11 The threshold for a Hamilton cycle

PIC

Theorem 11.1 (Pósa). If p≥16log⁡nn and G∼G(n,p) then G⊃Cn with high probability.

PIC

Let P=v1⋯vl be a path in a graph G. Let vivl∈G. Define the path Q by

Q=v1⋯vivlvl−1⋯vi+1.

We call this the rotation of P through the edge vivl.

PIC

We write for P,Q on the same vertex set P∼Q if I can obtain Q by applying a sequence of rotations to P. We define P∼P.

Define

X(P)={x∈V(P):x is the end of a path}.

Q and Q∼P.

X¯(P)=X(P)∪{vi:vi−1 or vi+1∈X(P)}.

Lemma 11.2. Let G be a graph, and let P⊃G be a path. Then

e(X(P),V(P)∖X¯(P))=0.

Proof. Let P be v1,v2,…,vl. Let x∈X(P) and let y∈V(P) and xy∈G. We show y∈X¯(P).

PIC

Let Px be a path Px∼P with endpoint x. Let y=vi.

Case 1: vi−1vi and vivi+1∈Px. We rotate through the edge xy from Px to form a path Q. Of course P¯∼Q. Note in this path Q, either vi−1 or vi+1∈X(P). This means vi∈X¯(P).

Case 2: If either vi−1vi or vivi+1∉Px then there was a first path P′, when we were rotating from P to Px, where we removed one of vi−1vi or vivi+1.

PIC

Note carefully that this means that in P′, one of vi−1,vi,vi+1 is an end of P. So vi∈X¯(P). □

Definition 11.3 (Expander). Say an n-vertex graph G is an expander if for all A,B⊂V(G) disjoint with 1≤|A|≤n6, |B|=n−3|A|, we have e(A,B)≠0.

Lemma 11.4. Let G be an expander on n vertices. Let P⊂G be a longest path. Then |X(P)|>n6 (note |P|≥|X(P)|).

Proof. If P is a longest path in G, we have that

e(X(P),V(G)∖P)=0,

otherwise we could rotate and extend P, contradicting maximality of P. By Lemma 11.2,

e(X(P),V(P)∖X¯(P))=0.

So

e(X(P),V(G)∖X(P)¯)=0.

Note

|X¯(P)|≤3|X(P)|

so use expander condition to see

|X(P)|>n6.□

Lemma 11.5. Let p≥12log⁡nn, G∼G(n,p). Then G is an expander with probability ≥1−n−2 for large enough n.

Proof.

ℙ(G not an expander)≤∑A⊂[n]|A|≤n6∑B⊂[n]|B|=n−3|A|ℙ(e(A,B)=0)≤∑k=1n6nkn−kn−3k(1−p)k(n−3k)≤∑k=1n6nkn2ke−pk(n−3k)≤∑k=1n6(enk)3ke−pk(n−3k)≤∑k=1n6(en3ke−p(n−3k))k≤∑k=1n6(en3kn−6)k≤n−3□

Lemma 11.6. If p≥12log⁡nn, G∼G(n,p). Then G⊃Pn with high probability.

Proof. Let Ei be the event that G−vi is an expander.

PIC

Let

Mi=“the longest path in G has the same length as the longest path in G−vi”.

Then

PIC

ℙ(G⁄⊃Pn)≤ℙ(⋃i=1nMi)≤nℙ(M1)≤nℙ(M1∩E1)+O(n−1)≤n⋅max⁡G−v∈M1∩E1ℙN(v1)(M1)+O(n−1)≤n(1−p)n6+O(n−1) Since a longest path P in G−v1 has |X(P)|≥n6. So if M1 is to hold, we cannot include any edges between v1 and X(P). To finish,
ne−pn6+O(n−1)=O(n−1).□

11.1 Sprinkling

Let G1∼G(n,p1), G2∼G(n,p2) on the same vertex set. Consider G=G1∪G2. Then G∼G(n,p), where p=1−(1−p1)(1−p2)≈p1+p2 for p1,p2 small.

PIC

Proof of G(n,p)⊃Cn when p>20log⁡nn. Let G1∼G(n,p1) where p1=12log⁡nn, G2∼G(n,p2) where p2=log⁡ log ⁡nn. Let G=G1∪G2: then G∼G(n,p) where p=(12+o(1))log⁡nn.

Define E to be the event that G1 is an expander. Define P to be the event that G1⊃Pn. Then

ℙ(G⁄⊃Cn)≤ℙ(G⁄⊃Cn∩E∩P)+ℙ(Ec)+ℙ(Pc)⏟o(1)≤max⁡G1∈E∩PℙG2(G⁄⊃Cn)+o(1)

Let P be a Hamiltonian path in G1.

PIC

Since G1 is an expander, we have |X(P)|>n6. So there are ≥n6 edges that, if included, will complete a Hamiltonian cycle by Pósa rotation lemma. So

ℙ(G⁄⊃Cn)≤(1−p)n6+o(1)→0.□