Mathias Schacht
November 2002
Outline
1 Introduction
Extremal Graph Theory Random Graph Theory
Extremal Random Graph Theory The Main Result
1 Introduction
Extremal Graph Theory Random Graph Theory
Extremal Random Graph Theory The Main Result
2 Proof of the Main Result
Szemer ´edi’s regularity lemma for sparse graphs
The Counting Lemma for subgraphs for random graphs Outline of the proof the Main Result
Outline
1 Introduction
Extremal Graph Theory Random Graph Theory
Extremal Random Graph Theory The Main Result
2 Proof of the Main Result
Szemer ´edi’s regularity lemma for sparse graphs
The Counting Lemma for subgraphs for random graphs Outline of the proof the Main Result
3 Proof of the Counting Lemma
Outline of the proof of the Counting Lemma for H = K4 − e The k-tuple Lemma
The “Pick-Up” Lemma
Extremal (Deterministic) Graph Theory
Motivation.
What is the structure of a H-free graph with a maximum number of edges, for some fixed graph H?
Extremal (Deterministic) Graph Theory
Motivation.
What is the structure of a H-free graph with a maximum number of edges, for some fixed graph H?
Which properties imply the existence of a certain fixed subgraph H?
Definition. ex(G, H) = max{e(F)|H 6⊆ F ⊆ G}
Results in Extremal Graph Theory
Definition. ex(G, H) = max{e(F)|H 6⊆ F ⊆ G} Observation. For every graph G and H
ex(G, H) ≥ e(G) − #{H0 ⊆ G} for every H0 ⊆ H.
Definition. ex(G, H) = max{e(F)|H 6⊆ F ⊆ G} Observation. For every graph G and H
ex(G, H) ≥ e(G) − #{H0 ⊆ G} for every H0 ⊆ H.
Theorem (Tur ´an 1941).
ex(Kn, Kl) =
1 − 1
l − 1 + O
1 n
n 2
Results in Extremal Graph Theory
Definition. ex(G, H) = max{e(F)|H 6⊆ F ⊆ G} Observation. For every graph G and H
ex(G, H) ≥ e(G) − #{H0 ⊆ G} for every H0 ⊆ H.
Theorem (Tur ´an 1941).
ex(Kn, Kl) =
1 − 1
l − 1 + O
1 n
n 2
Theorem (Erd ˝os–Stone–Simonovits 1946, 1966). For every graph H ex(Kn, H) = 1 − 1
χ(H) − 1 + o(1)
! n
2
.
Definition (G(n, p)).
For 0 ≤ p = p(n) ≤ 1, the binomial random graph G in G(n, p) has n vertices and each edge occurs independently with probability p.
Random Graph Theory
Definition (G(n, p)).
For 0 ≤ p = p(n) ≤ 1, the binomial random graph G in G(n, p) has n vertices and each edge occurs independently with probability p.
Equivalently, G(n, p) is the probabilty space over all labeled graphs on n vertices by setting
P(G) = pe(G)(1 − p)(
n
2)−e(G).
Definition (G(n, p)).
For 0 ≤ p = p(n) ≤ 1, the binomial random graph G in G(n, p) has n vertices and each edge occurs independently with probability p.
Equivalently, G(n, p) is the probabilty space over all labeled graphs on n vertices by setting
P(G) = pe(G)(1 − p)(
n
2)−e(G).
We say a property P holds asymptotically almost surely (a.a.s.) if P(G ∈ G(n, p) satisfies P) = 1 − o(1).
Extremal Random Graph Theory
For 0 ≤ p = p(n) ≤ 1 consider the random variable:
ex(G(n, p), H)
For 0 ≤ p = p(n) ≤ 1 consider the random variable:
ex(G(n, p), H)
Question. For which p does
ex(G(n, p), H) = 1 + 1
χ(H) − 1 + o(1)
!
pn 2
(1) hold a.a.s.?
Extremal Random Graph Theory
For 0 ≤ p = p(n) ≤ 1 consider the random variable:
ex(G(n, p), H)
Question. For which p does
ex(G(n, p), H) = 1 + 1
χ(H) − 1 + o(1)
!
pn 2
(1) hold a.a.s.?
Fact. If (1) holds a.a.s. for some p1, then it does a.a.s. for each p2 ≥ p1. Question. What is the minimal p = p(n)? What is the threshold?
Main Conjecture
Definition (2-density). Let H be a graph with v(H) ≥ 3 m2(H) = max
(e(H0) − 1 v(H0) − 2
H0 ⊆ H,|V (H0)| ≥ 3
)
.
Main Conjecture
Definition (2-density). Let H be a graph with v(H) ≥ 3 m2(H) = max
(e(H0) − 1 v(H0) − 2
H0 ⊆ H,|V (H0)| ≥ 3
)
.
Conjecture (Kohayakawa, Łuczak & R ¨odl 1997). Let H be a non-empty graph with v(H) ≥ 3 and let 0 ≤ p(n) ≤ 1 such that p(n) n−1/m2(H). Then a.a.s.
ex(G, H) = 1 + 1
χ(H) − 1 + o(1)
!
e(G) holds for G in G(n, p).
The Lower Bound ( p ( n ) n
−1/m2(H))
Recall. For every graph G and H
ex(G, H) ≥ e(G) − #{H0 ⊆ G} for every H0 ⊆ H.
The Lower Bound ( p ( n ) n
−1/m2(H))
Recall. For every graph G and H
ex(G, H) ≥ e(G) − #{H0 ⊆ G} for every H0 ⊆ H.
Observation.
p(n) n−1/m2(H) ⇔ p(n)n2 p(n)e(H0)nv(H0) for every H0 ⊆ H.
Definition (d-degenerate). A graph H is d-degenerate, if d(H) = max
H0⊆H δ(H0) ≤ d.
Main Result
Definition (d-degenerate). A graph H is d-degenerate, if d(H) = max
H0⊆H δ(H0) ≤ d.
Theorem (Main Result). For every graph H, real δ > 0, and p = p(n) (logn)4/n1/d(H),
a graph G in G(n, p) a.a.s. satisfies the following property:
If F is an arbitrary, not necessarily induced subgraph of G with e(F) ≥ 1 − 1
χ(H) − 1 + δ
!
pn 2
, then H ⊆ F.
Corollary (H = Kl). For
p(n) (logn)4 n
!1/(l−1)
G in G(n, p) a.a.s. satisfies
ex(G, Kl) =
1 + 1
l + o(1)
e(G).
Corollary (H = Kl). For
p(n) (logn)4 n
!1/(l−1)
G in G(n, p) a.a.s. satisfies
ex(G, Kl) =
1 + 1
l + o(1)
e(G).
Remark. KŁR–Conjecture for H = Kl: p(n)
1 n
2/(l+1)
is sufficient.
Related Work
H is a forest follows from a simple application of Szemer ´edi’s regu- larity lemma for sparse graphs
Related Work
H is a forest follows from a simple application of Szemer ´edi’s regu- larity lemma for sparse graphs
H = K3 follows from a result of Frankl and R ¨odl, 1986 H = C4 follows from a result of F ¨uredi 1994
Related Work
H is a forest follows from a simple application of Szemer ´edi’s regu- larity lemma for sparse graphs
H = K3 follows from a result of Frankl and R ¨odl, 1986 H = C4 follows from a result of F ¨uredi 1994
H = Cl proved by Haxell, Kohayakawa & Łuczak, 1995 and 1996
Related Work
H is a forest follows from a simple application of Szemer ´edi’s regu- larity lemma for sparse graphs
H = K3 follows from a result of Frankl and R ¨odl, 1986 H = C4 follows from a result of F ¨uredi 1994
H = Cl proved by Haxell, Kohayakawa & Łuczak, 1995 and 1996
H = K4 proved by Kohayakawa, Łuczak & R ¨odl, 1997
H = K5 follows from a stronger result proved by Gerke, Schickinger & Steger, 2002+
Related Work
H is a forest follows from a simple application of Szemer ´edi’s regu- larity lemma for sparse graphs
H = K3 follows from a result of Frankl and R ¨odl, 1986 H = C4 follows from a result of F ¨uredi 1994
H = Cl proved by Haxell, Kohayakawa & Łuczak, 1995 and 1996
H = K4 proved by Kohayakawa, Łuczak & R ¨odl, 1997
H = K5 follows from a stronger result proved by Gerke, Schickinger & Steger, 2002+
H = Kl Szab ´o and Vu proved it for p(n) n−1/(l−1.5), 2002+
Summary & Outline
1 Introduction
Extremal Graph Theory Random Graph Theory
Extremal Random Graph Theory The Main Result
1 Introduction
Extremal Graph Theory Random Graph Theory
Extremal Random Graph Theory The Main Result
2 Proof of the Main Result
Szemer ´edi’s regularity lemma for sparse graphs
The Counting Lemma for subgraphs for random graphs Outline of the proof the Main Result
3 Proof of the Counting Lemma
Outline of the proof of the Counting Lemma for H = K4 − e The k-tuple Lemma
The “Pick-Up” Lemma
Preliminary Definitions
Definition ((ε, p)-regular). For G = (V, E), ε > 0, U, W ⊆ V disjoint, we say (U,W) is (ε, p)-regular if for all U0 ⊆ U, W0 ⊆ W with |U0| ≥ ε|U| and |W0| ≥ ε|W| we have
eG(U0, W0)
p|U0||W0| − eG(U, W) p|U||W|
≤ ε.
Definition ((ε, p)-regular). For G = (V, E), ε > 0, U, W ⊆ V disjoint, we say (U,W) is (ε, p)-regular if for all U0 ⊆ U, W0 ⊆ W with |U0| ≥ ε|U| and |W0| ≥ ε|W| we have
eG(U0, W0)
p|U0||W0| − eG(U, W) p|U||W|
≤ ε.
Definition ((ξ, C)-bounded). For ξ > 0, C > 1, and 0 ≤ p ≤ 1 we say that G = (V, E) is (ξ, C)-bounded with respect to density p, if for all U0, W0 ⊆ V not necessarily disjoint with |U0|,|V 0| ≥ ξ|V | we have
eG(U0, W0) ≤ Cp |U0||W0| − |U0 ∩ W0| + 1 2
!
.
Szemer ´edi’s regularity lemma for sparse graphs
Theorem (Kohayakawa, R ¨odl 1997). For every ε > 0, C > 1, and t0 ≥ 1, there exist constants ξ = ξ(ε, C, t0) and T0 = T0(ε, C, t0) such that the vertex set of any graph G = (V, E) that is (ξ, C)-bounded with respect to density p can be partitioned V = V0 ∪ V1 ∪ · · · ∪ Vt such that
Theorem (Kohayakawa, R ¨odl 1997). For every ε > 0, C > 1, and t0 ≥ 1, there exist constants ξ = ξ(ε, C, t0) and T0 = T0(ε, C, t0) such that the vertex set of any graph G = (V, E) that is (ξ, C)-bounded with respect to density p can be partitioned V = V0 ∪ V1 ∪ · · · ∪ Vt such that
(i) t0 ≤ t ≤ T0
Szemer ´edi’s regularity lemma for sparse graphs
Theorem (Kohayakawa, R ¨odl 1997). For every ε > 0, C > 1, and t0 ≥ 1, there exist constants ξ = ξ(ε, C, t0) and T0 = T0(ε, C, t0) such that the vertex set of any graph G = (V, E) that is (ξ, C)-bounded with respect to density p can be partitioned V = V0 ∪ V1 ∪ · · · ∪ Vt such that
(i) t0 ≤ t ≤ T0
(ii) V0 ≤ ε|V | and |V1| = |V2| = · · · = |Vt|
Theorem (Kohayakawa, R ¨odl 1997). For every ε > 0, C > 1, and t0 ≥ 1, there exist constants ξ = ξ(ε, C, t0) and T0 = T0(ε, C, t0) such that the vertex set of any graph G = (V, E) that is (ξ, C)-bounded with respect to density p can be partitioned V = V0 ∪ V1 ∪ · · · ∪ Vt such that
(i) t0 ≤ t ≤ T0
(ii) V0 ≤ ε|V | and |V1| = |V2| = · · · = |Vt|
(iii) all but at most ε2t pairs (Vi, Vj) are (ε, p)-regular
Szemer ´edi’s regularity lemma for sparse graphs
Theorem (Kohayakawa, R ¨odl 1997). For every ε > 0, C > 1, and t0 ≥ 1, there exist constants ξ = ξ(ε, C, t0) and T0 = T0(ε, C, t0) such that the vertex set of any graph G = (V, E) that is (ξ, C)-bounded with respect to density p can be partitioned V = V0 ∪ V1 ∪ · · · ∪ Vt such that
(i) t0 ≤ t ≤ T0
(ii) V0 ≤ ε|V | and |V1| = |V2| = · · · = |Vt|
(iii) all but at most ε2t pairs (Vi, Vj) are (ε, p)-regular
A partition V = V0 ∪ V1 ∪ · · · ∪ Vt satisfying (i)–(iii) we call (ε, t)-regular partition.
• H = (V, E) is a fixed d-degenerate graph on h
• order V = {w1, . . . , wh} such that |Γ(wi) ∩ {w1, . . . , wi−1}| ≤ d
Set-up
• H = (V, E) is a fixed d-degenerate graph on h
• order V = {w1, . . . , wh} such that |Γ(wi) ∩ {w1, . . . , wi−1}| ≤ d
• t ≥ h is a fixed integer and n is sufficiently large
• 0 < α, ε < 1 are real constants
• H = (V, E) is a fixed d-degenerate graph on h
• order V = {w1, . . . , wh} such that |Γ(wi) ∩ {w1, . . . , wi−1}| ≤ d
• t ≥ h is a fixed integer and n is sufficiently large
• 0 < α, ε < 1 are real constants
• G is a graph in G(n, p)
• J ⊆ G is h-partite with V (J) = V1 ∪ · · · ∪ Vh, denote J[Vi, Vj] by Jij
Set-up
• H = (V, E) is a fixed d-degenerate graph on h
• order V = {w1, . . . , wh} such that |Γ(wi) ∩ {w1, . . . , wi−1}| ≤ d
• t ≥ h is a fixed integer and n is sufficiently large
• 0 < α, ε < 1 are real constants
• G is a graph in G(n, p)
• J ⊆ G is h-partite with V (J) = V1 ∪ · · · ∪ Vh, denote J[Vi, Vj] by Jij Moreover, consider the following assertions.
(I) |Vi| = m = n/t (II) pdn (logn)4
• H = (V, E) is a fixed d-degenerate graph on h
• order V = {w1, . . . , wh} such that |Γ(wi) ∩ {w1, . . . , wi−1}| ≤ d
• t ≥ h is a fixed integer and n is sufficiently large
• 0 < α, ε < 1 are real constants
• G is a graph in G(n, p)
• J ⊆ G is h-partite with V (J) = V1 ∪ · · · ∪ Vh, denote J[Vi, Vj] by Jij Moreover, consider the following assertions.
(I) |Vi| = m = n/t (II) pdn (logn)4
(III) for all 1 ≤ i < j ≤ h e(Jij) =
T = αpm2 {wi, wj} ∈ E(H) 0 {wi, wj} 6∈ E(H) (IV) Jij is (ε, p)-regular.
The Counting Lemma for subgraphs of random graphs
Lemma (Counting Lemma). For all reals α, σ > 0, positive integer d and every d-degenerate graph H on h vertices, there exists a real ε > 0 such that for every fixed integer t ≥ h a random graph G in G(n, p) satisfies the following property a.a.s.:
Every subgraph J ⊆ G satisfying conditions (I)–(IV) contains at least (1 − σ)(αp)e(H)mh
copies of H.
Outline of the Proof of the Main Result
Facts.
(P1) For G in G(n, p) a.a.s. e(G) = (1 + o(1))pn2.
Outline of the Proof of the Main Result
Facts.
(P1) For G in G(n, p) a.a.s. e(G) = (1 + o(1))pn2.
(P2) ∀ ξ > 0, C > 1 if p = p(n) 1/n a.a.s. G in G(n, p) is (ξ, C)- bounded with respect to density p.
Outline of the Proof of the Main Result
Facts.
(P1) For G in G(n, p) a.a.s. e(G) = (1 + o(1))pn2.
(P2) ∀ ξ > 0, C > 1 if p = p(n) 1/n a.a.s. G in G(n, p) is (ξ, C)- bounded with respect to density p.
Sketch of the proof.
• typical “regularity lemma” proof
Outline of the Proof of the Main Result
Facts.
(P1) For G in G(n, p) a.a.s. e(G) = (1 + o(1))pn2.
(P2) ∀ ξ > 0, C > 1 if p = p(n) 1/n a.a.s. G in G(n, p) is (ξ, C)- bounded with respect to density p.
Sketch of the proof.
• typical “regularity lemma” proof
• Let H be the fixed d-degenerate graph.
Outline of the Proof of the Main Result
Facts.
(P1) For G in G(n, p) a.a.s. e(G) = (1 + o(1))pn2.
(P2) ∀ ξ > 0, C > 1 if p = p(n) 1/n a.a.s. G in G(n, p) is (ξ, C)- bounded with respect to density p.
Sketch of the proof.
• typical “regularity lemma” proof
• Let H be the fixed d-degenerate graph.
• Let G be a random graph in G(n, p) satisfying the Facts, and δ > 0.
Outline of the Proof of the Main Result
Facts.
(P1) For G in G(n, p) a.a.s. e(G) = (1 + o(1))pn2.
(P2) ∀ ξ > 0, C > 1 if p = p(n) 1/n a.a.s. G in G(n, p) is (ξ, C)- bounded with respect to density p.
Sketch of the proof.
• typical “regularity lemma” proof
• Let H be the fixed d-degenerate graph.
• Let G be a random graph in G(n, p) satisfying the Facts, and δ > 0.
• Let F ⊆ G with more than (1 − 1/(χ(H) − 1) + δ)pn2 edges.
Outline of the Proof of the Main Result
Facts.
(P1) For G in G(n, p) a.a.s. e(G) = (1 + o(1))pn2.
(P2) ∀ ξ > 0, C > 1 if p = p(n) 1/n a.a.s. G in G(n, p) is (ξ, C)- bounded with respect to density p.
Sketch of the proof.
• typical “regularity lemma” proof
• Let H be the fixed d-degenerate graph.
• Let G be a random graph in G(n, p) satisfying the Facts, and δ > 0.
• Let F ⊆ G with more than (1 − 1/(χ(H) − 1) + δ)pn2 edges.
• Choose α and σ appropriately and the Counting lemma yields εCL.
Outline of the Proof of the Main Result
Facts.
(P1) For G in G(n, p) a.a.s. e(G) = (1 + o(1))pn2.
(P2) ∀ ξ > 0, C > 1 if p = p(n) 1/n a.a.s. G in G(n, p) is (ξ, C)- bounded with respect to density p.
Sketch of the proof.
• typical “regularity lemma” proof
• Let H be the fixed d-degenerate graph.
• Let G be a random graph in G(n, p) satisfying the Facts, and δ > 0.
• Let F ⊆ G with more than (1 − 1/(χ(H) − 1) + δ)pn2 edges.
• Choose α and σ appropriately and the Counting lemma yields εCL.
• Obtain an (ε, t)-regular partition P of F and ξ for appropriate chosen C, ε = ε(εCL) and t0 ≤ t by the regularity lemma for sparse graphs.
• Consider the cluster graph FCluster of F with respect to P, depending on ε and α.
Outline of the Proof of the Main Result (continued)
• Consider the cluster graph FCluster of F with respect to P, depending on ε and α.
• Use (P2) and “smart” choice of constants to show that e(FCluster) ≥ 1 − 1
χ(H) − 1 + δ0
! t
2
.
• Consider the cluster graph FCluster of F with respect to P, depending on ε and α.
• Use (P2) and “smart” choice of constants to show that e(FCluster) ≥ 1 − 1
χ(H) − 1 + δ0
! t
2
.
⇒ H ⊆ FCluster by Erd ˝os–Stone–Simonovits
Outline of the Proof of the Main Result (continued)
• Consider the cluster graph FCluster of F with respect to P, depending on ε and α.
• Use (P2) and “smart” choice of constants to show that e(FCluster) ≥ 1 − 1
χ(H) − 1 + δ0
! t
2
.
⇒ H ⊆ FCluster by Erd ˝os–Stone–Simonovits
• This copy of H translates to a J0 ⊆ F which almost satisfies the as- sumptions (set-up) of the Counting Lemma.
• Consider the cluster graph FCluster of F with respect to P, depending on ε and α.
• Use (P2) and “smart” choice of constants to show that e(FCluster) ≥ 1 − 1
χ(H) − 1 + δ0
! t
2
.
⇒ H ⊆ FCluster by Erd ˝os–Stone–Simonovits
• This copy of H translates to a J0 ⊆ F which almost satisfies the as- sumptions (set-up) of the Counting Lemma.
• “Massage” J0 to find a subgraph J ⊆ F satisfying the set-up of the Counting Lemma.
Outline of the Proof of the Main Result (continued)
• Consider the cluster graph FCluster of F with respect to P, depending on ε and α.
• Use (P2) and “smart” choice of constants to show that e(FCluster) ≥ 1 − 1
χ(H) − 1 + δ0
! t
2
.
⇒ H ⊆ FCluster by Erd ˝os–Stone–Simonovits
• This copy of H translates to a J0 ⊆ F which almost satisfies the as- sumptions (set-up) of the Counting Lemma.
• “Massage” J0 to find a subgraph J ⊆ F satisfying the set-up of the Counting Lemma.
⇒ H ⊆ F with probability 1 − o(1)
1 Introduction
Extremal Graph Theory Random Graph Theory
Extremal Random Graph Theory The Main Result
2 Proof of the Main Result
Szemer ´edi’s regularity lemma for sparse graphs
The Counting Lemma for subgraphs for random graphs Outline of the proof the Main Result
Summary & Outline
1 Introduction
Extremal Graph Theory Random Graph Theory
Extremal Random Graph Theory The Main Result
2 Proof of the Main Result
Szemer ´edi’s regularity lemma for sparse graphs
The Counting Lemma for subgraphs for random graphs Outline of the proof the Main Result
3 Proof of the Counting Lemma
Outline of the proof of the Counting Lemma for H = K4 − e The k-tuple Lemma
The “Pick-Up” Lemma
4
−
Outline of the proof of the Counting Lemma for H = K
4− e
4
−
Outline of the proof of the Counting Lemma for H = K
4− e
4
−
Outline of the proof of the Counting Lemma for H = K
4− e
4
−
Outline of the proof of the Counting Lemma for H = K
4− e
4
−
Outline of the proof of the Counting Lemma for H = K
4− e
4
−
Outline of the proof of the Counting Lemma for H = K
4− e
4
−
Outline of the proof of the Counting Lemma for H = K
4− e
4
−
Outline of the proof of the Counting Lemma for H = K
4− e
The k-tuple Lemma
Let G ∈ G(n, p) and B = (U, W;E) be a bipartite, not necessarily induced subgraph of G with |U| = m1 and |W| = m2. Let q = e(B)/(m1m2).
The k-tuple Lemma
Let G ∈ G(n, p) and B = (U, W;E) be a bipartite, not necessarily induced subgraph of G with |U| = m1 and |W| = m2. Let q = e(B)/(m1m2). For some fixed integer k ≥ 1 and real γ > 0 consider
B(k)(U, W;γ) = {b = {v1, . . . , vk} ∈ [W]k| |dB(b)− qkm1| ≥ γqkm1} where dB(b) is the size of the joint neighbourhood of b in B.
The k-tuple Lemma
Let G ∈ G(n, p) and B = (U, W;E) be a bipartite, not necessarily induced subgraph of G with |U| = m1 and |W| = m2. Let q = e(B)/(m1m2). For some fixed integer k ≥ 1 and real γ > 0 consider
B(k)(U, W;γ) = {b = {v1, . . . , vk} ∈ [W]k| |dB(b)− qkm1| ≥ γqkm1} where dB(b) is the size of the joint neighbourhood of b in B.
Lemma (k-tuple Lemma; Kohayakawa and R ¨odl 2002+). For every reals α, γ, η > 0, integer k ≥ 1, and function m0 = m0(n) such that pkm0 (logn)4 there exists a real constant ε > 0 for which a.a.s. G in G(n, p) satisfies the following property:
The k-tuple Lemma
Let G ∈ G(n, p) and B = (U, W;E) be a bipartite, not necessarily induced subgraph of G with |U| = m1 and |W| = m2. Let q = e(B)/(m1m2). For some fixed integer k ≥ 1 and real γ > 0 consider
B(k)(U, W;γ) = {b = {v1, . . . , vk} ∈ [W]k| |dB(b)− qkm1| ≥ γqkm1} where dB(b) is the size of the joint neighbourhood of b in B.
Lemma (k-tuple Lemma; Kohayakawa and R ¨odl 2002+). For every reals α, γ, η > 0, integer k ≥ 1, and function m0 = m0(n) such that pkm0 (logn)4 there exists a real constant ε > 0 for which a.a.s. G in G(n, p) satisfies the following property:
If B = (U, W;F) ⊆ G is such that (i) e(B) ≥ αe(G[U, W]),
(ii) B is (ε, p)-regular,
(iii) |U| = m1 ≥ m0 and |W| = m2 ≥ m0,
The k-tuple Lemma
Let G ∈ G(n, p) and B = (U, W;E) be a bipartite, not necessarily induced subgraph of G with |U| = m1 and |W| = m2. Let q = e(B)/(m1m2). For some fixed integer k ≥ 1 and real γ > 0 consider
B(k)(U, W;γ) = {b = {v1, . . . , vk} ∈ [W]k| |dB(b)− qkm1| ≥ γqkm1} where dB(b) is the size of the joint neighbourhood of b in B.
Lemma (k-tuple Lemma; Kohayakawa and R ¨odl 2002+). For every reals α, γ, η > 0, integer k ≥ 1, and function m0 = m0(n) such that pkm0 (logn)4 there exists a real constant ε > 0 for which a.a.s. G in G(n, p) satisfies the following property:
If B = (U, W;F) ⊆ G is such that (i) e(B) ≥ αe(G[U, W]),
(ii) B is (ε, p)-regular,
(iii) |U| = m1 ≥ m0 and |W| = m2 ≥ m0, then
|B(k)(U, W;γ)| ≤ η
m2 k
.
The “Pick-Up” Lemma
Set-up.
• k ≥ 2 fixed integer and m sufficently large, 1/m q = q(m) < 1, and set T = qm
Set-up.
• k ≥ 2 fixed integer and m sufficently large, 1/m q = q(m) < 1, and set T = qm
• V1, . . . , Vk pairwise disjoint sets of size m
The “Pick-Up” Lemma
Set-up.
• k ≥ 2 fixed integer and m sufficently large, 1/m q = q(m) < 1, and set T = qm
• V1, . . . , Vk pairwise disjoint sets of size m
• Consider the uniform probability space Ω = V1 × Vk
T
× · · · × Vk−11 × Vk T
Set-up.
• k ≥ 2 fixed integer and m sufficently large, 1/m q = q(m) < 1, and set T = qm
• V1, . . . , Vk pairwise disjoint sets of size m
• Consider the uniform probability space Ω = V1 × Vk
T
× · · · × Vk−11 × Vk T
• For Ri ∈ Vi×TVk and vk ∈ Vk set dR
i(vk) = |{vi ∈ Vi| {vi, vk} ∈ Ri}|
The “Pick-Up” Lemma
Set-up.
• k ≥ 2 fixed integer and m sufficently large, 1/m q = q(m) < 1, and set T = qm
• V1, . . . , Vk pairwise disjoint sets of size m
• Consider the uniform probability space Ω = V1 × Vk
T
× · · · × Vk−11 × Vk T
• For Ri ∈ Vi×TVk and vk ∈ Vk set dR
i(vk) = |{vi ∈ Vi| {vi, vk} ∈ Ri}|
• B ⊆ V1 × · · · × Vk
Set-up.
• k ≥ 2 fixed integer and m sufficently large, 1/m q = q(m) < 1, and set T = qm
• V1, . . . , Vk pairwise disjoint sets of size m
• Consider the uniform probability space Ω = V1 × Vk
T
× · · · × Vk−11 × Vk T
• For Ri ∈ Vi×TVk and vk ∈ Vk set dR
i(vk) = |{vi ∈ Vi| {vi, vk} ∈ Ri}|
• B ⊆ V1 × · · · × Vk
Definition (Π(ζ, µ, K,B)). For ζ, µ, K > 0 and B ⊆ V1 × · · · ×Vk we say Π(ζ, µ, K,B) holds for R = (R1, . . . , Rk−1) ∈ Ω if
V˜k(K) = {vk ∈ Vk | dR
i(vk) ≤ Kqm,∀1 ≤ i < k} and
B(R) = {b = (v1, . . . , vk) ∈ B | vk ∈ V˜k ∧ {vi, vk} ∈ Ri,∀1 ≤ i < k} satisfy the inequalities
|V˜k| ≥ (1 − µ)m
|B(R)| ≤ ζqk−1mk.
k
Π(ζ, µ, K,B) holds for R = (R1, . . . , Rk−1) ∈ Ω if V˜k(K) = {vk ∈ Vk | dR
i(vk) ≤ Kqm,∀1 ≤ i < k} and
B(R) = {b = (v1, . . . , vk) ∈ B | vk ∈ V˜k ∧ {vi, vk} ∈ Ri,∀1 ≤ i < k} satisfy the inequalities
|V˜k| ≥ (1 − µ)m
|B(R)| ≤ ζqk−1mk.
Lemma (Pick-Up Lemma). For every β, ζ, µ > 0 there exist η = η(β, ζ, µ) >
0, K = K(β, µ) > 0 and m0 such that if m ≥ m0 and
|B| ≤ ηmk, then
P(Π(ζ, µ, K,B) fails for R ∈ Ω) ≤ β(k−1)T.