Asymmetrically Weighted Dowker Persistence and Applications in Dynamical Systems

Let \delta_w be defined as above, and \delta'>\delta_w. Naturally, \mathfrak{D}^\star_{\delta_w}(G)\subseteq \mathfrak{D}^\star_{\delta'}(G). We show the other inclusion. Let \sigma\subseteq X be a simplex of \mathfrak{D}^\star_{\delta'}(G). It follows that there exists y\in X such that (y,x)\in R_{\delta'}(G) for all x\in \sigma. Thus \omega(y,x)\leq \delta' for all x\in \sigma. It follows that \omega(y,x)\leq \delta' for all x\in \sigma. Thus \omega(y,x)< \infty for all x\in \sigma, implying that \omega(y,x)\leq \delta_w for all x\in \sigma. Hence \sigma is a simplex of \mathfrak{D}^\star_{\delta_w}(G). This yields the desired inclusion.

Let \sigma be a simplex of \mathfrak{D}^\star_{\delta_\text{max}}(G) and K be a source dominating set of G. Since \sigma is a simplex of \mathfrak{D}^\star_{\delta_\text{max}}(G), it follows that there exists vertex h such that \omega(h,x)<\delta_{\text{max}} for all x\in \sigma. However, since h\in X it follows that either an element of the dominating set, k\in K, sees it with an edge of finite weight, or h=k\in K. In either case it follows that all edges from k to x\in \sigma also have finite weight. Thus \delta^* = \max_{x\in \sigma} \omega(k,x) is finite and (k,x) \in R_{\delta^*}(G) for all x\in \sigma.

Since the edge set of H is a subset of the edge set of G, it follows that for all \delta, any simplex of \mathfrak{D}^\star_\delta(H) must also be contained in \mathfrak{D}^\star_\delta(G). Thus \mathfrak{D}^\star_{\delta_\text{max}}(H)\subseteq \mathfrak{D}^\star_{\delta_\text{max}}(G)

Let \sigma\in \mathfrak{D}^\star_{\delta_\text{max}}(G). Then by Proposition 3 it follows that \sigma\in\mathfrak{D}^\star_{\delta}(H) for some \delta value. Thus \sigma\in \mathfrak{D}^\star_{\delta_\text{max}}(H). Hence \mathfrak{D}^\star_{\delta_\text{max}}(H)\supseteq \mathfrak{D}^\star_{\delta_\text{max}}(G).

We note that \delta_r as defined is also the edge weight bound of the the subgraph induced by K, which we denote by H.

Let \delta_e be the edge weight bound of G. The inequality \delta_r\leq \delta_e follows from the fact that the finite weight edges of H are a subset of those in G.

Let \sigma\in \mathfrak{D}^\star_{\delta_\text{max}}(G). By Proposition 3 there exists v\in K and \delta such that (v,x)\in R_\delta for all x\in \sigma. Since \delta_r is at least a big as \max_{x\in \sigma} \omega'(v,x) by construction, it follows \delta\leq \delta_r. Thus \sigma\in \mathfrak{D}^\star_{\delta_r}(G). It follows that \mathfrak{D}^\star_{\delta_\text{max}}(G) \subseteq \mathfrak{D}^\star_{\delta_r}(G). Thus \delta_r\geq \delta_{\text{max}}.

Let K=\{v_1,\dotsc, v_k\} be a minimal dominating set of G, with indexing set I=\{1,\dotsc, k\}. By Corollary 4 it follows \mathfrak{D}^\star_{\delta_\text{max}}(G) = \text{Cl}\left([N(v_1)] \cup \dotsc \cup [N(v_k)]\right). Let N be the nerve of \{|N(v_1)|, \dotsc , |N(v_k)|\}, that is the set of all J\subseteq I such that \cap_{i\in J}|N(v_i)| \neq \emptyset, where |\cdot| denotes the geometric realization of the simplex. Notice that N is a simplicial complex because it is closed under containment. By the nerve theorem, if all J\subseteq I have \cap_{i\in J}|N(v_i)| as empty or contractible, then N is homotopy equivalent to the union \mathfrak{D}^\star_{\delta_\text{max}}(G) = [N(v_1)] \cup \dotsc \cup [N(v_k)]. By the construction of the maximal Dowker complex, we know that any intersection of neighborhoods is entirely seen by some vertex (in its neighborhood) and thus gets fully connected. Thus the corresponding intersection of simplicial complexes is contractible and we are done.

By Proposition 7, we have that \mathfrak{D}^\star_{\delta_\text{max}}(G) is homotopic to the nerve of \{|N(v_1)|, \dotsc , |N(v_k)|\} which we will call N. This is a simplicial complex on a base set of k elements. Since an n simplex has n+1 vertices, it follows that N with k vertices cannot contain the boundary of a simplex of dimension k or higher. Thus N must have trivial homology for dimension k-1 and greater. Since N is homotopically equivalent to \mathfrak{D}^\star_{\delta_\text{max}}(G) it follows that \mathfrak{D}^\star_{\delta_\text{max}}(G) shares these trivial homology classes.

Let \alpha=\sum_i a_i\alpha_i. For some j, suppose that \alpha_j=[v,w] is such that \omega_G(v,w) > \delta and \omega_G(w,v) > \delta hold. By definition, both bounds hold in P(G) as well. It follows by construction that there exists a vertex x such that \omega_{P(G)}(x,v)\leq \delta and \omega_{P(G)}(x,w)\leq \delta. This implies the existence of the 2-chain [x,v,w] in the chain complex. Thus the first chain group contains

The addition of the boundary removes the term a_j\alpha_j from \alpha, instead leaving -a_j[x,w] and a_j[x,v] which correspond with edges of weight less than \delta in P(G).

Because it differs by a boundary, this new chain is homologous to \alpha. If we repeat this process for all elements of \alpha, we find that \alpha is homologous to \zeta=\sum_i z_i \zeta_i where \zeta_i=[x_i,v_i] implies \omega_{P(G)}(x_i,v_i)\leq \delta or \omega_{P(G)}(v_i,x_i)\leq \delta.

Now we find \eta, with edges corresponding to edges in G, that \zeta is homologous to. For some p, consider \zeta_p=[x_p,v_p]. As we saw previously this implies \omega_{P(G)}(x_p,v_p)\leq \delta or \omega_{P(G)}(v_p,x_p)\leq \delta. Without loss of generality, we assume the former. Since the edge weights of P(G) come from directed paths in G it follows that there exist vertices y_2,\dotsc,y_{m-1} such that with y_1=x_p and y_m=v_p we have

Since we assume edge weights are positive, it follows that \omega(y_j,y_{j+1})\leq \delta for all 1\leq j<m. It also follows that \omega_{P(G)}(y_1,y_j)\leq \delta for all 1\leq j<m and therefore the complex contains [x_i,y_1,\dotsc,y_m,v_i].

It follows that the first chain group contains

Thus

this replaces the term z_p\zeta_p in \zeta with a sum of 1 simplices corresponding to edges with weights less than \delta in G. Since the difference is again a boundary, this new chain is homologous to \zeta. If we again repeat this process for all the elements of \zeta, we find a new 1-chain \eta with elements \eta_j=[x_p,x_q] such that \omega_G(x_p,x_q)\leq \delta or \omega_G(x_q,x_p)\leq \delta. Additionally, the homology class [\alpha]=[\zeta]=[\eta].

Let the \Delta = [x_1,\dotsc,x_k] be the simplex in question, with labels reflecting the order of V(G). This simplex is included in the Dowker complex if and only if there exists a v\in V(G) such that \omega(v,x_i)<\delta for all i.

Suppose v is not a vertex of \Delta and by symmetry let v=0 without loss of generality. Notice that in a consistently oriented cycle graph, there exists a unique directed path between any two vertices. Further, given vertices with x_1<x_2<x_3 in the vertex order, x_2 lies on the unique directed path between x_1 and x_3, respecting that vertex order. Thus \omega(x_1,x_3)=\omega(x_1,x_2) + \omega(x_2,x_3). Therefore, since \omega(v=0,x_1)<\delta and \delta>\omega(v,x_i)=\omega(v,x_1)+\omega(x_1,x_i) for all i>1, it follows \omega(x_1,x_i)<\delta for i\geq 1.

Note that this same argument implies \omega(x_2,x_i)<\delta for i\geq2, and so on for all the vertices, implying the existence of sufficiently weighted edges for all pairs of vertices in \Delta.

We show the equivalence by proving that [v_1,\dotsc,v_m]\in A if and only if [\phi(v_1),\dotsc,\phi(v_m)]\in Nerv\left(\{U_1, \dotsc, U_{n-1},{U}_n\}\right) \cup \text{Cl}(\Delta_b).

( \Longrightarrow ) Let [v_1,\dotsc,v_m]\in A. Assume this is ordered so that i<j\implies\omega_{P(G)}(v_i,v_j)\leq \delta (we can do this by Lemma 11). We consider two cases.

(Case 1) Assume that v_m\neq n. From the claim \omega_{P(G)}(v_j,v_m)\leq \delta for all 1\leq j<m. Since v_m\in V(G') it follows \omega_{P(G')}(v_j,v_m)\leq \delta for all 1\leq j<m. Thus v_m\in\cap_{v_j\neq n} \phi(v_j). But if there exists 1\leq j<m such that v_j=n, the fact that \omega_{P(G)}(v_j,v_m)\leq \delta implies v_j\in U_n too. Thus v_m\in\cap_{j=1}^m \phi(v_j). Therefore [\phi(v_1),\dotsc,\phi(v_m)]\in\text{Nerv}\left(\{U_1,\dotsc,U_n\}\right).

(Case 2) Assume v_m=n. From the claim \omega_{P(G)}(v_j,v_m=n)\leq \delta for all 1\leq j<m. By construction \phi(v_j)\in \Delta_b for all 1\leq j< m. Thus [\phi(v_1),\dotsc,\phi(v_m=n)]\in \text{Cl}(\Delta_b).

( \Longleftarrow ) Let [\phi(v_1),\dotsc,\phi(v_m)]\in Nerv\left(\{U_1, \dotsc, U_{n-1},{U'}_n\}\right) \cup \text{Cl}(\Delta_b). By definition \cap_{i=1}^m \phi(v_i)\neq \emptyset or [\phi(v_1),\dotsc,\phi(v_m)]\in Cl(\Delta_b).

In the first case there exists v_p in this intersection such that \omega_{P(G)}(v_i,v_p)\leq \delta for 1\leq i\leq m. This works for v_i= n by our definition of U_n. Thus there exists v_q with largest edge weight to v_p, \omega_{P(G)}(v_q,v_p), which in turn implies that it sees all the other vertices in question and we have [v_1,\dotsc,v_p]\in A.

In the second case, by the definition of \Delta_b, for all 1\leq i\leq m the edge weight \omega_{P(G)}(v_i,v_m=n)\leq \delta. Thus, as before there exists v_p with largest edge weight \omega_{P(G)}(v_p,n) which can see all the other vertices in question and we have [v_1,\dotsc,v_p]\in A.

Consider the nerve of the neighborhood cover of A' augmented with an nth element, N=Nerv\left(\{U_1,\dotsc,U_n\equiv\text{Cl}([\{j\neq n | \omega_{P(G)}(n,j)\leq \delta\}\}])\right), and let B=\text{Cl}(\Delta_b). Since G and G' are cycle graphs, Proposition 10 has shown that each corresponding complex can have at most one nontrivial first homology class. Further, by Proposition 12 we have A\approx N\cup B, and by the nerve theorem we have homotopy equivalence between N and A'. Thus it is sufficient to show that N has nontrivial first homology if and only if N\cup B has nontrivial first homology.

Suppose \delta<\max_i \omega_G(i,i+1\pmod{n}). Then \delta<\max_i \omega_{G'}(i,i+1\pmod{n}). By the contrapositive of Proposition 10 it follows that neither A nor A' may have nontrivial first homology. Thus we assume \delta\geq \max_i \omega_G(i,i+1\pmod{n}).

(\implies) For the first implication, suppose there exists an H^1 cycle in N\cup B. By Proposition 10, we may choose a cycle representative from N\cup B= A of \gamma=\sum_{i=1}^n b_i e_i with \{e_i\}_{i=1}^n being the set of 1-simplices corresponding to the edges of G. Note these are all in the complex because \delta\geq \max_i \omega_G(i,i+1\pmod{n}). Because our assumption is that \delta\geq \omega(n-1,n)+\omega(n,1), it follows that U_{n-1}\cap U_1\neq \emptyset implying N includes e_{n-1}=[n-1,n]. This complex includes e_n=[1,n] since U_n\subseteq U_1, as well as all the other e_i by construction. Thus the first chain group of N contains \gamma, and since N is a subcomplex of N\cup B, it follows that \gamma is nonbounding in N as well.

( \Longleftarrow ) For the second implication we assume there exists an H^1 cycle in N and one does not exist in N\cup B.

It follows that the addition of B adds 2-simplices to N which allow for the non-bounding cycle of the nerve to be expressed as a boundary. We showed in Proposition 10 that any non-bounding cycle of a cycle graph is homologous to the outer cycle from the base graph. Let \gamma=\sum_{i=1}^n b_i e_i be the nontrivial cycle representative from N, where \{e_i\}_{i=1}^n is the set of 1-simplices corresponding to outer cycle edges, that is the edges of G. By assumption there exists a 2-chain of N\cup B = A, \alpha, such that \partial_2(\alpha)=\gamma.

Recall that \Delta_b is a simplex in A created by the vertices v_j which have n in their out \delta neighborhood. We note that the simplices in the closure of \Delta_b which do not include n are already in N since the weight function of P(G) is preserved in P(G') for edges not containing n. By construction, n is a sink in all the remaining simplices of B. With this in mind, it follows

If the boundary sum on the left is zero, then the boundaries from \Delta_b must have canceled themselves out and the right sum constitutes a 2-chain of N whose boundary is \gamma, a contradiction. So we assume that the left summand is not zero.

Suppose \sum_{\Delta'=[i,n]\in B}a'_{\Delta'} [i,n]=0. Notice that that any boundary containing [n-1,n] with nonzero coefficient must also contain [i,n] for some i. Thus this assumption implies a'_{[n-1,n]}=0. It follows the boundary of simplices from \Delta_b, represented in the left summand, is \sum_{\substack{\Delta'=[i,j]\in N\\b\leq i,j< n}}a'_{\Delta'} [i,j]. However, since \omega_{P(G)}(b,n-1)\leq \delta it follows that [b,b+1,\dotsc,n-1]\in N which implies this boundary is trivial in N. This also contradicts the assumption that \gamma is a nonbounding cycle in N since the sum of boundaries trivial in H^1 cannot make a nontrivial element.

Thus there exists some \Delta'=[i,n]\in B where b\leq i<n-1 and a'_{\Delta'}\neq 0 above. Then it follows that an element in the right summand must cancel this out. Therefore [i,n]\in B\cap N. Since [i,n]\in B we know \omega_{P(G)}(i,n)\leq \delta by construction. However, from [i,n]\in N it follows that \omega_{P(G)}(n,i)\leq \delta (by the construction of U_n). But now i and n can see each other and we have a contradiction.

(\Longleftarrow) Without loss of generality, assume \max\{ \omega(1,p), \omega(p,1) \} < \delta where p<n. It follows that 1 sees all vertices i\leq p-1 and p sees all vertices p\leq i\leq n. Thus U_1 and U_p cover the complex. Since their intersection is contractible, it follows that the complex is homotopy equivalent to Nerv\{U_1, U_p\}, which is a 1-simplex and itself contractible.

( \Longrightarrow ) We prove the claim by induction.

As our base case we take G to be the consistently oriented cycle graph on three vertices. The Dowker complex \mathfrak{D}_\delta(G) then contains a two simplex which fills the cycle when some vertex contains the other two in its \delta out neighborhood. If \delta is also above \max\{\omega_{P(G)}(1,2),\omega_{P(G)}(2,3),\omega_{P(G)}(3,1)\}, as is our assumption, then we have a bidirected edge.

Now assume that the claim is true for cycle graphs on n-1 vertices and let G be a cycle graph on n vertices such that no pair of vertices in P(G) see each other with weight under \delta.

For \max_i \omega_{P(G)}(i,i+1 \pmod{n})\leq \delta < \min_{i}\{\omega_{P(G)}(i,i+1\pmod{n})+\omega_{P(G)}(i+1\pmod{n},i+2\pmod{n})\} the lower bound allows the cycle of the form \sum_{i=1}^n a_ie_i to exist, and the upper bound implies the complex has no 2-simplices and thus the cycle cannot be a boundary.

Now assume \delta \geq \min_{i}\{\omega_{P(G)}(i,i+1\pmod{n})+\omega_{P(G)}(i+1\pmod{n},i+2\pmod{n})\}. By symmetry we may assume without loss of generality this minimum is achieved by \omega_{P(G)}(n-1\pmod{n},n)+\omega_{P(G)}(n,1). Let G' be the contraction of G with respect to vertex n. If no pair of vertices in P(G) see each other with this \delta, then the same is true for P(G'). Since G' is a cycle graph on n-1 vertices, our assumption indicates that its Dowker complex has a nontrivial H^1 cycle. Proposition 13 indicates that under these assumptions \mathfrak{D}_\delta(G') has a nontrivial H^1 cycle if and only if \mathfrak{D}_\delta(G) has a nontrivial H^1 cycle, which proves our claim for \mathfrak{D}_\delta(G).

By Proposition 10 we know that this homology group is trivial for \delta below the lower bound.

In Proposition 14 we found that there exists a nontrivial element of the first homology group for \delta above that range if and only if there do not exist i and j such that \omega(i,j),\omega(j,i)<\delta. The term \min_{i\neq j}\max \{\omega(i,j),\omega(j,i)\} indicates the first value of \delta at which this property would hold for some i and j. If \delta \geq \min_i \omega(i,i-1\pmod{n}), then a vertex has low weight edges too all vertices of the graph and the dowker complex is fully connected with trivial first homology. Thus if \delta satisfies the given inequalities, the first homology group has a nontrivial element.

Below the lower bound, the graph G does not contain a cycle with edge weights less than \delta, ignoring orientation. By Proposition 10 it follows the complex has trivial first homology in this case. This Proposition also guarantees that there is at most one first homology class.

Let \delta\geq\max_i\omega(i,i+1 \pmod{n}). For all k\in K, let A_k =\cup_{i\in \Gamma_k} U_i where \Gamma_k=\{i | \omega_{P(G)}(k,i)<\infty\} is the set of vertex indices which vertex k sees with finite weight. The set A_k is connected since \delta is greater than the maximum sequential edge weight, and if k sees a vertex, there exists a path with edges of weight less than \delta to it in G.

The set \{A_k\}_{k\in K} forms a cover by the fact that K is a dominating set. Since |K|\geq 3 it follows that each A_k is only incident to the cover elements on either side through a sink vertex (the orange vertices in Figure ?). Thus Nerv\{|A_k| | k\in K\} is a cycle of 1-simplices with |K| vertices. By the nerve theorem, this is homotopy equivalent to \mathfrak{D}_\delta(G), concluding the proof.

Following the assumption, let K=\{k_1,k_2\}. Since G is a cycle graph it follows that there must exist two sink vertices, s_1,s_2, which are each seen by k_1 and k_2. Without loss of generality, assume that 0=k_1<s_1<k_2<s_2.

Let A_1=\cup_{0<i<k_2}U_i and A_2=\cup_{k_2<i\leq n}U_i. We show that \mathfrak{D}_\delta(G) is homotopy equivalent to Nerv\{U_0,A_1,U_{k_2},A_2\}, and that this nerve has a nonbounding cycle precisely when \delta lies in the specified region.

By construction, we know that U_0\cup A_1\cup U_{k_2}\cup A_2=\cup_{0\leq i\leq n}U_i=\mathfrak{D}_\delta(G), implying that this set forms a proper cover.

by construction A_1 and A_2 are disjoint because all vertices 0<i<k_2 have no directed paths to the vertices k_2<j\leq n, and vice versa. At the same time U_0\cap U_{k_2}\subseteq\text{Cl}([s_1,s_2]) because s_1 and s_2 are the only vertices which both source vertices can see in P(G) for any positive weight.

If we consider another intersection, we see

because the union of out neighborhood intersections is over a directed path, implying its components are nested simplices. This resulting term is a simplex and thus contractible. Similar reasoning finds that U_0\cap A_2, U_{k_2}\cap A_1, and U_{k_2}\cap A_2 are contractible. The lower bound on \delta also guarantees that they are nonempty.

By the nerve theorem, we get the desired homotopy equivalence. From the nonempty intersections described in the previous paragraph, we find that this nerve contains a 4 vertex cycle. Whether this cycle is filled with two simplices or not depends on the other intersections. Since A_1\cap A_2=\emptyset, it follows that A_1\cap A_2\cap U_0=A_1\cap A_2\cap U_{k_2}=\emptyset. Thus the 4-cycle in the nerve is filled if and only if A_1\cap U_{k_2}\cap U_0 and A_2\cap U_{k_2}\cap U_0 are nonempty. from our previous conclusions from the structure of the underlying graph this means A_1\cap U_{k_2}\cap U_0=[s_1] and A_2\cap U_{k_2}\cap U_0=[s_2]. This implies \omega(0,s_1),\omega(0,s_2),\omega(k_2,s_1),\omega(k_2,s_2)\leq\delta. Thus we may conclude that the complex has a nontrivial element in the first homology if and only if

Since the source vertices are the vertices furthest away from the sinks, the above reduces to the desired inequality.

Since G is a cycle graph with one source and one sink, it must consist of two directed paths joining vertices k and s which are disjoint outside of their endpoints. By symmetry, assume without loss of generality that k=0. We proceed with cases.

(Case 1) Suppose \delta<\omega(k,s). Let W_1=U_1\cup \cdots\cup U_{s-1} and W_2=U_{n-1}\cup \cdots \cup U_{s+1}. We prove that \mathfrak{D}_\delta(G) is homotopy equivalent to Nerv\{W_1,W_2,U_0\}.

By construction and the assumption on \delta, W_1\cap W_2=[s]. Since s is a source vertex, it has a trivial out neighborhood. Thus W_1\cup W_2\cup U_0=\cup_{0\leq i\leq n}U_0 suggesting that these three sub-complexes form a cover of the complex.

From \delta<\omega(k,s), it follows that s\not\in U_o. Thus W_1\cap W_2\cap U_0=\emptyset. At the same time, our assumption on \delta gives that U_0\cap W_1 and U_0\cap W_2 are nonempty. Also

since these out neighborhood intersections along a directed path are nested. It follows that this intersection is a simplex and therefore contractible. The same reasoning follows for W_2.

By the Nerve theorem, it follows that \mathfrak{D}_\delta(G) is homotopy equivalent to Nerv\{W_1,W_2,U_0\}. Further, we’ve shown that this nerve contains 3 vertex cycle and since W_1\cap W_2\cap U_0=\emptyset it follows that it is not filled in and therefore not a boundary.

(Case 2) Suppose \delta\geq\omega(k,s). If there does not exist an i such that \max\{\omega(k,s),\omega(k,i),\omega(i,s)\}\neq \omega(k,s), then following the analysis of the previous case, we find the resulting nerve is a filled 2-simplex, and therefore has no nontrivial 1-homology.

If there exists i such that \max\{\omega(k,s),\omega(k,i),\omega(i,s)\}> \omega(k,s), then it follows that the weight sum of one path from k to s is smaller than that of the other. Thus for any i along the shorter path, \max\{\omega(k,s),\omega(k,i),\omega(i,s)\}= \omega(k,s) by construction of the weight \omega(k,s). Thus i is a vertex along the longer path, and by symmetry let us assume that it is some i<s. That is, we assume the longer path from k to s encompasses the vertices i<s.

Let j<s be the vertex on the longer path which maximizes \omega(j,s) less than \delta. If j\geq 1, let W_j=U_1\cup \cdots \cup U_{j-1}, and otherwise let W_j=[j]. We show homotopy equivalence to Nerv\{U_0,W_j,U_j\}. Notice that by assumption the vertices of the shorter path, i\geq s, are in U_0. Following the same reasoning as the previous case in regard to the intersection with a union of out neighborhoods along a path, we find that W_j intersects contractibly with U_0 and U_j.

If U_0\cap W_j\cap U_j=\emptyset then again by the Nerve theorem the complex is homotopy equivalent to Nerv\{U_0,W_j,U_j\}. We have also shown that this nerve is a triangle not filled with a 2-simplex, admitting a single nontrivial element in the the first homology.

Now we show that \delta is below our upper bound in this case. Because the intersection is empty it follows that j\not\in U_0. Thus \omega(k,j)>\delta. It follows that \max\{\omega(k,s),\omega(k,j),\omega(j,s)\}=\omega(k,j)>\delta. Since \omega(k,i)>\omega(k,j) for all j<i<s, it follows that

for all j<i\leq s. We conclude that a minimizing vertex m such that

must have index less than j, meaning m<j. However, in this case \omega(m,s)>\omega(j,s), and j was chosen so that \omega(j,s) is maximal, meaning that \omega(m,s)>\delta. It then follows

and so \delta is less than the proposed upper bound.

If U_0\cap W_j\cap U_j\neq \emptyset then j\in U_0. It follows \omega(k,s),\omega(k,j),\omega(j,s)\leq \delta, which implies that \delta\geq \max\{\omega(k,s),\omega(k,j),\omega(j,s)\}. Thus \delta\geq \min_{0<i<s}\max\{\omega(k,s),\omega(k,i),\omega(i,s)\}. We concluded before that for vertices on the shorter path, \min_{s<i<n}\max\{\omega(k,s),\omega(k,i),\omega(i,s)\}=\omega(k,s). Thus \min_{s<i<n}\max\{\omega(k,s),\omega(k,i),\omega(i,s)\}\leq \min_{0<i<s}\max\{\omega(k,s),\omega(k,i),\omega(i,s)\}\leq \delta which implies \delta is greater than the proposed upper bound.

The result follows immediately after applying Lemma 20 then Lemma 21.

Let A= \text{Cl}(\mathfrak{D}^\star_\delta(P(G_1\vee G_2))\setminus \mathfrak{D}^\star_\delta(P(G_1)\vee P(G_2))) and B=\mathfrak{D}^\star_\delta(P(G_1)\vee P(G_2)).

Our first claim is that A is contractible. Notice that the maximal simplices of A are not in \mathfrak{D}^\star_\delta(P(G_1)\vee P(G_2)). Thus they require edges between G_1\setminus x_0 and G_2\setminus x_0 in P(G_1\vee G_2). Since any path in G_1\vee G_2 inducing such an edge must pass through x_0, it follows that the maximal simplices of A all contain x_0. This induces a free face of every maximal simplex, which we may collapse with respect to. That is a maximal simplex \sigma=[\sigma_1,\cdots,x_0,\cdots,\sigma_n] implies [\sigma_1,\cdots,\hat{x_0},\cdots,\sigma_n] is free. Consider the complex A' where we perform a simplicial collapse with respect to this free face of every maximal simplex. It follows that the maximal simplices of A' also contain x_0. Thus we may inductively collapse the maximal simplices until only x_0 remains. This shows the desired contractibility.

Our second claim is that A\cap B is contractible. This is precisely simplices returned by the closure in the definition of A. Let \sigma=[\sigma_1,\dotsc,\sigma_n] be a maximal simplex of A, induced by a path from \sigma_1\in V(G_1)\setminus x_0 to \sigma_n\in v(G_2)\setminus x_0. This simplex implies \eta_1=[\sigma_1,\dotsc,x_0] and \eta_2=[x_0,\dotsc,\sigma_n] in A\cap B representing paths to and from x_0, respectively. Thus all maximal simplices of A\cap B contain x_0. Repeating the same argument as for A, we find that A\cap B is contractible.

Given that \tilde{H}_n(A)=\tilde{H}_n(A\cap B)=0, it follows by Mayer Vietoris that the following sequence is exact.

This implies the desired isomorphism

Notice that \mathfrak{D}^\star_\delta(P_1),\mathfrak{D}^\star_\delta(P_2)\subseteq \mathfrak{D}^\star_\delta(P_1\vee P_2). It follows that \mathfrak{D}^\star_\delta(P_1)\vee \mathfrak{D}^\star_\delta(P_2)\subseteq \mathfrak{D}^\star_\delta(P_1\vee P_2), where \vee denotes the simplicial wedge sum. We assume the simplicial and graph wedge sums identify the same vertex, x_0, and that that vertex is not isolated for our choice of \delta.

By construction x_0, in the graph P_1\vee P_2, is the only vertex with neighbors in P_1 and P_2, when they are seen as subsets of their wedge sum. Otherwise connectivity is unchanged. It follows that \mathfrak{D}^\star_\delta(P_1\vee P_2) = \mathfrak{D}^\star_\delta(P_1)\vee \mathfrak{D}^\star_\delta(P_2) \cup N_\delta(x_0) where N_\delta(x_0) is the \delta neighborhood simplex of x_0.

Notice now that the intersection (\mathfrak{D}^\star_\delta(P_1)\vee \mathfrak{D}^\star_\delta(P_2)) \cap N_\delta(x_0)=(\mathfrak{D}^\star_\delta(P_1)\cap N_\delta(x_0)\vee (\mathfrak{D}^\star_\delta(P_2)\cap N_\delta(x_0) ) since the base point x_0 is in the intersection. This is the wedge of the \delta neighborhood simplices of x_0 in the component complexes separately, and is clearly contractible.

Since the intersection is contractible, through a standard Mayer-Vietoris argument we can show that \tilde{H}_n(\mathfrak{D}^\star_\delta(P_1)\vee \mathfrak{D}^\star_\delta(P_2) \cup N_\delta(x_0))) \approx \tilde{H}_n(\mathfrak{D}^\star_\delta(P_1)\vee \mathfrak{D}^\star_\delta(P_2)).

Notice that the assumption on the base points guarantees that x_0 is not an isolated point in the complex (in particular, that the basepoints form good pairs with their parent complex). It follows from a standard result that \tilde{H}_n(\mathfrak{D}^\star_\delta(P_1)\vee \mathfrak{D}^\star_\delta(P_2))\approx \tilde{H}_n(\mathfrak{D}^\star_\delta(P_1))\oplus \tilde{H}_n(\mathfrak{D}^\star_\delta(P_2))(e.g. Corollary 2.25 in [Hatcher_AT]).

Since W is of finite length, it follows that all edges along it are finite as well. Consider the partition cell the starting vertex of W is in. We denote this partition cell with v_1 and refer to it interchangeably with the corresponding vertex of G^\pi. Because W spans V(G), it follows that there exists a first vertex outside of partition cell v_1. Let the partition cell containing this vertex be v_2. This process can be continued inductively to span all the vertices of G^\pi. Note that this walk may revisit the same vertices or edges multiple times before seeing new ones.

As discussed previously, represented as a graph G, a dynamical trajectory is simply a directed path. Thus P(G) has both a one element source dominating set (universal source) and a one element sink dominating set (universal sink). The binning procedure can be described as a reduction of the graph to an unnormalized quotient graph under some vertex partition \pi. By Proposition 23 it follows that there exists a vertex of G^\pi with a directed path to any other vertex, and a vertex which all other vertices have a directed path to. Thus P(G^\pi) must also have a universal source and sink. By Proposition 9 it follows that P(G^\pi) has contractible maximal source and sink complexes.

By Proposition 23, every directed path of h induces a directed walk in G^\pi. Let the h^\pi be the set of these walks.

Consider the first vertex, v_0, on some walk w\in h^\pi. Naturally G^\pi contains a directed path from v_0 to each vertex of w. Thus the out neighborhood of v_0 in P(G^\pi) contains all vertices of w.

By assumption the walks in h^\pi span the vertex set of G^\pi. Thus h^\pi induces a source dominating set of P(G^\pi) of size |h^\pi|=|h|.

The conclusion follows from Proposition 9.