PROPERTIES OF THE TURAN GRAPH

MICHAEL TWITO
July 20, 2018

1 Introduction
In this note we will discuss in the so called Turan graph denoted as Tr (n)
which is complete r-partite graph of maximum size (amongst all r-partite
graphs on n vertices). The uniqueness part is the interesting one in this
definition, why is there a single complete r-partite graph of maximum size?
For r-partite graph let n1 ≤ n2 ≤ ... ≤ nr dentoe the size of each partition.

2 Definition
The Turan graph is the complete r-partite n-vertex graph whose partition
classes are as equal as possible, i.e nr ≤ n1 + 1

3 Claims
Claim: Suppose G is an n-vertex complete r-partite graph with n2 ≥ n1 + 2.
Then ∃ an n-vertex complete r-partite graph H s.t E(H) > E(G)
Proof. Define H to be the n-vertext complete r-partite graph s.t n1 (H) =
n1 (G) + 1,n2 (H) = n2 (G) − 1, and ni (H) = ni (G) for 3 ≤ i ≤ r. Then
r
X r
X X
E(H) = (n1 (G)+1)(n2 (G)−1)+ (n1 (G)+1)ni )+ (n2 (G)−1)ni + ni nj
i=3 i=3 3≤i<j≤r

applying the same claulation on E(G) we get that E(H) - E(G) = 1.

Claim: The inequalities:
X
n1 ≤ n2 ≤ ... ≤ n1 + 1, ni = n
i

1
determine the ni ’s uniquely. In particular
n+i−1
ni = b c
r
holds for every i ∈ [r]: Define tr (n) := E(Tr (n)). Then the equality
! !
n X ni X n+i−1 n+i−1
− = tr (n) = b cb c
2 i 2 1≤i<j≤r r r

captures two approaches to determine tr (n)

Proof. Lets proove the claim by induction on n. For n < r the claim is trivial
as we may set n1 = n2 = ... = nr−n and ni = 1 for r − n + 1 ≤ i ≤ r. In
particular, if n = r − j for some 1 ≤ j ≤ r − 1 we in fact set
(
r−j+1 0, i<j+1
ni = b c=
r 1, i≥j+1
Now for the induction step given n1 , ..., nr as above define n0i = n1 − 1 Then
n01 ≤ n02 ≤ ... ≤ n0r ≤ n01 + 1, n01 = n − r
X

By th I.H
(n − r) + i − 1
n0i = b c
r
Recalling that ni = n0i + 1 and that n−r+i−1
r
+1 = n+i−1
r
completes the
proof.
Claim: Let r be a fixed integer then
!
1 n
tr (n) = (1 − + o(1))
r 2

Proof. We start with the lower bound

! !
r n 2 r r(r − 1)
E(Tr (n)) ≥ b c ≥ (n/r − 1)2 = ((n/r)2 − 2n/r + 1)
2 r 2 2

Observe
 
that now r(r−1)
2
(n/r)2 = r−1
r
n2 /2 = (1 − 1/r)n2 /2. Also recall that
n
2
= n2 /2 − n/2 so that n2 /2 = n2 + o(n2 ). Also r(r−1)
2
(2n/r + 1) = o(n2 )
 
It follows that E(Tr (n)) ≥ (1 − 1/r + o(1)) n2 For the upper bound note that
in our lower bound estimation we neglected at most one vertex per partition
class. When all of these neglected vertices account for at most rn = o(n2 )
edges in the graph. The upper bound then follows.

2
Claim: Let r > 0 be fixed. Then ∃n0 := n0 (r) s.t
!
1 n
tr (n) ≥ (1 − ) , ∀n ≥ n0
r 2

Observing Tr (n) we note that

δ(Tr (n)) = n − dn/re, ∆(Tr (n)) = n − bn/rc

hence, Tr (n) maximises δ(G) and minimises ∆(G) amongst all graphs with
tr (n) edges.
Proof. Let n = qr + s s.t 0 ≤ s ≤ r − 1.Assume s > 0(otherwise the
claim is trivial by the proof of the previous claim).This means that n1 =
n2 = ... = nr−s = q and nr−s+1 = ... = nr = q + 1. Indeed, we note that
(r − s)q + s(q + 1) = qr + s = n . Then:
! ! !
n q q+1
E(Tr (n)) ≥ − r(r − s) −s
2 2 2
!
n rq(q − 1) sq(q − 1) sq(q + 1)
= − + −
2 2 2 2
!
n rq(q − 1) sq
= − − (−(q − 1) + q + 1)
2 2 2
!
n rq(q − 1)
= − − sq
2 2
!
n n+s−r
= −q
2 2
!
n n−s n+s−r
= −( )( )
2 r 2
n2 sn rn sn s2
!
n sr
= − − + + + −
2 2r 2r 2r 2r 2r 2r
s2
! !
n 1 n n n s
= − + − + −
2 r 2 2 2r 2r 2
n n s 2 s(r−s)
We require that 2
− 2r + 2r − 2s > 0. This occurs when n > r−1
concluding
the proof.

3
Claim: Let G be an n-vertex graph with E(G) = tr (n). Then

δ(G) ≤ δ(Tr (n)) ≤ ∆(Tr (n)) ≤ ∆(G)

Proof. Define A := {v : deg(n) = δ(Tr (n))},B := {v : deg(n) = ∆(Tr (n))}

Then: X X
2tr (n) = (n − dn/re) + (n − bn/rc)
v∈A v∈B

For Tr (n) we know that |A| + |B| = n. Let G be an n-vertex graph with
E(G) = tr (n).If δ(G) > δ(Tr (n)) then δ(G) ≥ ∆(Tr (n)).In which case
2E(G) ≥ n∆(Tr (n)). However as 2E(G) = 2tr (n) this implies that n and r
are s.t δ(Tr (n)) = ∆(Tr (n)). But now δ(G) > δ(Tr (n)) = ∆(Tr (n)) implying
that E(G) > tr (n); a contradiction. For ∆(G) the same argument holds with
< Tr (n)