Degree lists and connectedness are $3$-reconstructible for graphs with at least seven vertices
Alexandr V. Kostochka, Mina Nahvi, Douglas B. West, Dara Zirlin

TL;DR
This paper proves that for graphs with at least seven vertices, the degree list and connectedness can be uniquely reconstructed from specific subgraph collections, extending previous results and establishing sharp thresholds.
Contribution
It establishes that degree lists and connectedness are 3-reconstructible for graphs with at least seven vertices, improving upon earlier 2-reconstructibility results.
Findings
Degree list is 3-reconstructible for n ≥ 7
Connectedness is 3-reconstructible for n ≥ 7
Results extend and sharpen previous 2-reconstructibility findings
Abstract
The -deck of a graph is the multiset of its subgraphs induced by vertices. A graph or graph property is -reconstructible if it is determined by the deck of subgraphs obtained by deleting vertices. We show that the degree list of an -vertex graph is -reconstructible when , and the threshold on is sharp. Using this result, we show that when the -deck also determines whether an -vertex graph is connected; this is also sharp. These results extend the results of Chernyak and Manvel, respectively, that the degree list and connectedness are -reconstructible when , which are also sharp.
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.
Degree lists and connectedness are -reconstructible for graphs
with at least seven vertices
Alexandr V. Kostochka , Mina Nahvi , Douglas B. West , Dara Zirlin University of Illinois at Urbana–Champaign, Urbana IL 61801, and Sobolev Institute of Mathematics, Novosibirsk 630090, Russia: [email protected]. Research supported in part by NSF grants DMS-1600592 and grants 18-01-00353A and 16-01-00499 of the Russian Foundation for Basic Research.University of Illinois at Urbana–Champaign, Urbana IL 61801: [email protected] Normal University, Jinhua, China 321004 and University of Illinois at Urbana–Champaign, Urbana IL 61801: [email protected]. Research supported by National Natural Science Foundation of China grant NNSFC 11871439.University of Illinois at Urbana–Champaign, Urbana IL 61801: [email protected].
Abstract
The -deck of a graph is the multiset of its subgraphs induced by vertices. A graph or graph property is -reconstructible if it is determined by the deck of subgraphs obtained by deleting vertices. We show that the degree list of an -vertex graph is -reconstructible when , and the threshold on is sharp. Using this result, we show that when the -deck also determines whether an -vertex graph is connected; this is also sharp. These results extend the results of Chernyak and Manvel, respectively, that the degree list and connectedness are -reconstructible when , which are also sharp.
MSC Codes: 05C60, 05C07
Key words: graph reconstruction, deck, reconstructibility, connected
1 Introduction
A card of a graph is a subgraph of obtained by deleting one vertex. Cards are unlabeled, so only the isomorphism class of a card is given. The deck of is the multiset of all cards of . A graph is reconstructible if it is uniquely determined by its deck. The famous Reconstruction Conjecture was first posed in 1942.
Conjecture 1.1** **(The Reconstruction Conjecture; Kelly [9, 10],
Ulam [21]).
Every graph having more than two vertices is reconstructible.
The two graphs with two vertices have the same deck. Graphs in many families are known to be reconstructible; these include disconnected graphs, trees, regular graphs, and perfect graphs. Surveys on graph reconstruction include [4, 5, 11, 12, 13].
Various parameters have been introduced to measure the difficulty of reconstructing a graph. Harary and Plantholt [8] defined the reconstruction number of a graph to be the minimum number of cards from its deck that suffice to determine it, meaning that no other graph has the same multiset of cards in its deck (surveyed in [2, 17]). Kelly looked in another direction, considering cards obtained by deleting more vertices. He conjectured a more detailed version of the Reconstruction Conjecture.
Conjecture 1.2** (Kelly [10]).**
For , there is an integer such that any graph with at least vertices is reconstructible from its deck of cards obtained by deleting vertices.
The original Reconstruction Conjecture is the claim .
A -card of a graph is an induced subgraph having vertices. The -deck of , denoted , is the multiset of all -cards. When discussing reconstruction from the -deck, we will refer to -cards simply as cards.
Definition 1.3**.**
A graph is -deck reconstructible if implies . A graph (or a graph invariant) is -reconstructible if it is determined by (agreeing on all graphs having that deck). The reconstructibility of , written , is the maximum such that is -reconstructible.
For an -vertex graph, “-deck reconstructible” and “-reconstructible” have the same meaning when . Kelly’s conjecture is that for any , all sufficiently large graphs are -reconstructible. Let and be the graphs obtained from the claw by subdividing one or two edges, respectively. The -vertex graphs and are not -reconstructible, since they have the same -deck. Having checked by computer that every graph with at least six and at most nine vertices is -reconstructible, McMullen and Radziszowski [15] asked whether . With computations up to nine vertices, Rivshin and Radziszowski [18] conjectured .
Some results about reconstruction have been extended to the context of reconstruction from the -deck. For example, almost every graph is reconstructible from any set of three cards in the deck of cards obtained by deleting one vertex (see [3, 7, 16]). Spinoza and West [19] proved more generally that for , almost all graphs are -reconstructible using only cards that omit vertices. Among other results, they also determined exactly for every graph with maximum degree at most .
Since each induced subgraph with vertices arises exactly times by deleting one vertex from a member of , we have the following.
Observation 1.4**.**
For any graph , the -deck determines the -deck .
By Observation 1.4, information that is -deck reconstructible is also -deck reconstructible when . This motivates the definition of reconstructibility; if is -reconstructible, then is also -reconstructible, so we seek the largest such .
Manvel [14] proved for that the -deck of an -vertex graph determines whether the graph satisfies the following properties: connected, acyclic, unicyclic, regular, and bipartite. For the first three of these properties, sharpness of the threshold on is shown by the graphs and mentioned above. Spinoza and West [19] extended Manvel’s result by showing that connectedness is -reconstructible when . Using a somewhat different approach, we extend their result.
Theorem 1.5**.**
For , connectedness is -reconstructible for -vertex graphs, and the threshold on is sharp.
The threshold is sharp because and have the same -deck. For general , the known upper and lower bounds on the threshold for to guarantee that connectedness of -vertex graphs is -reconstructible are quite far apart. Spinoza and West [19] proved that connectedness is -reconstructible when . As a lower bound, we know only that is needed, since and have the same -deck [19]. Indeed, is the only -vertex graph whose reconstructibility is known to be less than .
One of the first easy results in ordinary reconstruction is that the degree list of a graph with at least three vertices is -reconstructible. Manvel [14] showed that the degree list is reconstructible from the -deck when the maximum degree is at most . With no restriction on the maximum degree, Taylor showed that the degree list is reconstructible from the -deck when the number of vertices is not too much larger than , regardless of the value of the maximum degree.
Theorem 1.6** (Taylor [20]).**
If and , then the degree list of any -vertex graph is determined by its -deck, where
[TABLE]
and denotes the base of the natural logarithm. Thus the degree list is -reconstructible when .
For small , one can obtain exact thresholds. Chernyak [6] proved that the degree list is -reconstructible when ; again the example of and shows that this is sharp. We extend this to -reconstructibility.
Theorem 1.7**.**
For , any two graphs of order that have the same -deck have the same degree list, and this threshold on is sharp.
Again the example of and proves sharpness. We use Theorem 1.7 as a tool in the proof of Theorem 1.5. With Chernyak’s result being somewhat inaccessible, we also obtain it and Manvel’s result on -reconstructibility of connectedness as corollaries of our results.
2 -reconstructibility of degree lists
We begin with a basic counting tool used also by Manvel [14] and by Taylor [20]. In a graph , we refer to a vertex of degree as a -vertex.
Lemma 2.1**.**
Let denote the total number of -vertices over all cards in the -deck of an -vertex graph . Letting denote the number of -vertices in (and ),
[TABLE]
Proof.
In each card, each vertex counted by has degree at least in . When that degree is , the vertex in the reconstructed graph contributes exactly to the computation of . This contribution is [math] when ; the vertex then does not have enough nonneighbors in the full graph to occur with degree exactly in a card. Thus we require .
Corollary 2.2** (Manvel [14]).**
From the -deck of a graph and the numbers of vertices with degree for all at least , the degree list of the graph is determined.
Proof.
Since the -deck determines the -deck, using induction it suffices to show that knowing both and for determines . Simply solve for in the expression (1) for obtained by setting .
With these tools, we prove Theorem 1.7, which we restate.
Theorem** **(1.7).
For , any two graphs of order that have the same -deck have the same degree list, and this threshold on is sharp.
Proof.
For sharpness, the -decks of both and consist of five copies of , ten copies of , and five copies of .
Given , let be the -deck of an -vertex graph. We show that all reconstructions from have the same degree list.
Let and be reconstructions from . Since determines the -deck, we know the common number of edges in and ; let it be . We may assume , since otherwise we can analyze the complements of and .
We will use repeatedly the fact that any vertices whose degrees sum to at least are together incident to at least edges.
Let and be the numbers of -vertices in and , respectively, and let . The computation in (1) is valid using either or , producing the same value from . Hence the difference of the two instances of (1) yields
[TABLE]
since here . We will be interested in particular in the cases (dominating vertices on cards) and , which we write explicitly as
[TABLE]
and
[TABLE]
The observation of Manvel (Corollary 2.2) implies that if and have different degree lists, then for some with . Let be the largest such index. By symmetry, we may assume . We consider cases depending on the value of .
Case 1: . In this case and . By (3), . Since when , we have and . Now (4) implies . Thus has at least vertices, but requires . Hence and has degree list exactly , and has no vertices of degree or at least . Furthermore , so has exactly four vertices with degree and cannot reach the same degree-sum as .
Case 2: . Now and . Let . By (3), , so . With and and , this can only be satisfied when , , and . Since , the degree-sum is at most ; hence , which requires . Since we have obtained , (4) yields ; this requires , a contradiction when . Hence we conclude .
With , we have . Hence
[TABLE]
Substituting into (4) yields
[TABLE]
Since must be an integer, by (5) there are not many possibilities for . Let . Since , we have when is even, and when is odd. Also (2) simplifies to .
With and , the possibilities that remain for are , , and . Note that . In the even cases, . When , five -vertices are together incident to at least edges, which is more than .
When , four -vertices in are together incident to at least edges, which is the maximum allowed, so there can be no other edges or other -vertices, the four -vertices induce , and eight edges join these vertices to the rest. Since , in the vertices of degree at least already contribute to the degree-sum, so has no -vertex. With , also has no -vertex. Hence the degree list of is . With , applying (2) with now yields , a contradiction.
When , we have , and . Hence , and having equal degree-sum requires . Now has six vertices with degrees and has six vertices with degrees , and they each have one more vertex of the same odd degree. Since the degree list of must be realizable, the only choice is for and for . Now is realized only by adding three pendant edges to , so is a card in , which can be obtained from only on the four vertices of high degree. Thus consists of copies of and sharing one vertex, plus an isolated vertex. Being the union of three complete graphs, has no independent set of size , but does have such a set, so their -decks cannot be equal.
Case 3: . If , then . Since this exceeds when , we conclude .
Let . If , then (3) and together yield . The contribution to degree-sum in from vertices of degrees and is now at least , which exceeds when . Hence , but then having two -vertices in requires at least edges (more than ), a contradiction. Thus we may assume .
With , (3) yields . If , then . Dividing by and using yields , which requires . If , then simplifies (3) to , but and yield . If , then (3) simplifies to , but yields . Since , we conclude and . With at most edges, . Now has five cards that are . With only four edges not incident to its dominating vertex, cannot have five such cards. We conclude .
With , we now break into subcases by the value of . We have already proved . Let , so and (3) yields
[TABLE]
Substituting (7) into (4) yields
[TABLE]
Subcase 3.1: . If , then with the vertices of degrees and in are incident to at least edges. Hence ; this exceeds when . If , then , by (7). Also , since otherwise (7) yields . If , then , again too many vertices when (since ).
Hence . Now . This quantity exceeds when . For , we have , , , yielding degree-sum already , so has degree list , but degree forbids two isolated vertices. For , we have , so even degree-sum at most requires degree list . To avoid higher degree-sum, for . Hence for these values. Now having one -vertex requires to reach degree-sum , contradicting .
Subcase 3.2: . If , then when by (8), a contradiction. If , then , so . If , then , so . Setting and and in (8) yields , so . Now (7) yields , so . With by (4), we have . Now has at least vertices, a contradiction.
Hence . Since also , the number of edges in incident to vertices of degree at least is at least , which simplifies to and is more than when .
Subcase 3.3: . Note that . By (7), , so . By (8), . With , this yields when , a contradiction. Since , only remains.
With , the expressions above reduce to , , and , with . If , then . If , then has three vertices of degrees and such that the number of edges incident to them is at least , which equals and exceeds .
The remaining case is and , also and . Since we know the -deck, and have the same degree-sum; that is, . We have ; hence . Now and , which contradicts .
Subcase 3.4: . Here , so . The number of edges incident to vertices of degree at least in is at least , which exceeds when and equals it when . For with , (7) reduces to and (8) reduces to , which requires . Hence , which with gives degree-sum at least , contradicting .
Using Theorem 1.7, we present an alternative proof of the result by Chernyak on the threshold for -reconstructibility of the degree list.
Corollary 2.3** (Chernyak [6]).**
The degree list of an -vertex graph is -reconstructible whenever , and this is sharp.
Proof.
Since the -deck determines the -deck, it is immediate from Theorem 1.7 that the degree list is -reconstructible when . By the example of and , is not sufficient. It remains only to consider .
Let and be two -vertex graphs having the same -deck but different degree lists. Let (we know the -deck). Since the -deck determines the -deck of the complement and , we may assume . Define as in Theorem 1.7. That is, with , different degree lists in and require a largest with such that , and by symmetry we have . We use the equation for , which counts dominating vertices in the cards of the -deck:
[TABLE]
Case 1: . We have , because two -vertices in already force . Thus , by (9). If , then . If , then also and . However, . If , then has too many vertices.
Case 2: . Here . With , we have and . With degree-sum at most , the degree list of is with . Thus also and , so has only one vertex with degree exceeding . If , then and is a card, but is not contained in .
Hence , and the degree list of must be . The only such graph consists of a -cycle and a -cycle with one common vertex. Every card of has at most four edges. Whether is or , deleting from the two vertices of smallest degree eliminates at most two edges and leaves a card with five edges, a contradiction.
3 -reconstructibility of connectedness
Using Theorem 1.7, we prove Theorem 1.5. Again the example of and shows that the threshold on is sharp; they have the same -deck, but only one is connected.
Theorem** **(1.5).
For , connectedness is -reconstructible for -vertex graphs, and the threshold on is sharp.
Proof.
Suppose that -vertex graphs and have the same -deck , but that is connected and is disconnected. Let be the common number of edges in and . Let be the largest component in . Since is connected, it has a spanning tree . Since , has at least two connected cards. Thus has at least two connected cards, so has at least vertices.
By Theorem 1.7, and have the same degree list. Since is connected, cannot have an isolated vertex, so . If has a -vertex, then deleting it and the vertices of the small component in leaves a card in with edges. However, since is connected, it is not possible to delete three vertices in and only remove two edges. Hence has no -vertex, which means that and each have exactly two -vertices. Let and be the -vertices in , and let be the set of -vertices in .
Let be the number of -vertices in both and in . If , then has minimum degree at least . Deleting and one vertex of from now yields cards with minimum degree at least . Such cards can arise from only by deleting the two -vertices and one other vertex. Hence and have the same -deck. They must therefore have the same number of edges. However, has edges, while has edges. Thus .
To eliminate only three edges from when deleting three vertices, one must delete and a -vertex of . Thus is also the number of cards in with edges. We show the remaining possibilities for in Figure 1.
If and have the same neighbor, , then can have a card with edges only if has degree and the deleted set is . Hence in this case .
If and have different neighbors, then each of and is the end of a maximal path containing no vertices of degree larger than in ; call these paths and . We can only obtain a card with edges by deleting vertices from and vertices from , where . There are at most four choices for , so . In order to have exactly cards with edges, there must be a total of vertices of degree on and hence no -vertices elsewhere in (See Figure 1).
Now consider the cards of obtained by removing three vertices. When , the paths and together have at least four vertices of degree at most , so removing any three vertices of leaves a vertex of degree at most . Hence removing and a vertex of from must also leave a vertex of degree at most . This means that every vertex of has a neighbor of degree . In the two possibilities when , the one card of with edges may have no vertex of degree at most , but all other cards must have such a vertex. In this case every vertex of except possibly one has a neighbor of degree .
For , label and so that . Consider a card of with edges that is obtained by deleting , and the neighbor of , so has two vertices of degree and vertices of degree . Since all -vertices in are in , the other vertices in have degree at least . Note that must be a vertex-deleted subgraph of , since cards with edges are obtained from only by deleting and a vertex of . Since must have vertices of degree and none of degree , it must be formed from by adding one vertex of degree whose neighbors are the two -vertices in . Adding to form shows that the -vertices in lie along a single path. This means that only two vertices outside this path can have neighbors of degree . Since every vertex of must have a neighbor of degree , we conclude that has at most two vertices outside the path, but then those vertices cannot have degree greater than , a contradiction.
When , recall that every vertex in has a neighbor of degree (including the vertices of degree ). Each vertex of degree is a neighbor of only two vertices. Hence , so . Similarly, when , all but one vertex of has a neighbor of degree , so , yielding .
We have obtained contradictions in all cases, so such and do not exist.
Using Theorem 1.5, Manvel’s result on -reconstructibilty of connectedness follows quite easily.
Corollary 3.1** (Manvel [14]).**
For , connectedness of an -vertex graph is -reconstructible.
Proof.
Again and give sharpness, and Theorem 1.5 handles . Consider connected and disconnected -vertex graphs and with the same -deck.
By Corollary 2.3, and have the same degree list, so neither has isolated vertices. Since has a connected -card, has a -vertex component , and . Thus has only one connected -card.
Now must also have only one connected -card. Therefore every spanning tree of is a path, so is a path, but then has three connected -cards.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1]
- 2[2] K. J. Asciak, M. A. Francalanza, J. Lauri, and W. Myrvold, A survey of some open questions in reconstruction numbers, Ars Combin. 97 (2010), 443–456.
- 3[3] B. Bollobás, Almost every graph has reconstruction number three, J. Graph Theory 14 (1990), 1–4.
- 4[4] J. A. Bondy, A graph reconstructor’s manual, in Surveys in Combinatorics (Guildford, 1991) , Lond. Math. Soc. Lec. Notes 166 (Cambridge U. Press, 1991), 221–252.
- 5[5] J. A. Bondy and R. L. Hemminger, Graph reconstruction—a survey, J. Graph Theory 1 (1977), 227–268.
- 6[6] Zh. A. Chernyak, Some additions to an article by B. Manvel: ”Some basic observations on Kelly’s conjecture for graphs” (Russian), Vests ı ¯ ¯ italic-ı \overline{\i} Akad. Navuk BSSR Ser. Fīz.-Mat. Navuk (1982), 44–49, 126.
- 7[7] P. Chinn, A graph with p 𝑝 p points and enough distinct ( p − 2 ) 𝑝 2 (p-2) -order subgraphs is reconstructible, Recent Trends in Graph Theory Lecture Notes in Mathematics 186 (Springer, 1971).
- 8[8] F. Harary and M. Plantholt, The graph reconstruction number, J. Graph Theory 9 (1985), 451–454.
