Basic entanglement distillation with realistic noise

Entanglement distillation is a key component of modular quantum computing and long-range quantum communications. However, this powerful tool to reduce noise in entangled states is difficult to realize in practice for two main reasons. First, operations used to carry out distillation inject noise they seek to remove. Second, the extent to which distillation can work under realistic device noise is less well studied. In this work, we both simulate distillation using a variety of device noise models and perform distillation experiments on fixed-frequency IBM devices. We find reasonable agreement between experimental data and simulation done using Pauli and non-Pauli noise models. In our data, we find broad improvement when the metric of success for distillation is to improve average Bell fidelity under effective global depolarizing noise, remove coherent errors, or improve the Bell fidelity of mildly degraded Bell pairs. We pave the way to obtain broad improvement from distillation under a stricter, but practically relevant, metric: distill (physically) nonlocal Bell pairs with higher fidelity than possible to obtain with other available methods. Our results also help understand metrics and requirements for quantum devices to use entanglement distillation as a primitive for modular computing.