Triangles in Ks-saturated graphs with minimum degree t
Benjamin Cole
Department of Mathematics and Statistics,
California State University, Sacramento, [email protected]
Albert Curry
Department of Mathematics and Statistics,
California State University, Sacramento, [email protected]
David Davini
Department of Mathematics and Statistics,
University of California, Los Angeles, [email protected]
Craig Timmons
Department of Mathematics and Statistics, California State University, Sacramento.
Research is supported by a grant from the Simons Foundation #359419.
Abstract
For n≥15, we prove that the minimum number of triangles in an n-vertex
K4-saturated graph with minimum degree 4 is exactly 2n−4, and that there is a unique extremal
graph.
This is a triangle version of a result of
Alon, Erdős, Holzman, and Krivelevich from 1996.
Additionally, we show that for any s>r≥3 and t≥2(s−2)+1, there
is a Ks-saturated n-vertex graph with minimum degree t that has
(r−1s−2)2r−1n+cs,r,t copies of Kr. This shows that
unlike the number of edges, the number
of Kr’s (r>2) in a Ks-saturated graph
is not forced to grow with the minimum degree, except for possibly in lower order terms.
1 Introduction
Let F be a graph. A graph G is F-free if G does not contain F as a subgraph.
A graph G is F-saturated if G is F-free,
and adding a new edge to G creates a copy of F.
The minimum number of edges in an F-saturated graph with
n vertices is called the saturation number of F.
Write sat(n,F) for this minimum, so
[TABLE]
An n-vertex F-saturated
graph with sat(n,F) edges is called an extremal graph.
1.1 History and Previous Results
One of the most important results on graph saturation
is that for any graph F with at least one edge, there is a constant CF such that
sat(n,F)≤CFn. This was proved by Kászonyi and Tuza in 1986 [14],
and shows that saturation numbers are linear in n.
Since then, the study of saturation has become an established branch
of extremal graph theory.
Saturation numbers of hypergraphs and of random graphs have been studied
as well [4, 16, 20, 19].
The survey paper of Faudree, Faudree, and Schmitt [9] contains
many results and references.
For complete graphs, Erdős, Hajnal, and Moon [7] proved that for s≥3,
[TABLE]
Furthermore, there is a unique extremal graph which is, up to isomorphism,
Ks−2+Kn−s+2 (the join of a clique with s−2 vertices
and an independent set with n−s+2 vertices).
This graph has
minimum degree
s−2, and the minimum degree of a Ks-saturated graph is at least s−2 (since nonadjacent vertices must have
a Ks−2 in their common neighborhood).
A natural question
is to ask for the minimum number of edges in an F-saturated graph G
with δ(G)=t where t>s−2.
Given a graph F and
an integer t, let
[TABLE]
Duffus and Hanson [6] proved that sat2(n,K3)=2n−5 for n≥5, and
characterized the extremal graphs. They also showed that
sat3(n,K3)=3n−15 for n≥10.
For larger t and s, the bounds are not exact.
In 2014, Day [5], resolving a conjecture of Bollobás [13] from 1996,
showed that
[TABLE]
for any s≥3, t≥s−2. Here ct is a constant depending only on t.
Further discussion on saturated graphs with degree constraints can be found
in [9].
In 2014, Alon and Shikhelman [2]
introduced a very important generalization of Turán numbers which has since
been extensively studied ([8, 10, 12, 17] to name a few).
It is connected to the widely studied Turán problem
for Berge hypergraphs ([11, 18], for instance). Motivated by this generalization,
Kritschgau et. al. [15] defined an analogous generalization of saturation numbers.
For graphs H and F, let
sat(n,H,F)
be the minimum number of copies of H in an F-saturated graph
with n vertices. Observe that sat(n,K2,F)=sat(n,F).
In this paper, we introduce the function
[TABLE]
which is defined to be the minimum number of copies of H in an n-vertex
F-saturated graph with minimum degree t.
This function generalizes both satt(n,F) and sat(n,H,F).
The question we put forth is the following.
Question 1.1
Let H and F be graphs. What is the minimum number
of copies of H in an F-saturated n-vertex graph with minimum degree
t?
Before stating our results, let us recall a result from [15]
which we take as a starting point for our work.
In [15], the formula
[TABLE]
was proved, and it was shown that K2+Kn−2 is the unique extremal graph.
Bounds on sat(n,Kr,Ks) for
all s>r≥3 were also proved in [15]. However,
an asymptotic formula for sat(n,Kr,Ks) is known only
in the case when (r,s)=(3,4).
Motivated by the fact that (2) gives a formula for
sat(n,K3,K4),
we will study the function satt(n,K3,K4) in detail.
Our aim is to determine whether or not a statement similar to (1) holds
when counting K3’s in a K4-saturated graph with minimum degree at least t. First
let us state some results that answer Question 1.1
in certain cases. The first proposition we give is easy to prove using the fact
that in a Ks-saturated graph, any two nonadjacent vertices must have
a Ks−2 in their common neighborhood.
Proposition 1.2
Let s>r≥2 be integers. If G is an n-vertex Ks-saturated graph with
δ(G)=s−2, then G is isomorphic to Ks−2+Kn−s+2 and consequently, has
[TABLE]
copies of Kr.
In the special case (r,s)=(3,4) and δ(G)=2, we have exactly n−2 triangles.
A close look at the proof of (2) in [15] shows that
if G is not isomorphic to K2+Kn−2, then G has at
least n triangles. Now if G is K4-saturated and is not isomorphic to
K2+Kn−2, then one can prove that G
has minimum degree at least 3.
Thus, we see a small jump in the number of triangles when G is no longer allowed
to have a vertex of degree 2. It turns out that when
the minimum degree is 3, we do not just get 2 additional triangles, but we must get
and additional n−5 triangles.
Before formalizing this as a proposition, we need to introduce a new graph.
Let a1a2a3a4a5a1 be a cycle of length
five and let b be a vertex joined to each ai. Call this graph W.
Let m1,m3,m4 be positive integers
with m1+m3+m4=n−s+1. In W, replace b with a clique of size s−3, and for
i∈{1,3,4}, replace ai with an independent set of size mi. Vertices in this new graph
are adjacent if and only if the vertices they replaced are adjacent vertices in W.
Write Ws(m1,1,m3,m4,1) for this graph, which has appeared in the literature (see [3]).
Proposition 1.3
Let s≥3. If G is an n-vertex Ks-saturated
graph with δ(G)=s−1, then G is isomorphic to either (Ks−1−e)+Kn−s+1,
or Ws(m1,1,m3,m4,1) for some m1,m3,m4 with m1+m3+m4=n−s+1.
It follows from Proposition 1.3 that if G is an n-vertex K4-saturated graph
with minimum degree 3, then G has at least 2n−7 triangles, and equality holds
only if G is isomorphic to W4(m1,1,m3,1,1) for some
m1+m3=n−4. For a proof of Proposition 1.3, see [15].
1.2 New Results
In light of Day’s Theorem (1) and that
∣V(G)∣=n, G is K4-saturated, and δ(G)=2 ⇒ G has at least n−2 triangles,
∣V(G)∣=n, G is K4-saturated, and δ(G)=3 ⇒ G has at least 2n−7 triangles,
one may be tempted to conjecture that in general, δ(G)=t forces at least (t−1)n−O(1) triangles in
any n-vertex K4-saturated graph. This would then give a version of
Day’s result for triangles.
It turns out, rather surprisingly, that
for any t≥4 and n≥t+5, there is an n-vertex K4-saturated graph
that has minimum degree t and only 2n+2t−12 triangles.
We call this graph Ht(n) and it is defined in Section 2.
Our main theorem determines satt(n,K3,K4), and shows Ht(n) is the unique extremal graph.
Theorem 1.4
Let n≥14 and G be an n-vertex K4-saturated graph
with δ(G)=4.
Then G contains at least 2n−4 triangles.
Furthermore, if G contains exactly 2n−4 triangles,
then G is isomorphic to Ht(n).
For t≥4, the graph Ht(n) implies the following upper bound on satt(n,K3,K4).
Theorem 1.5
For integers t≥4 and n≥2t,
[TABLE]
We conjecture that the upper bound in Theorem 1.5 is best possible, and that
Ht(n) is the unique extremal graph for all sufficiently large n.
Conjecture 1.6
For any integer t≥4, there is an integer nt such that for all n≥nt,
[TABLE]
and Ht(n) is the unique extremal graph.
Theorem 1.4
shows that Conjecture 1.6 is correct when t=4.
The next theorem gives an upper bound for arbitrary s>r≥3.
Theorem 1.7
Let s>r≥3 and t≥2(s−2)+1 be integers. For n≥2(s−2)+2t,
[TABLE]
where Cs,r,t is a constant depending only on r, s, and t.
It would be interesting to determine if there is a constant cr,s, depending only on r and s, such that
for all n≥n(r,s,t) and t≥s−2, there is an n-vertex Ks-saturated graph with minimum degree t
having at most cr,sn+o(n) copies of Kr. Theorem 1.5 shows that
such a constant exists in the case when (r,s)=(3,4).
Theorem 1.7 covers all s>r≥3, but assumes t≥2(s−2)+1, and perhaps
the coefficient of n could be improved (as in Theorem 1.5
when (r,s)=(3,4)). We can improve this upper bound in the case (r,s)=(3,5) as the
following result shows.
Theorem 1.8
For any t≥8 and n≥t+30,
[TABLE]
where Ct is a constant dependent only on t.
Finally, a simple argument in the case r=3 gives a lower bound on satt(n,K3,Ks).
Proposition 1.9
Let s>3 and t≥6(2s−2) be integers. For
n≥2s−2,
[TABLE]
Proposition 1.9 implies satt(n,K3,K5)≥3(n−2) for t≥18. We suspect
the upper bound of Theorem 1.8 could give the correct coefficient of n.
In Section 2, we define the graphs that prove
Theorems 1.5, 1.7,
and Proposition 1.8.
In Section 3, we prove Theorem 1.4.
In Section 4, we prove Proposition 1.9.
2 Proof of Theorems 1.5,
1.7, and 1.8
A natural question one might ask is whether or not every edge e in a Ks-saturated graph
is in a triangle. It turns out that this is not necessarily true.
In fact, Ks-saturated graphs that have an edge in no triangle
contain certain subgraphs–what we call Ks-support structures–which pave the way for constructions proving Theorems 1.5 and 1.7. For this and the following sections, we will use kr(A) to denote the number of copies of Kr in A.
Definition 2.1
Let S(A,B) be a graph with vertex set A∪B, where A and B are disjoint sets. We call S(A,B) a pre-Ks-support-structure provided:
A* and B are both Ks−1-free,*
N(a)∩B* has a copy of Ks−2 for every vertex a∈A,*
N(b)∩A* has a copy of Ks−2 for every vertex b∈B, and*
S(A,B)* itself is Ks-free.*
If S(A,B) has the added condition that
the subgraph induced by A and the subgraph induced by B are Ks−1-saturated,
and adding any missing edge between A and B would create a Ks, then we call S(A,B) a Ks-support-structure.
Lemma 2.2
Given a pre-Ks-support-structure S(A,B), there is a Ks-support-structure S′(A,B) where S′(A,B) is a supergraph of S(A,B).
Proof. We add edges one at a time to S(A,B) arbitrarily, adding each missing edge e only if it does not create a Ks−1 in A, a Ks−1 in B, or a Ks in S(A,B).
We continue this process until there are no longer any edges to add. Certainly after at most (2∣A∪B∣) steps this process will end. Notice that with each edge added, the new graph is still a pre-Ks-support-structure, and in fact the final graph S′(A,B) is a Ks-support-structure.
Lemma 2.3
Let S(A,B) be a Ks-support-structure, and set N=t−min{∣B∣,δS(A,B)(A)} and M=t−min{∣A∣,δS(A,B)(B)}. Given integers
[TABLE]
there exists a Ks-saturated supergraph G of S(A,B) with order n and minimum degree t containing
nkr−1(A)+C∣A∣,∣B∣,t copies of Kr.
Proof. Define G as the graph obtained from S(A,B) by
taking two independent sets X={x1,x2,…,xn−∣A∣−∣B∣−M}
and Y={y1,y2,…,yM}, and joining X to A∪Y, and joining
Y to B∪X. Note that since S(A,B) is a Ks-support-structure, G is Ks-saturated. Next we show that the minimum degree of G is t. We have
[TABLE]
By symmetry, δG(X∪B)≥t.
Putting these two inequalities together, we see that δ(G)=min{δG(Y∪A),δG(X∪B)}=t.
We now count the number of Kr’s in G. Note that by G’s construction, the number of vertices not in X is dependent
only on ∣A∣, ∣B∣ and t. Thus, the number of copies of Kr that do not use a vertex in X is also dependent
only on these parameters.
Suppose K is a Kr that does contain a vertex in X. Then K can only contain one vertex in X, since X is an independent set. Also, K cannot contain any vertex in Y because r≥3 and Y is an independent set. Thus, all the vertices in K that are not in X must be in A. We conclude that the number of copies of Kr containing a vertex in X is
[TABLE]
We now define three graphs by defining pre-Ks-support structures, and then
applying Lemmas 2.2 and 2.3.
The graph Ht(n): Let t≥4 and n>2t be integers. Let A={a1,a2,a3,a4} and B={b123,b124,b134,b234}, and construct the graph S(A,B) with vertex set A∪B, and the following edges:
a1a2 and a3a4,
b123b234, b234b124, b124b134, and b134b123,
arbijk if and only if r∈{i,j,k}.
It is easy to verify that S(A,B) is a Ks-support-structure. Employing Lemmas 2.2 and 2.3,
we obtain a K4-saturated graph, which we call Ht(n), that has n-vertices,
minimum degree t, and containing 2n+2t−12 triangles.
Thus, Ht(n) proves Theorem 1.5.
Remark: The graph H4(n) appears briefly in [1], but the focus in that paper is on minimizing the number of edges in a K4-saturated graph with minimum degree 4. Up to isomorphism, there is a unique graph that minimizes the number of edges in a K4-saturated graph with minimum degree 4, but this graph does not minimize the number of triangles.
Next we construct the graph that proves Theorem 1.7.
The graph Fs,t(n): Let s>r≥3, t≥2(s−2)+1, and n≥2(s−2)+2t. Let A={a1,a2,…,a2(s−2)} and B={b1,b2,…,b2(s−2)}, and construct the graph S(A,B) with vertex
set A∪B, and the following edges:
aiaj if and only if i≡j+(s−2)(mod2(s−2)),
bibj if and only if i≡j+(s−2)(mod2(s−2)),
aibi(mod2(s−2)), aibi+1(mod2(s−2)), …, and aibi+(s−2)−1(mod2(s−2)).
Notice that the vertices in A and the vertices in B form two complete (s−2)-partite graphs with
2 vertices in each part.
We now show that S(A,B) is a pre-Ks-support-structure.
It is easy to verify that A and B are Ks−1-free, and that each a∈A and b∈B has a copy of Ks−2 in N(a)∩B and N(a)∩B, respectively.
Next we prove that S(A,B) is Ks-free.
It is enough to show that a1 does not lie in a Ks since S(A,B) is vertex transitive.
The neighborhood of a1 is the union of the two sets
A1={a2,a3,…,a2(s−2)}\{as−1}
and
B1={b1,b2,…,bs−2}.
Let H1 be the subgraph of S(A,B) induced by A1∪B1.
We will show H1 is Ks−1-free.
It
will be convenient to relabel the vertices in A1 as
as=1, as+1=2, as+2=3,…,a2(s−2)=s−3,
and
a2=s−2, a3=s−1,a4=s,…,as−2=2s−6.
With this relabeling, in the graph H1 the vertex bi (1≤i≤s−2) is adjacent to all vertices in
[TABLE]
Now suppose K is a clique in H1. Let bi1,bi2,…,biβ be the vertices in K
that are in B1, where 1≤i1<i2<⋯<iβ≤s−2.
By (3), if a∈{1,2,3,…,2s−6} is a vertex in A1∩K, then
[TABLE]
for ℓ=1,2,…,β. In particular, iβ≤a≤i1+s−4.
Adding −iβ+1 throughout leads to
[TABLE]
Therefore, there are at most s−β−2 vertices in A1 that are in K, so K contains at most
[TABLE]
vertices. We conclude that the neighborhood of a1 is Ks−1-free which implies S(A,B) is Ks-free.
Adding edges to S(A,B) using the process described in Lemma 2.2, we
obtain a Ks-support-structure S′(A,B). Notice that since A and B were already Kr−1-saturated, the only edges added were between A and B, so the subgraphs induced by A and B have not changed. Applying Lemma 2.3 to S′(A,B), we obtain a Ks-saturated graph Fs,t(n) with n vertices, minimum degree t, and
containing at most nkr−1(A)+Cr,s,t=(r−1s−2)2r−1n+Cr,s,t copies of Kr.
A computation, using the assumption that n≥2(s−2)+2t, shows that all vertices in Fs,t(n) have
degree at least t (vertices in X will have degree t). The graph Fs,t(n) proves Theorem 1.7.
The graph Rt(n): Let t>9, and n>2t+12. Let A be the disjoint union of the three K3’s
[TABLE]
The subscripts appearing in a given triangle form a class of parallel lines in an affine plane of order 3.
Next, let B be the disjoint union of triangles B1 through B9, where each Bi is given by
[TABLE]
Construct the graph S(A,B) by connecting each vertex aijk∈As to all vertices in the union Bi∪Bj∪Bk∪{bt,s:1≤t≤9}.
We now show that S(A,B) is a pre-K5-support structure. It is easy to verify that A and B are K4-free, and that each a∈A and b∈B has a copy of K3 in N(a)∩B and N(a)∩B, respectively. Also, since A and B are K4-free, a possible K5 in S(A,B) must have at least two vertices in A, say ai1j1k1,ai2j2k2∈As. However, by S(A,B)’s construction, their common neighborhood in B contains at most 1 vertex. Hence S(A,B) is K5-free.
Adding edges to S(A,B) using the process described in Lemma 2.2, we
obtain a Ks-support-structure S′(A,B). Importantly, since by S(A,B)’s construction every pair of non-adjacent vertices in A must have a K3 in their common neighborhood in B, no edges were added to A, so the subgraph induced by A has not changed. Applying Lemma 2.3 to S′(A,B), we obtain a K5-saturated graph Rt(n) with n vertices, minimum degree t, and
containing at most nk3(A)+Ct=9n+Ct copies of K3.
Remark: The construction of Rt(n)
can be generalized to create a Ks-support-structure Rs,t(n)
for any s≥3, but for brevity’s sake we have chosen to omit this. Notice that for s=5, R5,t(n)’s 9n+Ct copies of K3 is less than
F5,t(n)’s 12n+Ct copies of K3. This turns out to be an exception;
for s=6, R6,t(n) and F6,t(n) both have 24n+Ct copies of K3, and for s≥7, R7,t(n) contains more copies of K3 than F7,t(n). In general, Rs,t(n) will
have fewer copies of Kr than Fs,t(n) for s∈{4,5,…,2r−1+1}, but when s>2r−1+2, Rs,t(n) contains more copies of Kr than Fs,t(n). This suggests that a better construction may exist.
3 Proof of Theorem 1.4
In this section, we prove Theorem 1.4. The proof will be broken up into several
lemmas.
We write k3(G)
for the number of triangles in G. For x∈V(G), N(x) is the neighborhood of x.
Let us set up some specialized notation that will be used in the rest of this section.
This same notation was used in [1].
Setup and Notation:
For the remainder of Section 3,
let G be an n-vertex graph that is
K4-saturated, and has
δ(G)=4.
Let x∈V(G) be a vertex of degree 4, and let
[TABLE]
Let Y=V(G)\N[x].
If y∈Y, then N(y)∩N(x) must contain an edge so
that every vertex in Y is adjacent to
at least two vertices in N(x).
For S⊆{1,2,3,4}, let VS be the vertices
y∈Y such that y is a neighbor of xi if and only if i∈S.
For notational convenience, we will omit braces and commas so if, say
S={1,3,4}, then we write V134 rather than
V{1,3,4}.
Given sets S,T⊆{1,2,3,4}, write
[TABLE]
if all vertices in VS are adjacent to all vertices in VT,
and VS≁VT if no vertex in VS is adjacent to a vertex
in VT.
An important observation is that if VS=∅, then there must be a pair {i,j}⊆S such
that xi is adjacent to xj. Also,
if {xi:i∈S} forms an independent set in N(x), then VS=∅.
Because of this, we ignore all VS
for which {xi:i∈S} is an independent set in N(x).
Given a subset S⊂{1,2,3,4} with i∈/S for some i∈{1,2,3,4},
let TS,i be the set
of all T⊆{1,2,3,4} for which i∈T, and for any j,k∈S for which xj and xk are adjacent
vertices in N(x), we have ∣T∩{j,k}∣≤1.
Lemma 3.1** (Rules Lemma)**
Let S⊂{1,2,3,4} with i∈/S for some i∈{1,2,3,4}.
If VS=∅, then y is adjacent to some vertex in
[TABLE]
and in particular, this union is not empty.
Proof. Suppose y∈VS. Since i∈/S, the vertex y is not adjacent to xi where xi∈N(x).
There must be an edge αβ with both α and β in the intersection N(y)∩N(xi).
If α,β∈N(x), then {x,xi,α,β} is a K4 which is a contradiction.
Thus, we may assume that α∈Y, say α∈VT. Because α is adjacent to
xi, we have i∈T. If there is a j,k∈S for which xjxk is an edge in N(x)
and j,k∈T, then {y,α,xj,xk} is a K4.
We conclude that T must be some member of TS,i.
It will be convenient to represent Lemma 3.1 with an arrow diagram and
we refer to these as Rules. Assuming the set up and conclusion of Lemma 3.1 where
TS,i={T1,T2,…,Tℓ}, then we illustrate
using the figure below.
V_{S}$$V_{T_{\ell}}$$V_{T_{2}}$$V_{T_{1}}$$\vdots
Rule for VS and choice of i∈/S
We also use the notation VS→VT1∪VT2∪⋯∪VTℓ.
Note that the assertion of Lemma 3.1 is not that Tj=∅
for every Tj∈TS,i, but
only that at least one of these sets is not empty.
The first application of Lemma 3.1 will be
in the proof of Lemma 3.4.
First we prove two easier lemmas.
Lemma 3.2
The number of edges in N(x) is at least 2.
Proof of Lemma 3.2.
See Lemma 3.2 of [15] (the proof is not difficult).
Lemma 3.3
If n≥14 and the number of edges in N(x) is 2, then
[TABLE]
Furthermore, equality holds if and only if G is isomorphic to H4(n).
Proof of Lemma 3.3.
There are two possibilities for the edges in N(x).
Case 1: The two edges in N(x) form a path.
Suppose the edges in N(x) are x1x2 and x2x3.
Every vertex in Y must be adjacent to x2 since N(x)∩N(y) must contain
an edge for all y∈Y, and every edge in N(x) has x2 as an endpoint.
Therefore, x2 is adjacent to all vertices in G except for x4.
The intersection N(x3)∩N(x4) must contain an edge,
say y1y2. The edge y1y2 must have both endpoints
in Y since x is the only neighbor of x4 in N[x],
but then {y1,y2,x2,x3} is a
K4 in G (y1 and y2 are both adjacent to x2).
This is a contradiction.
Case 2: The two edges in N(x) share no endpoints.
Assume the edges in N(x) are x1x2 and x3x4.
Then V12 and V34 must be empty (see [1]).
Since N(y)∩N(x) must contain an edge for each y∈Y and
V12=V34=∅,
we can partition Y as follows:
[TABLE]
Note that every vertex in Y is adjacent to either both x1 and x2, or to both x3 and x4.
Suppose y∈V1234. We claim that the degree of y is exactly 4.
If y is adjacent to some
z∈Y\{y}, then we may assume, without loss of generality,
that z is adjacent to both x1 and x2. However, this implies
{y,z,x1,x2} is a K4 in G which is a contradiction.
Consequently, d(y)=4 whenever y∈V1234, and y lies in
exactly two triangles: yx1x2 and yx3x4.
We now focus on
[TABLE]
The intersection N(x1)∩N(x3)
must contain an edge αβ. This edge
has both endpoints in Y
since the only common neighbor of x1 and x3 in N[x]
is x.
Because α and β are adjacent vertices in Y,
neither can be in V1234. Thus,
α,β∈V123∪V134.
If α and β are both in V123, then {α,β,x1,x2} is a K4.
A similar argument shows α and β cannot both
be in V134. We may assume that α∈V123
and β∈V134, which shows V123=∅ and V134=∅.
By symmetry, V124 and V234 are not empty.
It is proved in [1] that G[Y′] is a complete bipartite graph with parts
V123∪V124 and V134∪V234.
This determines the graph G up to the number of vertices in the parts
V123, V124, V134, and V234 (which are all not empty),
and V1234 (which may be empty).
Let y123∈V123, and define y124, y134, and y234 similarly.
The subgraph G1′ of G induced by N[x]∪{y123,y124,y134,y234} has
9 vertices and 14 triangles; it is exactly one of the graphs appearing in the proof of Theorem 8 in [1]. These triangles are
xx1x2, xx3x4, x1x2y123, x1x2y124, x3x4y134, x3x4y234
y123y134x1, y123y134x3,
y134y124x1, y134y124x4,
y124y234x2, y124y234x4,
y123y234x2, y123y234x3.
We now estimate the number of triangles in G by adding back the remaining n−9 vertices in V(G)\V(G1′)
one by one. At each step, we count the number of triangles that contain the new added vertex
and vertices in V(G1′).
By counting triangles in this way, we never count the same triangle more than once.
If a vertex y is added to V1234, then we obtain exactly two new triangles: yx1x2 and
yx3x4. We can add any number of vertices to V1234 and the resulting graph
is K4-saturated.
Now suppose a vertex y is added to V123 (the other cases are the same by symmetry).
This creates at least five new triangles: yx1x2, yy134x1, yy134x3,
yy234x2, and yy234x3.
The conclusion is that
[TABLE]
We have k3(G)=2n−4
if and only if all of the n−9 vertices V(G)\V(G1′) are contained in V1234.
This is precisely the graph H4(n).
Method and Notation For Triangle Counting: The counting method used in the last three paragraphs of the proof of Lemma
3.3 will be used multiple times in the proof of Lemmas 3.4 and 3.5.
We find a small subgraph Gi′ of G, count the number of triangles in Gi′, and then count
the number of triangles created when a vertex y is added to Gi′. The crucial point is that
when a vertex y is added, we are only counting triangles that contain y and vertices in Gi′.
This means that we never count the same triangle more than once.
In an effort to be concise, we will always write y for the added vertex, and then list the triangles in the following way.
If y is added to VS and this creates triangles yα1β1, yα2β2,…,yαkβk,
we will write
[TABLE]
We preface this with the statement “When y is added to VS,” and then list the
triangles containing y using the notation shown above.
Lemma 3.4
Suppose n≥12.
If the number of edges in N(x) is 3, then
[TABLE]
Proof of Lemma 3.4.
Since N(x) must be triangle free, the three edges in N(x) either form a star or a path of length three.
Suppose first the edges in N(x) are x1x2, x1x3, and x1x4.
Since every edge in N(x) has x1 as an endpoint and every vertex y∈Y must
be joined to an edge in N(x), vertex x1 is adjacent to all vertices in G.
Let H be the subgraph of G induced by N(x1).
Then H is a K3-saturated (n−1)-vertex graph with δ(H)=3.
Since n≥11,
a result of Duffus and Hanson [6] implies that H has
at least 3(n−1)−15=3n−18 edges. Each of these edges forms a triangle with x and
so k3(G)≥3n−18.
Assume now the edges in N(x) are x1x2, x2x3, and x3x4.
First we show that V12 and V34 are empty.
We prove this in full, but then
we will make use of the results proved in [1] concerning adjacencies
between a VS and a VT.
If
y∈V12, then y is not adjacent to x4, and so there
must be an edge αβ in the intersection N(y)∩N(x4).
Since y and x4 have no common neighbors in N[x],
the endpoints of αβ lie in Y. If α or β is
adjacent to both x1 and x2,
then either {x1,x2,y,α} or
{x1,x2,y,β} is a K4.
The sets N(α)∩N(x) and N(β)∩N(x)
must contain at least one of the edges in N(x), other than x1x2.
All of the edges in N(x), other than x1x2, have x3 as an endpoint
and so α and β are both adjacent to x3, but then
{α,β,x3,x4} is a K4. The conclusion is that
V12=∅. By symmetry, V34=∅.
As in the proof of Lemma 3.3,
if y∈V1234, then d(y)=4 and N(y)={x1,x2,x3,x4}.
Let Y′=Y\V1234. Because V12=V34=∅, we have
[TABLE]
If y and z are
adjacent vertices with y,z∈V23∪V123∪V234,
then {y,z,x2,x3} is a K4. Thus,
[TABLE]
Similar arguments show that
[TABLE]
Summarizing, VS∼VT if the intersection S∩T does not
contain one of the pairs {1,2}, {2,3}, or {3,4}. Also,
VS≁VT if S∩T contains at least
one of the pairs {1,2}, {2,3}, or {3,4}.
We represent this using a graph.
A solid edge between
VS and VT indicates VS∼VT, a dashed edge indicates VS≁VT,
and no edge between VS and VT indicates that it is possible for a vertex in VS to be
adjacent or not adjacent to vertex in VT (this last possibility does not occur
here, but will occur in the proof of Lemma 3.5).
V_{123}$$V_{124}$$V_{134}$$V_{234}$$V_{23}
Adjacencies among the VS and VT
From Lemma 3.1, we have that two Rules for this graph are
V_{123}$$V_{124}$$V_{134}$$V_{234}$$V_{23}$$V_{123}$$V_{124}$$V_{134}$$V_{234}$$V_{23}
Rule 1 (on the left) and Rule 2 (on the right)
By symmetry (which is highlighted by the way we have chosen to show Rules 1 and 2),
we have V234→V124 and V134→V23∪V123∪V124.
We also refer to the assertion V234→V124 as Rule 1, and
the assertion V134→V23∪V123∪V124 as Rule 2.
We take a moment to carefully show how Lemma 3.1 gives Rule 2.
Choosing S={1,2,4} and
i=3, we find all T for which 3∈T, and ∣T∩{1,2}∣≤1.
These two conditions fail for {1,2,3}, but hold for {2,3}, {1,3,4}, and {2,3,4}.
Thus, V124→V23∪V134∪V234.
The structure of G is now determined up to the sizes of the parts
V1234, V23, V123, V124, V134, and V234.
If all of the vertices in Y belong to V1234, then
k3(G)≥3+3(n−5)=3n−12
and we are done. Let Y′:=Y\V1234 and assume Y′=∅.
Case 1: V23=∅.
Let y23∈V23. Since δ(G)≥4, y23 has at least two other neighbors
aside from x2 and x3. Let z1, z2 be two other neighbors of y23. These two neighbors
must be in the union V124∪V134.
First suppose z1:=z124∈V124 and z2:=z134∈V134.
Let G2′ be the subgraph of G induced by N[x]∪{y23,z124,z134}. This subgraph
has 8 vertices and 11 triangles (see the Appendix for the list of triangles).
When y is added to VS,
V23 : yx2x3,yz124x2,yz134x3, yz124z134
V123 : yx1x2,yx2x3,yx1z134,yx3z134 (V234 is the same by symmetry)
V124 : yx1x2,yx2y23,yz134y23 (V134 is the same by symmetry)
The conclusion is that
[TABLE]
Now suppose that z1:=z124 and z2:=z124′ are in V124, and
that V134=∅.
If V123=∅, then by Rule 2, V123→V134 implying V134=∅,
a contradiction. Therefore, V123=∅.
We consider two subcases.
Subcase 1: V234=∅.
Let z234∈V234 and G3′ be the subgraph of G induced by
N[x]∪{y23,z124,z124′,z234}.
This subgraph has 9 vertices and 14 triangles.
When y is added to VS,
V23 : yx2x3,yz124x2,yz124′x2
V124 : yx1x2,yy23x2,yz234x2,yz234x4
V234 : yx2x3,yx3x4,yz124x2,yz124′x2
We conclude that k3(G)≥k3(G3′)+3(n−9)=14+3(n−9)=3n−13.
Subcase 2: V234=∅.
The neighborhood x1x4 must contain an edge, say αβ. This edge cannot
use the vertex x since x1 is not adjacent to x3, and x4 is not adjacent to x2.
Furthermore, neither endpoint is in V23 because x1 is not adjacent to any vertex in
V23. We conclude that the endpoints of α and β must both be in
V124, but then {α,β,x1,x2} is a K4.
Case 2: V23=∅.
Here we will consider two subcases.
Subcase 1: V123=∅
Let y123∈V123. By Rule 1,
V123→V134 so there is a y134∈V134,
and y123 is adjacent to y134.
If V234=∅, say y234∈V234,
then by Rule 1, V234→V124. Let y124∈V124
be a neighbor of y234.
Let
G4′ be the subgraph of G induced by
N[x]∪{y123,y134,y234,y124}.
This graph has 9 vertices and 15 triangles. When y is added to VS,
V123 : yx1x2,yx2x3,yy134x1,yy134x3 (V234 is the same by symmetry)
V134 : yx3x4,yy123x1,yy123x3,yy124x4 (V124 is the same by symmetry)
Thus,
[TABLE]
Now suppose V234=∅.
Let G5′ be the subgraph induced by N[x]∪{y123,y134}. The graph G5′ has 7 vertices and 8 triangles.
When y is added to VS,
V123 : yx1x2,yx2x3,yy134x1,yy134x3
V134 : yx3x4,yy123x1,yy123x3
V124 : yx1x2,yy134x1,yy134x4
Thus, k3(G)≥k3(G′)+3(n−7)=8+3(n−7)=3n−13.
Subcase 2: V123=∅.
By symmetry, we may assume V234=∅ (otherwise we are
back in Subcase 1 with V234 replacing V123). Then all of the vertices of G not in
N[x] must be in V124∪V134∪V1234. Let y124∈V124.
By Rule 1, V124→V23∪V234∪V134, but
V23=V234=∅. Thus, there is a y134∈V134. If we take
G6′ to be the subgraph of G induced by N[x]∪{y124,y134}, then
G6′ has 6 vertices and 7 triangles.
When y is added to VS,
V124 : yx1x2,yy134x1,yy134x4 (V134 is the same by symmetry)
Therefore, k3(G)≥3(n−7)+7=3n−14.
Lemma 3.5
If n≥15 and the number of edges in N(x) is 4, then
[TABLE]
Proof of Lemma 3.5.
Since G is K4-free, N(x) must be triangle-free and so the four edges in N(x) form
a C4. Assume x1x2, x2x3, x3x4, and x4x1 are the four
edges in N(x).
The set
Y can be partitioned into the disjoint union
[TABLE]
The first step is to deal with vertices in V1234. If y∈V1234, then
y cannot have a neighbor in Y, otherwise we obtain a K4. Thus, N(y)={x1,x2,x3,x4}
for all y∈V1234. Such a vertex lies in four triangles: yx1x2, yx2x3, yx3x4,
and yx4x1.
We will therefore assume that V1234=∅ (if G′ is the subgraph of G obtained by
removing the vertices in Y1234 and we can prove the
result for G′, then the result easily follows for G).
For the rest of the proof, we focus on
[TABLE]
As stated in [1], the following relationships hold among these 8 sets. Recall a solid/dashed edge
indicates that all vertices of VS are adjacent/not adjacent to all vertices of VT.
V_{412}$$V_{123}$$V_{234}$$V_{341}$$V_{41}$$V_{12}$$V_{23}$$V_{34}
The only missing edges in this figure are those with endpoints in {V12,V23,V34,V41}. At this
stage, we do not have enough information to determine adjacencies between these sets.
Next, we apply Lemma 3.1 to obtain Rules 3 and 4.
V_{412}$$V_{41}$$V_{12}$$V_{341}$$V_{123}$$V_{34}$$V_{23}$$V_{234}$$V_{412}$$V_{41}$$V_{12}$$V_{341}$$V_{123}$$V_{34}$$V_{23}$$V_{234}
Rule 3 (on the left) and Rule 4 (on the right)
Again, the pictures are drawn to highlight the symmetry, and we use Rule 3 and Rule 4
accordingly (for example, the assertion V412→V23∪V234∪V34 is also
called Rule 4). Rule 3 is obtained by applying Lemma 3.1 with S={1,2} and
i=3, which gives V12→V23∪V234, and then
again with S={1,2} with i=4, which gives V12→V41∪V341.
The proof of Lemma 3.5 from this point forward will be divided into cases.
Case 1: V12∪V23∪V34∪V14=∅.
Since G has more than 5 vertices, we can assume there is some y123∈V123.
By Rule 4, V123→V41∪V341∪V34,
but V41=V34=∅, so y123 has a neighbor y134∈V134.
Hence, we know that V123=∅ and V134=∅.
Subcase 1: V234∪V124=∅.
Let G7′ be the subgraph of G induced by N[x]∪{y123,y134}. Then G7′ has
7 vertices and 10 triangles. When y is added to VS,
V123 : yx1x2,yx2x3,yy134x1,yy134x3 (V341 is the same by symmetry)
Thus, k3(G)≥10+4(n−7)=4n−18≥2n−2.
Subcase 2: All of V123, V134, V234, and V124 are not empty.
Before dealing with Subcase 2, we note that Subcases 1 and 2 do cover all possibilities.
This is because if V234=∅ (or V124=∅), then Rule 4 and
the fact that V12∪V23∪V34∪V41=∅ implies
V124=∅ (or V234=∅). Therefore,
it cannot be the case that exactly one of V123, V134, V234, V124 is empty.
Assume y234∈V234 and y412∈V412.
Let G8′ be the subgraph of G induced by
N[x]∪{y123,y234,y134,y412}.
This graph has 9 vertices and 16 triangles. When y is added to VS,
V123 : yx1x2,yx2x3,yy134x1,yy134x3 (V234, V134, V412 are the
same)
Thus, k3(G)≥16+4(n−9)=4n−20≥2n−2.
Case 2: V12∪V23∪V34∪V14=∅.
For Case 2 we need an additional rule which is proved in [1].
Rule 5: If F is the subgraph of G induced by V12∪V23∪V34∪V14,
then F is a 4-partite graph that is K4-saturated with respect to the parts.
In particular, Rule 5 implies that if one of the four parts V12,V23,V34,V14 is empty,
then the remaining parts induce a complete
1-partite, 2-partite, or 3-partite graph depending on the number of nonempty parts.
Subcase 1: V12=∅ and V23=V34=V14=∅.
Let y12∈V12. By Rule 3,
V12→V41∪V341 and
V12→V23∪V234, but
V41=V23=∅ so
there must be vertices y234∈V234 and
y134∈V134 that are both adjacent to y12.
Let G9′ be the subgraph of G induced
by N[x]∪{y12,y234,y134}.
Then G′ has 11 triangles and 8 vertices.
When y is added to VS,
V12 : yx1x2,yy234x2,yy134x1
V123 : yx1x2,yx2x3,yy134x1 (V412 is the same)
V234 : yx2x3,yx3x4,yy12x2 (V341 is the same)
We conclude that
[TABLE]
where we have used the assumption n≥15 for the last inequality.
Subcase 2: V12=∅ and V23=∅, V34=V14=∅.
Let y12∈V12 and y23∈V23.
By Rule 5, any vertex in V12 will be adjacent to all vertices in V23.
By Rule 3, V12→V41∪V341 so, since V41=∅,
there is a vertex y134∈V134. This vertex must be adjacent to both y12 and y23.
Let G10′ be the subgraph of G induced by N[x]∪{y12,y23,y134}.
Then G10′ has 8 vertices and 12 triangles. When y is added to VS,
V12 : yx1x2 yy23x2, yy134x1 (V23 is the same),
V123: yx1x2, yx2x3, yy134x1,yy134x4
V134 :yx1x4, yx3x4, yy12x1,
V234: yx2x3, yx3x4, yy12x2 (V412 is the same).
Therefore,
k3(G)≥12+3(n−8)=3n−12≥2n−2.
Subcase 3: V12=∅ and V34=∅, V23=V14=∅.
Let y12∈V12 and y34∈V34. By Rule 3,
V12→V23∪V234, V12→V41∪V341,
but V23=V14=∅. Thus,
there are y134∈V134 and y234∈V234 with
both y134 and y234 adjacent to y12.
Also by Rule 3, V34→V23∪V123 and
V34→∪V41∪V412. This implies
there are vertices y123∈V123 and y412∈V412
where both of these vertices are adjacent to y34. Let G11′ be
the subgraph of G induced by
N[x]∪{y12,y34,y134,y234,y123,y412}.
Then G′ has 11 vertices and has 22 triangles. When y is added to VS,
V12: yx1x2 yy134x1, yy234x2 (V34 is the same),
V123: yx1x2, yx2x3, yy34x3,
yy134x1, yy134x3
(V234, V134,
and V124 are the same).
Thus, k3(G)≥k3(G11′)+3(n−11)=3n−11≥2n−2.
Subcase 4: V12=∅ , V23=∅, V34=∅, V14=∅.
Let y12∈V12, y23∈V23, and y34∈V34.
Note that V12∪V23∪V34 is a complete 3-partite graph by Rule 5.
By Rule 3, y34→V14∪V124, but V14=∅.
Therefore, there is a y124∈V124 and this vertex is adjacent to
both y34 and y23.
Let G12′ be the subgraph of G induced by
N[x]∪{y12,y23,y34,y124}.
Then G12′ has 9 vertices and has 15 triangles. When y is added to VS,
V12 : yx1x2 yy23x2, yy23y34
V23 : yx2x3, yy34x3, yy124x2, yy34y12,
yy34y124
V34 : yx3x4, yy23x3, yy124x4, yy23y12,yy23y124
V123 : yx1x2 yx2x3, yy34x3
V234 : yx2x3, yx3x4, yy12y2,
V124 : yx1x2, yx1x4, yy23x2, yy34x4
V341: yx3x4, yx1x4, yy12x1.
Hence, k3(G)≥k3(G12′)+3(n−9)=3n−12.
Subcase 5: Each of V12, V23, V34, V14 is not empty.
For this subcase, we will count triangles in G in a different way than the previous subcases.
Let
X={x1,x2,x3,x4}, S=V12∪V23∪V34∪V41,
and
T=V123∪V234∪V341∪V412.
Claim 3.6
The number of triangles that contain at least one vertex in X is at least 2n−6.
Proof of Claim 3.6.
There are four triangles that contain x,
2∣T∣ triangles that contain two vertices in X and one in T,
and ∣S∣ triangles that contain two vertices in X and one in S.
By Rule 3, a vertex y12∈V12 lies in a triangle
of the form y12zx2 where z is some vertex in V23∪V234. Similar
statements hold for vertices in V23, V34, and V41. Altogether,
we have 4+2∣T∣+2∣S∣=4+2(n−5)=2n−6 triangles.
If ∣S∣=4, then we let Vij={yij} for (i,j)∈{(1,2),(2,3),(3,4),(4,1)}.
Since G is K4-free, Rule 5 implies that S induces a K4−e.
The two distinct cases, up to symmetry, are the missing edge e is
y12y23 or y12y34.
Suppose first e=y12y34. For any y123∈V123, we have
that y123y34y14 is a triangle. Likewise,
y234∈V234 implies y234y12y14 is a triangle,
y341∈V341 implies y341y23y12 is a triangle,
and
y412∈V234 implies y412y34y12 is a triangle.
The two triangles y12y23y34 and y12y41y34 have no vertex in X. The number
of triangles containing no vertex in X is at least
[TABLE]
Thus, by Claim 3.6, G contains at least 3n−13≥2n−2 triangles.
Now suppose e=y12y23. By Rule 3, y12 has a neighbor
in V23∪V234={y23}∪V234, but y12 is not adjacent to
y23 so V234=∅. Similarly, y23 has a neighbor in V12∪V412
but y23 is not adjacent to y12 so V412=∅.
For any y234∈V234, y234y12y41 is a triangle.
Likewise,
y412∈V412 implies y412y23y34 is a triangle,
and
y123∈V123 implies y123y34y41 is a triangle.
Therefore, there are at least
[TABLE]
triangles in G with no vertex in X. Since V234 and V412 are not empty,
[TABLE]
Hence, γ≥2+(n−9)−(n−11)=4. Combining this with Claim 3.6 gives
[TABLE]
The final possibility is if at least one of the parts V12, V23, V34, V41 contains
2 or more vertices. Assume ∣V12∣>1 and let y121,y122 be distinct vertices in
V12. Let y23∈V23, y34∈V34, and y41∈V41.
The set {y121,y23,y34,y41} cannot form a K4 and so by
Rule 5, this set of four vertices induces a K4−e. There are two triangles
using these vertices.
Similarly, there are two triangles using the vertices
{y122,y23,y34,y41} and regardless of which
pair of vertices is not adjacent in this set, one of the triangles must contain
y122. Thus, we have 3 distinct triangles contained in S. By Claim 3.6,
[TABLE]
This completes the proof of Lemma 3.5.
Combining Lemmas 3.2, 3.3, 3.4,
and 3.5 implies Theorem 1.4.
4 Proof of Proposition 1.9
Let s>3 and t≥6(2s−2) be integers.
Suppose G is an n-vertex Ks-saturated graph
with minimum degree t where n≥2s−2. We must show that G has at least
(2s−2)(n−2) triangles.
First assume every edge of G lies in a triangle. If t(e) is the number of triangles that contain the edge e,
then the number of triangles in G is
[TABLE]
Now assume there is an edge xy in G that does not lie in any triangle. Let A=N(x) and B=N(y).
Because xy is not in any triangle, A∩B=∅.
If a∈A, then N(y)∩N(a) must contain a copy of Ks−2.
The number of triangles that contain a and an edge from this Ks−2
is (2s−2). The same argument applies to a vertex b∈B.
Let C=V(G)\({x,y}∪A∪B).
If c∈C, then both N(c)∩N(x) and N(c)∩N(y) must
contain a copy of Ks−2. These two copies of Ks−2 cannot
share any edges because A∩B=∅. Thus, G must contain at least
[TABLE]
triangles. This completes the proof of Proposition 4.
5 Appendix
For G1′, the edges in N(x) are x1x2 and x3x4.
14 Triangles in G1′:
xx1x2, xx3x4,
y123x1x2, y124x1x2, y134x3x4, y234x3x4,
y123y134x1, y123y134x3,
y123y234x2, y123y234x3,
y124y134x1, y124y134x4,
y124y234x2, y124y234x4
For G2′ through G6′, the edges in N(x) are x1x2, x2x3, and x3x4.
V_{123}$$V_{124}$$V_{134}$$V_{234}$$V_{23}
11 Triangles in G2′: xx1x2, xx2x3, xx3x4, y23x2x3,
z124x1x2, z134x3x4,
y23z124x2,
y23z134x3, z124z134x1,
z124z134x4,
y23z124z134
14 Triangles in G3′: xx1x2, xx2x3, xx3x4, y23x2x3,
z124x1x2, z124′x1x2,
z234x2x3,
z234x3x4,
y23z124x2, y23z124′x2,
z124z234x2, z124z234x4, z124′z234x2, z124′z234x4
15 Triangles in G4′: xx1x2, xx2x3, xx3x4,
y123x1x2, y123x2x3, y124x1x2,
y234x2x3, y234x3x4,
y134x3x4,
y123x1y134, y123x3y134,
y124x1y134, y124x4y134,
y124x2y234,y124x4y234
8 Triangles in G5′: xx1x2, xx2x3, xx3x4,
y123x1x2, y123x2x3,
y134x3x4,
y123y134x1, y123y134x3
7 Triangles in G6′: xx1x2, xx2x3, xx3x4,
y124x1x2, y134x3x4, y124y134x1, y124y134x4
For G7′ through G12′, the edges in N(x) are x1x2, x2x3, x3x4, and x4x1.
V_{412}$$V_{123}$$V_{234}$$V_{341}$$V_{41}$$V_{12}$$V_{23}$$V_{34}
10 Triangles in G7′: xx1x2, xx2x3, xx3x4, xx4x1,
y123x1x2, y123x2x3, y134x3x4, y134x4x1,
y123y134x1, y123y134x3
16 Triangles in G8′: xx1x2, xx2x3, xx3x4, xx4x1,
y123x1x2, y123x2x3,
y412x1x2, y412x4x1,
y234x2x3, y234x3x4,
y134x3x4, y134x4x1,
y123y134x1, y123y134x3,
y412y234x2, y412y234x4
11 Triangles in G9′: xx1x2, xx2x3, xx3x4, xx4x1,
y12x1x2,
y134x3x4, y134x4x1,
y234x2x3, y234x3x4,
y12y134x1,
y12y234x2
12 Triangles in G10′: xx1x2, xx2x3, xx3x4, xx4x1,
y12x1x2,
y23x2x3,
y134x3x4, y134x4x1,
y12y134x1,
y23y134x1,
y12y23x2,
y12y23y134
22 Triangles in G11′: xx1x2, xx2x3, xx3x4, xx4x1,
y12x1x2, y34x3x4,
y123x1x2, y123x2x3,
y412x4x1, y412x1x2,
y234x2x3, y234x3x4,
y341x3x4, y341x4x1,
y12y134x1, y12y234x2,
y34y124x4, y34y123x3,
y134y123x1, y134y123x3,
y234y412x2, y234y412x4
15 Triangles in G12′: xx1x2, xx2x3, xx3x4, xx4x1,
y12x1x2, y23x2x3,
y34x3x4,
y124x1x2, y124x4x1,
y12y23x2, y23y34x3,
y124y23x2, y124y34x4,
y12y23y34,
y124y23y34