Stability in the Erd\H{o}s--Gallai Theorem on cycles and paths, II
Zolt\'an F\"uredi, Alexandr Kostochka, Ruth Luo, Jacques Verstra\"ete

TL;DR
This paper strengthens Kopylov's stability theorem for the Erd ext{o}s--Gallai cycle length problem, characterizing the structure of near-extremal 2-connected graphs with no long cycles.
Contribution
It completes a stability theorem for the Erd ext{o}s--Gallai problem, identifying the structure of graphs close to extremal edge counts for odd cycle length constraints.
Findings
Characterizes 2-connected graphs near extremal edge counts
Identifies extremal graphs with maximum edges under cycle constraints
Provides tight bounds for the number of edges in such graphs
Abstract
The Erd\H{o}s--Gallai Theorem states that for , any -vertex graph with no cycle of length at least has at most edges. A stronger version of the Erd\H{o}s--Gallai Theorem was given by Kopylov: If is a 2-connected -vertex graph with no cycle of length at least , then , where . Furthermore, Kopylov presented the two possible extremal graphs, one with edges and one with edges. In this paper, we complete a stability theorem which strengthens Kopylov's result. In particular, we show that for odd and all , every -vertex -connected graph with no cycle of length at least is a subgraph of one of the two extremal graphs or $e(G) \leq…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
Topicsadvanced mathematical theories · Mathematical Dynamics and Fractals
Stability in the Erdős–Gallai Theorem on cycles and paths, II111This paper started
at SQuaRES meeting of the American Institute of Mathematics.
Zoltán Füredi Alfréd Rényi Institute of Mathematics, Hungary. E-mail: [email protected]. Research was supported in part by grant K116769 from the National Research, Development and Innovation Office NKFIH, by the Simons Foundation Collaboration Grant #317487, and by the European Research Council Advanced Investigators Grant 267195.
Alexandr Kostochka University of Illinois at Urbana–Champaign, Urbana, IL 61801 and Sobolev Institute of Mathematics, Novosibirsk 630090, Russia. E-mail: [email protected]. Research of this author is supported in part by NSF grant DMS-1266016 and by grants 15-01-05867 and 16-01-00499 of the Russian Foundation for Basic Research.
Ruth Luo University of Illinois at Urbana–Champaign, Urbana, IL 61801. E-mail: [email protected].
Jacques Verstraëte Department of Mathematics, University of California at San Diego, 9500 Gilman Drive, La Jolla, California 92093-0112, USA. E-mail: [email protected]. Research supported by NSF Grant DMS-1101489.
Abstract
The Erdős–Gallai Theorem states that for , any -vertex graph with no cycle of length at least has at most edges. A stronger version of the Erdős–Gallai Theorem was given by Kopylov: If is a 2-connected -vertex graph with no cycle of length at least , then , where . Furthermore, Kopylov presented the two possible extremal graphs, one with edges and one with edges.
In this paper, we complete a stability theorem which strengthens Kopylov’s result. In particular, we show that for odd and all , every -vertex -connected graph with no cycle of length at least is a subgraph of one of the two extremal graphs or . The upper bound for here is tight.
Mathematics Subject Classification: 05C35, 05C38.
Keywords: Turán problem, cycles, paths.
1 Introduction
One of the basic Turán-type problems is to determine the maximum number of edges in an -vertex graph with no -vertex path. Erdős and Gallai [3] in 1959 proved the following fundamental result on this problem.
Theorem 1.1** (Erdős and Gallai [3]).**
*Fix . If is an -vertex graph that does not contain a path with vertices, then . *
When is divisible by , the bound is best possible. Indeed, the -vertex graph whose every component is the complete graph has edges and no -vertex paths. Also, if is an -vertex graph without a -vertex path , then by adding to a new vertex adjacent to all vertices of we obtain an -vertex graph with edges that contains no cycle of length or longer. Then Theorem 1.1 follows from another theorem of Erdős and Gallai:
Theorem 1.2** (Erdős and Gallai [3]).**
*Fix . If is an -vertex graph that does not contain a cycle of length at least , then . *
The bound of this theorem is best possible for divisible by . Indeed, any connected -vertex graph in which every block is a has edges and no cycles of length at least . In the 1970’s, some refinements and new proofs of Theorems 1.1 and 1.2 were obtained by Faudree and Schelp [4, 5], Lewin [9], Woodall [10], and Kopylov [8] – see [7] for more details. The strongest version was proved by Kopylov [8]. His result is stated in terms of the following graphs. Let and . The -vertex graph is as follows. The vertex set of is the union of three disjoint sets and such that , and , and the edge set of consists of all edges between and together with all edges in (Fig. 1 shows ). Let
[TABLE]
For a graph , let denote the length of a longest cycle in . Observe that : Since , any cycle of at length at least has at least vertices in . But as is independent and , also has to contain at least neighbors of the vertices in , while only vertices in have neighbors in . Kopylov [8] showed that the extremal -connected -vertex graphs with no cycles of length at least are and : the first has more edges for small , and the second — for large .
Theorem 1.3** (Kopylov [8]).**
Let and . If is an -vertex -connected graph with , then
[TABLE]
with equality only if or .
2 Main results
2.1 A previous result
Recently, three of the present authors proved in [6] a stability version of Theorems 1.2 and 1.3 for -vertex -connected graphs with , but the problem remained open for when . The main result of [6] was the following:
Theorem 2.1** (Füredi, Kostochka, Verstraëte [6]).**
Let and and . Let be a -connected -vertex graph . Then unless
[TABLE]
Note that
[TABLE]
The paper [6] also describes the -connected -vertex graphs with for all .
2.2 The essence of the main result
Together with [6], this paper gives a full description of the 2-connected -vertex graphs with and ‘many’ edges for all and . Our main result is:
Theorem 2.2**.**
Let and , so that . If is a -connected graph on vertices and , then either or
[TABLE]
Note that the case is trivial and the case was fully resolved in [6].
2.3 A more detailed form of the main result
In order to prove Theorem 2.2, we need a more detailed description of graphs satisfying (b) in the theorem that do not contain ‘long’ cycles.
Let . Each is defined by a partition and two vertices , such that , , is the empty graph, is a complete bipartite graph, and for every . Every member of is defined by a partition such that , , is a complete bipartite graph, and
— has more than one component,
— all components of are stars with at least two vertices each,
— there is a -element subset of such that ,
— for every component of with at least vertices, all leaves of have degree 2 in and are adjacent to the same vertex in .
The class is empty unless . Each graph has a -vertex set such that and is a star forest such that if a component of has more than two vertices then all its leaves have degree 2 in and are adjacent to the same vertex in . These classes are illustrated below:
We can refine Theorem 2.2 in terms of the classes as follows:
Theorem 2.3**.**
(Main Theorem)* Let , and . Let be an -vertex -connected graph with no cycle of length at least . Then or is a subgraph of a graph in , where*
[TABLE]
Since every graph in and many graphs in have a separating set of size (see Fig. 3), the theorem implies the following simpler statement for -connected graphs:
Corollary 2.4**.**
Let where . If is a -connected graph on vertices and , then either or or and is a subgraph of some graph such that each component of has at most 2 vertices.
3 The proof idea
3.1 Small dense subgraphs
First we define some more graph classes. For a graph and a nonnegative integer , we denote by the family of graphs obtained from by deleting at most edges.
Let denote the complete bipartite graph with partite sets and where and . Let , i.e., the family of subgraphs of with at least edges.
Let denote the complete bipartite graph with partite sets and where and . Let , i.e., the family of subgraphs of with at least edges.
Let denote the family of graphs obtained from a graph in by subdividing an edge with a new vertex , where and . Note that any member has at least edges between and and the pair is not an edge.
Let denote the complete bipartite graph with partite sets and where and . Take a graph from , select two non-empty subsets , with such that if , add two vertices and , join them to each other and add the edges from to the elements of , (). The class of obtained graphs is denoted by . The family consists of these graphs when , and . In particular, consists of exactly one graph, call it .
Graph has vertex set , where and are disjoint. Its edges are the edges of the complete bipartite graph and three extra edges , , and (see Fig. 4 below). Define as the (only) member of such that , , and . Let , which is defined only for .
Define \mathcal{F}(k):=\left\{\begin{array}[]{ll}\mathcal{F}_{0},&\mbox{if kis odd},\\ \mathcal{F}_{1}\cup\dots\cup\mathcal{F}_{4},&\mbox{ifk is even.}\end{array}\right.
3.2 Proof idea
For our proof, it will be easier to use the stronger induction assumption that the graphs in question contain certain dense graphs from . We will prove the following slightly stronger version of Theorem 2.3 which also implies Theorem 2.2.
Theorem 2.3′ Let , , and . Let be an -vertex -connected graph with no cycle of length at least . Then or
[TABLE]
The method of the proof is a variation of that of [6]. Also, when is close to , we use Kopylov’s disintegration method. We take an -vertex graph satisfying the hypothesis of Theorem 2.3′, and iteratively contract edges in a certain way so that each intermediate graph still satisfies the hypothesis. We consider the final graph of this process on vertices and show that satisfies Theorem 2.3′. Two results from [6] will be instrumental. The first is:
Lemma 3.1** (Main lemma on contraction [6]).**
Let and suppose and are -connected graphs such that and . If contains a subgraph , then also contains a subgraph .
This lemma shows that if contains a subgraph , then the original graph also contains a subgraph in . The second result (proved in Subsection 4.5 of [6]) is:
Lemma 3.2** ([6]).**
Let , and let be a -connected graph with and . If contains a subgraph , then is a subgraph of a graph in .
We will split the proof into the cases of small and large . The following observations can be obtained by simple calculations (for ):
[TABLE]
In the case of large we will contract an edge such that the new graph still has more than edges. In order to apply induction, we also need the number of edges to be greater than . To guarantee this, we pick the cutoffs for the two cases and (therefore ).
4 Tools
4.1 Classical theorems
Theorem 4.1** (Erdős [2]).**
Let and be integers, and
[TABLE]
Then every -vertex graph with and is hamiltonian.
Theorem 4.2** (Chvátal [1]).**
Let and be an -vertex graph with vertex degrees . If is not hamiltonian, then there is some such that and .
Theorem 4.3** (Kopylov [8]).**
If is -connected and is an -path of vertices, then .
4.2 Claims on contractions
A helpful tool will be the following lemma from [6] on contraction.
Lemma 4.4** ([6]).**
Let and let be an -vertex -connected graph. For every , there exists such that is -connected.
For an edge in a graph , let denote the number of triangles containing . Let . When we contract an edge in a graph , the degree of every either does not change or decreases by . Also the degree of in is at least . Thus
[TABLE]
Similarly,
[TABLE]
We will use the following analog of Lemma 3.3 in [6].
Lemma 4.5**.**
*Let be a positive integer. Suppose a -connected graph is obtained from a -connected graph by contracting edge into chosen using the following rules:
one of , say is a vertex of the minimum degree in ;
is the minimum among the edges incident with such that is -connected. (Such edges exist by Lemma 4.4). If has at least vertices of degree at most , then either or
also has a vertex of degree at most , and
has at least vertices of degree at most .*
Proof. Since is -connected, . Let denote the set of vertices of degree at most in . Then by (2), each is also in . Moreover, then by (i),
[TABLE]
Thus if , then (b) follows. But if , then by (2), also . So, again (b) holds.
If , then (a) holds by (2). So, if (a) does not hold, then
[TABLE]
Case 1: . Then by (4), . This in turn yields . Since is -connected, each is not a cut vertex. Furthermore, is not a cut set. If it was, because is a common neighbor of all neighbors of , all neighbors of must be in the same component as in . It follows that
[TABLE]
If for some , then by (6) and (i), we would contract the edge rather than . Thus and so either or is a cut vertex in , as claimed.
Case 2: . Then . This means and . So by (5), there is such that . Again (6) holds (for the same reason that ). Thus similarly for every and every . Hence and either or is a cut vertex in , as claimed.
4.3 A property of graphs in
A useful feature of graphs in is the following.
Lemma 4.6**.**
Let and . Let be an -vertex graph contained in with . Then contains a graph in .
Proof.
Assume the sets to be as in the definition of . We will use induction on .
Case 1: . If , then because . Thus, since , contains a subgraph in . Suppose now the lemma holds for all . If , then each is adjacent to every . Hence contains . If , then since is dominating and , there is with . Then , and we are done by induction.
Case 2: . Let . If then as in Case 1,
[TABLE]
i.e., . Since , contains a subgraph in . Suppose now the lemma holds for all . If , then there is with . Then , and we are done by induction.
Finally, suppose . So, each is adjacent to every and each of has at least neighbors in . Since , contains a member of . Thus contains a member of unless , and has a nonneighbor . But then , and so contains either or .
5 Proof of Theorem 2.3′
5.1 Contraction procedure
If , we iteratively construct a sequence of graphs where . In [6], the following Basic Procedure (BP) was used:
At the beginning of each round, for some , we have a -vertex -connected graph with .
[TABLE]
Remark 5.1. By definition, (BP3) applies only when . As observed in [6], if , then a -vertex graph with a -vertex set separating the graph into at least components cannot have for every edge . It also was calculated there that if , then any -vertex graph with such -vertex set and for every edge has at most edges and so cannot be .
In this paper, we also use a quite similar Modified Basic Procedure (MBP): start with a -connected, -vertex graph with and . Then
[TABLE]
5.2 Proof of Theorem 2.3′ for the case
Apply to the Modified Basic Procedure (MBP) starting from . By Remark 1, (BP3) was never applied, since . Therefore at every step, we only contracted an edge. Denote by the terminating graph of (MBP). Then is -connected and for each . By construction, after each contraction, we lose at most edges. It follows that .
If , then the same argument as in [6] gives us the following structural result:
Lemma 5.1** ([6]).**
Let and .
- •
If , then .
- •
*If , then or . *
If and , then contains a subgraph in . Otherwise, by Lemma 4.6, again has a subgraph in . Next, undo the contractions that were used in (MBP). At every step for , by Lemma 3.1, contains some subgraph . In particular, contains such a subgraph. Thus by Lemma 3.2, we get our result. So, below we assume
[TABLE]
Since , does not have a hamiltonian cycle. Denote the vertex degrees of . By Theorem 4.2, there exists some such that and . Let be the smallest such .
Because has vertices of degree at most , similarly to [2],
[TABLE]
For , only when or , and for , when or . If , then repeating the argument in [6] yields:
Lemma 5.2** ([6]).**
If then .
Then by Lemma 4.6, Lemma 3.1, and Lemma 3.2, and contains some subgraph in . So we may assume that
[TABLE]
Our next goal is to show that contains a large “core”, i.e., a subgraph with large minimum degree. For this, we recall the notion of disintegration used by Kopylov [8].
Definition: For a natural number and a graph , the -disintegration of a graph is the process of iteratively removing from the vertices with degree at most until the resulting graph has minimum degree at least . This resulting subgraph will be called the -core of . It is well known that is unique and does not depend on the order of vertex deletion.
Claim 5.3**.**
The -core of is not empty.
Proof of Claim 5.3: We may assume that for all , graph was obtained from by contracting edge , where . By Rule (MBP2), , provided that .
By definition, . So by Lemma 4.5 (applied several times), for each , because each is not a complete graph (otherwise it would have a hamiltonian cycle),
[TABLE]
To show that
[TABLE]
by (9) and (8), it is enough to observe that
[TABLE]
We will apply a version of -disintegration in which we first manually remove a sequence of vertices and count the number of edges they cover. By (10) and (MBP2), . Let . Then is a subgraph of . If in , then let , otherwise let . In both cases, . We continue in this way until : each time we delete from the unique survived vertex that was in the preimage of when we obtained from . Graph has vertices of degree at most . We additionally delete 2 such vertices and . Altogether, we have lost at most edges in the deletions.
Finally, apply -disintegration to the remaining graph on vertices. Suppose that the resulting graph is empty.
Case 1: . Then
[TABLE]
where edges are from and , and after deleting and , every vertex deleted removes at most edges, until we reach the final vertices which altogether span at most edges.
For ,
[TABLE]
which is nonnegative for . Therefore , a contradiction.
Similarly, if ,
[TABLE]
[TABLE]
which is nonnegative when .
Case 2: . Then for ,
[TABLE]
where the last inequality holds because .
Similarly, for ,
[TABLE]
This contradiction completes the proof of Claim 5.3.
For the rest of the proof of Theorem 2.3′, we will follow the method of Kopylov in [8] to show that . Let be the -closure of . That is, add edges to until adding any additional edges creates a cycle of length at least . In particular, for any non-edge of , there is an -path in with at least edges.
Because has a nonempty -core, and contains as a subgraph, also has a nonempty -core (which contains the -core of ). Let denote the -core of . We will show that
[TABLE]
Indeed, suppose there exists a nonedge in . Choose a longest path of whose terminal vertices and are nonadjacent. By the maximality of , every neighbor of in is in , similar for . Hence , and also (edges). By Kopylov’ Theorem 4.3, must have a cycle of length at least , a contradiction.
Therefore is a complete subgraph of . Let . Because every vertex in has degree at least , . Furthermore, if , then has a clique of size at least . Because is -connected, we can extend a -cycle of to include at least one vertex in , giving us a cycle of length at least . It follows that
[TABLE]
and therefore . Apply a weaker -disintegration to , and denote by the resulting graph. By construction, .
Case 1: There exists . Since , there exists a nonedge between a vertex in and a vertex in . Pick a longest path with terminal vertices and . Then , and therefore .
Case 2: . Then
[TABLE]
If , then , so by (12), , and is the complete graph with vertices. Let . If there is an edge in , then because is -connected, there exist two vertex-disjoint paths, and , from to such that and only intersect at the beginning and end of the paths. Let and be the terminal vertices of and respectively that lie in . Let be any -hamiltonian path of . Then is a cycle of length at least in , a contradiction.
Therefore is an independent set, and since is -connected, each vertex of has degree 2. Suppose there exists where . Let where it is possible that . Then we can find a cycle of that covers which contains edges and . Then is a cycle of length in . Thus for every , for some . I.e., , and thus .
5.3 Proof of Theorem 2.3′ for all
We use induction on with the base case . Suppose and for all , Theorem 2.3′ holds. Let be a -connected graph with vertices such that
[TABLE]
Apply one step of (BP). If (BP4) was applied (so neither (BP2) nor (BP3) applies to ), then (with defined as in the previous case). By Lemmas 5.1, 4.6, and 3.2, the theorem holds.
Therefore we may assume that either (BP2) or (BP3) was applied. Let be the resulting graph. Then , and is -connected.
Claim 5.4**.**
[TABLE]
Proof.
If (BP2) was applied, i.e., for some edge , then
[TABLE]
so (14) holds. Therefore we may assume that (BP3) was applied to obtain . Then and . So by (13),
[TABLE]
The right hand side of (15) equals which is at least for , proving the first part of (14).
We now show that also . Indeed, for ,
[TABLE]
[TABLE]
Similarly, for ,
[TABLE]
[TABLE]
Thus if , then (14) is proved. But if then by Remark 5.1, no graph to which (BP3) applied may have more than edges.
By (14), we may apply induction to . So satisfies either (a) , or (b) is contained in a graph in and contains a subgraph . Suppose first that satisfies (b). If (BP3) was applied to obtain from , then because contains a subgraph and , also contains . If (BP2) was applied, then by Lemma 3.1, contains a subgraph . In either case, Lemma 3.2 implies that is a subgraph of a graph in .
So we may assume that (a) holds, that is, is a subgraph of . Because , , and so has edges in at most triangles. Therefore (BP2) was applied to obtain , where . Let be an independent set of vertices of of size with for some . Since for every , we have that with equality only if where .
We want to show that . If not, suppose first that . Then there exists , and and are not adjacent in . Therefore was not in a triangle with and in , and hence , so the edge should have been contracted instead. Otherwise if , at least one of , say , is not . If , then for every , , therefore each such edge was in a triangle with in . Then , a contradiction.
Thus and . But for , we have , a contradiction.
Acknowledgment. The authors thank Zoltán Király for helpful comments.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] V. Chvátal, On Hamilton’s ideals. J. Combinatorial Theory Ser. B 12 (1972), 163–168.
- 2[2] P. Erdős, Remarks on a paper of Pósa, Magyar Tud. Akad. Mat. Kutató Int. Közl. 7 (1962), 227–229.
- 3[3] P. Erdős and T. Gallai, On maximal paths and circuits of graphs, Acta Math. Acad. Sci. Hungar. 10 (1959), 337–356.
- 4[4] R. J. Faudree and R. H. Schelp, Ramsey type results, Infinite and Finite Sets, Colloq. Math. J. Bolyai 10 , (ed. A. Hajnal et al.), North-Holland, Amsterdam, 1975, pp. 657–665.
- 5[5] R. J. Faudree and R. H. Schelp, Path Ramsey numbers in multicolorings, J. Combin. Theory Ser. B 19 (1975), 150–160.
- 6[6] Z. Füredi, A. Kostochka, and J. Verstraëte. Stability in the Erdős–Gallai Theorem on cycles and paths, JCTB 121 (2016), 197–228.
- 7[7] Z. Füredi and M. Simonovits, The history of degenerate (bipartite) extremal graph problems, Bolyai Math. Studies 25 pp. 169–264, Erdős Centennial (L. Lovász, I. Ruzsa, and V. T. Sós, Eds.) Springer, 2013. Also see: ar Xiv:1306.5167 .
- 8[8] G. N. Kopylov, Maximal paths and cycles in a graph, Dokl. Akad. Nauk SSSR 234 (1977), 19–21. (English translation: Soviet Math. Dokl. 18 (1977), no. 3, 593–596.)
