A lower bound on the order of the largest induced linear forest in
triangle-free planar graphs
François Dross
Université
de Montpellier, LIRMM
Mickael Montassier
Université
de Montpellier, LIRMM
Alexandre Pinlou
Université Paul-Valéry Montpellier 3,
LIRMM
Abstract
We prove that every triangle-free planar graph of order n and size m has an induced linear forest with at least 119n−2m vertices, and thus at least 115n+8 vertices. Furthermore, we show that there are triangle-free planar graphs on n vertices whose largest induced linear forest has order ⌈2n⌉+1.
1 Introduction
In this article, we only consider simple finite graphs. All considered planar graphs are supposed to be embedded in the plane.
We look into the problem of finding large induced forests in planar graphs. Albertson and Berman [3] conjectured that every planar graph admits an induced forest on at least half of its vertices. This conjecture, if true, would be tight, as shown by the disjoint union of copies of the complete graph on four vertices. One of the motivations of this conjecture is that it would imply that every planar graph admits an independent set on at least one fourth of its vertices, the only known proof of which relies on the Four Colour Theorem. However, this conjecture appears to be very hard to prove. The best known result for planar graphs is that every planar graph admits an induced forest on at least two fifths of its vertices. This is a consequence of the theorem of 5-acyclic colourability of planar graphs of Borodin [4].
The conjecture of Albertson and Berman has been proved and strengthened for smaller classes of graphs. For example, Hosono [7] showed that every outerplanar graph admits an induced forest on at least two thirds of its vertices, which is tight. Akiyama and Watanabe [1], and Albertson and Rhaas [2] independently conjectured that every bipartite planar graph admits an induced forest on at least five eighths of its vertices, which is tight. For triangle-free planar graphs (and thus in particular for bipartite planar graphs), it is proven that every triangle-free planar graph of order n and size m admits an induced forest of order at least (38n−7m)/44, and thus at least (6n+7)/11 [6].
An interesting variant of this problem is to look for large induced forests with bounded maximum degree. A forest with maximum degree 2 is called a linear forest.
The problem for linear forests was solved for outerplanar graphs by Pelsmajer [8]: every outerplanar graph admits an induced linear forest on at least four sevenths of its vertices, and this is tight. More generally, the problem for a forest of maximum degree at most d, with d≥2, was solved for graphs with treewidth at most k for all k by Chappel and Pelsmajer [5]. Their result in particular extends the results of Hosono and Pelsmayer on outerplanar graphs to series-parallel graphs, and generalises it to graphs of bounded treewidth.
In this paper we focus on linear forests. Chappel conjectured that every planar graph admits an induced linear forest on at least four ninths of its vertices. Again, this would be tight if true. Poh [9] proved that every planar graph can have its vertices be partitioned into three sets, each inducing a linear forest, and thus that every planar graph admits an induced linear forest on at least one third of its vertices. In this paper, we prove and strengthen Chappel’s conjecture for a smaller class of graphs, the class of triangle-free planar graphs. Observe that planar graphs with arbitrarily large girth can have an arbitrarily large treewidth, so in this setting the best result known to date is that every triangle-free planar graph admits an induced linear forest on at least one third of its vertices.
We prove the following theorem:
Theorem 1**.**
Every triangle-free planar graph of order n and size m admits an induced linear forest of order at least 119n−2m.
Thanks to Euler’s formula, we can derive the following corollary:
Corollary 2**.**
Every triangle-free planar graph of order n admits an induced linear forest of order at least 115n+8.
For a graph G=(V,E), and S⊂V, let G[S] denote the subgraph of G induced by S.
Note that we cannot hope to get a better lower bound than 2n+1. Indeed, we prove the following claim:
Claim 3**.**
For all integer n≥2, there exists a triangle-free planar graph of order n whose largest induced linear forest has order ⌈2n⌉+1.
Proof.
Let us build such a graph for n=2k. For odd n, adding an isolated vertex to the graph of order n−1 yields the result.
Let Gk be defined by Gk=(⋃1≤i≤k{ui,vi},⋃1≤i≤k−1 {uiui+1,uivi+1,viui+1,vivi+1}), as represented in Figure 1. Let us prove by induction on k≥1 that the largest induced linear forest of Gk has order k+1.
For k=1, G1 is the graph with two vertices and no edge, and G1 is its own largest induced linear forest, with order 2=k+1.
For k=2, G2 is a cycle of length 4, any three vertices of G2 induce a linear forest of order 3=k+1, and G2 is not a linear forest (thus it has no induced linear forest of order 4).
For k=3, {u1,v1,u3,v3} induces a linear forest in G3, and it is easy to check that no five vertices of G3 induce a linear forest.
Suppose k≥4. By induction hypothesis, Gk−1, Gk−2, and Gk−3 have a largest induced forest of order k, k−1, and k−2 respectively. Adding uk and vk to any induced linear forest of Gk−2 leads to an induced linear forest of Gk, thus Gk has an induced linear forest of order k+1. All that remains to prove is that Gk has no induced linear forest of order at least k+2.
Let F⊂V(Gk) be a set inducing a linear forest of Gk. Let us prove that ∣F∣≤k+1. As F\{uk,vk} induces a linear forest in Gk−1, we have ∣F\{uk,vk}∣≤k. Similarly, ∣F\{uk,vk,uk−1,vk−1}∣≤k−1 and ∣F\{uk,vk,uk−1,vk−1,uk−2,vk−2}∣≤k−2. If ∣{uk,vk}∩F∣≤1, then ∣F∣=∣{uk,vk}∩F∣+∣F\{uk,vk}∣≤k+1. Suppose now that ∣{uk,vk}∩F∣>1, i.e. {uk,vk}⊂F. At most one of uk−1 and vk−1 is in F, otherwise Gk[F] has a cycle. If {uk−1,vk−1}∩F=∅, then ∣F∣=∣{uk,vk}∩F∣+∣{uk−1,vk−1}∩F∣+∣F\{uk,vk,uk−1,vk−1}∣≤2+k−1=k+1. Assume now that {uk−1,vk−1}∩F=∅, w.l.o.g. uk−1∈F. We have {uk−2,vk−2}∩F=∅, otherwise uk−1 has degree at least 3 in Gk[F]. Hence, ∣F∣=∣{uk,vk}∩F∣+∣{uk−1,vk−1}∩F∣+∣{uk−2,vk−2}∩F∣+∣F\{uk,vk,uk−1,vk−1,uk−2,vk−2}∣≤2+1+k−2=k+1.
∎
2 Proof of Theorem 1
Consider a graph G=(V,E). For a set S⊂V, let G−S be the graph obtained from G by removing the vertices of S and all the edges incident to a vertex of S. If x∈V, then we denote G−{x} by G−x. For a set S of vertices such that S∩V=∅, let G+S denote the graph obtained from G by adding the vertices of S. If x∈/V, then we denote G+{x} by G+x. For a set F of pairs of vertices of G such that F∩E=∅, let G+F be the graph constructed from G by adding the edges of F. If e is a pair of vertices of G and e∈/E, then we denote G+{e} by G+e. If x∈V, then we denote the neighbourhood of x, that is the set of the vertices adjacent to x, by N(x). For a set S⊂V, we denote the neighbourhood of S, that is the set of vertices in V\S that are adjacent to at least an element of S, by N(S). We denote ∣V∣ by ∣G∣ and ∣E∣ by ∣∣G∣∣.
We call a vertex of degree d, at least d, and at most d, a d-vertex, a d+-vertex, and a d−-vertex respectively. We call a cycle of length l an l-cycle, and by extension a face of length l an l-face.
Let P4 be the class of triangle-free planar graphs. Let G=(V,E) be a counter-example to Theorem 1 with the minimum order. Let n=∣G∣ and m=∣∣G∣∣. We will use the schemes presented in Observations 4–6 many times throughout this paper.
Observation 4**.**
Let α, β, γ be integers satisfying α≥1, β≥0, γ≥0.
Let H∗∈P4 be a graph with ∣H∗∣=n−α and ∣∣H∗∣∣≤m−β.
By minimality of G, H∗ admits an induced linear forest of order at least 119(n−α)−112(m−β).
Given an induced linear forest F∗ of H∗ of order ∣F∗∣≥119(n−α)−112(m−β), if there is an induced linear forest F of G of order ∣F∣≥∣F∗∣+γ, then as ∣F∣<119n−112m, we have γ<119α−112β.
Observation 5**.**
Suppose L⊂V induces a linear forest in G, and M is a set of vertices such that M∩L=∅ and M⊃N(L).
Let G′=G−M−L. See Figure 2 (left) for an illustration.
By minimality of G, G′ admits a linear forest F′ with ∣F′∣≥119∣G′∣−112∣∣G′∣∣. Observe that F=G[V(F′)∪L] is an induced linear forest of G.
As G is a counter-example to Theorem 1, ∣F∣<119∣G∣−112∣∣G∣∣.
Therefore ∣L∣=∣F∣−∣F′∣<119(∣M∣+∣L∣)−112(∣∣G∣∣−∣∣G′∣∣).
Observation 6**.**
Suppose L⊂V induces a linear forest in G. Suppose there is a set of vertices M and two vertices u∈L and v such that M∩L=∅, {v}=N(L)∖M, and {u}=N(v)∩L.
Let G′=G−M−L. Suppose v is a 1−-vertex in G′ and u is a 1−-vertex in G[L]. See Figure 2 (right) for an illustration.
By minimality of G, G′ admits a linear forest F′ with ∣F′∣≥119∣G′∣−112∣∣G′∣∣. Observe that F=G[V(F′)∪L] is an induced linear forest of G.
As G is a counter-example to Theorem 1, ∣F∣<119∣G∣−112∣∣G∣∣.
Therefore ∣L∣=∣F∣−∣F′∣<119(∣M∣+∣L∣)−112(∣∣G∣∣−∣∣G′∣∣).
Now we want to prove some structural properties of G, so that we can later show that the counter-example G does not exist, and thus that Theorem 1 is true.
First note that G is connected, otherwise one of its components would be a smaller counter-example to Theorem 1. Then note that every vertex of G has degree at most 4. Otherwise, by considering a vertex of degree at least 5 and by Observation 4 applied to H∗=G−v with (α,β,γ)=(1,5,0) and F=F∗, we have 0<119−5112, a contradiction.
Let us define the notion of a chain of G (or simple chain) of G which is a quadruplet C=(P,N,u,v) such that:
P⊂V, N⊂V∖P, u∈P, and v∈V\(N∪P);
G[P] is a linear forest;
vertex u is a 1−-vertex of G[P], and N(v)∩P={u};
N(P)⊂N∪{v} in G;
vertex v is a 2−-vertex in G−(N∪P).
See Figure 3 (left) for an illustration. We will use the following notation for a chain C=(P,N,u,v) of G:
∣C∣=∣P∣+∣N∣;
G−C=G−(N∪P);
d(C) is the degree of v in G−C (thus d(C)≤2);
∣∣C∣∣=∣∣G∣∣−∣∣G−C∣∣.
We will now prove the following lemma:
Lemma 7**.**
For every chain C=(P,N,u,v) of G, ∣P∣<119∣C∣−112(∣∣C∣∣−21).
Proof.
Let us consider by contradiction a chain C=(P,N,u,v) such that ∣P∣≥119∣C∣−112(∣∣C∣∣−21) maximizing ∣C∣.
Suppose d(C)=0. Let G′=(G−C)−v. The set P∪{v} induces a linear forest, and its neighbourhood is a subset of N. By Observation 5 applied to L=P∪{v} and M=N, we have ∣P∣+1<119(∣C∣+1)−112∣∣C∣∣. As ∣P∣≥119∣C∣−112(∣∣C∣∣−21), it follows that 1<119−11221, a contradiction.
Suppose d(C)=1. The set P induces a linear forest, and its neighbourhood is a subset of N∪{v}. Furthermore, N(v)∩P={u}, N(u)∩(G−C)={v}, and u and v are 1-vertices in P and G−C respectively. By Observation 6 applied to L=P and M=N, we have ∣P∣<119∣C∣−112∣∣C∣∣, thus ∣P∣<119∣C∣−112(∣∣C∣∣−21), a contradiction.
Suppose d(C)=2. Let w0 and w1 be the neighbours of v in G−C.
Suppose one of the wi’s, say w0, has degree 1 in G−C. Let G′=(G−C)−{v,w0,w1}. The set P∪{v,w0} induces a linear forest and its neighbourhood is a subset of N∪{w1}. By Observation 5 applied to L=P∪{v,w0} and M=N∪{w1}, we have ∣P∣+2<119(∣C∣+3)−112(∣∣C∣∣+2). As ∣P∣≥119∣C∣−112(∣∣C∣∣−21), it follows that 2<1193−11225, a contradiction.
Suppose the wi’s both have degree 2 in G−C. Observe that they are not adjacent since G is triangle-free. The set P∪{v} induces a linear forest, and its neighbourhood is a subset of N∪{w0,w1}. Furthermore, N(w1)∩(P∪{v})={v}, N(v)∩((G−C)−{v,w0})={w1}, and v and w1 are 1-vertices in P∪{v} and G−C−{v,w0} respectively. By Observation 6 applied to L=P∪{v} and M=N∪{w0}, we have ∣P∣+1<119(∣C∣+2)−112(∣∣C∣∣+3). As ∣P∣≥119∣C∣−112(∣∣C∣∣−21), it follows that 1<1192−11227, a contradiction.
Suppose the wi’s both have degree 4 in G−C. Again they are not adjacent since G is triangle-free.
The set P∪{v} induces a linear forest, and its neighbourhood is a subset of N∪{w1,w0}. By Observation 5 applied to L=P∪{v} and M=N∪{w0,w1}, we have ∣P∣+1<119(∣C∣+3)−112(∣∣C∣∣+8). As ∣P∣≥119∣C∣−112(∣∣C∣∣−21), it follows that 1<1193−112217, a contradiction.
Suppose one of the wi’s, say w0, is a 3−-vertex in G−C and the other one is a 3+-vertex in G−C. Let C′=(P∪{v},N∪{w1},v,w0). Then C′ is a chain of G, and by maximality of ∣C∣, we have ∣P∣+1<119(∣C∣+2)−112(∣∣C∣∣+27). As ∣P∣≥119∣C∣−112(∣∣C∣∣−21), it follows that 1<1192−1124, a contradiction.
∎
Let us now define a new notion quite similar to the notion of chain. A double chain of G is a sextuplet C=(P,N,u0,u1,v0,v1), so that:
P⊂V, N⊂V∖P, u0∈P, u1∈P, v0∈V\(N∪P) and v1∈V\(N∪P);
v0=v1;
G[P] is a linear forest;
u0 and u1 are 1−-vertices of G[P] if they are distinct, a [math]-vertex of G[P] if they are equal, and for i∈{0,1}, N(vi)∩P={ui};
N(P)⊂N∪{v0}∪{v1};
v0 and v1 are 2−-vertices in G−(N∪P).
See Figure 3 (right) for an illustration. We will use the following notations for a double chain C=(P,N,u0,u1,v0,v1) of G:
∣C∣=∣P∣+∣N∣;
G−C=G−(N∪P);
d0(C) is the degree of v0 in G−C (thus d0(C)≤2);
d1(C) is the degree of v1 in G−C (thus d1(C)≤2);
∣∣C∣∣=∣∣G∣∣−∣∣G−C∣∣.
A double chain C=(P,N,u0,u1,v0,v1) of G such that v0 and v1 are on different components of G−C is called a separating double chain of G.
We will now prove the following lemmas:
Lemma 8**.**
For every double chain C=(P,N,u0,u1,v0,v1) of G, ∣P∣<119∣C∣−112(∣∣C∣∣−3).
Proof.
Let us consider by contradiction a double chain C=(P,N,u0,u1,v0,v1) such that ∣P∣≥119∣C∣−112(∣∣C∣∣−3) maximizing ∣C∣.
Suppose first that v0 and v1 are not adjacent.
Suppose d0(C)=0. Then (P∪{v0},N,u1,v1) is a simple chain of G. By Lemma 7, ∣P∣+1<119(∣C∣+1)−112(∣∣C∣∣−21). As ∣P∣≥119∣C∣−112(∣∣C∣∣−3), we have 1<119−25112, a contradiction.
Suppose d0(C)=1. Let w be the neighbour of v0 in G−C. Then (P∪{v0},N∪{w},u1,v1) is a simple chain of G. By Lemma 7, ∣P∣+1<119(∣C∣+2)−112(∣∣C∣∣+21). As ∣P∣≥119∣C∣−112(∣∣C∣∣−3), we have 1<1192−11227, a contradiction.
Suppose d0(C)=2. Let w0 and w1 be the neighbours of v0 in G−C.
Suppose one of the wi’s, say w0, has degree 1 in G−C.
Then (P∪{v0,w0},N∪{w1},u1,v1) is a simple chain of G. By Lemma 7, ∣P∣+2<119(∣C∣+3)−112(∣∣C∣∣+23). As ∣P∣≥119∣C∣−112(∣∣C∣∣−3), we have 2<1193−11229, a contradiction.
Suppose that the wi’s both have degree 2 in G−C. Note that they are not adjacent since G is triangle-free. They may, however, be adjacent to v1.
Suppose both of the wi’s are adjacent to v1. The set P∪{v0,v1} induces a linear forest in G, and its neighbourhood is a subset of N∪{w0,w1}. By Observation 5 applied to L=P∪{v0,v1} and M=N∪{w0,w1}, we have ∣P∣+2<119(∣C∣+4)−112(∣∣C∣∣+4). As ∣P∣≥119∣C∣−112(∣∣C∣∣−3), it follows that 2<1194−1127, a contradiction.
Therefore one of the wi’s, say w0, is not adjacent to v1. Let x be the neighbour of w0 in G−C distinct from v0. Suppose x has degree 4 in G−C−{v0,w1}. Now (P∪{v0,w0},N∪{w1,x},u1,v1) is a chain of G. By Lemma 7, ∣P∣+2<119(∣C∣+4)−112(∣∣C∣∣+213). As ∣P∣≥119∣C∣−112(∣∣C∣∣−3), we have 2<1194−112219, a contradiction. Therefore x is a 3−-vertex in G−C−{v0,w1}. Then (P∪{v0,w0},N∪{w1},w0,u1,x,v1) is a double chain of G, so by maximality of ∣C∣, ∣P∣+2<119(∣C∣+3)−112(∣∣C∣∣+4−3). As ∣P∣≥119∣C∣−112(∣∣C∣∣−3), we have 2<1193−1124, a contradiction.
Suppose that the wi’s are 4-vertices in G−C. Now (P∪{v0},N∪{w0,w1},u1,v1) is a chain of G. By Lemma 7, ∣P∣+1<119(∣C∣+3)−112(∣∣C∣∣+215). As ∣P∣≥119∣C∣−112(∣∣C∣∣−3), we have 1<1193−112221, a contradiction.
Suppose one of the wi, say w0, is a 3−-vertex in G−C and the other one is a 3+-vertex in G−C. Then (P∪{v0},N∪{w1},v0,u1,w0,v1) is a double chain of G. By maximality of ∣C∣, ∣P∣+1<119(∣C∣+2)−112(∣∣C∣∣+4−3). As ∣P∣≥119∣C∣−112(∣∣C∣∣−3), we have 1<1192−1124, a contradiction.
Now v0 and v1 are adjacent.
Suppose d0(C)=1 or d1(C)=1, say d0(C)=1. The set P∪{v0} induces a linear forest, and its neighbourhood is a subset of N∪{v1}. By Observation 5 applied to L=P∪{v0} and M=N∪{v1}, we have ∣P∣+1<119(∣C∣+2)−112(∣∣C∣∣+1). As ∣P∣≥119∣C∣−112(∣∣C∣∣−3), it follows that 1<1192−1124, a contradiction.
Now d0(C)=2 and d1(C)=2. Let w be the neighbour of v0 in (G−C)−v1. Note that w is not adjacent to v1, otherwise v0v1w would be a triangle in G.
Suppose w is a 2−-vertex in (G−C).
The set P∪{v0} induces a linear forest, and its neighbourhood is a subset of N∪{v1,w}. Furthermore, N(w)∩(P∪{v0})={v0}, N(v0)∩V((G−C)−{v1})={w}, and v0 and w are 1−-vertices in G[P∪{v0}] and (G−C)−{v0} respectively. By Observation 6 applied to L=P∪{v0} and M=N∪{v1}, we have ∣P∣+1<119(∣C∣+2)−112(∣∣C∣∣+3). As ∣P∣≥119∣C∣−112(∣∣C∣∣−3), it follows that 1<1192−1126, a contradiction.
Now w is a 3+-vertex in (G−C). The set P∪{v0} induces a linear forest, and its neighbourhood is a subset of N∪{v1,w}. By Observation 5 applied to L=P∪{v0} and M=N∪{w}, we have ∣P∣+1<119(∣C∣+3)−112(∣∣C∣∣+5). As ∣P∣≥119∣C∣−112(∣∣C∣∣−3), it follows that 1<1193−1128, a contradiction.
∎
Lemma 9**.**
For every separating double chain C=(P,N,u0,u1,v0,v1) of G, ∣P∣<119∣C∣−112(∣∣C∣∣−1).
Proof.
Let us consider by contradiction a separating double chain C=(P,N,u0,u1,v0,v1) such that ∣P∣≥119∣C∣−112(∣∣C∣∣−1) maximizing ∣C∣.
Suppose d0(C)=0. Then (P∪{v0},N,u1,v1) is a simple chain of G. By Lemma 7, ∣P∣+1<119(∣C∣+1)−112(∣∣C∣∣−21). As ∣P∣≥119∣C∣−112(∣∣C∣∣−1), we have 1<119−11221, a contradiction.
Suppose d0(C)=1. Let w be the neighbour of v0 in G−C. Suppose w is a 4+-vertex in G−C. Then (P∪{v0},N∪{w},u1,v1) is a simple chain of G. By Lemma 7, ∣P∣+1<119(∣C∣+2)−112(∣∣C∣∣+27). As ∣P∣≥119∣C∣−112(∣∣C∣∣−1), we have 1<1192−11229, a contradiction.
Now w is a 3−-vertex in G−C. Let C′=(P∪{v0},N,v0,u1,w,v1). One can see that C′ is a separating double chain of G, and by maximality of ∣C∣, ∣P∣+1<119(∣C∣+1)−112(∣∣C∣∣). As ∣P∣≥119∣C∣−112(∣∣C∣∣−1), we have 1<119−112, a contradiction.
Suppose d0(C)=2. Let w0 and w1 be the neighbours of v0 in G−C.
Suppose one of the wi’s, say w0, has degree 1 in G−C. We have a simple chain (P∪{v0,w0},N∪{w1},u1,v1). By Lemma 7, ∣P∣+2<119(∣C∣+3)−112(∣∣C∣∣+23). As ∣P∣≥119∣C∣−112(∣∣C∣∣−1), we have 2<1193−11225, a contradiction.
Suppose the wi’s both have degree 2 in G−C. Note that they are not adjacent since G is triangle-free. Let x be the second neighbour of w0 in G−C. Suppose x is a 4-vertex in G−C−{w1}. Then (P∪{v0,w0},N∪{w1,x},u1,v1) is a simple chain of G. By Lemma 7, ∣P∣+2<119(∣C∣+4)−112(∣∣C∣∣+213). As ∣P∣≥119∣C∣−112(∣∣C∣∣−1), we have 2<1194−112215, a contradiction. Now x is a 3−-vertex in G−C−{w1}, so (P∪{v0,w0},N∪{w1},w0,u1,x,v1) is a separating double chain of G. By maximality of ∣C∣, ∣P∣+2<119(∣C∣+3)−112(∣∣C∣∣+3). As ∣P∣≥119∣C∣−112(∣∣C∣∣−1), we have 2<1193−1124, a contradiction.
Suppose the wi’s have degree 4 in G−C. Again they are not adjacent since G is triangle-free. We have a simple chain (P∪{v0},N∪{w0,w1},u1,v1). By Lemma 7, ∣P∣+1<119(∣C∣+3)−112(∣∣C∣∣+215). As ∣P∣≥119∣C∣−112(∣∣C∣∣−1), we have 1<1193−112217, a contradiction.
Suppose one of the wi’s, say w0, is a 3−-vertex in G−C and the other one is a 3+-vertex in G−C. Then (P∪{v0},N∪{w1},v0,u1,w0,v1) is a separating double chain. By maximality of ∣C∣, ∣P∣+1<119(∣C∣+2)−112(∣∣C∣∣+3). As ∣P∣≥119∣C∣−112(∣∣C∣∣−1), we have 1<1192−1124, a contradiction.
∎
Let us now prove some lemmas on the structure of G.
Lemma 10**.**
Graph G has no 2−-vertex.
Proof.
As G is connected, if it has a [math]-vertex, then G is the graph with one vertex and it satisfies Theorem 1, a contradiction.
By contradiction, suppose u∈V is a 1-vertex. Let v be the neighbour of u. If v is a 3−-vertex in G, then ({u},∅,u,v) is a chain of G, thus by Lemma 7, 1<119−112(1−21), a contradiction.
Now v is a 4-vertex. Let H∗=G−{u,v}. Graph H∗ has n−2 vertices and m−4 edges. Adding vertex u to any induced linear forest of H∗ leads to an induced linear forest of G. By Observation 4 applied to (α,β,γ)=(2,4,1), 1<1192−1124, a contradiction.
Therefore G has no 1−-vertex.
Suppose now that u∈V is a 2-vertex.
Let v0 and v1 be the two neighbours of u.
Suppose v0 or v1, say v1, is a 3−-vertex. We have a simple chain ({u},{v0},u,v1). By Lemma 7, 1<1192−112(4−21), a contradiction.
Now v0 and v1 are 4-vertices. Let H∗=G−{u,v0,v1}. Graph H∗ has n−3 vertices and m−8 edges. Adding vertex u to any induced linear forest of H∗ leads to an induced linear forest of G. By Observation 4 applied to (α,β,γ)=(3,8,1), 1<1193−1128, a contradiction.
∎
Lemma 11**.**
Graph G has no 3-vertex adjacent to another 3-vertex and two 4-vertices.
Proof.
By contradiction, suppose G has a 3-vertex u, adjacent to a 3-vertex v and two 4-vertices w0 and w1. We have a simple chain ({u},{w0,w1},u,v). By Lemma 7, 1<1193−112(9−21), a contradiction.
∎
Lemma 12**.**
Graph G has no 3-vertex adjacent to two other 3-vertices and a 4-vertex.
Proof.
Let u be a 3-vertex adjacent to two 3-vertices v0 and v1, and to a 4-vertex w. Let x0 and x1 be the two neighbours of v0 distinct from u. Note that x0 and x1 are 3+-vertices in G by Lemma 10, and thus 1+-vertices in G′=G−{u,w,v0} since they are not adjacent to u. Note that x0 and x1 may be adjacent to w.
Suppose that x0 and x1 are 2+-vertices in G′, or that one is a 3-vertex and the other a 1+-vertex. We have a simple chain ({u,v0},{x0,x1,w},u,v1) in G. By Lemma 7, 2<1195−112(12−21), a contradiction.
Suppose one of the xi’s, say x0, is a 2+-vertex in G′, and the other one is a 1-vertex in G′. We have a double chain ({u,v0},{w,x0},u,v0,v1,x1). By Lemma 8, 2<1194−112(10−3), a contradiction.
Now the xi’s are 1-vertices in G′. By Lemma 10, the xi’s are 3-vertices in G, and thus are both adjacent to w. By planarity of G, one of the xi’s, say x0, is not adjacent to v1. Let y be the neighbour of x0 in G′. By Lemmas 10 and 11, y is a 3-vertex in G. We have a simple chain ({u,v0,x0},{w,v1,x1},x0,y). By Lemma 7, 3<1196−112(11−21, a contradiction.
∎
Lemma 13**.**
Graph G has no two adjacent 3-vertices.
Proof.
By Lemma 10, every vertex in G has degree 3 or 4. By Lemmas 11 and 12, there is no 3-vertex adjacent to a 3-vertex and a 4-vertex in G.
Suppose by contradiction that there are two adjacent 3-vertices in G. Then as G is connected, G only has 3-vertices.
Suppose there is a 4-cycle u0u1u2u3 in G. For all i, let vi be the third neighbour of ui. Since G has no triangle, the only vertices among the ui’s and vi’s that may not be distinct are v0 and v2 on the one hand, and v1 and v3 on the other hand. Suppose v0=v2 and v1=v3. Let H∗=G−{u0,u1,u2,u3}. Graph H∗ has n−4 vertices and m−8 edges. As v0 and v1 are 1-vertices in H∗, separated by u0u1u2u3 in G, adding vertices u0 and u1 to any induced linear forest of H∗ leads to an induced linear forest of G. By Observation 4 applied to (α,β,γ)=(4,8,2), we have 2<1194−1128, a contradiction. Now w.l.o.g. v0 and v2 are distinct. We have a double chain ({u0,u1,u2},{u3,v1},u0,u2,v0,v2). By Lemma 8, 3<1195−112(9−3), a contradiction.
Now there is no 4-cycle in G. Suppose there is a 5-cycle u0u1u2u3u4 in G. For all i, let vi be the third neighbour of ui. Now all the vi’s are distinct, otherwise there is a 4-cycle and we fall into the previous case. We have a double chain ({u0,u1,u2,u3},{u4,v1,v2},u0,u3,v0,v3). By Lemma 8, 4<1197−112(14−3), a contradiction.
Now G is a 3-regular planar graph with girth at least 6, which contradicts Euler’s formula.
∎
Lemma 14**.**
There is no 4-cycle with at least two 3-vertices in G.
Proof.
By contradiction, suppose there is such a 4-cycle u0u1u2u3. By Lemmas 10 and 13, this cycle has exactly two 3-vertices and two 4-vertices, and the two 3-vertices are not adjacent. W.l.o.g. u0 and u2 are 3-vertices, and u1 and u3 are 4-vertices.
Let v0 and v2 be the third neighbours of u0 and u2 respectively. By Lemma 13, v0 and v2 are 4-vertices.
Suppose that u0 and u2 have three neighbours in common, u1, u3, and v=v0=v2. Let H∗=G−{u0,u1,u2,u3,v}. Graph H∗ has n−5 vertices and m−12 edges. Adding vertices u0 and u2 to any induced linear forest of H∗ leads to an induced linear forest of G. By Observation 4 applied to (α,β,γ)=(5,12,2), 2<1195−11212, a contradiction.
Now v0 and v2 are distinct. Suppose that v0v2∈E. We have a chain ({u0,u2},{u1,u3,v2},u0,v0). By Lemma 7, 2<1195−112(13−21), a contradiction.
Now v0v2∈/E. Let H∗=G−{u0,u1,u2,u3,v0,v2}. Graph H∗ has n−6 vertices and m−16 edges. Adding vertices u0 and u2 to any induced linear forest of H∗ leads to an induced linear forest of G. By Observation 4 applied to (α,β,γ)=(6,16,2), 2<1196−11216, a contradiction.
∎
Lemma 15**.**
There is no 4-face with exactly one 3-vertex in G.
Proof.
By contradiction, suppose there is a 4-face u0u1u2u3, such that u0 is a 3-vertex and the other ui are 4-vertices. Let v be the third neighbour of u0. Note that v is a 4-vertex by Lemma 13.
Suppose first that vu2∈E. By planarity of G, {u0,v,u2} separates the vertices u1 and u3. Therefore ({u0},{v,u2},u0,u0,u1,u3) is a separating double chain of G. By Lemma 9, 1<1193−112(9−1), a contradiction.
Now vu2∈/E. Let w0 and w1 be the neighbours of u1 distinct from u0 and u2.
Suppose w0 and w1 are adjacent to u3. Let H∗=G−{u0,u1,u2,u3,v,w0,w1}. By Lemma 10, the wi’s are 3+-vertices, and since G is triangle-free, they cannot be adjacent to u2. Moreover, by planarity, at least one of the wi’s is not adjacent to v. This implies that H∗ has at most m−15 edges. Graph H∗ has n−7 vertices. Adding vertices u0, u1, and u3 to any induced linear forest of H∗ leads to an induced linear forest of G. By Observation 4 applied to (α,β,γ)=(7,15,3), 3<1197−11215, a contradiction.
Suppose that one of the wi’s, say w0, is adjacent to u3 and that the other one (w1) is not adjacent to u3.
Let w2 be the neighbour of u3 distinct from u0, u2, and w0.
Suppose w1 or w2, say w1, is a 3-vertex in G−{u0,u1,u2,u3,v,w0}. Suppose w2 is a 3-vertex in G−{u0,u1,u2,u3,v,w0,w1} (note that this implies that w1w2∈/E). Let H∗=G−{u0,u1,u2,u3,v,w0,w1,w2}. Graph H∗ has n−8 vertices and at most m−20 edges. Adding vertices u0, u1, and u3 to any induced linear forest of H∗ leads to an induced linear forest of G. By Observation 4 applied to (α,β,γ)=(8,20,3), 3<1198−11220, a contradiction.
Now w2 is a 2−-vertex in G−{u0,u1,u2,u3,v,w0,w1}, and thus ({u0,u1,u3}, {u2,v,w0,w1}, u3, w2) is a chain of G. By Lemma 7, 3<1197−112(17−21), a contradiction.
Suppose w1 and w2 are 2−-vertex in G−{u0,u1,u2,u3,v,w0}. We have a double chain ({u0,u1,u3},{u2,v,w0},u1,u3,w1,w2). By Lemma 8, 3<1196−112(14−3), a contradiction.
Suppose the wi’s are not adjacent to u3. Let us prove by contradiction that the wi’s are 2−-vertices in G−{u0,u1,u2,v}.
Suppose the wi’s are 3-vertices in G−{u0,u1,u2,v}. Then we have the following chain: ({u0,u1},{u2,v,w0,w1},u0,u3). By Lemma 7, 2<1196−112(18−21), a contradiction.
Now one of the wi’s, say w0, is a 2−-vertex in G−{u0,u1,u2,v}. Suppose w1 is a 3-vertex in G−{u0,u1,u2,v}. Then we have the following double chain: ({u0,u1},{u2,v,w1},u0,u1,u3,w0). By Lemma 8, 2<1195−112(15−3), a contradiction.
Now the wi’s are 2−-vertices in G−{u0,u1,u2,v}.
The wi’s are 3-vertices or 4-vertices in G, they are not adjacent to u0 and u2 since G is triangle-free, and by Lemma 14, if for some i, wi is adjacent to v, then wi is a 4-vertex. Therefore each of the wi’s is either a 3-vertex non-adjacent to v or a 4-vertex adjacent to v, and thus the wi’s are 2-vertices in G−{u0,u1,u2,v}.
Let x0 and x1 be the two neighbours of w0 in G−{u0,u1,u2,u3,v}. Let d be the sum of the degrees of x0 and x1 in G−{u0,u1,u2,v,w0,w1}.
Suppose d≥4. We have a simple chain ({u0,u1,w0},{u2,v,w1,x0,x1},u0,u3). By Lemma 7, 3<1198−112(20−21), a contradiction.
Suppose d≤3.
Suppose one of the xi, say x0, is a [math]-vertex in G−{u0,u1,u2,v,w0,w1}. We have a simple chain ({u0,u1,w0,x0},{u2,v,w1,x1},u0,u3). By Lemma 7, 4<1198−112(16−21), a contradiction.
Suppose one of the xi, say x0, is a 1-vertex in G−{u0,u1,u2,v,w0,w1} and the other one (x1) is a 1-vertex or a 2-vertex in G−{u0,u1,u2,v,w0,w1}.
Let us prove by contradiction that x0 is not adjacent to u3. Suppose x0 is adjacent to u3.
Suppose x0w1∈E. By Lemma 14, at least one of the wi’s, w0 say, is a 4-vertex, and thus is adjacent to v. By planarity, w1 is not adjacent to v, and thus w1 is a 3-vertex in G. In this case, x0 is adjacent exactly to w0, w1 and u3 since it is a 1-vertex in G−{u0,u1,u2,v,w0,w1} and G is triangle-free. Then x0 and w1 are adjacent 3-vertices in G, which contradicts Lemma 13.
Now x0w1∈/E. We have a simple chain ({u0,u1,x0},{u2,u3,v,w0},u1,w1). By Lemma 7, 3<1197−112(16−21), a contradiction.
Therefore we know that x0 is not adjacent to u3. Let y be the neighbour of x0 in G−{u0,u1,u2,v,w0,w1}. Suppose y is a 3-vertex in G−{u0,u1,u2,v,w0,w1,x0,x1}. Now the following quadruplet is a simple chain: ({u0,u1,w0,x0},{u2,v,w1,x1,y},u0,u3). By Lemma 7, 4<1199−112(21−21), a contradiction. Now y is a 2−-vertex in G−{u0,u1,u2,v,w0,w1,x0,x1}. Then ({u0,u1,w0,x0},{u2,v,w1,x1}, u0,x0,u3,y) is a double chain. By Lemma 8, 4<1198−112(18−3), a contradiction.
∎
For every face f of G, let l(f) denote the length of f, and let c4(f) denote the number of 4-vertices in f. For every vertex v, let d(v) be the degree of v. Let k be the number of faces of G, and for every 3≤d≤4 and every 4≤l, let kl be the number of l-faces and nd the number of d-vertices in G.
Each 4-vertex is in the boundary of at most four faces. Therefore the sum of the c4(f) over all the 4-faces and 5-faces is ∑f,4≤l(f)≤5c4(f)≤4n4.
Now, by Lemmas 10, 14, and 15, every 4-face of G has only 4-vertices in its boundary, so for each 4-face f, c4(f)=4. By Lemma 13, every 5-face of G has at least three 4-vertices, so for each 5-face f we have c4(f)≥3≥2.
Thus ∑f,l(f)=4c4(f)+∑f,l(f)=5c4(f)≥4k4+2k5. Thus 4n4≥4k4+2k5, and thus 2n4≥2k4+k5.
By Euler’s formula, we have:
[TABLE]
That contradiction ends the proof of Theorem 1.