Proof Generation for KECCAK & SUPER circuit fails with error: RootCircuit::new: AssertionFailure("e(lhs, g2)·e(rhs, -s_g2) == O") {'error': 'RootCircuit::new: AssertionFailure("e(lhs, g2)·e(rhs, -s_g2) == O")', 'result': 'FAILED'} #1335
Comments
AronisAt79
changed the title
|
proof generation for super circuit also fails similarly |
|
Look like this might be related to #1330 |
|
Close for inactivity |
|
KECCAK circuit proof in latest main is now passing |
|
Reopening the issue to track the failure of the Super Circuit with aggregation. |
|
The next step to continue debugging this issue is to update the zkevm-chain to use the latest release of halo2_proofs so that it can run with the current main of zkevm-circuits. On a second note, I think it would be great to add root circuit tests into the integration tests from the zkevm-circuits, so that we have easy to run integration tests that include the root circuit (I think currently we don't have any) |
### Description Introduce two new sets of tests in the Integration tests with the root circuit: one where the root circuit is verified with the MockProver and the other where a real proof is generated. The input proof to the root circuit is the output of each subcircuit test we had before. ### Issue Link Resolve #1334 Resolve #1460 ### Type of change - [x] Bug fix (non-breaking change which fixes an issue) - [x] New feature (non-breaking change which adds functionality) ### Contents Issue #1335 has been resolved by assigning the fixed columns in the ExpTable and CopyTable in their `load` methods. Previously they were either not assigned, or assigned dynamically. This was not affecting the SuperCircuit or the Exp/Copy Circuit because in those cases, the assignment is done with a different method. Only for SubCircuit tests where the SubCircuit requires the external table the `load` method is used. ### Rationale Testing a Root Circuit (no mater if it's with the MockProver or with a real proof) requires a real proof of the underlying circuit. This means that now for each SubCircuit we have 3 tests that require the SubCircuit proof: - normal real prover test - root mock prover test - root real prover test I implemented a cache to keep the SubCircuit proofs for each SubCircuit and block so that we have the possibility of running all the tests in a single run while reusing the SubCircuit proofs. ### How Has This Been Tested? I have only tested the SubCircuits (minus SuperCircuit) + Root Circuit via MockProver. - The machine I use has 64 GiB of memory, and that's not enough to generate a SuperCircuit proof - I have as a pending task to try the Root Circuit real prover But now we can run these tests under the github CI, so we can defer full testing after the PR is merged.
…ions#1335) * fix degree pow of rand table in agg circuit * change some log level from info to debug
What command(s) is the bug in?
No response
Describe the bug
Proof request for block consisting of single l2 transaction (eth withddraw)
Proof request options:
('{"jsonrpc":"2.0", "method":"proof", '
'"params":[{"block":37,"circuit":"keccak","aggregate":true,"mock":false,"mock_feedback":true,"rpc":"http://leader-testnet-geth:8545/", '
'"retry":false}], "id":37}')
Error:
RootCircuit::new: AssertionFailure("e(lhs, g2)·e(rhs, -s_g2) == O")
{'error': 'RootCircuit::new: AssertionFailure("e(lhs, g2)·e(rhs, -s_g2) == O")',
'result': 'FAILED'}
Circuit Commit: https://github.com/privacy-scaling-explorations/zkevm-circuits.git@branch=main
Chain Commit: eaab99aeb31b92ed8870548414c37148e18ac78d
Full logs:
https://zkevm-chain-testing.s3.eu-central-1.amazonaws.com/proveSingeCrossChainTx_keccak_True_372-maronis.tar.gz
Concrete steps to reproduce the bug
No response
The text was updated successfully, but these errors were encountered: