Quantum error correction is a method to protect quantum information from decoherence and errors, enhancing the accuracy of quantum computers.Quantum error correction is a set of methods that protect logical qubits (units of quantum information) from noise and decoherence (loss of quantum properties) by distributing information across redundant physical qubits. Unlike classical copying, nonlocal correlations are used here. Errors are detected by measuring stabilizers—special operators for which the code state is an eigenstate; such measurement does not destroy the encoded superposition. The simplest Shor code corrects any single error—bit flip or phase flip. The five-qubit code achieves the theoretical minimum for correcting a general error.
The concept arose from the realization of the fragility of quantum states and the impossibility of copying them (no-cloning theorem, 1982). In 1995, Peter Shor, using an analogy with classical codes and entanglement, proposed a nine-qubit code. Soon Andrew Steane developed a seven-qubit code based on linear algebra. Generalization led to the stabilizer formalism (Daniel Gottesman, 1996) and a family of codes, including topological ones, underlying fault-tolerant quantum computers.
QEC uses code distance d: the minimum number of errors that turn one codeword into another. A code with distance d corrects any combination of errors of weight t ≤ (d−1)/2. Stabilizer formalism: the set S — an abelian subgroup of Pauli operators on n qubits without −I. The code space is the common eigenspace with eigenvalue +1. An error E either commutes or anticommutes with the generators; anticommutation changes the sign and is detected. Measuring the generators yields a syndrome, by which the error is recovered up to a stabilizer. Error correction condition: for any two correctable errors E_a, E_b, the projector P satisfies P E_a† E_b P = c_{ab} P. CSS codes are often used, combining classical codes for bit and phase flips.
The foundation of reliable quantum computers; without it, algorithms like Shor's factoring would drown in noise.
The smallest code that corrects any single error uses only five qubits – and this is the theoretical limit; no four-qubit code can do it.
| J_{k\alpha} | variable | parameter being updated at iteration — parameter |
| \tau_\mu | variable | iteration step index — iteration index |
| \eta | variable | learning rate, step size — learning rate |
| \epsilon | variable | error or loss function — loss function |
| \frac{\partial}{\par | operator | computes the gradient of the error with respect to the parameter — partial derivative |
