EUF-CMA security of Fiat-Shamir with aborts #
Statistical CMA-to-NMA reduction for the Fiat-Shamir-with-aborts transform,
matching Theorem 3 of Barbosa et al. (CRYPTO 2023). Instantiates
FiatShamir.signHashQueryBound at the with-aborts signature type and exposes
cmaToNmaLoss plus euf_cma_bound / euf_cma_bound_perfectHVZK.
The scheme-specific NMA-to-hard-problem reduction lives with each concrete
scheme (e.g. MLDSA.nma_security).
The exact classical ROM statistical loss from the Fiat-Shamir-with-aborts
CMA-to-NMA reduction (Theorem 3, CRYPTO 2023), parameterized by the HVZK simulator
error ζ_zk.
The paper proves
`Adv_EUF-CMA(A) ≤ Adv_EUF-NMA(B)
- 2·qS·(qH+1)·ε/(1-p)
- qS·ε·(qS+1)/(2·(1-p)^2)
- qS·ζ_zk
- δ`
where:
qS: number of signing-oracle queriesqH: number of adversarial random-oracle queriesε: commitment-guessing boundp: effective abort probabilityζ_zk: total-variation error of the HVZK simulator for one signing transcriptδ: regularity failure probability
The qH + 1 term comes from applying the paper's hybrid bounds to the forging
experiment, which adds one final verification query to the random oracle.
Instances For
CMA-to-NMA reduction for Fiat-Shamir with aborts (Theorem 3, CRYPTO 2023).
For any EUF-CMA adversary A making at most qS signing-oracle queries and qH
random-oracle queries, there exists an NMA reduction such that:
Adv^{EUF-CMA}(A) ≤ Adv^{EUF-NMA}(B) + L
The reduction uses:
- The quantitative HVZK simulator
simto answer signing queries without the secret key - Commitment recoverability
recoverto map between the standard and commitment-recoverable variants of the signature scheme - Nested hybrid arguments over ROM reprogramming (accepted and rejected transcripts)
The statistical loss L involves the commitment guessing probability ε, the effective
abort probability p, the simulator error ζ_zk, the regularity failure probability δ,
and the query bounds qS, qH; it is captured here by cmaToNmaLoss.
The scheme-specific reduction from NMA to computational assumptions (e.g., MLWE +
SelfTargetMSIS for ML-DSA) is stated separately; see MLDSA.nma_security and
MLDSA.euf_cma_security.
Perfect-HVZK special case of euf_cma_bound, where the simulator contributes no
qS · ζ_zk loss term.