Why Qubits Need Error Correction
Quantum information is fragile: qubits decohere, gates are imperfect, and measurement collapses state, so raw qubits cannot run long computations.
The fragility problem
A qubit stores a continuous superposition a|0> + b|1> with |a|^2 + |b|^2 = 1. Unlike a classical bit sitting in a well-separated voltage level, this state is exposed to its environment. Stray electromagnetic fields, thermal photons, control-line noise, and imperfect gates all perturb the amplitudes a and b, and any leakage of information about the qubit into the environment destroys the coherence the computation depends on.
Two timescales quantify the decay. T1 is the energy-relaxation time, the characteristic time for |1> to decay toward |0>. T2 is the dephasing time, the time over which the relative phase between |0> and |1> randomizes. In current hardware these range from microseconds to milliseconds, while useful algorithms may require billions of gate operations.
Why classical tricks do not transfer
Classical computers achieve near-perfect reliability partly by using many electrons per bit and by copying and voting. A quantum state cannot be copied (see No-Cloning and QEC), and measuring a qubit to check it collapses the superposition. Naive redundancy is therefore forbidden.
- Errors are continuous, not just discrete flips: a and b can drift by tiny amounts.
- Measurement is destructive, so you cannot simply read a qubit to see if it is wrong.
- No-cloning blocks straightforward triplication of the state.
- Gates and idling both inject error, so the budget accumulates with circuit depth.
Quantum error correction (QEC) resolves all three obstacles at once. It spreads one logical qubit across many physical qubits, measures only carefully chosen joint parities that reveal errors without revealing the encoded data, and shows that a continuum of small errors can be digitized into a discrete set that a code can correct. This makes arbitrarily long quantum computation possible in principle, provided physical error rates fall below a threshold.
The rest of this category builds that machinery from repetition codes up through the surface code, fault tolerance, and the overhead required to run real algorithms.