Paper 2026/2936

Aggregating ML-DSA Signatures Without Recursion: A Cross-Field Argument for the Commitment Hash

Jipeng Zhang, National University of Singapore
Jiaheng Zhang, National University of Singapore
Abstract

Consensus protocols must compress tens of thousands of signatures per block, yet ML-DSA (FIPS 204) signatures are kilobytes long and lack native aggregation. Aerie aggregates Falcon signatures into a single succinct proof per batch without recursion, using a sum-check-based polynomial interactive oracle proof (PIOP) over the Akita lattice-based polynomial commitment. Extending this approach to ML-DSA is challenging: the hash that its verifier recomputes from the response and hint takes seven Keccak-f permutations, or nine for ML-DSA-87, which are expensive over prime fields. We address this bottleneck by proving ML-DSA's algebraic and Boolean components over separate fields, supporting all three FIPS 204 parameter sets. A PIOP over a 64-bit prime field handles the lattice relation, hint verification, and range checks, while Flock proves Keccak over GF(2). We introduce a cross-field parity bridge to enforce bitwise consistency between the two commitments. Using seeded sparse matrices, the bridge detects any fixed nonzero difference except with probability below 2^-144.6, independently of batch size. The aggregate certifies that every signature passes FIPS 204 verification. In the classical random-oracle model, each Fiat-Shamir challenge incurs at most 2^-128 soundness error per query, as in Aerie, and aggregate unforgeability reduces to the strong unforgeability of ML-DSA. On 16 AMD Zen 5 cores with AVX2, our prover aggregates 32,768 ML-DSA-44 signatures in 2.68 s, at 49.2 bytes per signature. For ML-DSA-87, it takes 4.13 s at 82.7 bytes per signature. Against the Johnson-regime variant of Alloul26, a prime-field ML-DSA-87 aggregator, our prover is 43x faster and our aggregates are 1.8x smaller.

Metadata
Available format(s)
PDF
Category
Cryptographic protocols
Publication info
Preprint.
Keywords
ML-DSAsignature aggregationpost-quantum signaturesSNARKsum-checklattice-based commitments
Contact author(s)
jp-zhang @ outlook com
jhzhang @ nus edu sg
History
2026-11-22: approved
2026-11-18: received
See all versions
Short URL
https://ia.cr/2026/2936
License
Creative Commons Attribution
CC BY

BibTeX

@misc{cryptoeprint:2026/2936,
      author = {Jipeng Zhang and Jiaheng Zhang},
      title = {Aggregating {ML}-{DSA} Signatures Without Recursion: A Cross-Field Argument for the Commitment Hash},
      howpublished = {Cryptology {ePrint} Archive, Paper 2026/2936},
      year = {2026},
      url = {https://eprint.iacr.org/2026/2936}
}
Note: In order to protect the privacy of readers, eprint.iacr.org does not use cookies or embedded third party content.