arXiv:1106.0958v1 [math.CO] 6 Jun 2011 eeaie h eutfral(o ipe 3-quasi-planar simple) for (not conjecture all gra for 3-quasi-planar result simple the for generalized conjecture this prove to first ainlSineFudto,GatN.200021-125287/1. No. Grant Foundation, Science National on yuigarsl fAkra 1.Vlr[0 rvdta ever that proved [20] Valtr [1]. Ackerman of result a using by A Introduction 1 a tmost at has olw rmElrsPlhda oml hteeytopolog every that the Formula graph Polyhedral topological Euler’s extremal from in follows problem classic gr a the been has then segments, straight-line as drawn graph are topological only is edges We graph common. the topological in If A point that self-loops. v at or through cross edges i pass properly tangencies, have must not to they may allowed then not they are but edges intersect, the Furthermore, to allowed arc are non-self-intersecting by arcs represented are edges its and de a tms 3 most at has edges nscin96o 5)sae htfrayfixed any for that states [5]) of p 9.6 are section graphs in 2-quasi-planar Hence 1 edges. crossing pairwise oooia graph topological ∗ n PL asne Email: Lausanne. EPFL, For idn h aiu ubro de natplgclgahwi graph topological a in edges of number maximum the Finding tteritreto) n19,Pc,Sarki n zgd [1 Szegedy and Shahrokhi, Pach, 1996, In intersection). their at simple oooia rp is graph topological eetyipoe frlarge (for improved recently htalsc rpscnana ot( most at contain graphs such all that cemn ucinand function Ackermann etcshsa most at has vertices k oooia rp is graph topological A ≥ ,Pc,Sarki n zgd 1]soe hteeysimpl every that showed [16] Szegedy and Shahrokhi, Pach, 5, k c qaipaa rp otisa most at contains graph -quasi-planar k k n 4. = de,where edges, cross sagahdani h ln uhta t etcsaerepres are vertices its that such plane the in drawn graph a is n − fteritrossaeapoint. a share interiors their if simple de.Atplgclgahis graph topological A edges. 6 [email protected] c k c n k k feeypi fisegsitreta otoc ete tavertex a at (either once most at intersect edges its of pair every if (log c sacntn htdpnsol on only depends that constant a is -quasi-planar k k k yFxadPc 8 to [8] Pach and Fox by ) sacntn htdpnsol on only depends that constant a is n qaipaa graphs planar -quasi ) 2 k − uut1,2018 13, August 4 h uhrgaeul cnwegstespotfo h Sw the from support the acknowledges gratefully author The . n nrwSuk Andrew de.Ti on a eipoe to improved be can bound This edges. simple log Abstract fi osntcontain not does it if 2 n 1 O )2 > k feeypi fisegsitreta otonce. most at intersect edges its of pair every if α n c k (log ( n ,every 0, ) oncigtecrepnigpit.The points. corresponding the connecting s ∗ rps eety cemn[]poe the proved [1] Ackerman Recently, graphs. n de,where edges, h.LtrPc,Radoiˇci´c, Pach, T´oth Later [14] and phs. n ) k (log 2 clgahon graph ical aa.A l ojcue(e Problem (see conjecture old An lanar. r se[,3 ,6 ,1,1,1,2].It 21]). 19, 15, 10, 8, 6, 4, 3, [2, (see ory e,i w de hr nitro point, interior an share edges two if .e., -quasi-planar k − 4 y n p is aph k ) k rie xetfrterendpoints. their for except ertices n qaipaa rp on graph -quasi-planar O de.Ti pe on was bound upper This edges. ]soe htevery that showed 6] . k -vertex (log osdrgah ihu parallel without graphs consider ariecosn de.A edges. crossing pairwise hafridncosn pattern crossing forbidden a th k α ) geometric k nti oe eshow we note, this In . ( gra ta.wr the were al. et Agarwal . n n fi osntcontain not does it if , eoe h inverse the denotes ) k etcsadn crossing no and vertices qaipaa geometric -quasi-planar e k qaipaa graph -quasi-planar w de fa of edges Two . ne ypoints by ented c n k n -vertex (log n or vertices n ) 2 k − iss k 8 graph contains at most O(n log n) edges. Later, he extended this result to simple topological graphs with edges drawn as x-monotone curves [21]. Pach, Radoiˇci´c, and T´oth showed that every n-vertex 4k−12 (not simple) k-quasi-planar graph has at most ckn(log n) edges, which can also be improved 4k−16 to ckn(log n) by a result of Ackerman [1]. Very recently, Fox and Pach [8] improved (for large k) the exponent in the polylogarithmic factor for simple topological graphs. They showed that every simple k-quasi-planar graph on n vertices has at most n(c log n/ log k)c log k edges, where c is an absolute constant. Our main result is the following.
Theorem 1.1. Let G = (V, E) be an n-vertex simple k-quasi-planar graph. Then E(G) 2 αck (n) | | ≤ (n log n)2 , where α(n) denotes the inverse Ackermann function and ck is a constant that depends only on k.
In the proof of Theorem 1.1, we apply results on generalized Davenport-Schinzel sequences. This method was used by Valtr [21], who showed that every n-vertex simple k-quasi-planar graph with ck edges drawn as x-monotone curves has at most 22 n log n edges, where c is an absolute constant. Our next theorem extends his result to (not simple) topological graphs with edges drawn with x-monotone curves, and moreover we obtain a slightly better upper bound.
Theorem 1.2. Let G = (V, E) be an n-vertex (not simple) k-quasi planar graph with edges drawn 3 as x-monotone curves. Then E(G) 2ck n log n, where c is an absolute constant. | |≤ 2 Generalized Davenport-Schinzel sequences
The sequence u = a1, a2, ..., am is called l-regular if any l consecutive terms are pairwise different. For integers l,t 2, the sequence ≥
S = s1,s2, ..., slt of length l t is said to be of type up(l,t) if the first l terms are pairwise different and for i = 1, 2, ..., l · s = s = s = = s . i i+l i+2l · · · i+(t−1)l For example,
a, b, c, a, b, c, a, b, c, a, b, c, would be an up(3, 4) sequence. By applying a theorem of Klazar on generalized Davenport-Schinzel sequences, we have the following.
Theorem 2.1 ([11]). For l 2 and t 3, the length of any l-regular sequence over an n-element ≥ ≥ alphabet that does not contain a subsequence of type up(l,t) has length at most
lt n l2(lt−3) (10l)10α (n). · · For l 2, the sequence ≥
S = s1,s2, ..., s3l−2
2 of length 3l 2 is said to be of type up-down-up(l), if the first l terms are pairwise different, and − for i = 1, 2, ..., l,
si = s2l−i = s(2l−2)+i. For example,
a, b, c, d, c, b, a, b, c, d, would be an up-down-up(4) sequence. Valtr and Klazar showed the following.
Lemma 2.2 ([12]). For l 2, the length of any l-regular sequence over an n-element alphabet ≥ containing no subsequence of type up-down-up(l) has length at most 2O(l)n.
For more results on generalized Davenport-Schinzel sequences, see [13, 18, 17].
3 Simple topological graphs
In this section, we will prove Theorem 1.1. For any partition of V (G) into two disjoint parts, V1 and V2, let E(V1,V2) denote the set of edges with one endpoint in V1 and the other endpoint in V2. The bisection width of a graph G, denoted by b(G), is the smallest nonnegative integer such that there is a partition of the vertex set V = V ˙ V with 1 V V 2 V for i = 1, 2, and 1 ∪ 2 3 · | | ≤ | i|≤ 3 · | | E(V ,V ) = b(G). We will use the following result by Pach et al. | 1 2 | Lemma 3.1 ([16]). If G is a graph with n vertices of degrees d1, ..., dn, then
n b(G) 7cr(G)1/2 + 2v d2, ≤ u i uXi=1 t where cr(G) denotes the crossing number of G.
Since n d2 2n E(G) holds for every graph, we have i=1 i ≤ | | P b(G) 7cr(G)1/2 + 3 E(G) n. (1) ≤ p| | Proof of Theorem 1.1. Let k 5 and f (n) denote the maximum number of edges in a simple ≥ k k-quasi-planar graph on n vertices. We will prove that
c f (n) (n log2 n)2α k (n) k ≤ 2 where c = 105 2k +2k. For sake of clarity, we do not make any attempts to optimize the value of k · c . We proceed by induction on n. The base case n< 7 is trivial. For the inductive step n 7, let k ≥ G = (V, E) be a simple k-quasi-planar graph with n vertices and m = fk(n) edges, such that the vertices of G are labeled 1 to n. The proof splits into two cases. Case 1. Suppose that cr(G) m2/(104 log2 n). By (1), there is a partition V (G) = V V with ≤ 1 ∪ 2 V , V 2n/3 and the number of edges with one vertex in V and one vertex in V is at most | 1| | 2|≤ 1 2
3 m b(G) 7cr(G)1/2 + 3√mn 7 + 3√mn. ≤ ≤ 100 log n Let n = V and n = V . Now if 7m/(100 log n) 3√mn, then we have 1 | 1| 2 | 2| ≤ m 43n log2 n ≤ and we are done since α(n) 2 and k 5. Therefore, we can assume 7m/(100 log n) > 3√mn, ≥ ≥ which implies m b(G) . (2) ≤ 7 log n By the induction hypothesis and equation (2), we have
m f (n )+ f (n )+ b(G) ≤ k 1 k 2 c c n log2(2n/3) 2α k (n) + n log2(2n/3) 2α k (n) + b(G) ≤ 1 2 c n log2(2n/3) 2α k (n) + m ≤ 7 log n c c c (n log2 n)2α k (n) 2n2α k (n) log n log(3/2) + n2α k (n) log2(3/2) + m ≤ − 7 log n which implies
2 1 c 2 log(3/2) log (3/2) m 1 (n log2 n)2α k (n) 1 + . − 7 log n ≤ − log n log2 n Hence
−1 2 −2 c 1 2 log(3/2) log n + log (3/2) log n c m (n log2 n)2α k (n) − (n log2 n)2α k (n). ≤ 1 1/(7 log n) ≤ − Case 2. Now suppose that cr(G) m2/(104 log2 n). By a simple averaging argument, there exists ≥ an edge e = uv such that at least 2m/(104 log2 n) other edges cross e. Fix such an edge e = uv, and let E′ denote the set of edges that cross e. We order the edges in E′ = e , e , ..., e ′ , in the order that they cross e from u to v. Now { 1 2 |E |} we create two sequences S = a , a , ..., a ′ and S = b , b , ..., b ′ as follows. For each e E′, 1 1 2 |E | 2 1 2 |E | i ∈ as we move along edge e from u to v and arrive at edge ei, we turn left and move along edge ei until we reach its endpoint ui. Then we set ai = ui. Likewise, as we move along edge e from u to v and arrive at edge ei, we turn right and move along edge ei until we reach its other endpoint vi. Then set b = v . Thus S and S are sequences of length E′ over the alphabet 1, 2, ..., n . See i i 1 2 | | { } Figure 1 for a small example. Now we need the following two lemmas. The first one is due to Valtr.
Lemma 3.2 ([21]). For l 1, at least one of the sequences S ,S defined above contains an ≥ 1 2 l-regular subsequence of length at least E′ /(4l). | |
4 v
v3 v 4
v v 1 5