Leakage Errors
Leakage is population escaping the two-level computational space into other levels, an error type standard correction codes do not handle well.
Outside the qubit
A qubit is defined on two chosen levels, but the physical system has more. In a transmon the 2 state sits only a few hundred MHz above the 1 state; in atoms and spins there are auxiliary levels used for control. Leakage is when an operation accidentally moves population into one of these non-computational states. Once there, the information is neither a valid 0 nor 1, and normal gates and error correction may not act on it correctly.
Where it comes from
- Fast or poorly shaped pulses that spectrally overlap the 1-2 transition
- CZ gates that intentionally use the 2 state and fail to fully return population
- Rydberg-state decay in neutral atoms
- Heating and off-resonant excitation in ions
Why it is dangerous
Error-correcting codes assume errors keep qubits inside the computational subspace, so a leaked qubit can silently corrupt a code cycle and even spread errors to neighbors during subsequent gates. Leakage also does not show up cleanly in some standard benchmarks, so it can be underestimated if not measured specifically.
Mitigation
- DRAG pulse shaping to suppress leakage during single-qubit gates
- Careful flux-pulse tuning for CZ gates to empty the 2 state
- Leakage-reduction units: circuits that return leaked population to the qubit space
- Choosing frequencies and anharmonicities that keep unwanted transitions far off resonance
Because leakage is a distinct, correlated, and code-breaking error, controlling it is a specific engineering goal alongside lowering ordinary gate error, and dedicated leakage-reduction is now a standard part of large error-correction demonstrations.