Computing singular simplicial homologies of digraphs and quivers Matthew Burfitt, Jie Wu, Stephen S.-T. Yau, Shing-Tung Yau Abstract The dynamics of large complex systems are predominately modeled through pairwise interactions, the principle underlying structure being a network of the form of a digraph or quiver. Significant success has been obtained in applying the homology of the directed flag complex to study digraphs arising as networks within numerous scientific disciplines. This homology of directed cliques enjoys relative ease of computation when compared to other digraph homologies, making it preferable for use in applications concerning large networks. By extending the ideas of singular simplicial homology to quivers in categories of different morphism types, several new singular simplicial homology theories have recently been constructed. Computationally efficient homologies for quivers have in general not previously been seriously considered. In this paper we develop further the homotopy theory of quivers necessary to derive functors that realise isomorphisms between the singular simplicial quiver homologies and the homologies of certain spaces. The simplicial chains of these spaces arise in a conveniently compact form that is at least as convenient as the directed flag complex for computations. Moreover, our constructions are natural with respect to the isomorphisms on homology making them suitable for applications in conjunction with persistent homology for practical use. In particular, for each of the singular simplicial homologies considered, we provide efficient algorithms for the computation of their persistent homology. 1 Introduction The investigation of complex systems is the study of a large number of distinct entries influencing each over though pairwise interactions from which emergent behavior results in global dynamics. Complex systems are ubiquitous, describing a vast verity of natural and manmade phenomenon, with greater prevalence resulting from the growing availability of large networked data sets. As a consequence of their importance, mathematical modeling of complex networks is extremely diverse [ 3 ] . However, network models are based universally on a collection of nodes and pairwise relations between them. Formally, a network has the mathematical structure of a graph, with greater flexibility provided by digraphs where pairwise relations are oriented between nodes. Notably, directed networks demonstrate phenomenon that cannot be captured by symmetric relations alone [ 1 ] . Significant examples of directed complex networks include food webs [ 4 ] , spreading of disease [ 36 ] , or neural connectivity both biological [ 16 ] and artificial [ 48 ] . Yet more structure can be encodes using quivers, allowing for multiple interactions of the same direction in the form of multiple edges and self interactions of a single node in the form of loops. In particular, both of these interaction types occur in circumstances such as gene regularly networks [ 37 ] , or representations of gauge theories [ 13 ] . The development of methods for the analysis of digraphs and quivers therefore forms a foundation for more sophisticated models including higher order interactions [ 14 ] , and networked dynamical systems [ 52 ] . Nevertheless, the graphical structure itself is in may circumstances sufficient [ 15 , 40 ] . Important large scale network properties often captured though local sub-networks combining to provide global structure [ 51 , 31 , 32 , 25 ] . Classical algebraic topology provides invariants of topological spaces, including singular homology, that extract global structure up to continuous deformations know as homotopies [ 42 , 24 ] . For computational purposes it is convenient to describe spaces as simplicial complexes, or more generally \Delta -sets. Given a weighted undirected network, a filtered sequences of Vietoris-Rips complexes can be obtained and persistent homology applied to capture multiscale network features. Using the graph metric induced by the weightings, persistent homology in demission zero is equivalent to single linkage hierarchical clustering [ 7 ] , describing network communities. In higher dimensions persistent homology features can be interpreted as measuring the scale of cycles and voids contained within the network, naturally formalising a description of higher-order structures. The fact that classical community structure alone is in general insufficient to capture all system properties [ 30 , 47 , 26 ] makes the detection of higher order structures particularly important. More generally, persistent homology is a widely applied method in the field of topological data analysis [ 44 ] , can be effectively computed [ 2 ] , and satisfies desirable stability proprieties [ 9 ] . When considering a directed graph, one obtains a space known as the directed flag complex by associating to each directed clique of vertices a simplex, as first described by Masulli and Villa [ 39 ] . Persistents homology can then be applied to extract global geometric structure from a weighted directed network though the topology of its associated filtered directed flag complex. More generally, once a filtered simplicial complex is obtained from a weighted directed network, other specialized methods can be developed for its analysis without the use of persistent homology [ 46 , 28 ] . Applications of the directed flag complex include the interpretation of brain activity [ 45 ] , dynamics of artificial neural networks [ 38 ] , protein-ligand binding affinity [ 53 ] , and the study of various interaction networks [ 29 ] . To enable the incorporation of persistent homology, in this work we primally consider weighted directed networks, allowing the further capacity to study networks with a real value associated to each interaction. While the directed flag complex is valued for the simplicity of its construction, in the case of directed networks it is not always immediately obvious to decide which directed sub-networks should be considered topologically trivial within the context of a given application. Therefore, it is harder to determine an appropriate filtered space to associate to the networks underlying digraph or quiver. Furthermore, major drawbacks of the the directed flag complex lie in the fact that it does not satisfy good homotopy [ 8 , §6] or stability properties [ 8 , §5] [ 50 , §5.2] . An alternative well developed approach to the homology of directed graphs is provided by the path homology of Grigor’yan, Lin, Muranov and Yau [ 19 ] , formulated around the idea of detecting cycles in equal length paths within a digraph. In contrast to the directed flag complex, path homology has been shown to satisfy variations of the most important properties of classical homology of spaces [ 20 , 23 , 18 ] , and has stable persistent homology when applied to weighted networks [ 11 , §5] . However, there is no agreed extensions of path homology to quivers [ 22 ] [ 27 , §12] , with any extension satisfying the desired homotopy invariance being trivial (see Remark 5.1 ). Moreover, effective computation of path homology faces major challenges [ 21 , §1.7 and §2.3] . The computational problems originate primarily from difficulties in explicitly describing a basis of the path homology chain complex. Consequently, efficient computation of path homology is presently limited to dimensions zero and one [ 12 ] . However, computation does not pose as significant a problem for singular homologies such as the directed flag complex, with developed highly parallelizable algorithms publicly available [ 35 ] making it possible to analyse large scale complex systems. For the purposes of applications, it would therefore be highly desirable to develop algorithms for singular homology theories with good theoretical proprieties that are also applicable to quivers in general, and that retain the computational advantages of the directed flag complex. The digraph consisting of a single clique on n+1 vertices 0,\dots,n with directed edges from any vertex to one with a larger value, may be considered a discrete directed analog of the usual geometric n -simplex. Given a category whose objects are quivers, a singular simplex is a morphism from the directed n -simplex to a given quiver. The singular homology of the quiver is then the homology of the chain complex freely generated by singular mappings in each dimension. In a recent paper Li, Muranov, Wu and Yau [ 33 ] gave three natural morphisms types: quiver mappings for which edges can be sent to edges or vertices, quiver homomorphisms which only send edges to edges, and the usual quiver inclusions. Using these three categories, three associated singular simplicial homologies H_{*}^{\Delta,m} , H_{*}^{\Delta,h} , and H_{*}^{\Delta,i} of quiver were introduced with associated functorial and homotopy proprieties provided. Notably, persistent H_{*}^{\Delta,m} coincides with the homology of the ordered tuple complex of a digraph introduced by Turner, who proved a stability theorem [ 50 , Theorem 21] in this case. In particular, strong homotopy invariance [ 33 , Theorem 5.8] and stability provide H_{*}^{\Delta,m} with a significant advantage over the homology of the directed flag complex when applied to complex networks. In this work we further develop the homotopy theoretic properties of the singular simplicial homologies H_{*}^{\Delta,m} , H_{*}^{\Delta,h} , and H_{*}^{\Delta,i} , providing efficient algorithms necessary to apply them to complex systems in practice. This is achieved by associating to each quiver G and singular homology a spaces such that its homology coincides with H_{*}^{\Delta,m}(G) , H_{*}^{\Delta,h}(G) , or H_{*}^{\Delta,i}(G) . Significantly, for a finite quiver the chain complexes of these spaces typically consist of a considerably smaller number of generators. Other approaches to singular homologies of digraphs considered in the context of applied topology are Dowker complexes [ 10 ] , the preorder homology of strongly connected components [ 50 , §6] , complexes of tournaments [ 17 ] and the homology of closure spaces [ 5 ] . We begin in Section 2 by introducing the necessary algebraic topology background and results from [ 33 ] . Furthermore, we provide in Section 2.5 a generalised directed flag functor to the category of \Delta -sets whose homology naturally coincides with H_{*}^{\Delta,i}(G) for an arbitrary quiver G . Under certain restrictions on G , the homologies H_{*}^{\Delta,m} , H_{*}^{\Delta,h} , and H_{*}^{\Delta,i} are the same, and these relationships are fully described in Section 3 . Explicitly, H_{*}^{\Delta,h} and H_{*}^{\Delta,i} agree when G has no loops, and all three homologies coincide when considering digraphs without double edges. Moreover, these results are seen to be strict through the construction of counter examples in the absence of the necessary conditions. The main constructions presented in this work begin in Section 5 , where we detail the necessary theoretical machinery to achieve efficient computation of H_{*}^{\Delta,m} . We initially describe in Section 5.1 the reduced digraph \bar{\mathcal{R}}(G) of a quiver G , which we show is a strong homotopy deformation retraction of G . As a consequence the strong homotopy invariance of H_{*}^{\Delta,m} now implies that H_{*}^{\Delta,m}(G) is not dependent on the quiver structure of G , being invariant under operations collapsing loops and multiple edges. This simplification leads in Section 5.2 to a proof that H_{*}^{\Delta,m}(G) is isomorphic to the homology of a construction we call the reduced directed flag complex \bar{\mathcal{F}}(G) . Moreover, \bar{\mathcal{F}} is a functor acting naturally with respect to this isomorphism, providing a suitable alternative framework for computation of persistent H_{*}^{\Delta,m} . In fact computing the homology of \bar{\mathcal{\mathcal{F}}}(G) is always simpler than computing H_{*}^{\Delta,m}(G) on its own, with H_{*}^{\Delta,m} chains in general being prohibitively large to compute the homology directly. This advantage is made explicit when we consider the computational complexity of the corresponding algorithms in Appendix A.4.1 . Building on the ideas of Section 5 , we develop analogous computational results for H_{*}^{\Delta,h} in Section 6 . However, in this case the situation is somewhat more complex. We make use of a new notion of local strong h -homotopy initially set-out in Section 4 , which H_{*}^{\Delta,h} is shown to be invariant under. The local strong h -homotopies are more flexible than the strong h -homotopies provided as an invariant of H_{*}^{\Delta,h} in [ 33 ] . Analogously to the reduced digraph \bar{\mathcal{R}} from Section 5 , local strong h -homotopies yield in 6.1 a deformation retraction of a quiver G to the partially reduced quiver \tilde{\mathcal{R}}(G) , which collapses multiple edges incident to vertices with a loop. This construction leads in Section 6.2 to a proof that H_{*}^{\Delta,m}(G) is isomorphic to the homology of a construction we call the partially reduced direct flag complex \tilde{\mathcal{F}}_{<}(G) , which in turn is applied to show H^{\Delta,h} is invariant under a weaker notion of weak local strong homotopy of quivers. Crucially, the \Delta -set \tilde{\mathcal{F}}_{<}(G) like \bar{\mathcal{F}} for H_{*}^{\Delta,m} provides a better intuitive description of the information captured by H_{*}^{\Delta,h} and is a basis for more efficient computation of persistent H_{*}^{\Delta,h} . In addition, the structure of \tilde{\mathcal{F}}_{<}(G) demonstrates the dependence of H_{*}^{\Delta,h} on both loops and multiple edges implying that unlike H_{*}^{\Delta,m} , H_{*}^{\Delta,h} detects the additional structure of the quiver not captured in its reduced digraph \bar{\mathcal{F}}(G) , while simultaneously satisfying a more flexible homotopy invariance proprieties than H_{*}^{\Delta,i} . Finally, based on the existing memory efficient parallelizable algorithms for computation of the directed flag complex of a digraph [ 35 ] , in Appendix A we detail the precise steps for the computation of the directed flag complex \mathcal{F}(G) , reduced directed flag complex \bar{\mathcal{F}}(G) , and partially reduced directed flag complex \tilde{\mathcal{F}}(G) of a quiver G . All algorithms are considered over filtered quivers so that they may be applied together with existing packages for persistent homology and demonstration code is made available at [ 6 ] . We additionally analyse the computational complexity of each algorithm, detailing the time savings when computing the persistent homology of \bar{\mathcal{F}}(G) and \tilde{\mathcal{F}}(G) over direct computation of persistent H_{*}^{\Delta,m}(G) and H_{*}^{\Delta,h}(G) , respectively. 2 Background We now provide the necessary background on simplicial homology and singular simplicial homologies of digraphs and quivers used during the remainder of this work. Throughout the section assume that n is a non-negative integer and R a commutative ring, unless stated otherwise. 2.1 Abstract simplicial complexes and \Delta -sets In this subsection we detail the properties of the models of space used in this work. We make use of abstract simplicial complexes and \Delta -sets as they provide flexible combinatorial structures that work well for computer computations. An abstract simplicial complex (S,V) consist of a vertex set V and a non-empty set of finite subsets S of V such that s\in S\;\text{and}\;s^{\prime}\subseteq s\implies s^{\prime}\in S. An s\in S with size n+1 is called an n -simplex of S . Let (S_{1},V_{1}) and (S_{2},V_{2}) be abstract simplicial complexes. A simplicial map of abstract simplicial complexes f\colon(S_{1},V_{1})\to(S_{2},V_{2}) is a map of sets f\colon V_{1}\to V_{2} such that \{v_{0},\dots,v_{n}\}\in S_{1}\implies\{f(v_{0}),\dots,f(v_{n})\}\in S_{2}. Denote the category of abstract simplicial complexes and simplicial maps by ASim . A \Delta -set or semi-simplicial set X consists of a sequence of sets \{X_{n}\}_{n=0}^{\infty} and face maps d_{i}^{n}\colon X_{n+1}\to X_{n} for each integer n\geq 0 and i=0,\dots,n+1 , such that d_{i}^{n}\circ d_{j}^{n+1}=d_{j-1}^{n}\circ d_{i}^{n+1} (1) where j=1,\dots,n+1 and i|\>u\neq v\}. We also denote an edge (u,v)\in E_{G} by u\to v . A quiver G=(V_{G},E_{G},s_{G},t_{G}) , consists of a non-empty set of vertices V_{G} and a set of edges E_{G} , as well as maps s_{G}\colon E_{G}\to V_{G}\;\;\;\text{and}\;\;\;t_{G}\colon E_{G}\to V_{G} called the source and target functions, respectively. We often drop the subscript G from s_{G} and t_{G} , as it is usually clear from the context which quiver they are associated to. A subquiver of G is a quiver H such that V_{H}\subseteq V_{G} , E_{H}\subseteq E_{G} and s_{H} , t_{H} are the restrictions of s_{G} , t_{G} to E_{H} . A subquiver H of G is full if E_{H}=\{e\in E_{G}\mid s_{G}(e)\in V_{H}\>\text{and}\>t_{G}(e)\in V_{H}\} . In addition, given a quiver G and vertices u,v\in V_{G} , denote by G_{u,v} the number of edges e\in E_{G} such that s_{G}(e)=u and t_{G}(e)=v . A digraph G naturally carries the structure of a quiver by setting s_{G}((u,v))=u\;\;\;\text{and}\;\;\;t_{G}((u,v))=v for each (u,v)\in E_{G} . Therefore, from now on we usually treat a digraph as a special case of a quiver without explicitly stating the application of the above identification. We also make use of the following additional terminology for quivers. A quiver G is said to have a double edge between u,v\in E_{G} if there are d_{1},d_{2}\in E_{G} such that s(d_{1})=t(d_{2})=u\neq v=t(d_{1})=s(d_{2}). A quiver G has double edges if it has at least one double edge between some pair of vertices. An edge d\in E_{G} is called a double edge if there is a double edge between s(d) and t(d) . A quiver G has a multiple edge between u\in V_{G} and v\in V_{G} if there are m_{1},m_{2}\in E_{G} such that m_{1}\neq m_{2},\;\;\;s(m_{1})=s(m_{2})=u,\;\;\;\text{and}\;\;\;t(m_{1})=t(m_{2})=v. A quiver G has multiple edges if it has at least one multiple edge between some pair of vertices. An edge m\in E_{G} is called a multiple edge if there is are multiple edges between s(m) and t(m) . Finally, a quiver G is said to have a loop at vertex v\in V_{G} , if there is an l\in E_{G} such that s(l)=t(l). A quiver G has loops if it has at least one loop at a vertex. An edge e\in E_{G} is said to have a loop , when either of the vertices s(e) or t(e) have a loop. As a quiver, a digraph can have double edges but not multiple edges or loops Throughout this work three different class of morphisms between quivers are considered. These morphisms are; quiver maps, quiver homomorphisms, and quiver inclusions, with each a specialisation of the preceding morphism type. More precisely, let G and G^{\prime} be quivers. A map of quivers m\colon G\to G^{\prime} is a pair of functions m_{V}\colon V_{G}\to V_{G^{\prime}}\;\;\;\text{and}\;\;\;m_{E}\colon E_{G}\to E_{G^{\prime}}\coprod V_{G^{\prime}} such that for every e\in E_{G} , either \displaystyle m_{E}(e)\in E_{G^{\prime}}\>\text{and}\>m_{V}(s(e))=s(m_{E}(e)),\>m_{V}(t(e))=t(m_{E}(e))\> or \displaystyle m_{E}(e)=m_{V}(s(e))=m_{V}(t(e))\in V_{G^{\prime}}. When the image of m_{E} lies in E_{G^{\prime}} , m is called a homomorphism . If in addition m_{V} and m_{E} are also injective, then m is an inclusion . When it is clear from the context what is meant, from now on we write m instead of m_{V} or m_{E} . Each choice of morphism type provides a different category of quivers, and each such category will be a subcategory of a category formed using a less strict morphism type. We denote the categories of quivers formed using inclusions, homomorphisms, and maps by \textbf{Quiv}_{i}\subset\textbf{Quiv}_{h}\subset\textbf{Quiv}_{m}, respectively. 2.3 Singular simplicial homologies of digraphs and quivers The homology theories presented in this subsection were defined in [ 33 ] alongside cubical homology theories on two types of singular n -cubes with respect to the three morphism types set out in the previous subsection. Throughout this section assume that G is a quiver and n\geq 0 an integer, unless otherwise stated. The n -dimensional directed simplex \Delta^{n} is a digraph on vertex set V_{\Delta^{n}}=\{0,\dots,n\} and edge set E_{\Delta^{n}}=\{(a,b)\in V_{\Delta^{n}}\times V_{\Delta^{n}}\>|\>aj|\>i=0,\dots,n,\>\text{the}\>v_{i}\>\text{are distinct,} \displaystyle 0\leq j\text{are integers, and}\>s(e_{j,k})=v_{j},\>t(e_{j,k})=v_{k}\}. The face map d^{n-1}_{t}(x) for t=0,\dots,n and x\in\mathcal{F}_{n}(G) is given by the subset of x obtained by removing v_{t} , all edges e_{a,t}\in x for a=0,\dots,t-1 , and all edges e_{t,b}\in x for b=t+1,\dots,n . As the elements of \mathcal{F}(G)_{n} coincidence precisely with the images of singular simplex inclusions, we immediately obtain that H_{*}^{\Delta,i}(G)=H_{*}(\mathcal{F}(G)). (5) Moreover, an inclusion of quivers \phi\colon G\to G^{\prime} induces an inclusion of \Delta -sets \mathcal{F}(\phi)\colon\mathcal{F}(G)\to F(G^{\prime})\;\;\;\text{given by}\;\;\;\mathcal{F}(\{v_{i},e_{l,k}\}_{\begin{subarray}{c}0\leq i\leq n\\ 0\leq jE_{G},\>V_{G^{\prime}} and E_{G^{\prime}} are disjoint sets. The strong box product of G and G^{\prime} denoted G\boxtimes G^{\prime} , is the quiver with \displaystyle V_{G\boxtimes G^{\prime}}=V_{G}\times V_{G^{\prime}},\;\;\;E_{G\boxtimes G^{\prime}}=(E_{G}\times V_{G^{\prime}})\cup(V_{G}\times E_{G^{\prime}})\cup(E_{G}\times E_{G^{\prime}}) and \displaystyle s_{G\boxtimes G^{\prime}}((e,v^{\prime})) \displaystyle=(s_{G}(e),v^{\prime}),\;\;\;s_{G\boxtimes G^{\prime}}((v,e^{\prime}))=(v,s_{G^{\prime}}(e^{\prime})),\;\;\;s_{G\boxtimes G^{\prime}}((e,e^{\prime}))=(s_{G}(e),s_{G^{\prime}}(e^{\prime})), \displaystyle t_{G\boxtimes G^{\prime}}((e,v^{\prime})) \displaystyle=(t_{G}(e),v^{\prime}),\;\;\;\>t_{G\boxtimes G^{\prime}}((v,e^{\prime}))=(v,t_{G^{\prime}}(e^{\prime})),\;\;\;\>t_{G\boxtimes G^{\prime}}((e,e^{\prime}))=(t_{G}(e),t_{G^{\prime}}(e^{\prime}))\; where v\in V_{G} , v^{\prime}\in V_{G^{\prime}} , e\in E_{G} , and e^{\prime}\in E_{G^{\prime}} . In the case of G\boxtimes I , there are two natural inclusions \displaystyle i_{0}\colon G\to G\boxtimes I\;\;\;\text{given by}\;\;\;v\mapsto(v,0),\;e\mapsto(e,0) and \displaystyle i_{1}\colon G\to G\boxtimes I\;\;\;\text{given by}\;\;\;v\mapsto(v,1),\;e\mapsto(e,1) (8) where v\in V_{G} and e\in E_{G} . A choice of quiver product and type of morphisms leads to the construction of digraph homotopies in the following manner. Definition 5 . Quiver maps f_{0},f_{1}\colon G\to H are 1 -step strong homotopic if there is a quiver map \displaystyle F\colon G\boxtimes I\to H\;\;\;\text{such that}\;\;\;F\circ i_{0}=f_{0},\;F\circ i_{1}=f_{1}. In this case we write f_{0}\simeq_{1}^{S}f_{1} . More generally, quiver maps f,g\colon G\to H are strong homotopic if they are related by the equivalence relation generated by 1 -step strong homotopies. In which case we write f\simeq^{S}g . Quivers G , G^{\prime} are strong homotopy equivalent if there are quiver maps \displaystyle f\colon G\to G^{\prime},\;h\colon G^{\prime}\to G\;\;\;\text{such that}\;\;\;h\circ f\simeq^{S}\text{id}_{G},\;f\circ h\simeq^{S}\text{id}_{G^{\prime}}. A quiver is called strongly contractable if it is strong homotopy equivalent to a quiver consisting of a single vertex and no edges. Similarly, we also obtain the construction of 1 -step strong h -homotopic , strong h -homotopic , \simeq^{Sh} , strong h -homotopy equivalent , and strongly h -contactable , by making the above definitions with respect to quiver homomorphisms rather than quiver maps. In the reminder of this work, we often develop the strong homotopy and strong h -homotopy cases in parallel indicating inside brackets the differences in terminology. Remark 2.1 . To be sure \simeq_{1}^{S} generates an equivalence relation with a well defined composition of homotopy classes, we need to check that the transitivity property. More precisely, given quiver maps f\colon G_{1}\to G_{2} , g,g^{\prime}\colon G_{2}\to G_{3} , and h\colon G_{3}\to G_{4} we require that g\simeq_{1}^{S}g^{\prime}\implies g\circ f\simeq_{1}^{S}g^{\prime}\circ f\;\;\;\text{and}\;\;\;h\circ g\simeq_{1}^{S}h\circ g^{\prime}. If F\colon G_{2}\boxtimes I\to G_{3} is a one step strong homotopy from g to g^{\prime} , then h\circ g\simeq_{1}^{S}h\circ g^{\prime} is realised by h\circ F . Furthermore, let F^{\prime}\colon G_{1}\boxtimes I\to G_{2}\boxtimes I be given by \displaystyle F^{\prime}((v,0)) \displaystyle=(f(v),0),\;F^{\prime}((v,1))=(f(v),1),\;F^{\prime}((e,0))=(f(e),0), \displaystyle F^{\prime}((e,1)) \displaystyle=(f(e),1),F^{\prime}((v,0\to 1))=(f(v),0\to 1),\;\text{and}\;F^{\prime}((e,0\to 1))=(f(e),0\to 1) for each v\in V_{G_{1}} and e\in E_{G_{1}} . Then we verify that g\circ f\simeq_{1}^{S}g^{\prime}\circ f using the map F\circ F^{\prime} . Moreover, when f is a homomorphism F^{\prime} is also a homomorphism. Therefore, transitivity of one step strong h -homotopies is obtained by precisely the same argument as above. A map r\colon G\to G is called a retraction onto subquiver H of G if r(V_{G})\subseteq V_{H} , r(E_{G})\subseteq V_{H}\coprod E_{H} and the restrictions of r to the vertices and edges of H are the identities. Retraction r\colon G\to G is called a strong ( h -)deformation retraction if there is a strong ( h -)homotopy from r to the identity on G . As in classical homotopy theory, the existence of a strong ( h -)deformation retraction of a quiver G onto a subquiver H implies that G is strong ( h -)homotopy equivalent to H . The next theorem is obtained from [ 33 , Theorem 5.8 and Theorem 5.17] . Theorem 2.3 ( [ 33 ] ) . The following homotopy invariance properties hold. (1) If quiver maps f,g\colon G\to G^{\prime} satisfy f\simeq^{S}g , then f_{*}=g_{*}\colon H_{*}^{\Delta,m}(G)\to H_{*}^{\Delta,m}(G). (2) If quiver homomorphisms f,g\colon G\to G^{\prime} satisfy f\simeq^{Sh}g , then f_{*}=g_{*}\colon H_{*}^{\Delta,h}(G)\to H_{*}^{\Delta,h}(G). As stated in [ 33 , §6] , it is straightforward to check that H_{*}^{\Delta,i} is invariant under isometries. In the case of digraphs, a more general construction of homotopies under which H_{*}^{\Delta,i} is invariant is provided for a certain class of triangle collapsing maps on directed flag complexes in [ 8 , Definition 5.13] . In particular, these homotopies are derived from homotopies of path complexes, which generalise the original notion of digraph homotopy. 3 Conditions on equivalence between singular simplicial morphisms and homologies In this section we examine the relationships between the singular simplicial homologies within the categories \textbf{Quiv}_{m} , \textbf{Quiv}_{h} , and \textbf{Quiv}_{i} . More precisely, we investigate under what conditions homologies H^{\Delta,m}_{*} , H^{\Delta,h}_{*} , H^{\Delta,i}_{*} coincide and provide examples to demonstrate when these singular simplicial homologies are in general distinct for quivers with and without double edges, multiple edges, or loops. The main results presented in the section largely follow directly from the realisation that under certain conditions the different types of singular simplices are in fact identical, and we do not require the construction of quasi-isomorphisms between the chain complexes. Throughout the section assume that G is a quiver and n\geq 0 is an integer, unless otherwise stated. Note that some results in this section require the axiom of choice if the quiver G contains an infinite number of edges. We first consider the case when the quiver has no loops. Proposition 3.1 . Let G be a quiver without loops, then a singular simplicial map f\colon\Delta^{n}\to G is an inclusion if and only if it is a homomorphism. In particular, H^{\Delta,i}_{*}(G)=H^{\Delta,h}_{*}(G). Proof. When n=0 homomorphisms f\colon\Delta^{0}\to G are the same as inclusions of a vertex. For n\geq 1 , suppose f\colon\Delta^{n}\to G is a homomorphism which is not an inclusion, then there must be a pair of distinct vertices sent to the same vertex under f . Any pair of distinct vertices in the n -simplex \Delta^{n} are connected by a directed edge in some direction. Therefore, as f is a homomorphism there must exist an edge of \Delta^{n} sent to a loop under f , which contradicts the assumption that G is a quiver without loops. ∎ Given a quiver with at least one loop, the statement of the proposition above immediately fails to hold, as in this case there is a homomorphism of \Delta^{1} onto any loop and this is not an inclusion. In particular, H^{\Delta,i}_{*} is unaffected by the removal or addition of loops in the quiver, while H^{\Delta,h}_{*} might vary. However, it is straightforward to see that all homomorphisms from \Delta^{n} to a loop in dimensions n\geq 1 generate an acyclic sub-chain complex of C^{\Delta,h}_{*}(G) . A demonstration that H^{\Delta,i}_{*} and H^{\Delta,h}_{*} differ in general is provided in Example 3.1 . Definition 6 . A singular simplex f\colon\Delta^{n}\to G is called degenerate if there is an i=1,\dots,n such that f(i-1)=f(i)=f(i-1\to i). Otherwise, f is called non-degenerate . Ignoring the additional condition on the image of the edge, the definition of degenerate singular simplices provided above coincides with the property satisfied by the elements of the conical basis corresponding to degenerated simplices in the usual chain complex associated to a \Delta -sets. Consequently, the C^{\Delta,m}_{*}(G) sub-chain complex C^{\Delta,dm}_{*}(G) generated by degenerate simplices is an acyclic direct summand of C^{\Delta,m}_{*}(G) . Therefore, we obtain that H^{\Delta,m}_{*}(G)=H_{*}\left(\frac{C^{\Delta,m}_{*}(G)}{C^{\Delta,dm}_{*}(G)}\right). (9) We now consider when H^{\Delta,m}_{*} coincides with H^{\Delta,i}_{*}(G) and H^{\Delta,h}_{*}(G) in general. Proposition 3.2 . Let G be a digraph without double edges, then non-degenerate singular simplicial maps coincide with singular simplicial homomorphisms and singular simplicial inclusions. In particular, \displaystyle H^{\Delta,i}_{n}(G)=H^{\Delta,h}_{n}(G)=H^{\Delta,m}_{n}(G). Proof. The equivalence between singular simplicial inclusions and homomorphisms follow immediately from Proposition 3.1 . Therefore, we need only prove that singular simplicial inclusions and non-degenerate singular simplicial maps coincide. When n=0 or n=1 all non-degenerate maps f\colon\Delta^{n}\to G are inclusions. For n\geq 2 , suppose that f\colon\Delta^{n}\to G is a non-degenerate map which is not an inclusion. In this case, there must be a pair of vertices a and c of \Delta^{n} such that aa+1 and there is a vertex b of \Delta^{n} such that an=0\>\text{or}\>n=2\\ 0&\text{otherwise}.\end{cases} Where the dimension 2 homology is generated by the sum of the image of the two singular 2 -simplices with 0 sent to w subtracted from the sum of the two singular 2 -simplices with 0 sent to w^{\prime} . When a loop is added at vertex u to obtain a quiver G^{\prime} from G , the chain complex C^{\Delta,i}_{*}(G^{\prime}) remains unchanged and H^{\Delta,i}_{n}(G^{\prime};\mathbb{Z})=H^{\Delta,i}_{n}(G;\mathbb{Z}) . However, there is a single additional singular generator of C^{\Delta,h}_{1}(G) lying on the loop, one additional non-degenerate singular 2 -simplex f and two additional non-degenerate singular 3 -simplices g and h , uniquely determined by their vertex images \displaystyle f(0) \displaystyle=u,\>f(1)=v,\>f(2)=u, \displaystyle g(0) \displaystyle=w,\>g(1)=u,\>g(2)=v,\>g(3)=u \displaystyle\text{and}\;h(0) \displaystyle=w^{\prime},\>h(1)=u,\>h(2)=v,\>h(3)=u. As a result, we have that H^{\Delta,h}_{n}(G^{\prime};\mathbb{Z})=0 for each n\geq 1 . In either case, the homology of the quivers G and G^{\prime} is acyclic under H^{\Delta,m}_{*} by Theorem 2.3 , as both G and G^{\prime} are strongly contractible. In addition, when the double edge between u and v in the original digraph G is replaced with a single edge, any homology in dimension 2 vanishes in all cases. Furthermore, the above homologies of G and G^{\prime} remain unchanged if the double edge v\to u is replaced with an additional edge u\to v forming a multiple edge. 4 Local strong h -homotopy We now develop new quiver homotopy theory that will be applied in the subsequent sections of this work. Motivated by the constrained conditions under which strong h -homotopies can be obtained [ 33 , Lemma 3.7] , the central construction of this section is a weaker notion called local strong h -homotopy under which H_{*}^{\Delta,h} remains invariant. This result greatly improves upon the H_{*}^{\Delta,h} strong h -homotopy invariance provided by [ 33 , Theorem 5.8] and serves as a key tool for revealing further properties of H_{*}^{\Delta,h} . Throughout the section assume that G is a quiver and n\geq 0 an integer, unless otherwise stated. We are not aware of the next definition having been made previously. However, the definition would be a special case of the homotopy mapping cylinder construction [ 18 ] if we were making use of the box products of digraphs rather the strong product of quivers. Definition 7 . Let G be a quiver and G^{\prime} a full subquiver. Then define the strong mapping cylinder M_{G^{\prime}\hookrightarrow G}^{S} of the inclusion G^{\prime}\hookrightarrow G to be the quiver with vertices V_{M_{G^{\prime}\hookrightarrow G}^{S}}=V_{G}\times\{0\}\cup V_{G^{\prime}}\times\{1\} and edges \displaystyle E_{M_{G^{\prime}\hookrightarrow G}^{S}}=(E_{G}\times\{0\})\cup E_{G^{\prime}\boxtimes I}\cup E(G,G^{\prime}) where \displaystyle E(G,G^{\prime})=\{e_{e_{G}}\>|\>e_{G}\in E_{G}\setminus E_{G^{\prime}},\;s(e_{G})\in V_{G^{\prime}}\;\text{and}\;t(e_{G})\notin V_{G^{\prime}},\; \displaystyle\text{or}\;s(e_{G})\notin V_{G^{\prime}}\;\text{and}\;t(e_{G})\in V_{G^{\prime}}\} such that s(e_{e_{G}})=\begin{cases}(s(e_{G}),1)\ &\text{if}\;s(e_{G})\in V_{G^{\prime}}\\ (s(e_{G}),0)&\text{otherwise,}\end{cases}\;\;\;t(e_{e_{G}})=\begin{cases}(t(e_{G}),1)\ &\text{if}\;t(e_{G})\in V_{G^{\prime}}\\ (t(e_{G}),0)&\text{otherwise.}\end{cases} Similarly to equation ( 8 ), we can define two natural inclusions into the strong mapping cylinder. The first being \displaystyle i_{0}\colon G\to V_{M_{G^{\prime}\hookrightarrow G}^{S}}\;\;\;\text{given by}\;\;\;v\mapsto(0,v),\;e_{G}\mapsto(0,e_{G}) where v\in V_{G} and e_{G}\in E_{G} and the second i_{1}\colon G\to M_{G^{\prime}\hookrightarrow G}^{S} given by \displaystyle i_{1}(v)= \displaystyle\begin{cases}(v,1)\ &\text{if}\;v\in V_{G^{\prime}}\\ (v,0)&\text{otherwise,}\end{cases} \displaystyle i_{1}(e_{G})= \displaystyle\begin{cases}(e_{G},1)\ &\text{if}\;e_{G}\in E_{G^{\prime}}\\ (e_{G},0)\ &\text{if}\;s(e_{G})\notin V_{G^{\prime}}\>\text{and}\>t(e_{G})\notin V_{G^{\prime}}\\ e_{e_{G}}&\text{otherwise.}\end{cases} The construction of M_{G^{\prime}\hookrightarrow G}^{S} is demonstrated in following example. Example 4.1 . Let G be the quiver u_{1} u_{2} u_{3} u_{4} u_{5} u_{6} and G^{\prime} the full subquiver with v_{G^{\prime}}=\{u_{3},u_{4}\} . Then M^{S}_{G^{\prime}\hookrightarrow G} is the following quiver. (u_{1},0) (u_{2},0) (u_{3},0) (u_{4},0) (u_{5},0) (u_{6},0) (u_{3},1) (u_{4},1) There is a retraction r\colon M_{G^{\prime}\hookrightarrow G}^{S}\to M_{G^{\prime}\hookrightarrow G}^{S} onto G , given by \displaystyle r((v,0)) \displaystyle=r((v,1))=(v,0), \displaystyle\text{and}\;r((e_{G},0)) \displaystyle=r((e_{G},1))=r((e_{G},0\to 1))=r(e_{e_{G}})=(e_{G},0) where v\in v_{G} and e_{G}\in E_{G} . However, as demonstrated by the next example, r\colon M_{G^{\prime}\hookrightarrow G}^{S}\to M_{G^{\prime}\hookrightarrow G}^{S} is not a strong h -retraction, which differs from the situation for mapping cylinders of topological spaces. Example 4.2 . Consider the line digraph I given in equation ( 7 ) with itself as a full subquiver, then M_{I\hookrightarrow I}^{S}\boxtimes I=(I\boxtimes I)\boxtimes I. However, there are no digraph homomorphisms from M_{I\hookrightarrow I}^{S}\boxtimes I to I as any digraph map is required to send at least one edge to a vertex. Therefore, r\colon M_{I\hookrightarrow I}^{S}\to M_{I\hookrightarrow I}^{S} is not a strong h -deformation retraction. More generally, it is shown [ 33 , Theorem 3.8] that a 1 -step strong h -homotopy can only exist when there is a directed closed walks contained in the quiver. We now provide a weaker notion of homotopy containing strong h -homotopy as a special case, under which we will show that H^{h,\Delta}_{*} remains invariant. Definition 8 . Two quiver maps f_{0},f_{1}\colon G_{1}\to G_{2} are 1 -step local strong h -homotopic if there is a full subquiver G^{\prime}_{1} of G_{1} and quiver homomorphism \displaystyle F\colon M^{S}_{G^{\prime}_{1}\hookrightarrow G_{1}}\to G_{2}\;\;\;\text{such that}\;\;\;F\circ i_{0}=f_{0},\;F\circ i_{1}=f_{1}. In which case we write f_{0}\simeq_{1}^{lSh}f_{1} . Quiver maps f and g are local strong h -homotopic if they are related by the equivalence relation generated by 1 -step local strong h -homotopyies. In which case we write f\simeq^{lSh}g . Quivers G_{1} , G_{2} are locally strong h -homotopy equivalent if there are quiver maps \displaystyle f\colon G_{1}\to G_{2},\;h\colon G_{2}\to G_{1}\;\;\;\text{such that}\;\;\;h\circ f\simeq^{lSh}\text{id}_{G_{1}},\;f\circ h\simeq^{lSh}\text{id}_{G_{2}}. A quiver is called locally strongly h -contactable if it is strong homotopy equivalent to a quiver consisting of a single vertex and no edges. We note that to fully justify the definition of local strong homotopies we must check that it is compatible with composition of homotopy classes. This can be achieved similarly to the construction in Remark 2.1 . Since strong h -homotopy is a special case of local strong h -homotopy, the next theorem is a stronger version of the homotopy invariance of H^{h,\Delta}_{*} with respect to strong h -homotopy from [ 33 , Theorem 5.17] . Theorem 4.1 . Let \phi,\varphi\colon G_{1}\to G_{2} be quiver homomorphisms such that \phi\simeq^{lSh}\varphi , then \phi_{*}=\varphi_{*}\colon H^{h,\Delta}_{*}(G_{1})\to H^{h,\Delta}_{*}(G_{2}). Proof. By the construction of local strong homotopies, it is sufficient to consider only 1 -step local strong h -homotopies. Therefore, assume \phi\simeq_{1}^{lSh}\varphi . This implies that there exists an full subquiver G^{\prime}_{1} of G_{1} and a quiver homomorphism F\colon M^{S}_{G_{1}^{\prime}\hookrightarrow G_{1}}\to G_{2} such that F\circ i_{0}=\phi , F\circ i_{1}=\varphi . Let f\colon\Delta^{n}\to G_{1} be a singular n -simplex homomorphism. When f(i)\in V_{G_{1}}\setminus V_{G^{\prime}_{1}} for each i=0,\dots,n , then by construction of M^{S}_{G_{1}^{\prime}\hookrightarrow G_{1}} and F we have \phi_{\#}\circ f=\varphi_{\#}\circ f . Alternatively, when f(i)\in V_{G^{\prime}_{1}} for each i=0,\dots,n , then \phi_{\#}\circ f and \varphi_{\#}\circ f are related by the boundary of a quiver construction analogous to the classical prism operator [ 24 , proof of Corollary 2.11] . More generally, we require the construction of a hybrid prism operator transitioning between the two cases above, which we obtain as follows. Let f_{M^{S}_{G^{\prime}_{1}\hookrightarrow G_{1}}} be the subquiver of M^{S}_{G_{1}^{\prime}\hookrightarrow G_{1}} on vertex set \{(f(i),0)\>|\>i=0,\dots,n\}\cup\{(f(i),1)\>|\>i=0,\dots,n\>\text{with}\>f(i)\in G^{\prime}_{1}\} and with edges \displaystyle(f(j\to k),0); \displaystyle(f(j\to k),1),\;(f(j\to k),0\to 1),\;\text{or}\;(f(i),0\to 1)\;\text{when}\;f(i),f(j),f(k)\in G^{\prime}_{1}; \displaystyle e_{f(j\to k)}\;\text{when}\;f(j)\in G^{\prime}_{1}\;\text{and}\;f(k)\notin G^{\prime}_{1},\;\text{or}\;f(j)\notin G^{\prime}_{1}\;\text{and}\;f(k)\in G^{\prime}_{1} for i=0,\dots,n and integers 0\leq jt\leq i\\ (f(t),\mathcal{I}_{f}(t))&\text{otherwise}\end{cases} for t=0,\dots,n+1 and on edges by g_{i}(j\to k)=\begin{cases}(f(j\to k),0)&\text{if}\>i\geq k\>\text{or}\>\mathcal{I}_{f}(j)=\mathcal{I}_{f}(k)=0\\ (f(j\to k),1)&\text{if}\>i\text{and}\>\mathcal{I}_{f}(j)=\mathcal{I}_{f}(k)=1\\ (f(j\to k),0\to 1)&\text{if}\>j\leq i\text{and}\>\mathcal{I}_{f}(j)=\mathcal{I}_{f}(k)=1\\ e_{f(j\to k)}&\text{otherwise}\end{cases} for integers 0\leq j\text{and}\;\>t(e)=v\;\;\;\text{or}\;\;\;s(e)=v\;\>\text{and}\;\>t(e)=u the edge e is not a multiple edge, and either (i) u has a loop or (ii) e is not a double edge. A subquiver G^{\prime} obtained from G by removing one or more degenerate loops from E_{G} is called a loop contraction of G . Definition 10 . A pair of quivers G_{1} , G_{2} are 1 -step weak local strong h -homotopy equivalent if (1) there are quiver homomorphisms f\colon G_{1}\to G_{2} , g\colon G_{2}\to G_{1} such that g\circ f\simeq_{1}^{lSh}\text{id}_{G_{1}} , f\circ g\simeq_{1}^{lSh}\text{id}_{G_{2}} ; (2) G_{1} is a loop contraction of G_{2} ; (3) or G_{2} is a loop contraction of G_{1} . In which case we write G_{1}\simeq^{wlSh}_{1}G_{2} . Quivers G_{1} and G_{2} are weak local strong h -homotopic if they are related by the equivalence relation generated by (1), (2) and (3) above, in which case we write G_{1}\simeq^{wlSh}G_{2} . A quiver is called weak local strongly h -contactable if it is weak local strong h -homotopy equivalent to a quiver on one vertex with no edges. We will prove at the end of Section 6.2 that H_{*}^{\Delta} is invariant under weak local strong h -homotopy of quivers. To complete the section we consider some partial results in this direction. In order make these statements, we introduce the following terminology. Let G be a quiver and l a loop in G and recall that we assume all coefficients lie in a commutative ring R . Then consider the C_{*}^{\Delta,h}(G) submodules, \displaystyle C_{*}^{\Delta,h,l}(G)=R[\{ \displaystyle f\colon\Delta^{n}\to G\>|\>f\>\text{is a homomorphism},\>f(a\to c)=l\>\text{for some}\>0\leq a\forall\>0\leq a\text{with}\>f(a\to c)=l\>\text{we have}\>f(b)=f(a)\}] and \displaystyle C_{*}^{\Delta,h,ld}(G)=R[\{ \displaystyle f\colon\Delta^{n}\to G\>|\>f\>\text{is a homomorphism},\>\text{and} \displaystyle\exists\>0\leq a\text{with}\>f(a\to c)=l\>\text{and}\>f(b)\neq f(a)\}]. Any singular simplex containing the loop l in its image lies in C_{*}^{\Delta,h,l}(G) or C_{*}^{\Delta,h,ld}(G) and C_{*}^{\Delta,h,l}(G)\cap C_{*}^{\Delta,h,ld}(G)=\emptyset . Crucially, when l is degenerate singular simplices containing the loop l in their image are precisely the generators of C_{*}^{\Delta,h}(G) that no longer exist after a loop contraction removing l . The key distinction between singular simplicies f\colon\Delta^{n}\to G generating C_{*}^{\Delta,h,l}(G) and C_{*}^{\Delta,h,ld}(G) is provided by the presence of a G subquiver of the form f(a)=f(c) f(b) l in the image of f when f\in C_{*}^{\Delta,h,ld}(G) . Such an image of a singular simplex homomorphism can only occur when l is incident to a double edge and in this case H_{*}^{\Delta,h}(G) can be altered after the removal of l if there is no loop at f(b) , as made explicit in Example 3.1 . The case of C_{*}^{\Delta,h,l}(G) alone is easier handle. To demonstrate this we make use of the following operations. Let f\colon\Delta^{n}\to G be a singular n -simplex homomorphism in C_{*}^{\Delta,h}(G) such that f(i) has a loop l for some i=0,\dots,n . Then define singular (n+1) -simplex s_{i,l}(f) in C_{n+1}^{\Delta,h}(G) , given on vertices by s_{i,l}(f)(t)=\begin{cases}f(t)&\text{if}\>t\leq i\\ f(i)&\text{if}\>t=i+1\\ f(t-1)&\text{otherwise}\end{cases} for t=0,\dots,n+1 and on edges by s_{i,l}(f)(j\to k)=\begin{cases}f(j\to k)&\text{if}\>k\leq i\\ f(j\to i)&\text{if}\>jk=i+1\\ l&\text{if}\>j=i,\>k=i+1\\ f(i\to k-1)&\text{if}\>j=i+1\\ f(j-1\to k-1)&\text{otherwise}\end{cases} for integers 0\leq jji=j\>\text{or}\>i+1=j\\ s_{i,l}(f\circ\delta_{j-1})&\text{if}\>j>i+1\end{cases} (10) for each j=0,\dots,n+1 . Lemma 4.2 . Let G be a quiver and l the unique loop in G at vertex v_{l}\in V_{G} with no multiple edges indecent to v_{l} . Then C_{*}^{\Delta,h,l}(G) is an acyclic sub-chain complex of C_{*}^{\Delta,h}(G) . Proof. Suppose that f is a singular n -simplex homomorphism in C_{n}^{\Delta,h,l}(G) . Then there are unique integers 0\leq a1 we also have that f\circ\delta_{i}\in C_{n-1}^{\Delta,h,l}(G) for i=a,\dots,c . If c-a=1 , then as there are no multiple edges incident to v_{l} f\circ\delta_{a}-f\circ\delta_{c}=0\in C_{n-1}^{\Delta,h,l}(G) . Therefore, C_{*}^{\Delta,h,l}(G) is a sub-chain subcomplex of C_{*}^{\Delta,h}(G) and it remains to show that C_{*}^{\Delta,h,l}(G) is acyclic. To this end, define h_{n}\colon C_{n}^{\Delta,h,l}(G)\to C_{n+1}^{\Delta,h,l}(G) by linearly extending h_{n}(f)=(-1)^{i}s_{i,l}(f) where f\colon\Delta^{n}\to G is a singular n -simplex homomorphism in C_{n}^{\Delta,h,l}(G) and i is the least integer from 0,\dots,n such that f(i)=v_{l} . Applying equation ( 10 ), for each singular n -simplex homomorphism f\in C_{n}^{\Delta,h,l}(G) we have (\partial_{n+1}\circ h_{n})(f)=(-1)^{i}\sum_{j=0}^{n+1}(-1)^{j}s_{i,l}(f)\circ\delta_{j}=(-1)^{i}\sum_{0\leq jv=m(n)\\ (v,m(n))&\text{otherwise}\end{cases} and \displaystyle F(e,0\to 1)=\begin{cases}m(n)&\text{if}\>s_{G}(e)=m(n)\\ (s(e),m(n))&\text{otherwise}\end{cases} where v\in V_{G} and e\in E_{G} . Therefore, r is a strong deformation retract of G onto vertex m(n) and G is strongly contactable as required. ∎ The next lemma determines an initial special case when C_{*}^{\Delta,h}(G) acyclic under certain assumptions. Lemma 4.4 . Let G be a quiver and h\colon\Delta^{n}\to G a homomorphism surjective on vertices such that h(n) has a loop. Then C_{*}^{\Delta,h}(G) is acyclic. Proof. With the modification that every edge that would have been sent to the vertex m(n) being instead sent to the loop at h(n) , the retraction r and strong homotopy given in the proof of Lemma 4.3 provide a strong h -deformation retraction to G subquiver G^{\prime} consisting of vertex h(n) and a loop at h(n) . Since C_{*}^{\Delta,h}(G^{\prime}) is acyclic, the strong homotopy invariance of H_{*}^{\Delta,h} from Theorem 2.3 now implies that C_{*}^{\Delta,h}(G) acyclic. ∎ We can now give a variation of the previous lemma under different assumptions. Lemma 4.5 . Let G be a quiver with no multiple edges, such that any double edge has a loop, there exists a digraph homomorphism h\colon\Delta^{n}\to G surjective on vertices, and for every edge of e\in E_{G} there exists some digraph homomorphism h_{e}\colon\Delta^{n}\to G surjective on vertices of G with e lying in its image. Then C_{*}^{\Delta,h}(G) is acyclic. Proof. When h(n) has a loop, G is acyclic by Lemma 4.4 . If h(n) has no loop and is not incident to a double edge in G , then let G^{\prime} be the quiver obtained from G by adding an additional loop l at h(n) . In this case, as h is surjective on vertices and there are no double edges indecent to h(n) , all edges incident to h(n) in G^{\prime} have target vertex h(n) . This implies that for any integer m\geq 0 and singular m -simplex homomorphism f\colon\Delta^{m}\to G^{\prime} , f(i)\neq h(n) for each i=0,\dots,m-1 . Hence, C_{*}^{\Delta,h,ld}(G^{\prime})=0 which implies C_{*}^{\Delta,h}(G^{\prime})=C_{*}^{\Delta,h}(G)+C_{*}^{\Delta,h,l}(G^{\prime}) as a chain complexes. Since G , hence, G^{\prime} contains no multiple edges, Lemma 4.2 implies that C_{*}^{\Delta,h,l}(G^{\prime}) is an acyclic summand of C_{*}^{\Delta,h}(G^{\prime}) . Therefore, H_{*}^{\Delta,h}(G)=H_{*}^{\Delta,h}(G^{\prime}) and C_{*}^{\Delta,h}(G) is acyclic as C_{*}^{\Delta,h}(G^{\prime}) is acyclic by Lemma 4.4 . The only remaining possibility is that h(n) has no loop and is incident to a double edge d such that s(d)=h(n) and t(d)\neq h(n) . Applying the conditions in the statement of the lemma, there exists a h_{d}\colon\Delta^{n}\to G surjective on vertices of G with d in its image. If h_{d}(n)=h(n) then as d lies in the image of h_{d} , there is an i=0,\dots,n-1 such that h_{d}(i)=s(d)=h(n) . This implies that h_{d}(i\to n) is a loop at h(n) , which contradicts the present assumption that h(n) has no loop. Therefore, h_{d}(n)\neq h(n) . By construction of singular simplex homomorphisms h and h_{d} there must be an edge from every element of V_{G}\setminus\{h(n)\} to h(n) and from every from every element of V_{G}\setminus\{h_{d}(n)\} to h_{d}(n) . This implies there is a double edge in G between h_{d}(n) and h(n) . Using the conditions of the lemma, since h(n) does not have loop h_{d}(n) must have a loop. Hence, C_{*}^{\Delta,h}(G) is acyclic by Lemma 4.4 with the singular simplex homomorphism h_{d} . ∎ 5 Computation of H^{\Delta,m}_{*} In this section, we provide an object-wise map from a quiver to a suitably simpler simplicial complex called the reduced directed flag complex, whose simplicial homology coincides with H^{\Delta,m}_{*} . In particular, this construction extends to a functor natural with recept to the isomorphism between homologies. The naturality of our result implies that persistent H^{\Delta,m}_{*} can be computed more efficiently using the persistent simplicial homology of the reduced directed flag complex. Throughout the section assume that G is a quiver and n\geq 0 is an integer, unless otherwise stated. Note also that some results in the section require the axiom of choice if the quiver G contains an infinite number of edges. 5.1 The reduced digraph of a quiver In this subsection, we show that the presence of loops and multiple edges does not effect the structure of H^{\Delta,m}_{*}(G) . More precisely, we show that the removal of loops and multiple edges realises a strong homotopy equivalence between any quiver and a digraph. The strong homotopy invariance of H_{*}^{\Delta,m} shown in [ 33 ] , then provides us with our desired result. Definition 11 . Let G be a quiver. Define the reduced digraph \bar{\mathcal{R}}(G) of the quiver G to have vertices V_{\bar{\mathcal{R}}(G)}=V_{G} and edges E_{\bar{\mathcal{R}}(G)}=\{\bar{\mathcal{R}}(G,u,v)\>|\>\exists\>e\in E_{G}\>\text{with}\>s(e)=u,\>t(e)=v\>\text{and}\>u\neq v\} where s(\bar{\mathcal{R}}(G,u,v))=u\;\;\;\text{and}\;\;\;t(\bar{\mathcal{R}}(G,u,v))=v. The reduced digraph construction extends to a functor \bar{\mathcal{R}}\colon\textbf{Quiv}_{m}\to\textbf{Quiv}_{m} by setting \bar{\mathcal{R}}(m)(v)=m(v)\;\;\;\text{and}\;\;\;\bar{\mathcal{R}}(m)(\bar{\mathcal{R}}(G,s(e),t(e)))=(\bar{\mathcal{R}}(G^{\prime},s(m(e)),t(m(e)))) (11) for quiver map m\colon G\to G^{\prime} , v\in V_{G} and e\in E_{G} . A weaker version of the following proposition for quiver homotopies with respect to the box product rather than strong homotopies can be found in course notes of Muranov [ 43 , Lecture 7] , and is expected to appear in a textbook on the subject of digraph homotopy and homology. Proposition 5.1 . Quivers G and \bar{\mathcal{R}}(G) are strong homotopy equivalent. Proof. Define f\colon G\to\bar{\mathcal{R}}(G) by f(v)=v\;\;\;\text{and}\;\;\;f(e)=\begin{cases}s(e)&\text{if}\;s(e)=t(e)\\ \bar{\mathcal{R}}(G,s(e),t(e))&\text{otherwise}\end{cases} for any v\in V_{G} and e\in E_{G} . Given distinct u,v\in E_{G} such that there is an e\in E_{G} with s(e)=u and t(e)=v , choose some edge e_{u,v}\in\{e\in E_{G}\>|\>s(e)=u\>\text{and}\>t(e)=v\}. Define g\colon\bar{\mathcal{R}}(G)\to G by g(v)=v\;\;\;\text{and}\;\;\;g(\bar{\mathcal{R}}(G,u,v))=e_{u,v}. By construction of \bar{\mathcal{R}}(G) , both f and g are well defined and f\circ g=\text{id}_{\bar{\mathcal{R}}(G)} . Moreover, g\circ f\simeq_{1}^{S}\text{id}_{G} as we have well defined F\colon G\boxtimes I\to G given on edges by \displaystyle F((v,0))=F((v,1))=F((v,0\to 1))=v, \displaystyle F((e,0))=\begin{cases}s(e)&\text{if}\;s(e)=t(e)\\ e_{u,v}&\text{otherwise,}\end{cases} and \displaystyle F((e,1))=F((e,0\to 1))=e for all v\in V_{G} and e\in E_{G} . ∎ In fact we have proved that \bar{\mathcal{R}}(G) is a strong deformation retraction of G . Note however, that the proof of the lemma above does not hold in the category of quivers and quiver homomorphism \textbf{Quiv}_{h} , as we are always required to send at least one edge of the form (v,0\to 1)\in E_{G\boxtimes I} for v\in V_{G} to a vertex. Remark 5.1 . As stated prior to Proposition 5.1 , the cartesian or box product G\square G^{\prime} of quivers G and G^{\prime} can be obtained similarly to the strong box product in Definition 4 by restricting only to edges E_{G\square G^{\prime}}=E_{G}\times V_{G^{\prime}}\cup V_{G}\times E_{G^{\prime}} . In precisely the same way as for strong homotopies in Definition 5 , we obtain a notion of homotopy of digraphs and quivers using \square in place of \boxtimes . In particular, Proposition 5.1 is equally valid for quivers homotopies, as less conditions are required to be checked in the poof. The path homology of a digraph is known to be a homotopy invariant of digraphs [ 20 ] . Therefore, any extension of path homology that is a homotopy invariant of quivers when applied to a quiver G would coincide with the path homology of the reduced digraph \bar{\mathcal{R}}(G) , which is a digraph by construction. Proposition 5.1 , has the following immediate consequence. Corollary 5.2 . Let G be a quiver, then H_{*}^{\Delta,m}(G)=H_{*}^{\Delta,m}(\bar{\mathcal{R}}(G)) naturally with respect to induced maps. Proof. It is shown in [ 33 , Theorem 5.6] that H_{*}^{\Delta,m} is a strong homotopy invariant. Hence the isomorphism of homology follows from Proposition 5.1 . The naturality of this isomorphism can be obtained directly from the functorial construction of \bar{\mathcal{R}} in equation ( 11 ) and induced maps of H_{*}^{\Delta,m} given in equation ( 4 ). ∎ 5.2 The reduced directed flag complex Motivated by computation, we now extend the construction of the reduced digraph functor \bar{\mathcal{R}} from the previous subsection and construct for any quiver a simplicial complex with H_{*}^{\Delta,m} isomorphic to its simplicial homology. This construction is functorial, providing a functor we call the reduced direct flag complex \bar{\mathcal{F}}\colon\textbf{Quiv}_{m}\to\textbf{ASim} . Furthermore, \bar{\mathcal{F}} is natural with respect to the isomorphisms between homologies. Definition 12 . Let G be a quiver. Then the reduced directed flag complex , \bar{\mathcal{F}}(G) is the abstract simplicial complex on vertex set V_{G} , with an n -simplex on the set of vertices v_{0},\dots,v_{n}\in V_{G} if and only if these is an inclusion f\colon\Delta^{n}\to G such that v_{0},\dots,v_{n} lie in the image of f . We note that for digraphs the reduced directed flag complex also appears in [ 41 ] where it is called the directed Vietoris-Rips complex. In this case, the reduced directed flag complex is considered in a different theoretical context related to the homology of digraphs as pseudotopological spaces. However, analogously to the direction of this section, the directed Vietoris-Rips complex was shown to have homology coinciding with the homology of a digraph as a certain pseudotopological spaces. The following example demonstrates the difference between the two notions of flag complex that have been introduced so far in this work. Example 5.1 . The reduced directed flag complex \bar{\mathcal{F}}(G) is not the same as the directed flag complex \mathcal{F}(G) (Definition 3 ) whenever multiple or double edges are present. For example, if G is a quiver that contains two vertices and precisely one double edge or multiple edge then \text{F(x)}(G) is a circle while \bar{\mathcal{F}}(G) is an interval. Moreover, neither \bar{\mathcal{F}}(G) or \mathcal{F}(G) in general coincide with the classical flag complex on the undirected graph containing G . For example, the directed cycle digraph C with V_{C}=\{u,v,w\} and edges u v w provides \bar{\mathcal{F}}(C) and \mathcal{F}(C) that contain no 2 -simplices, while the classical flag complex contains a 2 -simplex. Given a map of quivers \phi\colon G_{1}\to G_{2} we obtain an induced map \bar{\mathcal{F}}(\phi)\colon\bar{\mathcal{F}}(G_{1})\to\bar{\mathcal{F}}(G_{2}) given by \bar{\mathcal{F}}(\phi)(\{v_{0},\dots,v_{n}\})=\{\phi(v_{0}),\dots,\phi(v_{n})\} (12) for each simplex \{v_{0},\dots,v_{n}\} in \bar{\mathcal{F}}(G_{1}) . The induced maps make \bar{\mathcal{F}}\colon\textbf{Quiv}_{m}\to\textbf{ASim} a functor, as for any quiver maps \phi\colon G_{1}\to G_{2} and \varphi\colon G_{2}\to G_{3} we have \bar{\mathcal{F}}(\varphi\circ\phi)(\{v_{0},\dots,v_{n}\})=\{\varphi\circ\phi(v_{0}),\dots,\varphi\circ\phi(v_{n})\}=(\bar{\mathcal{F}}(\varphi)\circ\bar{\mathcal{F}}(\phi))(\{v_{0},\dots,v_{n}\}) (13) for each simplex \{v_{0},\dots,v_{n}\} in \bar{\mathcal{F}}(G_{1}) . Every map of abstract simplicial complexes is by construction provided by a function on its vertices, while a map of quivers is determined also by a function on its edges. This implies that \bar{\mathcal{F}} cannot be a faithful functor. However, \bar{\mathcal{F}} is faithful when restricted to digraphs, in which case there is a unique arrow in either direction between any pair of vertices. Furthermore, \bar{\mathcal{F}} is not a full functor even when restricted to digraphs. Consider for example the interval digraph I with a single arrow between two vertices. Then a digraph map switching the order of the vertices does not exist. However, a simplicial map switching the vertices of \bar{\mathcal{F}}(I) is well defined. The next example demonstrates that \bar{\mathcal{F}} is not surjective on objects, though \bar{\mathcal{F}} can realise any simplicial complex up to homotopy type. Example 5.2 . Consider the abstract simplicial complex (V,S) on vertex set V=\{v_{0},v_{1},v_{2},v_{3}\} with S=\{\{v_{0},v_{1}\}\{v_{0},v_{2}\},\{v_{0},v_{3}\}\{v_{1},v_{2}\},\{v_{1},v_{3}\},\{v_{2},v_{3}\}\} That is the 1 -skeleton of a 2 -simplex. We now attempt to construct a quiver G such that \bar{\mathcal{F}} is (V,S) . Since (V,S) contains no 2 simplices, all triples of vertices contained in V_{G}=V must be directed cycles (see the second part of Example 5.1 ), otherwise there is an inclusion of \Delta^{2} into G . However, once any two triples of vertices in V_{G} are assigned arrows in the form of directed cycles the remaining two tipples are forced to be non-cyclic. Moreover, any additional edges added to G only increases the number of possible inclusions \colon\Delta^{2}\to G . Therefore, there does not exist a quiver G such that \bar{\mathcal{F}}(G)=(V,S) . However, given an arbitrary abstract simplicial complex (V,S) , we can construct a quiver G such that \bar{\mathcal{F}}(G) and (V,S) have the same homotopy type as follows. Using a simplicial construction similar to the cubical one from [ 20 ] , we will consider the Barycentric subdivision B of (V,S) and form a digraph G_{B} such that \bar{\mathcal{F}}(G_{B})=B . Recall that the Barycentric subdivision B of (V,S) is the flag complex of the graphs whose vertices are S and with an edge between s_{1},s_{2}\in S if and only if s_{1}\subset s_{2} or s_{2}\subset s_{1} . In particular, B has the same homotopy type as (V,S) . The digraph G_{B} is obtained by setting V_{G_{B}} as the vertices of B and E_{G_{B}} to be in bijection with the edges of B assigning direction in each case from the smaller element of S to the larger element of S . In this case, inclusions f\colon\Delta^{n}\to G_{B} are in one to one correspondence with n -simplices of B . Where the bijection is given by assigning to f\colon\Delta^{n}\to G_{B} the simplex i(n) . Therefore, \bar{\mathcal{F}}(G_{B})=B by definition. The following theorem and subsequent corollary demonstrate that the computation of H_{*}^{\Delta,m}(G) can be reduced to computing the homology of the abstract simplicial complex \bar{\mathcal{F}}(\bar{\mathcal{R}}(G)) . In particular, for a finite quiver the size of a basis of C_{*}(\bar{\mathcal{F}}(\bar{\mathcal{R}}(G))) is in general far smaller than for C_{*}^{\Delta,m}(G) , leading to a considerably more efficient algorithm in Appendix A.5 . Theorem 5.3 . Let G be a quiver, then H_{*}^{\Delta,m}(G)=H_{*}(\bar{\mathcal{F}}(G)). Moreover, the above isomorphism on homology is natural with respect to \bar{\mathcal{F}} . Proof. Fix a total order on V_{G} . In particular, the homology of the abstract simplicial complex \bar{\mathcal{F}}(G) is independent of this choice. We will show that there are chain maps g\colon C^{\Delta,m}_{*}(G)\to C_{*}(\bar{\mathcal{F}}(G))\;\;\;\text{and}\;\;\;h\colon C_{*}(\bar{\mathcal{F}}(G))\to C^{\Delta,m}_{*}(G) that induce a chain homotopy equivalence between chain complexes C^{\Delta,m}_{*} and C_{*}(\bar{\mathcal{F}}(G)) . First we define the chain map g . Given distinct v_{0},\dots,v_{n}\in V_{G} , define \sigma_{v_{0},\dots,v_{n}}\in\Sigma_{n+1} to be the permutation in the symmetric group \Sigma_{n} on n elements such that v_{\sigma(0)},\dots,v_{\sigma(n)} is ordered under the total order chosen on V_{G} . Write also \text{sgn}(\sigma_{v_{0},\dots,v_{n}}) for the usual sign of the permutation. We may now define the function g\colon C^{\Delta,m}_{*}(G)\to C_{*}(\bar{\mathcal{F}}(G)) by linearly extending \displaystyle g(f)=\begin{cases}\text{sgn}(\sigma_{f(0),\dots,f(n)})\{f(0),\dots,f(n)\}&\text{if}\>f(0),\dots,f(n)\>\text{are distinct}\\ 0&\text{otherwise}\end{cases} where f\colon\Delta^{n}\to G is a singular n -simplex. By construction, (g\circ\partial_{n})(f)=\sum_{i=0}^{n}\alpha_{i}\{f(0),\dots,f(n)\}\setminus\{f(i)\}\;\;\;\text{and}\;\;\;(\partial_{n}\circ g)(f)=\sum_{i=0}^{n}\beta_{i}\{f(0),\dots,f(n)\}\setminus\{f(i)\} for some \alpha_{0},\dots,\alpha_{n},\beta_{0},\dots,\beta_{n}\in\{1,-1\} . Therefore, to show that g is a chain map it sufficient to check that \alpha_{i}=\beta_{i} for i=0,\dots,n . Furthermore, since \sigma_{f(0),\dots,f(n)}\in\Sigma_{n} is a composition of transpositions, it is sufficient to check the case when \sigma_{f(0),\dots,f(n)} is a transposition. Fix i=0,\dots,n . When \sigma_{f(0),\dots,f(n)} exchanges two elements not equal to f(i) , then \alpha_{i}=\beta_{i}=(-1)^{i+1} . Otherwise, \sigma_{f(0),\dots,f(n)} exchanges f(i) with f(j) , for some j=0,\dots,n and j\neq i . If ji , the argument is the same as above except that equation ( 14 ) is replaced with \sigma_{f(0),\dots,\widehat{f(i)},\dots,f(n)}=(f(i),f(i+1))\circ\cdots\circ(f(j-2),f(j-1)) which has sign \text{sgn}(\sigma_{f(0),\dots,\widehat{f(i)},\dots,f(n)})=(-1)^{j-i-1} . Therefore, g is a chain map. Next we define the chain map h . Consider distinct v_{0},\dots,v_{n}\in V_{G} such that there exists a singular simplex f\colon\Delta^{n}\to G with \{v_{0},\dots,v_{n}\} the image of f on vertices. Then for each such possible vertex set \{v_{0},\dots,v_{n}\} above, make a choice of singular simplex s_{v_{0},\dots,v_{n}}\colon\Delta^{n}\to G. Define the function h\colon C_{*}(\bar{\mathcal{F}}(G))\to C^{\Delta,m}_{*}(G) by linearly extending \displaystyle h(\{v_{0},\dots,v_{n}\})=\begin{cases}\text{sgn}(\sigma_{s_{v_{0},\dots,v_{n}}(0),\dots,s_{v_{0},\dots,v_{n}}(n)})s_{v_{0},\dots,v_{n}}&\begin{aligned} &\text{if}\>\exists\>f\colon\Delta^{n}\to G\>\text{with}\\ &\text{image}\>\{v_{0},\dots,v_{n}\}\\ &\text{on vertices}\end{aligned}\\ 0&\text{otherwise}.\end{cases} The function h can be shown to be a chain map by a similar argument used to show g was a chain map above. In addition, by construction, g\circ h=\text{id}_{C_{*}(\mathcal{F}(G))} . Therefore, it remains to show that h\circ g is chain homotopic to the identity on C^{\Delta,m}_{*}(G) . Given v_{0},\dots,v_{n}\in V_{G} , denote by G_{v_{0},\dots,v_{n}} the unique full subquiver of G containing vertices v_{0},\dots,v_{n} . To each singular simplex f\colon\Delta^{n}\to G associate the sub-chain complex C^{\Delta,m}_{*}(G_{f(0),\dots,f(n)}) . By Lemma 4.3 each G_{f(0),\dots,f(n)} is strongly contractible. Hence, using the strong homotopy invariance of H_{*}^{\Delta,m} from Theorem 2.3 , the chain complexes C^{\Delta,m}_{*}(G_{f(0),\dots,f(n)}) is acyclic and provides us with an acyclic carrier \varphi on the basis of singular simplices in C^{\Delta,m}_{*}(G) . Moreover, both h\circ g and \text{id}_{C^{\Delta,m}_{*}(G)} are carried by \varphi . Therefore, h\circ g and \text{id}_{C^{\Delta,m}_{*}(G)} are chain homotopic by the acyclic carrier theorem (Theorem 2.2 ). It remains to check that naturality of the isomorphism on homologies. To this end, let \phi\colon G_{1}\to G_{2} be a digraph map. Using the functoriality of H^{\Delta,m}_{*} from Proposition 2.1 , the map \phi induces a chain map \phi_{\#}\colon C^{\Delta,m}_{*}(G_{1})\to C^{\Delta,m}_{*}(G_{2}) as in equation ( 4 ) by linearly extending \phi_{\#}(f)=f\circ\phi for each singular simplex f\colon\Delta^{n}\to G . Similarly, the simplicial map F(\phi) defined in equation ( 12 ) induces the map F(\phi)_{\#}\colon C_{*}(F(G_{1}))\to C_{*}(F(G_{2})) by linearly extending \mathcal{F}(\phi)_{\#}(\{v_{0},\dots,v_{n}\})=(\{\phi(v_{0}),\dots,\phi(v_{n})\}) for each simplex \{v_{0},\dots,v_{n}\} in abstract simplicial complex F(G_{1}) . We now observe that the following diagram of chain maps commutes \textstyle{C_{*}^{\Delta,m}(G_{1})\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{g} \scriptstyle{\phi_{\#}} \textstyle{C_{*}^{\Delta,m}(G_{2})\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{g} \textstyle{C_{*}(F(G_{1}))\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{F(\phi)_{\#}} \textstyle{C_{*}(F(G_{2}))} as it commutes for each singular n -simplex, with both (g\circ\phi_{\#})(f) and (F(\phi)_{\#}\circ g)(f) being equal to \text{sgn}(\sigma_{(\phi\circ f)(0),\dots,(\phi\circ f)(n)})\{(\phi\circ f)(0),\dots,(\phi\circ f)(n)\} when (\phi\circ f)(0),\dots,(\phi\circ f)(n) are distinct and 0 otherwise. Therefore, the naturality condition follows as we have already shown that the chain map g induces the isomorphism on homology. ∎ 6 Computation of H^{\Delta,h}_{*} In this section we provide object-wise maps from a quiver to a suitably simpler \Delta -set called the partial directed flag complex, whose simplicial homology coincides with H^{\Delta,h}_{*} . In particular, the object level construction lifts to a natural functorial construction on chains, and hence, to a natural functorial construction on homology. The naturality of our result implies persistent H^{\Delta,h}_{*} can be computed more efficiently using the homology of the partial directed flag complex. Furthermore, using the partial directed flag complex, we are able to complete a proof of the weak local strong homotopy invariance of H^{\Delta,h}_{*} on quivers. Throughout the section assume that G is a quiver and n\geq 0 is an integer unless otherwise stated. Note that some results in this section require the axiom of choice if the quiver G contains an infinite number of edges. 6.1 The partially reduced quiver In this subsection, we provide the construction of the partially reduced quiver, which transitions locally between the reduced digraph described from Section 5.1 and the original quiver depending on the presence of loops. The partially reduced quiver is local strong homotopy equivalent to the quiver itself. Therefore, combined with the local strong homotopy invariance of H_{*}^{\Delta,h} shown in Section 4 we obtain a general simplification for the computation of H_{*}^{\Delta,h} that we extend in the next subsection. Definition 13 . Let G be a quiver. Then define the partially reduced quiver \tilde{\mathcal{R}}(G) to be the quiver with vertices V_{\tilde{\mathcal{R}}(G)}=V_{G} and edges E_{\tilde{\mathcal{R}}(G)}=E^{1}_{\tilde{\mathcal{R}}(G)}\cup E^{2}_{\tilde{\mathcal{R}}(G)} where \displaystyle E^{1}_{\tilde{\mathcal{R}}(G)} \displaystyle=\{e_{G}\in E_{G}\>|\>e_{G}\>\text{does not have a loop}\} \displaystyle E^{2}_{\tilde{\mathcal{R}}(G)} \displaystyle=\{\tilde{\mathcal{R}}(G,u,v)\>|\>u,v\in V_{G}\>\text{satisfying}\>\exists\>e\in E_{G}\>\text{with a loop such that}\>s(e)=u\>\text{and}\>t(e)=v\} (15) with \displaystyle s_{\tilde{\mathcal{R}}(G)}(e_{G})=s_{G}(e_{G}),\;t_{\tilde{\mathcal{R}}(G)}(e_{G})=t_{G}(e_{G}),\;s_{\tilde{\mathcal{R}}(G)}(\tilde{\mathcal{R}}(G,u,v))=u\;\text{and}\;t_{\tilde{\mathcal{R}}(G)}(\tilde{\mathcal{R}}(G,u,v))=v. In brief, the partially reduced quiver collapses all multiple edges with a loop to a single edge. The partially reduced quiver extends to a functor \tilde{\mathcal{R}}\colon\textbf{Quiv}_{h}\to\textbf{Quiv}_{h} by setting \displaystyle\tilde{\mathcal{R}}(h)(v)=h(v), \displaystyle\tilde{\mathcal{R}}(h)(e_{G})=\begin{cases}h(e_{G})&\text{if}\>h(e_{G})\>\text{has no loops}\\ \tilde{\mathcal{R}}(G^{\prime},s(e_{G}),t(e_{G}))&\text{otherwise,}\end{cases} (16) and \displaystyle\;\tilde{\mathcal{R}}(h)(\tilde{\mathcal{R}}(G,u,v))=\tilde{\mathcal{R}}(G^{\prime},h(u),h(v)) for quiver homomorphism h\colon G\to G^{\prime} , u,v\in V_{G} , e_{G}\in E^{1}_{\tilde{\mathcal{R}}(G)} , and \tilde{\mathcal{R}}(G,u,v)\in E^{2}_{\tilde{\mathcal{R}}(G)} . This is well defined as homomorphism h must send loops to loops. We now consider the local strong homotopy class of the partially reduced quivers. Theorem 6.1 . The quivers G and \tilde{\mathcal{R}}(G) are local strong h -homotopy equivalent. Proof. Define homomorphism f\colon G\to\tilde{\mathcal{R}}(G) by f(v)=v\;\;\;\text{and}\;\;\;f(e)=\begin{cases}e&\text{if}\;e\in E^{1}_{\tilde{\mathcal{R}}(G)}\\ \tilde{\mathcal{R}}(G,s_{G}(e),t_{G}(e))&\text{otherwise}\end{cases} which is well defined by the construction of E^{1}_{\tilde{\mathcal{R}}(G)} , E^{2}_{\tilde{\mathcal{R}}(G)} and \tilde{\mathcal{R}}(G,s_{G}(e),t_{G}(e)) in equation ( 13 ). Given u,v\in E_{G} such that there exists e\in E_{G} with a loop satisfying s_{G}(e)=u and t_{G}(e)=v , choose a unique element e_{u,v}\in\{e\in E_{G}\>|\>s(e)=u\>\text{and}\>t(e)=v\}. Define g\colon\tilde{\mathcal{R}}(G)\to G by g(v)=v,\;\;\;g(e_{G})=e_{G}\;\;\;\text{and}\;\;\;g(\tilde{\mathcal{R}}(G,u,v))=e_{u,v}. Homomorphism g is also well defined by the construction of E^{1}_{\tilde{\mathcal{R}}(G)} and E^{2}_{\tilde{\mathcal{R}}(G)} in equation ( 13 ). In particular, f\circ g=\text{id}_{\tilde{\mathcal{R}}(G)} . Let G^{\prime} be the full subquiver of G on vertices V_{G^{\prime}}=\{v\in V_{G}\>|\>v\>\text{has a loop}\} . Then define F\colon M^{S}_{G^{\prime}\hookrightarrow G}\to G given by \displaystyle F((v,0))=v,\;F((v^{\prime},1))=v^{\prime},\;F((e,0))=e,\;F((v^{\prime},0\to 1))=e_{v^{\prime},v^{\prime}}, \displaystyle F((e^{\prime},1))=F((e^{\prime},0\to 1))=e_{s(e^{\prime}),t(e^{\prime})},\;\text{and}\;F(e^{\prime\prime})=e_{s(e^{\prime\prime}),t(e^{\prime\prime})} for v\in V_{G} , v^{\prime}\in V_{G^{\prime}} , e\in E_{G} , e^{\prime}\in E_{G^{\prime}} , and e^{\prime\prime}\in E(G,G^{\prime}) as given in Definition 7 of the strong mapping cylinder M^{S}_{G^{\prime}\hookrightarrow G} of G and G^{\prime} . By construction, the homomorphism F is a well defend 1 -step local strong h -homotopy between \text{id}_{G} and g\circ f . Therefore, \text{id}_{G}\simeq^{lSh}_{1}g\circ h which completes the proof. ∎ In fact we have proved that \tilde{\mathcal{R}}(G) is a local strong deformation retraction of G and we obtain the following corollary. Corollary 6.2 . Let G be a quiver, then H_{*}^{\Delta,h}(G)=H_{*}^{\Delta,h}(\tilde{\mathcal{R}}(G)) naturally with respect to induced maps. Proof. The corollary follows from Theorem 6.1 and the local strong h -homotopy invariance of H_{*}^{\Delta,h} shown in Theorem 4.1 . The naturality of this isomorphism can be obtained directly from the construction of \tilde{\mathcal{R}} as a functor in equation ( 6.1 ) and induced maps for H_{*}^{\Delta,h} given in equation ( 4 ). ∎ 6.2 The partial directed flag complex Motivated by computation, we now extend the construction of the partially reduced quiver functor \tilde{\mathcal{R}} from the previous subsection and construct for any quiver G a \Delta -set called the partial directed flag complex with homology isomorphic to H^{\Delta,h}_{*}(G) . While the partial directed flag complex is not a functor, the composition with the chain functors becomes functorial. When the quiver G has no loops the partial directed flag complex coincides with \mathcal{F}(G) . If every vertex of G has a loop and a total order on V_{G} is selected, the partial directed flag complex coincides with \bar{\mathcal{F}}(G) as a \Delta -set. The construction of the partial directed flag complex allows us to complete the section by showing that H^{\Delta,h}_{*} is invariant under the weak local strong homotopy of quivers given in Definition 10 . The structure of the subsection is similar to Section 5.2 , though the construction of the reduced directed flag complex is considerably more detailed. We first set out the following terminology. Definition 14 . Let G be a quiver. A total order < on V_{G} is called loop maximal if u,v\in V_{G} such that u has a loop and v has no loop, then v|\>e_{G}\in E_{G}\>\text{has no loop}\} \displaystyle E^{2}_{\tilde{\mathcal{F}}_{<}(G)} \displaystyle=\{\{\tilde{\mathcal{F}}_{<}(u,v)\}\>|\>u,v\in V_{G},\>u\exists\>e\in E_{G}\>\text{with}\> (17) \displaystyle\;\;\;\;\;\;\;\;s(e)=u,\>t(e)=v\;\text{or}\>s(e)=v,\>t(e)=u,\>\text{and $e$ has a loop}\} with face maps given by \displaystyle d^{0}_{0}(\{e_{G}\})=t(e_{G}),\>d^{0}_{1}(\{e_{G}\})=s(e_{G}), and \displaystyle d^{0}_{0}(\{\tilde{\mathcal{F}}_{<}(u,v)\})=v,\>d^{0}_{1}(\{\tilde{\mathcal{F}}_{<}(u,v)\})=u. It can be useful to view the construction of the 1 -skeleton above as providing a function \tilde{\mathcal{F}}_{<}^{1}(G)\colon E_{G}\to\tilde{\mathcal{F}}_{<}(G)_{1} given by \tilde{\mathcal{F}}_{<}^{1}(G)(e)=\begin{cases}\{\tilde{\mathcal{F}}_{<}(s(e),t(e))\}&\text{if}\>e\>\text{has a loop and}\>s(e)e\>\text{has a loop and}\>t(e)|\>f\colon\Delta^{n}\to G\>\text{is an inclusion}\} (18) with face maps given by d^{n-1}_{k}\left(\{\alpha_{e_{i,j}}\}_{0\leq its(\phi(e_{i,j}))\neq t(\phi(e_{i,j}))\\ &\text{for each}\>0\leq i\{\tilde{\mathcal{F}}_{<_{G_{2}}}(\phi(u),\phi(v))\}&\text{if}\>\phi(u)<_{G_{2}}\phi(v)\\ -\{\tilde{\mathcal{F}}_{<_{G_{2}}}(\phi(v),\phi(u))\}&\text{otherwise}\end{cases} (21) with an analogous expressions for \tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\varphi)(\{\alpha_{\phi(e)}\}) . From these possibilities, we can conclude that \displaystyle\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\varphi\circ\phi)(\{\alpha_{e}\}) \displaystyle=\begin{cases}\;\;\>\{\tilde{\mathcal{F}}_{<_{G_{2}}}((\varphi\circ\phi)(u),(\varphi\circ\phi)(v))\}&\text{if}\>(\varphi\circ\phi)(u)<_{G_{3}}(\varphi\circ\phi)(v)\\ -\{\tilde{\mathcal{F}}_{<_{G_{2}}}((\varphi\circ\phi)(v),(\varphi\circ\phi)(u))\}&\text{otherwise}\end{cases} \displaystyle=\left(\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\varphi)\circ\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\phi)\right)(\{\alpha_{e}\}). where in the first case above either both \text{sgn}(\{\alpha_{e}\},\phi)=\text{sgn}(\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\phi)(\{\alpha_{e}\}),\varphi)=\pm 1 and in the second case \text{sgn}(\{\alpha_{e}\},\phi) and \text{sgn}(\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\phi)(\{\alpha_{e}\}),\varphi) have opposite signs. The remaining possibility is that e has no loop and \{\alpha_{e}\}=\{e\} . In this situation, we have \tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\phi)(\{\alpha_{e}\})=\begin{cases}\;\;\>\{\tilde{\mathcal{F}}_{<_{G_{2}}}(\phi(s(e_{G_{1}})),\phi(t(e_{G_{1}})))\}&\begin{aligned} &\text{if}\>\phi(e)\>\text{has a loop and}\\ &\{\phi(s(e))<_{G_{2}}\phi(t(e))\}\end{aligned}\\ -\{\tilde{\mathcal{F}}_{<_{G_{2}}}(\phi(t(e)),\phi(s(e)))\}&\begin{aligned} &\text{if}\>\phi(e)\>\text{has a loop and}\\ &\>\phi(t(e))<_{G_{2}}\phi(s(e))\end{aligned}\\ \;\;\>\{\phi(e)\}&\text{otherwise}\end{cases} (22) with an analogous expression for \tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\varphi)(\{e_{G_{2}}\}) where e_{G_{2}}\in E_{G_{2}} is an edge with no loop. It is necessary to check each of the possible combinations of cases arising for \phi and \varphi arising from equation ( 22 ), and the \varphi case of equation ( 21 ). Firstly, when \phi(e) has a loop, then as \varphi is a homomorphism (\varphi\circ\phi)(e) must have a loop. Therefore, using the \phi version of equation ( 22 ) and the \varphi version of equation ( 21 ), both \left(\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\varphi)\circ\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\phi)\right)(\{\alpha_{e}\}) and \tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\varphi\circ\phi)(\{\alpha_{e}\}) are equal to \displaystyle\left\{\tilde{\mathcal{F}}_{<_{G_{2}}}((\varphi\circ\phi)(s(e)),(\varphi\circ\phi)(t(e)))\right\}\;\text{or}\;-\left\{\tilde{\mathcal{F}}_{<_{G_{2}}}((\varphi\circ\phi)(t(e)),(\varphi\circ\phi)(s(e)))\right\} (23) depending on whether s(\phi(e))<_{G_{3}}t((\varphi\circ\phi)(e)) or t((\varphi\circ\phi)(e))<_{G_{3}}s(\phi(e)) , respectively. Similarly, we now consider the case when \phi(e) does not have a loop but (\varphi\circ\phi)(e) does have a loop. In this case, using the \phi and \varphi versions of equation ( 22 ), the resulting expressions for both \tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\varphi\circ\phi)(\alpha_{e}) and \left(\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\varphi)\circ\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\phi)\right)(\alpha_{e}) are the same as in equation ( 23 ), again depending on whether s((\varphi\circ\phi)(e))<_{G_{3}}t((\varphi\circ\phi)(e)) or t((\varphi\circ\phi)(e))<_{G_{3}}s((\varphi\circ\phi)(e)) , respectively. The last case to consider occurs when (\varphi\circ\phi)(e) does not have a loop, which by the property that homomorphisms send loops to loops implies that \phi(e) cannot have a loop. Therefore, by applying the \phi and \varphi versions of equation ( 22 ), we obtain \tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\varphi\circ\phi)(\alpha_{e})=\{(\varphi\circ\phi)(e_{G_{1}})\}=\left(\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\varphi)\circ\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}}(\phi)\right)(\alpha_{e}) as required. Finally, using the the second line of equation ( 6.2 ), fuctoriality for n\geq 2 on basis elements \alpha=\{\alpha_{e_{i,j}}\}_{0\leq it\leq n-m\\ b^{f}_{t+1}&\text{otherwise}.\end{cases} (24) for t=0,\dots,n . We also define s_{<}^{f}\in\tilde{\mathcal{F}}_{<}(G) given by s_{<}^{f}=f(0)\in\tilde{\mathcal{F}}_{<}(G)_{0}\;\;\;\text{when}\;\;\;n=0\;\;\;\text{and}\;\;\;s_{<}^{f}=\{\alpha_{f(i\to j)}\}_{0\leq if\>\text{is an inclusion}\\ 0&\text{otherwise}\end{cases} (25) and show that it is a chain map a follows. Suppose first that f\colon\Delta^{n}\to G is a singular simplex homomorphism that is not an inclusion, the exists 0\leq i|\>i=0,\dots,n\}, edges E_{G_{f}}=\{f^{\prime}(i\to j)\>|\>f^{\prime}\colon\Delta^{n}\to G\;\text{is a singular simplex inclusion inducing}\>s^{f}_{<}\>\text{and}\>0\leq i|\>0\leq i\{\tilde{\mathcal{F}}_{<_{1}}(f(0),f(1))\}&\text{if}\>f(0)<_{1}f(1)\\ -\{\tilde{\mathcal{F}}_{<_{1}}(f(1),f(0))\}&\text{otherwise}\end{cases} and (g\circ\phi_{\#})(f)=\begin{cases}\;\;\>\{\tilde{\mathcal{F}}_{<_{2}}((\phi\circ f)(0),(\phi\circ f)(1))\}&\text{if}\>(\phi\circ f)(0)<_{2}(\phi\circ f)(1)\\ -\{\tilde{\mathcal{F}}_{<_{2}}((\phi\circ f)(1),(\phi\circ f)(0))\}&\text{otherwise}.\end{cases} It now follows directly from the construction of \tilde{f}_{\{<_{1},<_{2}\}}(\phi) , that (\tilde{f}_{\{<_{1},<_{2}\}}(\phi\circ g))(f) agrees with the second equation above by equations ( 6.2 ) Secondly, when neither f(0) or f(1) have a loop and neither (\phi\circ f)(0) or (\phi\circ f)(1) have a loop, then (g\circ\phi_{\#})(f)=\{(e_{(\phi\circ f)(0\to 1)}\}=(\tilde{f}_{\{<_{1},<_{2}\}}(\phi\circ g))(f). Finally, when neither f(0) or f(1) have a loop and either (\phi\circ f)(0) or (\phi\circ f)(1) has a loop, then \displaystyle(g\circ\phi_{\#})(f) \displaystyle=\begin{cases}\;\;\>\{\tilde{\mathcal{F}}_{<_{2}}(\phi(f(0)),\phi(f(1)))\}&\phi(f(0))<_{2}\phi(f(1))\\ -\{\tilde{\mathcal{F}}_{<_{2}}(\phi(f(1)),\phi(f(0)))\}&\>\text{otherwise}.\end{cases} \displaystyle=(\tilde{f}_{\{<_{1},<_{2}\}}(\phi\circ g))(f) For n\geq 1 , using the same reasoning as the final paragraph of the proof of Proposition 6.3 , commutativity of diagram ( 26 ) on generators follows from the n=1 case. Applying the homology functor to diagram ( 26 ), we induce the commutative diagram on homology in the statement of the theorem. Using the commutativity of the diagram in homology and Proposition 6.3 , the fuctoriality of the partial flag functors extends g_{*} to a natural transformation between functors H_{*}^{\Delta,h} and H_{*}\circ\tilde{f}_{\{<_{G}\}_{G\in\mathcal{C}}} . In addition, invariance of partial flag functors with respect to the choices of loop maximal total orders follows also from the commutative diagram in homology, as any choice of such an order commutes up to isomorphism g_{*} with homomorphism \phi_{*} . Therefore, any choices of loop maximal total order must result in the same homomorphisms of graded modules between any pair of quivers, as required. ∎ To end the section, we apply Theorem 6.4 to demonstrate that the H_{*}^{\Delta,h} is invariant with respect to the weak local strong h -homotopy of quivers given in Definition 10 , differing from local strong h -homotopy by additionally allowing loop contractions in the sense of Definition 9 . Theorem 6.5 . The functor H^{h,\Delta}_{*} is invariant under weak local strong h -homotopy equivalence of quivers. Proof. Suppose that quivers G_{1} and G_{2} are weak local strong h -homotopy equivalent. By construction of weak local strong h -homotopy, it is sufficient to prove the statement of the theorem in the case G_{1}\simeq^{wlSh}_{1}G_{2} . Using Theorem 4.1 , we have that H^{h,\Delta}_{*}(G_{1})=H^{h,\Delta}_{*}(G_{2}) when the 1 -step homotopy arises from \phi\colon G_{1}\to G_{2} , \varphi\colon G_{2}\to G_{1} such that \varphi\circ\phi\simeq_{1}^{lSh}\text{id}_{G_{1}} , \phi\circ\varphi\simeq_{1}^{lSh}\text{id}_{G_{2}} . Therefore, without loss of generality it is sufficient to show that H^{h,\Delta}_{*}(G_{1})=H^{h,\Delta}_{*}(G_{2}) when G_{2} is a loop contraction of G_{1} . Moreover, we need only consider the case when a single degenerate loop l at some v_{l}\in V_{G} is removed, since the same conditions apply equally to all degenerate loops simultaneously. Let < be loop maximal total order on G_{1} . Then < is also a loop maximal total order on G_{2} . We will construct chain maps g\colon C_{*}(\tilde{\mathcal{F}}_{<}(G_{1}))\to C_{*}(\tilde{\mathcal{F}}_{<}(G_{2})) and h\colon C_{*}(\tilde{\mathcal{F}}_{<}(G_{2}))\to C_{*}(\tilde{\mathcal{F}}_{<}(G_{1})) satisfying h\circ g=\text{id}_{C_{*}(\tilde{\mathcal{F}}_{<}(G_{1}))} and g\circ h=\text{id}_{C_{*}(\tilde{\mathcal{F}}_{<}(G_{2}))} . This implies that C_{*}(\tilde{\mathcal{F}}_{<}(G_{2})) and C_{*}(\tilde{\mathcal{F}}_{<}(G_{2})) are chain homotopic. Therefore, by Theorem 6.4 , we obtain that H_{*}^{\Delta,h}(G_{1})=H_{*}(\tilde{\mathcal{F}}_{<}(G_{1}))=H_{*}(\tilde{\mathcal{F}}_{<}(G_{2}))=H_{*}^{\Delta,h}(G_{2}) as desired. Define g\colon C_{*}(\tilde{\mathcal{F}}_{<}(G_{1}))\to C_{*}(\tilde{\mathcal{F}}_{<}(G_{2})) and h\colon C_{*}(\tilde{\mathcal{F}}_{<}(G_{)})\to C_{*}(\tilde{\mathcal{F}}_{<}(G_{1})) on vertices by linearly extending the identity map on vertices and on n -simplices for n\geq 1 as follows. Given s_{1}\in\tilde{F}(G_{1})_{n} denote by f_{s_{1}} a choice singular n -simplex inclusion in G_{1} inducing s_{1} . Since f_{s_{1}} is an inclusion it contains no loops in is image. Hence, f_{s_{1}} is also a singular simplex inclusion with image in G_{2} . As any singular simplex inclusion induces some n -simplex in \tilde{F}(G_{1})_{2} , f_{s_{1}} induces simplex s_{2}\in\tilde{\mathcal{F}}(G_{2})_{n} . Crucially, since l is a degenerate loop, being adjacent to no multiple edges and any double edges it is adjacent to having a loop in G_{2} , any choice of f_{s_{1}} provides the same simplex s_{2} . Therefore, the simplex s_{2} is uniquely determined and we construct g by linearly extending g(s_{1})=s_{2} . The construction of h on n -simplices for n\geq 1 is achieved similarly with the roles of \tilde{F}(G_{1})_{n} and \tilde{F}(G_{2})_{n} reversed. It follows from the uniqueness of the construction of n -simplices above that (g\circ h)(s_{1})=s_{1} for any s_{1}\in\tilde{F}(G_{1})_{1} and (g\circ h)(s_{2})=s_{2} for any s_{2}\in\tilde{F}(G_{1})_{2} . Therefore, g and h are chain maps satisfying h\circ g=\text{id}_{C_{*}(\tilde{\mathcal{F}}_{<}(G_{1}))} and g\circ h=\text{id}_{C_{*}(\tilde{\mathcal{F}}_{<}(G_{2}))} , as required. ∎ Appendix A Algorithmic implementations In this appendix we detail algorithms that obtain from a filtered quiver a filtered \Delta -set corresponding to the images of the directed flag, reduced directed flag, and partial directed flag complexes. To investigate the efficiency of the algorithms we consider the computational complexity of each procedure. Through this analysis, we demonstrate that dependence on the number of n -simplices in any give dimension makes computation of the reduced directed flag and partial reduced directed flag complexes significantly more efficient than direct calculation of generators of the quasi-isomorphic chain complexes C^{\Delta,m}_{*}(G) and C^{\Delta,h}_{*}(G) , respectively. A working python demonstration of all algorithms detailed is available at [ 6 ] . Each proposed algorithm is structured with a main loops over the vertices v\in V_{G} . Therefore, all procedures can be parallel processed by assigning computations over each vertex v\in V_{G} to separate processing units. Moreover, the algorithms are designed to remain memory efficient, requiring only information on certain n -simplices and vertices of G at any given step. Once the filtered \Delta -sets are obtained, the results can be combined with [ 49 ] or [ 34 ] to compute persistent homology. Providing efficient computation of persistent H^{\Delta,i}_{n} , H^{\Delta,m}_{n} , and H^{\Delta,h}_{n} on filtered quivers. For the computational purposes of this appendix, an abstract simplicial complexes comes equipped with a total order on its vertices. In addition, quivers, simplicial complexes, and \Delta -sets are assumed to be finite. Throughout the appendix all indices begin at 0 , and as throughout this work G is a quiver and n a non-negative integer, unless stated otherwise. A.1 Filtered objects For any given quiver homology, additional information about the structure of the quiver with respect to an associated filtration can be extracted using persistent homology, the construction of which we detail in Appendix A.3 . Definition 16 . A filtered quiver is a quiver G together with real valued functions f_{V}\colon V_{G}\to\mathbb{R}\cup\{-\infty,\infty\} and f_{E}\colon E_{G}\to\mathbb{R}\cup\{-\infty,\infty\} such that f_{E}(e)\geq f_{V}(s(e)) and f_{E}(e)\geq f_{V}(t(e)) for each e\in E_{G} . Given t\in\mathbb{R}\cup\{\infty\} , the sublevel quivers G^{t} of a filtered quiver G are given by V_{G^{t}}=f_{V}^{-1}([-\infty,t)) and E_{G^{t}}=f_{E}^{-1}([-\infty,t)) . It is the changes in the homology of the increasing family of sublevel quivers that is measured by persistent homology. Definition 17 . A filtered simplicial complex is an abstract simplicial complex (S,V) together with a real valued function f\colon S\to\mathbb{R}\cup\{-\infty,\infty\} such that for any s,s^{\prime}\in S with s\subseteq s^{\prime} we have f(s)\leq f(s^{\prime}) . A filtered \Delta -set is a \Delta -set X together with real valued functions f_{n}\colon X_{n}\to\mathbb{R}\cup\{-\infty,\infty\} such that f_{n+1}(x_{n+1})\geq f_{n}(d_{i}^{n}(x_{n+1})) for each integer n\geq 0 , i=0,\dots,n+1 , and x_{n+1}\in X_{n+1} . The sublevel simplicial complexes (V^{t},S^{t}) of a filtered simplicial complex are given by S^{t}=f^{-1}([-\infty,t)) and V^{t}=\{v\in V\>|\>\{v\}\in S^{t}\} for each t\in\mathbb{R}\cup\{\infty\} . The sublevel \Delta -sets X^{t} of a filtered \Delta -set X are given by X_{n}^{t}=f_{n}^{-1}([-\infty,t)) for each t\in\mathbb{R}\cup\{\infty\} and integer n\geq 0 . As all quivers, simplicial complexes, and \Delta -sets considered in this section are finite, there are only finitely many values of t for which the sublevel quivers, simplicial complexes, or \Delta -sets of any such filtered object are distinct. A.2 Data structures Before detailing the quiver homology algorithms, we first set out the data structures used to store quivers, filtered quivers, simplices, filtered simplices, simplicial complexes, filtered simplicial complexes, \Delta -sets, and filtered \Delta -sets. Quivers were defined in Section 2.2 , and simplicial complexes and \Delta -sets in Section 2.1 earlier in this work. For the purposes of setting out the flag complex \mathcal{F} and partially reduced flag complex \tilde{\mathcal{F}} procedures, we additionally require a notion of k -partial (n+1) -dimensional simplices within a \Delta -set for k=0,\dots,n+1 . In particular, a 0 -partial (n+1) -simplex in X is an n -simplex of X and the boundary of an (n+1) -simplex provides an (n+1) -partial (n+1) -simplex in X . Formally, a k -partial n -simplex in X consists of an sequential collection of n -simplices subject to a subset of the face conditions from equation ( 1 ) and is defined precisely as follows. Definition 18 . Let X be a \Delta -set, n\geq 0 and k=0,\dots,n+1 . Then a k -partial (n+1) -simplex in X consists of n -simplices s_{n+1-k},\dots,s_{n+1}\in X_{n} satisfying d_{i}^{n-1}(s_{j})=d_{j-1}^{n-1}(s_{i}) for each i,j=n+1-k,\dots,n+1 with i