Clique vs. Independent Set

Similar documents
THE CIS PROBLEM AND RELATED RESULTS IN GRAPH THEORY

Monotone Circuit Lower Bounds from Resolution

Communication Lower Bounds via Query Complexity

A Composition Theorem for Conical Juntas

Nonnegative Rank vs. Binary Rank

Communication Lower Bounds via Critical Block Sensitivity

Nondeterminism LECTURE Nondeterminism as a proof system. University of California, Los Angeles CS 289A Communication Complexity

Partitions and Covers

The Alon-Saks-Seymour and Rank-Coloring Conjectures

Query-to-Communication Lifting for BPP

arxiv: v3 [cs.cc] 28 Jun 2015

On the tightness of the Buhrman-Cleve-Wigderson simulation

CS Foundations of Communication Complexity

Lecture 16: Communication Complexity

Approximation norms and duality for communication complexity lower bounds

Breaking the Rectangle Bound Barrier against Formula Size Lower Bounds

THE ALON-SAKS-SEYMOUR AND RANK-COLORING CONJECTURES. by Michael Tait

Communication Complexity

Chebyshev Polynomials, Approximate Degree, and Their Applications

Communication is bounded by root of rank

Post-selected classical query complexity

Complete problems for classes in PH, The Polynomial-Time Hierarchy (PH) oracle is like a subroutine, or function in

The Computational Complexity Column

Approximation Algorithms and Hardness of Approximation. IPM, Jan Mohammad R. Salavatipour Department of Computing Science University of Alberta

Structure of protocols for XOR functions

Polynomial-time Reductions

Kernelization Lower Bounds: A Brief History

Structure of protocols for XOR functions

CS Communication Complexity: Applications and New Directions

Lower bounds on the size of semidefinite relaxations. David Steurer Cornell

Polynomial Identity Testing and Circuit Lower Bounds

Information Complexity vs. Communication Complexity: Hidden Layers Game

CS Introduction to Complexity Theory. Lecture #11: Dec 8th, 2015

On the P versus NP intersected with co-np question in communication complexity

Non-Deterministic Time

Lifting Nullstellensatz to Monotone Span Programs over Any Field

A.Antonopoulos 18/1/2010

Communication vs information complexity, relative discrepancy and other lower bounds

Asymmetric Communication Complexity and Data Structure Lower Bounds

CSE200: Computability and complexity Space Complexity

CSC 2429 Approaches to the P vs. NP Question and Related Complexity Questions Lecture 2: Switching Lemma, AC 0 Circuit Lower Bounds

CS151 Complexity Theory. Lecture 13 May 15, 2017

CSC 5170: Theory of Computational Complexity Lecture 9 The Chinese University of Hong Kong 15 March 2010

Direct product theorem for discrepancy

14. Direct Sum (Part 1) - Introduction

arxiv: v1 [cs.cc] 12 Feb 2009

Complexity Theory VU , SS The Polynomial Hierarchy. Reinhard Pichler

Outline. Complexity Theory EXACT TSP. The Class DP. Definition. Problem EXACT TSP. Complexity of EXACT TSP. Proposition VU 181.

Essential facts about NP-completeness:

Complexity, P and NP

COMMUNICATION VS. COMPUTATION

Topics on Computing and Mathematical Sciences I Graph Theory (6) Coloring I

Lecture #14: NP-Completeness (Chapter 34 Old Edition Chapter 36) Discussion here is from the old edition.

Algebraic Problems in Computational Complexity

Direct product theorem for discrepancy

Circuits. Lecture 11 Uniform Circuit Complexity

Grothendieck Inequalities, XOR games, and Communication Complexity

COMP Analysis of Algorithms & Data Structures

Communication vs. Computation

Lecture 5: February 21, 2017

On The Hardness of Approximate and Exact (Bichromatic) Maximum Inner Product. Lijie Chen (Massachusetts Institute of Technology)

Umans Complexity Theory Lectures

Introduction to Advanced Results

Lecture 10: Learning DNF, AC 0, Juntas. 1 Learning DNF in Almost Polynomial Time

PROPERTY TESTING LOWER BOUNDS VIA COMMUNICATION COMPLEXITY

BBM402-Lecture 11: The Class NP

Evaluation of DNF Formulas

U.C. Berkeley CS278: Computational Complexity Professor Luca Trevisan 9/6/2004. Notes for Lecture 3

NP-COMPLETE PROBLEMS. 1. Characterizing NP. Proof

1 PSPACE-Completeness

ON QUANTUM-CLASSICAL EQUIVALENCE FOR COMPOSED COMMUNICATION PROBLEMS

Notes on Space-Bounded Complexity

Classes of Boolean Functions

Linear sketching for Functions over Boolean Hypercube

Limits to Approximability: When Algorithms Won't Help You. Note: Contents of today s lecture won t be on the exam

Lower Bounds for Dynamic Connectivity (2004; Pǎtraşcu, Demaine)

Lecture 20: conp and Friends, Oracles in Complexity Theory

The Polynomial Hierarchy

Information Complexity and Applications. Mark Braverman Princeton University and IAS FoCM 17 July 17, 2017

1.1 P, NP, and NP-complete

On the Isomorphism Problem for Decision Trees and Decision Lists

The Zero-Error Randomized Query Complexity of the Pointer Function

NP-Completeness. A language B is NP-complete iff B NP. This property means B is NP hard

A An Overview of Complexity Theory for the Algorithm Designer

CS 583: Algorithms. NP Completeness Ch 34. Intractability

Lecture 24: Randomized Complexity, Course Summary

arxiv: v1 [cs.cc] 6 Apr 2010

1. Introduction Recap

Lower Bound Techniques for Multiparty Communication Complexity

Notes on Space-Bounded Complexity

Cutting Plane Methods II

Multiparty Communication Complexity of Disjointness

CS 151 Complexity Theory Spring Solution Set 5

Lecture 20: Lower Bounds for Inner Product & Indexing

Lecture 15: Privacy Amplification against Active Attackers

Dynamic Programming on Trees. Example: Independent Set on T = (V, E) rooted at r V.

CS 395T Computational Learning Theory. Scribe: Mike Halcrow. x 4. x 2. x 6

Lecture notes on OPP algorithms [Preliminary Draft]

Transcription:

Lower Bounds for Clique vs. Independent Set Mika Göös University of Toronto Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

On page 6... Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

CIS problem [Yannakakis, STOC 88] G = ([n], E) Alice Bob Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

CIS problem [Yannakakis, STOC 88] G = ([n], E) Alice Clique x [n] of G Bob Independent set y [n] of G Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

CIS problem [Yannakakis, STOC 88] G = ([n], E) Alice Clique x [n] of G Bob Independent set y [n] of G Compute: CIS G (x, y) = x y Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Background Yannakakis s motivation: Size of LPs for the vertex packing polytope of G Breakthrough: [Fiorini et al., STOC ] Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Background Yannakakis s motivation: Size of LPs for the vertex packing polytope of G Breakthrough: [Fiorini et al., STOC ] Known bounds: G : NP cc (CIS G ) = log n (guess x y) Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Background Yannakakis s motivation: Size of LPs for the vertex packing polytope of G Breakthrough: [Fiorini et al., STOC ] Known bounds: G : G : NP cc (CIS G ) = log n (guess x y) conp cc (CIS G ) O(log n) Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Background Yannakakis s motivation: Size of LPs for the vertex packing polytope of G Breakthrough: [Fiorini et al., STOC ] Known bounds: G : G : NP cc (CIS G ) = log n (guess x y) conp cc (CIS G ) O(log n) Yannakakis s question: G : conp cc (CIS G ) O(log n)? Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Background Alon Saks Seymour conjecture: G : χ(g) bp(g) +? Yannakakis s question: G : conp cc (CIS G ) O(log n)? Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Background Alon Saks Seymour conjecture: G : χ(g) bp(g) +? [Huang Sudakov, ]: G : χ(g) bp(g) 6/5 Yannakakis s question: G : conp cc (CIS G ) O(log n)? Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Background Polynomial Alon Saks Seymour conjecture: G : χ(g) poly( bp(g))? Yannakakis s question: G : conp cc (CIS G ) O(log n)? Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Background Polynomial Alon Saks Seymour conjecture: G : χ(g) poly( bp(g))? [Alon Haviv] = = [Bousquet et al.] Yannakakis s question: G : conp cc (CIS G ) O(log n)? Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Background Polynomial Alon Saks Seymour conjecture: G : χ(g) poly( bp(g))? [Alon Haviv] = = [Bousquet et al.] Yannakakis s question: G : conp cc (CIS G ) O(log n)? Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Our result Main theorem G : conp cc (CIS G ) Ω(log.8 n) Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 4 / 4

Our result Main theorem G : conp cc (CIS G ) Ω(log.8 n) Prior bounds Measure Lower bound Reference P cc log n Kushilevitz, Linial, and Ostrovsky (999) conp cc 6/5 log n Huang and Sudakov () conp cc / log n Amano (4) conp cc log n Shigeta and Amano (4) Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 4 / 4

Our result Main theorem G : conp cc (CIS G ) Ω(log.8 n) Proof strategy: Query complexity Communication complexity Cf. lower bounds for log-rank [Nisan Wigderson, 995] Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 4 / 4

Models of communication F : X Y {, } Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 5 / 4

Models of communication NP cc Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 5 / 4

Models of communication UP cc Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 5 / 4

Models of communication CIS G is complete for UP cc : F CIS G UP cc (F) = UP cc (CIS G ) = log n Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 5 / 4

Proof strategy Restatement of Main theorem: F : X Y {, } conp cc (F) UP cc (F).8 Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 6 / 4

Proof strategy Restatement of Main theorem: F : X Y {, } conp cc (F) UP cc (F).8 Query separation: f : {, } n {, } conp dt ( f ) UP dt ( f ).8 Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 6 / 4

Proof strategy Restatement of Main theorem: F : X Y {, } conp cc (F) UP cc (F).8 Query separation: f : {, } n {, } conp dt ( f ) UP dt ( f ).8 Decision tree complexity measures: NP dt = DNF width = -certificate complexity conp dt = CNF width = -certificate complexity UP dt = Unambiguous DNF width Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 6 / 4

Proof strategy Restatement of Main theorem: F : X Y {, } conp cc (F) UP cc (F).8 Query separation: f : {, } n {, } conp dt ( f ) UP dt ( f ).8 Agenda: Step : Query separation Step : Simulation theorem [GLMWZ, 5] Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 6 / 4

Step : Query separation Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 7 / 4

Warm-up Example: Let f (x, x, x ) = iff x + x + x {, } UP dt ( f ) = because f x x x x x x conp dt ( f ) = because -input is fully sensitive Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 8 / 4

Warm-up Example: Let f (x, x, x ) = iff x + x + x {, } UP dt ( f ) = because f x x x x x x conp dt ( f ) = because -input is fully sensitive Recursive composition: f f ( ) := f ( ) f i+ ( ) := f ( f i ( ), f i ( ), f i ( )) x x x Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 8 / 4

Warm-up Example: Let f (x, x, x ) = iff x + x + x {, } UP dt ( f ) = because f x x x x x x conp dt ( f ) = because -input is fully sensitive Recursive composition: f f f f f ( ) := f ( ) f i+ ( ) := f ( f i ( ), f i ( ), f i ( )) x x x x 4 x 5 x 6 x 7 x 8 x 9 Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 8 / 4

Warm-up Example: Let f (x, x, x ) = iff x + x + x {, } UP dt ( f ) = because f x x x x x x conp dt ( f ) = because -input is fully sensitive Recursive composition: f f ( ) := f ( ) f i+ ( ) := f ( f i ( ), f i ( ), f i ( )) f x x x f x 4 x 5 x 6 f x 7 x 8 x 9 Hope: conp dt ( f i ) UP dt ( f i ) ( ) i Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 8 / 4

Warm-up Example: Problem! Let f (x, x, x ) = iff x + x + x {, } In order UP dt ( tof ) certify = because f i ( ) = f, x x x (should x x x be easy) might need conp dt to certify f ( f ) = because i ( ) = (should be hard) -input is fully sensitive Recursive composition: f f ( ) := f ( ) f i+ ( ) := f ( f i ( ), f i ( ), f i ( )) f x x x f x 4 x 5 x 6 f x 7 x 8 x 9 Hope: conp dt ( f i ) UP dt ( f i ) ( ) i Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 8 / 4

Warm-up Example: Problem! Let f (x, x, x ) = iff x + x + x {, } In order UP dt ( tof ) certify = because f i ( ) = f, x x x (should x x x be easy) might need conp dt to certify f ( f ) = because i ( ) = (should be hard) -input is fully sensitive Solution: Enlarge input/output alphabets Recursive composition: f x x x f f x 4 x 5 x 6 f : ({} Σ) n {} Σ f x 7 x 8 x 9 f ( ) := f ( ) f i+ ( ) := f ( f i ( ), f i ( ), f i ( )) Hope: conp dt ( f i ) UP dt ( f i ) ( ) i Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 8 / 4

Warm-up Example: Problem! Let f (x, x, x ) = iff x + x + x {, } In order UP dt ( tof ) certify = because f i ( ) = f, x x x (should x x x be easy) might need conp dt to certify f ( f ) = because i ( ) = (should be hard) -input is fully sensitive Solution: Enlarge input/output alphabets Recursive composition: f f : ({} Σ) n {} Σ f ( ) := f ( ) f i+ ( ) := f ( f i ( ), f i ( ), f i ( )) f f f Now: In order to certify f i ( ) = σ for σ Σ, only need to certify f i ( conp Hope: ) = σ for σ dt Σ( f i ) UP dt ( f x x x x 4 x 5 x 6 x 7 x 8 x i ) 9 ( ) i Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 8 / 4

Defining f Any two certificates in an UP dt decision tree intersect in variables = Finite projective planes! Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 9 / 4

Defining f Incidence ordering: Each node orders its incident edges using numbers from [] Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 9 / 4

Defining f Inputs to nodes: Pointer values from {} [] }{{} ( is a null pointer) =Σ Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 9 / 4

Defining f Defining f : ({} []) 7 {, } Say edge e is satisfied on input x iff all nodes v e point to e under x f (x) = iff x satisfies an edge Clearly UP dt ( f ) = Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 9 / 4

Defining f Problem! Certifying f ( ) = too easy! Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 9 / 4

Defining f Add input weights: f g 7 Gadget g is such that deciding if g( ) = i for i [] costs i queries Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 9 / 4

Defining f Add input weights: f g 7 Gadget g is such that deciding if g( ) = i for i [] costs i queries Else Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 9 / 4

Defining f Key properties: UP dt ( f g 7 ) = + + = 6 Certifying ( f g 7 )( ) = requires (# edges) = 7 queries Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 9 / 4

Defining f Key properties: UP dt ( f g 7 ) = + + = 6 Certifying ( f g 7 )( ) = requires (# edges) = 7 queries (Magic numerology: 57 6.8 ) Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 9 / 4

Defining f Key properties: UP dt ( f g 7 ) = + + = 6 Recursive composition Certifying ( f g 7 )( ) = Key trick: requires (# edges) = 7 queries ) From ({} Σ) (Magic n {, } numerology: 57 6.8 Construct ({} Σ) n {} {pointers} Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 9 / 4

Query separation: f : {, } n {, } conp dt ( f ) UP dt ( f ).8 Step : Simulation theorem from Rectangles Are Nonnegative Juntas Mika Göös, Shachar Lovett, Raghu Meka, Thomas Watson, and David Zuckerman (STOC 5) Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Composed functions f g n f Compose with gn f z z z z 4 z 5 g g g g g x y x y x y x 4 y 4 x 5 y 5 Examples: Set-disjointness: OR AND n Inner-product: XOR AND n Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Composed functions f g n f Compose with gn f z z z z 4 z 5 g g g g g x y x y x y x 4 y 4 x 5 y 5 Examples: In general: We choose: Set-disjointness: OR AND n Inner-product: XOR AND n g : {, } b {, } b {, } is a small gadget Alice holds x {, } bn Bob holds y {, } bn g = inner-product with b = Θ(log n) bits per party Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Approximation by juntas Conical d-junta: Nonnegative combination of d-conjunctions (Example:.4 z z +.66 z z +.5 z z ) Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Approximation by juntas Conical d-junta: Nonnegative combination of d-conjunctions (Example:.4 z z +.66 z z +.5 z z ) Main Structure Theorem: Suppose Π is cost-d randomised protocol for f g n Then there exists a conical d-junta h s.t. z dom f : Pr [ Π(x, y) accepts ] h(z) (x,y) (g n ) (z) Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Approximation by juntas Conical d-junta: Nonnegative combination of d-conjunctions (Example:.4 z z +.66 z z +.5 z z ) Main Structure Theorem: Suppose Π is cost-d randomised protocol for f g n Then there exists a conical d-junta h s.t. z dom f : Pr [ Π(x, y) accepts ] h(z) (x,y) (g n ) (z) Cf. Polynomial approximation [Razborov, Sherstov, Shi Zhu,... ]: Approximate poly-degree of AND = Θ( n) Approximate junta-degree of AND = Θ(n) Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Corollaries Simulation for NP: NP cc ( f g n ) = Θ(NP dt ( f ) b)... recall b = Θ(log n) Conical d-junta:.4 z z +.66 z z +.5 z z d-dnf: z z z z z z Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Corollaries Simulation for NP: NP cc ( f g n ) = Θ(NP dt ( f ) b)... recall b = Θ(log n) Trivially: UP cc ( f g n ) O(UP dt ( f ) b) Main theorem follows! Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Corollaries Simulation for NP: NP cc ( f g n ) = Θ(NP dt ( f ) b)... recall b = Θ(log n) NP MA Also covered! P BPP WAPP SBP PostBPP PP smooth rectangle corruption ext. discrepancy discrepancy Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 / 4

Summary Main result G : conp cc (CIS G ) Ω(log.8 n) Open problems Better separation for conp dt vs. UP dt? Simulation theorems for new models (e.g., BPP) Improve gadget size down to b = O() (Would give new proof of Ω(n) bound for set-disjointness) Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 4 / 4

Summary Main result G : conp cc (CIS G ) Ω(log.8 n) Open problems Better separation for conp dt vs. UP dt? Simulation theorems for new models (e.g., BPP) Improve gadget size down to b = O() (Would give new proof of Ω(n) bound for set-disjointness) Cheers! Mika Göös (Univ. of Toronto) Clique vs. Independent Set rd February 5 4 / 4