On torsion in eulerian magnitude homology of Erdös-Rényi random graphs
Abstract.
In this paper we investigate the regimes where an Erdös-Rényi random graph has torsion free eulerian magnitude homology groups. To this end, we start be introducing the eulerian Asao-Izumihara complex - a quotient CW-complex whose homology groups are isomorphic to direct summands of the graph eulerian magnitude homology group. We then proceed by producing a vanishing threshold for a shelling of eulerian Asao-Izumihara complex. This will lead to a result establishing the regimes where eulerian magnitude homology of Erdös-Rényi random graphs is torsion free.
1. Introduction
Magnitude, introduced by Leinster in [20], is an invariant for metric spaces that quantifies the number of effective points in the space. Hepworth and Willerton introduced magnitude homology for graphs as a categorification of magnitude [15], and this concept was later extended to metric spaces and enriched categories by Leinster and Shulman [21]. In recent years, various methods have been devised to calculate the magnitude homology groups [1, 5, 11, 15, 19].
Eulerian magnitude homology is a variant recently introduced by Giusti and Menara in [9] to highlight the connection between magnitude homology of simple graphs equipped with the path metric and their combinatorial structure. Here the authors introduce the complex of eulerian magnitude chains, which are supported by trails without repeated vertices. Then they describe the strong connections between the -eulerian magnitude homology groups and the graph’s structure. Further, in the context of Erdös-Rényi random graphs they derive a vanishing threshold for the limiting expected rank of the -eulerian magnitude homology in terms of the density parameter.
In this paper, we will make some progress towards investigating the presence of torsion in eulerian magnitude homology.
Torsion in standard magnitude homology was first studied by Kaneta and Yoshinaga in [19], where the authors have analyzed the structure and implications of torsion in magnitude homology. Torsion in the magnitude homology of graphs was also studied by Sazdanovic and Summers in [23] and by Caputi and Collari in [6].
In the present work, as a first step towards exploring whether graphs have torsion in their eulerian magnitude homology groups, we turn our attention to Erdös-Rényi model for random graphs. This model is the most extensively studied and utilized model for random graphs, and it represents the maximum entropy distribution for graphs with a given expected edge proportion. Random complexes originating from Erdös-Rényi graphs are widely studied in stochastic topology [16, 17, 18], and in studying this “unstructured” example our intent is to create a foundation for understanding the torsion in “structured” graphs.
Adapting the construction introduced by Asao and Izumihara in [2] to the context of eulerian magnitude homology, we able to produce for every pair of vertices two simplicial complexes and such that the homology of the quotient is isomorphic to a direct summand of the eulerian magnitude homology up to degree shift. Therefore, producing a shellability result of the complexes and will in turn determine a torsion-free result for . In Theorem 22 we achieve such shellability result for in terms of the density parameter. Further, in Corollary 27 we link the torsion-free result for eulerian magnitude homology groups stated in Theorem 24 with the vanishing threshold produced in [9, Thm. 4.4], determining sufficient conditions under which if eulerian magnitude homology is non-vanishing, then it is also torsion-free.
1.1. Outline
The paper is organized as follows. We start by recalling in Section 2 some general background about graphs, eulerian magnitude homology and shellability. In Section 3 we introduce the eulerian Asao-Izumihara complex. We then investigate in Section 4 the probability regimes in which the eulerian Asao-Izumihara complex is shellable, and we conclude by producing a vanishing threshold for torsion in eulerian magnitude homology groups. Finally, in Section 5 we propose extensions of the current work and identify open questions that could deepen the understanding of the topic.
2. Background
We begin by recalling relevant definitions and results. We assume readers are familiar with the general theory of simplicial homology (for a thorough exposition see [12]). Throughout the paper we adopt the notation and for common indexing sets.
2.1. Graph terminology and notation
An undirected graph is a pair where is a set of vertices and is a set of edges (unordered pairs of vertices). A walk in such a graph is an ordered sequence of vertices such that for every index there is an edge . A path is a walk with no repeated vertices. For the purposes of introducing eulerian magnitude homology we assume that all graphs are simple, i.e. they have no self-loops and no multiedges [22]. One can interpret the set of vertices of a graph as an extended metric space (i.e. a metric space with infinity allowed as a distance) by taking the path metric to be equal to the length of a shortest path in from to , if such a path exists, and taking if and lie in different components of .
Definition 1.
Let be a graph, and a non-negative integer. A -trail in is a -tuple of vertices for which and for every . The length of a -trail in is defined as the minimum length of a walk that visits in this order:
We call the vertices the landmarks, the starting point, and the ending point of the -trail.
2.2. Eulerian magnitude homology
The magnitude homology of a graph , , was first introduced by Hepworth and Willerton in [15], and the eulerian magnitude homology of a graph is a variant of it with a stronger connection to the subgraph structures of . Specifically, while the building blocks of standard magnitude homology are tuples of vertices where we ask that consecutive vertices are different, eulerian magnitude homology is defined starting from tuples of vertices where we ask that all landmarks are different.
Eulerian magnitude homology was recently introduced by Giusti and Menara in [9] and we recall here the construction.
Definition 2.
(Eulerian magnitude chain) Let be a graph. We define the -eulerian magnitude chain to be the free abelian group generated by trails such that for every and .
It is straightforward to demonstrate that the eulerian magnitude chain is trivial when the length of the path is too short to support the necessary landmarks.
Lemma 3 (c.f. [15, Proposition 10]).
Let be a graph, and non-negative integers. Then
Proof.
Suppose Then, there must exist a -trail in so that . However, as all vertices in the -trail must be distinct, for , so can be at most . ∎
Definition 4.
(Differential) Denote by the -tuple obtained by removing the -th vertex from the -tuple . We define the differential
to be the signed sum of chains corresponding to omitting landmarks without shortening the walk or changing its starting or ending points,
For a non-negative integer , we obtain the eulerian magnitude chain complex, given by the following sequence of free abelian groups and differentials.
Definition 5.
(Eulerian magnitude chain complex) We indicate as the following sequence of free abelian groups connected by differentials
The differential map used here is the one induced by standard magnitude, and it is shown in [15, Lemma 11] that the composition vanishes, justifying the name “differential” and allowing the definition the corresponding bigraded homology groups of a graph.
Definition 6.
(Eulerian magnitude homology) The -eulerian magnitude homology group of a graph is defined by
Notice that by construction we have the following proposition.
Proposition 8.
For , the following direct sum decomposition holds:
where is the subcomplex of generated by trails which start at and end at .
2.3. Shellable simplicial complexes
We recall the definition of shellable simplicial complex.
Definition 9 ([4, Definition 2.1]).
If is a finite simplicial complex, then a shelling of is an ordering of the facets (maximal faces) of such that is a non-empty union of facets of for . If has a shelling, we say it is shellable.
In other words, we ask that the last simplex meets the previous simplices along some union of top-dimensional simplices of the boundary of , so that can be built stepwise by introducing the facets one at a time and attaching each new facet to the complex previously built in the nicest possible fashion.
Suppose is a non-pure simplicial complex. In this case the first facet of a shelling is always of maximal dimension. In fact, if is shellable there is always a shelling in which the facets appear in order of decreasing dimension.
Lemma 10 ([4, Rearrangement lemma, 2.6]).
Let be a shelling of . Let be the rearrangement obtained by taking first all facets of dimension in the induced order, then all facets of dimension in the induced order, and continuing this way in order of decreasing dimension. Then this rearrangement is also a shelling.
Theorem 11 ([4, Theorem 2.9]).
Let be a simplicial complex, and let . Define . If is shellable, then so is for all .
Lemma 10 and Theorem 11 can be interpreted as providing a kind of “structure theorem”, describing how a general shellable complex is put together from pure shellable complexes. First there is the pure shellable complex generated by all facets of maximal size. Then ’s -skeleton, which is also shellable, is extended by shelling steps in dimension to obtain . Then ’s -skeleton is extended by shelling steps in dimension to obtain , and so on until all of has been constructed.
A shellable simplicial complex enjoys several strong properties of a combinatorial, topological and algebraic nature. Let it suffice here to mention that it is homotopy equivalent to a wedge sum of spheres, one for each spanning simplex of corresponding dimension [8].
3. Eulerian Asao-Izumihara complex
We introduced in this section the eulerian Asao-Izumihara complex.
Recall that the Asao-Izumihara complex is a CW complex which is obtained as the quotient of a simplicial complex divided by a subcomplex , and was proposed in [2] as a geometric approach to compute magnitude homology of general graphs.
Here we adapt this construction to the context of eulerian magnitude homology, providing a way of replacing the computation of the eulerian magnitude homology by that of simplicial homology.
Let us start by recalling the Asao-Izumihara complex.
Let be a connected graph and fix .
For any the set of walks with length which start with and end with is denoted by
Definition 12 (c.f.[2, Def. 4.1]).
Let G be a graph, and , .
Remark 13.
Following [2], we will denote by when there is no confusion.
It can also be easily seen that is a simplicial complex and is a subcomplex.
Theorem 14 (c.f.[2, Thm. 4.3]).
Let and . Then, the isomorphism
of chain complexes holds.
Corollary 15 (c.f.[2, Cor. 4.4]).
Let .
-
•
If , .
-
•
If , we also have
where denotes the reduced homology group.
Remark 16.
Notice while both and are subcomplexes of , in general . Indeed, say and are two adjacent vertices, then the tuple is an element of both and because it is a subtuple of , but it cannot be in . This type of example with consecutively repeated vertices is the only one that can be constructed to show that is a proper subset of , and in the context of eulerian magnitude homology it cannot arise because the tuples have all different vertices. Therefore when introducing the eulerian Asao-Izumihara complex it will possible to only rely on the (eulerian versions of the) complexes and .
Definition 17.
Let be the set of eulerian trails from to with length smaller than . That is, the set of all trails such that for every and
The set is clearly a simplicial complex, and the complex is a subcomplex of , see Figure 2 for an illustration.
Example 18.
Consider the same graph as in example 7. Suppose we choose and . Then we have and .
The following two results can be shown proceeding similarly to the proofs of [2, Thm. 4.3 and Cor. 4.4].
Theorem 19.
Let be vertices of a graph , and fix an integer . Then we can construct a pair of simplicial complexes which satisfies
Corollary 20.
Let . Then
Moreover, for , we also have
where denotes the reduced homology group.
4. Torsion in EMH of Erdős-Rényi random graphs
In this section we investigate the regimes where the eulerian magnitude homology of Erdős-Rényi random graphs is torision free.
Recall that the Erdős-Rényi (ER) model for random graphs, denoted as and first introduced in [7], is one of the most extensively studied and utilized models for random graphs. This model represents the maximum entropy distribution for graphs with a given expected edge proportion, making it a valuable null model across a wide array of scientific and engineering fields. Consequently, the clique complexes of ER graphs have garnered significant interest within the stochastic topology community [16, 17, 18].
Definition 21.
The Erdős-Rényi (ER) model is the probability space where is the discrete space of all graphs on vertices, and is the probability measure that assigns to each graph with edges probability
We can sample an ER graph on vertices with parameter by determining whether each of the potential edges is present via independent draws from a Bernoulli distribution with probability . In order to study the limiting behavior of these models as , it is often useful to change variables so that is a function of . Here we will take , , as in [9].
We will first prove in Section 4.1 that, under certain assumptions, the complex is shellable for every choice for . This will imply that is torsion free, and by Corollary 20 that is torsion free.
4.1. Homotopy type of the eulerian Asao-Izumihara complex
Recall from Section 3 that the eulerian Asao-Izumihara chain complex is the relative complex , where is the set of eulerian tuples such that , and is defined similarly. Fix and integer .
Theorem 22.
Let be an ER graph. Suppose the facets of are ordered in decreasing dimension. Then as is shellable asymptotically almost surely when
-
•
, if ,
-
•
, if for and for .
Proof.
Consider the facets of . Suppose they are ordered in decreasing dimension and say . There are some cases we need to consider.
-
(1)
If there is a single facet , then is homotopic to a sphere with and we are done.
-
(2)
Say there are two different maximal facets, and and suppose they have the same dimension .
If and differ in one vertex, then they intersect in a -face, and thus is a shelling.
If and differ in two vertices , then we need to distiguish the situations when and are adjacent and when they are not.-
(a)
If and are not adjacent, then we will have and , and by construction there exists a third facet such that is a shelling, see Figure 3.
-
(b)
If and are adjacent, then in order to construct a facet intersecting in a -face we need either the edge or the edge to be present (see Figure 4), and this happens with probability .
Now say and differ in vertices and, indicating the facets and only by the vertices they differ in, write and . Define a partition with such that two vertices belong to the same set if and only if they are adjacent in , see Figure 5. Call the corresponding partition for the vertices . Notice that for every . Indeed, suppose by contradiction this is not true. Then, because and have the same dimension, there exists such that and . But then it is possible to construct a visiting vertices from and thus having , contradicting the fact that and are maximal facets.
Then in this case we need for every set of adjacent vertices and a number of edges , , in order to create a shelling. Indeed, we need to be able to construct a sequence of facets by changing one vertex each time so that the intersection between the -th facet and the preceding facets is a - dimensional simplex, see Figure 5. Given the fact that we also require for every set a number of edges to connect the vertices in , we obtain that the probability of all the required edges existing is
With , , we get
Notice that we assumed and only when .
With , , we get
Since it holds also in this case that if and only if , we can conclude that we can construct a shelling when
-
(a)
-
(3)
Suppose now there are two different facets, and , and suppose .
Let . Following the structure theorem for non-pure shellable complexes provided by Lemma 10 and Theorem 11, in order to produce a shelling we need to extend the -skeleton of to by constructing a sequence of -dimensional facets by changing one vertex each time so that the intersection between the -th facet and the preceding facets is a -dimensional simplex.
If the simplices in the -skeleton of and differ in vertices, constructing such sequence is possible if we can find edges joining the vertices in which and differ. This happens with probability and therefore following the computations done in the previous point we get, for and ,
With , , we get
Again, from the fact that both inequalities and are true if and only if , we conclude that we can construct a shelling when
-
(4)
Suppose there are facets ordered in decreasing order with , then we only need to iterate the observations made in point (3).
That is, at each step we have a shelling when
Therefore, suppose . Then every smaller facet will be such that and we will have a shelling when
On the other hand, if let be the first facet in the sequence such that . Then we will have a shelling when
∎
Corollary 23.
Let be an ER graph. Suppose the facets of are ordered in decreasing dimension. Then as is shellable asymptotically almost surely when
-
•
, if ,
-
•
, if for and for .
It was shown in both [8] and [3] that a shellable simplicial complex has the homotopy type of a wedge of spheres.
Theorem 24.
Let be an ER graph. For any pair of vertices consider the eulerian Asao-Izumihara chain complex . Suppose the facets of and of are ordered in decreasing dimension. As , in the regimes where both and are shellable, is torsion free for every .
Proof.
In the regimes where both and are shellable we can assume
So, , and considering the long exact sequence
we see that
Recall that [9, Theorem 4.4] provides a vanishing threshold for the limiting expected rank of the -eulerian magnitude homology in terms of the density parameter in the contexts of Erdös-Rényi random graphs.
Theorem 25 ([9, Theorems 4.4]).
Let be an Erdös-Rényi random graph. Fix and let . As asymptotically almost surely.
Remark 26.
Notice that when the smallest facet of , , is such that , then is shellable when
Corollary 27.
Let be an Erdős-Rényi random graph. When the smallest facet of and the smallest facet of are such that , if is non-vanishing it is also torsion free.
5. Future directions
In this paper we investigated the regimes where an Erdös-Rényi random graph has torsion free eulerian magnitude homology groups.
While the results presented have provided significant insights into the problem, several aspects remain unexplored, offering fertile ground for continued research.
In this section, we propose extensions of the current work and identify open questions that could deepen the understanding of the topic.
5.1. The choice of
The result stated in Corollary 27 relies on the dimension of the minimal facet of and the minimal facet of being “close enough” to the parameter so that and for every other facet .
It is thus natural to ask, how do we choose so that ?
First, notice that the parameter cannot be too big with respect to the number of vertices . Specifically, cannot be of the order . Indeed, suppose we pick . The only way we can produce a facet inducing a path of such length is if we have a path graph on vertices , , and we visit vertex after vertex , , i.e. . Then . See Figure 6 for an illustration.
We conclude that a quadratic growth rate for with respect to is not appropriate.
On the other hand, setting we do not encounter the same problem as before. For example, consider the path graph in Figure 6. Choosing and we find two facets and . Both have dimension and thus .
Based on this computation, along with many other examples not displayed here, we make the following conjecture.
Conjecture 1.
Indicate the diameter of the graph by . There exists a linear function such that if , then .
5.2. Connection with the complex of injective words
A natural development of the work present in this paper (which we are already investigating) concerns a deterministic result about the presence of torsion in eulerian magnitude homology groups of graphs.
It is the author’s belief that this kind of result can be achieved by exploiting the strong connection between the eulerian magnitude chain complex and the complex of injective words.
An injective word over a finite alphabet is a sequence of distinct elements of .
Call the set of injective words on partially ordered by inclusion, and recall that the order complex of a poset , denoted , is the simplicial complex on the vertex set , whose -simplices are the chains of .
For example, if with the usual ordering, then is the standard -simplex.
Definition 28.
A complex of injective words is an order complex associated to a subposet .
Farmer [8] proved that if , then has the homology of a wedge of copies of the -sphere , where is the number of derangements (i.e. fixed point free permutations) in . The following result was obtained by Björner and Wachs in [3] as a strengthening of Farmer’s theorem.
Theorem 29 ([3]).
.
Let now the alphabet be the vertex set of a graph . Let be the set of injective words on the vertex set and denote by , the subset containing such that length of the walk in is less than . Then we have a filtration
(1) |
The following equivalence easily follows from the definition of the filtration of and the definition of the eulerian Asao-Izumihara complex ,
where denotes the geometric realization.
Further, the connection between the eulerian magnitude chain complex and the complex of injective words is strengthen by the following observation.
Hepworth and Roff [14] thoroughly analyzed in the context of directed graphs the magnitude-path spectral sequence (MPSS), a spectral sequence whose page is exactly standard magnitude homology, path homology [10] can be identified with a single axis of page , and whose target object is reachability homology [13].
Reproducing the computations proposed in [14, Section 2] using the filtration of the complex of injective words in 1, leads to a version of the MPSS where the page is exactly eulerian magnitude homology. Since the homology of the complex of injective words, as the target object, controls the behavior of the spectral sequence, we make the following conjecture.
Conjecture 2.
Let be a graph. The eulerian magnitude homology groups of , , are torsion free for every .
Acknowledgments
The author is thankful to Yasuhiko Asao, Luigi Caputi and Chad Giusti for helpful conversations throughout the development of this work.
References
- [1] Yasuhiko Asao and Sergei O Ivanov, Magnitude homology is a derived functor, arXiv preprint arXiv:2402.14466 (2024).
- [2] Yasuhiko Asao and Kengo Izumihara, Geometric approach to graph magnitude homology, Homology, Homotopy and Applications 23 (2021), no. 1, 297–310.
- [3] Anders Björner and Michelle Wachs, On lexicographically shellable posets, Transactions of the American Mathematical Society 277 (1983), no. 1, 323–341.
- [4] by same author, Shellable nonpure complexes and posets. i, Transactions of the American mathematical society 348 (1996), no. 4, 1299–1327.
- [5] Rémi Bottinelli and Tom Kaiser, Magnitude homology, diagonality, and median spaces, Homology, Homotopy and Applications 23 (2021), no. 2, 121–140.
- [6] Luigi Caputi and Carlo Collari, On finite generation in magnitude (co) homology, and its torsion, arXiv preprint arXiv:2302.06525 (2023).
- [7] Paul Erdős, Alfréd Rényi, et al., On the evolution of random graphs, Publ. Math. Inst. Hung. Acad. Sci 5 (1960), no. 1, 17–60.
- [8] Frank D Farmer, Cellular homology for posets, Math. Japon 23 (1978), no. 6, 79.
- [9] Chad Giusti and Giuliamaria Menara, Eulerian magnitude homology: subgraph structure and random graphs, arXiv preprint arXiv:2403.09248 (2024).
- [10] Alexander Grigor’yan, Rolando Jimenez, Yuri Muranov, and Shing-Tung Yau, Homology of path complexes and hypergraphs, Topology and its Applications 267 (2019), 106877.
- [11] Yuzhou Gu, Graph magnitude homology via algebraic morse theory, arXiv preprint arXiv:1809.07240 (2018).
- [12] Allen Hatcher, Algebraic topology, 2005.
- [13] Richard Hepworth and Emily Roff, The reachability homology of a directed graph, arXiv preprint arXiv:2312.01378 (2023).
- [14] by same author, Bigraded path homology and the magnitude-path spectral sequence, arXiv preprint arXiv:2404.06689 (2024).
- [15] Richard Hepworth and Simon Willerton, Categorifying the magnitude of a graph, arXiv preprint arXiv:1505.04125 (2015).
- [16] Matthew Kahle, Topology of random clique complexes, Discrete mathematics 309 (2009), no. 6, 1658–1671.
- [17] by same author, Random geometric complexes, Discrete & Computational Geometry 45 (2011), 553–573.
- [18] Matthew Kahle and Elizabeth Meckes, Limit theorems for betti numbers of random simplicial complexes, Homology, Homotopy and Applications 15 (2013), no. 1, 343–374.
- [19] Ryuki Kaneta and Masahiko Yoshinaga, Magnitude homology of metric spaces and order complexes, Bulletin of the London Mathematical Society 53 (2021), no. 3, 893–905.
- [20] Tom Leinster, The magnitude of metric spaces, Documenta Mathematica 18 (2013), 857–905.
- [21] Tom Leinster and Michael Shulman, Magnitude homology of enriched categories and metric spaces, Algebraic & Geometric Topology 21 (2021), no. 5, 2175–2221.
- [22] Tom Leinster, The magnitude of a graph, Mathematical Proceedings of the Cambridge Philosophical Society, vol. 166, Cambridge University Press, 2019, pp. 247–264.
- [23] Radmila Sazdanovic and Victor Summers, Torsion in the magnitude homology of graphs, Journal of Homotopy and Related Structures 16 (2021), no. 2, 275–296.