When should I use PEA?#
Advantages#
Probabilistic error amplification (PEA) can be useful when:
Provides higher accuracy than ZNE because it leverages a specified noise model rather than being noise-agnostic.
Requires lower sampling overhead than PEC.
Enables execution of deeper circuits than with ZNE, in cases where unitary folding would create circuits longer than qubit coherence times.
Reuses information learned from ZNE experiments to improve PEA performance.
Disadvantages#
PEA also has limitations:
Requires a reasonably accurate noise model and baseline noise estimate (e.g. by sparse Pauli–Lindblad tomography).
The sampling overhead can become large as the scale factor increases, since the one-norm of the representation grows and more samples are required.
The final extrapolation step can be sensitive to statistical noise and to the choice of scale factors.
In Mitiq, PEA currently supports local and global depolarizing noise models and assumes circuits can be decomposed into one- and two-qubit operations.
Example#
For a demonstration of PEA on superconducting hardware, see the study in [59], and for more information generally about tradeoffs find PEA on The QEM Zoo.