Add Rustiq-based synthesis for PauliEvolutionGate

[alexanderivrii]

Summary

Calls the Rustiq's synthesis algorithm for pauli networks natively from within Qiskit.

Together with #13295 addresses #12789.
This should be rebased on top of #13295, in order to support the plugin interface structure for PauliEvolutionGates.

Rustiq is available at https://github.com/smartiel/rustiq-core

How to experiment

One can experiment with the new functionality by creating a PauliEvolutionGate and directly calling the new PauliEvolutionSynthesisRustiq on this gate:

from qiskit.transpiler.passes.synthesis.hls_plugins import PauliEvolutionSynthesisRustiq

terms = ["IX", "ZZ"]
evo_gate = PauliEvolutionGate(SparsePauliOp(terms), time=0.12)
qct = PauliEvolutionSynthesisRustiq().run(evo_gate)

Alternatively, one can create a circuit with PauliEvolutionGates and specify "rustiq" via the synthesis config interface:

terms = ["IX", "ZZ"]
evo_gate = PauliEvolutionGate(SparsePauliOp(terms), time=0.12)
qc = QuantumCircuit(3)
qc.append(evo_gate, [1, 2])
config = HLSConfig(PauliEvolution=["rustiq"])
qct = transpile(qc, hls_config=config)

Details:

The Qiskit Rust function pauli_network_synthesis calls the greedy_pauli_network function from the rustiq-core (in particular, most of the code is fully in Rust). However, the latter function only returns a sequence of Clifford gates, without the single-qubit rotations. So pauli_network_synthesis also adapts the code from Simon's private rustiq-plugin repository that simulates the circuit and inserts rotations wherever required.

Preliminary Experiments:

To be updated (or maybe it would be even best to move this discussion to the issue #12789). After running some preliminary experiments on some specific examples (e.g. see #13285) and some hamlib examples (taken from https://github.com/Qiskit/benchpress/blob/main/benchpress/hamiltonian/hamlib/), rustiq can be amazingly powerful in many cases, greatly reducing the two-qubits gate-count and depth of the resulting circuit. On the other hand, there are also cases where Rustiq performs worse, i.e. creating a circuit with more gates.

Some Concerns:

• On some of the hamlib examples Rustiq is extremely fast, yet there are also examples where it takes more than 1 hour to complete.
• In some cases the resulting circuit is not fully deterministic. I am currently trying to understand why this happens.
• So far in all the cases the circuit produced by Rustiq is equivalent to the circuit produced by the default synthesis algorithm in Qiskit, up to the global phase, however the global phase difference is not always 0. This is after taking care of global phase updates related to all-I pauli rotation gates. I am also trying to understand why this happens.

It is possible that some additional changes to Rustiq itself might be required.

ToDo:

• Docstring improvements, release notes, new plugin documentation, documenting rustiq-specific options, etc.
• Tests: make sure that the output circuits are correct and are equivalent to the circuits produced by the default algorithm, make sure that the plugins run as intended

[qiskit-bot]

One or more of the following people are relevant to this code:

• @Cryoris
• @Qiskit/terra-core
• @ajavadia

[Cryoris]

I just had a brief look, but overall this looks good to me -- I'll give a detailed review later 🙂

[coveralls]

Pull Request Test Coverage Report for Build 11722744591

Details

• 226 of 278 (81.29%) changed or added relevant lines in 9 files are covered.
• 16 unchanged lines in 4 files lost coverage.
• Overall coverage decreased (-0.006%) to 88.794%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
qiskit/transpiler/passes/synthesis/hls_plugins.py	16	20	80.0%
crates/accelerate/src/synthesis/clifford/utils.rs	2	26	7.69%
crates/accelerate/src/synthesis/evolution/pauli_network.rs	161	185	87.03%

Files with Coverage Reduction	New Missed Lines	%
crates/accelerate/src/two_qubit_decompose.rs	1	92.12%
crates/qasm2/src/expr.rs	1	94.02%
crates/qasm2/src/lex.rs	2	92.23%
crates/qasm2/src/parse.rs	12	97.15%

Totals
Change from base Build 11719787153:	-0.006%
Covered Lines:	78038
Relevant Lines:	87887

---

💛 - Coveralls

[ShellyGarion]

LGTM - thanks!

[Cryoris]

Looks pretty good to me! I've left a bunch of comments but they are all very small 🙂

[alexanderivrii]

Thanks @Cryoris for a thorough review! I think I have addressed all of your review comments (and in a few cases explained why I don't want to address it). To keep track of your comments, I was marking them as resolved as I was going through them, so you will probably want to unresolve them to see if a deeper discussion is required.

[Cryoris]

LGTM thanks for the updates!