Softchains: Sidechains as a Soft Fork via Proof-of-Work Fraud Proofs
Originally posted on bitcoin-dev.
This post describes a fully decentralized two-way peg sidechain design. Activating new sidechains requires a soft fork, hence the name softchains. The key aspect is that all softchains are validated by everyone via Proof-of-Work Fraud Proofs (PoW FP) -- a slow but very efficient consensus mechanism that only requires the validation of disputed blocks. This does increase the validation burden of mainchain full nodes, but only by a minimal amount (~100MB per chain per year). It's similar to drivechains, but without the major downside of having to rely on miners, since all Bitcoin full node users can efficiently validate each sidechain.
Proof-of-Work Fraud Proofs
Last year I posted the idea of PoW FP to the Bitcoin mailing list (follow-up here). The idea is that we can use the existence of a fork in Bitcoin's PoW as evidence that a block might be invalid (i.e. a proof of potential fraud). Whenever this occurs, we download the block in question to verify whether it was valid (and available), and reject it if it was not. We forego the need for maintaining a UTXO set with UTXO set commitments (such as utreexo), by assuming that the commitment inside the last block to exist in both forks is valid. As a result, we only need to download as many blocks (and their corresponding UTXO set proofs) as there are orphans, which lowers the validation costs considerably compared to running a full node.
In the past 4 months, Forkmonitor has registered 11 stale and invalid blocks. Extrapolating from that data, a PoW FP node verifying Bitcoin consensus would have to download and verify a little over 100MB per year in order to have consensus guarantees that come close to that of a full node:
- All PoW headers (~4MB per year)
- 3 x 11 = 33 full blocks (~2MB x 33 = 66MB)
- UTXO merkle proofs (~1MB x 33 = 33MB with utreexo)
The reason consensus is considered slow, is because we need to allow time for a honest PoW minority to fork away from an invalid chain. If we assume only 1% of all miners are honest, this means consensus slows down by 100x. If you are normally satisfied waiting for 6 confirmations, you now need to wait 600 confirmations. The longer you wait, the fewer honest miners you need.
In order to have two-way pegged sidechains, you need a succinct method for proving to the mainchain that a peg-out is valid. PoW FP provides exactly that -- a low-bandwidth way of determining if a chain, and thus a peg-out, is valid. The slowness of PoW FP consensus is not an issue, as peg-outs can be made arbitrarily slow (e.g. one year).
The safest design would be a set of softchains that shares its consensus code with Bitcoin Core, with the addition of UTXO set commitments, and disabling non-taproot address types to minimize certain resource usage issues. All users validate the mainchain as usual with their full node, and all softchains are validated with PoW FP consensus. If a user is interested in directly using a specific softchain, they should run it as a full node in order to get fast consensus.
Peg-ins occur by freezing coins on the mainchain and assigning them to a softchain. Peg-outs occur by creating a mainchain transaction that points to a peg-out transaction on a softchain and waiting for a sufficient number of mainchain confirmations. If the peg-out transaction remains part of the softchain according to PoW FP consensus, the coins become spendable.
The peg-in/peg-out mechanism itself would require a soft fork (the exact design is an open question), and subsequently every softchain that gets activated will also require a soft fork.
Softchain consensus still requires a form of validation from mainchain users, which means that consensus bugs can have an adverse effect. In particular, if a softchain suffers from a non-deterministic consensus bug, it may be the case that a majority accepts a peg-out, while a minority rejects it. This specific scenario could cause a chain split in mainchain consensus. This is why it would be safest to base softchain designs on Bitcoin Core.
Similarly, it can theoretically be possible that a softchain gets a major reorg, invalidating a peg-out right as it would have become accepted on the mainchain, thus splitting consensus. The slow peg-out process makes this increasingly unlikely, but not impossible. One thing that might help (or perhaps only make it worse) is introducing a consensus rule that disallows reorgs that are bigger than half the peg-out time (e.g. half a year, if the peg-out is one year). This kind of rule does not actually solve this consensus problem, but instead pushes the problem forward so it plays out first on the softchain, giving time to take action before the problem affects the mainchain.
It is also important that each softchain produces a non-trivial amount of PoW, because if the difficulty is too low, the cost of creating forks and increasing the resource usage of PoW FP consensus goes down. It may therefore make sense to have a minimum accepted difficulty for softchain blocks (slowing down the chain when fees are not sufficient). Merged Mining could also help here, since that would allow the softchains to potentially receive the same hashrate as Bitcoin (assuming all miners participate), but of course this would also put an additional validation burden on miners.
It may turn out that the consensus risks outlined above make this prohibitively risky, but at the very least it seems worth exploring the possibilities. At a minimum it would provide more opt-in block space, and it could potentially open the door to chains with entirely different consensus rules.
Thank you for taking the time to read and comprehend my work. I will happily answer any questions and I look forward to any feedback on issues that I might have overlooked, and ideas on mitigating problems to ensure maximum safety.
Hopefully this will bring decentralized two-way peg sidechains one step closer to becoming a reality.
Happy new year, everyone.
-- Ruben Somsen
Some recent thoughts to briefly update people on my thinking. The softchains two-way peg idea, while possible in theory, is probably too exotic for Bitcoin. I now think the light client / fast IBD use case for the mainchain is a bit more realistic because:
Full nodes could skip IBD with this and then proceed to download blocks as normal. A light client could accept SPV proofs at the tip for faster confirmation of small payments (and accept a small reorg risk), and for larger payments it could download and verify the last X blocks to ensure correctness. Something like block filters will be needed as well to find relevant outputs.
The need for a UTXO set commitment (like utreexo) is the part that is probably the trickiest. There is a way to do it without a soft fork, but it's a slightly daring design and needs careful consideration to minimize DoS potential.