Pulse-Level Fractional Gate Lowering on IBM Heron
We evaluate whether pulse-level fractional gate decomposition (synthesizing continuous R_ZZ(θ) pulses directly instead of decomposing into discrete CZ pairs) improves physical execution fidelity on IBM Heron (ibm_marrakesh, Job da1tg6eg52gs73cm3gg0, 24,576 shots). While pulse schedules achieved structural gate duration reductions of up to 40%, physical execution yielded 2/6 conclusive wins (N4_B, N8_B) and 4/6 inconclusive cases within noise margins. The technique is classified as a mixed physical signal.
1. Motivation
Standard circuit compilation synthesizes arbitrary two-qubit interactions using discrete basis gates (such as CZ or CNOT). On physical superconducting qubits, direct calibrated fractional pulses can in theory reduce total microwave drive time and reduce decoherence.
2. Empirical Results on IBM Heron r2
| Case | Qubits | Fractional Fidelity | Discrete CZ Fidelity | Δ(F_frac - F_cz) | Outcome |
|---|---|---|---|---|---|
| N4_A | 4 | 0.9120 | 0.9085 | +0.0035 | Inconclusive |
| N4_B | 4 | 0.8950 | 0.8620 | +0.0330 | Physical Win |
| N6_A | 6 | 0.7840 | 0.7790 | +0.0050 | Inconclusive |
| N6_B | 6 | 0.7610 | 0.7580 | +0.0030 | Inconclusive |
| N8_A | 8 | 0.6850 | 0.6810 | +0.0040 | Inconclusive |
| N8_B | 8 | 0.6520 | 0.6190 | +0.0330 | Physical Win |
3. Conclusion & Boundary
While fractional pulse lowering showed definite physical improvements on selected parameterized instances, 4 out of 6 cases remained within statistical noise bounds. Universal fidelity superiority is not claimed.