Concurrent Non-malleable Commitments from any One-way Function

Similar documents
Constant-round Non-Malleable Commitments from Any One-Way Function

Constant-Round Non-Malleable Commitments from Any One-Way Function

Constant-Round Non-Malleable Commitments from Any One-Way Function

Parallel Coin-Tossing and Constant-Round Secure Two-Party Computation

Delayed-Input Non-Malleable Zero Knowledge and Multi-Party Coin Tossing in Four Rounds

Lecture Notes 20: Zero-Knowledge Proofs

Impossibility Results for Universal Composability in Public-Key Models and with Fixed Inputs

Constant-round Leakage-resilient Zero-knowledge from Collision Resistance *

Adaptive Hardness and Composable Security in the Plain Model from Standard Assumptions

Inaccessible Entropy and its Applications. 1 Review: Psedorandom Generators from One-Way Functions

Round-Optimal Fully Black-Box Zero-Knowledge Arguments from One-Way Permutations

Augmented Black-Box Simulation and Zero Knowledge Argument for NP

Non-Interactive ZK:The Feige-Lapidot-Shamir protocol

Non-Conversation-Based Zero Knowledge

Impossibility and Feasibility Results for Zero Knowledge with Public Keys

Lecture 11: Non-Interactive Zero-Knowledge II. 1 Non-Interactive Zero-Knowledge in the Hidden-Bits Model for the Graph Hamiltonian problem

Session 4: Efficient Zero Knowledge. Yehuda Lindell Bar-Ilan University

Lecture th January 2009 Fall 2008 Scribes: D. Widder, E. Widder Today s lecture topics

CMSC 858K Introduction to Secure Computation October 18, Lecture 19

Lecture 15 - Zero Knowledge Proofs

Probabilistically Checkable Arguments

1 Recap: Interactive Proofs

Zero Knowledge and Soundness are Symmetric

From Secure MPC to Efficient Zero-Knowledge

Parallel Repetition of Zero-Knowledge Proofs and the Possibility of Basing Cryptography on NP-Hardness

6.897: Selected Topics in Cryptography Lectures 7 and 8. Lecturer: Ran Canetti

Lecture 3,4: Universal Composability

Abstract. Often the core diculty in designing zero-knowledge protocols arises from having to

Notes on Zero Knowledge

Round-Efficient Multi-party Computation with a Dishonest Majority

An Epistemic Characterization of Zero Knowledge

Statistical WI (and more) in Two Messages

Cryptography CS 555. Topic 23: Zero-Knowledge Proof and Cryptographic Commitment. CS555 Topic 23 1

The Round-Complexity of Black-Box Zero-Knowledge: A Combinatorial Characterization

c 2010 Society for Industrial and Applied Mathematics

Improved OR-Composition of Sigma-Protocols

How to Construct Constant-Round. Zero-Knowledge Proof Systems for NP. Oded Goldreich y Ariel Kahan z. March Abstract

A New Approach to Black-Box Concurrent Secure Computation

Notes for Lecture 27

Concurrent Knowledge Extraction in the Public-Key Model

Lecture 17: Constructions of Public-Key Encryption

An Epistemic Characterization of Zero Knowledge

How many rounds can Random Selection handle?

Lecture 3: Interactive Proofs and Zero-Knowledge

On Constant-Round Concurrent Zero-Knowledge

Distinguisher-Dependent Simulation in Two Rounds and its Applications

Statistically Hiding Commitments and Statistical Zero-Knowledge Arguments from Any One-Way Function

Lecture 18: Message Authentication Codes & Digital Signa

Constant-Round Coin-Tossing With a Man in the Middle or Realizing the Shared Random String Model

Lecture 18: Zero-Knowledge Proofs

A Framework for Non-Interactive Instance-Dependent Commitment Schemes (NIC)

(Nearly) round-optimal black-box constructions of commitments secure against selective opening attacks

On the round complexity of black-box constructions of commitments secure against selective opening attacks

2 Message authentication codes (MACs)

of trapdoor permutations has a \reversed sampler" (which given ; y generates a random r such that S 0 (; r) = y), then this collection is doubly-enhan

Statistically Hiding Commitments and Statistical Zero-Knowledge Arguments from Any One-Way Function

1/p-Secure Multiparty Computation without an Honest Majority and the Best of Both Worlds

Constant-Round Concurrently-Secure rzk in the (Real) Bare Public-Key Model

Indistinguishability and Pseudo-Randomness

Computational security & Private key encryption

Finding Collisions in Interactive Protocols Tight Lower Bounds on the Round and Communication Complexities of Statistically Hiding Commitments

An Identification Scheme Based on KEA1 Assumption

A Full Characterization of Functions that Imply Fair Coin Tossing and Ramifications to Fairness

Non-Interactive Non-Malleability from Quantum Supremacy

MTAT Cryptology II. Zero-knowledge Proofs. Sven Laur University of Tartu

Statistical Zero-Knowledge Arguments for NP from Any One-Way Function

Lecture 11: Key Agreement

Generic and Practical Resettable Zero-Knowledge in the Bare Public-Key Model

Lecture 13: Private Key Encryption

Lecture 9 and 10: Malicious Security - GMW Compiler and Cut and Choose, OT Extension

In this chapter we discuss zero-knowledge proof systems. Loosely speaking, such proof

Contents 1 Introduction The Basics : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : A

A Transform for NIZK Almost as Efficient and General as the Fiat-Shamir Transform Without Programmable Random Oracles

Interactive and Noninteractive Zero Knowledge Coincide in the Help Model

Textbook Non-Malleable Commitments

Extracted from a working draft of Goldreich s FOUNDATIONS OF CRYPTOGRAPHY. See copyright notice.

CS 355: TOPICS IN CRYPTOGRAPHY

Perfect Zero-Knowledge Arguments for N P Using any One-Way. Permutation. Abstract

On the (Im)Possibility of Arthur-Merlin Witness Hiding Protocols

Interactive Zero-Knowledge with Restricted Random Oracles

Contents 1 Introduction 2 2 Formal Setting and General Observations Specication : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

Lecture 22. We first consider some constructions of standard commitment schemes. 2.1 Constructions Based on One-Way (Trapdoor) Permutations

Black-Box, Round-Efficient Secure Computation via Non-Malleability Amplification

Lecture 7: CPA Security, MACs, OWFs

Lecture 6. 2 Adaptively-Secure Non-Interactive Zero-Knowledge

Founding Cryptography on Smooth Projective Hashing

Magic Functions. In Memoriam Bernard M. Dwork

Lecture 2: Program Obfuscation - II April 1, 2009

Foundation of Cryptography, Lecture 7 Non-Interactive ZK and Proof of Knowledge

Block-wise Non-malleable Codes

Game-Theoretic Security for Two-Party Protocols

Notes on Complexity Theory Last updated: November, Lecture 10

Contents 1 Introduction Objects, specications, and implementations : : : : : : : : : : : : : : : : : : : : : : : : : : : : Indistinguishab

On the Security of Classic Protocols for Unique Witness Relations

6.897: Advanced Topics in Cryptography. Lecturer: Ran Canetti

On Achieving the Best of Both Worlds in Secure Multiparty Computation

Lecture 4: Hardness Amplification: From Weak to Strong OWFs

Weak Zero-Knowledge Beyond the Black-Box Barrier

Block Ciphers/Pseudorandom Permutations

Inaccessible Entropy

Transcription:

Concurrent Non-malleable Commitments from any One-way Function Margarita Vald Tel-Aviv University 1 / 67

Outline Non-Malleable Commitments Problem Presentation Overview DDN - First NMC Protocol Concurrent Non-Malleable Commitments Overview Denitions The connmccom Protocol Security Proof Summary 2 / 67

Outline Non-Malleable Commitments Problem Presentation Overview DDN - First NMC Protocol Concurrent Non-Malleable Commitments Overview Denitions The connmccom Protocol Security Proof Summary 3 / 67

Commitments ˆ String commitment protocol consists of two stages ˆ Commit stage: C and R exchange messages. At the end of this stage R has some information that represents v (R should gain no information on the value of v) ˆ The reveal stage: at the end of this stage R knows v (There should be only one string that C can reveal). 4 / 67

Commitment Schemes I The requirements of string commitment protocol: ˆ Secrecy - Computational hiding ˆ For every (expected) PPT machine R the following ensembles are computationally indistinguishable over{0, 1} n { } sta R C,R (v 1, z) v 1,v 2 {0,1} n,n N, z {0,1} { } sta R C,R (v 2, z) v 1,v 2 {0,1} n,n N, z {0,1} ˆ Where sta R C,R (v, z) denote random variable describing the output of R after receiving a commitment to v using C, R. 5 / 67

Commitment Schemes II ˆ Binding - Statistically binding ˆ Informally, the statistical-binding property asserts that, with overwhelming probability over the coin-tosses of the receiver R, the transcript of the interaction fully determines the value committed to by the sender. ˆ Instead we can require a computationally binding and statistically hiding commitment scheme ˆ It is a known result that statistically binding and statistically hiding commitment schemes don't exist ˆ We consider commitment schemes that are statistically-binding and computationally-hiding 6 / 67

Outline Non-Malleable Commitments Problem Presentation Overview DDN - First NMC Protocol Concurrent Non-Malleable Commitments Overview Denitions The connmccom Protocol Security Proof Summary 7 / 67

Non-Malleable Commitments I ˆ Non-Malleability is the case where the adversary sees a commitment to a specic value and doesn't succeeds to commit to a related value ˆ Intuitively we tend to think about commitments as sealed envelopes ˆ It appears to captures the notion of non-malleability ˆ In reality when we replace the envelopes with real commitments we see our intuition is faulty ˆ The basic denition of commitment schemes doesn't capture non-malleability. 8 / 67

Non-Malleable Commitments II ˆ Consider the following commitment scheme for m [1...p]: ˆ Let G be a group of prime order p 2 n and let g be a generator of G. ˆ This scheme is computationally hiding and perfectly binding as desired. ˆ Recall the DDH assumption 9 / 67

Non-Malleable Commitments III ˆ A commitment that was made using this scheme can be maul. ˆ Consider an adversary that upon receiving a commitment (a, b, c) proceeds as follows: ˆ Choose r 1, r 2 R Z p. ) ) ) ˆ Compute a = (g r 1 a, b = (g r 2 b and c = (g r 1 r 2 c ˆ Commit to (a, b, c ) ˆ Note: The adversary commits to ( ) g r 1 r 1, g r 2 r 2, g r 2 r 2 r 1 r1 m 10 / 67

Non-Malleable Commitments IV ˆ The rst to raise this notion were Dolev, Dvork and Naor in their paper Non-Malleable Cryptography (1991) ˆ Presented a non-malleable commitment scheme with O (log n) rounds ˆ The paper provides a very complex and subtle proof ˆ Based on the existence of OWF ˆ Uses ZKP as a building block 11 / 67

NMC (cont.) ˆ Intuitively a commitment scheme is non-malleable if a man-in-the-middle adversary that receives commitment to v will not be able to successfully commit to related value. C com(v1) A(z) com(v1) R Man-in-the-middle execution ˆ We compare between a Man-in-the-middle execution and an ideal execution ˆ Notation: Let C, R be a commitment scheme, R {0, 1} n {0, 1} n be a polynomial-time computable non-reexive relation (i.e., R(v, v) = 0), and let n N be a security parameter. 12 / 67

Man-in-the-middle Execution C com(v1) A(z) com(v1) R Man-in-the-middle execution The success of A is dened by: ˆ mim A <C,R> (R, v, z) = 1 i A produces a valid commitment to v such that R(v, v) = 1. 13 / 67

Ideal Execution S(z) com(v1) R Simulated execution The success of S is dened by: ˆ sim S <C,R> (R, v, z) = 1 if and only if S produces a valid commitment to v such that R(v, v) = 1. 14 / 67

Non-malleable Commitments Denition (Non-Malleable Commitment Scheme) A commitment scheme C, R is said to be liberal non-malleable if for every probabilistic polynomial-time man-in-the-middle adversary A, there exists a probabilistic expected polynomial time adversary S, such that for every non-reexive polynomial-time computable relation R {0, 1} n {0, 1} n, every v {0, 1} n, and every z {0, 1}, it holds that: Pr [mim A <C,R> (R, v, z) = 1 ] Pr [ sim S <C,R> (R, v, z) = 1 ] < ν (n) 15 / 67

Outline Non-Malleable Commitments Problem Presentation Overview DDN - First NMC Protocol Concurrent Non-Malleable Commitments Overview Denitions The connmccom Protocol Security Proof Summary 16 / 67

POK - First Attempt ˆ We wish to convert a malleable commitment scheme to a non-malleable one ˆ Our rst attempt would be to use a 3 rounds POK ˆ To ensure the committer knows the value he committed to ˆ 3 rounds POK consists of ˆ First step: Commitment ˆ Second step: Challenge ˆ Third step: Response 17 / 67

ˆ POK may be malleable Malleable Protocol ˆ Consider the following scenario: C Commit A Commit R Challenge Challenge Reply Reply POK ˆ Goal: ensure that for each execution there will be a triple for which no other triple is useful 18 / 67

Outline Non-Malleable Commitments Problem Presentation Overview DDN - First NMC Protocol Concurrent Non-Malleable Commitments Overview Denitions The connmccom Protocol Security Proof Summary 19 / 67

Concurrent Non-Malleable commitments ˆ A natural extension of Non-Malleability is one in which more than two invocations of the commitment protocol take place concurrently ˆ DDN scheme does not rule out joint dependencies between more than one individual pair ˆ Still two major questions remained open: ˆ Can concurrent non-malleable commitments be based solely on the existence of one-way functions? ˆ Does there exist concurrent non-malleable commitments with black-box proofs of security? 20 / 67

Concurrent NMC from any OWF ˆ These questions were answered fully in the recent paper Concurrent Non-Malleable Commitments from any One-Way Function [Lin Pass Venkitasubramaniam 2008] ˆ Proves the existence of statistically-binding commitment scheme that is concurrent non-malleable based solely on the existence of OWF and using only black-box techniques. ˆ Variant of the DDN protocol ˆ Provides stronger denition of non-malleability ˆ The commitment scheme uses O(n) communication rounds ˆ Provides simple proof of security 21 / 67

Concurrent NMC (Cont.) ˆ Consider man-in-the-middle adversaries that are participating in left and right interactions in which m = poly(n) commitments take place. ˆ We compare between Man-in-the-middle execution and the ideal execution ˆ Notation: Let C, R be a commitment scheme, n N be a security parameter. ˆ Let id i be the identier of committer C i ˆ Let id i the identier used by the adversary in committing to v i 22 / 67

Man-in-the-middle Execution C1 (id1) com(v1) (id1) com(v1) R1 Ci (idi) com(vi) (idm) com(vm) A(z) (idi) com(vi) (idm) com(vm) Ri Cm Rm Man-in-the-middle execution ˆ Let { } mim A <C,R> (v 1,..., v m, z) denote a random variable that describes the values v 1,..., v m and the view of A, in the above experiment. 23 / 67

Ideal Execution S(z) (id1) com(v1) (idi) com(vi) R1 Ri (idm) com(vm) Rm Simulated execution ˆ Let { } sim S C,R (1n, z) denote the random variable describing the values v 1,..., v m committed to by S, and the output view of 24 / 67

Identities ˆ We make extensive use of identities ˆ There are two ways to model the identity of the committer ˆ Assuming there is a name-space in which a committer is represented by a unique string ˆ Concatenation of the committer identity to the commitment 25 / 67

Outline Non-Malleable Commitments Problem Presentation Overview DDN - First NMC Protocol Concurrent Non-Malleable Commitments Overview Denitions The connmccom Protocol Security Proof Summary 26 / 67

Concurrent Non-Malleable Scheme I Denition (Concurrent Non-Malleable Commitment Scheme) A commitment scheme C, R is said to be concurrent non-malleable if every probabilistic polynomial-time man-in-the-middle adversary A that participates in at most m concurrent executions, there exists a probabilistic polynomial time adversary S such that the following ensembles are computationally indistinguishable over {0, 1} n : { } mim A <C,R> (v 1,..., v m, z) { } v 1,...,v m {0,1} n,n N, z {0,1} sim S C,R (1n, z) v 1,...,v m {0,1} n, n N, z {0,1} 27 / 67

Concurrent Non-Malleable Scheme II ˆ This is a stronger denition that considers not only the values the adversary commits to, but also the view of the adversary! ˆ For m = 1 any protocol that satises this denition also satises the denition of DDN. ˆ This is because the existence of a polynomial time computable relation R that violates the original denition of non-malleability could be used to distinguish between the values of mim A C,R (v, z) and sims C, R (1n, z). 28 / 67

Relaxed Notion relaxed notions of concurrent non-malleability: ˆ one-one non-malleable commitment: we consider only adversaries A that participate in one left and one right interaction. ˆ one-many: A participates in one left and many right. ˆ many-one: A participates in many left and one right. One-one DDN Many-one Trivial Trivial One-many LPV (stronger definition) Many-many 29 / 67

One-many implies many-many I Proposition 1. Let C, R be a one-many concurrent non-malleable commitment. Then C, R is also a concurrent non-malleable commitment. Proof ˆ Let A be a man-in-the-middle adversary that participates in at most m = p(n) concurrent executions. ˆ Simulator S on input (1 n, z) ˆ S incorporates A(z) and internally emulates all the left interactions for A by simply honestly committing to the string 0 n. ˆ Messages from the right interactions are instead forwarded externally. ˆ Finally S outputs the view of A. 30 / 67

One-many implies many-many II ˆ Suppose, for contradiction, there exists a polynomial-time distinguisher D that distinguishes mim A C,R (v 1,..., v m, z) and sim S C,R (1n, z) w.p 1 p(n). ˆ Consider the hybrid simulator S i that on input 1 n, z = v 1,..., v m, z, proceeds just as S, with the exception that in left interactions j i, it instead commits to v j. ˆ By a standard hybrid argument there exists an i [m] such that Pr a sim S i 1 C,R (1n,z ) Pr b sim S i C,R (1n,z ) [ D ( 1 n, z, a ) = 1 ] [ D ( 1 n, z, b ) = 1 ] 1 p(n)m 31 / 67

One-many implies many-many III ˆ Consider the one-many adversary à that on input z = z, n, i executes S i=1(1 n, z ) with the exception that the i th left interaction is forwarded externally. ˆ Consider, the function reconstruct: on input mimãcom(0 n, z) ˆ reconstructs the view view of A in the emulation by Ã, and sets ṽ i = v 1 if A did not copy the identity of any of the left interactions, and otherwise, ˆ nally outputs ṽ i,..., ṽ m, view. ˆ By construction, it follows that ˆ reconstruct(mimã C,R (0n, z)) = sim S i 1 C,R (1n, z ) ˆ reconstruct(mimã C,R (v i, z)) = sim S i C,R (1n, z ) Since reconstruct is polynomial-time computable, this contradicts the one-many non-malleability of C, R. 32 / 67

Special-sound proofs Denition (Special-sound proofs) A 3-round public-coin interactive proof for the language L NP with witness relation R L is special-sound with respect to R L, if for any two transcripts (α, β, γ) and (α, β, γ ) such that the initial messages α, α are the same but the challenges β, β are dierent, there is a deterministic procedure to extract the witness from the two transcripts and runs in polynomial time. 33 / 67

Witness Indistinguishability Denition (Witness-indistinguishability) Let P, V be an interactive proof system for a language L NP. We say that P, V is witness-indistinguishable for R L, if for every probabilistic polynomial-time interactive machine V and for every two sequences { } w 1 and { } x w 2, such that x L x w 1 x L x, w 2 x R L (x) for every x L, { ˆ P(w 1 x ), V (z) (x) } x L,z {0,1} { ˆ P(w 2 x ), V (z) (x) } x L,z {0,1} are computationally indistinguishable over x L. 34 / 67

Outline Non-Malleable Commitments Problem Presentation Overview DDN - First NMC Protocol Concurrent Non-Malleable Commitments Overview Denitions The connmccom Protocol Security Proof Summary 35 / 67

Protocol connmccom I Common Input: An identier id {0, 1} l (The identity of the committer). Private Input for Committer: A string v {0, 1} n. Stage 1: ˆ R uniformly chooses r {0, 1} n. ˆ R C: s = f (r) where f is a OWF. ˆ C aborts if s not in the range of f. Stage 2: ˆ C R: c = com(v, r ) where r R {0, 1} poly(n).. 36 / 67

Protocol connmccom II Stage 3: C R: 4l special-sound WI proofs of the statement either there exists values v, r s.t c = com(v, r ) or there exists a value r s.t s = f (r) with verier query of length 2n, in the following schedule: ˆ For j = 1 to l do: ˆ Execute design idj followed by Execute design 1=id j 37 / 67

One-Way Functions ˆ The protocol relies on the existence of one-way functions with eciently recognizable range ˆ The protocol can be easily modied to work with any arbitrary one-way function ˆ by simply providing a witness hiding proof that an element is in the range of the one-way function 38 / 67

Designs ˆ Description of the schedules used in Stage 3. ˆ The scheduling identical to DDN massage scheduling technique. 39 / 67

Statistically-binding Commitment Scheme Lemma C, R is a statistically-binding commitment scheme. ˆ Binding: Follows directly from the binding property of com. ˆ Hiding: Follows from the hiding property of com and the fact that Stage 3 of the protocol is WI 40 / 67

Outline Non-Malleable Commitments Problem Presentation Overview DDN - First NMC Protocol Concurrent Non-Malleable Commitments Overview Denitions The connmccom Protocol Security Proof Summary 41 / 67

Proof of Security Theorem C, R is one-many concurrent non-malleable. ˆ Let A be an adversary. ˆ Simulator S : on input (1 n, z) ˆ incorporates A(z) and internally emulates the left interaction by honestly committing to the string 0 n. ˆ Messages in the right interactions are instead forwarded externally. ˆ Finally, outputs the view of A. } {mim A C,R (v, z) ˆ Goal: Show { } mim A C,R (0n, z) v {0,1} n,n N,z {0,1} v {0,1} n,n N,z {0,1} 42 / 67

Modied Scheme Let Ĉ, ˆR be a commitment scheme as follow: ˆ Variant of C, R Protocol ˆ Modied Stage 3: ˆ The receiver can ask for an arbitrary number of special-sound WI designs ˆ The receiver chooses which design to execute in each iteration ˆ By sending bit i ˆ Note: Ĉ, ˆR is computationally hiding ˆ Same proof as for C, R 43 / 67

Proof (Cont.) ˆ Let Γ(A, z) denote the distribution of all joint views τ of A and the receivers in the right, such that A sends its rst message in the left interaction directly after receiving the messages in τ. ˆ Let Z(z, τ) = z τ ṽ 1... ṽ l where l [m] and ṽ 1,..., ṽ l are the values committed to by A(z) in τ. Lemma For every PPT adversary A, there exists an expected PPT adversary R such that the following ensembles are indistinguishable { over } {0, 1} : sta R Ĉ, ˆR (v, Z A (z)) } v {0,1} n,n N,z {0,1} {mim A C,R (v, z) v {0,1} n,n N,z {0,1} 44 / 67

Why does it help? Ĉ, ˆR ˆ According to the hiding property of ensembles are indistinguishable { } sta R Ĉ, ˆR (v, Z A (z)) { } sta R Ĉ, ˆR (0n, Z A (z)) the following v {0,1} n,n N,z,τ,z {0,1} v {0,1} n,n N,z,τ,z {0,1} ˆ Using the Lemma, we conclude that the following ensembles are indistinguishable: {mim A C,R (v, z) } { } mim A C,R (0n, z) ˆ This concludes the proof of theorem. v {0,1} n,n N,z {0,1} v {0,1} n,n N,z {0,1} 45 / 67

Proving the Lemma Description of R ˆ Input: R receives auxiliary input z = z τ ṽ 1... ṽ l. ˆ Procedure: R interacts externally as a receiver using Ĉ, ˆR. Internally it incorporates A(z) and emulates a one-many man-in-the-middle execution by simulating all right receivers and emulating the left C, R interaction by requesting the appropriate designs expected by A(z) using Ĉ, ˆR from outside. 46 / 67

Adversary R < C, R > C R* < C, R > A(z) Adversary R* ˆ R interacts externally as a receiver using Ĉ, ˆR. ˆ R simulating all right receivers and emulating the left C, R interaction. 47 / 67

ˆ Main Execution Phase: Feed the view in τ to A and all right receivers. Emulate all the interaction from τ and complete the execution with A. Let be the transcript of messages obtained. ˆ Rewinding Phase: attempt to extract all the values committed by A. ˆ Output phase: For every interaction k that is not convincing or if the identity of the right interaction is the same as the left interaction, set ˆv k =. Output (ˆv 1,..., ˆv m ) and the view from the Main Execution Phase. ˆ Finally, if it runs for more than 2 n steps, halt and output fail. 48 / 67

Rewinding Phase ˆ Goal: Ensure that there exist some point (called safe-point) where we can rewind A on the right interaction, without rewinding on the left. ˆ This is possible in two cases: ˆ If rewinding on the right does not cause A to request any new messages on the left. ˆ If rewinding on the right causes A to only request a new special-sound proofin this case R can perfectly emulate this new proof by simply requesting another design from Ĉ, ˆR. 49 / 67

Safe-points Denition (Safe-point) A prex ρ of a transcript is called a safe-point for interaction k, if there exists an accepting proof (α r, β r, γ r ) in the k th right interaction, such that: 1. α r occurs in ρ, but not β r (and γ r ). 2. for any proof (α l, β l, γ l ) in the left interaction, if α l occurs in ρ and β l not in ρ, then β l occurs after γ r. 50 / 67

Safe-point examples ˆ When rewinding from ρ, R can emulate the left proof by requesting a new design from Ĉ, ˆR. 51 / 67

Safe-point examples(cont.) ˆ R can simply re-send the third message of the left proof ˆ since it is determined by the rst two messages in the proof 52 / 67

Safe-point examples(cont.) ˆ No new message is requested by A, so the left interaction can be trivially emulated by doing nothing. 53 / 67

Non Safe-point example ˆ Note: The only case a right-interaction proof does not have a safe-point is if it is aligned with a left execution proof ˆ A can forward messages between the left and the right interactions. 54 / 67

Rewinding Phase Procedure I Rewinding Phase: For k = l + 1 to m, if interaction k is convincing and its identity is dierent from the left interaction do: ˆ In, nd the rst point ρ that is a safe-point for the interaction k; let the associated proof be (α ρ, β ρ, γ ρ ). ˆ Repeat until a second-proof transcript ( α ρ, β ρ, γ ρ) is obtained: Emulate the left interaction as in the Main-Execution Phase. For the left interaction: ˆ If A expects to get a new proof from the external committer: ˆ Emulate the proof by requesting for design 0 from outside committer. ˆ Forward one of the two proofs internally. 55 / 67

Rewinding Phase Procedure II ˆ If A sends a challenge for a proof whose rst message occurs in ρ: ˆ Cancel the execution, rewind to ρ and continue. ˆ If β ρ β ρ extract witness w from ( α ρ, β ρ, γ ρ) and (α ρ, β ρ, γ ρ ). Otherwise halt and output fail. ˆ If w = (v, r) is a valid commitment for interaction k, then set ˆv k = v. Otherwise halt and output fail. Note: since right interaction l + 1 to m all have their Stage 2 and 3 occurring after τ, none of the rewinding can make A request a new commitment from the external committer. 56 / 67

Safe-point Lemma I Lemma In any one-many man-in-the-middle execution with m right interactions, for any right interaction k [m], such that it has a dierent identity from the identity of the left interaction, there exists a safe-point for interaction k. ˆ Consider a one-many man-in-the-middle execution, where the identities in the left and right interaction are dierent. ˆ Assume for contradiction, that there is some right interaction k which does not have a safe-point ˆ Every prex of is not a safe-point for interaction k. ˆ Consider any proof (α r, β r, γ r ) in the right interaction k. Let ρ be the prex after which β r is sent immediately. 57 / 67

Safe-point Lemma II ˆ By assumption, there exists a proof (α l, β l, γ l ) in the left interaction, such that α l occurs before ρ, β l occurs after ρ and before γ l (β l occurs in between β r, and γ r ). ˆ In this case a left proof is associated with a right proof. ˆ Note: Each left proof can be associated with at most one right proof. ˆ Since interaction k doesn't have a safe-point, the proofs in the left and right interactions must match up each other one by one. ˆ There is a position j where the identities in the left and right interactions are dierent. 58 / 67

Safe-point Lemma III ˆ Let the j th bit in the left be b and that in the right be 1=b. ˆ In the j th round of Stage 3 of the protocol, the left interaction has design b followed by design 1=b; and the right interaction has design 1=b followed by design b. ˆ Since all the proofs are matched up, there must be a design 0 on the left that is matched with a design 1 on the right, 59 / 67

Safe-point Lemma IV ˆ Consider ρ to be the prex that includes all the message up until β l 1. ˆ Consider the second proof (α r 2, βr 2, γr 2 ); 60 / 67

Safe-point Lemma V ˆ there is no proof on the left having its rst message before ρ and its challenge before γ2 r at the same time. It is a contradiction to the assumption that there is no safe-point for that right interaction. 61 / 67

Output distribution I Claim: Assume that R does not output fail, then except with negligible probability, its output is identical to the values committed to by A in the right interactions combined with its view. ˆ Since in the Main Execution Phase, R feeds A messages according to the correct distribution, the view of A in the simulation by R is identical to the view of A in a real interaction. 62 / 67

Output distribution II ˆ For every convincing right interaction k > l that has a dierent identity, R rewinds that interaction and eventually will either output fail or a witness is extracted from the rewinding phase of R. ˆ Conditioned on R not outputting fail, by the statistical-binding property, except with negligible probability the witnesses extracted by R are the values committed to by A. 63 / 67

Failure Probability I Claim: R outputs fail with negligible probability. R outputs fail only in the following cases: ˆ R runs for more than 2n steps: The expected running time of R is poly(n). Therefore, the probability that R runs more than 2 n steps is at most poly(n) 2 n (Markov). 64 / 67

Failure Probability II ˆ The same proof transcript is obtained from some safe-point: This case occurs if R picks some challenge β in the Rewinding Phase that appeared as a challenge in the Main Execution Phase. ˆ As R runs for at most 2 n steps, it picks at most 2 n challenges. Furthermore, the length of each challenge is 2n. Therefore, the probability that a β is picked twice is at most 2 n 2 2n (Union bound). ˆ Since there are at most polynomially many challenges picked in the Main Execution Phase, the probability that it outputs fail in this case is negligible. 65 / 67

Failure Probability III ˆ The witness extracted is not a valid decommitment: Suppose, the witness extracted is not the decommitment information, then by the special-sound property it follows that it must be a value r such that f (r) = s. ˆ If this happens with non-negligible probability, then we can invert the one-way function f. ˆ Given A, z and v; A on input y, picks uniformly at random from Γ(A, z) and proceeds identically as R with inputs τ,z = z τ... with the exception that it picks a random right interaction, say k, and feeds y as the Stage 1 message in that interaction. ˆ On the left interaction it honestly commits to the string v using Ĉ, ˆR. ˆ Finally, if f (r ) = y then A outputs r. 66 / 67

Questions? Thank you for listening! 67 / 67