Pointwise reduction and zero-knowledge simulation for the 3-copy GMW protocol #
gmwViewReduction is one uniform PRG-challenge test whose ordinary input contains the unary
security parameter, instance, witness, auxiliary word and challenge. For every valid tuple,
gmwView_advantage_le bounds the real/simulated view advantage by
(7 / 8) ^ (n + 1) plus dyadicSize (n + 1) * dyadicSize (totalLen (f a)) times this test's
PRG advantage at seed parameter n + 7. No security or efficient-generation assumption is used
in that finite bound. The final view test receives the instance and auxiliary word in both games.
The simulator gmwSimulator remains witness-free.
isPPTOn_gmwViewReduction charges the complete encoded input, both hybrid selections, all
surrounding samples and the malicious verifier's adaptive computation and retained private state.
The existing uniform-family theorem gmwView_indistinguishable_gmwSimulator follows from this
finite bound and certificate under uniform PRG security, for deterministic polynomial-time
instance/witness/auxiliary families.
gmwView_negligible_of_advice_prg instead permits arbitrary families with three separate
polynomial encoding-length bounds, under an explicit stronger PRG premise against uniform
full-input tests supplied those families as advice. This module does not derive that premise
from uniform one-wayness or HILL. Unbounded auxiliary input with a clock independent of its
length, witness-level Levin reductions, and sequential soundness amplification remain separate.
The proof uses four distribution transformations:
- Guess a uniform 3-bit challenge in each of
n + 1independent attempts. The probability of no success is(7 / 8) ^ (n + 1)(advantage_repeatAttempt_le). - Factor the commitment seeds into active seeds and pseudorandom unopened pads without
changing the attempt distribution (
eval_game0Pads_eq_eval_game0). - Apply
sequenceTest_advantageonce to attempts and once to unopened pads. The resulting product of dyadic-range losses is exact (gmwViewReduction_advantage), including zero pads. - Uniform unopened pads eliminate the witness from each attempt
(
eval_game1_eq_eval_gmwSimAttempt), giving the existing witness-free simulator.
Main definitions #
Cslib.Crypto.GMW.gmwRealWithPads: a guessed-challenge attempt with supplied unopened pads.Cslib.Crypto.GMW.gmwRealPRGPads,Cslib.Crypto.GMW.gmwRealUniformPads: its two pad sources.Cslib.Crypto.GMW.gmwViewReduction: the single-challenge pointwise reduction.Cslib.Crypto.GMW.AdvicePRG: explicit advice-supplied PRG premise at seed sizen + 7for arbitrary polynomially bounded instance, witness, and auxiliary-input families.
Main results #
Cslib.Crypto.GMW.gmwView_advantage_le: unconditional finite view-advantage reduction.Cslib.Crypto.GMW.isPPTOn_gmwViewReduction: strict full-input polynomial-time certificate.Cslib.Crypto.GMW.gmwView_negligible_of_advice_prg: the stronger advice-PRG consequence.Cslib.Crypto.GMW.gmwView_indistinguishable_gmwSimulator: the existing uniform-family result.
References #
Step 1: Replacing the verifier's challenge in runTwoRound by a random guess #
Step 2: Introducing unopened pads via naorSelectPads #
Run one attempt with pre-sampled unopened pads pads and active flags
allOpenFlags φ ch_guess bits.
Instances For
Eliminating the witness: masking unopened committed bits and eliminating the witness #
A pointwise finite reduction #
A single PRG-challenge test for the complete GMW view. The ordinary input is the security
parameter n, instance a, witness w, auxiliary word z, and PRG challenge. Two independently
sampled dyadic indices select an attempt and an unopened pad. The remaining attempts and pads
are generated by this program, including the verifier's adaptive queries and private randomness.
The simulator remains witness-free; only the reduction is given w. No security or validity
assumption is needed to define this program. Its PRG challenge has seed parameter n + 7.
Instances For
Pointwise finite GMW view reduction for every valid instance, witness and auxiliary word.
The two dyadic factors are exact hybrid losses; the remaining term bounds failure of all
n + 1 challenge guesses. The PRG challenge uses the protocol's shifted seed size n + 7.
This is an unconditional reduction identity and statistical bound, not a PRG-security premise.
In particular no efficient generation of a, w or z is assumed. The test receives the same
public instance and auxiliary word in both views, and the simulator receives no witness.
The reduction has a strict uniform clock polynomial in its entire ordinary input. In
particular the encoded witness and auxiliary word are charged, without requiring either to be
generated efficiently from n. The certificate includes both hybrid indices, all surrounding
samples, and the verifier interaction; the verifier may retain its private coins in its output.
Auxiliary-input consequence under an explicitly stronger PRG premise. The three families
may be arbitrary, subject to their separate polynomial encoding-length bounds. hPRG supplies
those families as ordinary advice to every full-input PPT challenge test, at seed size n + 7.
It is not a consequence established here of uniform PRG security or uniform one-wayness.
Both the verifier and final test have clocks polynomial in their complete encoded inputs. This does not cover unbounded auxiliary words with clocks bounded only in the security parameter and common-input length. Witness-level Levin reductions and sequential amplification are separate from this single-execution masked-3SAT view theorem.
Explicit advice-PRG premise at GMW seed size n + 7 for an instance encoding enc:
generator has output length length, is computable in polynomial time, and its output at seed
length n + 7 is indistinguishable from uniform strings of length length (n + 7) against every
full-input uniform strict PPT test supplied with arbitrary polynomially bounded instance, witness,
and auxiliary-input families (a, w, z).
This stronger advice-supplied premise is not derived here from uniform PseudorandomGenerator
security or uniform OneWay security.
Instances For
Under AdvicePRG enc generator length, the real GMW interaction view and gmwSimulator
have negligible distinguishing advantage against every full-input uniform strict PPT distinguisher
across all polynomially bounded valid instance, witness, and auxiliary-input families.
Zero-knowledge simulation of the 3-copy GMW protocol. For any uniform strict oracle PPT
malicious verifier Vstar and any deterministic polynomial-time families of instances a, valid
ThreeSAT witnesses w, and auxiliary inputs z, the real interaction view of Vstar with
gmwProver is computationally indistinguishable from gmwSimulator.