Short answer: Leakage is when a qubit leaves its two computational states and ends up in a higher energy level. In error correction it matters because a leaked qubit can disturb its neighbours and create correlated errors, which error correction assumes will be rare. Experiments and simulations on superconducting qubits report that active leakage removal can limit this. These are research results on small codes.
Why leakage is a problem for error correction
A key assumption of quantum error correction (QEC) is that physical errors are sufficiently uncorrelated in space and time. A 2023 Nature Physics paper explains that a qubit in a leakage state can induce errors on several neighbouring qubits and can even cause them to leak as well. During an error-corrected algorithm, leakage builds over time and spreads through multi-qubit interactions. The result is correlated errors that degrade the exponential suppression of logical error with scale. A 2021 PRX Quantum paper adds that QEC codes are designed for errors inside the computational subspace and give no inherent protection against leakage.
A reset protocol tested on a bit-flip code
An arXiv preprint from 2021 (2102.06131) reports a reset protocol that returns a transmon qubit to the ground state from all relevant higher levels. The authors test it with the bit-flip stabilizer code, a simplified version of the surface code, and study how leakage accumulates during error correction. With the protocol they report lower logical error rates and improved scaling and stability of error suppression as the number of qubits grows.
Leakage removal in every cycle
The 2023 Nature Physics paper reports a distance-3 surface code and a distance-21 bit-flip code on a quantum processor in which leakage is removed from all qubits in each cycle. This shortens the lifetime of leakage and limits its spread. The authors report a tenfold reduction in the steady-state leakage population of the data qubits, and an average leakage population below 1 x 10^-3 across the device. They state that adding the removal process to a code circuit would prevent leakage from inducing correlated errors across cycles, and they describe the result as resolving a key challenge for QEC at scale. That last statement is the authors' own conclusion, and larger codes have not been shown in these papers.
Leakage-reduction units in simulation
The PRX Quantum paper proposes two leakage-reduction units (LRUs) for a transmon-based surface code that need no extra hardware or QEC-cycle time compared with earlier proposals. For data qubits, a microwave drive moves leakage to the readout resonator, where it decays quickly. For ancilla qubits, a pulse between the 1 and 2 levels is applied depending on the measurement outcome. Density-matrix simulations of a distance-3 surface code show the average leakage lifetime falling to almost one QEC cycle, even with limited LRU fidelity, with a significant reduction in the logical error rate. This is simulation, and the authors describe it as opening the prospect of near-term QEC demonstrations.
What the results do and do not show
The experiments cover small codes on superconducting hardware, and the simulation covers a distance-3 code. The numbers come from each paper's own setup and are not directly comparable. Results for larger codes and other qubit types are outside what these papers report.
FAQ
What is leakage in a qubit?
Population moving out of the two computational states into other energy levels.
Why is leakage worse than an ordinary error?
It can spread to neighbouring qubits and make errors correlated, which QEC assumes are limited.
How much did leakage removal reduce leakage?
The 2023 paper reports a tenfold reduction in steady-state leakage population of the data qubits.
Do the results come from experiments or simulations?
Both. The reset protocol and the 2023 removal were experiments; the LRU scheme was simulated.
Does this mean error correction at scale is solved?
No. The papers cover small codes, and the claim of resolving the challenge is the authors' own.
All field notes