Back to Research Overview
RESEARCH NOTE • TIER A LEAD RESEARCH
PHYSICAL QPU RESEARCH NOTE • EXP-05-MQT-BENCH

Cross-Compiler Benchmarking on IBM Heron: Q-Psi vs Qiskit O3 and TKET

AUTHORS: Q-Psi Research Team
•
BACKEND: ibm_marrakesh (Heron r2)
•
STATUS: MIXED COMPETITIVENESS
ABSTRACT

We evaluate the Q-Psi State-Space Compiler against industrial compiler toolchains Qiskit O3 and TKET across the standard MQT Bench suite on IBM Heron (ibm_marrakesh, 156 programmable qubits, Job da1ul7e3kjvs73877tag, 36,864 shots). Q-Psi compiled all 18 benchmark circuits into valid QPU executables. Physical fidelity measurements demonstrated workload-dependent competitiveness: Qiskit O3 held a small aggregate advantage (mean delta \(-0.0067\)), while Q-Psi's apparent aggregate advantage over TKET (\(+0.0398\)) was driven entirely by a severe routing failure in TKET on vqe_su2_n8 (\(+0.2696\) delta). Omitting vqe_su2_n8 shifts the Q-Psi vs. TKET delta to \(-0.0062\).

1. Experimental Setup & Workload Suite

The benchmark suite comprised 6 standard algorithmic families from the Munich Quantum Toolkit (MQT Bench): dj_n6, ghz_n8, qft_n6, graphstate_n8, vqe_su2_n8, and wstate_n6. Each circuit was compiled by Q-Psi, Qiskit (Optimization Level 3), and TKET (Default Routing), producing 18 physical QPU circuits evaluated with 2,048 shots each.

2. Physical Fidelity Comparison

CircuitQ-Psi FidelityQiskit O3 FidelityTKET FidelityΔ(Q-Psi - Qiskit)Δ(Q-Psi - TKET)
dj_n60.96120.96450.9630-0.0033-0.0018
ghz_n80.88400.89120.8890-0.0072-0.0050
qft_n60.71200.73150.7240-0.0195-0.0120
graphstate_n80.84100.84500.8435-0.0040-0.0025
vqe_su2_n80.68450.68100.4149+0.0035+0.2696
wstate_n60.79500.80450.8050-0.0095-0.0100

3. Leave-One-Out Sensitivity Analysis

AGGREGATE DELTA SENSITIVITY
Q-Psi vs Qiskit O3 (All 6):-0.0067 [-0.0123, -0.0010]
Q-Psi vs TKET (All 6):+0.0398 [+0.0329, +0.0457]
Q-Psi vs TKET (Excluding vqe_su2_n8):-0.0062 [-0.0120, -0.0006]

The sensitivity audit confirms that Q-Psi does not hold an aggregate advantage over TKET across general circuits; the positive delta was an artifact of TKET's heavy-hex routing layout on vqe_su2_n8.

4. Conclusion & Boundary

FORMAL STATUS: MIXED COMPETITIVENESS

Q-Psi demonstrates valid compilation and competitive execution across industrial benchmarks, but does NOT establish general superiority over Qiskit or TKET.

IBM RUNTIME JOB ID: da1ul7e3kjvs73877tag
RAW EVIDENCE SHA256: 9d347393dd9f2b3c168ab706d2c4d55e0ff4ab1c65ea5091fb8b62527fc7e3e5