Toda s theorem in bounded arithmetic with parity quantifiers and bounded depth proof systems with parity gates

Similar documents
Collapsing modular counting in bounded arithmetic and constant depth propositional proofs

Formalizing Randomized Matching Algorithms

On Transformations of Constant Depth Propositional Proofs

Bounded Arithmetic, Constant Depth Proofs, and st-connectivity. Sam Buss Department of Mathematics U.C. San Diego

Unprovability of circuit upper bounds in Cook s theory PV

A polytime proof of correctness of the Rabin-Miller algorithm from Fermat s Little Theorem

Provability of weak circuit lower bounds

Proofs with monotone cuts

Bounded Arithmetic, Expanders, and Monotone Propositional Proofs

Regular Resolution Lower Bounds for the Weak Pigeonhole Principle

Expander Construction in VNC 1

I. Introduction to NP Functions and Local Search

Random Resolution Refutations

Model theory of bounded arithmetic with applications to independence results. Morteza Moniri

On extracting computations from propositional proofs (a survey)

INSTITUTE OF MATHEMATICS THE CZECH ACADEMY OF SCIENCES. Random resolution refutations. Pavel Pudlák Neil Thapen

Lecture 11: Measuring the Complexity of Proofs

Algorithms for Satisfiability beyond Resolution.

Foundations of Proof Complexity: Bounded Arithmetic and Propositional Translations. Stephen Cook and Phuong Nguyen c Copyright 2004, 2005, 2006

Proof Complexity and Computational Complexity

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

LIFTING LOWER BOUNDS FOR TREE-LIKE PROOFS

NP search problems in low fragments of bounded arithmetic

Uniform Proofs of ACC Representations

RESOLUTION OVER LINEAR EQUATIONS AND MULTILINEAR PROOFS

Lecture 13: Polynomial-Size Frege Proofs of the Pigeonhole Principle

1 PSPACE-Completeness

Upper and Lower Bounds for Tree-like. Cutting Planes Proofs. Abstract. In this paper we study the complexity of Cutting Planes (CP) refutations, and

On the Automatizability of Resolution and Related Propositional Proof Systems

Complexity of propositional proofs: Some theory and examples

The Strength of Multilinear Proofs

Organization. Informal introduction and Overview Informal introductions to P,NP,co-NP and themes from and relationships with Proof complexity

Resolution and the Weak Pigeonhole Principle

LOGICAL STRENGTH OF COMPLEXITY THEORY AND A FORMALIZATION OF THE PCP THEOREM IN BOUNDED ARITHMETIC

Proof Complexity of Quantified Boolean Formulas

Stanford University CS254: Computational Complexity Handout 8 Luca Trevisan 4/21/2010

Some open problems in bounded arithmetic and propositional proof complexity (research proposal paper)

Chapter 1 : The language of mathematics.

Circuit principles and weak pigeonhole variants

The provably total NP search problems of weak second order bounded arithmetic

1 Circuit Complexity. CS 6743 Lecture 15 1 Fall Definitions

Total NP Functions I: Complexity and Reducibility

A The NP Search Problems of Frege and Extended Frege Proofs

INSTITUTE OF MATHEMATICS THE CZECH ACADEMY OF SCIENCES. Incompleteness in the finite domain. Pavel Pudlák

arxiv: v1 [cs.cc] 10 Aug 2007

An Introduction to Proof Complexity, Part II.

PROOF COMPLEXITY IN ALGEBRAIC SYSTEMS AND BOUNDED DEPTH FREGE SYSTEMS WITH MODULAR COUNTING

15-855: Intensive Intro to Complexity Theory Spring Lecture 7: The Permanent, Toda s Theorem, XXX

On the computational content of intuitionistic propositional proofs

On Proofs About Threshold Circuits and Counting Hierarchies (Extended Abstract)

Logic: The Big Picture

A Tight Karp-Lipton Collapse Result in Bounded Arithmetic

Definition: Alternating time and space Game Semantics: State of machine determines who

INSTITUTE OF MATHEMATICS THE CZECH ACADEMY OF SCIENCES. A trade-off between length and width in resolution. Neil Thapen

Characterizing Propositional Proofs as Non-Commutative Formulas

Quantified Propositional Calculus and a Second-Order Theory for NC 1

Lecture 7: Recursive saturation

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

Partial Collapses of the Σ 1 Complexity Hierarchy in Models for Fragments of Bounded Arithmetic

Gödel s Incompleteness Theorems

Effectively Polynomial Simulations

Umans Complexity Theory Lectures

Sample Problems for all sections of CMSC250, Midterm 1 Fall 2014

Definition: Alternating time and space Game Semantics: State of machine determines who

This is logically equivalent to the conjunction of the positive assertion Minimal Arithmetic and Representability

Essential facts about NP-completeness:

Arithmetical Hierarchy

Arithmetical Hierarchy

Packet #2: Set Theory & Predicate Calculus. Applied Discrete Mathematics

Comp487/587 - Boolean Formulas

ALGEBRAIC PROOFS OVER NONCOMMUTATIVE FORMULAS

Logical Closure Properties of Propositional Proof Systems

DRAFT. Complexity of counting. Chapter 8

Logic. Propositional Logic: Syntax

Part I: Propositional Calculus

Notes for Lecture 3... x 4

Propositional and Predicate Logic - V

Computational Complexity: A Modern Approach. Draft of a book: Dated January 2007 Comments welcome!

Logical Closure Properties of Propositional Proof Systems

STUDIES IN ALGEBRAIC AND PROPOSITIONAL PROOF COMPLEXITY. Iddo Tzameret

Victoria Gitman and Thomas Johnstone. New York City College of Technology, CUNY

CSCI3390-Lecture 18: Why is the P =?NP Problem Such a Big Deal?

The quantifier complexity of polynomial-size iterated definitions in first-order logic

Notes on Complexity Theory Last updated: October, Lecture 6

NP-Completeness Part II

Algebraic Proof Complexity: Progress, Frontiers and Challenges

Tecniche di Verifica. Introduction to Propositional Logic

Krivine s Intuitionistic Proof of Classical Completeness (for countable languages)

2 BEAME AND PITASSI systems of increasing complexity. This program has several important side eects. First, standard proof systems are interesting in

Non-automatizability of bounded-depth Frege proofs

Lecture 7: The Polynomial-Time Hierarchy. 1 Nondeterministic Space is Closed under Complement

Bounded Arithmetic vs. Propositional Proof Systems vs. Complexity Classes (depth oral survey)

Between proof theory and model theory Three traditions in logic: Syntactic (formal deduction)

CS 151 Complexity Theory Spring Solution Set 5

Proof Theory and Subsystems of Second-Order Arithmetic

Parallel Time and Proof Complexity. Klaus Aehlig

2 Evidence that Graph Isomorphism is not NP-complete

Propositional and Predicate Logic. jean/gbooks/logic.html

Propositional Logic: Models and Proofs

Math 267a - Propositional Proof Complexity. Lecture #1: 14 January 2002

Transcription:

1 / 17 Toda s theorem in bounded arithmetic with parity quantifiers and bounded depth proof systems with parity gates Leszek Kołodziejczyk University of Warsaw/UCSD (joint work with Sam Buss and Konrad Zdanowski) Logical Approaches to Barriers in Complexity II Cambridge, March 01

/ 17 Introduction Major problem in propositional proof complexity: lower bounds (ideally, exponential) on bounded depth proofs with mod gates: (φ 1,..., φ n ) = an odd number of φ i have value 1. Bounded depth Frege with mod gates = AC 0 []-Frege. Related problem in bounded arithmetic: interesting independence result for T ( )(α), bounded arithmetic with a parity quantifier.

/ 17 Introduction Major problem in propositional proof complexity: lower bounds (ideally, exponential) on bounded depth proofs with mod gates: (φ 1,..., φ n ) = an odd number of φ i have value 1. Bounded depth Frege with mod gates = AC 0 []-Frege. Related problem in bounded arithmetic: interesting independence result for T ( )(α), bounded arithmetic with a parity quantifier. More modest aim: better understanding of AC 0 []-Frege and T ( )(α) (and analogues for prime p ).

3 / 17 Toda s Theorem (A version of) Toda s Theorem: PH( ), the polynomial hierarchy with a parity quantifier, collapses to BP P. Observation: the relativized version of this can be seen as a collapse of AC 0 [] circuits to a very simple form, with quasipolynomial increase in size.

3 / 17 Toda s Theorem (A version of) Toda s Theorem: PH( ), the polynomial hierarchy with a parity quantifier, collapses to BP P. Observation: the relativized version of this can be seen as a collapse of AC 0 [] circuits to a very simple form, with quasipolynomial increase in size. Can something similar be done for AC 0 []-Frege proofs?

4 / 17 Collapsing AC 0 []-Frege? Maciel-Pitassi 1998: simulation of AC 0 []-Frege by proofs of simple form, but the simulating system has exact counting (threshold) gates. New development since then (Jeřábek 004-9): bounded arithmetic has reasonable notions of approximate cardinality and probabilistic complexity classes. So: why not try to prove Toda in bounded arithmetic with parity quantifiers, and see what that says about AC 0 []-Frege proofs?

5 / 17 Plan for rest of talk There won t be any really interesting proofs in this talk. There won t even be too many pictures/diagrams. So, the above is offered as a form of compensation.

6 / 17 Bounded arithmetic: a very quick review Σ b i class of arithmetic formulas corresponding to Σ p i. T i induction for Σb i formulas. T = i Ti. PV induction for polytime properties ( right notion of T 0 ). For new predicate α (oracle), Σ b i (α) and Ti (α) can be defined. Paris-Wilkie translation: translates arithmetic formulas (with α) into families of propositional formulas, and proofs in T i (α) into uniform families of fixed-depth quasipolynomial size proofs. (atoms in α variables, quantifiers /, etc.)

7 / 17 Approximate counting in bounded arithmetic swphp(γ) surjective WPHP for function class Γ: no function f Γ is surjection a a(1 + 1/(log a)). (in many contexts, ruling out a a suffices.) APC 1 = PV + swphp(fp). APC = T 1 + swphp(fpnp ). APC 1 is contained in T. It can approximate the size of polytime set X n up to 1/poly(n) fraction of n. APC can do the same for X P NP, while for X NP it finds surjections witnessing m X m + m/polylog(m). It is contained in T 3.

8 / 17 Bounded arithmetic with a parity quantifier Two ways of adding the new quantifier: T ( ): add x < y to the usual language, induction available for all bounded formulas. T i, P : allow x < y only in front of polytime formulas. T i, P has i induction.

9 / 17 Toda s Theorem in APC P L BP P if for some polytime functions u(x), f (x, r), x L Pr r<u(x) [f (x, r) / SAT] < 1/4, x / L Pr r<u(x) [f (x, r) SAT] < 1/4, where probabilities stated using approximate counting. Theorem Every formula can be assigned a BP P representation which is provably correct in APC P. As a consequence, T ( ) is conservative over APC P. The theorem smoothly relativizes to a new oracle α.

10 / 17 Toda in APC P : comments on proof Essentially a formalization of the textbook proof. Induction on formula complexity, some technicalities involved. The base case uses a version of the Valiant-Vazirani Theorem: SAT is probabilistically reducible to Unique-SAT. One point in the proof of V-V: given propositional formula φ, if S is the set of satisfying assignments for φ, then for some k, This seems to need APC P. k S k+1.

11 / 17 Back to propositional proofs The proof system PCK i : lines are cedents of / formulas of depth i with literals replaced by low-degree polynomials over F ( low = logarithmic in the proof size). Intended meaning of the PCK 1 line f 1, f f 3 f 4 is: f 1 is 0 or f, f 3, f 4 all are. So, constant 1 plays the role of. (Btw, low-degree polynomials s of small conjunctions, it s just that algebraic rules are sometimes less clumsy than boolean. So PCK i is a subsystem of AC 0 []-Frege.)

1 / 17 Propositional proofs: rules ψ, ψ Axiom Γ, ψ i, where i I Γ, i I ψ i ( ) Γ, ψ i, all i I Γ, i I ψ i ( ) Γ Γ, (weakening) Γ, ψ Γ, ψ (cut) Γ Γ, f Γ, fg ( ) Γ, f Γ, g (+) Γ, f + g ( is DeMorgan negation, and f is 1 + f.)

Propositional proofs: correspondence An arithmetic formula A(x, α) has propositional translations A n, with variables for bits of α, meaning A(n, α) holds. Theorem A provable in A have qpoly size refutations in i (α) T i, P (α) i+1 (α) Ti, P (α) 1 (α) T 1, P (α) 1 (α) PV P (α) 13 / 17

Propositional proofs: correspondence An arithmetic formula A(x, α) has propositional translations A n, with variables for bits of α, meaning A(n, α) holds. Theorem A provable in A have qpoly size refutations in i (α) T i, P (α) PCK i, treelike PCK i 1 i+1 (α) Ti, P (α) treelike PCK i 1 1 (α) T 1, P (α) 1 (α) PV P (α) 13 / 17

Propositional proofs: correspondence An arithmetic formula A(x, α) has propositional translations A n, with variables for bits of α, meaning A(n, α) holds. Theorem A provable in A have qpoly size refutations in i (α) T i, P (α) PCK i, treelike PCK i 1 i+1 (α) Ti, P (α) treelike PCK i 1 1 (α) T 1, P (α) polylog degree Polynomial Calculus 1 (α) PV P (α) 13 / 17

Propositional proofs: correspondence An arithmetic formula A(x, α) has propositional translations A n, with variables for bits of α, meaning A(n, α) holds. Theorem A provable in A have qpoly size refutations in i (α) T i, P (α) PCK i, treelike PCK i 1 i+1 (α) Ti, P (α) treelike PCK i 1 1 (α) T 1, P (α) polylog degree Polynomial Calculus 1 (α) PV P (α) polylog degree Nullstellensatz 13 / 17

14 / 17 Propositional proofs: collapse Corollary For proofs of simple enough formulas ( small ), AC 0 []-Frege is quasipolynomially simulated by PCK 1. Proof. by conservativity, T 3 (α) proves reflection for AC0 []-Frege: every provable formula is true. so PCK 1 refutes Reflection. by substituting bits of an actual AC 0 []-Frege proof of φ, we get a PCK 1 refutation of φ is false. modulo cosmetic changes, that is a refutation of φ.

15 / 17 Propositional proofs: collapse (cont d) Corollary For proofs of simple enough formulas ( small ), AC 0 []-Frege is quasipolynomially simulated by PCK 1.

15 / 17 Propositional proofs: collapse (cont d) Corollary For proofs of simple enough formulas ( small ), AC 0 []-Frege is quasipolynomially simulated by PCK 1. Using conservativity over APC P instead of T 3, P, we get: Corollary For proofs of simple enough formulas ( small ), AC 0 []-Frege is quasipolynomially simulated by treelike PCK 0 extended by axioms corresponding to swphp(fp NP ( P)). (Using partial conservativity of swphp over so-called retraction WPHP, one could even replace treelike PCK 0 by polylog degree Polynomial Calculus, but the extra axioms become less natural.)

16 / 17 The picture right now T 1, P (α) polylog degree PC, treelike PCK 1 APC P (α) treelike PCK 1 + swphp PV P (α) polylog degree NS APC P 1 (α) polylog degree NS + swphp

The picture right now T 1, P (α) polylog degree PC, treelike PCK 1 APC P (α) treelike PCK 1 + swphp PV P (α) polylog degree NS APC P 1 (α) polylog degree NS + swphp The pigeonhole PHP n+1 n (α) is independent from: T 1, P (α), by known Polynomial Calculus lower bounds, PV P (α) + swphp(fp(α)), by combining Nullstellensatz lower bounds with switching lemma techniques. Seems within reach to extend this to T 1, P (α) + swphp(fp NP (α)), but swphp for functions involving P seems very difficult to deal with. 16 / 17

17 / 17 Problems with an approach to lower bounds Let φ be a small formula. Want to show: φ has no refutations in low degree Nullstellensatz + swphp(fp( P)) axioms. The swphp axiom says c < t not in F([0, t)) ; has polylog many new variables for bits of c. For any assignment to the φ variables, almost all assignments to the new variables make the axiom true. So maybe...

17 / 17 Problems with an approach to lower bounds Let φ be a small formula. Want to show: φ has no refutations in low degree Nullstellensatz + swphp(fp( P)) axioms. The swphp axiom says c < t not in F([0, t)) ; has polylog many new variables for bits of c. For any assignment to the φ variables, almost all assignments to the new variables make the axiom true. So maybe... However: take a suitably constructed low degree approximation φ to φ. This has polylog many new variables, and for any assignment to the old variables, almost all assignments to the new variables make φ true. But φ joined with φ is refutable in low degree Nullstellensatz!