AI summary by TwinMind
Lecture 5: The Kilian-Micali Protocol, Part 1
MIT OpenCourseWare
1h 44m 35s
Recap of Previous Lectures
- Last week was crypto day, missed previous session
- Two weeks ago: discussed using GKR protocol to construct PCPs (Probabilistically Checkable Proofs)
- PCP concept: take NP language proof/witness, extend it (make longer), but verify by reading only few locations
- Construction approach: convert NP problem to 3-SAT, use GKR protocol on low-depth circuit checking satisfying assignment
PCP Construction from GKR Protocol
- Take 3-SAT instance, create circuit C_φ that checks if assignment satisfies formula
- Circuit has low depth (log depth for 3-SAT)
- Generate PCP by running GKR protocol "in head" on this circuit
- Write down all possible prover messages for every verifier query
- Include low-degree extension of witness in PCP
- Results in large PCP but verifiable with polylog queries
Soundness Issues with Basic PCP Construction
- Problem: malicious prover might not provide valid low-degree extension
- If prover gives high-degree function instead of low-degree extension, soundness breaks
- Solution: add low-degree test to verify witness extension has low degree
- Low-degree test checks if function is close to low-degree (agrees on 1-ε fraction of points)
Transition to Computational Soundness
- Moving from information-theoretic to computational security
- Interactive Arguments: computationally sound proof systems
- Security parameter λ governs computational assumptions
- Soundness only against polynomial-time (or T(λ)-time) adversaries
- Completeness: honest prover with witness convinces verifier with probability 1
- Computational Soundness: any poly(λ)-size cheating prover succeeds with negligible probability
Security Parameter Considerations
- Challenge: if λ = polylog(n), cheating prover limited to polylog(n) time
- May seem too weak since can't even read input of size n
- Solutions:
- Set λ = n^ε for small ε > 0
- Use stronger cryptographic assumptions (subexponential security)
- Assume adversaries bounded by 2^(λ^ε) rather than poly(λ)
Collision-Resistant Hash Functions with Local Opening
- Five algorithms: Gen, Eval, Open, Verify
- Gen(1^λ): generates hash key HK
- Eval(HK, X): outputs hash value V ∈ {0,1}^λ for arbitrary-length input X
- Open(HK, X, i): outputs bit b = X_i and opening π of size poly(λ)
- Verify(HK, V, i, b, π): accepts/rejects opening
- Correctness: honest openings always verify
- Collision Resistance: adversary cannot open same position to both 0 and 1
Kilian-Micali Protocol Construction
- Succinct interactive argument for NP using PCP + collision-resistant hash with local opening
- Protocol steps:
- Verifier generates hash key HK, sends to prover
- Prover computes PCP, sends V = Eval(HK, PCP)
- Verifier runs PCP verifier, generates queries i₁,...,iₗ
- Prover opens PCP at queried locations: sends PCP[i₁],...,PCP[iₗ] with openings
- Verifier checks PCP verification and all openings
Communication Complexity
- Hash value: λ bits
- Query indices: polylog bits each
- PCP values and openings: poly(λ) bits total
- Total communication: poly(λ) (independent of input size n)
- Verifier runtime: linear in input size (must read instance) + poly(λ) for verification
Completeness Analysis
- Straightforward: honest prover with valid witness
- PCP has perfect completeness
- Hash function has perfect correctness
- Overall protocol has perfect completeness
Soundness Analysis Approach
- Assume cheating prover P* succeeds with probability ε ≥ 1/poly(λ)
- Construct adversary A that breaks collision resistance
- A runs P* many times with same hash value but different queries
- Two cases:
- Find collision (same position opened to 0 and 1) → break hash function
- No collisions → can reconstruct accepting PCP for false statement → contradiction
Soundness Proof Issues
- Rewinding necessary: must run cheating prover multiple times
- Runtime concern: adversary A runs in time poly(n) · poly(1/ε)
- Only works if T(λ) ≥ n (security parameter assumption strength)
- Alternative: Barak-Goldreich 2002 proof avoids full PCP reconstruction
Homework Assignment Details
- Due date moved from after Thanksgiving to before (this weekend)
- Will include low-degree testing component
- Attack construction against unsound PCP variant
- Extension available upon request
Technical Implementation Notes
- PCP verifier queries can be sent as randomness rather than explicit indices
- Enables later non-interactive conversion via Fiat-Shamir paradigm
- First message (hash key) need not be random, later messages should be random
- Practical implementations often commit only to witness, then run interactive GKR
