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

The maximum Wiener index of a uniform hypergraph

Stijn Cambie, Ervin Győri, Nika Salia11footnotemark: 1, Casey Tompkins22footnotemark: 2, James Tuite Extremal Combinatorics and Probability Group (ECOPRO), Institute for Basic Science (IBS), Daejeon, KR, supported by the Institute for Basic Science (IBS-R029-C4), E-mail: [email protected], [email protected]Alfréd Rényi Institute of Mathematics, Budapest, HU, supported by the National Research, Development and Innovation Office NKFIH, grants K132696, K135800 and SNN-135643, E-mail: [email protected],[email protected]Open University, Walton Hall, Milton Keynes, UK, supported by EPSRC grant EP/W522338/1 and London Mathematical Society grant ECF-2021-27. E-mail: [email protected]
Abstract

The Wiener index of a (hyper)graph is calculated by summing up the distances between all pairs of vertices. We determine the maximum possible Wiener index of a connected nn-vertex kk-uniform hypergraph and characterize for every nn all hypergraphs attaining the maximum Wiener index.

1 Introduction

The Wiener index [16], also known as the total distance, is a well-studied parameter in graph theory. For a given graph GG, the Wiener index of the graph is denoted by W(G)W(G) where

W(G)={u,v}V(G)dG(u,v).W(G)=\sum_{\{u,v\}\subseteq V(G)}d_{G}(u,v).

The Wiener index of a graph with a fixed number of vertices is linearly related to the average distance of the graph, which is itself another important metric parameter in random graph theory. Additionally, it is related to parameters of random walks such as cover cost and Kemeny’s constant [8, 12]. Its generalization of Steiner-Wiener index [13] is related to the average traveling salesman problem distance [5].

It is natural to investigate the extremal properties of the Wiener index. Among the most basic results on the total distance of a graph, there are folklore results that for every connected graph GG and a tree TT with nn vertices, we have

W(Kn)W(G)W(Pn) and W(Sn)W(T)W(Pn).W(K_{n})\leq W(G)\leq W(P_{n})\mbox{~{}and~{}}W(S_{n})\leq W(T)\leq W(P_{n}).

Here KnK_{n} is the nn-vertex complete graph, PnP_{n} is a path with nn vertices and SnS_{n} is a star with nn vertices. Proofs can be found in [4].

While these statements and proofs might be straightforward and easy, the relationship between the Wiener index and other graph parameters or specific classes of graphs is still being explored. For example among all graphs with a given diameter and order the minimum Wiener index is determined in [15]. In [3] the maximum Wiener index is determined up to the asymptotics of the second-order term. The authors in [6, 7, 9] have studied planar graphs and determined sharp bounds for the maximum of the Wiener index in Apollonian planar graphs and maximal planar graphs.

In this work, we study the Wiener index of kk-uniform hypergraphs. Generalizing fundamental results in graph theory to hypergraphs gives a broader insight into the problem. A few examples of such extensions of classical results towards the hypergraph setting are given in [14, 11]. A kk-uniform hypergraph \mathcal{H} of order nn can be represented as a pair (V,E)(V,E) where VV is a vertex set with nn elements and E(Vk)E\subseteq\binom{V}{k} is a family of kk-sets, called hyperedges or just edges. Note that in the case of k=2k=2, a 22-uniform hypergraph is a graph. In order to introduce a distance between the vertices of a hypergraph, at first we need to introduce a concept of paths for hypergraphs. In this work we follow the celebrated definition of Berge [1]: a Berge path of a hypergraph \mathcal{H} is a sequence of distinct vertices and hyperedges v0h1v1h2v2vd1hdvdv_{0}h_{1}v_{1}h_{2}v_{2}\ldots v_{d-1}h_{d}v_{d} of \mathcal{H} where vi1,vihiv_{i-1},v_{i}\in h_{i} for every 1id1\leq i\leq d for some dd. A hypergraph is connected if there is a Berge path for every pair of vertices. The length of a Berge path is the number of hyperedges in the Berge path. Naturally the distance between two vertices u,vu,v is the length of the shortest Berge path containing uu and vv in the hypergraph \mathcal{H}, and it is denoted by d(u,v)d_{\mathcal{H}}(u,v). Note that when the ground hypergraph is known from the context we write d(u,v)d(u,v) instead of d(u,v)d_{\mathcal{H}}(u,v). The Wiener index of a connected hypergraph \mathcal{H} is defined as {u,v}V()d(u,v).\sum_{\{u,v\}\subseteq V(\mathcal{H})}d_{\mathcal{H}}(u,v).

A connected hypergraph \mathcal{H} is called a hyper-tree when it can be represented by a tree TT in the graph sense, on the same vertex set, where every edge of \mathcal{H} induces a connected subgraph of TT, equivalent formulations are collected in [2, Chap. 5, Sec. 4]. When restricting to linear kk-uniform hyper-trees (connected hypergraphs with n=m(k1)+1n=m(k-1)+1 vertices and mm edges), it was shown in [10] that for every such hyper-tree \mathcal{H}, W(Snk)W()W(Pnk)W(S_{n}^{k})\leq W(\mathcal{H})\leq W(P_{n}^{k}), where SnkS_{n}^{k} and PnkP_{n}^{k} are the linear path and star, see Figure 1. Furthermore, the corresponding extremal hyper-trees are unique.

Figure 1: The loose star S134S_{13}^{4} and path P134P_{13}^{4}

Considering arbitrary connected kk-uniform hypergraphs, or hyper-trees, the extremal hypergraphs for maximizing/minimizing the Wiener index are not always unique. Determining the lower bound for the Wiener index of kk-uniform hypergraphs is trivial, as all distances between the distinct vertices are at least 11.

Proposition 1.

Any connected kk-uniform hypergraph \mathcal{H} satisfies (n2)=W(Knk)W().\binom{n}{2}=W(K_{n}^{k})\leq W(\mathcal{H}).

We remark that, in huge contrast with the graph case, equality can also be attained by very sparse hypergraphs.

Remark 2.

Equality in Proposition 1 occurs if and only if for every pair of vertices there is a hyperedge incident to both. This is the case when the hypergraph corresponds with a projective plane. This is an example of a sparse hypergraph minimizing the Wiener index. Also hyper-trees can attain equality when k3k\geq 3, e.g. the hyper-tree with underlying tree SnS_{n} and every subtree of order kk being a hyperedge.

On the other hand, maximizing the Wiener index for kk-uniform hypergraphs is a relatively complex problem. In this note, we determine the maximum possible Wiener index of a connected nn-vertex kk-uniform hypergraph and characterize all hypergraphs attaining the maximum. First, we define the class of extremal hypergraphs. For integers aa and bb with aba\leq b, let [a,b][a,b] be the set {a,a+1,,b}\{a,a+1,\ldots,b\} and [a][a] be the set [1,a][1,a]. For given integers nn and kk with 0k<n0\leq k<n, let ss and rr be integers such that n=ks+rn=ks+r and 0r<k0\leq r<k. For every rr such that r0r\neq 0, let Pnk=(V,E)P_{n}^{k}=(V,E) be the following tight-path. Let the vertex set of PnkP_{n}^{k} be [n][n] and the edge set be E={[(i1)k+1,ik]:1is}{[r+(i1)k+1,ik+r]:1is}.E=\left\{[(i-1)k+1,ik]\colon 1\leq i\leq s\right\}\cup\left\{[r+(i-1)k+1,ik+r]\colon 1\leq i\leq s\right\}. For r=0r=0, and every integer xx, such that 0<x<k0<x<k, we define a kk-uniform hypergraph Pn,xk=(V,E)P_{n,x}^{k}=(V,E) where V=[n]V=[n] and E={[(i1)k+1,ik]:1is}{[(i1)k+1+x,ik+x]:1is1}.E=\{[(i-1)k+1,ik]\colon 1\leq i\leq s\}\cup\{[(i-1)k+1+x,ik+x]\colon 1\leq i\leq s-1\}. See Figure 2 for examples of both PnkP_{n}^{k} and Pn,xkP_{n,x}^{k}. Our main result can now be stated as follows.

Figure 2: The tight paths P134P_{13}^{4} and P12,24P_{12,2}^{4} for k=4k=4
Theorem 1.

Let \mathcal{H} be a connected kk-uniform hypergraph of order nkn\geq k. If kn,k\nmid n, then W()W(Pnk)W(\mathcal{H})\leq W(P_{n}^{k}) with equality if and only if =Pnk.\mathcal{H}=P_{n}^{k}. If knk\mid n, then W()W(Pn,1k)W(\mathcal{H})\leq W(P_{n,1}^{k}) with equality if and only if =Pn,xk\mathcal{H}=P_{n,x}^{k} for some 0<x<k.0<x<k.

2 Proof of Theorem 1

Let =(V,E)\mathcal{H}=(V,E) be an nn-vertex kk-uniform connected hypergraph with maximum Wiener index. Then we may assume that for every edge hh of \mathcal{H}, the hypergraph =(V,E\h)\mathcal{H}^{\prime}=(V,E\backslash h)111We abuse notation and write E\hE\backslash h instead of E\{h}E\backslash\{h\}. is not connected, otherwise we would remove the edge and the Wiener index would not decrease. Proving Theorem 1 for edge-minimal hypergraphs is sufficient since adding an edge to one of the extremal hypergraphs from Theorem 1 decreases the Wiener index. The following lemma shows that there is a hyperedge hh in EE, such that hypergraph =(V,E\h)\mathcal{H}^{\prime}=(V,E\backslash h) an contains at most one connected component of size greater than one.

Lemma 3.

Let =(V,E)\mathcal{H}=(V,E) be a kk-uniform hypergraph such that the deletion of any edge disconnects the hypergraph. Then there is an edge hh of \mathcal{H} such that the hypergraph =(V,E\h)\mathcal{H}^{\prime}=(V,E\backslash h) contains at most one connected component of size greater than one.

Proof.

Let hh be a hyperedge of \mathcal{H} such that the size of the largest connected component of =(V,E\h)\mathcal{H}^{\prime}=(V,E\backslash h) is the maximum. Then either we are done or =(V,E\h)\mathcal{H}^{\prime}=(V,E\backslash h) contains two connected components 1\mathcal{H}_{1} and 2\mathcal{H}_{2} each containing a hyperedge such that 1\mathcal{H}_{1} is a component with the maximum size. Let h2h_{2} be a hyperedge of 2\mathcal{H}_{2}, the hypergraph (V,E\h2)(V,E\backslash h_{2}) contains a component of size larger than H1H_{1} since it contains a component containing 1\mathcal{H}_{1} with all vertices of hh as well, a contradiction. ∎

By Lemma 3 and the edge-minimality of \mathcal{H}, there is an edge h1h_{1} of \mathcal{H} such that =(V,E\h1)\mathcal{H}^{\prime}=(V,E\backslash h_{1}) consists of a connected component of order nn-\ell on a vertex set VVV^{\prime}\subset V and \ell isolated vertices for some 1<k1\leq\ell<k. Let V\mathcal{H}^{\prime}_{V^{\prime}} be the largest connected component of \mathcal{H}^{\prime}.

Now we proceed by induction, noting that the theorem trivially holds for n<2kn<2k. Assume Theorem 1 holds for every order strictly smaller than nn. Let vv be an isolated vertex of \mathcal{H}^{\prime} such that uVd(u,v)\sum_{u\in V^{\prime}}d_{\mathcal{H}^{\prime}}(u,v) is the maximum possible. Then we have

W()W(V)+(2)+uVd(u,v).W(\mathcal{H})\leq W(\mathcal{H}^{\prime}_{V^{\prime}})+\binom{\ell}{2}+\ell\sum_{u\in V^{\prime}}d(u,v). (1)

By induction, we have an upper bound on W(V).W(\mathcal{H}^{\prime}_{V^{\prime}}). For this, we compute the Wiener index of the paths PnkP_{n}^{k} and Pn,xkP_{n,x}^{k}. When n=ks+r,n=ks+r, for some integers ss and rr such that 0r<k0\leq r<k, the Wiener index of the path equals

f(s,k,r)\displaystyle f(s,k,r) =k2s33+rks2+r2s+(k23k)s6+(r2),\displaystyle=\frac{k^{2}s^{3}}{3}+rks^{2}+r^{2}s+\frac{(k^{2}-3k)s}{6}+\binom{r}{2},

Note that the Wiener index of Pn,xkP_{n,x}^{k} is independent of the choice of xx.

We also need to find an upper bound for uVd(u,v)\sum_{u\in V^{\prime}}d_{\mathcal{H}}(u,v). To this end, we define the following two functions:

g1(,k,s,r)\displaystyle g_{1}(\ell,k,s,r) =ks2+s+r(2s+1),\displaystyle=ks^{2}+\ell s+r(2s+1),
g2(,k,s,r)\displaystyle g_{2}(\ell,k,s,r) =ks2+s+2rs.\displaystyle=ks^{2}+\ell s+2rs.
Claim 4.

Let n=ks+rn-\ell=ks^{\prime}+r^{\prime} for integers ss^{\prime} and rr^{\prime} for 0r<k0\leq r^{\prime}<k, then we have

uVd(u,v){g1(,k,s,r) if krg2(,k,s,r) if k>r\sum_{u\in V^{\prime}}d(u,v)\leq\begin{cases}g_{1}(\ell,k,s^{\prime},r^{\prime})\mbox{ if }k-\ell\leq r^{\prime}\\ g_{2}(\ell,k,s^{\prime},r^{\prime})\mbox{ if }k-\ell>r^{\prime}\\ \end{cases}
Proof.

Let dd be the eccentricity of vv, i.e. d=maxuVd(u,v).d=\max_{u\in V^{\prime}}d_{\mathcal{H}}(u,v). For di1d\geq i\geq 1, let nin_{i} be the number of vertices in \mathcal{H}^{\prime} which are at distance ii from vv. Let ni=0n_{i}=0 for i>d.i>d. Now by definition, we have uVd(u,v)=i=1dini\sum_{u\in V^{\prime}}d(u,v)=\sum_{i=1}^{d}in_{i} and i=1dni=(n)=ks+r\sum_{i=1}^{d}n_{i}=(n-\ell)=ks^{\prime}+r^{\prime}. Note that n1kn_{1}\geq k-\ell and ni+ni+1kn_{i}+n_{i+1}\geq k for every 1id1,1\leq i\leq d-1, since for every such ii there is an edge whose vertices are all at distance ii and i+1i+1 from v.v.

First, we bound the eccentricity. Let uu be a vertex for which d(u,v)=dd(u,v)=d and hh an edge containing the vertex uu. When 0r<k,0\leq r^{\prime}<k-\ell, either d2sd\leq 2s^{\prime} or we have n1++n2s1(k)+(s1)k>k(s1)+rn_{1}+\dots+n_{2s^{\prime}-1}\geq(k-\ell)+(s^{\prime}-1)k>k(s^{\prime}-1)+r^{\prime} and hence hh cannot be disjoint of these vertices at distance at most 2s12s^{\prime}-1 from vv, thus we have d2s.d\leq 2s^{\prime}. Now we conclude

i=1dini\displaystyle\sum_{i=1}^{d}in_{i} =2s(i=12sni)j=12s1(i=1jni)\displaystyle=2s^{\prime}\left(\sum_{i=1}^{2s^{\prime}}n_{i}\right)-\sum_{j=1}^{2s^{\prime}-1}\left(\sum_{i=1}^{j}n_{i}\right)
2s(n)(s(k)+j=12s1j2k)\displaystyle\leq 2s^{\prime}(n-\ell)-\left(s^{\prime}(k-\ell)+\sum_{j=1}^{2s^{\prime}-1}\mathopen{}\left\lfloor\frac{j}{2}\right\rfloor\mathclose{}k\right)
=2s(ks+r)(s(k)+s(s1)k)\displaystyle=2s^{\prime}(ks^{\prime}+r^{\prime})-\left(s^{\prime}(k-\ell)+s^{\prime}(s^{\prime}-1)k\right)
=s2k+2sr+s.\displaystyle=s^{\prime 2}k+2s^{\prime}r^{\prime}+\ell s^{\prime}.

In the case that kr<k,k-\ell\leq r^{\prime}<k, we similarly have n1++n2sksn_{1}+\dots+n_{2s^{\prime}}\geq ks^{\prime} and again hh cannot be disjoint from at least ksks^{\prime} vertices in \mathcal{H}^{\prime}, thus d2s+1.d\leq 2s^{\prime}+1. The computation now analogously results into i=1dinis2k+(2s+1)r+s.\sum_{i=1}^{d}in_{i}\leq s^{\prime 2}k+(2s^{\prime}+1)r^{\prime}+\ell s^{\prime}.

Finally, it is sufficient to prove that the upper bound for the right-hand side of Inequality (1) is always bounded by the claimed upper bound. For n=ks+rn-\ell=ks+r, we have

f(s+1,k,r+k)\displaystyle f(s+1,k,r+\ell-k) (f(s,k,r)+g1(,k,s,r)+(2))=(kr)20 if kr,\displaystyle-\left(f(s,k,r)+\ell g_{1}(\ell,k,s,r)+\binom{\ell}{2}\right)=(k-\ell-r)^{2}\geq 0\>\>\mbox{ if }k-\ell\leq r,
f(s,k,r+)\displaystyle f(s,k,r+\ell) (f(s,k,r)+g2(,k,s,r)+(2))=r0 if k>r,\displaystyle-\left(f(s,k,r)+\ell g_{2}(\ell,k,s,r)+\binom{\ell}{2}\right)=\ell r\geq 0\>\>\>\quad\quad\quad\quad\mbox{ if }k-\ell>r,

which also has been verified by Maple222See https://github.com/StijnCambie/WienerHypergraph.

Equality holds if k=+rk=\ell+r or r=0r=0. If kn,k\mid n, equality occurs if and only if k=+rk=\ell+r and for every fixed 0<<k0<\ell<k the only hypergraph attaining the maximum Wiener index is Pn,kP_{n,\ell}^{k}. If kn,k\nmid n, equality occurs if and only if r=0r=0, since 0<<k0<\ell<k and by the induction hypothesis the unique connected nn-vertex kk-uniform hypergraph maximizing the Wiener index is PnkP_{n}^{k}.

References

  • [1] C. Berge. Graphs and hypergraphs. North-Holland, 1973.
  • [2] C. Berge. Hypergraphs, volume 45 of North-Holland Mathematical Library. North-Holland Publishing Co., Amsterdam, 1989. Combinatorics of finite sets, Translated from the French.
  • [3] S. Cambie. An asymptotic resolution of a problem of Plesník. Journal of Combinatorial Theory, Series B, 145:341–358, 2020.
  • [4] S. Cambie. Extremal aspects of distances and colourings in graphs. PhD thesis, Radboud University, 2022.
  • [5] S. Cambie. The average solution of a TSP instance in a graph. arXiv e-prints, page arXiv:2209.03409, Sept. 2022.
  • [6] Z. Che and K. L. Collins. An upper bound on Wiener indices of maximal planar graphs. Discrete Applied Mathematics, 258:76–86, 2019.
  • [7] É. Czabarka, P. Dankelmann, T. Olsen, and L. A. Székely. Wiener index and remoteness in triangulations and quadrangulations. Discrete Mathematics & Theoretical Computer Science, 23, 2021.
  • [8] A. Georgakopoulos and S. Wagner. Hitting times, cover cost, and the Wiener index of a tree. J. Graph Theory, 84(3):311–326, 2017.
  • [9] D. Ghosh, E. Győri, A. Paulos, N. Salia, and O. Zamora. The maximum Wiener index of maximal planar graphs. Journal of Combinatorial Optimization, 40(4):1121–1135, 2020.
  • [10] H. Guo, B. Zhou, and H. Lin. The Wiener index of uniform hypergraphs. MATCH Commun. Math. Comput. Chem., 78(1):133–152, 2017.
  • [11] E. Győri and N. Lemons. Hypergraphs with no cycle of a given length. Combin. Probab. Comput., 21(1-2):193–201, 2012.
  • [12] J. Jang, S. Kim, and M. Song. Kemeny’s constant and Wiener index on trees. arXiv e-prints, page arXiv:2209.11271, Sept. 2022.
  • [13] X. Li, Y. Mao, and I. Gutman. The Steiner Wiener index of a graph. Discuss. Math. Graph Theory, 36(2):455–465, 2016.
  • [14] D. Mubayi. A hypergraph extension of Turán’s theorem. J. Combin. Theory Ser. B, 96(1):122–134, 2006.
  • [15] J. Plesník. On the sum of all distances in a graph or digraph. Journal of Graph Theory, 8(1):1–21, 1984.
  • [16] H. Wiener. Structural determination of paraffin boiling points. J. Amer. Chem. Soc. , 69:17–20, 1947.