Elusive problems in extremal graph theory Andrzej Grzesik (Krakow) - - PowerPoint PPT Presentation

elusive problems in extremal graph theory
SMART_READER_LITE
LIVE PREVIEW

Elusive problems in extremal graph theory Andrzej Grzesik (Krakow) - - PowerPoint PPT Presentation

Elusive problems in extremal graph theory Andrzej Grzesik (Krakow) Dan Kr al (Warwick) L aszl o Mikl os Lov asz (MIT/Stanford) Monash 24/4/2017 1 Tur an Problems Maximum edge-density of H -free graph 1


slide-1
SLIDE 1

Elusive problems in extremal graph theory

Andrzej Grzesik (Krakow) Dan Kr´ al’ (Warwick) L´ aszl´

  • Mikl´
  • s Lov´

asz (MIT/Stanford)

Monash 24/4/2017

1

slide-2
SLIDE 2

Tur´ an Problems

  • Maximum edge-density of H-free graph
  • Mantel’s Theorem (1907):

1 2 for H = K3 (K n

2 , n 2 )

  • Tur´

an’s Theorem (1941):

ℓ−2 ℓ−1 for H = Kℓ (K

n ℓ−1 ,..., n ℓ−1 )

  • Erd˝
  • s-Stone Theorem (1946):

χ(H)−2 χ(H)−1

  • extremal examples unique up to o(n2) edges

2

slide-3
SLIDE 3

Edge vs. Triangle Problem

  • Minimum density of K3 for a specific edge-density
  • determined by Razborov (2008), Kαn,...,αn,(1−kα)n
  • extensions by Nikiforov (2011) and Reiher (2016) for Kℓ
  • Pikhurko and Razborov (2017) gave extremal examples

generally not unique, can be made unique by Kn = 0

3

slide-4
SLIDE 4

Another example

  • Minimum sum of densities of K3 and K3
  • Goodman’s Bound (1959): K3 + K3 ≥ 1

4

every n/2-regular graph is a minimizer

  • minimizer can be made unique K3 = 0, or K3 = 0, or

C4 = 1/16 (Erd˝

  • s-R´

enyi random graph Gn,1/2)

4

slide-5
SLIDE 5

This Talk

  • Conjecture (Lov´

asz 2008, Lov´ asz and Szegedy 2011) Every finite feasible set Hi = di, i = 1, . . . , k, can be extended to a finite feasible set with an asymptotically unique structure.

  • Every extremal problem has a finitely forcible optimum.
  • Theorem (Grzesik, K., Lov´

asz Jr.): FALSE

5

slide-6
SLIDE 6

Limits of dense graphs

  • d(H, G) = probability |H|-vertex subgraph of G is H
  • a sequence (Gn)n∈N of graphs is convergent

if d(H, Gn) converges for every H

  • examples: Kn, Kαn,n, blow ups G[Kn]

Erd˝

  • s-R´

enyi random graphs Gn,p, planar graphs

  • graphon W : [0, 1]2 → [0, 1], s.t. W(x, y) = W(y, x)

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

6

slide-7
SLIDE 7

Finitely forcible graph limits

  • a graphon W is finitely forcible if there exist

H1, . . . , Hk and d1, . . . , dk such that W is the only graphon with the expected density of Hi equal to di

  • ⇔ the only graphon minimizing αjd(H′

j, W)

density calculation: (Hi − di)2 = αjH′

j

  • Lov´

asz, S´

  • s (2008): Step graphons are finitely forcible.

7

slide-8
SLIDE 8

Statement of Problem

  • Conjecture (Lov´

asz 2008, Lov´ asz and Szegedy 2011): Every extremal problem min αjd(Hj, W) has a finitely forcible optimal solution.

  • extremal graph theory problem →

finitely forcible optimal solution → “simple structure” gives new bounds on old problems

  • Conjectures (Lov´

asz and Szegedy): The space T(W) of a finitely forcible W is compact. The space T(W) has finite dimension.

8

slide-9
SLIDE 9

Finitely forcible graph limits

  • Theorem (Cooper, K., Martins):

Every graphon is a subgraphon of a finitely forcible graphon.

A B C D E F G P Q R A B C D E F G P Q R WF

9

slide-10
SLIDE 10

Main result

  • Theorem (Grzesik, K., Lov´

asz Jr.) ∃ graphon family W, graphs Hi, reals di, i = 1, . . . , m W ∈ W ⇔ d(Hi, W) = di for i = 1, . . . , m no graphon in W is finitely forcible

A B C DA DB DC DD DE DF DG E F G H A B C DA DB DC DD DE DF DG E F G H

10

slide-11
SLIDE 11

Some details of the proof

  • graphons WP (⃗

z), ⃗ z ∈ [0, 1]N ⃗ z satisfies polynomial inequalities in P (e.g. z1 + z2

2 ≤ 1)

  • construct Ji ⊆ [0, 1], inequalities P and inj. maps fi

fi(x1, . . . , xi) = (z1, . . . , z (i+1)(i+2)

2

) (x1) → (z1, z2), (x1, x2) → (z1, z2, z3, z4, z5), etc. d(Hi, WP (⃗ z)) = gi(x1, . . . , xi) if xk ∈ Jk, k ∈ N each Ji has positive measure

  • ⇒ no graphon in WP (⃗

z) is finitely forcible

11

slide-12
SLIDE 12

Some details of the proof

  • graphons WP (⃗

z), ⃗ z ∈ [0, 1]N ⃗ z satisfies polynomial inequalities in P (e.g. z1 + z2

2 ≤ 1)

  • independent of P: there exist graphs H1, . . . , Hk

there exist polynomials q1, . . . , qℓ in d(Hi, W)

  • for every P: there exist reals α1, . . . , αℓ

WP (⃗ z) are precisely graphons satisfying qi = αi

  • analysis of the dependance of d(Hi, WP (⃗

z)) on P approximation of maps fi by polynomial inequalities

12

slide-13
SLIDE 13

Possible extensions

  • techniques universal to prove more general results

equalize other functions than subgraph densities

  • Theorem (Grzesik, K., Lov´

asz Jr.) ∃ graphon family W, graphs Hi, reals di, i = 1, . . . , m W ∈ W ⇔ d(Hi, W) = di for i = 1, . . . , m no graphon in W is finitely forcible all graphons in W have the same entropy

  • extremal problems with no typical structure

13

slide-14
SLIDE 14

Thank you for your attention!

14