| leanXMSS aggregation | 1600/s |
| leanSPHINCS aggregation | 500/s |
| leanDA commitment | 3,2 MiB/s |
| 2-to-1 recursion | 140 ms |
| proving KECCAK | 700K/s |
| cheap cycles | 7,6M/s |
leanVM is designed for security:
- 125-bit ROM (62-bit QROM) soundness
- no proximity gap conjecture
- end-to-end formal verification
- a traditional hash function (Keccak-256)
- leanVM is not yet stable
- Formal statements are still under review
- The Lean statements only describe the spec. Proving the Rust implem is faithful to the spec is still WIP.
leanVM is not (yet) production ready.
machine: M4 Max MacBook Pro (12 performance cores, 4 efficiency cores, 48GB RAM, CPU only, i.e. no metal GPU)
The leanXMSS parameters are specified in leanXMSS.pdf, with a security proof in Lean 4. This is not the definitive parameter set.
cargo leanvm aggregate --xmss 1400 --log-inv-rate 2 --repeat 3aggregation, 1,400 leanXMSS signatures
cycles (VM steps) : 1,147,761 = 2^20.13
details : XOR 2^17.999 (22.8%) KECCAK 2^17.994 (22.7%) MUL 2^17.855 (20.7%) SET 2^17.365 (14.7%) DEREF 2^17.3 (14.1%) JUMP 2^15.81 (5.0%) MEMORY 2^21.081 BYTECODE 2^18.812 TOTAL_COMMITTED 2^25.589
proof size : 209.4 KiB
proving time : 0.885 s ± 0.8% peak memory 4.995 GiB
per signature : 1,581.591 signatures/s
verifying : 4.522 ms
The leanSPHINCS parameters are specified in leanSPHINCS.pdf, with a security proof in Lean 4. This is not the definitive parameter set.
cargo leanvm aggregate --sphincs 500 --log-inv-rate 1 --repeat 3aggregation, 500 leanSPHINCS signatures
cycles (VM steps) : 2,757,351 = 2^21.395
details : XOR 2^19.755 (32.1%) MUL 2^19.435 (25.7%) SET 2^19.391 (24.9%) KECCAK 2^17.976 (9.3%) DEREF 2^17.195 (5.4%) JUMP 2^16.058 (2.5%) MEMORY 2^22.043 BYTECODE 2^18.812 TOTAL_COMMITTED 2^26.422
proof size : 294.1 KiB
proving time : 0.986 s ± 0.4% peak memory 4.574 GiB
per signature : 507.34 signatures/s
verifying : 5.341 ms
cargo leanvm aggregate --blobs 16 --log-inv-rate 1 --repeat 3aggregation, 16 blobs
cycles (VM steps) : 3,020,842 = 2^21.527
details : XOR 2^20.098 (37.1%) MUL 2^19.811 (30.4%) DEREF 2^19.206 (20.0%) JUMP 2^18.052 (9.0%) SET 2^16.043 (2.2%) KECCAK 2^15.088 (1.2%) MEMORY 2^21.799 BYTECODE 2^18.812 TOTAL_COMMITTED 2^26.306
proof size : 289.9 KiB
proving time : 0.608 s ± 3.4% peak memory 4.279 GiB
blob throughput : 26.299 blobs/s, 3.287 MiB/s
verifying : 5.714 ms
cargo leanvm recursion --n 2 --xmss-per-leaf 1400 --leaf-log-inv-rate 2 --log-inv-rate 1 --repeat 3recursion 2→1, over leaves of 1,400 leanXMSS signatures, leaf rate = 1/4, recursion rate = 1/2
cycles (VM steps) : 431,087 = 2^18.718
details : MUL 2^17.185 (34.6%) XOR 2^16.726 (25.1%) DEREF 2^16.641 (23.7%) SET 2^15.409 (10.1%) JUMP 2^13.797 (3.3%) KECCAK 2^13.747 (3.2%) MEMORY 2^19.394 BYTECODE 2^18.812 TOTAL_COMMITTED 2^23.611
proof size : 270.5 KiB
proving time : 0.146 s ± 3.9% peak memory 0.917 GiB
verifying : 6.921 ms
cargo leanvm hash --log-n 17 --repeat 3Keccak GKR batch proving, 131,072 permutations, a state committed every 12 rounds
witness : 9.8 ms ± 32.7% 5.5%
commit : 18.1 ms ± 12.6% 10.0%
output claim : 2.2 ms ± 22.3% 1.2%
layers and openings : 126.6 ms ± 2.8% 70.2%
pcs opening : 23.7 ms ± 3.9% 13.1%
------------------------------------------
prove TOTAL (witness included) : 180.4 ms ± 4.5%
verify : 2.4 ms
throughput : 726,763 permutations/s ± 4.5%
proof size : 237.7 KiB (1639 scalars, 212.1 KiB of Merkle openings)
cargo leanvm fibonacci --n 2000000 --log-inv-rate 1 --repeat 3Fibonacci (in the exponent, i.e. modulo 2^64 - 1), N = 2,000,000
cycles (VM steps) : 2,040,016
details : MUL 2^20.944 (98.9%) SET 2^13.288 (0.5%) DEREF 2^12.967 (0.4%) JUMP 2^10.968 (0.1%) XOR 2^10.966 (0.1%) MEMORY 2^20.959 BYTECODE 2^10.017 TOTAL_COMMITTED 2^25.014
proof size : 262.6 KiB
proving : 0.267 s ± 2.1% 7,641,814 cycles/s peak memory 1.733 GiB
verifying : 3.85 ms
- 128-bit binary field (degree-2 tower over the 64-bit field)
- WHIR PCS, aka Ligerito, using affine combiners
- Hashcaster 2 proving Keccak via GKR and RMFE
- Binius/Binius64 ring switching, M3 arithmetisation, and more (see DP23 and DP24)
- leanerVM by Nicolas Schleicher and others
- leanDA FV by Benedikt Wagner
- VCVio, ArkLib, CompPoly, (Verified zk(E)VM) Clean (zksecurity)