Evaluating the resilience of variational quantum algorithms to leakage noise

As we are entering the era of constructing practical quantum computers, suppressing inevitable noise to accomplish reliable computational tasks will be primary goal. Leakage noise, as amplitude population leaking outside qubit subspace, is a particularly damaging source error that correction approaches cannot handle. However, impact this on performance variational algorithms (VQAs), type near-term naturally resistant variety noises, yet unknown. Here, {we consider typical scenario with widely used hardware-efficient ansatz and emergence leakage in two-qubit gates}, observing generally reduces expressive power VQAs. Furthermore, benchmark influence VQAs real-world learning tasks. Results show that, both for data fitting classification, has negative training process final outcomes. Our findings give strong evidence vulnerable most cases, implying must effectively suppressed order achieve computing applications, whether long-term error-correcting computing.