18)} 0 ⟶ S α α β γ i S α α γ γ j S α β γ γ ⟶ 0 , (18)
with S α α β γ ≅ coker d S_{\alpha\alpha}^{\beta\gamma}\cong\operatorname{coker}d S α α β γ ≅ coker d and S α β γ γ ≅ ker d S_{\alpha\beta}^{\gamma\gamma}\cong\ker d S α β γ γ ≅ ker d , by (17) . Equation (17) implies another family of exact sequences.
Lemma 4.4 The sequence
0 ⟶ S β δ ξ ϵ → ℓ S γ δ ξ ϵ → d S α β γ δ → ℓ S α β ξ ϵ ⟶ 0 (19) 0\longrightarrow S_{\beta\delta}^{\xi\epsilon}\xrightarrow{\ell}S_{\gamma\delta}^{\xi\epsilon}\xrightarrow{d}S_{\alpha\beta}^{\gamma\delta}\xrightarrow{\ell}S_{\alpha\beta}^{\xi\epsilon}\longrightarrow0 \tag*{(19)} 0 ⟶ S β δ ξ ϵ ℓ S γ δ ξ ϵ d S α β γ δ ℓ S α β ξ ϵ ⟶ 0 (19) is exact for all sextuples ( α , β , γ , δ , ϵ , ζ ) ∈ Q 6 (\alpha,\beta,\gamma,\delta,\epsilon,\zeta)\in\mathbf{Q}^{6} ( α , β , γ , δ , ϵ , ζ ) ∈ Q 6 .
Proof. For d : S γ δ ξ ϵ → S α β γ δ d:S_{\gamma\delta}^{\xi\epsilon}\to S_{\alpha\beta}^{\gamma\delta} d : S γ δ ξ ϵ → S α β γ δ we have the exact sequence 0\longrightarrow\ker d\lhook\joinrel\longrightarrow S_{\gamma\delta}^{\xi\epsilon}\xrightarrow{d}S_{\alpha\beta}^{\gamma\delta}\xrightarrow{\pi}\operatorname{coker}d\longrightarrow0 . By Definition 4.1(iii), ℓ : S β δ ξ ϵ → S γ δ ξ ϵ \ell:S_{\beta\delta}^{\xi\epsilon}\to S_{\gamma\delta}^{\xi\epsilon} ℓ : S β δ ξ ϵ → S γ δ ξ ϵ is injective with image ker d \ker d ker d , and ℓ : S α β γ δ → S α β ξ ϵ \ell:S_{\alpha\beta}^{\gamma\delta}\to S_{\alpha\beta}^{\xi\epsilon} ℓ : S α β γ δ → S α β ξ ϵ is surjective with kernel im d \operatorname{im}d im d (apply (iii) to the octuple ( α , α , α , β , γ , δ , ϵ , ζ ) (\alpha,\alpha,\alpha,\beta,\gamma,\delta,\epsilon,\zeta) ( α , α , α , β , γ , δ , ϵ , ζ ) ), which proves the lemma. □
As a consequence of Lemma 4.4 we can build exact triangles from the short exact sequences in (18) .
Lemma 4.5 Let S S S be a spectral system over Q \mathbf{Q} Q . Then, the triangles
S α α β γ → i S α α γ γ ↖ d ↙ j S α β γ γ (20) \begin{array}{ccc}
S_{\alpha\alpha}^{\beta\gamma} & \xrightarrow{i} & S_{\alpha\alpha}^{\gamma\gamma}\\
& \nwarrow d & \swarrow j\\
& S_{\alpha\beta}^{\gamma\gamma} & \tag*{(20)}
\end{array} S α α β γ i ↖ d S α β γ γ S α α γ γ ↙ j (20) are exact, for all triples ( α , β , γ ) ∈ Q 3 (\alpha,\beta,\gamma)\in\mathbf{Q}^{3} ( α , β , γ ) ∈ Q 3 .
Proof. By Lemma 4.4 the sequence
0 ⟶ S α β γ γ → j ′ ′ S β β γ γ → d S α α β β → i ′ ′ S α α β γ ⟶ 0 (21) 0\longrightarrow S_{\alpha\beta}^{\gamma\gamma}\xrightarrow{j''}S_{\beta\beta}^{\gamma\gamma}\xrightarrow{d}S_{\alpha\alpha}^{\beta\beta}\xrightarrow{i''}S_{\alpha\alpha}^{\beta\gamma}\longrightarrow0 \tag*{(21)} 0 ⟶ S α β γ γ j ′′ S β β γ γ d S α α β β i ′′ S α α β γ ⟶ 0 (21)
is exact for all triples ( α , β , γ ) ∈ Q 3 (\alpha,\beta,\gamma) \in\mathbf{Q}_3 ( α , β , γ ) ∈ Q 3 . By combining (18) with (21) we obtain the diagram
\begin{CD}
0 @>>> S_{\alpha\alpha}^{\beta\gamma} @>{i'}>> S_{\alpha\alpha}^{\gamma\gamma} @>{j'}>> S_{\alpha\beta}^{\gamma\gamma} @>>> 0 \\
@. @A{i''}AA @AA{i}A @VV{j''}V @. \\
@. S_{\alpha\alpha}^{\beta\beta} @>{d}<< S_{\beta\beta}^{\gamma\gamma} @. @.
\end{CD}
Using the composition property in (ii) we obtain i = i ′ i ′ ′ i=i'i'' i = i ′ i ′′ and j = j ′ ′ j ′ j=j''j' j = j ′′ j ′ . Since the horizontal sequence and U U U -shaped sequence from Lemma 4.4 are exact we have
ker i ′ = 0 , ker j ′ = im i ′ , and im j ′ = S α β γ γ . \ker i'=0,\quad\ker j'= \operatorname{im}i',\quad\text{and}\quad\operatorname{im}j'=S_{\alpha\beta}^{\gamma\gamma}. ker i ′ = 0 , ker j ′ = im i ′ , and im j ′ = S α β γ γ .
ker j ′ ′ = 0 , ker d = im j ′ ′ , ker i ′ ′ = im d , and im i ′ ′ = S α α β γ . \ker j''=0,\quad\ker d=\operatorname{im}j'',\quad\ker i''=\operatorname{im}d,\quad\text{and}\quad\operatorname{im}i''=S_{\alpha\alpha}^{\beta\gamma}. ker j ′′ = 0 , ker d = im j ′′ , ker i ′′ = im d , and im i ′′ = S α α β γ .
Now consider ker i \ker i ker i , i.e. i ′ i ′ ′ ( x ) = 0 i'i''(x)=0 i ′ i ′′ ( x ) = 0 if and only if i ′ ′ ( x ) = 0 i''(x)=0 i ′′ ( x ) = 0 since i ′ i' i ′ is injective. Therefore, ker i = ker i ′ ′ = im d \ker i=\ker i''=\operatorname{im}d ker i = ker i ′′ = im d . Similarly, ker j = ker j ′ ′ j ′ \ker j=\ker j''j' ker j = ker j ′′ j ′ , j ′ ′ j ′ ( x ) = 0 j''j'(x)=0 j ′′ j ′ ( x ) = 0 if and only if j ′ ( x ) = 0 j'(x)=0 j ′ ( x ) = 0 since j ′ ′ j'' j ′′ is injective. Therefore, ker j = ker j ′ ′ j ′ = ker j ′ = im i ′ = im i ′ ′ i ′ = im i \ker j=\ker j''j'=\ker j'=\operatorname{im}i'=\operatorname{im}i''i'=\operatorname{im}i ker j = ker j ′′ j ′ = ker j ′ = im i ′ = im i ′′ i ′ = im i , since i ′ ′ i'' i ′′ is surjective. Finally, ker d = im j ′ ′ = im j ′ ′ j ′ = im j \ker d=\operatorname{im}j''=\operatorname{im}j''j'=\operatorname{im}j ker d = im j ′′ = im j ′′ j ′ = im j , since j ′ j' j ′ is surjective, which proves that the triangle is exact. □ \square □
For a spectral system S S S over Q \mathbf{Q} Q we define the E E E -terms E α β = S α α β β E_{\alpha}^{\beta}=S_{\alpha\alpha}^{\beta\beta} E α β = S α α β β , which then define a Cartan-Eilenberg system E ( S ) \mathrm{E}(S) E ( S ) over Q \mathbf{Q} Q , with connecting homomorphisms k = d : S β β γ γ → S α α β β k=d:S_{\beta\beta}^{\gamma\gamma}\to S_{\alpha\alpha}^{\beta\beta} k = d : S β β γ γ → S α α β β . From Definition 4.1 (iii) it follows that
S α β γ δ ≅ ker : S β β γ γ → d S α α β β im : S γ γ δ δ → d S β β γ γ = ker : E β γ → k E α β im : E γ δ → k E β γ = H ( S β β γ γ ) = H ( E β γ ) , S_{\alpha\beta}^{\gamma\delta}\cong\frac{\ker:S_{\beta\beta}^{\gamma\gamma}\xrightarrow{d}S_{\alpha\alpha}^{\beta\beta}}{\operatorname{im}:S_{\gamma\gamma}^{\delta\delta}\xrightarrow{d}S_{\beta\beta}^{\gamma\gamma}}
=\frac{\ker:E_{\beta}^{\gamma}\xrightarrow{k}E_{\alpha}^{\beta}}{\operatorname{im}:E_{\gamma}^{\delta}\xrightarrow{k}E_{\beta}^{\gamma}}
=H(S_{\beta\beta}^{\gamma\gamma})=H(E_{\beta}^{\gamma}), S α β γ δ ≅ im : S γ γ δ δ d S β β γ γ ker : S β β γ γ d S α α β β = im : E γ δ k E β γ ker : E β γ k E α β = H ( S β β γ γ ) = H ( E β γ ) ,
by using octuples ( α , α , β , β , γ , γ , δ , δ ) ∈ Q 8 (\alpha,\alpha,\beta,\beta,\gamma,\gamma,\delta,\delta)\in\mathbf{Q}_8 ( α , α , β , β , γ , γ , δ , δ ) ∈ Q 8 . The conditions for Cartan-Eilenberg systems follow directly from Definition 4.1 and Lemma 4.5.
Example 4.6 In a linear order P \mathbf{P} P given by 1 < 2 < ⋯ < p < ⋯ < n 1<2<\cdots<p<\cdots<n 1 < 2 < ⋯ < p < ⋯ < n , where n = # P n=\#\mathbf{P} n = # P , the lattice of down-sets Q = O ( P ) \mathbf{Q}=\mathcal{O}(\mathbf{P}) Q = O ( P ) is given by 0 ⪯ α 1 ⪯ α 2 ⪯ ⋯ ⪯ α n ≺ 1 0\preceq\alpha_1\preceq\alpha_2\preceq\cdots\preceq\alpha_n\prec1 0 ⪯ α 1 ⪯ α 2 ⪯ ⋯ ⪯ α n ≺ 1 ,16 where α p = { 1 , ⋯ , p } \alpha_p=\{1,\cdots,p\} α p = { 1 , ⋯ , p } , 0 = ∅ 0=\varnothing 0 = ∅ , and 1 = { 1 , ⋯ , n } 1=\{1,\cdots,n\} 1 = { 1 , ⋯ , n } . For technical reasons it is convenient to take p ∈ Z p\in\mathbf{Z} p ∈ Z for indexing the S S S -terms and define
α p = { 0 for p ≤ 0 , 1 for p ≥ n . \alpha_p=
\begin{cases}
0 & \text{for }p\leq0,\\
1 & \text{for }p\geq n.
\end{cases} α p = { 0 1 for p ≤ 0 , for p ≥ n .
The S S S -terms S α p − r α p α p − 1 α p α p − 1 + r S_{\alpha_{p-r}^{\alpha_p\alpha_{p-1}}}^{\alpha_p\alpha_{p-1+r}} S α p − r α p α p − 1 α p α p − 1 + r are the standard terms in a spectral sequence, where r r r is the page number.
The S S S -terms for Cartan-Eilenberg systems
For a Cartan-Eilenberg system E \mathrm{E} E over Q \mathbf{Q} Q we follow [16 ] for the construction of an associated spectral system S ( E ) \mathrm{S}(\mathrm{E}) S ( E ) . Define the S S S -terms:
S α β γ δ = ker : E β γ → k E α β im : E γ δ → k E β γ , ∀ ( α , β , γ , δ ) ∈ Q 4 . (22) S_{\alpha\beta}^{\gamma\delta}=
\frac{\ker:E_{\beta}^{\gamma}\xrightarrow{k}E_{\alpha}^{\beta}}
{\operatorname{im}:E_{\gamma}^{\delta}\xrightarrow{k}E_{\beta}^{\gamma}},
\quad\forall(\alpha,\beta,\gamma,\delta)\in\mathbf{Q}_4. \tag*{(22)} S α β γ δ = im : E γ δ k E β γ ker : E β γ k E α β , ∀ ( α , β , γ , δ ) ∈ Q 4 . (22)
Here k k k denotes the connecting homomorphisms of E \mathrm{E} E . For ( α , β , γ , δ ) ⪯ ( α ′ , β ′ , γ ′ , δ ′ ) (\alpha,\beta,\gamma,\delta)\preceq(\alpha',\beta',\gamma',\delta') ( α , β , γ , δ ) ⪯ ( α ′ , β ′ , γ ′ , δ ′ ) in Q 4 \mathbf{Q}_4 Q 4 the homomorphisms ℓ : E γ γ ′ → E β ′ β ′ \ell:E_{\gamma}^{\gamma'}\to E_{\beta'}^{\beta'} ℓ : E γ γ ′ → E β ′ β ′ induce homomorphisms:
ℓ : S α β γ δ → S α ′ β ′ γ ′ δ ′ \ell:S_{\alpha\beta}^{\gamma\delta}\to S_{\alpha'\beta'}^{\gamma'\delta'} ℓ : S α β γ δ → S α ′ β ′ γ ′ δ ′
v. [16 Sect. 4] . The connecting homomorphisms k k k on the E E E -terms define differentials d : S γ δ ξ ζ → S α β γ δ d:S_{\gamma\delta}^{\xi\zeta}\to S_{\alpha\beta}^{\gamma\delta} d : S γ δ ξ ζ → S α β γ δ on the S S S -terms for sextuples ( α , β , γ , δ , ϵ , ζ ) ∈ Q 6 (\alpha,\beta,\gamma,\delta,\epsilon,\zeta)\in Q_6 ( α , β , γ , δ , ϵ , ζ ) ∈ Q 6 . The differentials commute with the ℓ \ell ℓ -homomorphisms: ℓ ∘ d = d ∘ ℓ \ell\circ d=d\circ\ell ℓ ∘ d = d ∘ ℓ . The S S S -terms together with the above defined differentials yield a spectral system denoted by S ( E ) \mathbf{S}(E) S ( E ) , v. [16 Sect. 4] . A morphism h : S → S ′ h:S\to S' h : S → S ′ of spectral systems over Q Q Q consists of homomorphisms h : S α β γ δ → S α β ′ γ δ h:S_{\alpha\beta}^{\gamma\delta}\to S_{\alpha\beta}^{\prime\gamma\delta} h : S α β γ δ → S α β ′ γ δ that commute with the homomorphisms ℓ \ell ℓ and with the differentials d d d ; h h h is an isomorphism if all these homomorphisms are isomorphisms. Combining this with Lemma 4.5 we obtain the following duality theorem:
Theorem 4.7 Let S S S be a spectral system and let E E E be a Cartan–Eilenberg system, both over a poset Q Q Q . Then,
S ( E ( S ) ) ≅ S , and E ( S ( E ) ) = E . S(E(S))\cong S,\qquad\text{and}\qquad E(S(E))=E. S ( E ( S )) ≅ S , and E ( S ( E )) = E .
Proof Let E E E be a Cartan–Eilenberg system. By (22) ,
S α α β β ( E ) = ker ( E α β ⟶ k E α α ) / im ( E β β ⟶ k E α β ) = E α β , S_{\alpha\alpha}^{\beta\beta}(E)=\ker\left(E_{\alpha}^{\beta}\mathrel{\mathop{\longrightarrow}^{k}}E_{\alpha}^{\alpha}\right)/\operatorname{im}\left(E_{\beta}^{\beta}\mathrel{\mathop{\longrightarrow}^{k}}E_{\alpha}^{\beta}\right)=E_{\alpha}^{\beta}, S α α β β ( E ) = ker ( E α β ⟶ k E α α ) / im ( E β β ⟶ k E α β ) = E α β ,
since E α α = E β β = 0 E_{\alpha}^{\alpha}=E_{\beta}^{\beta}=0 E α α = E β β = 0 , and the induced homomorphisms ℓ \ell ℓ and differentials d d d on these terms are the homomorphisms ℓ \ell ℓ and k k k of E E E . This implies that E ( S ( E ) ) = E E(S(E))=E E ( S ( E )) = E .
Let S S S be a spectral system and E = E ( S ) E=E(S) E = E ( S ) , i.e. E α β = S α α β β E_{\alpha}^{\beta}=S_{\alpha\alpha}^{\beta\beta} E α β = S α α β β . For ( α , β , γ , δ ) ∈ Q 4 (\alpha,\beta,\gamma,\delta)\in Q_4 ( α , β , γ , δ ) ∈ Q 4 , Remark 4.2 with the octuple ( α , α , β , β , γ , γ , δ , δ ) (\alpha,\alpha,\beta,\beta,\gamma,\gamma,\delta,\delta) ( α , α , β , β , γ , γ , δ , δ ) gives isomorphisms
h : S α β γ δ → ≅ H ( S β β γ γ ) = S α β γ δ ( E ) , x ↦ [ ℓ ( y ) ] , h:S_{\alpha\beta}^{\gamma\delta}\xrightarrow{\cong}H(S_{\beta\beta}^{\gamma\gamma})=S_{\alpha\beta}^{\gamma\delta}(E),\qquad x\mapsto[\ell(y)], h : S α β γ δ ≅ H ( S β β γ γ ) = S α β γ δ ( E ) , x ↦ [ ℓ ( y )] ,
where y ∈ S α β γ γ y\in S_{\alpha\beta}^{\gamma\gamma} y ∈ S α β γ γ with ℓ ( y ) = x \ell(y)=x ℓ ( y ) = x . On the E E E -terms, h h h is the identity. Since h h h is defined by homomorphisms ℓ \ell ℓ , it commutes with the homomorphisms ℓ \ell ℓ , by (ii). For the differentials, (iv) applied to ( α , α , γ , γ , ϵ , ϵ ) ≤ ( α , β , γ , δ , ϵ , ζ ) (\alpha,\alpha,\gamma,\gamma,\epsilon,\epsilon)\leq(\alpha,\beta,\gamma,\delta,\epsilon,\zeta) ( α , α , γ , γ , ϵ , ϵ ) ≤ ( α , β , γ , δ , ϵ , ζ ) gives d ∘ ℓ = ℓ ∘ k d\circ\ell=\ell\circ k d ∘ ℓ = ℓ ∘ k on E γ ϵ = S γ γ ϵ ϵ E_{\gamma}^{\epsilon}=S_{\gamma\gamma}^{\epsilon\epsilon} E γ ϵ = S γ γ ϵϵ , where ℓ : E γ ϵ → S γ δ ϵ ζ \ell:E_{\gamma}^{\epsilon}\to S_{\gamma\delta}^{\epsilon\zeta} ℓ : E γ ϵ → S γ δ ϵ ζ is surjective by (v) and (iii). The same relation holds in S ( E ) S(E) S ( E ) , which satisfies (iv) and has the same E E E -terms and connecting homomorphisms k k k . Since h h h commutes with ℓ \ell ℓ and is the identity on E E E -terms, it follows that h ∘ d = d ∘ h h\circ d=d\circ h h ∘ d = d ∘ h . Hence h : S → S ( E ( S ) ) h:S\to S(E(S)) h : S → S ( E ( S )) is an isomorphism of spectral systems. □ \square □
If Q = L Q=L Q = L is a modular lattice, then a spectral system is said to be excisive if E ( S ) E(S) E ( S ) is an excisive Cartan–Eilenberg system over L L L . In terms of the S S S -terms this is equivalent, for L = O ( P ) L=O(P) L = O ( P ) , to
ℓ : S α β γ δ → ≅ S α ′ β ′ γ ′ δ ′ , ∀ ( α , β , γ , δ ) ≤ ( α ′ , β ′ , γ ′ , δ ′ ) , \ell:S_{\alpha\beta}^{\gamma\delta}\xrightarrow{\cong}S_{\alpha'\beta'}^{\gamma'\delta'},\qquad\forall(\alpha,\beta,\gamma,\delta)\leq(\alpha',\beta',\gamma',\delta'), ℓ : S α β γ δ ≅ S α ′ β ′ γ ′ δ ′ , ∀ ( α , β , γ , δ ) ≤ ( α ′ , β ′ , γ ′ , δ ′ ) ,
with β \ α = β ′ \ α ′ \beta\backslash\alpha=\beta'\backslash\alpha' β \ α = β ′ \ α ′ , γ \ β = γ ′ \ β ′ \gamma\backslash\beta=\gamma'\backslash\beta' γ \ β = γ ′ \ β ′ , and δ \ γ = δ ′ \ γ ′ \delta\backslash\gamma=\delta'\backslash\gamma' δ \ γ = δ ′ \ γ ′ , v. [16 Lem. 4.16] . In other words, in an excisive S S S -system the S S S -terms S α β γ δ S_{\alpha\beta}^{\gamma\delta} S α β γ δ only depend on the convex sets β \ α \beta\backslash\alpha β \ α , γ \ β \gamma\backslash\beta γ \ β , and δ \ γ \delta\backslash\gamma δ \ γ .
Convergence and extensions
A spectral system S S S over Q Q Q may have various convergence properties. A poset Q Q Q is bounded if it contains unique elements 0 0 0 and 1 1 1 such that 0 ≤ q ≤ 1 0\leq q\leq1 0 ≤ q ≤ 1 for all $q\in Q.17 T h i s a l l o w s t h e c o n s t r u c t i o n o f g r a d e d v e r s i o n s o f t h e p a g e − 1 This allows the construction of graded versions of the page-1 T hi s a l l o w s t h eco n s t r u c t i o n o f g r a d e d v er s i o n so f t h e p a g e − 1 E-terms, v. [16 ] . In this setting the page-\inftyt e r m s a r e g i v e n b y terms are given by t er m s a r e g i v e nb y S_{\alpha\alpha}^{\beta 1}. T h e p o s e t . The poset . T h e p ose t Qi n d u c e s t h e f o l l o w i n g f i l t e r i n g o f induces the following filtering of in d u ces t h e f o l l o w in g f i l t er in g o f E_0^1$:
α ↦ F α E 0 1 = im ( E 0 α → i E 0 1 ) , with α ≤ β ⟹ F α E 0 1 ⊆ F β E 0 1 , \alpha\mapsto F_\alpha E_0^1=\operatorname{im}\left(E_0^\alpha\xrightarrow{i}E_0^1\right),\qquad\text{with }\alpha\leq\beta\Longrightarrow F_\alpha E_0^1\subseteq F_\beta E_0^1, α ↦ F α E 0 1 = im ( E 0 α i E 0 1 ) , with α ≤ β ⟹ F α E 0 1 ⊆ F β E 0 1 ,
which follows from the composition of the homomorphisms i i i , v. Defn. 3.1 (ii). The poset filtering of E 0 1 E_0^1 E 0 1 yields graded versions of E 0 1 E_0^1 E 0 1 . Following [16 ] the spectral system yields the short exact sequences, for α ≤ β \alpha\le\beta α ≤ β :
\begin{aligned}
0 &\longrightarrow S_{00}^{\alpha1} \longrightarrow S_{00}^{\beta1} \longrightarrow S_{0\alpha}^{\beta1} \longrightarrow0 \\
&\phantom{\longrightarrow}\cong\phantom{\longrightarrow}\cong\phantom{\longrightarrow}\cong\\
0 &\longrightarrow F_{\alpha}E_{0}^{1} \lhook\joinrel\longrightarrow F_{\beta}E_{0}^{1} \longrightarrow F_{\beta}E_{0}^{1}/F_{\alpha}E_{0}^{1} \longrightarrow0
\end{aligned}
A short exact sequence describes an extension F β E 0 1 F_{\beta}E_{0}^{1} F β E 0 1 of F β E 0 1 / F α E 0 1 F_{\beta}E_{0}^{1}/F_{\alpha}E_{0}^{1} F β E 0 1 / F α E 0 1 by F α E 0 1 F_{\alpha}E_{0}^{1} F α E 0 1 . Extensions are captured by the abelian group Ext R 1 ( S 0 α β 1 , S 00 α 1 ) \operatorname{Ext}_{R}^{1}(S_{0\alpha}^{\beta1},S_{00}^{\alpha1}) Ext R 1 ( S 0 α β 1 , S 00 α 1 ) . For example if S 0 α β 1 S_{0\alpha}^{\beta1} S 0 α β 1 is projective, then Ext R 1 ( S 0 α β 1 , S 00 α 1 ) = 0 \operatorname{Ext}_{R}^{1}(S_{0\alpha}^{\beta1},S_{00}^{\alpha1})=0 Ext R 1 ( S 0 α β 1 , S 00 α 1 ) = 0 , and therefore the exact sequence is split. This happens for example if we consider spectral systems in the category Vect K _K K .
A similar construction exists for the quotients E 0 1 / F α E 0 1 E_{0}^{1}/F_{\alpha}E_{0}^{1} E 0 1 / F α E 0 1 , for Q = O ( P ) Q=O(P) Q = O ( P ) . Define
α ↦ F α E 0 1 = coim : E 0 1 → j E α 1 = coker : E 0 α → i E 0 1 , α ‾ = P \ α . \alpha\mapsto F_{\alpha}E_{0}^{1}=\operatorname{coim}:E_{0}^{1}\xrightarrow{j}E_{\alpha}^{1}=\operatorname{coker}:E_{0}^{\alpha}\xrightarrow{i}E_{0}^{1},\qquad\overline{\alpha}=P\backslash\alpha. α ↦ F α E 0 1 = coim : E 0 1 j E α 1 = coker : E 0 α i E 0 1 , α = P \ α .
Following [16 ] we obtain the short exact sequences, for α ≤ β \alpha\le\beta α ≤ β :
0 ⟶ S 0 α β 1 ⟶ S 0 α 11 ⟶ S 0 β 11 ⟶ 0 ⟶ ≅ ⟶ ≅ ⟶ ≅ 0 ⟶ F β E 0 1 / F α E 0 1 ⟶ F α ‾ E 0 1 ⟶ F β ‾ E 0 1 ⟶ 0 \begin{aligned}
0 &\longrightarrow S_{0\alpha}^{\beta1} \longrightarrow S_{0\alpha}^{1 1} \longrightarrow S_{0\beta}^{1 1} \longrightarrow0 \\
&\phantom{\longrightarrow}\cong\phantom{\longrightarrow}\cong\phantom{\longrightarrow}\cong\\
0 &\longrightarrow F_{\beta}E_{0}^{1}/F_{\alpha}E_{0}^{1} \longrightarrow F_{\overline{\alpha}}E_{0}^{1} \longrightarrow F_{\overline{\beta}}E_{0}^{1} \longrightarrow0
\end{aligned} 0 0 ⟶ S 0 α β 1 ⟶ S 0 α 11 ⟶ S 0 β 11 ⟶ 0 ⟶ ≅ ⟶ ≅ ⟶ ≅ ⟶ F β E 0 1 / F α E 0 1 ⟶ F α E 0 1 ⟶ F β E 0 1 ⟶ 0
Remark 4.8 The same filtering can be applied to any E E E -term. This way we obtain grading information about a given E E E -term. Given a quadruple ( α , β , γ , δ ) ∈ Q 4 (\alpha,\beta,\gamma,\delta)\in Q_{4} ( α , β , γ , δ ) ∈ Q 4 , then
β ↦ F β E α δ = im : E α β → i E α δ ≃ S α α β δ , and F γ E α δ / F β E α δ ≃ S α β γ δ . \beta\mapsto F_{\beta}E_{\alpha}^{\delta}=\operatorname{im}:E_{\alpha}^{\beta}\xrightarrow{i}E_{\alpha}^{\delta}\simeq S_{\alpha\alpha}^{\beta\delta},\qquad\text{and}\qquad F_{\gamma}E_{\alpha}^{\delta}/F_{\beta}E_{\alpha}^{\delta}\simeq S_{\alpha\beta}^{\gamma\delta}. β ↦ F β E α δ = im : E α β i E α δ ≃ S α α β δ , and F γ E α δ / F β E α δ ≃ S α β γ δ .
Consider the short exact sequences
0 ⟶ S α β γ ε → i S α β δ ε → j S α γ δ ε ⟶ 0 , 0\longrightarrow S_{\alpha\beta}^{\gamma\varepsilon}\xrightarrow{i}S_{\alpha\beta}^{\delta\varepsilon}\xrightarrow{j}S_{\alpha\gamma}^{\delta\varepsilon}\longrightarrow0, 0 ⟶ S α β γ ε i S α β δ ε j S α γ δ ε ⟶ 0 ,
from (16) . The extensions of S α γ δ ε S_{\alpha\gamma}^{\delta\varepsilon} S α γ δ ε by S α β γ ε S_{\alpha\beta}^{\gamma\varepsilon} S α β γ ε are described by the extension groups Ext R 1 ( S α γ δ ε , S α β γ ε ) \operatorname{Ext}_{R}^{1}(S_{\alpha\gamma}^{\delta\varepsilon},S_{\alpha\beta}^{\gamma\varepsilon}) Ext R 1 ( S α γ δ ε , S α β γ ε ) , which are related by associated distinguished triangles in the derived category D h ( Mod R ) \mathbf{D}_{h}(\operatorname{Mod}_{R}) D h ( Mod R ) (of complexes):
S α β γ ε → i S α β δ ε → j S α γ δ ε → κ S α β γ ε [ 1 ] , (23) S_{\alpha\beta}^{\gamma\varepsilon}\xrightarrow{i}S_{\alpha\beta}^{\delta\varepsilon}\xrightarrow{j}S_{\alpha\gamma}^{\delta\varepsilon}\xrightarrow{\kappa}S_{\alpha\beta}^{\gamma\varepsilon}[1], \tag*{(23)} S α β γ ε i S α β δ ε j S α γ δ ε κ S α β γ ε [ 1 ] , (23)
where the S S S -terms are regarded as complexes concentrated at n = 0 n=0 n = 0 , with trivial differentials, and κ ∈ Hom D h ( Mod R ) ( S α γ δ ε , S α β γ ε [ 1 ] ) ≃ Ext R 1 ( S α γ δ ε , S α β γ ε ) \kappa\in\operatorname{Hom}_{\mathbf{D}_{h}(\operatorname{Mod}_{R})}(S_{\alpha\gamma}^{\delta\varepsilon},S_{\alpha\beta}^{\gamma\varepsilon}[1])\simeq\operatorname{Ext}_{R}^{1}(S_{\alpha\gamma}^{\delta\varepsilon},S_{\alpha\beta}^{\gamma\varepsilon}) κ ∈ Hom D h ( Mod R ) ( S α γ δ ε , S α β γ ε [ 1 ]) ≃ Ext R 1 ( S α γ δ ε , S α β γ ε ) which is not induced by an actual homomorphism in general. If S α γ δ ε S_{\alpha\gamma}^{\delta\varepsilon} S α γ δ ε is projective, then κ = 0 \kappa=0 κ = 0 . Otherwise, to determine κ \kappa κ we consider a projective resolution for S α γ δ ε S_{\alpha\gamma}^{\delta\varepsilon} S α γ δ ε , denoted by
P ∙ → ≃ π S α γ δ ε , P_{\bullet}\xrightarrow[\simeq]{\pi}S_{\alpha\gamma}^{\delta\varepsilon}, P ∙ π ≃ S α γ δ ε ,
which is a quasi-isomorphism and S α γ δ ε ∈ Ch ( Mod R ) S_{\alpha\gamma}^{\delta\varepsilon}\in\operatorname{Ch}(\operatorname{Mod}_{R}) S α γ δ ε ∈ Ch ( Mod R ) , concentrated at $n=0. Consider the commutative diagram:
⋯ ⟶ P n → d n ⋯ ⟶ P 2 → d 2 P 1 → d 1 P 0 → π S α γ δ ε ⟶ 0 ⟶ ↓ 0 ⟶ ↓ ⟶ ↓ 0 ⟶ ↓ ⟶ ↓ ⋯ ⟶ 0 ⟶ ⋯ ⟶ 0 ⟶ S α β γ ε → i S α β δ ε → j S α γ δ ε ⟶ 0 \begin{aligned}
\cdots&\longrightarrow P_{n}\xrightarrow{d_{n}}\cdots\longrightarrow P_{2}\xrightarrow{d_{2}}P_{1}\xrightarrow{d_{1}}P_{0}\xrightarrow{\pi}S_{\alpha\gamma}^{\delta\varepsilon}\longrightarrow0\\
&\phantom{\longrightarrow}\downarrow0\phantom{\longrightarrow}\phantom{\downarrow}\phantom{\longrightarrow}\downarrow0\phantom{\longrightarrow}\phantom{\downarrow}\phantom{\longrightarrow}\mathrel{\phantom{\downarrow}} \\
\cdots&\longrightarrow0\longrightarrow\cdots\longrightarrow0\longrightarrow S_{\alpha\beta}^{\gamma\varepsilon}\xrightarrow{i}S_{\alpha\beta}^{\delta\varepsilon}\xrightarrow{j}S_{\alpha\gamma}^{\delta\varepsilon}\longrightarrow0
\end{aligned} ⋯ ⋯ ⟶ P n d n ⋯ ⟶ P 2 d 2 P 1 d 1 P 0 π S α γ δ ε ⟶ 0 ⟶ ↓ 0 ⟶↓⟶ ↓ 0 ⟶↓⟶ ↓ ⟶ 0 ⟶ ⋯ ⟶ 0 ⟶ S α β γ ε i S α β δ ε j S α γ δ ε ⟶ 0
Since P 0 P_{0} P 0 is projective and j j j is surjective there exists a lift f 0 : P 0 → S α β δ ε f_{0}:P_{0}\to S_{\alpha\beta}^{\delta\varepsilon} f 0 : P 0 → S α β δ ε such that j f 0 = π jf_{0}=\pi j f 0 = π . Observe that j f 0 d 1 = π d 1 = 0 jf_{0}d_{1}=\pi d_{1}=0 j f 0 d 1 = π d 1 = 0 and therefore im ( f 0 d 1 ) ⊆ ker j = im i \operatorname{im}(f_{0}d_{1})\subseteq\ker j=\operatorname{im}i im ( f 0 d 1 ) ⊆ ker j = im i . The map i i i is injective and since f 0 d 1 f_{0}d_{1} f 0 d 1 maps into im i \operatorname{im}i im i , the map f 1 : = i − 1 f 0 d 1 f_{1}:=i^{-1}f_{0}d_{1} f 1 := i − 1 f 0 d 1 is well-defined. The map f 1 : P 1 → S α β γ ε f_{1}:P_{1}\to S_{\alpha\beta}^{\gamma\varepsilon} f 1 : P 1 → S α β γ ε fits in the commutative square above. Since i f 1 d 2 = f 0 d 1 d 2 = 0 if_{1}d_{2}=f_{0}d_{1}d_{2}=0 i f 1 d 2 = f 0 d 1 d 2 = 0 and i i i is injective, f 1 d 2 = 0 f_{1}d_{2}=0 f 1 d 2 = 0 , so that f 1 f_{1} f 1 provides an actual chain map P ∙ → S α β γ ε [ 1 ] P_{\bullet} \to S_{\alpha\beta}^{\gamma\varepsilon}[1] P ∙ → S α β γ ε [ 1 ] . In a diagram this reads:
\begin{CD}
\cdots@>>> P_n @>{d_n}>> \cdots@>>> P_2 @>{d_2}>> P_1 @>{d_1}>> P_0 \\
@. @V0VV @. @V0VV @. @V{f_1}VV @. @V0VV \\
\cdots@>>> 0 @>>> \cdots@>>> 0 @>>> S_{\alpha\beta}^{\gamma\varepsilon} @>>> 0
\end{CD}
The morphism κ : S α γ δ ε → S α β γ ε [ 1 ] \kappa:S_{\alpha\gamma}^{\delta\varepsilon}\to S_{\alpha\beta}^{\gamma\varepsilon}[1] κ : S α γ δ ε → S α β γ ε [ 1 ] is represented by the homotopy class of the chain map f 1 : P ∙ → S α β γ ε [ 1 ] f_1:P_{\bullet}\to S_{\alpha\beta}^{\gamma\varepsilon}[1] f 1 : P ∙ → S α β γ ε [ 1 ] , composed with the inverse of the quasi-isomorphism π \pi π . As we pointed out, κ \kappa κ represents a class in Ext R 1 ( S α γ δ ε , S α β γ ε ) \operatorname{Ext}^{1}_{R}(S_{\alpha\gamma}^{\delta\varepsilon},S_{\alpha\beta}^{\gamma\varepsilon}) Ext R 1 ( S α γ δ ε , S α β γ ε ) , and κ = 0 \kappa=0 κ = 0 if and only if f 1 ∼ 0 f_1\sim0 f 1 ∼ 0 , which is equivalent to f 1 − 0 = σ 0 d 1 + 0 ⋅ σ 0 f_1-0=\sigma_0d_1+0\cdot\sigma_0 f 1 − 0 = σ 0 d 1 + 0 ⋅ σ 0 , i.e. f 1 = σ 0 d 1 f_1=\sigma_0d_1 f 1 = σ 0 d 1 for some σ 0 : P 0 → S α β γ ε \sigma_0:P_0\to S_{\alpha\beta}^{\gamma\varepsilon} σ 0 : P 0 → S α β γ ε . Consequently, κ ≠ 0 \kappa\ne0 κ = 0 if and only if the short exact sequence (16) does not split. The homomorphism (in Dh ( Mod R ) \operatorname{Dh}(\operatorname{Mod}_{R}) Dh ( Mod R ) ) κ : S α γ δ ε → S α β γ ε [ 1 ] \kappa:S_{\alpha\gamma}^{\delta\varepsilon}\to S_{\alpha\beta}^{\gamma\varepsilon}[1] κ : S α γ δ ε → S α β γ ε [ 1 ] will be referred to as a (higher order) derived differential between S S S -terms.
More explicitly, for an arbitrary ring R R R the following statements are equivalent:
κ = 0 \kappa=0 κ = 0 ;
f 1 = σ 0 d 1 f_1=\sigma_0d_1 f 1 = σ 0 d 1 for some σ 0 : P 0 → S α β γ ε \sigma_0:P_0\to S_{\alpha\beta}^{\gamma\varepsilon} σ 0 : P 0 → S α β γ ε ;
the short exact sequence in (16) splits, i.e. there exists a homomorphism s : S α γ δ ε → S α β δ ε s:S_{\alpha\gamma}^{\delta\varepsilon}\to S_{\alpha\beta}^{\delta\varepsilon} s : S α γ δ ε → S α β δ ε with j s = id js=\operatorname{id} j s = id , or, equivalently, a homomorphism r : S α β δ ε → S α β γ ε r:S_{\alpha\beta}^{\delta\varepsilon}\to S_{\alpha\beta}^{\gamma\varepsilon} r : S α β δ ε → S α β γ ε with r i = id ri=\operatorname{id} r i = id .
The equivalence of (i) and (ii) was shown above. The equivalence of (i) and (iii) holds since κ \kappa κ represents the class of the extension (16) in Ext R 1 ( S α γ δ ε , S α β γ ε ) \operatorname{Ext}^{1}_{R}(S_{\alpha\gamma}^{\delta\varepsilon},S_{\alpha\beta}^{\gamma\varepsilon}) Ext R 1 ( S α γ δ ε , S α β γ ε ) , and this class vanishes if and only if the extension splits, v. [25 Thm. 3.4.3] .
Remark 4.9 If S α γ δ ε S_{\alpha\gamma}^{\delta\varepsilon} S α γ δ ε is projective, then the projective resolution can be chosen trivial and κ = 0 \kappa=0 κ = 0 follows immediately. Indeed, P 0 ≅ S α γ δ ε P_0\cong S_{\alpha\gamma}^{\delta\varepsilon} P 0 ≅ S α γ δ ε and P n = 0 P_n=0 P n = 0 , d n = 0 d_n=0 d n = 0 for n ≥ 1 n\ge1 n ≥ 1 . This implies that f 1 = 0 f_1=0 f 1 = 0 .
Spectral nets
In this paper we restrict ourselves to Cartan–Eilenberg systems over a finite distributive lattice L L L , which we represent, using Birkhoff’s representation theorem, as the lattice O ( P ) \operatorname{O}(P) O ( P ) of down-sets of a finite poset P P P . Throughout this section, E E E is an excisive Cartan–Eilenberg system over O ( P ) \operatorname{O}(P) O ( P ) , v. Sect. 3.1, and S = S ( E ) S=S(E) S = S ( E ) is the associated spectral system, v. Thm. 4.7 . Since E E E is excisive also S S S is excisive. A way to obtain the “limiting” S S S -terms in an excisive spectral system is to define a generalization of spectral sequences to lattices, which we will refer to as a spectral net. A maximal chain K K K in O ( P ) \operatorname{O}(P) O ( P ) is a linearly ordered subset of O ( P ) \operatorname{O}(P) O ( P ) (regarded as a poset) that is not contained in a larger chain in O ( P ) \operatorname{O}(P) O ( P ) . The length of a maximal chain K K K is equal to the number of elements ∣ P ∣ |P| ∣ P ∣ of P P P , since α ⊊ β \alpha\subsetneq\beta α ⊊ β is a cover in O ( P ) \operatorname{O}(P) O ( P ) if and only if β ∖ α \beta\setminus\alpha β ∖ α is a singleton. Therefore all maximal chains in O ( P ) \operatorname{O}(P) O ( P ) have the same length ∣ P ∣ |P| ∣ P ∣ (Jordan–Dedekind chain condition). As shown below, the number of maximal chains in O ( P ) \operatorname{O}(P) O ( P ) is equal to the number of linear extensions of P P P .18 Let K ⊆ O ( P ) K\subseteq\operatorname{O}(P) K ⊆ O ( P ) be a maximal chain. If K K K is represented as O ( P ′ ) \operatorname{O}(P') O ( P ′ ) , then the inclusion O ( P ′ ) → O ( P ) \operatorname{O}(P')\to\operatorname{O}(P) O ( P ′ ) → O ( P ) is an injective lattice homomorphism. Since P P P is a sublattice of O ( P ) \operatorname{O}(P) O ( P ) which contains 0 0 0 and 1 1 1 , the dual order-preserving map P → P ′ P\to P' P → P ′ is surjective. Since ∣ P ′ ∣ = ∣ K ∣ − 1 = ∣ P ∣ |P'|=|K|-1=|P| ∣ P ′ ∣ = ∣ K ∣ − 1 = ∣ P ∣ , the map P → P ′ P\to P' P → P ′ is bijective (in general not an order-isomorphism) and P ′ P' P ′ is a linear extension of the order on P P P . Conversely, a linear extension p 1 , … , p n p_1,\ldots,p_n p 1 , … , p n of P P P yields the maximal chain of down-sets { p 1 , … , p k } \{p_1,\ldots,p_k\} { p 1 , … , p k } , k = 0 , … , n k=0,\ldots,n k = 0 , … , n . This gives a one-to-one correspondence between maximal chains in O ( P ) O(P) O ( P ) and linear extensions of P P P . Define the following exact couple:
D 1 → i D 1 k ↑ ↓ j \begin{CD}
D^1 @>i>> D^1 \\
@A k AA @VV j V \\
& E^1 &
\end{CD} D 1 k ⏐ ↑ i D 1 ↓ ⏐ j
D 1 = ⨁ K ⊆ O ( P ) ⨁ α ∈ K E 0 α , E 1 = ⨁ K ⊆ O ( P ) ⨁ α ∈ K ∖ { 0 } E α ∗ α . D^1=\bigoplus_{K\subseteq O(P)}\ \bigoplus_{\alpha\in K} E^\alpha_0,\qquad E^1=\bigoplus_{K\subseteq O(P)}\ \bigoplus_{\alpha\in K\setminus\{0\}} E^{\alpha}_{\alpha^*}. D 1 = K ⊆ O ( P ) ⨁ α ∈ K ⨁ E 0 α , E 1 = K ⊆ O ( P ) ⨁ α ∈ K ∖ { 0 } ⨁ E α ∗ α .
where K K K ranges over the maximal chains in O ( P ) O(P) O ( P ) , and α ∗ \alpha^* α ∗ is the unique immediate predecessor of α \alpha α in K K K . The homomorphisms i i i , j j j and k k k are those of the exact triangles of E E E for the pairs α ∗ ⊆ α \alpha^*\subseteq\alpha α ∗ ⊆ α , v. Sect. 3.1, i.e. i : E 0 α → E 0 α ∗ i:E^{\alpha}_0\to E^{\alpha^*}_0 i : E 0 α → E 0 α ∗ , j : E 0 α ∗ → E α ∗ α j:E^{\alpha^*}_0\to E^{\alpha}_{\alpha^*} j : E 0 α ∗ → E α ∗ α and k : E α ∗ α → E 0 α ∗ k:E^{\alpha}_{\alpha^*}\to E^{\alpha^*}_0 k : E α ∗ α → E 0 α ∗ .19 The composition d 1 = j k d^1=jk d 1 = j k defines a (first order) differential d 1 : E 1 → E 1 d^1:E^1\to E^1 d 1 : E 1 → E 1 , and by the theory of exact couples, consecutive evaluation of homology with respect to the induced differentials yields the higher pages and differentials, i.e. E r + 1 = H ( E r , d r ) E^{r+1}=H(E^r,d^r) E r + 1 = H ( E r , d r ) , with d r = j i 1 − r k d^r=ji^{1-r}k d r = j i 1 − r k , where i 1 − r i^{1-r} i 1 − r denotes taking a preimage under i r − 1 i^{r-1} i r − 1 , which is well defined on E r E^r E r , r ≥ 1 r\geq1 r ≥ 1 , cf. Sect. 1.2 and [25 Sect. 5.9] .
Since E E E is excisive, there exist R R R -modules G p C ≅ E α β G_pC\cong E^\beta_\alpha G p C ≅ E α β , for all β ∖ α = { p } \beta\setminus\alpha=\{p\} β ∖ α = { p } and for all p ∈ P p\in P p ∈ P , with isomorphisms that are compatible with the homomorphisms ℓ \ell ℓ . For example, G p C = E α β G_pC=E^\beta_{\alpha} G p C = E α β , where β = ↓ p \beta=\mathord{\downarrow}p β = ↓ p is the principal down-set generated by p p p and α = β ′ = ↓ p ∖ { p } \alpha=\beta'=\mathord{\downarrow}p\setminus\{p\} α = β ′ = ↓ p ∖ { p } , with the isomorphisms ℓ : E α β → E α ′ β ′ \ell:E^\beta_\alpha\to E^{\beta'}_{\alpha'} ℓ : E α β → E α ′ β ′ , v. Sect. 3.1. Define
C 1 : = ⨁ P → P ′ ⨁ p ′ ∈ P ′ G p ′ C , C^1:=\bigoplus_{P\to P'}\ \bigoplus_{p'\in P'}G_{p'}C, C 1 := P → P ′ ⨁ p ′ ∈ P ′ ⨁ G p ′ C ,
where the outer sum ranges over all linear extensions P ′ P' P ′ of $P determined by the choice of maximal chain K K K . The isomorphisms from the excisive property can be used to define a differential d 1 d^1 d 1 on C 1 C^1 C 1 , as expressed in the diagram below:
\begin{CD}
E^\gamma_\beta@>{d_{\gamma\beta\alpha}}>> E^\beta_\alpha\\
@V{\ell}V{\cong}V @VV{\cong}V{\ell} \\
E^{\gamma'}_{\beta'} @>{d_{\gamma'\beta'\alpha'}}>> E^{\beta'}_{\alpha'} \\
@V{\cong}VV @VV{\cong}V \\
G_pC @>{d^1_{qp}}>> G_qC \tag*{(24)}
\end{CD}
Here γ ∖ β = γ ′ ∖ β ′ = { p } \gamma\setminus\beta=\gamma'\setminus\beta'=\{p\} γ ∖ β = γ ′ ∖ β ′ = { p } , β ∖ α = β ′ ∖ α ′ = { q } \beta\setminus\alpha=\beta'\setminus\alpha'=\{q\} β ∖ α = β ′ ∖ α ′ = { q } and ( α , β , γ ) ≤ ( α ′ , β ′ , γ ′ ) (\alpha,\beta,\gamma)\leq(\alpha',\beta',\gamma') ( α , β , γ ) ≤ ( α ′ , β ′ , γ ′ ) . The square commutes by the naturality of the differentials, and the triangles commute since the isomorphisms with G p C G_pC G p C and G q C G_qC G q C are compatible with ℓ \ell ℓ . Two such triples are both below their join, which is again such a triple. Hence d q p 1 d^1_{qp} d q p 1 does not depend on the choice of ( α , β , γ ) (\alpha,\beta,\gamma) ( α , β , γ ) . The isomorphisms G p C ≅ E α ∗ α G_pC\cong E^{\alpha}_{\alpha^*} G p C ≅ E α ∗ α , with α ∖ α ∗ = { p } \alpha\setminus\alpha^*=\{p\} α ∖ α ∗ = { p } , yield an isomorphism C 1 ≅ E 1 C^1\cong E^1 C 1 ≅ E 1 , under which d 1 d^1 d 1 on C 1 C^1 C 1 corresponds to d 1 = j k d^1=jk d 1 = j k on E 1 E^1 E 1 . Indeed, for consecutive elements α ⊆ β ⊆ γ \alpha\subseteq\beta\subseteq\gamma α ⊆ β ⊆ γ of K K K , the composition j k : E α β → E 0 β → E α β jk:E^\beta_\alpha\to E^\beta_0\to E^\beta_\alpha j k : E α β → E 0 β → E α β is the differential d γ β α d_{\gamma\beta\alpha} d γ β α , by the naturality of the exact triangles. Via this isomorphism the higher pages and differentials of the exact couple define C r C^r C r and d r d^r d r . The result is the (fundamental) spectral net C r + 1 = H ( C r , d r ) C^{r+1}=H(C^r,d^r) C r + 1 = H ( C r , d r ) for S S S , consisting of sub-quotients isomorphic to the S S S -terms and associated differentials. If K K K is written as α m \alpha_m α m , m ∈ Z m\in\mathbb{Z} m ∈ Z , as in Example 4.6, then the summand of C r C^r C r that corresponds to α m ∖ α m − 1 \alpha_m \setminus\alpha_{m-1} α m ∖ α m − 1 , 1 ≤ m ≤ ∣ P ∣ 1 \le m \le|P| 1 ≤ m ≤ ∣ P ∣ , is isomorphic to S α m − r ∖ α m − 1 α m ∖ α m − 1 + r S^{\alpha_m \setminus\alpha_{m-1}+r}_{\alpha_m-r \setminus\alpha_{m-1}} S α m − r ∖ α m − 1 α m ∖ α m − 1 + r , cf. Sect. 1.2. Since O ( P ) \operatorname{O}(P) O ( P ) is a finite lattice, the spectral net converges to C ∞ C^\infty C ∞ , i.e. C r = C ∞ C^r = C^\infty C r = C ∞ for r ≥ ∣ P ∣ r \ge|P| r ≥ ∣ P ∣ , which determines the S S S -terms S α β 1 S^{\beta1}_{\alpha} S α β 1 for all β ∖ α = { p } \beta\setminus\alpha= \{p\} β ∖ α = { p } and for all p ∈ P p \in P p ∈ P , and computes the graded object of E 0 1 E^1_0 E 0 1 with respect to the filtrations α ↦ F α E 0 1 \alpha\mapsto F_\alpha E^1_0 α ↦ F α E 0 1 , α ∈ K \alpha\in K α ∈ K , v. Sect. 4.3. For the fundamental spectral net no requirements on the ring R R R are needed. By considering sublattices of O ( P ) \operatorname{O}(P) O ( P ) we obtain associated spectral nets that reveal other higher differentials and E E E -terms, v. Defn. 5.1 below. Since a spectral net is defined over a partial order, there exist various additional relations between differentials, which are explained in detail in § 6.1 .
Definition 5.1 Let O ( Q ) ⊆ O ( P ) \operatorname{O}(Q) \subseteq\operatorname{O}(P) O ( Q ) ⊆ O ( P ) be a sublattice, where, by Birkhoff’s representation theorem, Q Q Q is the poset of join-irreducible elements of the sublattice. The restriction of E E E to O ( Q ) \operatorname{O}(Q) O ( Q ) is an excisive Cartan–Eilenberg system over O ( Q ) \operatorname{O}(Q) O ( Q ) , since meets and joins in O ( Q ) \operatorname{O}(Q) O ( Q ) are those in O ( P ) \operatorname{O}(P) O ( P ) . The spectral net over O ( Q ) \operatorname{O}(Q) O ( Q ) , or subnet , is the fundamental spectral net of this restriction, i.e. the above construction with Q Q Q in place of P P P , where the elements of O ( Q ) \operatorname{O}(Q) O ( Q ) are regarded as elements of O ( P ) \operatorname{O}(P) O ( P ) . For an interval O ( Q ) = [ α , β ] = { γ ∈ O ( P ) ∣ α ⊆ γ ⊆ β } \operatorname{O}(Q) = [\alpha,\beta] = \{\gamma\in\operatorname{O}(P) \mid\alpha\subseteq\gamma\subseteq\beta\} O ( Q ) = [ α , β ] = { γ ∈ O ( P ) ∣ α ⊆ γ ⊆ β } it is called the spectral net of the E E E -term E α β E^\beta_\alpha E α β .
Let 0 Q 0_Q 0 Q and 1 Q 1_Q 1 Q be the bottom and top elements of O ( Q ) \operatorname{O}(Q) O ( Q ) . The spectral net over O ( Q ) \operatorname{O}(Q) O ( Q ) converges to the graded object of E 0 Q 1 Q E^{1_Q}_{0_Q} E 0 Q 1 Q with respect to the filtrations along the maximal chains in O ( Q ) \operatorname{O}(Q) O ( Q ) . Its page 1 consists of the E E E -terms E α β E^\beta_\alpha E α β for covers α ⊊ β \alpha\subsetneq\beta α ⊊ β in O ( Q ) \operatorname{O}(Q) O ( Q ) , and the differentials between them. An interval [ α , β ] [\alpha,\beta] [ α , β ] is isomorphic to O ( β ∖ α ) \operatorname{O}(\beta\setminus\alpha) O ( β ∖ α ) via γ ↦ γ ∖ α \gamma\mapsto\gamma\setminus\alpha γ ↦ γ ∖ α , i.e. Q ≅ β ∖ α Q \cong\beta\setminus\alpha Q ≅ β ∖ α , which is a convex subset of P P P . In this case page 1 consists of the G p C G_pC G p C , p ∈ β ∖ α p \in\beta\setminus\alpha p ∈ β ∖ α , with the differentials d q p 1 d^1_{qp} d q p 1 of the fundamental spectral net. For other sublattices, β ∖ α \beta\setminus\alpha β ∖ α need not be a singleton for a cover α ⊊ β \alpha\subsetneq\beta α ⊊ β in O ( Q ) \operatorname{O}(Q) O ( Q ) , and page 1 contains E E E -terms and differentials which do not occur on page 1 of the fundamental spectral net.
Properties of first and higher order differentials
In this section we look at various properties of higher differentials and spectral matrices, and their interpretation for Morse representations. In this section the results are stated in the category of R R R -modules.
Perspective pairs
6.1 Perspective pairs. In [11 ] entries in a connection matrix can sometimes be computed using the fact that a connection matrix, or spectral matrix, is a differential. In principle a spectral matrix recovers all information about the spectral system. The interpretation of the entries of connection matrices in [11 ] only applies to the flow-order of adjacent Morse sets. In this section we demonstrate that such information can also be obtained from the spectral system as properties of a spectral net.
A perspective pair in O ( P ) \operatorname{O}(P) O ( P ) is a pair of triples ( α , β , δ ) (\alpha,\beta,\delta) ( α , β , δ ) , ( α , γ , δ ) ∈ O ( P ) 3 (\alpha,\gamma,\delta) \in\operatorname{O}(P)_3 ( α , γ , δ ) ∈ O ( P ) 3 such that β ∖ α = δ ∖ γ \beta\setminus\alpha= \delta\setminus\gamma β ∖ α = δ ∖ γ and γ ∖ α = δ ∖ β \gamma\setminus\alpha= \delta\setminus\beta γ ∖ α = δ ∖ β , which is equivalent to α = β ∩ γ \alpha= \beta\cap\gamma α = β ∩ γ and δ = β ∪ γ \delta= \beta\cup\gamma δ = β ∪ γ , v. Fig. 6.1. We assume in addition that there exist tail elements α − ⊊ α \alpha_- \subsetneq\alpha α − ⊊ α and δ ⊊ δ + \delta\subsetneq\delta_+ δ ⊊ δ + in O ( P ) \operatorname{O}(P) O ( P ) . We have the following exact triangles:
Figure 6.1.
Figure 6.1 (PDF p. 20). A perspective pair β γ α = δ γ \beta\gamma\alpha=\delta\gamma β γ α = δ γ and γ α = δ β \gamma\alpha=\delta\beta γ α = δ β [left] and the embedding into O(P) [right].
and isomorphisms (due to the excisive property):
ℓ : E α β → ≅ E γ β , j ′ i : E α β ⟶ E γ δ , \ell:E^\beta_\alpha\xrightarrow{\cong} E^\beta_\gamma,\qquad j'i:E^\beta_\alpha\longrightarrow E^\delta_\gamma, ℓ : E α β ≅ E γ β , j ′ i : E α β ⟶ E γ δ ,
ℓ ′ : E α γ → ≅ E γ β , j ′ i ′ : E α γ ⟶ E β δ . \ell':E^\gamma_\alpha\xrightarrow{\cong} E^\beta_\gamma,\qquad j'i':E^\gamma_\alpha\longrightarrow E^\delta_\beta. ℓ ′ : E α γ ≅ E γ β , j ′ i ′ : E α γ ⟶ E β δ .
By the functoriality of the homomorphisms ℓ \ell ℓ , v. Defn. 3.1 (ii), we have j ′ i = ℓ j'i=\ell j ′ i = ℓ and j ′ i ′ = ℓ ′ j'i'=\ell' j ′ i ′ = ℓ ′ , i.e.
ℓ − 1 j ′ i = id on E α β , ℓ ′ − 1 j ′ i ′ = id on E α γ , \ell^{-1}j'i=\operatorname{id}\text{ on }E^\beta_\alpha,\qquad\ell'^{-1}j'i'=\operatorname{id}\text{ on }E^\gamma_\alpha, ℓ − 1 j ′ i = id on E α β , ℓ ′ − 1 j ′ i ′ = id on E α γ ,
j ′ i ℓ − 1 = id on E γ δ , j ′ i ′ ℓ ′ − 1 = id on E β δ . j'i\ell^{-1}=\operatorname{id}\text{ on }E^\delta_\gamma,\qquad j'i'\ell'^{-1}=\operatorname{id}\text{ on }E^\delta_\beta. j ′ i ℓ − 1 = id on E γ δ , j ′ i ′ ℓ ′ − 1 = id on E β δ .
v. [21 ] . Consequently, i , i ′ i,i' i , i ′ are injective and j , j ′ j,j' j , j ′ are surjective. Hence we obtain split short exact sequences
0 ⟶ E α β → i E α δ → j E β δ ⟶ 0 0\longrightarrow E^\beta_\alpha\xrightarrow{i}E^\delta_\alpha\xrightarrow{j}E^\delta_\beta\longrightarrow0 0 ⟶ E α β i E α δ j E β δ ⟶ 0
0 ⟶ E α γ → i ′ E α δ → j ′ E γ δ ⟶ 0 0\longrightarrow E^\gamma_\alpha\xrightarrow{i'}E^\delta_\alpha\xrightarrow{j'}E^\delta_\gamma\longrightarrow0 0 ⟶ E α γ i ′ E α δ j ′ E γ δ ⟶ 0
For the connecting homomorphisms k k k and k ′ k' k ′ this implies: k = 0 k=0 k = 0 and k ′ = 0 k'=0 k ′ = 0 . For the differentials we have the following anti-commutativity relation:
Theorem 6.1 Consider a perspective pair given by ( α , β , δ ) , ( α , γ , δ ) ∈ O ( P ) 3 (\alpha,\beta,\delta),(\alpha,\gamma,\delta)\in\mathbf{O}(\mathbf{P})_3 ( α , β , δ ) , ( α , γ , δ ) ∈ O ( P ) 3 with tail elements α − \alpha_- α − and δ + \delta_+ δ + . Then, the differentials d δ + δ β : E δ δ + → E β δ d_{\delta_+\delta_\beta}:E^{\delta_+}_\delta\to E^\delta_\beta d δ + δ β : E δ δ + → E β δ and d δ + δ γ : E δ δ + → E γ δ d_{\delta_+\delta_\gamma}:E^{\delta_+}_\delta\to E^\delta_\gamma d δ + δ γ : E δ δ + → E γ δ and d β α α − : E α β → E α − α d_{\beta\alpha\alpha_-}:E^\beta_\alpha\to E^\alpha_{\alpha_-} d β α α − : E α β → E α − α and d γ α α − : E α γ → E α − α d_{\gamma\alpha\alpha_-}:E^\gamma_\alpha\to E^\alpha_{\alpha_-} d γ α α − : E α γ → E α − α satisfy
d β α α − ℓ − 1 d δ + δ γ + d γ α α − ℓ ′ − 1 d δ + δ β = 0 (25) d_{\beta\alpha\alpha_-}\ell^{-1}d_{\delta_+\delta_\gamma}+d_{\gamma\alpha\alpha_-}\ell'^{-1}d_{\delta_+\delta_\beta}=0 \tag*{(25)} d β α α − ℓ − 1 d δ + δ γ + d γ α α − ℓ ′ − 1 d δ + δ β = 0 (25)
Proof. Consider the differentials
E δ δ + → d δ + δ α E α δ → d δ α α − E α − α , with d δ α α − d δ + δ α = 0 E^{\delta_+}_\delta\xrightarrow{d_{\delta_+\delta_\alpha}}E^\delta_\alpha\xrightarrow{d_{\delta_\alpha\alpha_-}}E^\alpha_{\alpha_-},\qquad\text{with }d_{\delta_\alpha\alpha_-}d_{\delta_+\delta_\alpha}=0 E δ δ + d δ + δ α E α δ d δ α α − E α − α , with d δ α α − d δ + δ α = 0
The two factorizations of E α δ E^\delta_\alpha E α δ give the diagrams
E δ δ + → d δ + δ α E α δ = i E α β ⊕ i ′ ℓ ′ − 1 E β δ E^{\delta_+}_\delta\xrightarrow{d_{\delta_+\delta_\alpha}}E^\delta_\alpha=iE^\beta_\alpha\oplus i'\ell'^{-1}E^\delta_\beta E δ δ + d δ + δ α E α δ = i E α β ⊕ i ′ ℓ ′ − 1 E β δ
E δ δ + → d δ + δ α E α δ = i ′ E α γ ⊕ i ℓ − 1 E γ δ E^{\delta_+}_\delta\xrightarrow{d_{\delta_+\delta_\alpha}}E^\delta_\alpha=i'E^\gamma_\alpha\oplus i\ell^{-1}E^\delta_\gamma E δ δ + d δ + δ α E α δ = i ′ E α γ ⊕ i ℓ − 1 E γ δ
so that j d δ + δ α = d δ + δ β j\,d_{\delta_+\delta_\alpha}=d_{\delta_+\delta_\beta} j d δ + δ α = d δ + δ β and j ′ d δ + δ α = d δ + δ γ j'\,d_{\delta_+\delta_\alpha}=d_{\delta_+\delta_\gamma} j ′ d δ + δ α = d δ + δ γ . Note that i ′ ℓ ′ − 1 E β δ = i ′ E α γ i'\ell'^{-1}E^\delta_\beta=i'E^\gamma_\alpha i ′ ℓ ′ − 1 E β δ = i ′ E α γ and i ℓ − 1 E γ δ = i E α β i\ell^{-1}E^\delta_\gamma=iE^\beta_\alpha i ℓ − 1 E γ δ = i E α β , so that both splittings are the decomposition E α δ = i E α β ⊕ i ′ E α γ E^\delta_\alpha=iE^\beta_\alpha\oplus i'E^\gamma_\alpha E α δ = i E α β ⊕ i ′ E α γ .
Write d δ + δ α = ( d 1 d 2 ) d_{\delta_+\delta_\alpha}=\begin{pmatrix}d_1\\d_2\end{pmatrix} d δ + δ α = ( d 1 d 2 ) with respect to the splitting E α δ = i E α β ⊕ i ′ ℓ ′ − 1 E β δ E^\delta_\alpha=iE^\beta_\alpha\oplus i'\ell'^{-1}E^\delta_\beta E α δ = i E α β ⊕ i ′ ℓ ′ − 1 E β δ . Then
d δ + δ β x = j d δ + δ α x = j ( d 1 x + d 2 x ) = j d 2 x , d_{\delta_+\delta_\beta}x=j\,d_{\delta_+\delta_\alpha}x=j(d_1x+d_2x)=jd_2x, d δ + δ β x = j d δ + δ α x = j ( d 1 x + d 2 x ) = j d 2 x ,
since the splitting
0 ⟶ i E α β ⟶ E α δ ⟶ i ′ ℓ ′ − 1 E β δ ⟶ 0 0 \longrightarrow iE^\beta_\alpha\longrightarrow E^\delta_\alpha\longrightarrow i'\ell'^{-1}E^\delta_\beta\longrightarrow0 0 ⟶ i E α β ⟶ E α δ ⟶ i ′ ℓ ′ − 1 E β δ ⟶ 0
identifies the projection.
The map j : i ′ ℓ ′ − 1 E β δ → ≅ E β δ j:i'\ell'^{-1}E^\delta_\beta\xrightarrow{\cong} E^\delta_\beta j : i ′ ℓ ′ − 1 E β δ ≅ E β δ has inverse j − 1 = i ′ ℓ ′ − 1 j^{-1}=i'\ell'^{-1} j − 1 = i ′ ℓ ′ − 1 , hence
d 2 = j − 1 d δ + δ β = i ′ ℓ ′ − 1 d δ + δ β . d_2=j^{-1}d_{\delta+\delta\beta}=i'\ell'^{-1}d_{\delta+\delta\beta}. d 2 = j − 1 d δ + δ β = i ′ ℓ ′ − 1 d δ + δ β .
Symmetrically (using the other splitting and j ′ j' j ′ ), d 1 = i ℓ − 1 d δ + δ γ d_1=i\ell^{-1}d_{\delta+\delta\gamma} d 1 = i ℓ − 1 d δ + δ γ . Therefore
d δ + δ α = ( i ℓ − 1 d δ + δ γ i ′ ℓ ′ − 1 d δ + δ β ) . d_{\delta+\delta\alpha}=
\begin{pmatrix}
i\ell^{-1}d_{\delta+\delta\gamma}\\
i'\ell'^{-1}d_{\delta+\delta\beta}
\end{pmatrix}. d δ + δ α = ( i ℓ − 1 d δ + δ γ i ′ ℓ ′ − 1 d δ + δ β ) .
As for the second differential we argue as follows. Write d δ α α − = ( d 1 ′ d 2 ′ ) d_{\delta\alpha\alpha-}=(d'_1\ d'_2) d δ α α − = ( d 1 ′ d 2 ′ ) with respect to the splitting E α δ = i E α β ⊕ i ′ ℓ ′ − 1 E β δ E^\delta_\alpha=iE^\beta_\alpha\oplus i'\ell'^{-1}E^\delta_\beta E α δ = i E α β ⊕ i ′ ℓ ′ − 1 E β δ . As before we have the commutative diagrams:
\begin{CD}
iE^\beta_\alpha\oplus i'\ell'^{-1}E^\delta_\beta=E^\delta_\alpha@>{d_{\delta\alpha\alpha-}}>> E^\alpha_{\alpha-}\\
@A{i}AA @AA{\parallel}A\\
E^\beta_\alpha@>{d_{\beta\alpha\alpha-}}>> E^\alpha_{\alpha-}
\end{CD}
\qquad
\begin{CD}
i'E^\gamma_\alpha\oplus i\ell^{-1}E^\delta_\gamma=E^\delta_\alpha@>{d_{\delta\alpha\alpha-}}>> E^\alpha_{\alpha-}\\
@A{i'}AA @AA{\parallel}A\\
E^\gamma_\alpha@>{d_{\gamma\alpha\alpha-}}>> E^\alpha_{\alpha-}
\end{CD}
we get d δ α α − i = d β α α − d_{\delta\alpha\alpha-}i=d_{\beta\alpha\alpha-} d δ α α − i = d β α α − , i.e. d β α α − x = d δ α α − i ( x ) = d 1 ′ i ( x ) d_{\beta\alpha\alpha-}x=d_{\delta\alpha\alpha-}i(x)=d'_1i(x) d β α α − x = d δ α α − i ( x ) = d 1 ′ i ( x ) . Since i : E α β → ≅ i E α β i:E^\beta_\alpha\xrightarrow{\cong}iE^\beta_\alpha i : E α β ≅ i E α β ,
d β α α − i − 1 = d 1 ′ , d γ α α − i ′ − 1 = d 2 ′ , d_{\beta\alpha\alpha-}i^{-1}=d'_1,\qquad d_{\gamma\alpha\alpha-}i'^{-1}=d'_2, d β α α − i − 1 = d 1 ′ , d γ α α − i ′ − 1 = d 2 ′ ,
so
d δ α α − = ( d β α α − i − 1 d γ α α − i ′ − 1 ) . d_{\delta\alpha\alpha-}=\left(d_{\beta\alpha\alpha-}i^{-1}\quad d_{\gamma\alpha\alpha-}i'^{-1}\right). d δ α α − = ( d β α α − i − 1 d γ α α − i ′ − 1 ) .
The relation d δ α α − d δ + δ α = 0 d_{\delta\alpha\alpha-}d_{\delta+\delta\alpha}=0 d δ α α − d δ + δ α = 0 yields
( d β α α − i − 1 d γ α α − i ′ − 1 ) ( i ℓ − 1 d δ + δ γ i ′ ℓ ′ − 1 d δ + δ β ) = 0 , \left(d_{\beta\alpha\alpha-}i^{-1}\quad d_{\gamma\alpha\alpha-}i'^{-1}\right)
\begin{pmatrix}
i\ell^{-1}d_{\delta+\delta\gamma}\\
i'\ell'^{-1}d_{\delta+\delta\beta}
\end{pmatrix}
=0, ( d β α α − i − 1 d γ α α − i ′ − 1 ) ( i ℓ − 1 d δ + δ γ i ′ ℓ ′ − 1 d δ + δ β ) = 0 ,
and therefore
d β α α − ℓ − 1 d δ + δ γ + d γ α α − ℓ ′ − 1 d δ + δ β = 0 , d_{\beta\alpha\alpha-}\ell^{-1}d_{\delta+\delta\gamma}
+d_{\gamma\alpha\alpha-}\ell'^{-1}d_{\delta+\delta\beta}=0, d β α α − ℓ − 1 d δ + δ γ + d γ α α − ℓ ′ − 1 d δ + δ β = 0 ,
which proves the theorem. □
If we consider the fundamental spectral net then perspective pairs yield relations on the differentials. This is best illustrated with the example in § 1.2 . In the lattice O ( N 5 ) \mathrm{O}(N_5) O ( N 5 ) we have three perspective pairs. We start with two immediate pairs: (i) β , δ , ϵ \beta,\delta,\epsilon β , δ , ϵ and ζ \zeta ζ , with tail elements α \alpha α and 1 1 1 , and (ii) α , β , γ \alpha,\beta,\gamma α , β , γ and δ \delta δ , with tail elements 0 0 0 and ζ \zeta ζ . For (i) and (ii) we obtain the following relation on the differentials in Figure 6.2 :
(i):
0 ⋅ d 35 1 + d 24 1 d 45 1 = d 24 1 d 45 1 = 0 ; 0\cdot d^1_{35}+d^1_{24}d^1_{45}=d^1_{24}d^1_{45}=0; 0 ⋅ d 35 1 + d 24 1 d 45 1 = d 24 1 d 45 1 = 0 ;
(ii):
d 13 1 ⋅ 0 + d 12 1 d 24 1 = d 12 1 d 24 1 = 0. d^1_{13}\cdot0+d^1_{12}d^1_{24}=d^1_{12}d^1_{24}=0. d 13 1 ⋅ 0 + d 12 1 d 24 1 = d 12 1 d 24 1 = 0.
Here d 23 1 = 0 d^1_{23}=0 d 23 1 = 0 and d 34 1 = 0 d^1_{34}=0 d 34 1 = 0 , since the connecting homomorphisms k k k and k ′ k' k ′ of a perspective pair vanish: d 23 1 = d δ β α d^1_{23}=d_{\delta\beta\alpha} d 23 1 = d δ β α for (ii) and d 34 1 = d ζ δ β d^1_{34}=d_{\zeta\delta\beta} d 34 1 = d ζ δ β for (i). Figure 6.2 shows the paths of the differentials induced by the perspective pairs. For (i) in red and for (ii) in blue. Observe that in Figure 1.1 [left] the associated orders do not necessarily form a diamond. The diamond in a perspective pair is found in the lattice of down-sets.
A third diamond can be found in the down-set lattice O ( N 5 ) \mathrm{O}(N_5) O ( N 5 ) , v. Figure 1.1 [right], given by (iii) α , γ , ϵ \alpha,\gamma,\epsilon α , γ , ϵ and ζ \zeta ζ , with tail elements 0 0 0 and 1 1 1 . In Figure 6.3 the path is indicated in beige. This pair involves, besides differentials on page 1 of the fundamental spectral net, differentials from and to the convex set { 2 , 4 } \{2,4\} { 2 , 4 } . These are differentials on page 1 of the spectral net over the sublattice { 0 , α , γ , ϵ , ζ , 1 } \{0,\alpha,\gamma,\epsilon,\zeta,1\} { 0 , α , γ , ϵ , ζ , 1 } , v.
Defn. 5.1:
d 13 1 d 35 1 + d 1 , 24 1 d 24 , 5 1 = 0. \mathrm{d}^{1}_{13}\mathrm{d}^{1}_{35}+\mathrm{d}^{1}_{1,24}\mathrm{d}^{1}_{24,5}=0. d 13 1 d 35 1 + d 1 , 24 1 d 24 , 5 1 = 0.
The example shows that in the fundamental spectral net there are additional relations between differentials due to perspective pairs. In Example 8.3 we will apply these additional relations to an example discussed in [11 ] .
Remark 6.2 The perspective pairs in (i) and (ii) in the above example are elementary because they give an immediate relation between differentials on page-1 of the fundamental spectral net. In [11 ] such perspective pairs are related to adjacent elements in the order P. In particular the argument about additional connecting orbits between non-adjacent elements in [11, Fig. 5.1], cf. [17 ] , can be repeated using perspective pairs. We emphasize that the existence of a spectral matrix, or connection matrix, is not required. In [4 ] we will study perspective pairs in general for first and higher order differentials.
Derived differentials
If we consider spectral systems of vector spaces, i.e. R = K R=\mathbb{K} R = K is a field, then the short exact sequences in (16) are split exact and therefore the differentials carry all information. However, for ring coefficients the short exact sequences in (16) are not necessarily split exact. The extension problem can be solved via the information in the spectral system as discussed in § 4.3 .
The differentials in a spectral net do not suffice in general. We define derived differentials in this setting which carry the information about non-split extensions. In § 7 the latter is applied to the problem of connecting orbits in Morse representations.
Consider a subsystem of one exact triangle:
0 → E β γ → k E α β → 0 j ↑ i ↑ E α γ (26) \begin{CD}
0 @>>> E^{\gamma}_{\beta} @>k>> E^{\beta}_{\alpha} @>>> 0 \\
@. @A j AA @. @A i AA \\
@. E^{\gamma}_{\alpha} @. @. \tag*{(26)}
\end{CD} 0 E β γ j ⏐ ↑ E α γ k E α β 0 i ⏐ ↑ (26)
On the second (and final) page we obtain:
0 → 0 S α β γ γ → 0 S α α β γ → 0 0 0 \xrightarrow{0} S^{\gamma\gamma}_{\alpha\beta}
\xrightarrow{0} S^{\beta\gamma}_{\alpha\alpha}
\xrightarrow{0} 0 0 0 S α β γ γ 0 S α α β γ 0 0
The associated short exact sequence of S S S -terms is given by (16) :
0 ⟶ S α α β γ → i S α α γ γ → j S α β γ γ ⟶ 0. 0 \longrightarrow S^{\beta\gamma}_{\alpha\alpha}
\xrightarrow{i} S^{\gamma\gamma}_{\alpha\alpha}
\xrightarrow{j} S^{\gamma\gamma}_{\alpha\beta}
\longrightarrow0. 0 ⟶ S α α β γ i S α α γ γ j S α β γ γ ⟶ 0.
This yields a distinguished triangle in the derived category D h ( Mod R ) \mathop{\mathrm{Dh}}(\operatorname{Mod}_{R}) Dh ( Mod R ) :
S α α β γ → i E α γ → j S α β γ γ → κ S α α β γ [ 1 ] , S^{\beta\gamma}_{\alpha\alpha}
\xrightarrow{i} E^{\gamma}_{\alpha}
\xrightarrow{j} S^{\gamma\gamma}_{\alpha\beta}
\xrightarrow{\kappa} S^{\beta\gamma}_{\alpha\alpha}[1], S α α β γ i E α γ j S α β γ γ κ S α α β γ [ 1 ] ,
where κ : S α β γ γ → S α α β γ [ 1 ] \kappa:S^{\gamma\gamma}_{\alpha\beta}\to S^{\beta\gamma}_{\alpha\alpha}[1] κ : S α β γ γ → S α α β γ [ 1 ] is the derived differential of the triangle (26) . If β \ α = { p } \beta\backslash\alpha=\{p\} β \ α = { p } and γ \ β = { q } \gamma\backslash\beta=\{q\} γ \ β = { q } are singletons, then k k k is the first order differential d p q 1 d^{1}_{pq} d pq 1 , and κ \kappa κ is called a first order derived differential . The derived differential allows us to characterize when the exact triangle in (26) is split, i.e. k = 0 k=0 k = 0 and the short exact sequence 0 → E α β → E α γ → E β γ → 0 0\to E^{\beta}_{\alpha}\to E^{\gamma}_{\alpha}\to E^{\gamma}_{\beta}\to0 0 → E α β → E α γ → E β γ → 0 splits. In particular, then E α γ ≅ E α β ⊕ E β γ E^{\gamma}_{\alpha}\cong E^{\beta}_{\alpha}\oplus E^{\gamma}_{\beta} E α γ ≅ E α β ⊕ E β γ , but the converse does not hold in general, v. Remark 6.5. We consider the pair ( k , κ ) ∈ Hom ( E α β , E α β ) × Hom D h ( Mod R ) ( S α β γ γ , S α α β γ [ 1 ] ) (k,\kappa)\in\operatorname{Hom}(E^{\beta}_{\alpha},E^{\beta}_{\alpha})\times\operatorname{Hom}_{\mathop{\mathrm{Dh}}(\operatorname{Mod}_{R})}(S^{\gamma\gamma}_{\alpha\beta},S^{\beta\gamma}_{\alpha\alpha}[1]) ( k , κ ) ∈ Hom ( E α β , E α β ) × Hom Dh ( Mod R ) ( S α β γ γ , S α α β γ [ 1 ]) .
Theorem 6.3 Let ι ( a ) = ( a , 0 ) \iota(a)=(a,0) ι ( a ) = ( a , 0 ) and π ( a , c ) = c \pi(a,c)=c π ( a , c ) = c . If ( k , κ ) = 0 (k,\kappa)=0 ( k , κ ) = 0 , then there exists an isomorphism ψ : E α β ⊕ E β γ → E α γ \psi:E^{\beta}_{\alpha}\oplus E^{\gamma}_{\beta}\to E^{\gamma}_{\alpha} ψ : E α β ⊕ E β γ → E α γ with ψ ι = i \psi\iota=i ψ ι = i and j ψ = π j\psi=\pi j ψ = π . In particular, E α γ ≅ E α β ⊕ E β γ E^{\gamma}_{\alpha}\cong E^{\beta}_{\alpha}\oplus E^{\gamma}_{\beta} E α γ ≅ E α β ⊕ E β γ in the exact triangle (26) . Conversely, if ψ : E α β ⊕ E β γ → ≅ E α γ \psi:E^{\beta}_{\alpha}\oplus E^{\gamma}_{\beta}\xrightarrow{\cong}E^{\gamma}_{\alpha} ψ : E α β ⊕ E β γ ≅ E α γ satisfies ψ ι = i \psi\iota=i ψ ι = i and j ψ = π j\psi=\pi j ψ = π , then, ( k , κ ) = 0 (k,\kappa)=0 ( k , κ ) = 0 .
Proof. The condition k = 0 k=0 k = 0 implies that S α α β γ ≅ E α β S^{\beta\gamma}_{\alpha\alpha}\cong E^{\beta}_{\alpha} S α α β γ ≅ E α β and S α β γ γ ≅ E β γ S^{\gamma\gamma}_{\alpha\beta}\cong E^{\gamma}_{\beta} S α β γ γ ≅ E β γ , which yields the short exact sequence
0 ⟶ E α β → i E α γ → j E β γ ⟶ 0. 0\longrightarrow E^{\beta}_{\alpha}
\xrightarrow{i} E^{\gamma}_{\alpha}
\xrightarrow{j} E^{\gamma}_{\beta}
\longrightarrow0. 0 ⟶ E α β i E α γ j E β γ ⟶ 0.
Since the derived differential κ : E β γ → E α β [ 1 ] \kappa:E^{\gamma}_{\beta}\to E^{\beta}_{\alpha}[1] κ : E β γ → E α β [ 1 ] is trivial, this sequence splits, v. Sect. 4.3, i.e. there exists s : E β γ → E α γ s:E^{\gamma}_{\beta}\to E^{\gamma}_{\alpha} s : E β γ → E α γ with j s = id js=\operatorname{id} j s = id . Then ψ ( a , c ) = i ( a ) + s ( c ) \psi(a,c)=i(a)+s(c) ψ ( a , c ) = i ( a ) + s ( c ) is an isomorphism with ψ ι = i \psi\iota=i ψ ι = i and j ψ = π j\psi=\pi j ψ = π , and we conclude that E α γ ≅ E α β ⊕ E β γ E^{\gamma}_{\alpha}\cong E^{\beta}_{\alpha}\oplus E^{\gamma}_{\beta} E α γ ≅ E α β ⊕ E β γ .
For the converse we consider the exact sequences:
\begin{CD}
\cdots@>>> E^{\beta}_{\alpha} @>i>> E^{\gamma}_{\alpha} @>j>> E^{\gamma}_{\beta} @>k>> E^{\beta}_{\alpha} @>>> \cdots\\
@. @A\operatorname{id}AA @A\psi\cong AA @A\operatorname{id}AA @A\operatorname{id}AA \\
\cdots@>>> E^{\beta}_{\alpha} @>\iota>> E^{\beta}_{\alpha}\oplus E^{\gamma}_{\beta} @>\pi>> E^{\gamma}_{\beta} @>k>> E^{\beta}_{\alpha} @>>> \cdots
\end{CD}
From the bottom exact sequence we derive that ker k = E β γ \ker k=E^{\gamma}_{\beta} ker k = E β γ and im k = ker ι = 0 \operatorname{im}k=\ker\iota=0 im k = ker ι = 0 and thus k = 0 k=0 k = 0 . This implies that κ = 0 \kappa=0 κ = 0 since the short exact sequence 0 → E α β → E α β ⊕ E β γ → E β γ → 0 0\to E^{\beta}_{\alpha}\to E^{\beta}_{\alpha}\oplus E^{\gamma}_{\beta}\to E^{\gamma}_{\beta}\to0 0 → E α β → E α β ⊕ E β γ → E β γ → 0 corresponds to the trivial class in Ext R 1 ( E β γ , E α β ) \operatorname{Ext}^{1}_{R}(E^{\gamma}_{\beta},E^{\beta}_{\alpha}) Ext R 1 ( E β γ , E α β ) . □ \square □
The negation of the above result can be formulated as follows:
Corollary 6.4 The conditions k ≠ 0 k\ne0 k = 0 , or κ ≠ 0 \kappa\ne0 κ = 0 , imply that there is no isomorphism ψ : E α β ⊕ E β γ → E α γ \psi:E^{\beta}_{\alpha}\oplus E^{\gamma}_{\beta}\to E^{\gamma}_{\alpha} ψ : E α β ⊕ E β γ → E α γ with ψ ι = i \psi\iota=i ψ ι = i and j ψ = π j\psi=\pi j ψ = π .
Proof. The statement is the contrapositive of the second part of Theorem 6.3 .
Remark 6.5 If R R R is commutative and Noetherian and the E E E -terms are finitely generated, then κ ≠ 0 \kappa\ne0 κ = 0 or κ ≠ 0 \kappa\ne0 κ = 0 implies E γ δ ≇ E α β ⊕ E β γ E_{\gamma}^{\delta} \not\cong E_{\alpha}^{\beta} \oplus E_{\beta}^{\gamma} E γ δ ≅ E α β ⊕ E β γ . For κ = 0 \kappa= 0 κ = 0 this follows from [18 Thm. 1] . Over such a ring, a short exact sequence of finitely generated modules whose middle term is isomorphic to the direct sum of its outer terms splits. For κ ≠ 0 \kappa\ne0 κ = 0 it follows by comparing lengths after localizing at a minimal prime of the support of im k \operatorname{im} k im k . Without these assumptions the statement fails. For R = Z R = \mathbb{Z} R = Z , the sequence 0 → Z → Z ⊕ ( Z / 2 ) ( N ) → Z / 2 ⊕ ( Z / 2 ) ( N ) → 0 0 \to\mathbb{Z} \to\mathbb{Z} \oplus(\mathbb{Z}/2)^{(N)} \to\mathbb{Z}/2 \oplus(\mathbb{Z}/2)^{(N)} \to0 0 → Z → Z ⊕ ( Z /2 ) ( N ) → Z /2 ⊕ ( Z /2 ) ( N ) → 0 , with n ↦ ( 2 n , 0 ) n \mapsto(2n,0) n ↦ ( 2 n , 0 ) and ( n , x ) ↦ ( n ‾ , x ) (n,x) \mapsto(\overline{n},x) ( n , x ) ↦ ( n , x ) , does not split and occurs in (26) with κ = 0 \kappa= 0 κ = 0 and κ ≠ 0 \kappa\ne0 κ = 0 , but its middle term is isomorphic to the direct sum of its outer terms.
Remark 6.6 Let R R R be hereditary, for example a principal ideal domain, and let all E E E -terms E α β E_{\alpha}^{\beta} E α β with β \ α \beta\backslash\alpha β \ α a singleton be projective. Then κ = 0 \kappa= 0 κ = 0 for every triangle (26) with β \ α \beta\backslash\alpha β \ α and γ \ β \gamma\backslash\beta γ \ β singletons. Indeed, S α β γ δ = ker κ S_{\alpha\beta}^{\gamma\delta} = \ker\kappa S α β γ δ = ker κ is then a submodule of the projective module E β β E_{\beta}^{\beta} E β β , hence projective, v. Rmk. 4.9. For a chain-generated E E E -system whose differential module is a chain complex C C C vanishing in negative degrees, with subquotients C α β = F β C / F α C C_{\alpha}^{\beta} = F_{\beta}C/F_{\alpha}C C α β = F β C / F α C consisting of projective modules, such as E ( T ( U ) ) E(\mathrm{T}(\mathrm{U})) E ( T ( U )) in Sect. 3.2, the same holds over an arbitrary ring R R R .
Higher differentials
The term higher differential is used to indicate differentials between S S S -terms, other than those between E E E -terms.
Definition 6.7 Let P be a poset. Two elements p , q ∈ P p,q \in\mathrm{P} p , q ∈ P are connected inside a convex set I ∈ C o ( P ) I \in\mathrm{Co}(\mathrm{P}) I ∈ Co ( P ) if they belong to the same connected component of the Hasse diagram of I I I , regarded as an undirected graph. Equivalently, p p p and q q q are connected via a path { q k } ⊆ I \{q_k\} \subseteq I { q k } ⊆ I of covering relations
q = q 0 ≻ q 1 ≻ ⋯ ≻ q k ≻ ⋯ ≻ q m − 1 ≻ q m = p , m ≥ 1 , (27) q = q_0 \succ q_1 \succ\cdots\succ q_k \succ\cdots\succ q_{m-1} \succ q_m = p,\qquad m \ge1, \tag*{(27)} q = q 0 ≻ q 1 ≻ ⋯ ≻ q k ≻ ⋯ ≻ q m − 1 ≻ q m = p , m ≥ 1 , (27)
where each step q k ≻ q k + 1 q_k \succ q_{k+1} q k ≻ q k + 1 indicates either q k < q k + 1 q_k < q_{k+1} q k < q k + 1 , or q k > q k + 1 q_k > q_{k+1} q k > q k + 1 .
For example in Figure 1.1 the elements 2 and 3 are connected in P = N 5 \mathrm{P} = \mathrm{N}_5 P = N 5 via the covering path 2 ≻ 1 ≺ 3 2 \succ1 \prec3 2 ≻ 1 ≺ 3 , and 1 and 5 are connected in N 5 \mathrm{N}_5 N 5 via the covering path 1 ≺ 3 ≺ 5 1 \prec3 \prec5 1 ≺ 3 ≺ 5 . The following theorem relates disconnectedness of elements to a property of differentials.
Theorem 6.8 Let S S S be an excisive spectral system over O ( P ) \mathrm{O}(\mathrm{P}) O ( P ) . Let p , q ∈ P p,q \in\mathrm{P} p , q ∈ P . Then, the differentials
d : S γ δ ϵ ζ → S α β γ δ , ∀ ( α , β , γ , δ , ϵ , ζ ) ∈ O ( P ) 6 , \mathrm{d}: S_{\gamma\delta}^{\epsilon\zeta} \to S_{\alpha\beta}^{\gamma\delta},\qquad\forall(\alpha,\beta,\gamma,\delta,\epsilon,\zeta) \in\mathrm{O}(\mathrm{P})_6, d : S γ δ ϵ ζ → S α β γ δ , ∀ ( α , β , γ , δ , ϵ , ζ ) ∈ O ( P ) 6 , with ϵ \ δ = { p } \epsilon\backslash\delta= \{p\} ϵ \ δ = { p } and γ \ β = { q } \gamma\backslash\beta= \{q\} γ \ β = { q } , and such that p p p and q q q are not connected inside ϵ \ β \epsilon\backslash\beta ϵ \ β , are trivial, i.e. d = 0 \mathrm{d} = 0 d = 0 .
Proof. Using the commutative squares for S S S -terms as given by Definition 4.1(iv) we obtain:
S γ δ ϵ ϵ → d † S β β γ δ i ↓ ↓ i d S γ δ ϵ ζ → d ‡ S β β γ δ i d ↑ ↑ j S γ δ ϵ ζ → d S α β γ δ ⟹ S γ δ ϵ ϵ → d † S β β γ δ i ↓ ↑ j S γ δ ϵ ζ → d S α β γ δ (28) \begin{CD}
S_{\gamma\delta}^{\epsilon\epsilon} @>{\mathrm{d}^{\dagger}}>> S_{\beta\beta}^{\gamma\delta} \\
@V{\mathrm{i}}VV @VV{\mathrm{id}}V \\
S_{\gamma\delta}^{\epsilon\zeta} @>{\mathrm{d}^{\ddagger}}>> S_{\beta\beta}^{\gamma\delta} \\
@A{\mathrm{id}}AA @AA{\mathrm{j}}A \\
S_{\gamma\delta}^{\epsilon\zeta} @>{\mathrm{d}}>> S_{\alpha\beta}^{\gamma\delta}
\end{CD}
\quad\Longrightarrow\quad
\begin{CD}
S_{\gamma\delta}^{\epsilon\epsilon} @>{\mathrm{d}^{\dagger}}>> S_{\beta\beta}^{\gamma\delta} \\
@V{\mathrm{i}}VV @AA{\mathrm{j}}A \\
S_{\gamma\delta}^{\epsilon\zeta} @>{\mathrm{d}}>> S_{\alpha\beta}^{\gamma\delta} \tag*{(28)}
\end{CD} S γ δ ϵϵ i ↓ ⏐ S γ δ ϵ ζ id ⏐ ↑ S γ δ ϵ ζ d † d ‡ d S β β γ δ ↓ ⏐ id S β β γ δ ⏐ ↑ j S α β γ δ ⟹ S γ δ ϵϵ i ↓ ⏐ S γ δ ϵ ζ d † d S β β γ δ ⏐ ↑ j S α β γ δ (28)
Therefore, d † = j d i d^\dagger= jdi d † = j d i and, since i i i is surjective and j j j is injective, v. Defn. 4.1 (iii), d = 0 d = 0 d = 0 if and only if d † = 0 d^\dagger= 0 d † = 0 . It is enough to prove the theorem for d † d^\dagger d † .
Define I I I as the path connected component of q ∈ ϵ \ β q \in\epsilon\backslash\beta q ∈ ϵ \ β , that is, the set of all points q ′ q' q ′ that are connected to q q q inside ϵ \ β \epsilon\backslash\beta ϵ \ β . Now, define J = ( ϵ \ β ) \ I J = (\epsilon\backslash\beta)\backslash I J = ( ϵ \ β ) \ I . Define ω − = ↓ J \omega^- = \mathord{\downarrow} J ω − = ↓ J , and ω + = ↓ I \omega^+ = \mathord{\downarrow} I ω + = ↓ I . This gives the chain:
ω − ∩ ω + ⊆ β ⊆ γ ⊆ δ ⊆ ϵ , ϵ \ β ⊆ ω − ∪ ω + . \omega^- \cap\omega^+ \subseteq\beta\subseteq\gamma\subseteq\delta\subseteq\epsilon,\qquad\epsilon\backslash\beta\subseteq\omega^- \cup\omega^+. ω − ∩ ω + ⊆ β ⊆ γ ⊆ δ ⊆ ϵ , ϵ \ β ⊆ ω − ∪ ω + .
Here ω − ∩ ω + ⊆ β \omega^- \cap\omega^+ \subseteq\beta ω − ∩ ω + ⊆ β holds because I I I and J J J are unions of connected components of the convex set ϵ \ β \epsilon\backslash\beta ϵ \ β : an element of ω − ∩ ω + \omega^- \cap\omega^+ ω − ∩ ω + which is not in β \beta β lies in ϵ \ β \epsilon\backslash\beta ϵ \ β and below elements of I I I and of J J J , and would therefore connect I I I and J J J . Moreover, ϵ \ β = I ∪ J ⊆ ω − ∪ ω + \epsilon\backslash\beta= I\cup J \subseteq\omega^-\cup\omega^+ ϵ \ β = I ∪ J ⊆ ω − ∪ ω + by construction. In particular, p ∈ J p\in J p ∈ J , q ∈ I q\in I q ∈ I , and no element of I I I is comparable to an element of J J J .
Define β ± = β ∩ ω ± \beta^\pm= \beta\cap\omega^\pm β ± = β ∩ ω ± , γ ± = γ ∩ ω ± \gamma^\pm= \gamma\cap\omega^\pm γ ± = γ ∩ ω ± , δ ± = δ ∩ ω ± \delta^\pm= \delta\cap\omega^\pm δ ± = δ ∩ ω ± , and ϵ ± = ϵ ∩ ω ± \epsilon^\pm= \epsilon\cap\omega^\pm ϵ ± = ϵ ∩ ω ± . From the excisive property for E E E we conclude that
S γ δ ϵ ϵ ≅ S γ − δ − ϵ − ϵ − ⊕ S γ + δ + ϵ + ϵ + , S β β γ δ ≅ S β − β − γ − δ − ⊕ S β + β + γ + δ + , S_{\gamma\delta}^{\epsilon\epsilon}\cong S_{\gamma^-\delta^-}^{\epsilon^-\epsilon^-}\oplus S_{\gamma^+\delta^+}^{\epsilon^+\epsilon^+},\qquad
S_{\beta\beta}^{\gamma\delta}\cong S_{\beta^-\beta^-}^{\gamma^-\delta^-}\oplus S_{\beta^+\beta^+}^{\gamma^+\delta^+}, S γ δ ϵϵ ≅ S γ − δ − ϵ − ϵ − ⊕ S γ + δ + ϵ + ϵ + , S β β γ δ ≅ S β − β − γ − δ − ⊕ S β + β + γ + δ + ,
through the natural isomorphisms. Indeed, let χ ′ ⊆ χ \chi'\subseteq\chi χ ′ ⊆ χ be elements of the chain β ⊆ γ ⊆ δ ⊆ ϵ \beta\subseteq\gamma\subseteq\delta\subseteq\epsilon β ⊆ γ ⊆ δ ⊆ ϵ . Then χ \ χ ′ ⊆ I ∪ J \chi\backslash\chi'\subseteq I\cup J χ \ χ ′ ⊆ I ∪ J , and the down-sets χ ′ ∪ ( χ ∩ I ) \chi'\cup(\chi\cap I) χ ′ ∪ ( χ ∩ I ) and χ ′ ∪ ( χ ∩ J ) \chi'\cup(\chi\cap J) χ ′ ∪ ( χ ∩ J ) form a perspective pair with χ ′ \chi' χ ′ and χ \chi χ , so that E χ ′ χ ≅ E χ ′ χ ′ ∪ ( χ ∩ I ) ⊕ E χ ′ χ ′ ∪ ( χ ∩ J ) E_{\chi'}^\chi\cong E_{\chi'}^{\chi'\cup(\chi\cap I)}\oplus E_{\chi'}^{\chi'\cup(\chi\cap J)} E χ ′ χ ≅ E χ ′ χ ′ ∪ ( χ ∩ I ) ⊕ E χ ′ χ ′ ∪ ( χ ∩ J ) by Sect. 6.1. By excision, E χ ′ χ ′ ∪ ( χ ∩ I ) ≅ E χ ′ + χ + E_{\chi'}^{\chi'\cup(\chi\cap I)}\cong E_{\chi'^+}^{\chi^+} E χ ′ χ ′ ∪ ( χ ∩ I ) ≅ E χ ′ + χ + and E χ ′ χ ′ ∪ ( χ ∩ J ) ≅ E χ ′ − χ − E_{\chi'}^{\chi'\cup(\chi\cap J)}\cong E_{\chi'^-}^{\chi^-} E χ ′ χ ′ ∪ ( χ ∩ J ) ≅ E χ ′ − χ − . These isomorphisms are compatible with the homomorphisms ℓ \ell ℓ , which yields the isomorphisms for the S S S -terms. The following squares are commutative
S γ − δ − ϵ − ϵ − → ℓ ϵ − S γ δ ϵ ϵ d − ↓ ↓ d † S β − β − γ − δ − → ℓ β − S β β γ δ S γ + δ + ϵ + ϵ + → ℓ ϵ + S γ δ ϵ ϵ d + ↓ ↓ d † S β + β + γ + δ + → ℓ β + S β β γ δ \begin{CD}
S_{\gamma^-\delta^-}^{\epsilon^-\epsilon^-} @>{\ell_\epsilon^-}>> S_{\gamma\delta}^{\epsilon\epsilon}\\
@V{d^-}VV @VV{d^\dagger}V\\
S_{\beta^-\beta^-}^{\gamma^-\delta^-} @>{\ell_\beta^-}>> S_{\beta\beta}^{\gamma\delta}
\end{CD}
\qquad
\begin{CD}
S_{\gamma^+\delta^+}^{\epsilon^+\epsilon^+} @>{\ell_\epsilon^+}>> S_{\gamma\delta}^{\epsilon\epsilon}\\
@V{d^+}VV @VV{d^\dagger}V\\
S_{\beta^+\beta^+}^{\gamma^+\delta^+} @>{\ell_\beta^+}>> S_{\beta\beta}^{\gamma\delta}
\end{CD} S γ − δ − ϵ − ϵ − d − ↓ ⏐ S β − β − γ − δ − ℓ ϵ − ℓ β − S γ δ ϵϵ ↓ ⏐ d † S β β γ δ S γ + δ + ϵ + ϵ + d + ↓ ⏐ S β + β + γ + δ + ℓ ϵ + ℓ β + S γ δ ϵϵ ↓ ⏐ d † S β β γ δ
Under direct sum we obtain:
S γ − δ − ϵ − ϵ − ⊕ S γ + δ + ϵ + ϵ + → ℓ ϵ − + ℓ ϵ + S γ δ ϵ ϵ d − ⊕ d + ↓ ↓ d † S β − β − γ − δ − ⊕ S β + β + γ + δ + → ℓ β − + ℓ β + S β β γ δ \begin{CD}
S_{\gamma^-\delta^-}^{\epsilon^-\epsilon^-}\oplus S_{\gamma^+\delta^+}^{\epsilon^+\epsilon^+} @>{\ell_\epsilon^-+\ell_\epsilon^+}>> S_{\gamma\delta}^{\epsilon\epsilon}\\
@V{d^-\oplus d^+}VV @VV{d^\dagger}V\\
S_{\beta^-\beta^-}^{\gamma^-\delta^-}\oplus S_{\beta^+\beta^+}^{\gamma^+\delta^+} @>{\ell_\beta^-+\ell_\beta^+}>> S_{\beta\beta}^{\gamma\delta}
\end{CD} S γ − δ − ϵ − ϵ − ⊕ S γ + δ + ϵ + ϵ + d − ⊕ d + ↓ ⏐ S β − β − γ − δ − ⊕ S β + β + γ + δ + ℓ ϵ − + ℓ ϵ + ℓ β − + ℓ β + S γ δ ϵϵ ↓ ⏐ d † S β β γ δ
Due to linearity and commutativity we have that
d † ( ℓ ϵ − + ℓ ϵ + ) ( a , b ) = d † ℓ ϵ − ( a ) + d † ℓ ϵ + ( b ) = ℓ β − d − ( a ) + ℓ β + d + ( b ) = ( ℓ β − + ℓ β + ) ( d − ⊕ d + ) ( a , b ) \begin{aligned}
d^\dagger(\ell_\epsilon^-+\ell_\epsilon^+)(a,b)=d^\dagger\ell_\epsilon^-(a)+d^\dagger\ell_\epsilon^+(b)=\ell_\beta^-d^-(a)+\ell_\beta^+d^+(b) \\
&=(\ell_\beta^-+\ell_\beta^+)(d^-\oplus d^+)(a,b)
\end{aligned} d † ( ℓ ϵ − + ℓ ϵ + ) ( a , b ) = d † ℓ ϵ − ( a ) + d † ℓ ϵ + ( b ) = ℓ β − d − ( a ) + ℓ β + d + ( b ) = ( ℓ β − + ℓ β + ) ( d − ⊕ d + ) ( a , b )
and since both ℓ ϵ − + ℓ ϵ + \ell_\epsilon^-+\ell_\epsilon^+ ℓ ϵ − + ℓ ϵ + and ℓ β − + ℓ β + \ell_\beta^-+\ell_\beta^+ ℓ β − + ℓ β + are the natural isomorphisms, the differential d † d^\dagger d † transforms to d † ~ = ( ℓ β − + ℓ β + ) − 1 d † ( ℓ ϵ − + ℓ ϵ + ) \widetilde{d^\dagger}=(\ell_\beta^-+\ell_\beta^+)^{-1}d^\dagger(\ell_\epsilon^-+\ell_\epsilon^+) d † = ( ℓ β − + ℓ β + ) − 1 d † ( ℓ ϵ − + ℓ ϵ + ) , given by:
d † ~ = [ d − 0 0 d + ] : S γ − δ − ϵ − ϵ − ⊕ S γ + δ + ϵ + ϵ + ⟶ S β − β − γ − δ − ⊕ S β + β + γ + δ + . \widetilde{d^\dagger}=
\begin{bmatrix}
d^- & 0\\
0 & d^+
\end{bmatrix}
:S_{\gamma^-\delta^-}^{\epsilon^-\epsilon^-}\oplus S_{\gamma^+\delta^+}^{\epsilon^+\epsilon^+}
\longrightarrow
S_{\beta^-\beta^-}^{\gamma^-\delta^-}\oplus S_{\beta^+\beta^+}^{\gamma^+\delta^+}. d † = [ d − 0 0 d + ] : S γ − δ − ϵ − ϵ − ⊕ S γ + δ + ϵ + ϵ + ⟶ S β − β − γ − δ − ⊕ S β + β + γ + δ + .
v. [16 Lem. 4.18] . Since ϵ = δ ∪ { p } \epsilon=\delta\cup\{p\} ϵ = δ ∪ { p } and p ∉ ω + p\notin\omega^+ p ∈ / ω + , it follows that ϵ + = δ + \epsilon^+=\delta^+ ϵ + = δ + . Similarly, since γ = β ∪ { q } \gamma=\beta\cup\{q\} γ = β ∪ { q } , with q ∉ ω − q\notin\omega^- q ∈ / ω − , it follows that γ − = β − \gamma^-=\beta^- γ − = β − . Consequently,
S γ + δ + ϵ + ϵ + = 0 , and S β − β − γ − δ − = 0. S_{\gamma^+\delta^+}^{\epsilon^+\epsilon^+}=0,\qquad\text{and}\qquad S_{\beta^-\beta^-}^{\gamma^-\delta^-}=0. S γ + δ + ϵ + ϵ + = 0 , and S β − β − γ − δ − = 0.
and thus d † ~ = 0 \widetilde{d^\dagger}=0 d † = 0 , i.e. d † = 0 d^\dagger=0 d † = 0 , which proves the theorem.
Corollary 6.10. For any pair of parallel elements p ∥ q p\|q p ∥ q , the differential
d : E γ ϵ → E γ δ \mathsf{d}:E_{\gamma}^{\epsilon}\to E_{\gamma}^{\delta} d : E γ ϵ → E γ δ with ϵ \ δ = { p } \epsilon\backslash\delta=\{p\} ϵ \ δ = { p } , and δ \ γ = { q } \delta\backslash\gamma=\{q\} δ \ γ = { q } , is trivial.
Proof. Recall that E δ δ ϵ = E δ δ ϵ ϵ E_{\delta\delta}^{\epsilon}=E_{\delta\delta}^{\epsilon\epsilon} E δ δ ϵ = E δ δ ϵϵ and E γ δ = E γ γ δ δ E_{\gamma}^{\delta}=E_{\gamma\gamma}^{\delta\delta} E γ δ = E γ γ δ δ . By assumption, δ = γ ∪ { q } \delta=\gamma\cup\{q\} δ = γ ∪ { q } , and ϵ = δ ∪ { p } \epsilon=\delta\cup\{p\} ϵ = δ ∪ { p } . By construction ϵ \ γ = { p , q } \epsilon\backslash\gamma=\{p,q\} ϵ \ γ = { p , q } , and since p ∥ q p\|q p ∥ q , then p p p and q q q are disconnected in ϵ \ γ \epsilon\backslash\gamma ϵ \ γ , and thus d = 0 \mathsf{d}=0 d = 0 by Theorem 6.8 .
Corollary 6.12. Let S S S be an excisive spectral system over O ( P ) \mathsf{O}(\mathsf{P}) O ( P ) . Suppose that a differential
d : S γ δ ϵ ζ → S α β γ δ , for some ( α , β , γ , δ , ϵ , ζ ) ∈ O ( P ) 6 , \mathsf{d}:S_{\gamma\delta}^{\epsilon\zeta}\to S_{\alpha\beta}^{\gamma\delta},\qquad\text{for some }(\alpha,\beta,\gamma,\delta,\epsilon,\zeta)\in\mathsf{O}(\mathsf{P})_6, d : S γ δ ϵ ζ → S α β γ δ , for some ( α , β , γ , δ , ϵ , ζ ) ∈ O ( P ) 6 , with ϵ \ δ = { p } \epsilon\backslash\delta=\{p\} ϵ \ δ = { p } , p ≠ q p\ne q p = q and γ \ β = { q } \gamma\backslash\beta=\{q\} γ \ β = { q } , is non-trivial, i.e. d ≠ 0 \mathsf{d}\ne0 d = 0 . Then, there is a path of covering relations in ϵ \ β \epsilon\backslash\beta ϵ \ β such that q ≺ q 1 ≺ ⋯ ≺ q k ≺ ⋯ ≺ q m − 1 ≺ p q\prec q_1\prec\cdots\prec q_k\prec\cdots\prec q_{m-1}\prec p q ≺ q 1 ≺ ⋯ ≺ q k ≺ ⋯ ≺ q m − 1 ≺ p , with m ≥ 1 m\ge1 m ≥ 1 .
Proof. The hypothesis that d ≠ 0 \mathsf{d}\ne0 d = 0 implies that p , q p,q p , q are connected, which is equivalent to the existence of a path q ≺ q 1 ≺ ⋯ ≺ q m − 1 ≺ p q\prec q_1\prec\cdots\prec q_{m-1}\prec p q ≺ q 1 ≺ ⋯ ≺ q m − 1 ≺ p inside ϵ \ β \epsilon\backslash\beta ϵ \ β . Note that γ = β ∪ { q } \gamma=\beta\cup\{q\} γ = β ∪ { q } and ϵ = δ ∪ { p } \epsilon=\delta\cup\{p\} ϵ = δ ∪ { p } . Suppose there exists p ′ ∈ ϵ p'\in\epsilon p ′ ∈ ϵ such that p ≺ p ′ p\prec p' p ≺ p ′ . Then, p ′ ∈ δ p'\in\delta p ′ ∈ δ and thus p ∈ δ p\in\delta p ∈ δ , a contradiction. Consequently, p p p is maximal in ϵ \epsilon ϵ and thus maximal in ϵ \ β \epsilon\backslash\beta ϵ \ β . Suppose there exists q ′ ∈ ϵ \ β q'\in\epsilon\backslash\beta q ′ ∈ ϵ \ β such that q ′ ≺ q q'\prec q q ′ ≺ q . Then, q ′ ∈ γ q'\in\gamma q ′ ∈ γ and thus q ′ ∈ β q'\in\beta q ′ ∈ β , a contradiction. Consequently, q q q is minimal in ϵ \ β \epsilon\backslash\beta ϵ \ β . Hence the first step of the path is q ≺ q 1 q\prec q_1 q ≺ q 1 , and the last step is q m − 1 ≺ p q_{m-1}\prec p q m − 1 ≺ p . Hence, q ≺ q 1 ≺ ⋯ ≺ q m − 1 ≺ p q\prec q_1\prec\cdots\prec q_{m-1}\prec p q ≺ q 1 ≺ ⋯ ≺ q m − 1 ≺ p . If m = 1 m=1 m = 1 , this implies q ≺ p q\prec p q ≺ p . □ \square □
A path in Corollary 6.12 is called a directed path from p p p to q q q , and p p p is said to be directly path connected to q q q inside ϵ \ β \epsilon\backslash\beta ϵ \ β (not a symmetric relation!).
Example 6.13. Consider P and O(P) as given in Figure 6.4 . The elements 1 and 3 are connected or not, depending on the choices of ϵ \epsilon ϵ and β \beta β . First we take 1 \ 0 = P 1\backslash0=P 1\0 = P , in which case 1 and 3 are connected, where 1 is minimal and 3 is maximal in P. The directed path is given by 1 ≺ 2 ′ ≺ 2 ≺ 3 1\prec2'\prec2\prec3 1 ≺ 2 ′ ≺ 2 ≺ 3 . If we take 1 \ β 1\backslash\beta 1\ β , ζ \ 0 \zeta\backslash0 ζ \0 or ζ \ β \zeta\backslash\beta ζ \ β , then 1 and 3 cannot be connected.
Figure 6.4. The N N N -poset P [left] and its lattice of down-sets O(P) [right].
Figure 6.5. In the notation of Theorem 6.8 , the gray areas represent ϵ \epsilon ϵ and the red areas represent β \beta β . The difference ϵ ∖ β \epsilon\setminus\beta ϵ ∖ β is also shaded gray. From left to right we consider the cases 1 ∖ 0 1\setminus0 1 ∖ 0 , ζ ∖ 0 \zeta\setminus0 ζ ∖ 0 , 1 ∖ β 1\setminus\beta 1 ∖ β and ζ ∖ β \zeta\setminus\beta ζ ∖ β used in Example 6.13 . The red nodes and red edges do not belong to ϵ ∖ β \epsilon\setminus\beta ϵ ∖ β . Black nodes and black edges do belong to ϵ ∖ β \epsilon\setminus\beta ϵ ∖ β . Only in the leftmost figure can we connect nodes 1 and 3 through black lines.
Remark 6.14 All results described above are also valid for convex sets ϵ ∖ δ \epsilon\setminus\delta ϵ ∖ δ and γ ∖ β \gamma\setminus\beta γ ∖ β and not just single elements in P, where “p and q are not connected” is replaced by “no element of ϵ ∖ δ \epsilon\setminus\delta ϵ ∖ δ is connected to an element of γ ∖ β \gamma\setminus\beta γ ∖ β inside ϵ ∖ β \epsilon\setminus\beta ϵ ∖ β ”.
Connecting orbits for Morse representations
For attractor-repeller pairs the connecting homomorphism in the long exact sequence of Conley indices contains information about connecting orbits, i.e. if the connecting homomorphism is non-trivial, then the set of connecting orbits between attractor and repeller is non-void, v. [5 , 24 ] and Lemma 7.1 below. This idea is translated to connection matrices by Franzosa for flow-orders, v. [11 Prop. 5.3] . In Franzosa’s approach non-zero entries of a connection matrix associated with adjacent Morse sets of the flow-order imply connections between the adjacent Morse sets. In [2 ] Bartłomiejczyk uses attractor-repeller pairs to relate first order differentials in spectral sequences to connecting orbits between adjacent Morse sets in any linear Morse representation. We show that the differentials in a spectral system also carry information about connecting orbits between Morse sets. Throughout this section, ( M , ≤ ) (M,\leq) ( M , ≤ ) is a Morse representation, and the S S S -terms and differentials are those of the spectral system S = S ( E ( M ) ) \mathbf{S}=\mathbf{S}(E(M)) S = S ( E ( M )) , v. Sect. 3.2 and Thm. 4.7 . Since E ( M ) E(M) E ( M ) is excisive, so is S \mathbf{S} S .
Lemma 7.1 Let α ⊆ β ⊆ γ \alpha\subseteq\beta\subseteq\gamma α ⊆ β ⊆ γ in O ( M ) \mathcal{O}(M) O ( M ) , and let ( A , A ∗ ) (A,A^*) ( A , A ∗ ) be the attractor-repeller pair in Inv ( G γ ∖ α ∪ ) \operatorname{Inv}(G_{\gamma\setminus\alpha}\cup) Inv ( G γ ∖ α ∪ ) given by A = Inv ( G β ∖ α ∪ ) A=\operatorname{Inv}(G_{\beta\setminus\alpha}\cup) A = Inv ( G β ∖ α ∪ ) and A ∗ = Inv ( G γ ∖ β ∪ ) A^*=\operatorname{Inv}(G_{\gamma\setminus\beta}\cup) A ∗ = Inv ( G γ ∖ β ∪ ) . If C ( A ∗ , A ) = ∅ C(A^*,A)=\varnothing C ( A ∗ , A ) = ∅ , then there exists an isomorphism ψ : E α β ⊕ E β γ → E α γ \psi:E^\beta_\alpha\oplus E^\gamma_\beta\to E^\gamma_\alpha ψ : E α β ⊕ E β γ → E α γ with ψ ι = i \psi\iota=i ψ ι = i and j ψ = π j\psi=\pi j ψ = π , so that ( k , κ ) = 0 (k,\kappa)=0 ( k , κ ) = 0 , v. Thm. 6.3 . In particular, if the connecting homomorphism k : E β γ → E α β k:E^\gamma_\beta\to E^\beta_\alpha k : E β γ → E α β is non-trivial, then C ( A ∗ , A ) ≠ ∅ C(A^*,A)\ne\varnothing C ( A ∗ , A ) = ∅ .
Proof. The chain 0 ⊆ α ⊆ β ⊆ γ ⊆ 1 0\subseteq\alpha\subseteq\beta\subseteq\gamma\subseteq1 0 ⊆ α ⊆ β ⊆ γ ⊆ 1 is a sublattice of O ( M ) \mathcal{O}(M) O ( M ) . It defines a coarser Morse representation ( M † , ≤ † ) (M^\dagger,\leq^\dagger) ( M † , ≤ † ) , v. Sect. 2, whose Morse sets are the non-empty sets among Inv ( F α ∪ ) \operatorname{Inv}(F_\alpha\cup) Inv ( F α ∪ ) , A A A , A ∗ A^* A ∗ and Inv ( G 1 ∖ γ ∪ ) \operatorname{Inv}(G_{1\setminus\gamma}\cup) Inv ( G 1 ∖ γ ∪ ) , linearly ordered in this order. Its lattice of down-sets is identified with the chain, so that β ∖ α = { A } \beta\setminus\alpha=\{A\} β ∖ α = { A } and γ ∖ β = { A ∗ } \gamma\setminus\beta=\{A^*\} γ ∖ β = { A ∗ } , and by Remark 3.2 and Birkhoff duality, E ( M † , ≤ † ) E(M^\dagger,\leq^\dagger) E ( M † , ≤ † ) is the restriction of E ( M ) E(M) E ( M ) to the chain. Let ≤ ‡ \leq^\ddagger ≤ ‡ be the order obtained from ≤ † \leq^\dagger ≤ † by removing the relation A < † A ∗ A<^\dagger A^* A < † A ∗ , so that A ∥ A ∗ A\mathbin{\|}A^* A ∥ A ∗ . The condition in Theorem 2.1 for ≤ ‡ \leq^\ddagger ≤ ‡ differs from the one for ≤ † \leq^\dagger ≤ † only for orbits from A ∗ A^* A ∗ to A A A , and C ( A ∗ , A ) = ∅ C(A^*,A)=\varnothing C ( A ∗ , A ) = ∅ . Hence ( M † , ≤ ‡ ) (M^\dagger,\leq^\ddagger) ( M † , ≤ ‡ ) is again a Morse representation for ϕ \phi ϕ , and by Remark 3.2 and Birkhoff duality we obtain an injection E ( M † , ≤ ‡ ) ↪ E ( M † , ≤ † ) E(M^\dagger,\leq^\ddagger)\hookrightarrow E(M^\dagger,\leq^\dagger) E ( M † , ≤ ‡ ) ↪ E ( M † , ≤ † ) . Besides the chain, O ( M † , ≤ ‡ ) \mathcal{O}(M^\dagger,\leq^\ddagger) O ( M † , ≤ ‡ ) contains the down-set β ′ \beta'\! β ′ with β ′ ∖ α = { A ∗ } \beta'\setminus\alpha=\{A^*\} β ′ ∖ α = { A ∗ } and γ ∖ β ′ = { A } \gamma\setminus\beta'=\{A\} γ ∖ β ′ = { A } .
Since β ′ ∖ α = γ ∖ β = { A ∗ } \beta'\setminus\alpha=\gamma\setminus\beta=\{A^*\} β ′ ∖ α = γ ∖ β = { A ∗ } , the triples ( α , β , γ ) (\alpha,\beta,\gamma) ( α , β , γ ) and ( α , β ′ , γ ) (\alpha,\beta',\gamma) ( α , β ′ , γ ) form a perspective pair in O ( M † , ≤ ‡ ) \mathcal{O}(M^\dagger,\leq^\ddagger) O ( M † , ≤ ‡ ) ; in the notation of Sect. 6.1, γ \gamma γ and δ \delta δ are replaced by β ′ \beta' β ′ and γ \gamma γ . Since β ′ ∖ α = γ ∖ β \beta'\setminus\alpha=\gamma\setminus\beta β ′ ∖ α = γ ∖ β and γ ∖ β ′ = β ∖ α \gamma\setminus\beta'=\beta\setminus\alpha γ ∖ β ′ = β ∖ α , excision gives isomorphisms ℓ : E α β → E β ′ γ \ell:E^\beta_\alpha\to E^\gamma_{\beta'} ℓ : E α β → E β ′ γ and ℓ ′ : E α β ′ → E β γ \ell':E^{\beta'}_\alpha\to E^\gamma_\beta ℓ ′ : E α β ′ → E β γ with j ′ i ′ = ℓ j'i'=\ell j ′ i ′ = ℓ and j i ′ = ℓ ′ ji'=\ell' j i ′ = ℓ ′ , where i ′ : E α β ′ → E α γ i':E^{\beta'}_\alpha\to E^\gamma_\alpha i ′ : E α β ′ → E α γ and j ′ : E α γ → E β ′ γ j':E^\gamma_\alpha\to E^\gamma_{\beta'} j ′ : E α γ → E β ′ γ . Hence ℓ − 1 j ′ i ′ = id \ell^{-1}j'i'=\operatorname{id} ℓ − 1 j ′ i ′ = id , so that i i i is injective and k = 0 k=0 k = 0 , and j ′ ℓ ℓ − 1 = id j'\ell^{\ell-1} = \operatorname{id} j ′ ℓ ℓ − 1 = id , so that s = i ′ ℓ ℓ − 1 s = i'\ell^{\ell-1} s = i ′ ℓ ℓ − 1 is a section of j j j . Then ψ ( a , c ) = i ( a ) + s ( c ) \psi(a,c) = i(a) + s(c) ψ ( a , c ) = i ( a ) + s ( c ) is the required isomorphism, v. the proof of Thm. 6.3. The tail elements of Sect. 6.1 are not needed. Via the injection E ( M † , ⪯ ′ ) ↪ E ( M † , ⪯ ) \mathrm{E}(M^{\dagger},\preceq') \hookrightarrow\mathrm{E}(M^{\dagger},\preceq) E ( M † , ⪯ ′ ) ↪ E ( M † , ⪯ ) the same holds in E ( M † , ⪯ ′ ) \mathrm{E}(M^{\dagger},\preceq') E ( M † , ⪯ ′ ) , that is, in E ( M ) \mathrm{E}(M) E ( M ) .
Connecting orbits via higher order differentials
Corollary 6.12 yields an immediate result for a flow-order.
Theorem 7.2 Let ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) be a flow-order. Suppose the differential
d : S β γ δ δ 1 → S 0 α β γ , \mathrm{d}: S_{\beta\gamma}^{\delta\delta_1} \to S_{0\alpha}^{\beta\gamma}, d : S β γ δ δ 1 → S 0 α β γ , is non-trivial, with β ∖ α = { M } \beta\setminus\alpha= \{M\} β ∖ α = { M } and δ ∖ γ = { M ′ } \delta\setminus\gamma= \{M'\} δ ∖ γ = { M ′ } . Then, there exists a directed path { M k } k = 0 n ⊆ δ ∖ α \{M_k\}_{k=0}^{n} \subseteq\delta\setminus\alpha { M k } k = 0 n ⊆ δ ∖ α from M ′ = M n M' = M_n M ′ = M n to M = M 0 M = M_0 M = M 0 , such that for all 1 ≤ k ≤ n 1 \leq k \leq n 1 ≤ k ≤ n either C ( M k , M k − 1 ) ≠ ∅ C(M_k,M_{k-1}) \ne\varnothing C ( M k , M k − 1 ) = ∅ , or C ( M k − 1 , M k ) ≠ ∅ C(M_{k-1},M_k) \ne\varnothing C ( M k − 1 , M k ) = ∅ , depending on whether M k > M k − 1 M_k > M_{k-1} M k > M k − 1 , or M k < M k − 1 M_k < M_{k-1} M k < M k − 1 respectively.
Proof. Since d ≠ 0 \mathrm{d} \ne0 d = 0 , Corollary 6.12 implies that there exists a directed path { M k } \{M_k\} { M k } from M ′ M' M ′ to M M M . Since ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) is a flow-order, each such covering relation yields a connecting orbit between M k − 1 M_{k-1} M k − 1 and M k M_k M k , where the direction is determined by the order. Indeed, if M k − 1 < M k M_{k-1} < M_k M k − 1 < M k , then in the characterization of flow-orders after Defn. 2.2 the chain of connecting orbits from M k M_k M k to M k − 1 M_{k-1} M k − 1 has no intermediate elements, so that C ( M k , M k − 1 ) ≠ ∅ C(M_k,M_{k-1}) \ne\varnothing C ( M k , M k − 1 ) = ∅ .
Theorem 7.2 may be interpreted as the spectral systems analogue of [11 Prop. 5.3] . For arbitrary Morse representations the higher differentials in spectral systems still contain information about connecting orbits.
Theorem 7.3 Let ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) be a Morse representation and suppose the differential
d : S β γ δ δ 1 → S 0 α β γ , \mathrm{d}: S_{\beta\gamma}^{\delta\delta_1} \to S_{0\alpha}^{\beta\gamma}, d : S β γ δ δ 1 → S 0 α β γ , is non-trivial, with β ∖ α = { M } \beta\setminus\alpha= \{M\} β ∖ α = { M } and δ ∖ γ = { M ′ } \delta\setminus\gamma= \{M'\} δ ∖ γ = { M ′ } . Then, there exists a sequence { M k } k = 0 n ⊆ δ ∖ α \{M_k\}_{k=0}^{n} \subseteq\delta\setminus\alpha { M k } k = 0 n ⊆ δ ∖ α from M ′ = M n M' = M_n M ′ = M n to M = M 0 M = M_0 M = M 0 , with M 0 < M 1 M_0 < M_1 M 0 < M 1 and M n − 1 < M n M_{n-1} < M_n M n − 1 < M n , such that for all 1 ≤ k ≤ n 1 \leq k \leq n 1 ≤ k ≤ n either C ( M k , M k − 1 ) ≠ ∅ C(M_k,M_{k-1}) \ne\varnothing C ( M k , M k − 1 ) = ∅ , or C ( M k − 1 , M k ) ≠ ∅ C(M_{k-1},M_k) \ne\varnothing C ( M k − 1 , M k ) = ∅ , depending on whether M k > M k − 1 M_k > M_{k-1} M k > M k − 1 , or M k < M k − 1 M_k < M_{k-1} M k < M k − 1 respectively.
Proof. Let ⪯ ′ \preceq' ⪯ ′ be the flow-order obtained from ⪯ \preceq ⪯ by removing order relations, v. the characterization after Defn. 2.2. By Theorem 2.1, ( M , ⪯ ′ ) (M,\preceq') ( M , ⪯ ′ ) is again a Morse representation for ϕ \phi ϕ , the identity map is an order-surjection ( M , ⪯ ′ ) ↠ ( M , ⪯ ) (M,\preceq') \twoheadrightarrow(M,\preceq) ( M , ⪯ ′ ) ↠ ( M , ⪯ ) , and O ( M , ⪯ ) \mathrm{O}(M,\preceq) O ( M , ⪯ ) is a sublattice of O ( M , ⪯ ′ ) \mathrm{O}(M,\preceq') O ( M , ⪯ ′ ) . By Remark 3.2 and Birkhoff duality we obtain an injection E ( M , ⪯ ) ↪ E ( M , ⪯ ′ ) \mathrm{E}(M,\preceq) \hookrightarrow\mathrm{E}(M,\preceq') E ( M , ⪯ ) ↪ E ( M , ⪯ ′ ) , so that d \mathrm{d} d is also a non-trivial differential for ( M , ⪯ ′ ) (M,\preceq') ( M , ⪯ ′ ) . Theorem 7.2 for the flow-order ( M , ⪯ ′ ) (M,\preceq') ( M , ⪯ ′ ) yields a directed path { M k } k = 0 n ⊆ δ ∖ α \{M_k\}_{k=0}^{n} \subseteq\delta\setminus\alpha { M k } k = 0 n ⊆ δ ∖ α of ( M , ⪯ ′ ) (M,\preceq') ( M , ⪯ ′ ) from M ′ = M n M' = M_n M ′ = M n to M = M 0 M = M_0 M = M 0 , with connecting orbits as stated. By Corollary 6.12, the first and the last relation in this path are M 0 ≺ ′ M 1 M_0 \prec' M_1 M 0 ≺ ′ M 1 and M n − 1 ≺ ′ M n M_{n-1} \prec' M_n M n − 1 ≺ ′ M n , since M M M is minimal and M ′ M' M ′ is maximal in δ ∖ α \delta\setminus\alpha δ ∖ α . Hence M 0 < M 1 M_0 < M_1 M 0 < M 1 and M n − 1 < M n M_{n-1} < M_n M n − 1 < M n . Since ⪯ ′ ⊆ ⪯ \preceq' \subseteq\preceq ⪯ ′ ⊆⪯ , this proves the theorem.
If two Morse sets are adjacent in a Morse representation Theorem 7.3 yields a connecting orbit whenever d : S β β γ 1 → S 0 α β β \mathrm{d}: S_{\beta\beta}^{\gamma_1} \to S_{0\alpha}^{\beta\beta} d : S β β γ 1 → S 0 α β β is non-trivial, which is an extension of Bartłomiejczyk’s approach via spectral sequences. By Remark 6.11 and Theorem 7.3 we obtain:
Corollary 7.4 Let α ⪯ β ⪯ γ \alpha\preceq\beta\preceq\gamma α ⪯ β ⪯ γ and β ∖ α = { M } \beta\setminus\alpha= \{M\} β ∖ α = { M } and γ ∖ β = { M ′ } \gamma\setminus\beta= \{M'\} γ ∖ β = { M ′ } . Suppose the connecting homomorphism
k : E β γ → E α β , k: E_{\beta}^{\gamma} \to E_{\alpha}^{\beta}, k : E β γ → E α β , is non-trivial. Then, the set of connecting orbits C ( M ′ , M ) C(M',M) C ( M ′ , M ) from M ′ M' M ′ to M M M is non-empty.
Proof. The connecting homomorphism k k k is the differential d : S β β γ γ → S α α β β \mathrm{d}: S_{\beta\beta}^{\gamma\gamma} \to S_{\alpha\alpha}^{\beta\beta} d : S β β γ γ → S α α β β . By Remark 6.11 , k ≠ 0 k \ne0 k = 0 if and only if d : S β β γ γ → S α α β β \mathrm{d}: S_{\beta\beta}^{\gamma\gamma} \to S_{\alpha\alpha}^{\beta\beta} d : S β β γ γ → S α α β β is non-trivial. Theorem 7.3 , with ( α , β , β , γ ) (\alpha,\beta,\beta,\gamma) ( α , β , β , γ ) in place of ( α , β , γ , δ ) (\alpha,\beta,\gamma,\delta) ( α , β , γ , δ ) , applies, and since γ ∖ α = { M , M ′ } \gamma\setminus\alpha= \{M,M'\} γ ∖ α = { M , M ′ } , it yields M < M ′ M<M' M < M ′ and C ( M ′ , M ) ≠ ∅ C(M',M)\ne\varnothing C ( M ′ , M ) = ∅ . This also follows from Lemma 7.1 .
The next result reveals how non-trivial higher differentials in a spectral system S S S contain information about connecting orbits to and from specific Morse sets.
Theorem 7.5 Let ( M , ≤ ) (M,\leq) ( M , ≤ ) be a Morse representation and suppose the differential
d : S γ δ ϵ ζ → S α β γ δ \mathrm{d}:S_{\gamma\delta}^{\epsilon\zeta}\to S_{\alpha\beta}^{\gamma\delta} d : S γ δ ϵ ζ → S α β γ δ is non-trivial, with γ ∖ β = { M } \gamma\setminus\beta=\{M\} γ ∖ β = { M } and ϵ ∖ δ = { M ′ } \epsilon\setminus\delta=\{M'\} ϵ ∖ δ = { M ′ } . Then,
there exists a Morse set M ′ ′ ∈ δ ∖ β M''\in\delta\setminus\beta M ′′ ∈ δ ∖ β , M ′ ′ < M ′ M''<M' M ′′ < M ′ , such that C ( M ′ , M ′ ′ ) ≠ ∅ C(M',M'')\ne\varnothing C ( M ′ , M ′′ ) = ∅ ;
there exists a Morse set M ′ ′ ′ ∈ ϵ ∖ γ M'''\in\epsilon\setminus\gamma M ′′′ ∈ ϵ ∖ γ , M < M ′ ′ ′ M<M''' M < M ′′′ , such that C ( M ′ ′ ′ , M ) ≠ ∅ C(M''',M)\ne\varnothing C ( M ′′′ , M ) = ∅ .
Proof. By Remark 6.11 , the differential d : S γ δ ϵ ζ → S α β γ δ \mathrm{d}:S_{\gamma\delta}^{\epsilon\zeta}\to S_{\alpha\beta}^{\gamma\delta} d : S γ δ ϵ ζ → S α β γ δ is non-trivial. Theorem 7.3 , with ( β , γ , δ , ϵ ) (\beta,\gamma,\delta,\epsilon) ( β , γ , δ , ϵ ) in place of ( α , β , γ , δ ) (\alpha,\beta,\gamma,\delta) ( α , β , γ , δ ) , yields a sequence { M k } k = 0 n ⊆ ϵ ∖ β \{M_k\}_{k=0}^{n}\subseteq\epsilon\setminus\beta { M k } k = 0 n ⊆ ϵ ∖ β from M ′ = M n M'=M_n M ′ = M n to M = M 0 M=M_0 M = M 0 , with M 0 < M 1 M_0<M_1 M 0 < M 1 and M n − 1 < M n M_{n-1}<M_n M n − 1 < M n , and with connecting orbits between consecutive elements. Hence (i) holds with M ′ ′ = M n − 1 M''=M_{n-1} M ′′ = M n − 1 : C ( M ′ , M ′ ′ ) ≠ ∅ C(M',M'')\ne\varnothing C ( M ′ , M ′′ ) = ∅ , M ′ ′ < M ′ M''<M' M ′′ < M ′ , and M ′ ′ ∈ ( ϵ ∖ β ) ∖ { M ′ } = δ ∖ β M''\in(\epsilon\setminus\beta)\setminus\{M'\}=\delta\setminus\beta M ′′ ∈ ( ϵ ∖ β ) ∖ { M ′ } = δ ∖ β . Similarly, (ii) holds with M ′ ′ ′ = M 1 M'''=M_1 M ′′′ = M 1 : C ( M ′ ′ ′ , M ) ≠ ∅ C(M''',M)\ne\varnothing C ( M ′′′ , M ) = ∅ , M < M ′ ′ ′ M<M''' M < M ′′′ , and M ′ ′ ′ ∈ ( ϵ ∖ β ) ∖ { M } = ϵ ∖ γ M'''\in(\epsilon\setminus\beta)\setminus\{M\}=\epsilon\setminus\gamma M ′′′ ∈ ( ϵ ∖ β ) ∖ { M } = ϵ ∖ γ .
Remark 7.6 If M ∥ M ′ M\parallel M' M ∥ M ′ in Theorem 7.5 , then the Morse sets M ′ ′ M'' M ′′ and M ′ ′ ′ M''' M ′′′ in (i) and (ii) are different, since M ′ ′ = M ′ ′ ′ M''=M''' M ′′ = M ′′′ would give M < M ′ ′ ′ = M ′ ′ < M ′ M<M'''=M''<M' M < M ′′′ = M ′′ < M ′ . Hence there exist two different Morse sets M ′ ′ ′ > M M'''>M M ′′′ > M and M ′ ′ < M ′ M''<M' M ′′ < M ′ , and connections C ( M ′ ′ ′ , M ) ≠ ∅ C(M''',M)\ne\varnothing C ( M ′′′ , M ) = ∅ and C ( M ′ , M ′ ′ ) ≠ ∅ C(M',M'')\ne\varnothing C ( M ′ , M ′′ ) = ∅ . Example 8.2 will illustrate that considering spectral sequences as opposed to a spectral net will not reveal most of the orbit structure.
Remark 7.7 The following property of spectral systems gives a direct proof of Theorem 7.5 . Let α p \alpha_p α p , p ∈ Z p\in\mathbb{Z} p ∈ Z , be a maximal chain in O ( M ) O(M) O ( M ) as in Example 4.6, with
S p k = S α p α p − 1 α p − k α p − 1 − k S_p^k=S_{\alpha_p\alpha_{p-1}}^{\alpha_{p-k}\alpha_{p-1-k}} S p k = S α p α p − 1 α p − k α p − 1 − k
and
d p k : S p k → S p − k k \mathrm{d}_p^k:S_p^k\to S_{p-k}^k d p k : S p k → S p − k k
and let r ≥ 1 r\geq1 r ≥ 1 . For 1 ≤ k ≤ r 1\leq k\leq r 1 ≤ k ≤ r consider the differentials
d k : S α p α p − 1 α p − k α p − 1 − k → S α p − k α p − 1 − k α p − r − k α p − r − 1 . \mathrm{d}^k:S_{\alpha_p\alpha_{p-1}}^{\alpha_{p-k}\alpha_{p-1-k}}\to S_{\alpha_{p-k}\alpha_{p-1-k}}^{\alpha_{p-r-k}\alpha_{p-r-1}}. d k : S α p α p − 1 α p − k α p − 1 − k → S α p − k α p − 1 − k α p − r − k α p − r − 1 .
For k = r k=r k = r this is the differential d p r : S p r → S p − r r \mathrm{d}_p^r:S_p^r\to S_{p-r}^r d p r : S p r → S p − r r of the spectral sequence of the chain, and for k = 1 k=1 k = 1 it is the first order differential d α p α p − 1 α p − r − 1 : E α p − 1 α p → E α p − r − 1 α p − 1 \mathrm{d}_{\alpha_p\alpha_{p-1}\alpha_{p-r-1}}:E_{\alpha_{p-1}}^{\alpha_p}\to E_{\alpha_{p-r-1}}^{\alpha_{p-1}} d α p α p − 1 α p − r − 1 : E α p − 1 α p → E α p − r − 1 α p − 1 . For 1 < k < r 1<k<r 1 < k < r , d k \mathrm{d}^k d k is a differential in the spectral net which is not part of the spectral sequence. These differentials satisfy
d k + 1 ≠ 0 ⟹ d k ≠ 0 , 1 ≤ k ≤ r − 1. \mathrm{d}^{k+1}\ne0\quad\Longrightarrow\quad\mathrm{d}^k\ne0,\qquad1\leq k\leq r-1. d k + 1 = 0 ⟹ d k = 0 , 1 ≤ k ≤ r − 1.
This follows from the commutative diagram
S p k = S α p α p − 1 α p − k α p − 1 − k → d k S α p − k α p − 1 − k α p − r − k α p − r − 1 j ↓ d ~ ↓ i ↑ S α p α p − 1 α p − k α p − 1 − k → d ~ S α p − k α p − 1 − k α p − r − k − 1 α p − r − 1 i ↑ i ↑ S p k + 1 = S α p α p − 1 α p − k − 1 α p − 1 − k → d k + 1 S α p − k − 1 α p − 1 − k α p − r − k − 1 α p − r − 1 \begin{CD}
S_p^k @= S_{\alpha_p\alpha_{p-1}}^{\alpha_{p-k}\alpha_{p-1-k}} @>{\mathrm{d}^k}>> S_{\alpha_{p-k}\alpha_{p-1-k}}^{\alpha_{p-r-k}\alpha_{p-r-1}} \\
@V{j}VV @V{\tilde{\mathrm{d}}}VV @A{i}AA \\
S_{\alpha_p\alpha_{p-1}}^{\alpha_{p-k}\alpha_{p-1-k}} @>{\tilde{\mathrm{d}}}>> S_{\alpha_{p-k}\alpha_{p-1-k}}^{\alpha_{p-r-k-1}\alpha_{p-r-1}} \\
@A{i}AA @A{i}AA \\
S_p^{k+1} @= S_{\alpha_p\alpha_{p-1}}^{\alpha_{p-k-1}\alpha_{p-1-k}} @>{\mathrm{d}^{k+1}}>> S_{\alpha_{p-k-1}\alpha_{p-1-k}}^{\alpha_{p-r-k-1}\alpha_{p-r-1}}
\end{CD} S p k j ↓ ⏐ S α p α p − 1 α p − k α p − 1 − k i ⏐ ↑ S p k + 1 d ~ S α p α p − 1 α p − k α p − 1 − k d ~ ↓ ⏐ S α p − k α p − 1 − k α p − r − k − 1 α p − r − 1 i ⏐ ↑ S α p α p − 1 α p − k − 1 α p − 1 − k d k d k + 1 S α p − k α p − 1 − k α p − r − k α p − r − 1 i ⏐ ↑ S α p − k − 1 α p − 1 − k α p − r − k − 1 α p − r − 1
The maps i i i are injective and j j j is surjective, v. Defn. 4.1 (iii) and (v), and the squares commute by Defn. 4.1 (iv). The upper square gives d k = i d ~ j \mathrm{d}^k=i\tilde{\mathrm{d}}j d k = i d ~ j , so that d k ≠ 0 \mathrm{d}^k\ne0 d k = 0 if and only if d ~ ≠ 0 \tilde{d} \ne0 d ~ = 0 . This is also Remark 6.11 , since d k d^k d k and d ~ \tilde{d} d ~ differ only in the labels ζ \zeta ζ and α \alpha α . The lower square gives d ~ i = i d k + 1 \tilde{d}i = i d^{k+1} d ~ i = i d k + 1 , that is, d k + 1 d^{k+1} d k + 1 is the restriction of d ~ \tilde{d} d ~ to subterms. If d α k + 1 ≠ 0 d^{k+1}_\alpha\ne0 d α k + 1 = 0 , then d ~ ( i a ) = i d k + 1 a ≠ 0 \tilde{d}(i a) = i d^{k+1}a \ne0 d ~ ( ia ) = i d k + 1 a = 0 , since i i i is injective, so that d ~ ≠ 0 \tilde{d} \ne0 d ~ = 0 , and hence d k ≠ 0 d^k \ne0 d k = 0 . Starting with d r = d p r d^r = d^r_p d r = d p r and descending through k = r − 1 , … , 1 k = r-1,\ldots,1 k = r − 1 , … , 1 , a non-trivial differential d p r d^r_p d p r forces a staircase of non-trivial differentials d r − 1 , … , d 1 d^{r-1},\ldots,d^1 d r − 1 , … , d 1 in the spectral net. Along the staircase the sources run through the terms S p k S^k_p S p k of the spectral sequence, from S p r S^r_p S p r down to S p 1 = E α p − 1 α p = HC ( M p ) S^1_p = E^{\alpha_p}_{\alpha_{p-1}} = \operatorname{HC}(M_p) S p 1 = E α p − 1 α p = HC ( M p ) , while the targets grow from S p − r r S^r_{p-r} S p − r r to E α p − r α p − r − 1 E^{\alpha_{p-r-1}}_{\alpha_{p-r}} E α p − r α p − r − 1 , the Conley index of the Morse set of { M p − r , … , M p − 1 } \{M_{p-r},\ldots,M_{p-1}\} { M p − r , … , M p − 1 } . Similarly, the differentials
d ∗ k : S α p − r α p − r + k − 1 α p α p + k − 1 ⟶ S α p − r − k α p − r − 1 α p − r α p − r + k − 1 , 1 ≤ k ≤ r , d_*^k:S^{\alpha_p\alpha_{p+k-1}}_{\alpha_{p-r}\alpha_{p-r+k-1}}\longrightarrow S^{\alpha_{p-r}\alpha_{p-r+k-1}}_{\alpha_{p-r-k}\alpha_{p-r-1}},\qquad1\leq k\leq r, d ∗ k : S α p − r α p − r + k − 1 α p α p + k − 1 ⟶ S α p − r − k α p − r − 1 α p − r α p − r + k − 1 , 1 ≤ k ≤ r ,
connect d ∗ r = d p r d_*^r=d_p^r d ∗ r = d p r with the first order differential d ∗ 1 = d α p α p − r α p − r − 1 : E α p − r α p ⟶ E α p − r − 1 α p − r d_*^1=d_{\alpha_p\alpha_{p-r}\alpha_{p-r-1}}:E^{\alpha_p}_{\alpha_{p-r}}\longrightarrow E^{\alpha_{p-r}}_{\alpha_{p-r-1}} d ∗ 1 = d α p α p − r α p − r − 1 : E α p − r α p ⟶ E α p − r − 1 α p − r , and d ∗ k + 1 ≠ 0 d_*^{k+1}\ne0 d ∗ k + 1 = 0 implies d ∗ k ≠ 0 d_*^k\ne0 d ∗ k = 0 ; here d ∗ k + 1 d_*^{k+1} d ∗ k + 1 is induced by the corresponding differential d ~ \tilde{d} d ~ on quotients. By Lemma 7.1 , the two first order differentials yield connecting orbits from M p M_p M p to the Morse set of { M p − r , … , M p − 1 } \{M_{p-r},\ldots,M_{p-1}\} { M p − r , … , M p − 1 } and from the Morse set of { M p − r + 1 , … , M p } \{M_{p-r+1},\ldots,M_p\} { M p − r + 1 , … , M p } to M p − r M_{p-r} M p − r . Together with Theorem 2.1 this gives Theorem 7.5 .
Connecting orbits via derived differentials
Corollary 7.4 gives a statement about first order differentials and the existence of connecting orbits. It may happen that k = 0 k=0 k = 0 even though connecting orbits may be forced by the spectral system, v. Ex. 8.4. In this case the first order derived differential may detect such connecting orbits.
Let α ⊆ β ⊆ γ \alpha\subseteq\beta\subseteq\gamma α ⊆ β ⊆ γ and β ∖ α = { M } \beta\setminus\alpha=\{M\} β ∖ α = { M } and γ ∖ β = { M ′ } \gamma\setminus\beta=\{M'\} γ ∖ β = { M ′ } . In this case the Morse set associated to γ ∖ α \gamma\setminus\alpha γ ∖ α is denoted by M γ ∖ α = Inv ( G γ ∖ α U ) M_{\gamma\setminus\alpha}=\operatorname{Inv}(G_{\gamma\setminus\alpha}U) M γ ∖ α = Inv ( G γ ∖ α U ) and
M γ ∖ α = M ∪ M ′ ∪ C ( M ′ , M ) . M_{\gamma\setminus\alpha}=M\cup M'\cup C(M',M). M γ ∖ α = M ∪ M ′ ∪ C ( M ′ , M ) .
The Conley indices are given by HC ( M ) = E α β \operatorname{HC}(M)=E^\beta_\alpha HC ( M ) = E α β , HC ( M ′ ) = E β γ \operatorname{HC}(M')=E^\gamma_\beta HC ( M ′ ) = E β γ and HC ( M γ ∖ α ) = E α γ \operatorname{HC}(M_{\gamma\setminus\alpha})=E^\gamma_\alpha HC ( M γ ∖ α ) = E α γ .
Lemma 7.8 C ( M ′ , M ) = ∅ C(M',M)=\varnothing C ( M ′ , M ) = ∅ implies that there exists an isomorphism ψ : E α β ⊕ E β γ ⟶ E α γ \psi:E^\beta_\alpha\oplus E^\gamma_\beta\longrightarrow E^\gamma_\alpha ψ : E α β ⊕ E β γ ⟶ E α γ with ψ ι = i \psi\iota=i ψ ι = i and j ψ = π j\psi=\pi j ψ = π , v. Thm. 6.3 . In particular, E α γ ≅ E α β ⊕ E β γ E^\gamma_\alpha\cong E^\beta_\alpha\oplus E^\gamma_\beta E α γ ≅ E α β ⊕ E β γ .
Proof. This is Lemma 7.1 with β ∖ α = { M } \beta\setminus\alpha=\{M\} β ∖ α = { M } and γ ∖ β = { M ′ } \gamma\setminus\beta=\{M'\} γ ∖ β = { M ′ } , for which, in the notation of Lemma 7.1 , A = M A=M A = M and A ∗ = M ′ A^*=M' A ∗ = M ′ .
Remark 7.9 To prove Lemma 7.8 we can also use the property that the Conley index of a disjoint union is isomorphic to the direct sum of the Conley indices. In general E α γ ≅ E α β ⊕ E β γ E^\gamma_\alpha\cong E^\beta_\alpha\oplus E^\gamma_\beta E α γ ≅ E α β ⊕ E β γ does not imply that C ( M ′ , M ) = ∅ C(M',M)=\varnothing C ( M ′ , M ) = ∅ as is shown in Example 8.4.
Using derived differentials in combination with Lemma 7.8 allows us to prove the following extension of Corollary 7.4:
Theorem 7.10 Let α ⊆ β ⊆ γ \alpha\subseteq\beta\subseteq\gamma α ⊆ β ⊆ γ and β ∖ α = { M } \beta\setminus\alpha=\{M\} β ∖ α = { M } and γ ∖ β = { M ′ } \gamma\setminus\beta=\{M'\} γ ∖ β = { M ′ } . Consider the connecting homomorphism (or first order differential) and the first order derived differential
κ : E β γ ⟶ E α β , κ : S α γ γ ⟶ S α β γ [ 1 ] . \kappa:E^\gamma_\beta\longrightarrow E^\beta_\alpha,\qquad\kappa:S^{\gamma\gamma}_\alpha\longrightarrow S^{\beta\gamma}_\alpha[1]. κ : E β γ ⟶ E α β , κ : S α γ γ ⟶ S α β γ [ 1 ] . If ( κ , κ ) ≠ 0 (\kappa,\kappa)\ne0 ( κ , κ ) = 0 , then the set of connecting orbits C ( M ′ , M ) C(M',M) C ( M ′ , M ) from M ′ M' M ′ to M M M is non-empty.
Proof. By Corollary 6.4 the condition ( κ , κ ) ≠ 0 (\kappa,\kappa)\ne0 ( κ , κ ) = 0 implies that there is no isomorphism ψ : E α β ⊕ E β γ ⟶ E α γ \psi:E^\beta_\alpha\oplus E^\gamma_\beta\longrightarrow E^\gamma_\alpha ψ : E α β ⊕ E β γ ⟶ E α γ with ψ ι = i \psi\iota=i ψ ι = i and j ψ = π j\psi=\pi j ψ = π . By the contrapositive of Lemma 7.8 this implies that C ( M ′ , M ) ≠ ∅ C(M',M)\ne\varnothing C ( M ′ , M ) = ∅ .
:::
In the case that k = 0 k=0 k = 0 the derived differential is given by κ : E β γ → E α β [ 1 ] \kappa:E^\gamma_\beta\to E^\beta_\alpha[1] κ : E β γ → E α β [ 1 ] in the derived category, which justifies the terminology derived differential and reveals the higher order information hidden in the exact triangles. The occurrence of derived differentials for differentials between S S S -terms in general is a subject of future study, cf. [4 ] .
Examples
In this section we give some examples on how spectral nets are utilized to compute homology and find connecting orbits. The apparatus of spectral nets serves both purposes.
Example 8.1 Consider a flow ϕ \phi ϕ on the 2-disk D 2 D^2 D 2 as depicted in Figure 8.1 . The flow exhibits a stable periodic orbit M 1 M_1 M 1 , a degenerate periodic orbit M 2 M_2 M 2 and an unstable spiral M 3 M_3 M 3 . Figure 8.2 displays a filtering of trapping regions, which is a 4-element lattice isomorphic to O ( P ) O(P) O ( P ) , where P P P is the linear poset 1 < 2 < 3 1<2<3 1 < 2 < 3 . Specifically, F 0 D 2 = ∅ F_0D^2=\varnothing F 0 D 2 = ∅ , F α D 2 F_\alpha D^2 F α D 2 and F β D 2 F_\beta D^2 F β D 2 are annuli and F 1 D 2 = D 2 F_1D^2=D^2 F 1 D 2 = D 2 is the whole disk.
Figure 8.1. The flow ϕ \phi ϕ with Morse tessellation[left], the Morse representation M[middle], and the down-sets O(M)[right].
The associated Morse tiles G p D 2 G_pD^2 G p D 2 , p = 1 , 2 , 3 p=1,2,3 p = 1 , 2 , 3 , give isolating neighborhoods for the Morse sets M 1 M_1 M 1 , M 2 M_2 M 2 and M 3 M_3 M 3 respectively. The filtering may be regarded as a map O ( M ) → Set ( D 2 ) O(M)\to\operatorname{Set}(D^2) O ( M ) → Set ( D 2 ) , with O ( M ) = { 0 , α , β , 1 } O(M)=\{0,\alpha,\beta,1\} O ( M ) = { 0 , α , β , 1 } , α = { M 1 } \alpha=\{M_1\} α = { M 1 } , β = { M 1 , M 2 } \beta=\{M_1,M_2\} β = { M 1 , M 2 } . This implies that the flow allows a Morse representation M M M given by M 1 < M 2 < M 3 M_1<M_2<M_3 M 1 < M 2 < M 3 . We emphasize that at this point we do not know the latter is a flow-order. In this example we can describe the Cartan–Eilenberg system E ( M ) E(M) E ( M ) with ring coefficients R = Z R=\mathbb{Z} R = Z . For the filtering we have:
E 0 α = H ( F α D 2 ) ≅ Z ⊕ Z , E 0 β = H ( F β D 2 ) ≅ Z ⊕ Z , E 0 1 = H ( D 2 ) ≅ Z . E^\alpha_0=H(F_\alpha D^2)\cong\mathbb{Z}\oplus\mathbb{Z},\qquad E^\beta_0=H(F_\beta D^2)\cong\mathbb{Z}\oplus\mathbb{Z},\qquad E^1_0=H(D^2)\cong\mathbb{Z}. E 0 α = H ( F α D 2 ) ≅ Z ⊕ Z , E 0 β = H ( F β D 2 ) ≅ Z ⊕ Z , E 0 1 = H ( D 2 ) ≅ Z . Other E E E -terms that can be evaluated are
H C ( M 1 ) = E 0 α ≅ Z ⊕ Z ; H C ( M 2 ) = E α β = H ( F β D 2 , F α D 2 ) ≃ 0 ; H C ( M 3 ) = E 1 β = H ( F 1 D 2 , F β D 2 ) ≅ Z . \begin{aligned}
HC(M_1)&=E^\alpha_0\cong\mathbb{Z}\oplus\mathbb{Z};\\
HC(M_2)&=E^\beta_\alpha=H(F_\beta D^2,F_\alpha D^2)\simeq0;\\
HC(M_3)&=E^\beta_1=H(F_1D^2,F_\beta D^2)\cong\mathbb{Z}.
\end{aligned} H C ( M 1 ) H C ( M 2 ) H C ( M 3 ) = E 0 α ≅ Z ⊕ Z ; = E α β = H ( F β D 2 , F α D 2 ) ≃ 0 ; = E 1 β = H ( F 1 D 2 , F β D 2 ) ≅ Z . If we invoke the spectral net approach most of the information about connecting orbits can be obtained from this data without further knowledge of the Cartan–Eilenberg system. Consider the page-1 spectral sequence based on the Conley index data:
0 ⟶ H C ( M 3 ) → d 23 1 H C ( M 2 ) → d 12 1 H C ( M 1 ) ⟶ 0 0\longrightarrow HC(M_3)\xrightarrow{d^1_{23}}HC(M_2)\xrightarrow{d^1_{12}}HC(M_1)\longrightarrow0 0 ⟶ H C ( M 3 ) d 23 1 H C ( M 2 ) d 12 1 H C ( M 1 ) ⟶ 0
Figure 8.2. The filtering O(M) → Set(D 2 D^2 D 2 ), with the periodic orbits highlighted.
If we incorporate the information about the Conley indices and differentials we obtain:
0 ⟶ Z → 0 0 → 0 Z ⊕ Z ⟶ 0 0 \longrightarrow\mathbb{Z} \xrightarrow{0} 0 \xrightarrow{0} \mathbb{Z}\oplus\mathbb{Z} \longrightarrow0 0 ⟶ Z 0 0 0 Z ⊕ Z ⟶ 0
Computing homology we obtain the page-2, where the middle differential is still to be determined:
⋯ ⟶ 0 ⟶ Z ⟶ Z ⊕ Z ⟶ 0 ⟶ ⋯ \begin{aligned}
\cdots&\longrightarrow0 \longrightarrow\mathbb{Z} \longrightarrow\mathbb{Z}\oplus\mathbb{Z} \longrightarrow0 \longrightarrow\cdots
\end{aligned} ⋯ ⟶ 0 ⟶ Z ⟶ Z ⊕ Z ⟶ 0 ⟶ ⋯
By a classical result of spectral sequence theory, v. Sect. 4.3, the infinity page ⨁ p S p ∞ \bigoplus_p S_p^\infty ⨁ p S p ∞ is isomorphic to Gr E 1 0 = Gr H ( D 2 ) \operatorname{Gr}E_1^0=\operatorname{Gr}H(D^2) Gr E 1 0 = Gr H ( D 2 ) , and rank ( Gr E 1 0 ) = rank ( H ( D 2 ) ) = 1 \operatorname{rank}(\operatorname{Gr}E_1^0)=\operatorname{rank}(H(D^2))=1 rank ( Gr E 1 0 ) = rank ( H ( D 2 )) = 1 . If d 13 2 = 0 d_{13}^2=0 d 13 2 = 0 , then Gr E 1 0 ≅ Z ⊕ Z ⊕ Z \operatorname{Gr}E_1^0\cong\mathbb{Z}\oplus\mathbb{Z}\oplus\mathbb{Z} Gr E 1 0 ≅ Z ⊕ Z ⊕ Z , which is a contradiction. We conclude that d 13 2 ≠ 0 d_{13}^2\ne0 d 13 2 = 0 and d 13 2 d_{13}^2 d 13 2 is an injective map. Consequently, the system converges at page-3 and d r = 0 d^r=0 d r = 0 for r ≥ 3 r\ge3 r ≥ 3 . If we keep track of the grading of homology, then the disk has non-trivial homology only in dimension 0. Using the grading on homology, and ordering the summands by decreasing degree, HC ( M 1 ) = HC 1 ( M 1 ) ⊕ HC 0 ( M 1 ) ≅ Z ⊕ Z \operatorname{HC}(M_1)=\operatorname{HC}_1(M_1)\oplus\operatorname{HC}_0(M_1)\cong\mathbb{Z}\oplus\mathbb{Z} HC ( M 1 ) = HC 1 ( M 1 ) ⊕ HC 0 ( M 1 ) ≅ Z ⊕ Z and HC ( M 3 ) = HC 2 ( M 3 ) ≅ Z \operatorname{HC}(M_3)=\operatorname{HC}_2(M_3)\cong\mathbb{Z} HC ( M 3 ) = HC 2 ( M 3 ) ≅ Z , we obtain, up to the choice of generators:
d 13 2 [ 1 0 ] : Z ⟶ Z ⊕ Z , (29) d_{13}^2
\begin{bmatrix}
1\\
0
\end{bmatrix}
:\mathbb{Z}\longrightarrow\mathbb{Z}\oplus\mathbb{Z}, \tag*{(29)} d 13 2 [ 1 0 ] : Z ⟶ Z ⊕ Z , (29)
because the generators in dimensions 1 and 2 have to be annihilated. Since in this example all E E E -terms are given we can compute all differentials explicitly.
With this example we want to illustrate that with spectral nets/sequences information about connecting orbits may be obtained without knowing the E E E -system (S S S -system, or spectral matrix). Note that,
d 13 2 : S α β 11 ⟶ S 00 α β ≠ 0 d_{13}^2:S_{\alpha\beta}^{11}\longrightarrow S_{00}^{\alpha\beta}\ne0 d 13 2 : S α β 11 ⟶ S 00 α β = 0
which is a second order differential, and we emphasize that we are not assuming that ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) is a flow-order. By Theorem 7.5 there exists a Morse set M ′ ′ ′ ∈ { M 2 , M 3 } M'''\in\{M_2,M_3\} M ′′′ ∈ { M 2 , M 3 } such that C ( M ′ ′ ′ , M 1 ) ≠ ∅ C(M''',M_1)\ne\varnothing C ( M ′′′ , M 1 ) = ∅ and a Morse set M ′ ′ ∈ { M 1 , M 2 } M''\in\{M_1,M_2\} M ′′ ∈ { M 1 , M 2 } such that C ( M 3 , M ′ ′ ) ≠ ∅ C(M_3,M'')\ne\varnothing C ( M 3 , M ′′ ) = ∅ . The periodic orbit M 2 M_2 M 2 is an invariant circle which separates M 3 M_3 M 3 from M 1 M_1 M 1 in D 2 D^2 D 2 , so that no orbit goes from M 3 M_3 M 3 to M 1 M_1 M 1 , i.e. C ( M 3 , M 1 ) = ∅ C(M_3,M_1)=\varnothing C ( M 3 , M 1 ) = ∅ . Hence, the only possible Morse set is M ′ ′ = M ′ ′ ′ = M 2 M''=M'''=M_2 M ′′ = M ′′′ = M 2 and therefore C ( M 2 , M 1 ) ≠ ∅ C(M_2,M_1)\ne\varnothing C ( M 2 , M 1 ) = ∅ and C ( M 3 , M 2 ) ≠ ∅ C(M_3,M_2)\ne\varnothing C ( M 3 , M 2 ) = ∅ . The latter also proves that ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) is flow-order, v. Defn. 2.2 .
This example shows that minimal algebraic information provides insight into Morse representations and even flow-orders. In particular, higher order differentials play a role in finding connecting orbits. Observe that algebraically the Morse set M 2 M_2 M 2 is “trivial”. Under small perturbations the connections found by Theorem 7.5 may become a connection from M 3 M_3 M 3 to M 1 M_1 M 1 . In the next example we adjust Example 8.1 slightly such that higher order differentials are used for finding connections between topologically non-trivial Morse sets.
Example 8.2 Consider a similar flow ϕ \phi ϕ on the 2-disk D 2 D^2 D 2 as shown in Figure 8.3 . We consider a Morse representation consisting of { M 1 , M 2 , M 2 ′ , M 3 } \{M_1,M_2,M'_2,M_3\} { M 1 , M 2 , M 2 ′ , M 3 } where M 1 M_1 M 1 and M 2 M_2 M 2 are stable periodic orbits, M 2 ′ M'_2 M 2 ′ is an unstable periodic orbit, and M 3 M_3 M 3 is an unstable spiral. We consider a filtering of D 2 D^2 D 2 based on the down-sets of the Morse sets, v. Figure 8.3 [right]. The filtering determines a Morse representation ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) , v. Figure 8.3 [middle]. As before we do not assume that ( M , ⪯ ) (M,\preceq) ( M , ⪯ ) is a flow-order. For the
Figure 8.3. The flow ϕ \phi ϕ with Morse tessellation[left], the Morse representation M[middle], and the down-sets O(M)[right].
Cartan–Eilenberg system E ( M ) E(M) E ( M ) (with ring coefficients R = Z R=\mathbb{Z} R = Z ) we consider the E E E -terms associated with the lattice of down-sets O ( M ) O(M) O ( M ) :
E 0 α = H ( F α D 2 ) ≅ Z ⊕ Z , E 0 β = H ( F β D 2 ) ≅ Z ⊕ Z , E 0 γ = H ( F γ D 2 ) ≅ Z ⊕ Z ⊕ Z ⊕ Z , E_0^\alpha=H(F_\alpha D^2)\cong\mathbb{Z}\oplus\mathbb{Z},\quad E_0^\beta=H(F_\beta D^2)\cong\mathbb{Z}\oplus\mathbb{Z},\quad E_0^\gamma=H(F_\gamma D^2)\cong\mathbb{Z}\oplus\mathbb{Z}\oplus\mathbb{Z}\oplus\mathbb{Z}, E 0 α = H ( F α D 2 ) ≅ Z ⊕ Z , E 0 β = H ( F β D 2 ) ≅ Z ⊕ Z , E 0 γ = H ( F γ D 2 ) ≅ Z ⊕ Z ⊕ Z ⊕ Z ,
E 0 δ = H ( F δ D 2 ) ≅ Z , E 0 ϵ = H ( F ϵ D 2 ) ≅ Z ⊕ Z , E 0 ζ = H ( F ζ D 2 ) ≅ Z ⊕ Z ⊕ Z . E_0^\delta=H(F_\delta D^2)\cong\mathbb{Z},\quad E_0^\epsilon=H(F_\epsilon D^2)\cong\mathbb{Z}\oplus\mathbb{Z},\quad E_0^\zeta=H(F_\zeta D^2)\cong\mathbb{Z}\oplus\mathbb{Z}\oplus\mathbb{Z}. E 0 δ = H ( F δ D 2 ) ≅ Z , E 0 ϵ = H ( F ϵ D 2 ) ≅ Z ⊕ Z , E 0 ζ = H ( F ζ D 2 ) ≅ Z ⊕ Z ⊕ Z .
and the homology of the whole space E 0 1 = H ( D 2 ) ≅ Z E_0^1=H(D^2)\cong\mathbb{Z} E 0 1 = H ( D 2 ) ≅ Z . The Conley indices of the Morse sets are given by
H C ( M 1 ) = E 0 α ≅ E δ 0 ≅ E β 0 ≅ Z ⊕ Z , H C ( M 2 ′ ) = E γ 0 ≅ E ϵ 1 ≅ Z ⊕ Z ; HC(M_1)=E_0^\alpha\cong E_\delta^0\cong E_\beta^0\cong\mathbb{Z}\oplus\mathbb{Z},\quad HC(M'_2)=E_\gamma^0\cong E_\epsilon^1\cong\mathbb{Z}\oplus\mathbb{Z}; H C ( M 1 ) = E 0 α ≅ E δ 0 ≅ E β 0 ≅ Z ⊕ Z , H C ( M 2 ′ ) = E γ 0 ≅ E ϵ 1 ≅ Z ⊕ Z ;
H C ( M 2 ) = E 0 β ≅ E α 1 ≅ Z ⊕ Z , H C ( M 3 ) = E 0 β ≅ E γ 1 ≅ E ϵ 1 ≅ Z . HC(M_2)=E_0^\beta\cong E_\alpha^1\cong\mathbb{Z}\oplus\mathbb{Z},\quad HC(M_3)=E_0^\beta\cong E_\gamma^1\cong E_\epsilon^1\cong\mathbb{Z}. H C ( M 2 ) = E 0 β ≅ E α 1 ≅ Z ⊕ Z , H C ( M 3 ) = E 0 β ≅ E γ 1 ≅ E ϵ 1 ≅ Z .
In total there are 23 (strict) E E E -terms which are not all computed. The lattice O ( M ) O(M) O ( M ) has 5 maximal chains and there are 3 different perspective pairs consisting of covering relations, v. Sect. 6.1. The objective of this example is to illustrate that higher order differentials may contain information about connecting orbits in case first order differentials do not.
As in Section 1.2, page-1 of the spectral net is given by:
0 ⟶ E ϵ 1 → 0 E γ 1 → d γ α E α 0 → 0 E 0 α ⟶ 0 , 0 ⟶ E ζ 1 → 0 E γ 1 → d γ β E β 0 → 0 E 0 β ⟶ 0 , E ϵ 1 → d ϵ γ α E α 0 , E ζ 1 → d ζ γ α E α 0 , E δ 1 → d δ β E β 0 , E ϵ 1 → d ϵ γ β E β 0 , E ζ 1 → d ζ β α E α 0 , E δ 1 → d δ β α E α 0 . \begin{aligned}
0&\longrightarrow E_\epsilon^1 \xrightarrow{0} E_\gamma^1 \xrightarrow{d_{\gamma\alpha}} E_\alpha^0 \xrightarrow{0} E_0^\alpha\longrightarrow0,\\
0&\longrightarrow E_\zeta^1 \xrightarrow{0} E_\gamma^1 \xrightarrow{d_{\gamma\beta}} E_\beta^0 \xrightarrow{0} E_0^\beta\longrightarrow0,\\
E_\epsilon^1&\xrightarrow{d_{\epsilon\gamma\alpha}} E_\alpha^0,\quad E_\zeta^1\xrightarrow{d_{\zeta\gamma\alpha}}E_\alpha^0,\quad E_\delta^1\xrightarrow{d_{\delta\beta}}E_\beta^0,\\
E_\epsilon^1&\xrightarrow{d_{\epsilon\gamma\beta}} E_\beta^0,\quad E_\zeta^1\xrightarrow{d_{\zeta\beta\alpha}}E_\alpha^0,\quad E_\delta^1\xrightarrow{d_{\delta\beta\alpha}}E_\alpha^0.
\end{aligned} 0 0 E ϵ 1 E ϵ 1 ⟶ E ϵ 1 0 E γ 1 d γ α E α 0 0 E 0 α ⟶ 0 , ⟶ E ζ 1 0 E γ 1 d γ β E β 0 0 E 0 β ⟶ 0 , d ϵ γ α E α 0 , E ζ 1 d ζ γ α E α 0 , E δ 1 d δ β E β 0 , d ϵ γ β E β 0 , E ζ 1 d ζ β α E α 0 , E δ 1 d δ β α E α 0 .
The dotted arrows are the isomorphisms from excision and by Corollary 6.10 all differentials in the squares with isomorphisms are trivial, v. Figure 8.3 . Moreover, the differentials d 1 ζ δ d_{1\zeta\delta} d 1 ζ δ and d ϵ γ β d_{\epsilon\gamma\beta} d ϵ γ β , and the differentials d ζ γ α d_{\zeta\gamma\alpha} d ζ γ α and d δ β 0 d_{\delta\beta0} d δ β 0 are isomorphic. To illustrate the role of higher order differentials we denote the differential from H C ( M 2 ′ ) \mathrm{HC}(M'_2) HC ( M 2 ′ ) to H C ( M 2 ) \mathrm{HC}(M_2) HC ( M 2 ) by d 22 ′ 1 d^{1}_{22'} d 2 2 ′ 1 . The latter can be computed via d ϵ γ α d_{\epsilon\gamma\alpha} d ϵ γ α . To do so we unroll the exact triangles for M 2 M_2 M 2 and M 2 ′ M'_2 M 2 ′ as long exact sequences using the standard grading of singular homology:
H C 2 ( M 2 ′ ) → j H C 1 ( M 2 ′ ) → k H C 0 ( M 2 ′ ) k ↓ k ↓ k ↓ H 1 ( F γ D 2 ) ← i H 1 ( F ϵ D 2 ) ← i H 0 ( F γ D 2 ) j ↓ j ↓ j ↓ H C 1 ( M 2 ) → k H 0 ( F α D 2 ) → j H C 0 ( M 2 ) \begin{CD}
\mathrm{HC}_2(M'_2) @>{j}>> \mathrm{HC}_1(M'_2) @>{k}>> \mathrm{HC}_0(M'_2) \\
@V{k}VV @V{k}VV @V{k}VV \\
\mathrm{H}_1(F_\gamma D^2) @<{i}<< \mathrm{H}_1(F_\epsilon D^2) @<{i}<< \mathrm{H}_0(F_\gamma D^2) \\
@V{j}VV @V{j}VV @V{j}VV \\
\mathrm{HC}_1(M_2) @>{k}>> \mathrm{H}_0(F_\alpha D^2) @>{j}>> \mathrm{HC}_0(M_2)
\end{CD} HC 2 ( M 2 ′ ) k ↓ ⏐ H 1 ( F γ D 2 ) j ↓ ⏐ HC 1 ( M 2 ) j i k HC 1 ( M 2 ′ ) k ↓ ⏐ H 1 ( F ϵ D 2 ) j ↓ ⏐ H 0 ( F α D 2 ) k i j HC 0 ( M 2 ′ ) k ↓ ⏐ H 0 ( F γ D 2 ) j ↓ ⏐ HC 0 ( M 2 )
and determining all homomorphisms gives:
Z → [ 1 1 ] Z ⊕ Z ← [ 1 − 1 ] Z [ 1 0 ] ↓ ↓ [ 1 0 ] Z → 0 Z ← 0 Z \begin{CD}
\mathbb{Z} @>{\begin{bmatrix}1\\1\end{bmatrix}}>> \mathbb{Z}\oplus\mathbb{Z} @<{\begin{bmatrix}1&-1\end{bmatrix}}<< \mathbb{Z} \\
@V{\begin{bmatrix}1\\0\end{bmatrix}}VV @. @VV{\begin{bmatrix}1\\0\end{bmatrix}}V \\
\mathbb{Z} @>{0}>> \mathbb{Z} @<{0}<< \mathbb{Z}
\end{CD} Z [ 1 0 ] ↓ ⏐ Z [ 1 1 ] 0 Z ⊕ Z Z [ 1 − 1 ] 0 Z ↓ ⏐ [ 1 0 ] Z
The differential from H C ( M 2 ′ ) \mathrm{HC}(M'_2) HC ( M 2 ′ ) to H C ( M 2 ) \mathrm{HC}(M_2) HC ( M 2 ) is given by the composition j k jk j k . Taking the direct sum over the grading, with the summands ordered by decreasing degree, we obtain:
d 22 ′ 1 ∼ d ϵ γ α = [ 0 1 0 0 0 0 0 1 ] ⋅ [ 1 0 1 0 0 1 0 1 ] = [ 1 0 0 1 ] , d 12 ′ 1 = [ 1 0 0 1 ] , d 23 1 = [ 1 0 ] . d^{1}_{22'} \sim d_{\epsilon\gamma\alpha} =
\begin{bmatrix}
0&1&0&0\\
0&0&0&1
\end{bmatrix}
\cdot
\begin{bmatrix}
1&0\\
1&0\\
0&1\\
0&1
\end{bmatrix}
=
\begin{bmatrix}
1&0\\
0&1
\end{bmatrix},
\qquad
d^{1}_{12'}=
\begin{bmatrix}
1&0\\
0&1
\end{bmatrix},
\qquad
d^{1}_{23}=
\begin{bmatrix}
1\\
0
\end{bmatrix}. d 2 2 ′ 1 ∼ d ϵ γ α = [ 0 0 1 0 0 0 0 1 ] ⋅ 1 1 0 0 0 0 1 1 = [ 1 0 0 1 ] , d 1 2 ′ 1 = [ 1 0 0 1 ] , d 23 1 = [ 1 0 ] .
Page-1 of the (fundamental) spectral net can be displayed as follows:
0 → H C ( M 3 ) → 0 H C ( M 2 ′ ) → d 22 ′ 1 H C ( M 2 ) → 0 H C ( M 1 ) → 0 ↓ d 12 ′ 1 ↑ ↓ d 23 1 d 12 ′ 1 ↑ ↓ 0 → H C ( M 2 ′ ) → 0 H C ( M 3 ) → 0 H C ( M 1 ) → 0 H C ( M 2 ) → 0 ↓ d 12 ′ 1 ↓ d 23 1 ↑ H C ( M 1 ) → 0 H C ( M 3 ) (30) \begin{CD}
0 @>>> \mathrm{HC}(M_3) @>{0}>> \mathrm{HC}(M'_2) @>{d^{1}_{22'}}>> \mathrm{HC}(M_2) @>{0}>> \mathrm{HC}(M_1) @>>> 0 \\
@. @VVV @A{d^{1}_{12'}}AA @VV{d^{1}_{23}}V @A{d^{1}_{12'}}AA @VVV @. \\
0 @>>> \mathrm{HC}(M'_2) @>{0}>> \mathrm{HC}(M_3) @>{0}>> \mathrm{HC}(M_1) @>{0}>> \mathrm{HC}(M_2) @>>> 0 \\
@. @VV{d^{1}_{12'}}V @. @VVV @A{d^{1}_{23}}AA @. @. \\
@. \mathrm{HC}(M_1) @>{0}>> \mathrm{HC}(M_3) @. @. \tag*{(30)}
\end{CD} 0 0 HC ( M 3 ) ↓ ⏐ HC ( M 2 ′ ) ↓ ⏐ d 1 2 ′ 1 HC ( M 1 ) 0 0 0 HC ( M 2 ′ ) d 1 2 ′ 1 ⏐ ↑ HC ( M 3 ) HC ( M 3 ) d 2 2 ′ 1 0 HC ( M 2 ) ↓ ⏐ d 23 1 HC ( M 1 ) ↓ ⏐ 0 0 HC ( M 1 ) d 1 2 ′ 1 ⏐ ↑ HC ( M 2 ) d 23 1 ⏐ ↑ 0 ↓ ⏐ 0 (30)
One can follow 5 distinct paths in the above diagram that provide a page-1 of a spectral sequence associated with the 5 maximal chains in O ( M ) \mathrm{O}(M) O ( M ) . These are all possible linear Morse representations in this setting. We will not consider perspective pairs in page-1 in this example since they do not provide additional information about differentials. In principle the first order differentials yield connecting orbits when considering different paths: C ( M 2 ′ , M 1 ) ≠ ∅ C(M_{2'},M_1) \ne\varnothing C ( M 2 ′ , M 1 ) = ∅ , C ( M 2 ′ , M 2 ) ≠ ∅ C(M_{2'},M_2) \ne\varnothing C ( M 2 ′ , M 2 ) = ∅ , and C ( M 3 , M 2 ) ≠ ∅ C(M_3,M_2) \ne\varnothing C ( M 3 , M 2 ) = ∅ . We want to illustrate that higher order differentials can also fulfill this role.
Let us start with the path
2 < 3 < 1 < 2 ′ , 2<3<1<2', 2 < 3 < 1 < 2 ′ ,
which corresponds to the maximal chain: 0 ⊑ β ⊑ δ ⊑ ζ ⊑ 1 0 \sqsubseteq\beta\sqsubseteq\delta\sqsubseteq\zeta\sqsubseteq1 0 ⊑ β ⊑ δ ⊑ ζ ⊑ 1 . From Diagram (8.2) (bottom row) we obtain page-1:
⋯ ⟶ Z ⊕ Z → i d Z ⊕ Z → 0 Z → [ 1 0 ] Z ⊕ Z ⟶ ⋯ \cdots\longrightarrow\mathbb{Z}\oplus\mathbb{Z} \xrightarrow{\mathrm{id}} \mathbb{Z}\oplus\mathbb{Z} \xrightarrow{0} \mathbb{Z} \xrightarrow{\begin{bmatrix}1\\0\end{bmatrix}} \mathbb{Z}\oplus\mathbb{Z} \longrightarrow\cdots ⋯ ⟶ Z ⊕ Z id Z ⊕ Z 0 Z [ 1 0 ] Z ⊕ Z ⟶ ⋯
Computing homology yields page-2:
⋯ → 0 0 → 0 0 → 0 Z ⟶ ⋯ \cdots\xrightarrow{0} 0 \xrightarrow{0} 0 \xrightarrow{0} \mathbb{Z} \longrightarrow\cdots ⋯ 0 0 0 0 0 Z ⟶ ⋯
This maximal chain stabilizes at page-2. The higher order differentials do not play a role for this linear Morse representation. There are two attractor-repeller pairs with M 2 ′ , M 1 M_{2'},M_1 M 2 ′ , M 1 and M 3 , M 2 M_3,M_2 M 3 , M 2 , and since the differentials are non-trivial, connections are found by Corollary 7.4. No additional information about connections may be obtained from this Morse representation. However, the spectral net allows for more paths.
Consider the linear order
2 < 1 < 3 < 2 ′ , 2<1<3<2', 2 < 1 < 3 < 2 ′ ,
which is associated with the maximal chain 0 ⊑ β ⊑ γ ⊑ ζ ⊑ 1 0 \sqsubseteq\beta\sqsubseteq\gamma\sqsubseteq\zeta\sqsubseteq1 0 ⊑ β ⊑ γ ⊑ ζ ⊑ 1 . Diagram (8.2) (middle row) yields the page-1:
⋯ ⟶ Z ⊕ Z → 0 Z → 0 Z ⊕ Z → 0 Z ⊕ Z ⟶ ⋯ \cdots\longrightarrow\mathbb{Z}\oplus\mathbb{Z} \xrightarrow{0} \mathbb{Z} \xrightarrow{0} \mathbb{Z}\oplus\mathbb{Z} \xrightarrow{0} \mathbb{Z}\oplus\mathbb{Z} \longrightarrow\cdots ⋯ ⟶ Z ⊕ Z 0 Z 0 Z ⊕ Z 0 Z ⊕ Z ⟶ ⋯
Computing homology and establishing the non-trivial second order differentials d 12 2 d_{12}^{2} d 12 2 and d 23 2 d_{23}^{2} d 23 2 from the earlier data via j 1 i − 1 k j_{1}^{i-1}k j 1 i − 1 k , yields:
⋯ ⟶ Z ⊕ Z → 0 Z → [ 1 0 ] Z ⊕ Z → 0 Z ⊕ Z ⟶ ⋯ \cdots\longrightarrow\mathbb{Z}\oplus\mathbb{Z} \xrightarrow{0} \mathbb{Z} \xrightarrow{\begin{bmatrix}1\\0\end{bmatrix}} \mathbb{Z}\oplus\mathbb{Z} \xrightarrow{0} \mathbb{Z}\oplus\mathbb{Z} \longrightarrow\cdots ⋯ ⟶ Z ⊕ Z 0 Z [ 1 0 ] Z ⊕ Z 0 Z ⊕ Z ⟶ ⋯
Computing homology reveals that the spectral sequence stabilizes at page-3. Here we have the opposite scenario, where all differentials on page-1 are trivial, and some differentials on page-2 are non-trivial. By Theorem 7.5, since d : S γ ζ 11 → S β γ 0 d:S_{\gamma\zeta}^{11}\to S_{\beta\gamma}^{0} d : S γ ζ 11 → S β γ 0 and d : S β γ 21 → S β γ 0 d:S_{\beta\gamma}^{21}\to S_{\beta\gamma}^{0} d : S β γ 21 → S β γ 0 are not trivial, we obtain connections. For the former from M 2 ′ M_{2'} M 2 ′ to either M 1 , M 3 ∈ ζ \ β M_1,M_3\in\zeta\backslash\beta M 1 , M 3 ∈ ζ \ β and reciprocally from M 2 ′ , M 3 ∈ 1 \ γ M_{2'},M_3\in1\backslash\gamma M 2 ′ , M 3 ∈ 1\ γ to M 1 M_1 M 1 . Since M 2 ∥ M 3 M_2\mathbin{\|}M_3 M 2 ∥ M 3 , we conclude that C ( M 2 ′ , M 1 ) ≠ ∅ C(M_{2'},M_1)\ne\varnothing C ( M 2 ′ , M 1 ) = ∅ , and since M 1 ∥ M 3 M_1\mathbin{\|}M_3 M 1 ∥ M 3 , we conclude again that C ( M 2 ′ , M 1 ) ≠ ∅ C(M_{2'},M_1)\ne\varnothing C ( M 2 ′ , M 1 ) = ∅ . With analogous reasoning we obtain from the second non-trivial differential that C ( M 3 , M 2 ) ≠ ∅ C(M_3,M_2)\ne\varnothing C ( M 3 , M 2 ) = ∅ .
Finally, for the top row in Diagram (8.2) we have the linear order
1 < 2 < 2 ′ < 3 , 1<2<2'<3, 1 < 2 < 2 ′ < 3 ,
which corresponds to the maximal chain 0 ⊑ α ⊑ γ ⊑ ϵ ⊑ 1 0 \sqsubseteq\alpha\sqsubseteq\gamma\sqsubseteq\epsilon\sqsubseteq1 0 ⊑ α ⊑ γ ⊑ ϵ ⊑ 1 . This gives the page-1 spectral sequence
⋯ ⟶ Z → 0 Z ⊕ Z → i d Z ⊕ Z → 0 Z ⊕ Z ⟶ ⋯ \cdots\longrightarrow\mathbb{Z} \xrightarrow{0} \mathbb{Z}\oplus\mathbb{Z} \xrightarrow{\mathrm{id}} \mathbb{Z}\oplus\mathbb{Z} \xrightarrow{0} \mathbb{Z}\oplus\mathbb{Z} \longrightarrow\cdots ⋯ ⟶ Z 0 Z ⊕ Z id Z ⊕ Z 0 Z ⊕ Z ⟶ ⋯
The first order differential d 22 1 d^{1}_{22} d 22 1 yields a connection from M 2 ′ M_{2}^{\prime} M 2 ′ to M 2 M_{2} M 2 . Computing homology gives page-2, where all differentials are trivial:
⋯ Z → 0 0 → 0 Z ⊕ Z → 0 ⋯ \cdots\quad\mathbb{Z} \xrightarrow{0} 0 \xrightarrow{0} \mathbb{Z}\oplus\mathbb{Z} \xrightarrow{0} \cdots ⋯ Z 0 0 0 Z ⊕ Z 0 ⋯
Page-3 has a non-trivial differential since the spectral sequence has to converge at page-4. The differential d 13 3 d^{3}_{13} d 13 3 need not be computed; it is non-trivial for the above reason.
⋯ Z → 0 0 → [ 1 0 ] Z ⊕ Z → 0 ⋯ \cdots\quad\mathbb{Z} \xrightarrow{0} 0 \xrightarrow{\left[\begin{smallmatrix}1\\0\end{smallmatrix}\right]} \mathbb{Z}\oplus\mathbb{Z} \xrightarrow{0} \cdots ⋯ Z 0 0 [ 1 0 ] Z ⊕ Z 0 ⋯
The differential
d : S α ϵ 11 → S 00 α ϵ d:S^{11}_{\alpha\epsilon}\to S^{\alpha\epsilon}_{00} d : S α ϵ 11 → S 00 α ϵ
is non-trivial. From Theorem 7.3 we conclude that the following sets C ( M 2 ′ , M 1 ) C(M_{2}^{\prime},M_{1}) C ( M 2 ′ , M 1 ) , C ( M 2 ′ , M 2 ) C(M_{2}^{\prime},M_{2}) C ( M 2 ′ , M 2 ) , and C ( M 3 , M 2 ) C(M_{3},M_{2}) C ( M 3 , M 2 ) are non-empty. Indeed, the sequence in Theorem 7.3 goes from M 3 M_{3} M 3 to M 1 M_{1} M 1 through comparable consecutive elements, and because M 1 ∥ M 2 M_{1}\mathbin{\|}M_{2} M 1 ∥ M 2 , M 1 ∥ M 3 M_{1}\mathbin{\|}M_{3} M 1 ∥ M 3 and M 2 ′ ∥ M 3 M_{2}^{\prime}\mathbin{\|}M_{3} M 2 ′ ∥ M 3 , it passes through the pairs { M 1 , M 2 ′ } \{M_{1},M_{2}^{\prime}\} { M 1 , M 2 ′ } , { M 2 ′ , M 2 } \{M_{2}^{\prime},M_{2}\} { M 2 ′ , M 2 } and { M 2 , M 3 } \{M_{2},M_{3}\} { M 2 , M 3 } . Theorem 7.5 gives two of these connections directly. There is a connection from M 3 M_{3} M 3 to one of M 1 M_{1} M 1 , M 2 M_{2} M 2 , M 2 ′ ∈ ϵ M_{2}^{\prime}\in\epsilon M 2 ′ ∈ ϵ , and reciprocally, from one of M 3 M_{3} M 3 , M 2 ′ M_{2}^{\prime} M 2 ′ , M 2 ∈ 1 ∖ α M_{2}\in1\setminus\alpha M 2 ∈ 1 ∖ α to M 1 M_{1} M 1 . Given that M 3 ∥ M 2 ′ M_{3}\mathbin{\|}M_{2}^{\prime} M 3 ∥ M 2 ′ , M 3 ∥ M 1 M_{3}\mathbin{\|}M_{1} M 3 ∥ M 1 and M 2 ∥ M 1 M_{2}\mathbin{\|}M_{1} M 2 ∥ M 1 , that implies that C ( M 3 , M 2 ) ≠ ∅ C(M_{3},M_{2})\ne\varnothing C ( M 3 , M 2 ) = ∅ and C ( M 2 ′ , M 1 ) ≠ ∅ C(M_{2}^{\prime},M_{1})\ne\varnothing C ( M 2 ′ , M 1 ) = ∅ . With the remaining two linear orders we obtain the same results. The above results have shown that ( M , ≤ ) (M,\leq) ( M , ≤ ) as given in Figure 8.3 [middle] is a flow-order.
Example 8.3 In [11 ] the following system of differential equations is considered:
x ˙ = y , y ˙ = − θ y − x ( x − 1 / 3 ) ( 1 − x ) , θ ˙ = ε ( θ ′ − θ ) ( θ ′ ′ − θ ) , \dot{x}=y,\qquad\dot{y}=-\theta y-x(x-1/3)(1-x),\qquad\dot{\theta}=\varepsilon(\theta^{\prime}-\theta)(\theta^{\prime\prime}-\theta), x ˙ = y , y ˙ = − θ y − x ( x − 1/3 ) ( 1 − x ) , θ ˙ = ε ( θ ′ − θ ) ( θ ′′ − θ ) ,
with x , y , θ ∈ R x,y,\theta\in\mathbb{R} x , y , θ ∈ R and ε ≪ 1 \varepsilon\ll1 ε ≪ 1 , and 0 < θ ′ ≪ 1 0<\theta^{\prime}\ll1 0 < θ ′ ≪ 1 , θ ′ ′ ≫ 1 \theta^{\prime\prime}\gg1 θ ′′ ≫ 1 . The detailed description can be found in [11 Ex. 6.2] , v. Figure 8.5 . At θ = θ ′ \theta=\theta^{\prime} θ = θ ′ and at θ = θ ′ ′ \theta=\theta^{\prime\prime} θ = θ ′′ the system is known. The Conley indices at the six rest points are isomorphic to Z \mathbb{Z} Z . We consider Morse representations ( M , ≤ ) (M,\leq) ( M , ≤ ) , where the Morse sets are the six fixed points as indicated in Figure 8.5 , and we describe M M M via a poset ( P , ≤ ) (P,\leq) ( P , ≤ ) . The orbits given in Figure 8.5 are known to exist which force order relations on ( M , ≤ ) ≅ ( P , ≤ ) (M,\leq)\cong(P,\leq) ( M , ≤ ) ≅ ( P , ≤ ) . From this point on we will describe M M M via P P P . From the given orbits the following sets are down-sets in P P P , and form a perspective pair in O ( P ) O(P) O ( P ) :
Figure 8.5. Some flow-lines for the dynamical system given above. The spectral methods imply that d 26 1 ≠ 0 d^{1}_{26}\ne0 d 26 1 = 0 , which implies that a connecting orbit from p = 6 p=6 p = 6 to p = 2 p=2 p = 2 exists. We denote this with the dotted arrow.
α = { 1 } , β = { 1 , 2 } , γ = { 1 , 4 } , δ = { 1 , 2 , 4 } . \alpha=\{1\},\qquad\beta=\{1,2\},\qquad\gamma=\{1,4\},\qquad\delta=\{1,2,4\}. α = { 1 } , β = { 1 , 2 } , γ = { 1 , 4 } , δ = { 1 , 2 , 4 } .
We do not know whether a connection between p = 2 p=2 p = 2 and p = 6 p=6 p = 6 exists. As tail elements we take α − = ∅ \alpha_{-}=\varnothing α − = ∅ and δ + = { 1 , 2 , 4 , 6 } \delta_{+}=\{1,2,4,6\} δ + = { 1 , 2 , 4 , 6 } . The order relations and the two possible sublattices are shown in Figure 8.4 .
Figure 8.4. The order relations in P between the elements 1, 2, 4 and 6 [left]. The elements 2 and 6 may, or may not be ordered. The two possible sublattices of O(P), the first one supposing 2 ≤ 6 2 \le6 2 ≤ 6 [middle] and the second one for 2 ∥ 6 2\mathrel{\|}6 2 ∥ 6 [right]. For both scenarios we have the square in the diagram indicating a perspective pair.
The Conley indices (with R = Z R=\mathbb{Z} R = Z ) are used to define G p C ≅ Z G_{p}C\cong\mathbb{Z} G p C ≅ Z for all p ∈ P p\in P p ∈ P . For the differentials between E E E -terms we have:
d β α α − = d 12 1 : G 2 C → G 1 C , ℓ − 1 d δ + δ γ = d 26 1 : G 6 C → G 2 C ; d_{\beta\alpha\alpha_{-}}=d^{1}_{12}:G_{2}C\to G_{1}C,\qquad\ell^{-1}d_{\delta_{+}\delta\gamma}=d^{1}_{26}:G_{6}C\to G_{2}C; d β α α − = d 12 1 : G 2 C → G 1 C , ℓ − 1 d δ + δ γ = d 26 1 : G 6 C → G 2 C ;
d γ α α − = d 14 1 : G 4 C → G 1 C , ℓ ′ − 1 d δ + δ β = d 46 1 : G 6 C → G 4 C . d_{\gamma\alpha\alpha_{-}}=d^{1}_{14}:G_{4}C\to G_{1}C,\qquad\ell^{\prime-1}d_{\delta_{+}\delta\beta}=d^{1}_{46}:G_{6}C\to G_{4}C. d γ α α − = d 14 1 : G 4 C → G 1 C , ℓ ′ − 1 d δ + δ β = d 46 1 : G 6 C → G 4 C .
By Theorem 6.1 we conclude that
d 12 1 d 26 1 + d 14 1 d 46 1 = 0. d^{1}_{12}d^{1}_{26}+d^{1}_{14}d^{1}_{46}=0. d 12 1 d 26 1 + d 14 1 d 46 1 = 0.
A direct analysis of the system shows that E α − β = E α − γ = E β δ + = 0 E^{\beta}_{\alpha_-}=E^{\gamma}_{\alpha_-}=E^{\delta_+}_{\beta}=0 E α − β = E α − γ = E β δ + = 0 , and the exact sequences of the triples α − ⊆ α ⊆ β \alpha_- \subseteq\alpha\subseteq\beta α − ⊆ α ⊆ β , α − ⊆ α ⊆ γ \alpha_- \subseteq\alpha\subseteq\gamma α − ⊆ α ⊆ γ and β ⊆ δ ⊆ δ + \beta\subseteq\delta\subseteq\delta_+ β ⊆ δ ⊆ δ + show that d β α α − d_{\beta\alpha\alpha_-} d β α α − , d γ α α − d_{\gamma\alpha\alpha_-} d γ α α − and d δ + δ β d_{\delta_+\delta\beta} d δ + δ β are isomorphisms, i.e. d 12 1 d^{1}_{12} d 12 1 , d 14 1 d^{1}_{14} d 14 1 and d 46 1 d^{1}_{46} d 46 1 are isomorphisms which implies that d 26 1 ≠ 0 d^{1}_{26}\ne0 d 26 1 = 0 . Since d δ + δ γ = ℓ d 26 1 ≠ 0 d_{\delta_+\delta\gamma}=\ell\,d^{1}_{26}\ne0 d δ + δ γ = ℓ d 26 1 = 0 , Corollary 7.4, applied to γ ⊆ δ ⊆ δ + \gamma\subseteq\delta\subseteq\delta_+ γ ⊆ δ ⊆ δ + with δ ∖ γ = { 2 } \delta\setminus\gamma=\{2\} δ ∖ γ = { 2 } and δ + ∖ δ = { 6 } \delta_+\setminus\delta=\{6\} δ + ∖ δ = { 6 } , implies that there exists a connecting orbit from M 6 M_6 M 6 to M 2 M_2 M 2 . Therefore, 2 < 6 2<6 2 < 6 and hence the sublattice in Figure 8.4 [middle] is the correct sublattice. In [11 Ex. 6.2] this fact is obtained
via a connection matrix argument. With this example we show that the same conclusion can be obtained using spectral nets.
Example 8.4 [11 ] discusses the case of a flow on the Möbius strip S ≅ I 2 / ∼ S\cong I^2/\sim S ≅ I 2 / ∼ . We now show how derived differentials recognize connecting orbits in the case we use integral homology. Consider a flow φ \varphi φ on S S S with an attractor-repeller pair decomposition as displayed in Figure 8.6 . The attractor-repeller pair decomposition yields an E E E -system with poset P = { 1 , 2 } \mathrm{P}=\{1,2\} P = { 1 , 2 } , 1 < 2 1<2 1 < 2 and M 1 = A M_1=A M 1 = A (blue) and M 2 = A ∗ M_2=A^* M 2 = A ∗ (red). The lattice of down-sets is given by O ( P ) = { 0 , α , 1 } \mathop{\mathrm{O}}(\mathrm{P})=\{0,\alpha,1\} O ( P ) = { 0 , α , 1 } , where 0 = ∅ 0=\varnothing 0 = ∅ , α = { 1 } \alpha=\{1\} α = { 1 } and 1 = { 1 , 2 } 1=\{1,2\} 1 = { 1 , 2 } . The E E E -terms are given by the Conley indices: E 0 0 = H C ( A ) ≅ H ( S 1 ) E^0_0=\mathrm{HC}(A)\cong H(S^1) E 0 0 = HC ( A ) ≅ H ( S 1 ) , E α 1 = H C ( A ∗ ) ≅ H ( S , ∂ S ) E^1_\alpha=\mathrm{HC}(A^*)\cong H(S,\partial S) E α 1 = HC ( A ∗ ) ≅ H ( S , ∂ S ) and E 0 1 = H C ( S ) ≅ H ( S 1 ) E^1_0=\mathrm{HC}(S)\cong H(S^1) E 0 1 = HC ( S ) ≅ H ( S 1 ) .
Figure 8.6. The Möbius strip S S S with the middle periodic orbit repelling (red) and the boundary is an attracting periodic orbit (blue).
The exact triangle
E 0 α → i E 0 1 k ↑ ↓ j E α 1 \begin{CD}
E_{0}^{\alpha} @>i>> E_{0}^{1} \\
@A{k}AA @VV{j}V \\
E_{\alpha}^{1}
\end{CD} E 0 α k ⏐ ↑ E α 1 i E 0 1 ↓ ⏐ j
may be expressed as long exact sequence of graded Conley indices:
0 ⟶ H C 2 ( A ) ⟶ H C 2 ( S ) ⟶ H C 2 ( A ∗ ) ⟶ H C 1 ( A ) ⟶ H C 1 ( S ) ⟶ H C 1 ( A ∗ ) ⟶ H C 0 ( A ) ⟶ H C 0 ( S ) ⟶ H C 0 ( A ∗ ) ⟶ 0 0 \longrightarrow\mathrm{HC}_{2}(A) \longrightarrow\mathrm{HC}_{2}(S) \longrightarrow\mathrm{HC}_{2}(A^{*})
\longrightarrow\mathrm{HC}_{1}(A) \longrightarrow\mathrm{HC}_{1}(S) \longrightarrow\mathrm{HC}_{1}(A^{*})
\longrightarrow\mathrm{HC}_{0}(A) \longrightarrow\mathrm{HC}_{0}(S) \longrightarrow\mathrm{HC}_{0}(A^{*}) \longrightarrow0 0 ⟶ HC 2 ( A ) ⟶ HC 2 ( S ) ⟶ HC 2 ( A ∗ ) ⟶ HC 1 ( A ) ⟶ HC 1 ( S ) ⟶ HC 1 ( A ∗ ) ⟶ HC 0 ( A ) ⟶ HC 0 ( S ) ⟶ HC 0 ( A ∗ ) ⟶ 0
Let us start with R = Z R = \mathbb{Z} R = Z -coefficients. The Conley indices are given by:
Table 2.
If we implement these modules in the long exact sequence we obtain:
0 ⟶ 0 ⟶ 0 ⟶ 0 0 \longrightarrow0 \longrightarrow0 \longrightarrow0 0 ⟶ 0 ⟶ 0 ⟶ 0
Z ⟶ Z ⟶ Z 2 \mathbb{Z} \longrightarrow\mathbb{Z} \longrightarrow\mathbb{Z}_{2} Z ⟶ Z ⟶ Z 2
Z ⟶ Z ⟶ 0 ⟶ 0 \mathbb{Z} \longrightarrow\mathbb{Z} \longrightarrow0 \longrightarrow0 Z ⟶ Z ⟶ 0 ⟶ 0
The connecting homomorphism k 1 = 0 k_{1}=0 k 1 = 0 since Hom ( Z 2 , Z ) = 0 \operatorname{Hom}(\mathbb{Z}_{2},\mathbb{Z})=0 Hom ( Z 2 , Z ) = 0 , and k 2 = 0 k_{2}=0 k 2 = 0 since H C 2 ( A ∗ ) = 0 \mathrm{HC}_{2}(A^{*})=0 HC 2 ( A ∗ ) = 0 , and therefore k = 0 k=0 k = 0 . The above exact triangle then becomes a short exact sequence
0 ⟶ E 0 α → i E 0 1 → j E α 1 ⟶ 0 0 \longrightarrow E_{0}^{\alpha} \xrightarrow{i} E_{0}^{1} \xrightarrow{j} E_{\alpha}^{1} \longrightarrow0 0 ⟶ E 0 α i E 0 1 j E α 1 ⟶ 0
which is non-split. By Lemma 7.8 this implies that C ( A ∗ , A ) ≠ ∅ C(A^{*}, A) \ne\varnothing C ( A ∗ , A ) = ∅ . The latter is based on the algebraic interpretation of Morse representations. However, purely in terms of algebra of spectral systems this is manifested in terms of the derived differential κ : E α 1 → E 0 α [ 1 ] \kappa\colon E_{\alpha}^{1} \to E_{0}^{\alpha}[1] κ : E α 1 → E 0 α [ 1 ] . To compute κ \kappa κ we use a projective resolution for $\mathbb{Z}_{2} given by ⋯ → 0 → Z → d 1 / × 2 Z → ϵ / mod 2 Z 2 → 0 \cdots\to0 \to\mathbb{Z} \xrightarrow{d_1\,/\,\times2} \mathbb{Z} \xrightarrow{\epsilon\,/\,\operatorname{mod} 2} \mathbb{Z}_2 \to0 ⋯ → 0 → Z d 1 / × 2 Z ϵ / mod 2 Z 2 → 0 . In the diagram below we choose f 0 = id f_0=\operatorname{id} f 0 = id and consequently f 1 = id f_1=\operatorname{id} f 1 = id :
⋯ ⟶ 0 → 0 Z → × 2 Z → mod 2 Z 2 ⟶ 0 ↓ 0 ⋯ ⋯ ⋯ ↓ f 1 ⋯ ⋯ ⋯ ↓ f 0 ∥ ⋯ ⟶ 0 ⟶ Z → i Z → j Z 2 ⟶ 0 \begin{array}{ccccccccc}
\cdots& \longrightarrow& 0 & \xrightarrow{0} & \mathbb{Z} & \xrightarrow{\times2} & \mathbb{Z} & \xrightarrow{\operatorname{mod} 2} & \mathbb{Z}_2 \longrightarrow0 \\
&& \downarrow0 && \mathrel{\cdots\cdots\cdots}\downarrow f_1 && \mathrel{\cdots\cdots\cdots}\downarrow f_0 && \Vert\\
\cdots& \longrightarrow& 0 & \longrightarrow& \mathbb{Z} & \xrightarrow{i} & \mathbb{Z} & \xrightarrow{j} & \mathbb{Z}_2 \longrightarrow0
\end{array} ⋯ ⋯ ⟶ ⟶ 0 ↓ 0 0 0 ⟶ Z ⋯⋯⋯ ↓ f 1 Z × 2 i Z ⋯⋯⋯ ↓ f 0 Z mod 2 j Z 2 ⟶ 0 ∥ Z 2 ⟶ 0
The diagram for P ∙ → Z [ 1 ] P_\bullet\to\mathbb{Z}[1] P ∙ → Z [ 1 ] is given by
⋯ ⟶ 0 → 0 Z → × 2 Z ↓ 0 ↓ id ↓ 0 ⋯ ⟶ 0 ⟶ Z ⟶ 0 \begin{array}{ccccccc}
\cdots& \longrightarrow& 0 & \xrightarrow{0} & \mathbb{Z} & \xrightarrow{\times2} & \mathbb{Z} \\
&& \downarrow0 && \downarrow\operatorname{id} && \downarrow0 \\
\cdots& \longrightarrow& 0 & \longrightarrow& \mathbb{Z} & \longrightarrow& 0
\end{array} ⋯ ⋯ ⟶ ⟶ 0 ↓ 0 0 0 ⟶ Z ↓ id Z × 2 ⟶ Z ↓ 0 0
which is a non-trivial representative of κ \kappa κ in Hom K h ( Mod Z ) ( P ∙ , Z [ 1 ] ) \operatorname{Hom}_{\mathrm{Kh}(\operatorname{Mod}_{\mathbb{Z}})}(P_\bullet,\mathbb{Z}[1]) Hom Kh ( Mod Z ) ( P ∙ , Z [ 1 ]) : a null-homotopy would be a homomorphism h : Z → Z h:\mathbb{Z}\to\mathbb{Z} h : Z → Z with h ∘ ( × 2 ) = id h\circ(\times2)=\operatorname{id} h ∘ ( × 2 ) = id , which does not exist. This shows that the derived differential is non-trivial. In [11 ] the equivalent of the derived differential κ \kappa κ is a non-trivial entry in the C C C -connection matrix. This example illustrates that spectral systems also record the connecting orbits algebraically.
The same calculations can be carried out in different coefficient rings providing different conclusions. For example if we use the field coefficients in R = Z 2 R=\mathbb{Z}_2 R = Z 2 we obtain
Table 1.
In this case the derived differential is necessarily trivial since all modules are free. One concludes in this case that k ≠ 0 k\ne0 k = 0 , which by Theorem 7.10 algebraically records the existence of a connecting orbit. If we use R = Z 4 R=\mathbb{Z}_4 R = Z 4 the theory in [11 ] is not applicable since Z 4 \mathbb{Z}_4 Z 4 is not a PID. However, the analysis here still provides a non-trivial connecting homomorphism, since k 2 : H C 2 ( A ∗ ; Z 4 ) ≅ Z 2 → H C 1 ( A ; Z 4 ) ≅ Z 4 k_2:\mathrm{HC}_2(A^*;\mathbb{Z}_4)\cong\mathbb{Z}_2\to\mathrm{HC}_1(A;\mathbb{Z}_4)\cong\mathbb{Z}_4 k 2 : HC 2 ( A ∗ ; Z 4 ) ≅ Z 2 → HC 1 ( A ; Z 4 ) ≅ Z 4 is injective, as well as a non-trivial derived differential, and therefore an algebraic manifestation of a connecting orbit. If we consider R = R R=\mathbb{R} R = R , or for example R = Z 3 R=\mathbb{Z}_3 R = Z 3 the Conley index H C ( A ∗ ) \mathrm{HC}(A^*) HC ( A ∗ ) is trivial and the connecting orbit cannot be detected algebraically. The derived differential may also contain information about the connecting orbit in the Morse representation besides its existence. This is why considering different coefficient rings may be useful.
Disclosure of AI use
This manuscript was written without using generative AI. Before submission of the article the content was carefully checked by the authors using Claude Opus 5.5 Extra. After using this tool, the authors have reviewed the suggested edits to the content. The authors take full responsibility for the content of the publication.
Acknowledgements The authors wish to thank Thomas Rot for pointing out the subject of spectral systems and for many inspiring discussions on the subject.