UPA protocol specification
Version 1.2.0
Overview
Application developers register Groth16 verification keys (VKs) for their circuits with the UpaVerifier
contract (through the IUpaProofReceiver
interface). Upon registration, each VK is assigned a (the keccak hash of the VK).
Application Clients submit proofs and public inputs (PIs) to the UpaVerifier
contract as tuples, where is expected to be a (compressed) proof of knowledge that is an instance of the circuit with circuit id .
A single call to the contract submits an ordered list (of any size up to some implementation-defined maximum ) of these tuples. This ordered list of tuples is referred to as a Submission. Submissions of more than 1 proof allow the client to amortize the cost of submitting proofs. Note that there is no requirement for the s to match. A single Submission may contain proofs for multiple application circuits.
Each tuple in a submission is assigned:
- a unique proof id (equal to the Keccak hash of the circuit ID and PIs)
Each submission is assigned:
a Submission Id , computed as the Merkle root of the list of s, padded to the nearest power of 2 with
bytes32(0)
.a submission index , a simple incrementing counter of submissions, used later for censorship resistance.
Note that:
for submissions that consist of a single proof, , whereas
for submissions of multiple proofs, each proof is referred to by along with an index (or location) of the proof within the submission. Where required, a Merkle proof can be used to show that a proof with is indeed at the given index within the submission .
The proof and public inputs are not stored on-chain. The aggregator monitors for transactions submitting proofs to the contract and pulls this information from the transaction calldata. The contract stores information about the submission (including , and some further metadata), indexed by the .
There is a single Aggregator that puts together batches of proofs with increasing submission index values. The proofs in a batch must be ordered exactly as they appear within submissions. Aggregated batches do not need to align with submissions- a batch may contain multiple submissions, and a submission may span multiple batches. If a submission contains any invalid proofs, the entire submission is considered invalid. The aggregator may skip only invalid submissions. If the Aggregator skips a valid submission, it will be punished (see Censorship Resistance).
Once a submission is verified by the UPA contract, its submission id is marked as verified. Applications can confirm that an individual proof id is verified by providing a ProofReference
, which is a Merkle proof that the proof id was included in a verified submission. Note that for proofs in a multi-proof submission with , the contract does not mark the proof as verified until the entire submission has been verified.
Once the UPA contract marks a proof (or the submission containing a proof) as verified, an application client can submit a transaction to the application contract (optionally with some ProofReference
metadata), and the application contract can verify the existence of an associated ZKP as follows:
The application computes the public inputs for the proof, exactly as it would in the absence of UPA.
The application contract calls
isProofVerified
on theUpaVerifier
contract, passing in the public inputs , the circuit Id , and aProofReference
(when required).The
UpaVerifier
contract computes from the public inputs and then checks thatProofReference
contains a valid Merkle proof that belongs to a verified submission.The
UpaVerifier
returnstrue
if it has a record of a valid proof for , andfalse
otherwise.
Application contracts can also verify the existence of multiple ZKPs belonging to the same submission. In this case:
Application contract computes an array of public inputs where the -th entry corresponds to the -th proof of a submission with .
Application contract submits an array of tuples to the UPA contract.
The UPA contract computes the (unique) corresponding to the submitted array of circuit ids and public inputs.
The UPA contract returns 1 if it has verified the submission (i.e. it has verified all of the proofs within ), and 0 otherwise.
Note that in this case, there is no need to submit a ProofReference
.
Protocol
Circuit registration
Before submitting proofs on-chain, the application developer submits a transaction calling the registerVK
method to the UPA contract (through the IUpaProofReceiver
interface), passing their verification key .
The circuit's is computed as
where denotes a domain tag derived from a string describing the context, such as UPA Groth16 circuit id
(See the Universal Batch Verifier specification for details.)
is stored on the contract (for censorship resistance) in a mapping indexed by , and the aggregator is notified via an event. This will be used to reference the circuit for future operations.
Application proof submission
The application client creates the parameters for its smart contract as normal, including one or more proofs and public inputs . It then passes these, along with the relevant (pre-registered) circuit Ids , to the submit
method on the IUpaProofReceiver
interface, paying the aggregation fee in ether:
The UpaProofReceiver.submit
method:
computes for .
computes a
proofDigest
for each proof, ascomputes the submission Id as the Merkle root of the list (padded as required to the nearest power of 2)
computes the
digestRoot
as the Merkle root of the list (again padded as required to the nearest power of 2)rejects the tx if an entry for already exists
assigns a to the submission (using a single incrementing counter)
assigns a to each (using a single incrementing counter)
emits an event for each proof, including
updates the contract state to record the fact that a submission with id has been made, mapping it to
digestRoot
, , and the block number at submission time.
NOTE: Proof data itself does not appear in the input data used to compute . This is because when the proof is verified by the application, the application does not have access to (and does not require) any proof data. The application is only verifying the existence of a valid proof for the given circuit and public inputs.
NOTE: Application authors must ensure that the public inputs to their ZKPs contain some element that is hard to compute without the corresponding private witness (and in general this will already be the case for sound protocols, in order to prevent replay attacks). If the set of public inputs can be predicted by a malicious party, that malicious party can submit an invalid proof for the public inputs, preventing submission of further (valid) proofs for that same set of public inputs.
Aggregated proof submission
There is a single (permissioned) Aggregator that submits aggregated proofs to the Upa.verifyAggregatedProof
method. Each aggregated proof attests to the validity of a batch of application proofs. In return, the aggregator can claim submission fees (for on-chain submissions). An aggregated batch may contain proofs from both on-chain and off-chain submissions, as well as dummy proofs which are used to fill partial batches.
proof
- An aggregated proof for the validity of this batch.
proofIds
- The list of proofIds that are verified by the aggregated proofproof
. These are assumed to be arranged in the order: [On-chain, Dummy, Off-chain]. Furthermore, it is assumed that if there are dummy proofIds in this batch, these appear after the last proof in a submission. I.e. where dummy proof ids are used, the on-chain proof ids do not end with a partial submission.
numOnChainProofs
- The number of proofIds that were from on-chain submissions. This count includes dummy proofs.
submissionProofs
- An array of 0 or more Merkle proofs, each showing that some of the entries inproofIds
belong to a specific multi-proof on-chain submission. These are required as we do not have a map fromproofId
tosubmissionId
orsubmissionIdx
. See the algorithm below for details.
offChainSubmissionMarkers
- Represents abool[]
marking each off-chain member ofproofIds
with a 0 or 1. A proofId is marked with a 1 precisely when the proofId is the last one in an off-chain submission. Thisbool[]
is packed into auint256
to compress calldata.
The UpaVerifier
contract:
checks that
proof
is valid forproofIds
for each in
proofIds
:skips if it corresponds to a dummy proof,
checks that has been submitted to the contract, and that proofs appear in the aggregated batch in the order of submission (see below)
marks as valid (see below)
if is the last proof in a submission , emit an event indicating that the submission has been verified
Specifically, the algorithm for verifying (in the correct order) submissions of proofIds
and marking them as verified, is as follows.
State: the contract holds
a dynamic array
uint16[] numVerifiedInSubmission
of counters, where the -th entry corresponds to the number of proofs that have been verified (in order) of the submission withthe submission index
lastVerifiedSubmissionIdx
of the last submission from which a proof was verified.
Given a list of proofIds
and submissionProofs
, the contract verifies that proofIds
appear in previous submissions as follows:
For each in
proofIds
:If corresponds to a dummy proof, then the rest of the proofs in the batch are assumed to be dummy proofs. No more proofs from this batch will be marked as valid.
Attempt to lookup the submission data (see Proof Submission) for a submission with Id . If such a submission exists:
The proof was submitted as a single-proof submission. The contract extracts the from the submission data and then checks that is greater than or equal to
nextSubmissionIdxToVerify
. If not reject the transaction.The entry
numVerifiedInSubmission[
]
should logically be 0 (this can be sanity checked by the contract). Set this entry to 1Update
nextSubmissionIdxToVerify
in contract state
Otherwise (if no submission data was found for )
the proof is expected to be part of a multi-proof submission with
nextSubmissionIdxToVerify
.Note that if a previous aggregated proof verified some subset, but not all, of the entries in the submission,
nextSubmissionIdxToVerify
would still refer to the partially verified submission at this stage. In this case,numVerifiedInSubmission[
]
should contain the number of entries already verified.
Take the next entry in
submissionProofs
. This includes the following information:the for the submission to be verified
a Merkle "interval" proof for a contiguous set of entries from that submission.
Determine the number
m
of entries inproofIds
, including the current , that belong to this submission, as follows:Let
numProofIdsRemaining
be the number of entries (including ) still unchecked inproofIds
.Look up the submission data for , in particular and .
Let
numUnverifiedFromSubmission =
- numVerifiedInSubmission[
]
.The number
m
of entries fromproofIds
to consider as part of is given byMin(numUnverifiedFromSubmission, numProofIdsRemaining)
.
Use the submission Id and the Merkle "interval" proof from the submission proof, to check that the hashes of the
m
next entries fromproofIds
(including ) indeed belong to the submission . Reject the transaction if this check fails.Increment the entry
numVerifiedInSubmission[
]
bym
, indicating thatm
additional proofs from the submission have been verified.update
nextSubmissionIdxToVerify
in the contract state
NOTE: The arguments
offChainSubmissionMarkers
andnumOnchainProofs
are there for future off-chain submission support. For now, aggregators call this function withnumOnchainProofs = BATCH_SIZE
, which will skip the off-chain logic of this function.
Proof verification by the application
The application client now creates the transaction calling the application's smart contract to perform the business logic. Since the proof has already been submitted to UPA, the proof is not required in this transaction. If the proof was submitted as part of a multi-entry submission, the client must compute and send a ProofReference
structure indicating which submission the proof belongs to, and its "location" (or index) within it.
The application contract computes the public inputs, exactly as it otherwise would under normal operation, and queries the isProofVerified
on the UpaVerifier
contract (using the ProofReference
if given) to confirm the existence of a corresponding verified proof.
For proofs from single-entry submissions, the UPA provides the entry points:
For proofs from multi-entry submissions, the UPA provides entry points:
The UPA contract:
receives or computes from the public inputs
(using the
ProofReference
if necessary) confirms that belongs to a submission .Checks if there was an on-chain submission for , and if so reads the stored submission index and the total number of proofs
numProofs
contained in the submission . If it finds thatnumVerifiedInSubmission[
] == numProofs
then the submission was verified, and therefore so was the proof .
The application contract can also look up the verification status of entire submissions by computing the corresponding (nested) array of public inputs. The contract can then either use a submissionId computed from this array, or the array itself, to query the submission's status in the UPA contract.
The UPA provides the entry points:
The UPA contract:
receives or computes from the public inputs
Looks up the number of proofs
numProofsInSubmission
in and then checks ifnumVerifiedInSubmission[
] = numProofsInSubmission
.
Censorship resistance
A censorship event is considered to have occurred for a submission with Id (with submission index , consisting of entries) if all of the following are satisfied:
a submission with Id has been made, and all proofs in the submission are valid for the corresponding public inputs and circuit Ids
some of the entries in remain unverified, namely
numVerifiedInSubmission[
] <
one or more proofs from a submission with index greater than (the submission index of the submission with id ) have been included in an aggregated batch
namely, there exists s.t.
numVerifiedInSubmission[
] > 0
Note that, if one or more entries in a submission are invalid, aggregators are not obliged to verify any proofs from that submission.
Censorship by the Aggregator can be proven by a claimant, by calling the method:
providing:
the valid tuple , or
circuitId
,proof
andpublicInputs
, the claimed next unverified entry in the submissionor
submissionId
or
laterSubmissionIdx
A Merkle proof that (computed from and belongs to the submission (at the "next index" - see below)
A Merkle proof that belongs to the submission's
proofDigest
entry (at the "next index" - see below)
On receipt of a transaction calling this method, the contract:
checks that the conditions above hold and that the provided proof has indeed been skipped
looks up the verification key using and performs the full proof verification for . The transaction is rejected if the proof is not valid or if the verification key hasn't been registered.
increments the stored count
numVerifiedInSubmission[
]
The aggregator is punished only when all proofs in the submission have been shown to be valid. As such, after the above, the contract:
checks the condition
numVerifiedInSubmission[
] == n
(wheren
is the number of proofs in the original submission ).if this final condition holds then validity of all proofs in the submission has been shown and the aggregator is punished.
Note: proofDigest
is used here to prevent malicious clients from submitting invalid proofs, forcing aggregators to skip their proofs, and then later providing valid proofs for the same public inputs. This would otherwise be an attack vector since is not dependent on the proof data.
Collecting Aggregation Fees
The application contract pays an aggregation fee at submission time. These fees are held in the UPA contract. In order for the aggregator to claim the fees for a given submission, the UPA contract must have verified that submission.
The aggregator collects fees in two steps. First it calls
which stores the current value of lastSubmittedSubmissionIdx
and allocates all fees collected up to now to be claimable by the aggregator once it has verified the submission at lastSubmittedSubmissionIdx
(which implies that all previous submissions have also been verified). Once the aggregator has done this, it can call
to withdraw the previously allocated fees.
Circuit Statements
Batches of application proofs are verified in a batch verify circuit.
A keccak circuit computes all s and s of application proofs appearing in the batch verify proof, along with a final digest (the keccak hash of these s, used to reduce the public input size of the outer circuit below).
A collection of batch verify proofs along with the keccak proof for their s, s and final digest is verified in an outer circuit.
On-chain verification of an outer circuit proof thereby attests to the validity of application proofs with given s.
- inner batch size. Application proofs per batch verify circuit.
- outer batch size. Number of batch-verify circuits per outer proof.
- the maximum number of public inputs for an application circuit.
Batch Verify Circuit: Groth16 batch verifier
The batch verify circuit corresponds to the following relation:
Public inputs:
where
is the public inputs to the -th proof
is after zero-padded to extend it to length
- application verification keys, each padded to length
Witness values:
- application proofs
Statement:
for
where
is the truncation of the size verification key to a verification key of size , and
is the truncation of the public inputs to an array of size
Keccak Circuit: ProofIds and Final Digest
Computes the for each entry in each application proof in one or more verify circuit proofs.
Public inputs:
where
is the public inputs to the -th proof
is after zero-padded to extend it to length
- application verification keys, each padded to length
(digest, which consists of 32 bytes and is represented by two field elements)
Witness values: (none)
Statement:
Outer Circuit: Recursive verification of Batch Verifier and Keccak circuits
This step aggregates batch verify proofs as well as a single corresponding Keccak proof .
Public Inputs:
- final 32-byte public input digest, encoded as
- overall KZG accumulator, encoded as 12 = 4 * \texttt{num_limbs} points of
Witness values:
: the number of public inputs, the padded verifying key, and padded public inputs for the -th application proof in the -th BV proof.
for BV proofs
the Keccak proof for the public inputs
, and
"Equivalent Statement": (actual statement is shown as multiple sub-statements, given below)
All BV proofs are valid, and therefore there exist valid application proofs for each : for
Keccak proof is valid, and therefore is the final digest for all application PIs and vk hashes:
Actual Statement:
"Succinct" Plonk verification () namely "GWC Steps 1-11" using Shplonk, without final pairing, for random challenge scalar :
Verification: The EVM verifier does the following, given .
for random challenge scalar
Note that:
The same witness values are used to verify and , implying that is indeed the commitment to all application public inputs and circuit IDs.
The outer circuit does not include the pairing checks, therefore its statement is not that the BV/Keccak proofs are valid, but rather that they have been correctly accumulated into a single KZG accumulator . Checking that , for random scalar , therefore implies their validity.
In the case there is a Pedersen commitment point for proofs coming from e.g. gnark, the statements of the batch verifier and keccak circuits are a bit different. For each application proof:
[Batch verifier circuit] The Pedersen proof is verified: , where
is the Pedersen verification key (which is part of the corresponding app ).
is the Pedersen commitment point and the corresponding Pedersen proof of knowledge.
[Keccak circuit] The last public input is computed as the keccak hash of the commitment point: . Note that this last public input is not used in the computation of the proof Id.
Last updated