Techniques for identifying errors in quantum systems without full correction.Quantum error detection refers to methods that identify when errors have occurred in a quantum system, without necessarily correcting them on the spot. Unlike full quantum error correction, which requires many physical qubits to encode a single logical qubit and actively fix errors, error detection is lighter: it uses parity checks or other measurements to flag that an error has happened, so that the computation can be restarted or the affected part discarded. This is especially useful in near-term quantum devices where resources are limited and full correction is not yet feasible.
The concept emerged alongside quantum error correction in the mid-1990s, with early work by Peter Shor and Andrew Steane. The distinction between detection and correction was formalized in the late 1990s, and recent experiments on superconducting and trapped-ion platforms have demonstrated real-time error detection.
The approach typically involves entangling the data qubits with auxiliary 'ancilla' qubits. By measuring the ancillas, one can infer whether certain types of errors (like bit flips or phase flips) have occurred on the data qubits, without disturbing the quantum information itself. If an error is detected, the system can be reinitialized or the specific operation repeated. The key is that the measurement collapses only the ancilla, not the data, so the quantum state survives unless an error was actually present. This is often implemented using stabilizer codes, where the ancilla measurements correspond to checking the eigenvalues of stabilizer operators.
Error detection is used in current quantum processors to validate the output of computations, especially in algorithms like variational quantum eigensolvers, where a detected error can trigger a re-run of the circuit.
In 2023, a team at Google Quantum AI demonstrated a quantum error detection protocol that caught errors in just 384 nanoseconds—faster than the time it takes for a typical error to propagate, allowing them to discard faulty runs before they corrupt the result.





