Shamir s Theorem. Johannes Mittmann. Technische Universität München (TUM)

Similar documents
IP=PSPACE. Adi Shamir. Applied Mathematics Department The Weizmann Institute of Science Rehovot, Israel

Lecture 19: Interactive Proofs and the PCP Theorem

Lecture 14: IP = PSPACE

1 PSPACE-Completeness

Complexity Theory. Jörg Kreiker. Summer term Chair for Theoretical Computer Science Prof. Esparza TU München

Interactive Proofs 1

The Proof of IP = P SP ACE

Introduction to Interactive Proofs & The Sumcheck Protocol

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.

2 Evidence that Graph Isomorphism is not NP-complete

Reasoning with Quantified Boolean Formulas

Theorem 11.1 (Lund-Fortnow-Karloff-Nisan). There is a polynomial length interactive proof for the

Lecture 8. MINDNF = {(φ, k) φ is a CNF expression and DNF expression ψ s.t. ψ k and ψ is equivalent to φ}

POLYNOMIAL SPACE QSAT. Games. Polynomial space cont d

Database Theory VU , SS Complexity of Query Evaluation. Reinhard Pichler

Lecture : PSPACE IP

Algorithms & Complexity II Avarikioti Zeta

Quantified Boolean Formulas Part 1

Author s address: Applied Mathematics Department, The Weizmann Institute of Science, Rehovot,

CS151 Complexity Theory. Lecture 14 May 17, 2017

LOGIC PROPOSITIONAL REASONING

198:538 Complexity of Computation Lecture 16 Rutgers University, Spring March 2007

Comp487/587 - Boolean Formulas

CSE200: Computability and complexity Space Complexity

Lecture 8: Complete Problems for Other Complexity Classes

Propositional Logic: Evaluating the Formulas

CS151 Complexity Theory. Lecture 13 May 15, 2017

CSE 555 HW 5 SAMPLE SOLUTION. Question 1.

Umans Complexity Theory Lectures

Space is a computation resource. Unlike time it can be reused. Computational Complexity, by Fu Yuxi Space Complexity 1 / 44

Ταουσάκος Θανάσης Αλγόριθμοι και Πολυπλοκότητα II 7 Φεβρουαρίου 2013

Lecture 26. Daniel Apon

Chapter 9. PSPACE: A Class of Problems Beyond NP. Slides by Kevin Wayne Pearson-Addison Wesley. All rights reserved.

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

Computational Complexity IV: PSPACE

Quantified Boolean Formulas: Complexity and Expressiveness

Part 1: Propositional Logic

9. PSPACE. PSPACE complexity class quantified satisfiability planning problem PSPACE-complete

CS156: The Calculus of Computation

CSC 2429 Approaches to the P versus NP Question. Lecture #12: April 2, 2014

First-Order Logic. 1 Syntax. Domain of Discourse. FO Vocabulary. Terms

Lecture 12: Interactive Proofs

Real Interactive Proofs for VPSPACE

Introduction to Complexity Theory. Bernhard Häupler. May 2, 2006

Rational Proofs with Multiple Provers. Jing Chen, Samuel McCauley, Shikha Singh Department of Computer Science

Interactive protocols & zero-knowledge

Space and Nondeterminism

SOLUTION: SOLUTION: SOLUTION:

Informal Statement Calculus

9. PSPACE 9. PSPACE. PSPACE complexity class quantified satisfiability planning problem PSPACE-complete

Conjunction: p q is true if both p, q are true, and false if at least one of p, q is false. The truth table for conjunction is as follows.

6.841/18.405J: Advanced Complexity Wednesday, February 12, Lecture Lecture 3

COMPLEXITY THEORY. Lecture 17: The Polynomial Hierarchy. TU Dresden, 19th Dec Markus Krötzsch Knowledge-Based Systems

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

De Morgan s a Laws. De Morgan s laws say that. (φ 1 φ 2 ) = φ 1 φ 2, (φ 1 φ 2 ) = φ 1 φ 2.

Lecture 11: Proofs, Games, and Alternation

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

6.841/18.405J: Advanced Complexity Wednesday, April 2, Lecture Lecture 14

Outline. Complexity Theory. Introduction. What is abduction? Motivation. Reference VU , SS Logic-Based Abduction

Lecture 20: PSPACE. November 15, 2016 CS 1010 Theory of Computation

Worst-Case Upper Bound for (1, 2)-QSAT

Interactive protocols & zero-knowledge

A brief introduction to Logic. (slides from

Bounded-width QBF is PSPACE-complete

CS Lecture 29 P, NP, and NP-Completeness. k ) for all k. Fall The class P. The class NP

A New 3-CNF Transformation by Parallel-Serial Graphs 1

Complexity Theory. Knowledge Representation and Reasoning. November 2, 2005

Technische Universität München Summer term 2010 Theoretische Informatik August 2, 2010 Dr. J. Kreiker / Dr. M. Luttenberger, J. Kretinsky SOLUTION

Principles of Knowledge Representation and Reasoning

CSE 1400 Applied Discrete Mathematics Definitions

Definition 1. NP is a class of language L such that there exists a poly time verifier V with

Validating QBF Invalidity in HOL4

x 2 +2 x 2 x 3 2x 2 +2x +1 (mod 5).

Lecture 9: Polynomial-Time Hierarchy, Time-Space Tradeoffs

Winter 2011 Josh Benaloh Brian LaMacchia

CS 580: Algorithm Design and Analysis. Jeremiah Blocki Purdue University Spring 2018

Lecture 17: Interactive Proof Systems

Polynomial Hierarchy

Approximate Verification

Real Interactive Proofs for VPSPACE

INAPPROX APPROX PTAS. FPTAS Knapsack P

Lecture 22: PSPACE

PSPACE COMPLETENESS TBQF. THURSDAY April 17

Tecniche di Verifica. Introduction to Propositional Logic

First-Order Logic First-Order Theories. Roopsha Samanta. Partly based on slides by Aaron Bradley and Isil Dillig

1 First-order logic. 1 Syntax of first-order logic. 2 Semantics of first-order logic. 3 First-order logic queries. 2 First-order query evaluation

6.045 Final Exam Solutions

THE BLACK-BOX QUERY COMPLEXITY OF POLYNOMIAL SUMMATION

Complexity. Complexity Theory Lecture 3. Decidability and Complexity. Complexity Classes

Solution of Exercise Sheet 12

Chapter 2. Reductions and NP. 2.1 Reductions Continued The Satisfiability Problem (SAT) SAT 3SAT. CS 573: Algorithms, Fall 2013 August 29, 2013

Theory of Computation Space Complexity. (NTU EE) Space Complexity Fall / 1

CMSC Discrete Mathematics SOLUTIONS TO SECOND MIDTERM EXAM November, 2005

Propositional Logic: Models and Proofs

Chapter 1 : The language of mathematics.

Topic 1: Propositional logic

Notes on Complexity Theory Last updated: October, Lecture 6

1 Randomized Computation

-bit integers are all in ThC. Th The following problems are complete for PSPACE NPSPACE ATIME QSAT, GEOGRAPHY, SUCCINCT REACH.

Transcription:

IP = PSPACE Shamir s Theorem Johannes Mittmann Technische Universität München (TUM) 4 th Joint Advanced Student School (JASS) St. Petersburg, April 2 12, 2006 Course 1: Proofs and Computers Johannes Mittmann IP = PSPACE JASS 2006 1

Outline 1 Introduction 2 Polynomial Space Quantified Satifiability PSPACE-Completeness 3 Shamir s Theorem Arithmetization Reduction to a Finite Field Polynomials and Simple Expressions The Interactive Protocol Johannes Mittmann IP = PSPACE JASS 2006 2

Introduction History Papadimitriou (1983): IP PSPACE. Toda (1989): PH P #P. Nisan (Nov. 27, 1989): PH MIP. Lund, Fortnow, Karloff, Nisan (Dec. 13, 1989): PH IP. Shamir (Dec. 26, 1989): PSPACE IP. Figure: Adi Shamir Johannes Mittmann IP = PSPACE JASS 2006 3

Polynomial Space Polynomial Space PSPACE = k>0 SPACE(n k ). L NL P NP PSPACE. Johannes Mittmann IP = PSPACE JASS 2006 4

Polynomial Space Quantified Satifiability Quantified Boolean Formulas Definition Let X = {x 1, x 2,... } be an alphabet of Boolean variables. They can take the two truth values true and false. A quantified Boolean expression (QBF) φ is defined inductively by 1 a Boolean variable x i, or an expression of the form 2 φ 1 3 φ 1 φ 2 4 φ 1 φ 2 (negation), (disjunction), (conjunction), 5 x i φ 1 (existential quantification), 6 x i φ 1 (universal quantification), where φ 1 and φ 2 are quantified Boolean expressions. Johannes Mittmann IP = PSPACE JASS 2006 5

Polynomial Space Quantified Satifiability Proposition Let φ and ψ be QBFs. Then 1 (φ ψ) φ ψ. (De Morgan s Laws) 2 (φ ψ) φ ψ. 3 ( x i φ) x i φ. 4 ( x i φ) x i φ. 5 ( φ) φ. Johannes Mittmann IP = PSPACE JASS 2006 6

Polynomial Space Quantified Satifiability Proposition Let φ and ψ be QBFs. Then 1 (φ ψ) φ ψ. (De Morgan s Laws) 2 (φ ψ) φ ψ. 3 ( x i φ) x i φ. 4 ( x i φ) x i φ. 5 ( φ) φ. 6 x i (φ ψ) ( x i φ) ( x i ψ). 7 x i (φ ψ) ( x i φ) ( x i ψ). Johannes Mittmann IP = PSPACE JASS 2006 6

Polynomial Space Quantified Satifiability Proposition Let φ and ψ be QBFs. Then 1 (φ ψ) φ ψ. (De Morgan s Laws) 2 (φ ψ) φ ψ. 3 ( x i φ) x i φ. 4 ( x i φ) x i φ. 5 ( φ) φ. 6 x i (φ ψ) ( x i φ) ( x i ψ). 7 x i (φ ψ) ( x i φ) ( x i ψ). 8 If x i does not appear free in ψ, x i (φ ψ) ( x i φ) ψ. 9 If x i does not appear free in ψ, x i (φ ψ) ( x i φ) ψ. Johannes Mittmann IP = PSPACE JASS 2006 6

Polynomial Space Quantified Satifiability Proposition Let φ and ψ be QBFs. Then 1 (φ ψ) φ ψ. (De Morgan s Laws) 2 (φ ψ) φ ψ. 3 ( x i φ) x i φ. 4 ( x i φ) x i φ. 5 ( φ) φ. 6 x i (φ ψ) ( x i φ) ( x i ψ). 7 x i (φ ψ) ( x i φ) ( x i ψ). 8 If x i does not appear free in ψ, x i (φ ψ) ( x i φ) ψ. 9 If x i does not appear free in ψ, x i (φ ψ) ( x i φ) ψ. 10 If xj does not appear in φ, x i φ x j φ[x i x j ]. Johannes Mittmann IP = PSPACE JASS 2006 6

Polynomial Space Quantified Satifiability Prenex Normal Form φ = x 1 x 2 x 3... Q n x n ψ, Q n {, }. Johannes Mittmann IP = PSPACE JASS 2006 7

Polynomial Space Quantified Satifiability Prenex Normal Form φ = x 1 x 2 x 3... Q n x n ψ, Q n {, }. Proposition Any QBF φ can be transformed to an equivalent one in prenex normal form. Johannes Mittmann IP = PSPACE JASS 2006 7

Polynomial Space Quantified Satifiability Quantified Satisfiability QSAT = { φ : φ is a true QBF in prenex CNF }. Johannes Mittmann IP = PSPACE JASS 2006 8

Polynomial Space PSPACE-Completeness PSPACE-Completeness Theorem (Stockmeyer/Meyer 1973) QSAT is PSPACE-complete. Johannes Mittmann IP = PSPACE JASS 2006 9

Polynomial Space PSPACE-Completeness PSPACE-Completeness Theorem (Stockmeyer/Meyer 1973) QSAT is PSPACE-complete. Proof. It suffices to show: 1 QSAT PSPACE. 2 For all L PSPACE : L log QSAT. Johannes Mittmann IP = PSPACE JASS 2006 9

Polynomial Space PSPACE-Completeness Proof sketch of (1). φ = x 1 x 2 x 3... Q n x n ψ(x 1,..., x n ) x 1 = false x 1 = true x 2 x 3... ψ(0, x 2,... ) x 2 x 3... ψ(1, x 2,... ). x 2 = false x 2 = true.. x 3... ψ(0, 0, x 3,... ) x 3... ψ(0, 1, x 3,... ). Q n Q n... ψ(0,..., 0) ψ(0,..., 1) ψ(0,..., 1, 0) ψ(0,..., 1, 1) Johannes Mittmann IP = PSPACE JASS 2006 10

Polynomial Space PSPACE-Completeness Proof sketch of (1). φ = x 1 x 2 x 3... Q n x n ψ(x 1,..., x n ) x 1 = false x 1 = true x 2 x 3... ψ(0, x 2,... ) x 2 x 3... ψ(1, x 2,... ). x 2 = false x 2 = true.. x 3... ψ(0, 0, x 3,... ) x 3... ψ(0, 1, x 3,... ).... ψ(0,..., 0) ψ(0,..., 1) ψ(0,..., 1, 0) ψ(0,..., 1, 1) Johannes Mittmann IP = PSPACE JASS 2006 10

Polynomial Space PSPACE-Completeness Algorithm: Truth(φ) 1: if φ is quantifier-free then 2: return truth value of φ 3: end if 4: denote φ = Q 1 x 1... Q n x n ψ(x 1,..., x n ) 5: b 0 Truth(Q 2 x 2... Q n x n ψ(false, x 2,..., x n )) 6: b 1 Truth(Q 2 x 2... Q n x n ψ(true, x 2,..., x n )) 7: if Q 1 = then 8: return b 0 b 1 9: else 10: return b 0 b 1 11: end if Johannes Mittmann IP = PSPACE JASS 2006 11

Polynomial Space PSPACE-Completeness Proof sketch of (2). Let L PSPACE. Johannes Mittmann IP = PSPACE JASS 2006 12

Polynomial Space PSPACE-Completeness Proof sketch of (2). Let L PSPACE. L is decidable by a Turing machine M. Johannes Mittmann IP = PSPACE JASS 2006 12

Polynomial Space PSPACE-Completeness Proof sketch of (2). Let L PSPACE. L is decidable by a Turing machine M. For input x consider the configuration graph of M. Johannes Mittmann IP = PSPACE JASS 2006 12

Polynomial Space PSPACE-Completeness Proof sketch of (2). Let L PSPACE. L is decidable by a Turing machine M. For input x consider the configuration graph of M. 2 m configurations, where m = O(n k ). Johannes Mittmann IP = PSPACE JASS 2006 12

Polynomial Space PSPACE-Completeness Proof sketch of (2). Let L PSPACE. L is decidable by a Turing machine M. For input x consider the configuration graph of M. 2 m configurations, where m = O(n k ). Reachability method: ψ i (X, Y) is true configuration Y can be reached from configuration X in 2 i steps, for i = 0,..., m. Johannes Mittmann IP = PSPACE JASS 2006 12

Polynomial Space PSPACE-Completeness Proof sketch of (2). Let L PSPACE. L is decidable by a Turing machine M. For input x consider the configuration graph of M. 2 m configurations, where m = O(n k ). Reachability method: ψ i (X, Y) is true configuration Y can be reached from configuration X in 2 i steps, for i = 0,..., m. Required QBF: ψ m (A, B). Johannes Mittmann IP = PSPACE JASS 2006 12

Polynomial Space PSPACE-Completeness Proof sketch of (2). ψ 0 (A, B) can be written in DNF. Johannes Mittmann IP = PSPACE JASS 2006 13

Polynomial Space PSPACE-Completeness Proof sketch of (2). ψ 0 (A, B) can be written in DNF. Bad idea: ψ i+1 = Z [ ψ i (A, Z) ψ i (Z, B) ]. Johannes Mittmann IP = PSPACE JASS 2006 13

Polynomial Space PSPACE-Completeness Proof sketch of (2). ψ 0 (A, B) can be written in DNF. Bad idea: ψ i+1 = Z [ ψ i (A, Z) ψ i (Z, B) ]. Savitch s trick: ψ i+1 = Z X Y [( (X = A Y = Z) (X = Z Y = B) ) ψ i (X, Y) ]. Johannes Mittmann IP = PSPACE JASS 2006 13

Polynomial Space PSPACE-Completeness Proof sketch of (2). ψ 0 (A, B) can be written in DNF. Bad idea: ψ i+1 = Z [ ψ i (A, Z) ψ i (Z, B) ]. Savitch s trick: ψ i+1 = Z X Y [( (X = A Y = Z) (X = Z Y = B) ) ψ i (X, Y) ]. Convert to prenex DNF. Johannes Mittmann IP = PSPACE JASS 2006 13

Polynomial Space PSPACE-Completeness Proof sketch of (2). ψ 0 (A, B) can be written in DNF. Bad idea: ψ i+1 = Z [ ψ i (A, Z) ψ i (Z, B) ]. Savitch s trick: ψ i+1 = Z X Y [( (X = A Y = Z) (X = Z Y = B) ) ψ i (X, Y) ]. Convert to prenex DNF. L log QSAT. Johannes Mittmann IP = PSPACE JASS 2006 13

Polynomial Space PSPACE-Completeness Proof sketch of (2). ψ 0 (A, B) can be written in DNF. Bad idea: ψ i+1 = Z [ ψ i (A, Z) ψ i (Z, B) ]. Savitch s trick: ψ i+1 = Z X Y [( (X = A Y = Z) (X = Z Y = B) ) ψ i (X, Y) ]. Convert to prenex DNF. L log QSAT. PSPACE = copspace. Johannes Mittmann IP = PSPACE JASS 2006 13

Shamir s Theorem Theorem (Shamir 1992) IP = PSPACE. Johannes Mittmann IP = PSPACE JASS 2006 14

Shamir s Theorem Theorem (Shamir 1992) IP = PSPACE. Proof. : Optimal prover strategy: Traverse the tree of all possible interactions between Alice and Bob. Compute the probabilities of acceptance. Johannes Mittmann IP = PSPACE JASS 2006 14

Proof Outline Proof. : It suffices to show QSAT IP, since IP is closed under reductions, and QSAT PSPACE-complete. Johannes Mittmann IP = PSPACE JASS 2006 15

Proof Outline Proof. : It suffices to show QSAT IP, since IP is closed under reductions, and QSAT PSPACE-complete. Outline: Arithmetization: φ A φ. Reduction to F p. Polynomials and simple expressions. Interactive protocol that decides QSAT. Johannes Mittmann IP = PSPACE JASS 2006 15

Arithmetization Arithmetization Quantified Boolean expression φ Arithmetization A φ Conversion rules: QBF φ Arithmetization A φ true 1 false 0 x i X z i Z x i 1 z i ψ 1 ψ 2 A ψ1 + A ψ2 ψ 1 ψ 2 A ψ1 A ψ2 x i ψ 1 z i =0 A ψ x i ψ 1 z i =0 A ψ Johannes Mittmann IP = PSPACE JASS 2006 16

Arithmetization Example φ = x 1 [ x1 x 2 x 3 (x 1 x 2 ) x 3 ]. A φ = 1 z 1 =0 A φ is called Σ-Π expression. [ (1 z 1 ) + 1 1 z 2 =0 z 3 =0 ] (z 1 z 2 + z 3 ). Johannes Mittmann IP = PSPACE JASS 2006 17

Arithmetization Lemma Let φ be a closed QBF with negation only over variables. Then φ is true A φ > 0. Johannes Mittmann IP = PSPACE JASS 2006 18

Arithmetization Lemma Let φ be a closed QBF with negation only over variables. Then φ is true A φ > 0. Proof. Induction on the structure of a (not necessarily closed) QBF φ. φ = x i : φ is true x i = true A φ = z i = 1 > 0. Johannes Mittmann IP = PSPACE JASS 2006 18

Arithmetization Proof (continued). φ = x i : φ is true x i = false A φ = 1 z i = 1 0 > 0. Johannes Mittmann IP = PSPACE JASS 2006 19

Arithmetization Proof (continued). φ = x i : φ is true x i = false A φ = 1 z i = 1 0 > 0. φ = ψ 1 ψ 2 : φ is true ψ 1 is true or ψ 2 is true A ψ1 > 0 or A ψ2 > 0 A φ = A ψ1 + A ψ2 > 0.... Johannes Mittmann IP = PSPACE JASS 2006 19

Arithmetization Problem: Exponential Arithmetizations Example φ = x 1 x 2 x k 1 x k (x k x k ). 1 1 1 1 [ A φ = zk + (1 z k ) ] = 2 2k 1. z 1 =0 z 2 =0 z k 1 =0 z k =0 Johannes Mittmann IP = PSPACE JASS 2006 20

Arithmetization Problem: Exponential Arithmetizations Example φ = x 1 x 2 x k 1 x k (x k x k ). 1 1 1 1 [ A φ = zk + (1 z k ) ] = 2 2k 1. z 1 =0 z 2 =0 z k 1 =0 z k =0 exponentially many bits needed! Johannes Mittmann IP = PSPACE JASS 2006 20

Arithmetization Problem: Exponential Arithmetizations Example φ = x 1 x 2 x k 1 x k (x k x k ). 1 1 1 1 [ A φ = zk + (1 z k ) ] = 2 2k 1. z 1 =0 z 2 =0 z k 1 =0 z k =0 exponentially many bits needed! Idea: reduction to a finite field. Johannes Mittmann IP = PSPACE JASS 2006 20

Reduction to a Finite Field Lemma Let A φ be a Σ-Π expression of length n. Then A φ 2 2n. Johannes Mittmann IP = PSPACE JASS 2006 21

Reduction to a Finite Field Lemma Let A φ be a Σ-Π expression of length n. Then A φ 2 2n. Proof. Induction on the structure of φ. φ = ( )x i : A φ 1 2 21. Johannes Mittmann IP = PSPACE JASS 2006 21

Reduction to a Finite Field Lemma Let A φ be a Σ-Π expression of length n. Then A φ 2 2n. Proof. Induction on the structure of φ. φ = ( )x i : A φ 1 2 21. φ = ψ 1 ψ 2, {, }: A φ 2 2l 2 2m = 2 2l +2 m 2 2n (l + m n). Johannes Mittmann IP = PSPACE JASS 2006 21

Reduction to a Finite Field Lemma Let A φ be a Σ-Π expression of length n. Then A φ 2 2n. Proof. Induction on the structure of φ. φ = ( )x i : A φ 1 2 21. φ = ψ 1 ψ 2, {, }: A φ 2 2l 2 2m = 2 2l +2 m 2 2n (l + m n). φ = Qx i ψ, Q {, }: A φ 2 2m 2 2m = 2 2 2m = 2 2m+1 2 2n (m < n). Johannes Mittmann IP = PSPACE JASS 2006 21

Reduction to a Finite Field Number of Primes Lemma For n 3, n π(n) n. Johannes Mittmann IP = PSPACE JASS 2006 22

Reduction to a Finite Field Number of Primes Lemma For n 3, n π(n) n. Proof. π(n) n p n p 1 p n n i=2 i 1 i = n/ n n. Johannes Mittmann IP = PSPACE JASS 2006 22

Reduction to a Finite Field The Chinese Remainder Theorem Theorem (Sun Tzu) Let a 1,..., a k Z, and let p 1,..., p k N >0 be pairwise coprime. Then the system of simultaneous congruences x = a 1 (mod p 1 ). x = a k (mod p k ) has a unique solution x modulo k i=1 p i. Johannes Mittmann IP = PSPACE JASS 2006 23

Reduction to a Finite Field Reduction to F p Proposition For every Σ-Π expression A 0 of length n, there exists a prime p [ 2 n, 2 3n] such that A 0 (mod p). Johannes Mittmann IP = PSPACE JASS 2006 24

Reduction to a Finite Field Proof. Denote by p 1,..., p k all primes in [ 2 n, 2 3n]. Johannes Mittmann IP = PSPACE JASS 2006 25

Reduction to a Finite Field Proof. Denote by p 1,..., p k all primes in [ 2 n, 2 3n]. k = π ( 2 3n) π ( 2 n) 2 3n 2 n > 2 n. Johannes Mittmann IP = PSPACE JASS 2006 25

Reduction to a Finite Field Proof. Denote by p 1,..., p k all primes in [ 2 n, 2 3n]. k = π ( 2 3n) π ( 2 n) 2 3n 2 n > 2 n. Suppose A = 0 (mod p i ) i. Johannes Mittmann IP = PSPACE JASS 2006 25

Reduction to a Finite Field Proof. Denote by p 1,..., p k all primes in [ 2 n, 2 3n]. k = π ( 2 3n) π ( 2 n) 2 3n 2 n > 2 n. Suppose A = 0 (mod p i ) i. Chinese Remainder Theorem: A = 0 mod k p i > 2 2n. i=1 Johannes Mittmann IP = PSPACE JASS 2006 25

Reduction to a Finite Field Proof. Denote by p 1,..., p k all primes in [ 2 n, 2 3n]. k = π ( 2 3n) π ( 2 n) 2 3n 2 n > 2 n. Suppose A = 0 (mod p i ) i. Chinese Remainder Theorem: A = 0 mod k p i > 2 2n. i=1 A 2 2n = A = 0, contradiction. Johannes Mittmann IP = PSPACE JASS 2006 25

Polynomials and Simple Expressions Functional and Randomized Form Definition Let A be a Σ-Π expression. The functional form A is defined by eliminating the leftmost 1 1 z i =0 symbol in A, and can be considered as a polynomial q(z i ) Z[z i ]. z i =0 or The randomized form of A is A (z i = r), where r R F p is a random number supplied by the verifier. Johannes Mittmann IP = PSPACE JASS 2006 26

Polynomials and Simple Expressions Problem: Exponentially High Degree Example φ = x 1 x 2... x k (x 1 x 2 x k ). 1 1 1 A φ = (z 1 + z 2 + + z k ). z 1 =0 z 2 =0 z k =0 = deg q(z 1 ) = 2 k 1. Johannes Mittmann IP = PSPACE JASS 2006 27

Polynomials and Simple Expressions Problem: Exponentially High Degree Example φ = x 1 x 2... x k (x 1 x 2 x k ). 1 1 1 A φ = (z 1 + z 2 + + z k ). z 1 =0 z 2 =0 z k =0 = deg q(z 1 ) = 2 k 1. dense polynomial of exponentially high degree! Johannes Mittmann IP = PSPACE JASS 2006 27

Polynomials and Simple Expressions Problem: Exponentially High Degree Example φ = x 1 x 2... x k (x 1 x 2 x k ). 1 1 1 A φ = (z 1 + z 2 + + z k ). z 1 =0 z 2 =0 z k =0 = deg q(z 1 ) = 2 k 1. dense polynomial of exponentially high degree! Idea: elimination of universal quantifiers. Johannes Mittmann IP = PSPACE JASS 2006 27

Polynomials and Simple Expressions Simple Expressions Definition A QBF φ is called simple, if any occurrence of a variable is separated by at most one universal quantifier from its point of quantification. Johannes Mittmann IP = PSPACE JASS 2006 28

Polynomials and Simple Expressions Simple Expressions Definition A QBF φ is called simple, if any occurrence of a variable is separated by at most one universal quantifier from its point of quantification. Example φ = x 1 x 2 x 3 [ (x1 x 2 ) x 4 (x 2 x 3 x 4 ) ] is simple. ψ = x 1 x 2 [ (x1 x 2 ) x 3 ( x 1 x 3 ) ] is not simple. Johannes Mittmann IP = PSPACE JASS 2006 28

Polynomials and Simple Expressions Lemma Let φ be a simple QBF of length n, and let q(z i ) be the polynomial of the functional form of A φ. Then deg q(z i ) 2n. Johannes Mittmann IP = PSPACE JASS 2006 29

Polynomials and Simple Expressions Lemma Let φ be a simple QBF of length n, and let q(z i ) be the polynomial of the functional form of A φ. Then Proof. deg q(z i ) 2n. The degree of quantifier-free subexpressions in z i is bounded its size. Johannes Mittmann IP = PSPACE JASS 2006 29

Polynomials and Simple Expressions Lemma Let φ be a simple QBF of length n, and let q(z i ) be the polynomial of the functional form of A φ. Then Proof. deg q(z i ) 2n. The degree of quantifier-free subexpressions in z i is bounded its size. Summations cannot change the degree. Johannes Mittmann IP = PSPACE JASS 2006 29

Polynomials and Simple Expressions Lemma Let φ be a simple QBF of length n, and let q(z i ) be the polynomial of the functional form of A φ. Then Proof. deg q(z i ) 2n. The degree of quantifier-free subexpressions in z i is bounded its size. Summations cannot change the degree. Products can at most double the degree. Johannes Mittmann IP = PSPACE JASS 2006 29

Polynomials and Simple Expressions Lemma Let φ be a simple QBF of length n, and let q(z i ) be the polynomial of the functional form of A φ. Then Proof. deg q(z i ) 2n. The degree of quantifier-free subexpressions in z i is bounded its size. Summations cannot change the degree. Products can at most double the degree. In simple expressions this can happen at most once. Johannes Mittmann IP = PSPACE JASS 2006 29

Polynomials and Simple Expressions Lemma Any QBF φ can be transformed in logarithmic space to an equivalent simple expression. Johannes Mittmann IP = PSPACE JASS 2006 30

Polynomials and Simple Expressions Lemma Any QBF φ can be transformed in logarithmic space to an equivalent simple expression. Proof. Consider φ =... Qx i... x j ψ(x i ), Q {, }. Johannes Mittmann IP = PSPACE JASS 2006 30

Polynomials and Simple Expressions Lemma Any QBF φ can be transformed in logarithmic space to an equivalent simple expression. Proof. Consider φ =... Qx i... x j ψ(x i ), Transformation: Q {, }. φ =... Qx i... x j x i (x i x i ) ψ(x i ) [ =... Qx i... x j x i (xi x i ) ( x i x i ) ] ψ(x i ). Johannes Mittmann IP = PSPACE JASS 2006 30

Polynomials and Simple Expressions Lemma Any QBF φ can be transformed in logarithmic space to an equivalent simple expression. Proof. Consider φ =... Qx i... x j ψ(x i ), Transformation: Q {, }. φ =... Qx i... x j x i (x i x i ) ψ(x i ) [ =... Qx i... x j x i (xi x i ) ( x i x i ) ] ψ(x i ). O(n 2 ) steps. Johannes Mittmann IP = PSPACE JASS 2006 30

The Interactive Protocol Protocol Setup Prover Verifier Choose p [ 2 n, 2 3n]. Compute a A φ (mod p). p, a Verify a 0, p [ 2 n, 2 3n], and p Primes. Johannes Mittmann IP = PSPACE JASS 2006 31

The Interactive Protocol Split Step A = A 1 + A 2 or A = A 1 A 2, where A 2 starts with the leftmost 1 z i =0 or 1 z i =0 symbol: A 2 = 1... or A 2 = z i =0 1... z i =0 Johannes Mittmann IP = PSPACE JASS 2006 32

The Interactive Protocol Simplification Step Prover Verifier Compute q(z i ) of A. q(z i ) A A 2. a a a 1 (mod p), or a a/a 1 (mod p). Verify a = q(0) + q(1) (mod p), or a = q(0) q(1) (mod p). r Choose r R F p. A A (z i = r) (mod p). a q(r) (mod p). Johannes Mittmann IP = PSPACE JASS 2006 33

The Interactive Protocol Example (Round 1) A A φ = a 2. 1 z 1 =0 [ (1 z 1 ) + 1 1 z 2 =0 z 3 =0 ] (z 1 z 2 + z 3 ), Johannes Mittmann IP = PSPACE JASS 2006 34

The Interactive Protocol Example (Round 1) A A φ = a 2. 1 z 1 =0 A = (1 z 1 ) + q(z 1 ) = z 2 1 + 1. [ (1 z 1 ) + 1 z 2 =0 z 3 =0 1 1 z 2 =0 z 3 =0 1 (z 1 z 2 + z 3 ), ] (z 1 z 2 + z 3 ), Johannes Mittmann IP = PSPACE JASS 2006 34

The Interactive Protocol Example (Round 1) A A φ = a 2. 1 z 1 =0 A = (1 z 1 ) + q(z 1 ) = z 2 1 + 1. [ (1 z 1 ) + Verify a = 2 = 2 1 = q(0) q(1). 1 z 2 =0 z 3 =0 1 1 z 2 =0 z 3 =0 1 (z 1 z 2 + z 3 ), ] (z 1 z 2 + z 3 ), Johannes Mittmann IP = PSPACE JASS 2006 34

The Interactive Protocol Example (Round 1) A A φ = a 2. 1 z 1 =0 A = (1 z 1 ) + q(z 1 ) = z 2 1 + 1. [ (1 z 1 ) + Verify a = 2 = 2 1 = q(0) q(1). 1 z 2 =0 z 3 =0 A A (z 1 = 3) = (1 3) + a q(3) = 10. 1 1 z 2 =0 z 3 =0 1 (z 1 z 2 + z 3 ), 1 z 2 =0 z 3 =0 ] (z 1 z 2 + z 3 ), 1 (3z 2 + z 3 ), Johannes Mittmann IP = PSPACE JASS 2006 34

The Interactive Protocol Example (Round 2) 1 1 A A 2 = (3z 2 + z 3 ), z 2 =0 z 3 =0 a a a 1 = 10 ( 2) = 12. Johannes Mittmann IP = PSPACE JASS 2006 35

The Interactive Protocol Example (Round 2) 1 1 A A 2 = (3z 2 + z 3 ), z 2 =0 z 3 =0 a a a 1 = 10 ( 2) = 12. 1 A = (3 z 2 + z 3 ), z 3 =0 q(z 2 ) = 9z 2 2 + 3z 2. Johannes Mittmann IP = PSPACE JASS 2006 35

The Interactive Protocol Example (Round 2) 1 1 A A 2 = (3z 2 + z 3 ), z 2 =0 z 3 =0 a a a 1 = 10 ( 2) = 12. 1 A = (3 z 2 + z 3 ), z 3 =0 q(z 2 ) = 9z 2 2 + 3z 2. Verify a = 12 = 0 + 12 = q(0) + q(1). Johannes Mittmann IP = PSPACE JASS 2006 35

The Interactive Protocol Example (Round 2) 1 1 A A 2 = (3z 2 + z 3 ), z 2 =0 z 3 =0 a a a 1 = 10 ( 2) = 12. 1 A = (3 z 2 + z 3 ), z 3 =0 q(z 2 ) = 9z 2 2 + 3z 2. Verify a = 12 = 0 + 12 = q(0) + q(1). 1 A A (z 2 = 2) = (z 3 + 6), z 3 =0 a q(2) = 9 4 + 3 2 = 42. Johannes Mittmann IP = PSPACE JASS 2006 35

The Interactive Protocol Example (Round 3) 1 A (z 3 + 6), z 3 =0 a 42. Johannes Mittmann IP = PSPACE JASS 2006 36

The Interactive Protocol Example (Round 3) 1 A (z 3 + 6), z 3 =0 a 42. A = z 3 + 6, q(z 3 ) = z 3 + 6. Johannes Mittmann IP = PSPACE JASS 2006 36

The Interactive Protocol Example (Round 3) 1 A (z 3 + 6), z 3 =0 a 42. A = z 3 + 6, q(z 3 ) = z 3 + 6. Verify a = 42 = 6 7 = q(0) q(1). Johannes Mittmann IP = PSPACE JASS 2006 36

The Interactive Protocol Example (Round 3) 1 A (z 3 + 6), z 3 =0 a 42. A = z 3 + 6, q(z 3 ) = z 3 + 6. Verify a = 42 = 6 7 = q(0) q(1). A A (z 3 = 5) = 5 + 6 = 11, a q(5) = 5 + 6 = 11. Johannes Mittmann IP = PSPACE JASS 2006 36

The Interactive Protocol Example (Round 3) 1 A (z 3 + 6), z 3 =0 a 42. A = z 3 + 6, q(z 3 ) = z 3 + 6. Verify a = 42 = 6 7 = q(0) q(1). A A (z 3 = 5) = 5 + 6 = 11, a q(5) = 5 + 6 = 11. Verify A = 11 = a, and accept. Johannes Mittmann IP = PSPACE JASS 2006 36

The Interactive Protocol Correctness of the Interactive Proof Theorem 1 When φ is true and Alice is honest, Bob will always accept the proof. 2 When φ is false, Bob accepts the proof with negligible probability. Johannes Mittmann IP = PSPACE JASS 2006 37

The Interactive Protocol Correctness of the Interactive Proof Theorem 1 When φ is true and Alice is honest, Bob will always accept the proof. 2 When φ is false, Bob accepts the proof with negligible probability. Proof of completeness. Alice is able to provide all the polynomials. Bob will always accept. Johannes Mittmann IP = PSPACE JASS 2006 37

The Interactive Protocol Proof of soundness. Suppose A φ = 0 and still Alice claims an a 0. Johannes Mittmann IP = PSPACE JASS 2006 38

The Interactive Protocol Proof of soundness. Suppose A φ = 0 and still Alice claims an a 0. Alice must supply a wrong polynomial q (z i ) in the i-th round. Johannes Mittmann IP = PSPACE JASS 2006 38

The Interactive Protocol Proof of soundness. Suppose A φ = 0 and still Alice claims an a 0. Alice must supply a wrong polynomial q (z i ) in the i-th round. 0 q(z i ) q (z i ) has at most 2n roots. Johannes Mittmann IP = PSPACE JASS 2006 38

The Interactive Protocol Proof of soundness. Suppose A φ = 0 and still Alice claims an a 0. Alice must supply a wrong polynomial q (z i ) in the i-th round. 0 q(z i ) q (z i ) has at most 2n roots. Probability of a false positive: Pr[error in the i-th round] 2n p 2n 2 n. Johannes Mittmann IP = PSPACE JASS 2006 38

The Interactive Protocol Proof of soundness. Suppose A φ = 0 and still Alice claims an a 0. Alice must supply a wrong polynomial q (z i ) in the i-th round. 0 q(z i ) q (z i ) has at most 2n roots. Probability of a false positive: After m n rounds: Pr[error in the i-th round] 2n p Pr[error] = 1 1 2n 2 n. m Pr[no error in the i-th round] i=1 ( 1 2n 2 n ) n. Johannes Mittmann IP = PSPACE JASS 2006 38

References For further reading: L. J. Stockmeyer and A. R. Meyer: Word problems requiring exponential time. Proc. 5th ACM Symp. on the Theory of Computing, pp. 1-9, 1973. L. Babai: E-mail and the unexpected power of interaction. Structure in Complexity Theory Conf., pp. 30-44, 1990. Adi Shamir: IP=PSPACE. Journal of the ACM, 39 (4), pp. 869-877, 1992. Christos H. Papadimitriou: Computational Complexity. Addison Wesley, Reading, 1994. Johannes Mittmann IP = PSPACE JASS 2006 39

Exercise Exercise Exercise Show that #P IP. Hint. Arithmetization of the #P-complete counting problem #SAT. Johannes Mittmann IP = PSPACE JASS 2006 40

Exercise Solution Arithmetization: 3-CNF φ Arithmetization A φ x i X z i Z ψ 1 A ψ ψ 1 ψ 2 1 (1 A ψ1 )(1 A ψ2 ) ψ 1 ψ 2 A ψ1 A ψ2 Then φ(x 1,..., x n ) has exactly K satisfying assignments, iff 1 1 K = A φ (z 1,..., z n ). z 1 =0 z n=0 Johannes Mittmann IP = PSPACE JASS 2006 41