This section instantiates the re-randomised protocol of §5 over
the ciphersuite FROST(Pallas, BLAKE2b-512) and composes it with the Orchard
protocol. After the ciphersuite comes the section’s main result: the
aggregate is an ordinary RedPallas spend-authorisation signature under the
shifted group key, and consensus accepts it for an Action exactly when the
effective randomiser is the Action’s randomiser. There follow the sign
normalisation that makes a threshold group key a spend validating key; the
key tree of ZIP 2005 for an
Orchard-protocol threshold key, with the lemma that its common secrets are
simulable; the order of randomiser, randomised validating key, signature
digest and proof that the main result forces; and the threshold construction
of an Action in that order. Throughout,
For a point P of the Pallas group write
The ciphersuite FROST(Pallas, BLAKE2b-512) is the following instance of Definition 4.1, for re-randomised signing.
Group. The Pallas group E (F p Pallas) , of
prime order
| B = G SpendAuth = GroupHash (z.cash:Orchard, G) |
(Ironwood Guide, §“GroupHash, domain separation, and
nothing-up-my-sleeve generators”; Definition “Spend-authorisation and
binding-signature instances”), ZIP 312’s G Orchard .
A random scalar is uniform on
Elements.
Scalars.
Hashes. BLAKE2b-512 with a 16 -byte
personalisation,
Signature encoding and verification as in Definition 4.1, part (e), with these data.
Randomiser hash. The role H R of Table 2.
The data are those of ZIP 312, “FROST(Pallas, BLAKE2b-512)”, which takes
the group, the order and the encoding repr P from the
protocol specification, §“Pallas and Vesta”, and the base point from
§“Spend Authorization Signature (Sapling and Orchard)”. The generic element
serialisation of RFC 9591, Section 3.1, fails on the identity; this one, as
ZIP 312 states it, does not. Three consequences for
Theorem 4.11 follow. Its case (2),
The two texts conflict. RFC 9591, Section 3.1, requires every element serialisation to fail on the identity; ZIP 312 defines this suite’s as repr P and is silent on the identity; and the reddsa crate, which implements the suite, fails on it. Under the reading of RFC 9591 the serialisation of part (b) is not total: cases (2) and (3) of Theorem 4.11 and the identity case of the shifted group key in Theorem 5.11 apply to this suite, the bound of case (2) being that stated after Theorem 4.11; and Proposition 7.8, which assumes a total serialisation, does not cover it. ZIP 312 does not settle which reading holds. This volume follows the text of ZIP 312: each statement that this suite’s serialisation is total, and each result resting on one, holds under that reading only.
| Role | Personalisation | Output |
|---|---|---|
| H 1 , binding factor | FROST_RedPallasR | ToScalar |
| H 2 , challenge | Zcash_RedPallasH | ToScalar |
| H 3 , nonce | FROST_RedPallasN | ToScalar |
| H 4 , message pre-hash | FROST_RedPallasM | 64 -byte digest |
| H 5 , commitment-list pre-hash | FROST_RedPallasC | 64 -byte digest |
| H R , randomiser | FROST_RedPallasA | ToScalar |
The reduction ToScalar is ZIP 312’s “interpreting the 64 bytes as a little-endian integer, and reducing the resulting integer modulo G.Order()”. ZIP 312 records the equality of H 2 and H ⊛ (protocol specification, §“RedDSA, RedJubjub, and RedPallas”).
Assume Assumption 4.2 in its form for a suite built from
one hash function under distinct personalisations: the map
the hashes H 1 , H 2 , H 3 , H 4 , H 5 and H R are independent random oracles;
each value of H 1 , H 2 , H 3 and H R at a new input is within statistical distance
| δ 512 := p Vesta / 2 512 < 2 − 257 |
of a uniform scalar, so the suite meets Assumption 4.2
with
the hashes H 4 and H 5 are random oracles onto the 64 -byte strings, so Q queries find a collision of either with probability at most Q 2 / 2 513 ;
(i) The six personalisations are pairwise distinct and all 16 bytes
long, hence prefix-free; Crypto Guide, §“Domain separation and
personalisation”, Proposition “Domain separation yields independent
oracles”, gives independence. (ii) A value at a new input is
ToScalar of a uniform 64 -byte string, whose distance from
uniform is at most p Vesta / 2 512 by Math Guide, §“Uniform
sampling and the bias of modular reduction”, Proposition “Bias of modular
reduction”, with
At
Let valid key material (Definition 2.5) with group key
Y be given over FROST(Pallas, BLAKE2b-512), let
| σ := ser G (R) ∥ ser F q (z) |
(RFC 9591, Appendix A) is accepted by the validation of the
spend-authorisation instance of RedPallas (Ironwood Guide,
Construction “RedPallas”, part 3, with base G SpendAuth )
under
Theorem 5.11 gives
| [z] B = R + [c] X with c = H 2 (ser G (R) ‖ ser G (X) ‖ m) |
(RFC 9591, Section 4.6), its only exception in this suite being the zero nonce, since ser G is total. Validation of σ under X on m proceeds as follows.
The first 32 bytes are
The last 32 bytes give
Validation computes
H ⊛ (R ̄ ‖ X ̄ ‖ m) , where
The equation
How R was produced does not enter validation. These are the conditions that ZIP 312, “Rationale”, lists. □
ZIP 312, “FROST(Pallas, BLAKE2b-512)”, states that the suite is meant to produce signatures “indistinguishable from RedPallas Orchard Spend Authorization Signatures”. This volume reads that phrase as the statement of format and validation above; the distributional statement, for randomisers of Construction 5.6, is Proposition 7.24.
Let
| Y + [α ^] G SpendAuth = rk ⟺ α ^ = α in F p Vesta . |
Consequently:
if
if
the consensus rule
| 1 / p Vesta + δ 512 + q h ⋅ 2 − 256, |
with q h the number of queries to H R made by the party that fixed the key (Lemma 5.7).
Hence, up to these probabilities, consensus accepts the aggregate as the Action’s spend-authorisation signature if and only if the effective randomiser equals the Action’s α .
The map
For FROST(Pallas, BLAKE2b-512) the term
No consensus rule refers to FROST or to a threshold key. The rules that bear
on spend authorisation,
The spend validating key ak P of an Orchard-protocol key has
even y -coordinate: the last bit, the y ~ bit, of
repr P (ak P) is 0 . For keys with
Let shares sk j , coefficient commitments
The polynomial − f has degree at most
On the toy curve of Example 4.12, the negated material has
Input: valid key material output by Construction 3.2 or by Construction 3.6. Every party computes PK from public data (Lemma 2.6).
If
Otherwise, if the last bit of repr P (PK) is 1 , every participant P j replaces its share sk j by − sk j , and every party replaces each A k , each PK j and PK by its negative.
The test reads public data only, so all parties decide identically without
interaction. Output: the resulting key material and
Validity of the output and even y of ak P follow from
Lemma 6.6;
The Ironwood Guide constructs only keys with
This is the branch
Run a key generation (§3) and
Construction 6.7; let
The participants privately agree on a spending key sk , a 32 -byte string drawn uniformly and independently of the key material.
Every participant computes
| nk | := H nk (sk) = ToBase (PRFexpand sk ([0x07])), | ||
| qsk | := H qsk (sk) = truncate 32 (PRFexpand sk ([0x0C])), | ||
| qk | := H qk (qsk) | ||
| = BLAKE3 . derive _ key (Zcash ZIP 2005 qk-derivation v1, qsk, 32), | |||
| r ivk | := H qk rivk _ ext (ak, nk) | ||
| = ToScalar (PRFexpand qk ([0x0D] ‖ I2LEOSP 256 (ak) ‖ I2LEOSP 256 (nk))), |
where H qk is the key-derivation mode of the BLAKE3 hash function with the displayed context string and a 32 -byte output, ZIP 2005’s derivation, of which no property is used below.
The remainder of § 4.2.3 is unchanged:
The protocol MUST ensure that all participants obtain the same ak , nk and r ivk , which ZIP 2005 calls rivk _ ext .
The spend authorising key
Steps (3) and (5), the agreement of step (2) and the rejection of step (4) quote ZIP 2005, “Usage with FROST” and “Changes to the Protocol Specification”, § 4.2.3; the uniform, independent draw of sk in step (2) and keeping ak P on rejection in step (4) are this volume’s reading of “privately agree” and of the repetition with a new sk . Step (1) meets the even- y clause of that section by Construction 6.7. Attempts of step (4) are independent, sk being fresh on each; the volume proves no bound on their rejection probability.
ZIP 2005 names two sources of ak P for
The Orchard pool is spend-only by consensus: no transaction increases its
value, and every Orchard-pool Action creates its note at the address of the
note it consumes (ZIP 258, Draft, “Consensus rules from NU6.3 activation”;
Ironwood Guide, §“The Orchard protocol and its two pools”). By
consensus, therefore, only a party able to authorise spends for an address
can create an Orchard-pool note to it, the consumed side of the Action being
possibly a dummy note at that address (ZIP 326, “Rationale for
key-generation restrictions”). ZIP 326 (Draft), “Wallet key-generation
restrictions”, forbids conforming wallets to send Orchard-pool funds to a
key with
ZIP 2005, “Usage with FROST”, states its constraints for keys whose
ak is derived jointly by distributed key generation, and this volume
reads its MUST as binding those keys. A dealer that shares an ask
derived from a spending key, which ZIP 312, “Key Generation”, permits,
produces an ordinary key with
For Construction 6.8, the tuple
For a dealer-shared key with
(a) Every component of V is computed from sk and ak , and the rejection test reads only sk and ak ; the construction draws sk independently of the key material, as the simulating adversary does. (b) By the Ironwood Guide’s Lemma “Independence of domain-separated expansions” (“Pseudorandom functions and field reductions”), under Assumption “Pseudorandomness of PRF expansion” and with the three distinct leading bytes 0x06 , 0x07 , 0x08 , the triple (ask, nk, r ivk) is replaced by independent uniform values at the stated cost, each reduction distance being below 2 − 257 . After the replacement nk and r ivk are independent of ask , hence of ak P , and the simulator draws them and applies the rejection test, which reads ak , nk and r ivk . No property of H qk is used. □
Section 7 uses the lemma to give the corrupt participants the common secrets. Class (Definition 1.1): ZIP 2005’s constraints on FROST key generation are specified (ZIP 2005, Proposed); how the participants privately agree on sk is designed but unspecified: ZIP 2005, “Usage with FROST”, requires only that they agree privately, no specification gives a protocol for it, and the revision of ZIP 312 proposed in pull request 895 publishes one.
Consider a signing run over an Orchard-protocol threshold key, re-randomised as in Definition 5.3 by an effective randomiser α ^ , whose aggregate consensus accepts as the spend-authorisation signature of an Action with randomiser α and randomised validating key rk , in a transaction with signature digest m . Except with the probabilities of Corollary 6.4 and of Proposition 5.5:
the randomiser
the key rk is fixed before m , which commits to it (Ironwood Guide, §“Transaction digests and signatures”, Definition “Signature digest”);
the digest m is fixed before any Round Two share, being the message of the binding factors and of the challenge;
the Action proof takes α as auxiliary input and rk as instance, so it follows (i), and it is unconstrained with respect to m and to Round Two, proofs being authorising data outside the digest.
(i) is Corollary 6.4. (ii) and (iii) are the data dependencies of the cited definition and of Construction 4.8, as run in step (3) of Construction 5.10: ρ i takes H 4 (m) and c takes m . (iv) is the input of H R in Construction 5.6. (v) follows from the Ironwood Guide’s Definition “Authorising-data digest; effecting and authorising data” and “Construction of a transaction”, where signing can precede proving (ZIP 244, “Authorizing Data Commitment”, field A.3a). A randomiser that is a function of m , as under Construction 5.4 with the digest as message, would by (i) and (ii) be a fixed point, which Proposition 5.5 bounds; so that derivation admits no order except with that negligible probability. □
Class (Definition 1.1): the order of Proposition 6.12 under Construction 5.6 is designed but unspecified; no specification composes ZIP 312 with a transaction format.
ZIP 312 derives the randomiser in Round Two (“Randomizer Generation” and
“Round Two - Signature Share Generation”), and a derived
randomiser equals an α fixed earlier, as Corollary 6.4
requires, only with probability at most
The single-signer construction is cited, not rebuilt. Each Action draws
α by GenRandom , sets
The transaction builder acts as Coordinator: it creates the Action proof and so already controls the privacy of the transaction (ZIP 312, “Threat Model”). For each Action that consumes a note of an Orchard-protocol threshold key (Construction 6.8), in the order of Proposition 6.12, it proceeds as follows.
It runs Round One (Construction 5.10, step (1)) with a signing set I of at least t participants.
It sets
Once every Action of the transaction is fixed, it computes the one signature digest m of the transaction.
At any point after step (3), it proves the Action statement with α as auxiliary input (Ironwood Guide, §“The Action statement”).
These steps replace the draw of α and the single-holder signature in the cited construction; every other Action, dummy spends included, is built as there. No party holds rsk . The builder alone makes the binding signature (Ironwood Guide, §“The binding signature”). Every threshold-spent Action has its own α and rk and its own signing run, with fresh nonces and randomiser, all on the one digest m ; since m commits to the rk of every Action, the Round One and the randomiser of every threshold-spent Action precede m , and hence every Round Two (Proposition 6.12, (i) to (iv), applied to each Action). Class (Definition 1.1): designed but unspecified; the composition of ZIP 312 with the transaction is this volume’s own.
The circuit and the consensus rules are unchanged (Remark 6.5). Both pools of the Orchard protocol use the same RedPallas spend authorisation, so one flow serves an Action of either pool (Ironwood Guide, §“The Orchard protocol and its two pools”; ZIP 229, “Reuse of the Orchard protocol with minimal changes”); which pool’s notes a key may hold is fixed in §6.4.