QuantumClifford.jl
| Documentation |
|
| Continuous integration |
|
| Code coverage |
|
| Static analysis with |
|
A Julia package for working with quantum stabilizer states and Clifford circuits that act on them. Graphs states are also supported. The package is already very fast for the majority of common operations, but there are still many low-hanging fruits performance-wise. See the detailed suggested readings & references page for background on the various algorithms.
To install it use:
] add QuantumCliffordWorks efficiently with pure and mixed stabilizer states of thousands of qubits as well as sparse or dense Clifford operations acting upon them.
Provides canonicalization, projection, and generation operations, as well as partial traces.
julia> P"X" * P"Z"
-iY
julia> P"X" ⊗ P"Z"
+ XZ
julia> S"-XX
+ZZ"
- XX
+ ZZ
julia> tCNOT * S"-XX
+ZZ"
- X_
+ _Z
The code is vectorized and multithreaded.
Fast, in-place, allocation free implementations.
Quick Benchmarks (click to expand)
Comparison against other Clifford simulators
The only other simulator of similar performance I know of is Stim. In particular, Stim implements convenient tracking of Pauli frames, that makes simulating the performance of error correcting codes blazingly fast (which are possible in QuantumClifford.jl, but no convenient interface is provided for that yet).
The "low level" functionality is of similar performance in Stim and QuantumClifford but different tradeoffs are made at the higher levels: to multiply in-place 1M-qubit Pauli operators Stim needs 16μs while QuantumClifford.jl needs 14μs. The difference is inconsequential and depends on compilers and hardware.
Of note is that Stim achieved this performance through high-quality C++ SIMD code of significant sophistication, while QuantumClifford.jl is implemented in pure Julia.
Multiplying two 1 gigaqubit Paulis in 32 ms
julia> a = random_pauli(1_000_000_000);
julia> b = random_pauli(1_000_000_000);
julia> @benchmark QuantumClifford.mul_left!(a,b)
BenchmarkTools.Trial: 155 samples with 1 evaluation.
Range (min … max): 32.074 ms … 32.425 ms ┊ GC (min … max): 0.00% … 0.00%
Time (median): 32.246 ms ┊ GC (median): 0.00%
Time (mean ± σ): 32.247 ms ± 63.427 μs ┊ GC (mean ± σ): 0.00% ± 0.00%
▃ ▃▃ ▄ ▆▄▄▄██▃ ▃▄▁▆█▃▃ ▃ ▁
▄▁▁▄▁▁▄▆▁▁▄▆▄▆▇▇█▄▄██▄█▆███████▇███████▆█▆▄▄▄▁▄█▁▄▄▁▄▁▁▁▁▁▄ ▄
32.1 ms Histogram: frequency by time 32.4 ms <
Memory estimate: 0 bytes, allocs estimate: 0.
Canonicalization of a random 1000-qubit stabilizer in 22 ms
julia> @benchmark canonicalize!(s) setup=(s=random_stabilizer(1000))
BenchmarkTools.Trial: 226 samples with 1 evaluation.
Range (min … max): 21.938 ms … 22.680 ms ┊ GC (min … max): 0.00% … 0.00%
Time (median): 22.025 ms ┊ GC (median): 0.00%
Time (mean ± σ): 22.057 ms ± 115.247 μs ┊ GC (mean ± σ): 0.00% ± 0.00%
▂▂ █▃▃▂
▄▇███████▆▇▆█▆▄▄▄▄▄▅▄▃▃▁▄▃▃▃▃▃▃▁▁▁▁▁▃▁▁▁▁▁▃▁▃▁▁▁▁▃▁▁▁▁▁▁▁▁▃▃ ▃
21.9 ms Histogram: frequency by time 22.6 ms <
Memory estimate: 32 bytes, allocs estimate: 1.
Gate application (500 CNOT gates on 1000 qubits) in 7 ms
julia> @benchmark apply!(s, gate) setup=(s=random_stabilizer(1000); gate=tensor_pow(tCNOT,500))
BenchmarkTools.Trial: 564 samples with 1 evaluation.
Range (min … max): 6.602 ms … 17.719 ms ┊ GC (min … max): 0.00% … 0.00%
Time (median): 8.411 ms ┊ GC (median): 0.00%
Time (mean ± σ): 8.865 ms ± 1.836 ms ┊ GC (mean ± σ): 0.00% ± 0.00%
▂ ▁ █
██▆▆▆▄▅▃▄▄▄▄▅▇█▇▆▄▅▃▄▃▃▃▃▃▃▁▂▂▃▁▃▃██▃▂▃▂▂▂▂▂▂▂▂▁▃▃▃▁▂▂▂▂▂▂ ▃
6.6 ms Histogram: frequency by time 13.7 ms <
Memory estimate: 13.84 KiB, allocs estimate: 111.
Sparse gate application to only specified qubits in a 1000 qubit tableau in 4 μs
julia> @benchmark apply!(s, sCNOT(32,504)) setup=(s=random_stabilizer(1000))
BenchmarkTools.Trial: 10000 samples with 9 evaluations.
Range (min … max): 3.373 μs … 252.630 μs ┊ GC (min … max): 0.00% … 53.27%
Time (median): 3.766 μs ┊ GC (median): 0.00%
Time (mean ± σ): 3.892 μs ± 2.525 μs ┊ GC (mean ± σ): 0.35% ± 0.53%
▃▆█▅▁
▁▁▁▂▃▇█████▅▃▂▂▃▃▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▂
3.37 μs Histogram: frequency by time 6.07 μs <
Memory estimate: 96 bytes, allocs estimate: 2.
Measuring a dense 1000 qubit Pauli operator in 74 μs
julia> s=random_destabilizer(1000); p=random_pauli(1000);
julia> @benchmark project!(_s,_p) setup=(_s=copy(s);_p=copy(p)) evals=1
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
Range (min … max): 69.030 μs … 144.963 μs ┊ GC (min … max): 0.00% … 0.00%
Time (median): 73.799 μs ┊ GC (median): 0.00%
Time (mean ± σ): 73.639 μs ± 4.118 μs ┊ GC (mean ± σ): 0.00% ± 0.00%
▂ ▁█▁
▃██▆▄▃▃▃▄▆▅▄▃▃███▃▂▂▂▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▂▂▂▂▂▂▂▂▂▂▂▂▂ ▃
69 μs Histogram: frequency by time 92.8 μs <
Memory estimate: 480 bytes, allocs estimate: 4.
Measuring a single qubit in a 1000 qubit tableau in 50 μs
julia> s=MixedDestabilizer(random_destabilizer(1000));
julia> @benchmark projectY!(_s,42) setup=(_s=copy(s)) evals=1
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
Range (min … max): 46.928 μs … 88.046 μs ┊ GC (min … max): 0.00% … 0.00%
Time (median): 49.934 μs ┊ GC (median): 0.00%
Time (mean ± σ): 49.776 μs ± 2.623 μs ┊ GC (mean ± σ): 0.00% ± 0.00%
▁ ▄█▂
▂▂▆██▄▄▄▄▆███▅▃▂▂▂▂▂▂▂▂▂▂▂▂▂▁▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▂▂▂▂▁▂ ▃
46.9 μs Histogram: frequency by time 63.8 μs <
Memory estimate: 464 bytes, allocs estimate: 5.
Benchmarks executed on a Ryzen Zen1 8-core CPU.
More detailed benchmarks can be seen at github.com/Krastanov/QuantumCliffordBenchmarksLog.