This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Graphs with many hamiltonian paths

Erik Carlson Willem Fletcher MurphyKate Montee Chi Nguyen Jarne Renders  and  Xingyi Zhang
Abstract.

A graph is hamiltonian-connected if every pair of vertices can be connected by a hamiltonian path, and it is hamiltonian if it contains a hamiltonian cycle. We construct families of non-hamiltonian graphs for which the ratio of pairs of vertices connected by hamiltonian paths to all pairs of vertices approaches 1. We then consider minimal graphs that are hamiltonian-connected. It is known that any order-nn graph that is hamiltonian-connected must have 3n/2\geq 3n/2 edges. We construct an infinite family of graphs realizing this minimum.

1. Introduction

A hamiltonian path in a graph is a path which visits every vertex exactly once. A graph is hamiltonian if it admits a hamiltonian cycle, and homogeneously traceable if every vertex of GG is the starting vertex of a hamiltonian path. If every pair of vertices in GG is connected with a hamiltonian path, then GG is hamiltonian-connected. This class of graphs was introduced by Ore [Ore63] in 1963, and has been well studied since then. Every hamiltonian-connected graph must be hamiltonian (as long as it has at least 3 vertices), and every hamiltonian graph must be homogeneously traceable.

It is natural to explore the reverse implications; more generally, one might ask ‘how many hamiltonian paths can a non-hamiltonian graph contain?’ Every cycle graph is hamiltonian but not hamiltonian-connected, and the Petersen graph is (perhaps unsurprisingly) an example of a homogeneously traceable non-hamiltonian (HTNH) graph. In fact, Chartrand, Gould, and Kapoor show in [CGK79] that there exist HTNH graphs of any order n9n\geq 9. Hu and Zhan extended this in [HZ22] to find families of regular HTNH graphs. In all of these papers, the authors construct graphs by starting with some base graph and building new graphs (with larger vertex sets) inductively.

In this paper, we consider graphs with ‘many’ hamiltonian paths, in the sense that the number of pairs of vertices connected by a hamiltonian path is large. We say that a graph GG is kk-pair-strung if kk pairs of vertices can be connected by a hamiltonian path. Thus a hamiltonian-connected graph of order nn is (n2)\binom{n}{2}-pair-strung, and a homogeneously traceable graph is (at least) (n/2)(n/2)-pair-strung.

In Section 2 we consider non-hamiltonian graphs which are nevertheless highly pair-strung; in particular, we let rGr_{G} denote the fraction of pairs of vertices of GG which are connected by a hamiltonian path, so that in a graph GG with order nn,

rG=max{k(n2)|G is k-pair-strung}.r_{G}=\max\left.\left\{\frac{k}{\binom{n}{2}}\;\right|\;G\mbox{ is }k\mbox{-pair-strung}\right\}.

We first show that there exists a family of non-hamiltonian graphs 𝒢\mathcal{G} for which supG𝒢{rG}=1\sup_{G\in\mathcal{G}}\{r_{G}\}=1. We then investigate possible values of rGr_{G}; we describe an explicit construction that produces families of graphs for which rGr_{G} approaches k1k\frac{k-1}{k}. This construction depends on finding a collection of edges in a non-hamiltonian graph for which every edge is connected by a hamiltonian path containing the other edges. We call such edge sets H-path connected (see Definition 2.9, Theorem 2.14).

In Section 3 we consider another extreme; rather than limiting ourselves to non-hamiltonian graphs, we construct an infinite family of hamiltonian graphs containing a minimal edge set (see Theorem 3.2).

1.1. Acknowledgements

The authors would like to thank Xingzhi Zhan for alerting them to [CGK79, Sku84]; a previous version of this manuscript reproduced several of their results.

2. Nearly Hamiltonian-Connected Non-Hamiltonian Graphs

In this section, we consider graphs that have many pairs of vertices connected by a hamiltonian path. We will quantify this in the following definition.

Definition 2.1.

A graph GG is kk-pair-strung if kk pairs of the vertices in GG have a hamiltonian path between them.

Note that the existence of HTNH graphs implies that there exist non-hamiltonian graphs which are at least (n/2)(n/2)-pair-strung. Unsurprisingly, we can also find non-hamiltonian graphs with fewer hamiltonian paths.

Example 2.2.

Let n>m>0n>m>0. Consider the graph GG obtained by attaching a path with nm+1n-m+1 vertices, Pnm+1,P_{n-m+1}, to a complete graph on mm vertices so that KmPnm+1={v}K_{m}\cap P_{n-m+1}=\{v\} is one of the endpoints of Pnm+1P_{n-m+1}. (This is illustrated in Figure 1). Let ww be the other endpoint of Pnm+1P_{n-m+1}. Then every hamiltonian path in GG begins (or ends) at ww, and ends (or begins) at one of the vertices in V(Km)\{v}.V(K_{m})\backslash\{v\}. In particular, exactly mm vertices of GG are an endpoint of a hamiltonian path, and exactly mm pairs of vertices of GG are are connected by a hamiltonian path. Furthermore, GG is non-hamiltonian since no cycle can contain ww.

vvPnm+1P_{n-m+1}wwKmK_{m}
Figure 1. The graph illustrated here has nn vertices. Every vertex in KmK_{m} except vv is an endpoint of a hamiltonian path, and all hamiltonian paths begin (or end) at ww.

For a graph, GG, we define the pair connected ratio, rGr_{G}, as the fraction of pairs of vertices that have a hamiltonian path between them. More precisely, we have the following definition:

Definition 2.3.

Let GG be a graph of order nn. Let the pair connected ratio of GG be defined as

rG=max{k(n2)|G is k-pair-strung}.r_{G}=\max\left.\left\{\frac{k}{\binom{n}{2}}\;\right|\;G\mbox{ is }k\mbox{-pair-strung}\right\}.

For any nn, let Rn=sup{rG|G is a non-hamiltonian graph of order n}R_{n}=\sup\{r_{G}|G\mbox{ is a non-hamiltonian graph of order }n\}, and let R=sup{Rn|n3}R=\sup\{R_{n}|n\geq 3\}.

We will first establish an upper bound on RnR_{n}.

Theorem 2.4.

For n>3n>3, Rnn2nR_{n}\leq\frac{n-2}{n}.

Proof.

Consider a non-hamiltonian graph, GG, with at least one hamiltonian path γ\gamma and n>3n>3 vertices. Label the vertices v1,v2,,vnv_{1},v_{2},\dots,v_{n} in order along γ\gamma. If γ\gamma is the only hamiltonian path in GG then rG=2n(n1)<n2nr_{G}=\frac{2}{n(n-1)}<\frac{n-2}{n} when n>3n>3. Suppose instead that GG has at least one other hamiltonian path, γ=vi1,vi2,,vin\gamma^{\prime}=v_{i_{1}},v_{i_{2}},\dots,v_{i_{n}}. If i1=ki_{1}=k and in=k±1i_{n}=k\pm 1, then γ\gamma is a hamiltonian cycle. Thus any hamiltonian path in GG can not start and end on vertices with labels differing by one. The total number of pairs of vertices which are non-adjacent on γ\gamma is at most (n1)(n2)2\frac{(n-1)(n-2)}{2}. Dividing by the number of pairs of vertices, (n2)\binom{n}{2}, we get rGn2nr_{G}\leq\frac{n-2}{n}. Thus Rnn2n.R_{n}\leq\frac{n-2}{n}.

It is clear from the definition that R1R\leq 1; in fact, we can show that R=1R=1 by considering maximally non-hamiltonian (MNH) graphs; that is, a graph GG where GG is non-hamiltonian but G+eG+e is hamiltonian for every edge ee in the complement of GG.

Theorem 2.5.

R = 1.

Proof.

Note that if GG is MNH, then every pair of non-adjacent vertices in GG must be connected by a hamiltonian path. Therefore an MNH graph GG with nn vertices and mm edges must be ((n2)m)\left(\binom{n}{2}-m\right)-pair-strung. There exist MNH graphs of order n19n\geq 19 with 3n/2\lceil 3n/2\rceil edges (see [CES92, CE83, Lin+97]). Let {Gn}\{G_{n}\} be a family of such graphs. Then we have:

limnrGn=limnn(n1)3nn(n1)=1.\lim_{n\to\infty}r_{G_{n}}=\lim_{n\to\infty}\frac{n(n-1)-3n}{n(n-1)}=1.

The rest of this section offers an alternative proof of Theorem 2.5, and is motivated by the following question:

Question 2.6.

What are the possible values of rGr_{G}?

We conjecture that for every rational number q[0,1)q\in[0,1) there is some graph GG so that rG=q.r_{G}=q.

Conjecture 2.7.

For every nn there exists a number p(n)p(n) so that for every kp(n)k\leq p(n) there exists a kk-pair-strung graph of order nn which is not (k+1)(k+1)-pair-strung, and there is no kk-pair-strung non-hamiltonian graph of order nn for any k>p(n)k>p(n).

Using this language, we can ask further questions.

Question 2.8.

How does p(n)p(n) grow as a function of nn?

For small values of nn we have the data shown in Figure 2, which were obtained by computer search.

nn 4 5 6 7 8 9 10 11
p(n)p(n) 2 4 6 9 13 18 25 34
Figure 2. Value of p(n)p(n) for small numbers nn.

For the rest of this section we describe a construction that produces a family of graphs with rGn1nr_{G}\to\frac{n-1}{n} for any natural number nn.

Definition 2.9.

A set of edges S={e1,e2,,en}S=\{e_{1},e_{2},\dots,e_{n}\} in a graph GG is H-path connected if for every iji\neq j, there is a hamiltonian path with its endpoints in eie_{i} and eje_{j} and which contains every edge in S{ei,ej}.S-\{e_{i},e_{j}\}.

Finding sets of H-path connected edges is, in general, computationally difficult. One example of such a set is any collection of five pairwise disjoint edges in the Petersen graph. A larger example is demonstrated below.

Example 2.10.

Consider the graph GG illustrated in Figure 3. This graph is the smallest known almost hypohamiltonian graph (see [Zam15]). That is, it is non-hamiltonian and for all but one vertex vv, the graph GvG-v is hamiltonian. (In this case, the exceptional vertex is the central vertex.) The set of seven red edges indicated in Figure 3 is an H-path connected set of edges. The hamiltonian paths realizing this are indicated in green.

Refer to caption
Figure 3. A graph with an H-path connected edge set of size seven (indicated in red). The hamiltonian paths connecting each pair of edges are shown in green. Note that some of the pairs of edges are not illustrated because of the reflective symmetry of the graph.

We use a non-hamiltonian graph containing an H-path connected set of edges to construct a family of non-hamiltonian graphs which are highly pair-strung as follows.

Construction 2.11.

Let GG be a non-hamiltonian graph and let kk\in\mathbb{N}. Let S={e1,,en}S=\{e_{1},\dots,e_{n}\} be a set of H-path connected edges. Label the endpoints of each edge eie_{i} with vi,wiv_{i},w_{i}. We construct GkG_{k} as follows: To each edge eie_{i}, attach a complete graph on kk vertices, labeled KiK^{i}, so that every vertex of KiK^{i} is adjacent to viv_{i} and wiw_{i}.

We first prove some properties of the graphs GkG_{k}.

Lemma 2.12.

Let GG be a non-hamiltonian with a set of edges SS which is H-path connected. For each kk\in\mathbb{N} the graph GkG_{k} as in Construction 2.11 is non-hamiltonian.

Proof.

Assume (toward contradiction) that GkG_{k} is hamiltonian. Then there exists a hamiltonian cycle λ\lambda^{\prime} in GkG_{k}. Up to cyclic permutations, we may assume that the first vertex of λ\lambda^{\prime} is v1v_{1}. By construction, we know that vertices in each KiK^{i} are only adjacent to viv_{i} and wiw_{i} in GG. Therefore, between any pair of vertices viv_{i} and wiw_{i}, λ\lambda^{\prime} must pass through all the vertices of KiK^{i}. So up to reversing the orientation of λ\lambda^{\prime} we may assume that the next vertex of λ\lambda^{\prime} is in K1K^{1}, and the first vertex of λ\lambda^{\prime} after leaving K1K^{1} is w1w_{1}.

Consider the path λ\lambda in GG obtained by replacing every subpath of λ\lambda^{\prime} connecting viv_{i} to wiw_{i} with the edge eie_{i}. This is a hamiltonian path in GG, and since the last vertex of λ\lambda^{\prime} was adjacent to v1v_{1} and not in K1K^{1}, it is a hamiltonian cycle in GG. But GG is not hamiltonian, so this is a contradiction. ∎

Lemma 2.13.

Let GG be non-hamiltonian, SS a set of H-path connected edges, and GkG_{k} constructed as in Construction 2.11. There is a hamiltonian path in GkG_{k} from each vertex in KiK^{i} to each vertex in KjK^{j}, for any iji\neq j.

Proof.

Since ei,ejSe_{i},e_{j}\in S and SS is H-path connected, we know there exists a hamiltonian path in GG starting from an endpoint of eie_{i} and ending at an endpoint of eje_{j} which contains all the other edges in SS. Call this hamiltonian path ρ\rho. Then there exists a hamiltonian path ρ\rho^{\prime} in GkG_{k} obtained by replacing each edge ee_{\ell} with a hamiltonian path in {v,w}K\{v_{\ell},w_{\ell}\}\cup K^{\ell} (and possibly appending paths in Ki{wi}K^{i}\cup\{w_{i}\} and Kj{vj}K^{j}\cup\{v_{j}\} as needed). ∎

Theorem 2.14.

If GG is a non-hamiltonian graph containing a set of n>1n>1 edges which is H-path connected, then rGkn1n\displaystyle r_{G_{k}}\to\frac{n-1}{n}.

Proof.

Let GG be a non-hamiltonian graph of order mm containing an H-path connected set of nn edges. By Lemma 2.12 the graphs GkG_{k} are all non-hamiltonian, and by Lemma 2.13 GkG_{k} contains at least (n2)k2\binom{n}{2}k^{2} hamiltonian paths.

Thus we can bound rGkr_{G_{k}} from below as follows:

rGk\displaystyle\displaystyle r_{G_{k}} (n2)k2(m+nk2)\displaystyle\geq\frac{\displaystyle\binom{n}{2}k^{2}}{\displaystyle\binom{m+nk}{2}}
=n(n1)k2(m+nk)(m1+nk)\displaystyle=\frac{n(n-1)k^{2}}{(m+nk)(m-1+nk)}
=n(n1)k2n2k2+2mnknk+m2m.\displaystyle=\frac{n(n-1)k^{2}}{n^{2}k^{2}+2mnk-nk+m^{2}-m}.

As kk approaches infinity, this approaches n1n\frac{n-1}{n}.

On the other hand, we can also bound rGkr_{G_{k}} from above. Notice that there is no hamiltonian path connecting two vertices in the same complete subgraph KiK^{i}; indeed, if there were such a path then GkG_{k} would contain a hamiltonian cycle. So

rGk\displaystyle r_{G_{k}} (m+nk2)(k2)n(m+nk2)\displaystyle\leq\frac{\displaystyle\binom{m+nk}{2}-\binom{k}{2}n}{\displaystyle\binom{m+nk}{2}}
=(m+nk)(m1+nk)k(k1)n(m+nk)(m1+nk)\displaystyle=\frac{(m+nk)(m-1+nk)-k(k-1)n}{(m+nk)(m-1+nk)}
=(n2n)k2+(2mn)k+m2mn2k2+2mnknk+m2m.\displaystyle=\frac{(n^{2}-n)k^{2}+(2mn)k+m^{2}-m}{n^{2}k^{2}+2mnk-nk+m^{2}-m}.

As kk approaches infinity, this approaches n1n\frac{n-1}{n}. So

limkrGk=n1n,\lim_{k\to\infty}r_{G_{k}}=\frac{n-1}{n},

as desired. ∎

Note that the expression n(n1)k2n2k2+2mnknk+m2m\frac{n(n-1)k^{2}}{n^{2}k^{2}+2mnk-nk+m^{2}-m} is increasing in kk, so the asymptotic bound on rGkr_{G_{k}} may not be achieved in a finite graph. However, if we are more careful in counting the number of pairs of vertices connected by hamiltonian paths, we may find a maximum at a finite value of kk. We can see this explicitly by taking GG to be the Petersen graph.

Proposition 2.15.

Let PP denote the Petersen graph, and let {Pk}k\{P_{k}\}_{k\in\mathbb{N}} denote the family of graphs constructed as in Construction 2.11. We have rPk=10k2+40k+20(10+5k2).r_{P_{k}}=\frac{10k^{2}+40k+20}{\binom{10+5k}{2}}. In particular, rPkr_{P_{k}} is maximized at k=18k=18, where rP18=.804¯.r_{P_{18}}=.8\overline{04}.

v1v_{1}v2v_{2}v3v_{3}v4v_{4}v5v_{5}w1w_{1}w4w_{4}w2w_{2}w5w_{5}w3w_{3}e1e_{1}e2e_{2}e3e_{3}e4e_{4}e5e_{5}
Figure 4. The Petersen graph, with an H-path connected set of edges highlighted in red.
Proof.

Consider the Petersen graph, PP. Pick a perfect matching in PP with edges {e1,e2,,e5}\{e_{1},e_{2},\dots,e_{5}\}. One can verify that this is an H-path connected set of edges. Label one vertex of eie_{i} by viv_{i} and the other vertex by wiw_{i}, as in Figure 4. By Lemma 2.12 PkP_{k} is non-hamiltonian, and by Lemma 2.13 there is a hamiltonian path connecting any pair of vertices in distinct subgraphs Ki,KjK^{i},K^{j}.

Claim 2.16.

Let xV(Ki)x\in V(K^{i}) and yV(P){vi,wi}y\in V(P)-\{v_{i},w_{i}\}. There is a hamiltonian path in PkP_{k} connecting xx to yy.

Proof.

We may assume that xV(K1)x\in V(K^{1}). For each i{2,3,4,5}i\in\{2,3,4,5\} let γi\gamma_{i} be a hamiltonian path in KiK^{i}. Let γ1\gamma_{1} be a path in K1K^{1} containing every vertex except xx. There are two cases: Either the shortest path from yy to a vertex in K1K^{1} has edge length 22, or it has edge length 33.

Suppose that the shortest path from yy to a vertex in K1K^{1} has length 33. We can assume without loss of generality that y=w2y=w_{2}. The following path is hamiltonian:

γ=x,γ1,v1,w1,w4,γ4,v4,v5,γ5,w5,w3,γ3,v3,v2,γ2,y.\gamma=x,\gamma_{1},v_{1},w_{1},w_{4},\gamma_{4},v_{4},v_{5},\gamma_{5},w_{5},w_{3},\gamma_{3},v_{3},v_{2},\gamma_{2},y.

If instead the shortest path from yy to a vertex in K1K^{1} has length 22, we can assume without loss of generality that y=v2y=v_{2}. The following path is hamiltonian:

γ=x,w1,γ1,v1,v5,γ5,w5,w3,γ3,v3,v4,γ4,w4,w2,γ2,y.\gamma=x,w_{1},\gamma_{1},v_{1},v_{5},\gamma_{5},w_{5},w_{3},\gamma_{3},v_{3},v_{4},\gamma_{4},w_{4},w_{2},\gamma_{2},y.

Claim 2.17.

Let xx a vertex in V(Ki){vi,wi}V(K^{i})\amalg\{v_{i},w_{i}\} and yy a vertex in V(Ki)V(K^{i}). There is no hamiltonian path between xx and yy in PkP_{k}.

Proof.

Suppose x,yV(Ki){vi,wi}x,y\in V(K^{i})\amalg\{v_{i},w_{i}\}. Since x,yx,y are adjacent, any hamiltonian path from xx to yy would imply the existence of a hamiltonian cycle. By Lemma 2.12 this is impossible. ∎

Claim 2.18.

Let x,yV(P)x,y\in V(P), k>0k>0. There is a hamiltonian path in PkP_{k} between xx and yy if and only if

  1. (1)

    x,yx,y are not adjacent in PP, and

  2. (2)

    exactly one of x,yx,y lies in {w1,w2,w3,w4,w5}\{w_{1},w_{2},w_{3},w_{4},w_{5}\}.

Proof.

Suppose that x,yx,y are adjacent in PP. By Lemma 2.12 there is no hamiltonian path from xx to yy.

Suppose instead that xx and yy are not adjacent in PP. We may assume that x=v1x=v_{1} and yV(P){w1,v2,v5}.y\in V(P)-\{w_{1},v_{2},v_{5}\}. Up to symmetry of PP there are two cases: either y=w2y=w_{2} or y=w3.y=w_{3}. Let γi\gamma_{i} denote a hamiltonian path in KiK^{i}.

If y=w2y=w_{2}, the path

γ=x,γ1,w1,w4,γ4,v4,v5,γ5,w5,w3,γ3,v3,v2,γ2,x\gamma=x,\gamma_{1},w_{1},w_{4},\gamma_{4},v_{4},v_{5},\gamma_{5},w_{5},w_{3},\gamma_{3},v_{3},v_{2},\gamma_{2},x

is hamiltonian in PkP_{k}.

If y=w3y=w_{3}, the path

γ=x,γ1,w1,w4,γ4,v4,v5,γ5,w5,w2,γ2,v2,v3,γ3,y\gamma=x,\gamma_{1},w_{1},w_{4},\gamma_{4},v_{4},v_{5},\gamma_{5},w_{5},w_{2},\gamma_{2},v_{2},v_{3},\gamma_{3},y

is hamiltonian.

Suppose that there exists a hamiltonian path λ\lambda^{\prime} in PkP_{k} from xx to yy. Note that for any jj, the minimal subpath of λ\lambda^{\prime} containing all the vertices of KjK^{j} must have edge length exactly k1k-1. Let γj\gamma_{j} denote these subpaths in λ\lambda^{\prime}. Note that λ\lambda^{\prime} thus includes either vi,γi,wiv_{i},\gamma_{i},w_{i} or wi,γi,viw_{i},\gamma_{i},v_{i} as a subpath for each i[5]i\in[5]. Since 55 is odd and the first vertex of λ\lambda^{\prime} is v1v_{1}, the final vertex of λ\lambda^{\prime} must be wjw_{j} for some j[5]j\in[5]. Thus there is no hamiltonian path in PkP_{k} from viv_{i} to vjv_{j} for any i,j[5]i,j\in[5]. Similarly, there is no hamiltonian path in PkP_{k} from wiw_{i} to wjw_{j} for any i,j[5]i,j\in[5].

This covers all the possible pairs of vertices, so for any k>0,k>0, PkP_{k} is 10k2+40k+2010k^{2}+40k+20-pair-strung, and it is not kk^{\prime}-pair-strung for any k>10k2+40k+20k^{\prime}>10k^{2}+40k+20. Therefore

rPk=10k2+40k+20(10+5k2).r_{P_{k}}=\frac{10k^{2}+40k+20}{\binom{10+5k}{2}}.

We now describe a family of graphs which contain large H-path connected sets.

Theorem 2.19.

Let G=Kn,n+1G=K_{n,n+1} be the complete bipartite graph with bipartitioned vertices {a1,,an}{b1,,bn+1}\{a_{1},\dots,a_{n}\}\cup\{b_{1},\dots,b_{n+1}\}. Then the set

S={(a1,b1),,(an,bn)}S=\{(a_{1},b_{1}),\dots,(a_{n},b_{n})\}

is an H-path connected set in GG.

Proof.

Let 1i<jn1\leq i<j\leq n. The path

bi,ai,bi1,ai1,,b1,a1,bn+1,an,bn,an1,bn1,,aj+1,bj+1,ai+1,bi+1,,aj,bj.b_{i},a_{i},b_{i-1},a_{i-1},\dots,b_{1},a_{1},b_{n+1},a_{n},b_{n},\\ a_{n-1},b_{n-1},\dots,a_{j+1},b_{j+1},a_{i+1},b_{i+1},\dots,a_{j},b_{j}.

is hamiltonian and contains all edges in SS, as needed (see Figure 5 for an example). ∎

Putting together Theorem 2.14 and Theorem 2.19, we get the following result, which provides an alternative proof that R=1.R=1.

Corollary 2.20.

For all n>1n>1, there is a family of graphs GkG_{k} so that rGkn1nr_{G_{k}}\to\frac{n-1}{n} as kk\to\infty. In particular, R=1R=1.

Refer to caption
Figure 5. Hamiltonian path in K8,9K_{8,9} connecting edges (a3,b3)(a_{3},b_{3}) and (a6,b6)(a_{6},b_{6}).

3. A Minimal Hamiltonian-Connected Graph

In the previous section, we demonstrated the existence of graphs that are as close to being hamiltonian-connected as possible without being hamiltonian. In this section, we explore an alternative extreme: we construct a family of hamiltonian-connected (and therefore hamiltonian) graphs that have a minimal number of edges and minimal number of minimum degree vertices. This partially answers a question of Modalleliyan and Omoomi ([MO16]), which asked whether there exist minimally hamiltonian-connected graphs (i.e. hamiltonian-connected graphs such that the removal of any edge results in a non-hamiltonian-connected graph) with maximal vertex degree Δ\Delta so that 3Δn/2.3\leq\Delta\leq\lceil n/2\rceil. While preparing this paper, Zhan [Zha22] completely answered this question in the affirmative. In particular, he constructs a cubic graph which is minimally hamiltonian-connected. However, our construction is sufficiently distinct from Zhan’s that we include it out of interest to the field. Our construction is highly symmetric and inductive.

We first note that a minimal hamiltonian-connected graph of order nn must have at least 3n2\frac{3n}{2} edges. This was first proven by Moon [Moo65].

Theorem 3.1 ([Moo65]).

If GG is a hamiltonian-connected graph with n>3n>3 vertices, then every vertex has degree at least 33, and GG must have at least 3n2\frac{3n}{2} edges.

Proof.

Suppose that some vertex vv of GG has degree 2. Let u,wu,w be the two vertices adjacent to vv. If γ\gamma is a hamiltonian path in GG between uu and ww, it contains vv and therefore contains u,v,wu,v,w as a subpath. Thus γ=u,v,w\gamma=u,v,w. But since GG has more than 3 vertices, γ\gamma is not hamiltonian. Therefore every vertex of GG has degree at least 3, so the number of edges in GG is at least 3n2.\frac{3n}{2}.

Now we construct a family of graphs of even order m=2nm=2n which have 3n3n edges. We will show that these graphs are hamiltonian-connected. Note that since every hamiltonian-connected graph is hamiltonian, such graphs must contain a hamiltonian cycle as a subgraph. So we begin our construction with the cycle on 2n2n vertices, C2nC_{2n}, where n0mod3n\equiv 0\mod 3.

Label the vertices of the cycle C2nC_{2n} by 0,1,2,,2n10,1,2,\dots,2n-1 in order. Add edges (0,n),(1,n+2),(2,n+1)(0,n),(1,n+2),(2,n+1), and continue in this pattern in groups of three until every vertex has degree 3.

More precisely, for n0mod3n\equiv 0\mod 3, we define Γn\Gamma_{n} to be the graph with vertex set V(Γn)=[2n]V(\Gamma_{n})=[2n] and edge set

E(Γn)=E(C2n){(3i,3i+n),(3i+1,3i+n+2),(3i+2,3i+n+1)}i=0n1.E(\Gamma_{n})=E(C_{2n})\cup\{(3i,3i+n),(3i+1,3i+n+2),(3i+2,3i+n+1)\}_{i=0}^{n-1}.

The graphs Γ3,Γ6,Γ9\Gamma_{3},\Gamma_{6},\Gamma_{9} constructed in this way are illustrated in Figure 6.

Refer to caption
Figure 6. Shown here are Γ3,Γ6\Gamma_{3},\Gamma_{6} and Γ9\Gamma_{9}.
Theorem 3.2.

For any n0mod3n\equiv 0\mod 3, Γn\Gamma_{n} is a hamiltonian-connected graph of order 2n2n with 3n3n edges.

One standard tool to demonstrate that a graph is hamiltonian-connected is to show that it has sufficiently many edges, or vertices of sufficiently high degree (c.f. [Ore63, Fau+89, Wei93, KLZ12]). Since the graphs Γn\Gamma_{n} are sparse and contain no vertex of degree >3>3, none of these results apply. We will instead show that Γn\Gamma_{n} is hamiltonian-connected by considering several cases and inducting on nn.

To emphasize the relationship between Γn\Gamma_{n} and Γn+3\Gamma_{n+3}, we relabel the vertices of Γn+3\Gamma_{n+3} by 0,1,,n1,a,b,c,n,n+1,,2n1,α,γ,β0,1,\dots,n-1,a,b,c,n,n+1,\dots,2n-1,\alpha,\gamma,\beta. By construction, Γn+3\Gamma_{n+3} contains all the edges in Γn\Gamma_{n} except for (n1,n)(n-1,n) and (2n1,0)(2n-1,0), as well as edges (a,α),(b,β),(c,γ),(a,\alpha),(b,\beta),(c,\gamma), (n1,a),(c,n),(2n1,α),(β,0).(n-1,a),(c,n),(2n-1,\alpha),(\beta,0). This is illustrated in Figure 7. Partition the vertex set of Γn+3\Gamma_{n+3} into new vertices NN, proximal vertices PP, and orbital vertices OO, where

N\displaystyle N ={a,b,c,α,β,γ},\displaystyle=\{a,b,c,\alpha,\beta,\gamma\},
P\displaystyle P ={0,n2,n1,n,2n1,2n2},\displaystyle=\{0,n-2,n-1,n,2n-1,2n-2\},
O\displaystyle O ={1,2,,n3,n+1,n+2,,2n3}.\displaystyle=\{1,2,\dots,n-3,n+1,n+2,\dots,2n-3\}.
Refer to caption
Figure 7. Part of the graph Γn+3\Gamma_{n+3} is shown here. It can be obtained from Γn\Gamma_{n} by inserting vertices a,b,c,α,β,γa,b,c,\alpha,\beta,\gamma and edges (a,α),(b,β),(c,γ)(a,\alpha),(b,\beta),(c,\gamma), along with the edges needed to complete the cycle. Vertices indicated with squares are in OO, with circles are in PP, and with triangles are in NN.

We first prove a special case. We will use rotational symmetry of Γn+3\Gamma_{n+3} and this special case to prove the other cases.

Lemma 3.3.

Suppose that Γn\Gamma_{n} is hamiltonian-connected, and n12n\geq 12. For any distinct vertices u,vOu,v\in O, there exists a hamiltonian path in Γn+3\Gamma_{n+3} from uu to vv.

Proof.

By the inductive hypothesis, there exists a hamiltonian path γ\gamma in Γn\Gamma_{n} from uu to vv, which necessarily contains all of the vertices of PP.

Since the six vertices of PP are not endpoints of γ\gamma, and since there are no edges from {2n1,n1}\{2n-1,n-1\} to any vertex of OO, γ\gamma must enter and leave PP at vertices in {0,n2,n,2n2}\{0,n-2,n,2n-2\}. Let γP\gamma_{P} be the subpath(s) of γ\gamma restricted to the induced subgraph of Γn\Gamma_{n} on PP. Then γP\gamma_{P} has endpoints in {0,n2,n,2n2}\{0,n-2,n,2n-2\}, so it is a single connected path or two connected paths.

We first suppose γP\gamma_{P} is a single connected path. There are 6 possible pairs of endpoints of γP\gamma_{P}. In each case, there are up to 2 possible paths connecting the endpoints. For example, consider the endpoints 0 and 2n22n-2 (see Figure 8). Since n,n2n,n-2 are not endpoints of γP\gamma_{P}, the edges (n,n+1)(n,n+1) and (n2,n3)(n-2,n-3) are not contained in γ\gamma. Then since γP\gamma_{P} contains nn and nn is not an endpoint of γ\gamma, γP\gamma_{P} must include edges (0,n)(0,n) and (n,n1).(n,n-1). Similarly, γP\gamma_{P} must contain edges (2n1,n2)(2n-1,n-2) and (n1,n2)(n-1,n-2). Since γP\gamma_{P} contains (n,n1)(n,n-1) and (n1,n2)(n-1,n-2), it can not contain (n1,2n2)(n-1,2n-2). Similarly, it does not contain (2n1,0).(2n-1,0). Since γP\gamma_{P} is a path, it must contain the edge (2n2,2n1)(2n-2,2n-1), and in particular it must be the highlighted path indicated in Figure 8.

One can make a similar argument for the other endpoint pairs. This leaves 6 possible paths for γP\gamma_{P}. By a similar argument, when γP\gamma_{P} contains two connected subpaths there are 2 cases. These are all illustrated in Figure 9.

Refer to caption
Figure 8. Example case of 0 and 2n22n-2 as endpoints of γP\gamma_{P}. The only possible path is shown, as described in Lemma 3.3

Each of the possibilities for γP\gamma_{P} has a corresponding path γP\gamma_{P}^{\prime} in Γn+3\Gamma_{n+3}, as shown in Figure 9, which meets every vertex in NN. We can obtain a hamiltonian path in Γn+3\Gamma_{n+3} by substituting γP\gamma_{P} in γ\gamma with γP\gamma_{P}^{\prime}. ∎

Refer to caption
Figure 9. The 8 possibilities of γP\gamma_{P} on the left and the corresponding paths γP\gamma_{P}^{\prime} in Γn+3\Gamma_{n+3} on the right, as described in Lemma 3.3.
Proof of Theorem 3.2.

It can be verified (e.g. by computer program) that Γk\Gamma_{k} is hamiltonian-connected for k=3,6,9,12.k=3,6,9,12. Suppose that Γn\Gamma_{n} is hamiltonian-connected. Let u,vu,v be a pair of distinct vertices in Γn+3\Gamma_{n+3}. We will prove that there is a hamiltonian-path connecting them by considering several cases, depending on the vertex sets that u,vu,v belong to.

  1. Case 1:

    u,vOu,v\in O. Apply Lemma 3.3.

  2. Case 2:

    u,vNPu,v\in N\cup P. Rotate Γn+3\Gamma_{n+3} clockwise along the cycle C2n+6C_{2n+6} by 6 vertices. Note that this is a automorphism of Γn+3\Gamma_{n+3}. Then both uu and vv are now in OO, so there is a hamiltonian path γ\gamma from uu to vv by Case 1.

  3. Case 3:

    uO,vNu\in O,v\in N. If v{α,a}v\in\{\alpha,a\}, rotate Γn+3\Gamma_{n+3} counterclockwise by 3 vertices. Then vOv\in O, and either uOu\in O or u{n,c,b,0,β,γ}u\in\{n,c,b,0,\beta,\gamma\}. If uOu\in O, then by Case 1 there is a hamiltonian path vv to uu. Otherwise, rotate a further 6 vertices counterclockwise. Then uOu\in O and vOv\in O, so by Case 1 we can find a (u,v)(u,v) hamiltonian path.

  4. Case 4:

    uO,vPu\in O,v\in P. Rotate Γn+3\Gamma_{n+3} so that vv is in NN. By the Cases 2 and 3 there is a (u,v)(u,v) hamiltonian path, regardless of the set containing uu.

While this theorem is restricted to graphs of order 2n2n where n0mod3n\equiv 0\mod 3, a small adaptation to the construction produces graphs of arbitrary even order. In particular, to produce the graph of order 2(n1)2(n-1), first produce the graph of order 2n2n, then delete the edge (β,b)(\beta,b) and merge the two edges that share degree-2 vertices. To produce the graph of order 2(n2)2(n-2) delete the edge (γ,c)(\gamma,c) and merge the two edges that share degree-2 vertices. We believe that these are all hamiltonian-connected, and that a similar case-analysis approach will provide a proof. We have omitted this for the sake of space. We have verified hamiltonicity of these graphs for n84n\leq 84 by adapting the program hamiltonicityChecker (see [Goe+22, Goe+24]), which has methods for determining the existence of a hamiltonian path between two specified vertices.

References

  • [CE83] Lane Clark and Roger Entringer “Smallest maximally non-Hamiltonian graphs” In Period. Math. Hung. 14.1, 1983, pp. 57–68
  • [CES92] Lane Clark, Roger Entringer and Henry Shapiro “Smallest maximally non-Hamiltonian graphs. II” In Graphs Combin. 8.3, 1992, pp. 225–231
  • [CGK79] Gary Chartrand, Ronald J. Gould and S.. Kapoor “On homogeneously traceable non-Hamiltonian graphs” In Second International Conference on Combinatorial Mathematics (New York, 1978) 319, Ann. New York Acad. Sci., 1979, pp. 130–135
  • [Fau+89] Ralph Faudree, Ronald Gould, Michael Jacobson and Richard Schelp “Neighborhood unions and Hamiltonian properties in graphs” In J. Comb. Theory. Ser. B 47.1, 1989, pp. 1–9
  • [Goe+22] Jan Goedgebeur, Jarne Renders, Gábor Wiener and Carol T. Zamfirescu “K2-Hamiltonian Graphs”, 2022 URL: https://github.com/JarneRenders/K2-Hamiltonian-Graphs
  • [Goe+24] Jan Goedgebeur, Jarne Renders, Gábor Wiener and Carol T. Zamfirescu “K2-Hamiltonian Graphs: II” In J. Graph Theory 105.4, 2024, pp. 580–611
  • [HZ22] Yanan Hu and Xingzhi Zhan “Regular homogeneously traceable nonhamiltonian graphs” In Discrete Appl. Math. 310, 2022, pp. 60–64
  • [KLZ12] Zhao Kewen, Hong-Jian Lai and Ju Zhou “Hamiltonian-connected graphs with large neighborhoods and degrees” In Missouri J. Math. Sci. 24.1, 2012, pp. 54–66
  • [Lin+97] Xiaohui Lin, Wenzhou Jiang, Chengxue Zhang and Yuansheng Yang “On smallest maximally non-Hamiltonian graphs” In Ars Combin. 45, 1997, pp. 263–270
  • [MO16] Maliheh Modalleliyan and Behnaz Omoomi “Critical Hamiltonian connected graphs” In Ars Combin. 126, 2016, pp. 13–27
  • [Moo65] John W. Moon “On a Problem of Ore” In Math. Gaz. 49.367 Cambridge University Press, 1965, pp. 40–41 DOI: 10.2307/3614234
  • [Ore63] Øystein Ore “Hamilton connected graphs” In J. Math. Pures Appl. (9) 42, 1963, pp. 21–27
  • [Sku84] Zdzisław Skupień “Homogeneously traceable and Hamiltonian connected graphs” In Demonstr. Math. 17.4, 1984, pp. 1051–1067
  • [Wei93] Bing Wei “Hamiltonian paths and Hamiltonian connectivity in graphs” Graph theory (Niedzica Castle, 1990) In Discrete Math. 121.1-3, 1993, pp. 223–228
  • [Zam15] Carol T. Zamfirescu “On hypohamiltonian and almost hypohamiltonian graphs” In J. Graph Theory 79.1, 2015, pp. 63–81
  • [Zha22] Xingzhi Zhan “The maximum degree of a minimally hamiltonian-connected graph” In Discrete Math. 345.12, 2022, pp. Paper No. 113159\bibrangessep5