Directed Execution & Selective Pauli Twirling on Beta Hardware
We evaluate whether risk-mass layer selection can reduce randomized compiling overhead on IBM Directed Execution beta mode without fidelity degradation (ibm_marrakesh, Job da1u5o63kjvs73877cm0, 24,576 shots). Selective twirling twirled only 41.1% of CZ layers (a 58.9% reduction in twirling overhead) and achieved 1 win against unmitigated raw circuits. However, non-inferiority compared to full Pauli twirling remained statistically inconclusive across 4 tested circuits.
1. Protocol & Motivation
Full Pauli twirling inserts random single-qubit gates before and after every two-qubit gate layer, requiring multiple randomized circuit variants. Our selective twirling algorithm prioritizes twirling on high-risk gate layers while executing low-risk layers deterministically.
2. Empirical Status
While overhead reduction was mathematically verified (58.9% fewer twirled layers), physical fidelity non-inferiority was inconclusive due to high variance across randomized compilation runs on the beta execution endpoint.