Source-linked AI summary
Phase Gadget Synthesis for Shallow Circuits
Alexander Cowtan, Silas Dilkes, Ross Duncan, Will Simmons, Seyon Sivarajah
TL;DR
NISQ constraints make circuit depth and two-qubit-gate count important optimisation targets. The paper develops phase- and Pauli-gadget methods in the ZX-calculus and reports substantial average reductions on quantum-chemistry circuits, while leaving architecture-aware routing for future work.
Problem
NISQ hardware restricts circuit execution through limited qubits, coherence time, and gate fidelity, motivating optimisation of depth and two-qubit-gate count.
Method
The paper uses ZX-calculus phase-gadget structures to optimise phase and Pauli gadgets, then synthesise them into executable gate circuits.
Results
t|ket⟩ achieved average reductions of 54.5% in ∧X count and 57.7% in two-qubit-gate depth on the benchmark circuits.
Takeaways & Limitations
Phase-gadget resynthesis and Pauli-gadget optimisation can substantially reduce physical circuit costs in quantum-chemistry benchmarks.
Takeaways & Limitations
The reported empirical comparisons omit architectural constraints, and the effect on routing to fixed connectivity remains future work.
Abstract
from arXiv · showhide
We give an overview of the circuit optimisation methods used by tket, a compiler system for quantum software developed by Cambridge Quantum Computing Ltd. We focus on a novel technique based around phase gadgets, a family of multi-qubit quantum operations which occur naturally in a wide range of quantum circuits of practical interest. The phase gadgets have a simple presentation in the ZX-calculus, which makes it easy to reason about them. Taking advantage of this, we present an efficient method to translate the phase gadgets back to CNOT gates and single qubit operations suitable for execution on a quantum computer with significant reductions in gate count and circuit depth. We demonstrate the effectiveness of these methods on a quantum chemistry benchmarking set based on variational circuits for ground state estimation of small molecules.
1 Introduction
NISQ hardware imposes tight qubit, coherence, fidelity, and circuit-depth constraints, making automated optimisation essential. The paper presents t|ket⟩ methods centred on Pauli gadgets derived from phase-gadget structures for reducing physical-circuit costs.
- Motivation: NISQ devices limit executable circuits through scarce qubits, short coherence times, and imperfect gate fidelity.These restrictions constrain circuit depth before errors overwhelm computation.
- Motivation: Automated circuit optimisation is therefore essential for extracting value from NISQ hardware.
- Compiler context: t|ket⟩ retargets circuits to different quantum devices while satisfying architectural constraints and translating them to required gate sets.It supports Qiskit, Cirq, and PyQuil frameworks.
- Optimisation goals: t|ket⟩ targets total circuit depth and two-qubit-gate count because both serve as proxies for total physical error rate.
- Contribution: The paper introduces optimisation based on Pauli gadgets, which exploit symmetric structures in exponentials of Pauli strings.Pauli gadgets arise naturally when Hamiltonians are decomposed into Pauli tensors and Trotterised.
2 Circuit Optimisations
t|ket⟩ combines local rewriting with circuit representations and decomposition techniques that target single-qubit operations, two-qubit subcircuits, and Clifford structure. Its optimisation passes can improve circuits directly or enable stronger later transformations.
- Rewriting strategy: Circuit optimisation replaces recognised connected subcircuits with equivalent forms that are cheaper or enable later optimisation passes.
- Circuit representation: t|ket⟩ represents circuits as non-planar maps that preserve operation connectivity while abstracting qubit permutations and timing information.This representation admits non-commutative operations such as the ∧X gate.
- Single-qubit optimisation: Single-qubit operations can be fused and Euler-decomposed in a basis chosen to enable commutations or native-gate translation.Examples include ZXZ, XZX, and ZYZ forms.
- Two-qubit optimisation: The KAK decomposition reduces long two-qubit gate sequences to a canonical form requiring at most three ∧X gates.
- Multi-qubit optimisation: t|ket⟩ combines Clifford reduction rules with a novel optimisation technique for phase gadgets and Pauli gadgets.
3 Phase Gadgets
Phase gadgets are recursively defined multi-qubit unitary structures with simple ZX-calculus descriptions and useful decomposition, commutation, and fusion laws. Their qubit-permutation freedom enables balanced, context-aware synthesis that reduces depth and can expose further cancellations.
- Definition and role: Phase gadgets are macroscopic circuit structures that are easy to identify, synthesise, and manipulate algebraically.They are introduced as recursively defined Z-phase gadget unitary maps.
- Gadget variants: The Z-phase gadget is the paper’s default gadget, while X- and Y-phase gadgets are related by colour duality and conjugation.The paper focuses on Z-phase gadgets and refers to them simply as phase gadgets.
- Synthesis laws: The decomposition law provides a canonical circuit synthesis, while permutation invariance gives freedom to choose optimisation-friendly qubit orderings.
- Depth reduction: A naive ∧X ladder has depth 2(n −1), whereas a balanced tree achieves depth 2⌈logn⌉ for an n-qubit phase gadget.Both forms still use 2(n−1) ∧X gates.
- Context-aware optimisation: Balanced-tree synthesis exposes more ∧X gates to later optimisation and can align gates between consecutive phase gadgets.Trotterised quantum-chemistry circuits contain sequences of phase gadgets separated by single-qubit Clifford layers.
4 Pauli Gadgets
Pauli gadgets generalize phase gadgets to arbitrary Pauli tensors, supporting ZX-calculus reasoning about their synthesis and interactions. The resulting rules characterize when gadgets commute, when they obey an Euler relation, and how shared structure reduces CNOT cost.
- Definitions: Pauli gadgets represent exponentials of arbitrary Pauli tensors by conjugating phase gadgets with Clifford operators.They may be extended to full circuit width while acting trivially on selected qubits.
- Interactions: Pauli gadgets are difficult for phase-polynomial methods because Pauli evolution operators need not commute.The paper contrasts commuting cases with noncommuting examples such as X- and Z-based evolutions.
- Interactions: Clifford and ∧X commutation rules enable Pauli-gadget transformations and are derivable in the ZX-calculus.The displayed rule sets are sufficient for the paper’s optimisation procedure, although they are not exhaustive.
- Synthesis: A canonical Clifford unitary can transform Pauli-gadget strings while balanced-tree constructions expose cancellations for later optimisation.The construction uses layers of Clifford and ∧X gates, with alternative configurations chosen for circuit context.
- Interactions: Any two Pauli gadgets either commute or satisfy an Euler equation.If their reduced intersection is trivial or uses the same basis, they commute; otherwise the Euler relation applies.
- Synthesis: 2(|s|+|t|−|r|−⌊|u|/2⌋−2) ∧X gates suffice for a consecutive gadget pair under the stated shared-substring construction.Here r is the maximum common substring and u is the remaining intersection subset; depth analysis is left for future work.
5 Optimisation Example
The example optimises a hydrogen-molecule variational circuit by detecting phase-gadget structure, resynthesising it, and applying Clifford optimisation. This reduces two-qubit gates from 12 to 6 and arranges them in four layers.
- The circuit is a small region of a Unitary Coupled Cluster ansatz for estimating H2 ground-state energy, with parameters α and β variationally optimised.
- Clifford commutation and optimisation rules further reduce aligned CNOT pairs between the parametrised gates.
- 6 two-qubit gates arranged in 4 layers result after exploiting commuting Pauli gadgets and reducing the remaining pair to single-qubit rotations on different qubits.
6 Results
The evaluation compares t|ket⟩, Qiskit, and PyZX on variational quantum-chemistry circuits for small-molecule ground-state estimation. t|ket⟩ achieves the largest average reductions in both two-qubit gate count and two-qubit-gate depth, while larger cases face runtime and routing-related limitations.
- The benchmark contains Unitary Coupled Cluster circuits for H2, LiH, CH2, and C2H4 across several qubit mappings and chemical basis functions.
- The benchmark uses QASM files generated with Qiskit Chemistry, while Table 1 records original and optimised two-qubit counts and depths for each compiler.
- 54.5% average reduction in CNOT count is achieved by t|ket⟩, compared with 21.3% for Qiskit and 16.3% for PyZX.
- 57.7% average reduction in two-qubit-gate depth is achieved by t|ket⟩, compared with 21.8% for Qiskit and 30.8% for PyZX.
- All implementations encounter runtime scaling issues, leaving results for some larger circuits unobtainable.
- The empirical comparison imposes no architectural constraints, and routing effects remain future work because macroscopic resynthesis can change the interaction graph.
A Proof for Lemma 4.3
The proof derives Pauli-gadget transformation rules from ZX-calculus rewrite rules. It combines spider fusion, colour change, π-phase rules, and Hadamard-related identities to establish the required Clifford interactions.
- Green vertices commute through Z components and red vertices through X components using spider fusion and colour-change rules.
- The π-phase properties additionally use the π-copy/elimination and phase-inversion rules.
- Single-qubit Clifford gates are passed through Pauli gadgets by applying these rules in sequence.
B Proof for Lemma 4.4
The proof extends the gadget-rewriting rules to controls and Pauli components using ZX-calculus identities. Hadamard decomposition handles Y components by reducing them to the X case.
- A CNOT control commutes through a Z component of a Pauli gadget using spider fusion.
- For X components, Hadamards are removed with colour change and CNOT pairs are introduced using identity and Hopf rules.
- The Y-component rule introduces green phase vertices and Hadamards through Hadamard decomposition, reducing the proof to the X case.