béla bollobás memphis guy kindler microsoft imre leader cambridge ryan o’donnell microsoft

13
Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft

Upload: bernice-woods

Post on 20-Jan-2016

215 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft

Béla BollobásMemphis

Guy KindlerMicrosoft

Imre LeaderCambridge

Ryan O’DonnellMicrosoft

Page 2: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft

Q: How many

vertices need be

deleted to block

non-trivial cycles?

Page 3: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft

(with “L1 edge structure”)

Q: How many

vertices need be

deleted to block

non-trivial cycles?

Upper bound: d ¢

md−1

Upper bound: d ¢

md−1

Page 4: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft

(with “L1 edge structure”)

Q: How many

vertices need be

deleted to block

non-trivial cycles?

Upper bound: d ¢

md−1

Lower bound: 1 ¢

md−1

A: ? ¢ md−1

Lower bound:

Page 5: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft

Motivation

Upper:

Lower: m

2 ¢

m

¢

m

Best:

Page 6: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft
Page 7: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft
Page 8: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft

tiling of with period

(with discretized boundary)

Page 9: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft

tiling of with period

(with discretized boundary)

Page 10: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft

0

m

# of vertices:

Theorem 1:

upper bound, for d = 2r.

(Hadamard matrix)

In dimension d = 2r…

Page 11: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft

Motivation

• “L1 structure”:

• [SSZ04]: Asymptotically tight lower bound.

(Yields integrality gap for DIRECTED MIN MULTICUT.)

• Our Theorem 2: Exactly tight lower bound.

• Edge-deletion version: Our original motivation.

Connected to quantitative aspects of Raz’s Parallel Repetition Theorem.

Page 12: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft

Open questions

• Obviously, better upper/lower bounds for various versions?

(L1 / L1, vertex deletion / edge deletion)

• Continuous, Euclidean version:

“What tiling of with period has minimal surface area?”

Trivial upper bound: d

Easy lower bound:

No essential improvement known.

Best for d = 2:

Page 13: Béla Bollobás Memphis Guy Kindler Microsoft Imre Leader Cambridge Ryan O’Donnell Microsoft