Discretization of Errors

Physical noise is continuous: a stray field rotates a qubit by an infinitesimal angle, an amplitude leaks by a small amount $\epsilon$, the phase drifts. It seems hopeless to correct a continuum of possible errors with a finite procedure. The remarkable resolution — the digitisation of quantum errors — is that if a code corrects a discrete basis of errors, it automatically corrects every continuous error built from that basis. This is what makes quantum error correction possible at all.

The single-qubit Pauli basis

Any $2\times 2$ matrix is a complex linear combination of the four Pauli operators

which form a basis for operators on one qubit. Hence any single-qubit operation $E$ — unitary or not — can be written

A tiny over-rotation $R_x(\theta) = \cos(\theta/2)\,I - i\sin(\theta/2)\,X$ is literally a superposition of "no error" ($I$) and a bit-flip ($X$). Correcting $I$ and $X$ therefore covers the whole one-parameter family.

$n$ qubits: the Pauli group

On $n$ qubits the analogous basis is the set of Pauli strings

of which there are $4^n$. They span all $2^n \times 2^n$ operators, so any error operator $E_a = \sum_P c_{a,P}\, P$ expands in this basis. We classify a Pauli string by its weight: the number of tensor factors that are not $I$. Weight-$1$ errors are single-qubit faults, and an error-correcting code is judged by the largest weight it can handle.

The discretization theorem

Suppose a code corrects the set of Pauli errors $\{P_a\}$ — that is, the $P_a$ satisfy the Knill–Laflamme conditions. Now let a general error act, with Kraus operators each expanded as $E_k = \sum_a c_{k,a} P_a$ in this corrected basis. The error channel sends a codeword $|\psi\rangle$ to a mixture of terms $P_a |\psi\rangle$. The syndrome measurement projects onto one of the orthogonal error subspaces $V_a = P_a \mathcal{C}$ — and this projection collapses the superposition of errors onto a single $P_a$. Once collapsed, the known recovery for $P_a$ restores $|\psi\rangle$ exactly.

Why $\{X, Z\}$ (plus $I$) suffices per qubit

Because $Y = iXZ$, correcting bit-flips ($X$) and phase-flips ($Z$) on each qubit automatically handles $Y$ (a combined bit-and-phase flip). So the entire program of single-qubit error correction reduces to two channels of work: detect and fix $X$ errors, and — in a rotated basis — detect and fix $Z$ errors. This is precisely the logic behind the Shor and CSS code constructions later in the module.

A caution

Discretization makes errors finite but does not make them free. Each Pauli error still costs the code "room": the syndrome subspaces $V_a$ must be mutually orthogonal, which limits how many errors a code of fixed size can correct. That accounting — distance, the Hamming bound, and the singleton bound — is the quantitative heart of the lessons that follow.

The takeaway

The $4^n$ Pauli strings form an operator basis, so every physical error is a linear combination of them. A code that corrects a chosen set of Pauli errors corrects every error spanned by that set, because syndrome measurement collapses the continuous error onto a single, already-correctable Pauli. Continuous quantum noise is, for the purposes of correction, effectively digital.

Sign in on the full site to ask questions and join the discussion.