Wiener Index and Remoteness in Triangulations and Quadrangulations

Let $G$ be a a connected graph. The Wiener index of a connected graph is the sum of the distances between all unordered pairs of vertices. We provide asymptotic formulae for the maximum Wiener index of simple triangulations and quadrangulations with given connectivity, as the order increases, and make conjectures for the extremal triangulations and quadrangulations based on computational evidence. If $\overline{\sigma}(v)$ denotes the arithmetic mean of the distances from $v$ to all other vertices of $G$, then the remoteness of $G$ is defined as the largest value of $\overline{\sigma}(v)$ over all vertices $v$ of $G$. We give sharp upper bounds on the remoteness of simple triangulations and quadrangulations of given order and connectivity.

as the arithmetic mean of the distances between all unordered pairs of distinct vertices) have been proved since.
The notation we use in this paper is as follows. If G is a graph, then we denote its vertex set and edge set by V (G) and E(G). By n(G) and m(G) we mean the order and size of G, defined as |V (G)| and |E(G)|, respectively. The eccentricity e(v) of a vertex v is the distance to a vertex farthest from v, i.e., e(v) = max u∈V (G) d G (v, u). The largest and the smallest of the eccentricities of the vertices of G are the diameter and the radius of G, respectively. The neighborhood of a vertex v of G is the set of vertices adjacent to v, it is denoted by N G (v), and the cardinality |N G (v)| is the degree of v, which we denote by deg G (v). If i is an integer with 0 ≤ i ≤ e(v), then N i (v) denotes the set of all vertices at distance exactly i from v, and we write n i (v) for |N i (v)|. If there is no danger of confusion, we often omit the subscript G or the argument G or v. If A, B ⊆ V (G), then m(A, B) denotes the number of edges that join a vertex in A to a vertex in B, and G[A] denotes the subgraph of G induced by A. If w is a vertex of G and A ⊆ V (G), then a (w, A)-fan is a set of |A| paths from w to A, where any two paths have only w in common. If G is connected and not complete, then the connectivity of G, κ(G), is the smallest number of vertices whose deletion renders G disconnected. The (not necessarily simple) plane graph G is a triangulation (resp. quadrangulation) if every face is a triangle (resp. 4-cycle). A simple triangulation (resp. simple quadrangulation) is a triangulation (resp. quadrangulation) whose underlying graph is simple. i.e. has no multiple edges. The graph H is a minor of the graph G, if H can be obtained from a subgraph of G by edge-contractions.
By C n , K n and K n we mean the cycle, the complete graph, and the edgeless graph on n vertices, respectively. If G and H are graphs then G + H denotes the graph obtained from the union of G and H by adding edges joining every vertex of G to every vertex of H. if n = 3k + 2, which they conjectured to be optimal. (Note that this sequence is present in the On-Line Encyclopedia of Integer Sequences [37] under A014125, which is the bisection of A001400. The displayed closed form is due to Bruno Berseli [37].) We [9] announced that this conjecture is asymptotically true before the paper [8] was submitted. Che and Collins [8] verified this conjecture for simple triangulations of order not exceeding 10. Using computer, we verified this conjecture for simple triangulations of order not exceeding 18, see Table 1 in [10], an earlier version of this paper. Very recently, Debarun Ghosh, Ervin Győri, Addisu Paulos, Nika Salia, Oscar Zamora verified this conjecture [22]. In this paper we prove a generalization of this conjecture asymptotically. Note that every simple triangulation is 3-connected, but a simple triangulation cannot be 6-connected because of the number of edges. Our main theorem (Theorem 2) proves that for any 3 ≤ κ ≤ 5, the Wiener index of any κconnected simple triangulation of order n is at most 1 6κ n 3 + O(n 5/2 ). We also prove in Theorem 3 that for any 2 ≤ κ ≤ 3, the Wiener index of any κ-connected simple quadrangulation of order n is at most We provide constructions matching the upper bounds of Theorems 2 and 3 for the maximum Wiener index of triangulations and quadrangulations of given connectivity. We do more, as we exhibit triangulations and quadrangulations following patterns by the residue of the order n modulo κ, which we conjecture as realizers of the maximum Wiener index. Our conjectures are based on extensive computations. We detail next these conjectures.
We constructed 4-connected simple triangulations with Wiener index see Figures 8,9,10,11. This proves that Theorem 2 is also asymptotically tight for κ = 4. Furthermore, we conjecture that the repetition of the obvious pattern in these figures provide the extremal triangulations. Using computer, we verified this conjecture for simple triangulations of order not exceeding 22, see Table 2. We constructed 5-connected simple triangulations with Wiener index 4Éva Czabarka, Peter Dankelmann, Trevor Olsen, László A. Székely † see Figures 12,13,15,16,17. This proves that Theorem 2 is also asymptotically tight for κ = 5. Furthermore, we conjecture that the repetition of the obvious pattern in these figures provide the extremal triangulations. We arrived to these conjectures using computer and also guesswork regarding the pattern. Therefore these conjectures for the 5-connected case are less supported with computational evidence than other conjectures in this paper, as we were able to do the computation only up to the order 32, see Table 3. The issue is that the pattern slowly develops, and orders following the same pattern differ by 5-therefore we do not have sufficiently many data points to have a very convincing conjecture.
We are indebted to Paul Kainen, who after hearing about our triangulation results, asked whether we can prove similar results for simple quadrangulations. Recall that any simple quadrangulation is 2-connected, but no simple quadrangulation is 4-connected. We conjecture that the maximum Wiener index of a simple quadrangulation of order n is based on Figures 18,19. Furthermore, we conjecture that the repetition of the obvious pattern in these figures provide the extremal quadrangulations. Using computer, we verified this conjecture for simple quadrangulations of order not exceeding 20, see Table 4. We conjecture that the maximum Wiener index of a 3-connected simple quadrangulation of order n is if n = 3k + 14 based on Figures 20,21,22. Furthermore, we conjecture that the repetition of the obvious pattern in these figures provide the extremal quadrangulations. Using computer, we verified this conjecture for simple quadrangulations of order not exceeding 28, see Table 5. Section 5 contains the conjectures stated so far in the form of drawings for some fixed order, but with emphasis on the general pattern: the red colored part is the repeated pattern. Even more, we conjecture based on computational evidence that those drawings not only provide the maximum Wiener index, but for sufficiently large n they are unique with this property.
We remark here that the result of [22] described by formula (2) does not hold for non-simple triangulations. For the construction of non-simple triangulations with asymptotically larger Wiener indices, see Figure 1. In fact, we conjecture that these constructions are optimal for non-simple triangulations. The non-simple quadrangulation on Figure 2 has a larger Wiener index than conjectured best simple quadrangulation on Figure 18, but difference is not in the leading term. For the rest of the paper, under the terms triangulation and quadrangulation we will always understand simple triangulation and quadrangulation.
Che and Collins noted [8] that the minimum Wiener index of a triangulation of order n is a trivial problem, as Euler's formula determines the number of edges, and there are constructions, in which every pair of vertices are at most distance two. The situation is analogous for quadrangulations. For minimizers, see Figure 3.
There is second research direction of this paper, in addition to the Wiener index. We give bounds on the total distance σ(v) and the average distance σ(v) of a vertex v, defined as the sum and the average, respectively, of the distances from v to all other vertices. Bounds on σ(v) were obtained, for example, in   [3] [19] and [43]. Of particular interest is the maximum value over all v ∈ V (G) of σ(v) in a graph G, usually referred to as the remoteness ρ(G), of G. It was shown by Zelinka [43] and, independently, by Aouchiche and Hansen [2] that the remoteness is at most n 2 . For graphs of given minimum degree δ these bounds were improved in [15] by a factor of about 3 δ+1 . For more recent results on remoteness see, for example, [16], and [42].
In this paper we give sharp upper bounds on remoteness of triangulations and quadrangulations with given connectivity in Corollary 1 and Proposition 2. The bounds are sharp in Proposition 2 and Corollary 1 by Figures 5 through 12 and Figures 14 through 22. It is not difficult to compute the distances on those figures from the black vertex to the remaining vertices and show that the sum of distances from the black vertex meets the upper bound for remoteness. Details will be provided in the Ph.D. dissertation of the third author. Our results show that the maximum remoteness among triangulations and quadrangulations of prescribed connectivity κ is achieved on those ones that are conjectured to maximize the Wiener index, except for 5-connected triangulations of order n = 5k + 3. There are, however, lots of different realizations of the maximum of remoteness in all classes that we investigate, except among quadrangulations.

Upper Bounds on the Remoteness of Triangulations and Quadrangulations
In this section we present bounds on the remoteness of triangulations and quadrangulations. A sharp upper bound on the remoteness of a triangulation of given order was given by Che and Collins [8]. We give corresponding bounds for 4-connected and 5-connected triangulations, as well as for 2-connected and 3-connected quadrangulations. We begin by stating a sharp bound on the distance of an arbitrary vertex in a κ-connected graph of given order due to Favaron, Kouider and Mahéo [21], from which we will derive some of our bounds.

Proposition 1. [21]
Let G be a κ-connected graph of order n, and x an arbitrary vertex of G. Then Every simple triangulation is 3-connected, and every simple quadrangulation is 2-connected. Proposition 1 yields thus the following sharp bounds for the remoteness of 3-connected and 4-connected triangulations and 2-connected quadrangulations.
where ε n = 0 if n ≡ 1 (mod 2), and ε n = 1 4(n−1) if n ≡ 0 (mod 2). ✷ Proposition 1 also yields good bounds for the remoteness of 5-connected triangulations and 3-connected quadrangulations. These bounds are however not sharp for all values of n. In order to obtain sharp bounds we need some additional terminology and results from [1].
Let v be a fixed vertex of a connected plane graph and i ∈ N with i < e(v). We say that a vertex such that w and w ′ share a face of G, and w and w ′′ also share a face of G.
Proof: (a) Assume that G is a 5-connected simple triangulation, v is a vertex of G, and n d−1 = 5, where d is the eccentricity of v. This implies that N d−1 is a minimum cutset of G. Hence, since G is a triangulation, N d−1 induces a cycle C of length 5. We first show that the vertices in N d are all inside C, or all outside C.
Suppose not. Then there exist vertices a, b ∈ N d such that a is inside C, and b is outside Any two of these three fans share only the vertices of N d−1 . Indeed, other than vertices in , and the edges in F v − N d−1 to three single vertices yields a graph that contains 3K 1 + C 5 as a subgraph. Hence G contains 3K 1 + C 5 as a minor. Contracting three consecutive vertices of the 5-cycle shows that this implies that G contains K 3 +3K 1 as a minor, which contradicts the planarity of G. This contradiction proves (7).
By (7) we may assume that all vertices of N d are inside the cycle C.
We now bound the sum of the degrees of the vertices in N d . Let H be the plane graph obtained from It is easy to verify that this implies x∈N d deg G (x) < 5n d whenever n d > 1. But since G is 5-connected, every vertex of G has degree at least five. Hence we conclude that n d = 1, which proves (a).
(b) Let G be a 3-connected simple quadrangulation, v a vertex of G, and d = e(v). To prove the first statement assume that Since G is a quadrangulation and thus bipartite, the set {w, w, w ′′ } is independent in G. Since G is 3-connected, the vertices w, w ′ , w ′′ have a neighbor in N d and are thus active. By Lemma 1, w and w ′ share a face, and so do w and w ′′ , as well as w ′ and w ′′ . Hence we can add edges ww ′ , ww ′′ and ww ′′ to G to obtain a plane graph (but not a quadrangulation). Let C be the cycle consisting of the edges ww ′ , w ′ w ′′ , w ′′ w. A proof similar to that in (a) shows that the vertices of N d are all inside C, or all outside C. Without loss of generality we assume the former. We now bound the sum of the degrees of the vertices in This implies x∈N d deg G (x) < 3n d whenever n d > 1. But since G is 3-connected, every vertex of G has degree at least three. Hence we conclude that n d = 1, which proves the first statement of (b). To prove the second statement of (b) assume that n d−2 = 3 and n d−1 = 4. Suppose to the contrary that The same arguments as in the proof of the first statement of (b) show that we can add the edges ww ′ , ww ′′ , w ′ w ′′ to G to obtain a plane graph, such that these three edges form a cycle C, and that the vertices in N d−1 ∪ N d are either all inside C or all outside C, without loss of generality we assume the former. Let H be the plane graph obtained from
All n i are positive integers, n 0 = 1 and d i=0 n i = n. Since G is 5-connected we also have n i ≥ 5 for all i ∈ {1, 2, . . . , d − 1}. To bound F (n 0 , n 1 , . . . , n d ) from above we assume that n is fixed, and that d ′ ∈ N and X max (n) = (n ′ 0 , n ′ 1 , . . . , n ′ d ′ ) maximise the function F among all integers d and sequences X that satisfy these constraints. We first note that n ′ 1 = n ′ 2 = · · · = n ′ d−1 = 5. Indeed, if n ′ i > 5 for some i with 1 ≤ i ≤ d ′ − 1, then decreasing n ′ i by 1 and increasing n ′ i+1 by 1 yields a new sequence X ′ that satisfies the above constraints and for which F (X ′ ) = F (X max (n)) + 1, contradicting the choice of X max (n). Also, if n ′ d ′ > 5, then decreasing n ′ d ′ by 1, appending a new entry n ′ d ′ +1 = 1 at the end and increasing d ′ by 1 yields a sequence that satisfies the requirement but whose F -value is greater, again a contradiction to the choice of X max (n). Therefore, if q and r are positive integers with 1 ≤ r ≤ 5 such that n − 1 = 5q + r, then the unique sequence maximising F subject to the above constraints is X max (n) = (1, 5, 5, . . . , 5, r), where the entry 5 appears exactly q times. If r = 1, then it is easy to see that the unique sequence with the second largest F -value satisfying the constraints is the sequence where the entry 5 appears exactly q − 1 times. CASE 1: n ≡ 2 (mod 5).

Upper Bounds on the Wiener Index of Triangulations and Quadrangulations
In this section we present asymptotically sharp upper bounds on the Wiener index of simple triangulations and simple quadrangulations, and improved bounds for simple 4-connected and 5-connected triangulations as well as simple 3-connected quadrangulations.
In the statements and proofs of our results we use the following notation. If S is a separating cycle of a plane graph G, then we denote the set of vertices inside S by A, and the set of vertices outside S by B. We often use S also for the set of vertices on this cycle, and we further let a := |A|, b := |B| and s := |S|. The following separator theorem by Miller is an important tool for the proof of our bounds.
Theorem 1. ([32]) If G is a 2-connected plane graph of order n whose faces have length at most ℓ, then G has a separating cycle S of length at most 2 2⌊ℓ/2⌋n, such that a, b ≤ 2 3 n. We now define a plane graph which will be used in the proof of the main result of this section. Definition 1. For p ∈ N with p ≥ 3 let F p be the plane graph constructed as follows. Let C = u 0 , u 1 , . . . , u p−1 , u 0 be a cycle of length p. Inside C we add a cycle C ′ = v 0 , v 1 , . . . , v 2p−1 v 0 of length 2p and edges u i v 2i−1 , u i v 2i , u i v 2i+1 for i = 0, 1, . . . , p − 1, with indices taken modulo p for the u i and modulo 2p for the v i . Inside C ′ we add a cycle C ′′ = w 0 , w 1 , . . . , w 2p−1 , w 0 of length 2p and edges v i w i , v i w i+1 for i = 0, 1, . . . , 2p − 1, with all indices taken modulo 2p. Inside C ′′ we add a new vertex z and join it to every vertex of C ′′ . The graph F 4 is shown in Figure 4. We define F ′ p to be a plane graph with the same vertex and edge set as F p , but with the cycle C ′ outside the cycle C, the cycle C ′′ outside the cycle C ′ , and z lying in the unbounded face whose boundary is C ′′ . Proof: (a) It is easy to verify that any two vertices of F p are joined by five internally disjoint paths, hence F p is 5-connected.
(b) and (c) follow directly from F p being 5-connected. for every κ-connected simple triangulation of order n.
Proof: Our proof is by induction on n. Define D := max{D 1 , D 2 }, where D 1 is the smallest real x for which the inequality W (G) ≤ 1 6κ n 3 + xn 5/2 holds for all κ-connected simple triangulations G of order at most 10 4 , and D 2 is the smallest real x for which 8.1 + 0.76x ≤ x holds. We prove by induction on n that for all simple triangulations G of order n, Now (9) holds for all n ≤ 10 4 by the choice of D. Let n > 10 4 . By our induction hypothesis we may assume that (9) holds for all graphs of order less than n.
Since G is 2-connected, it follows by Theorem 1 that G contains a separating cycle S = t 0 t 1 . . . t s−1 t 0 with a, b ≤ 2 3 n, where A, B, a, b, s are as in Theorem 1 and above it. Let H be the simple triangulation obtained from the plane graph G − A as follows. We first delete all edges between non-consecutive vertices of S that run inside the cycle S. Inside S we insert the graph F s by identifying the cycles S and C, specifically t i ∈ S with u i ∈ V (F s ) for i = 0, 1, . . . , s − 1. Clearly, H is a simple triangulation of order b + 5s + 1. Similarly let K be the simple triangulation of order a + 5s + 1 obtained from the plane graph G − B by deleting all edges between non-consecutive vertices of S that run outside the cycle S and inserting F ′ s (a copy of F s ) into the unbounded face, bounded by the vertices of S, by identifying t i ∈ S with u i ∈ V (F ′ s ) for i = 0, 1, . . . , s − 1. For an illustration, see Figure 4. We claim that H and K are κ-connected.
We prove (10) only for H, the proof for K is analogous. Let u, v be two arbitrary vertices of H. It suffices to show that there exist κ internally disjoint (u, v)-paths in H. First assume that both, u and v, are in V (F s ), then it follows from Lemma 3(a) and κ ≤ 5 that there are κ internally disjoint (u, v)-paths in F s , and thus in H. Now assume that exactly one of the two vertices, say u, is in V (F s ). Fix a vertex w ∈ A. It follows from the κ-connectedness of G that in G there exist κ internally disjoint (w, v)-paths P 1 , P 2 , . . . , P κ . For i = 1, 2, . . . , κ let w i be the last vertex of P i on C, and let P ′ i be the (w i , v)-section of P i . By Lemma 3(b), F s contains a (u, {w 1 , . . . , w κ })-fan F . Then F together with P ′ 1 , . . . , P ′ κ yields a collection of κ internally disjoint (u, v)-paths in H. Finally assume that both, u and v, are not in V (F s ). Then it follows from the κ-connectedness of G that there exists internally disjoint (u, v)-paths P 1 , P 2 , . . . , P κ in G. If none of these contains a vertex in V (F s ), then P 1 , P 2 , . . . , P κ form a collection of κ internally disjoint (u, v)-paths in H. If some of the paths, P 1 , . . . , P k say, contain a vertex of V (F s ), then let a i and a ′ i be the first and last vertex, respectively, of P i in V (F s ). Let M = {a 1 , . . . , a k } and M ′ = {a ′ 1 , . . . , a ′ k }. By Lemma 3(c), F s contains k disjoint paths Q 1 , . . . , Q k from M to M ′ . Then the (u, a i )-sections and the (a ′ i , v)-sections of the paths P i together with Q 1 , . . . , Q k and the paths P k+1 , . . . , P κ form a collection of κ internally disjoint (u, v)-paths in H. This proves (10).
The two graphs H and K have exactly the vertices in S in common. We now bound the Wiener index of G in terms of the Wiener indices of H and K, and the total distance of z in H and K.
Indeed, for any two vertices x and y of G that are both in B ∪ S, we have d G (x, y) ≤ d H (x, y) + s 2 since a shortest (x, y)-path in H either contains only vertices in B ∪ S, in which case it is also a path in G, or it contains vertices in V (F s ) − S, in which case replacing the segment between the first and last occurrence of a vertex in V (F s ) − S in the path by a segment of the cycle S that contains at most s/2 vertices yields an (x, y)-path in G. Similarly, if x and y are both in A ∪ S, then d G (x, y) ≤ d K (x, y) + s 2 . Finally, if x ∈ A and y ∈ B, then we can obtain an (x, y)-path in G from the concatenation of an (x, z)-path in K and a (z, y)-path in H by replacing z with a segment of S containing at most s/2 vertices. This proves (11).
Since n ≥ 10 4 , we have 338 κ n 2 + 8788 3κ n 3/2 < 2n 5/2 . Also, 13 κ + 0.76 D + 1 + 2 −1/2 < 6.1 + 0.76 D, and so The following bound on the Wiener index of simple quadrangulations is proved in a similar way. The only difference is that a slightly modified version Q p of the plane graph F p is used in the proof. For an even p with p ≥ 4 let Q p be the plane graph obtained from a cycle C = u 0 , u 1 , . . . , u p−1 , u 0 of length p, inside which we add a cycle C ′ = v 0 , v 1 , . . . , v p−1 , v 0 of length p and edges u i v i for i = 0, 1, . . . , p − 1, inside which we add a vertex z and joint it to all v i with i even. It is easy to verify that a 3-connected quadrangualation with the insertion of Q p stays 3-connected. Apart from this difference, the proof of Theorem 3 follows closely that of Theorem 2, hence we omit the proof. for every κ-connected simple quadrangulation G of order n. ✷ The leading coefficients in the bounds in Theorems 2 and 3 are optimal. This is shown by the graphs in

Computational Results and Conjectures
This section contains numerous figures and tables summarizing months of computer searches. None of this would have been possible without the help provided by Plantri, a program that generates triangulations and quadrangulation on numerous surfaces. For each category of problem (triangulations, 4-connected triangulations, 5-connected triangulations, quadrangulations and 3-connected quadrangulations) there is a table, which summarizes the largest Wiener index and remoteness found for a given order in that category, along with "Count", telling how many graphs attain the optimal value. Note that remoteness in this section is not normalized to keep the calculations in the domain of integers. In other words, in the Tables we show (n − 1)ρ(G) under the name of "Remoteness". Our Wiener index findings match those of [8] for triangulations. The number of isomorphism classes that our code searched matches the numbers in [5], [6], [7], [31], [36], verifying that the values that the search provides are in fact maximal. In each figure below, purple edges represent the repeating pattern and the black node marks a vertex which maximizes the remoteness. The computational evidence suggests that for sufficiently large order, the maximum Wiener index is uniquely realized in every category, while remoteness is not, except for quadrangulations.