137
Part Three · D · Error correction and fault tolerance

Quantum codes: Shor, Steane, CSS

In the early 1990s many physicists believed quantum error correction was impossible, for three good reasons. You cannot copy a quantum state to make redundant copies (no-cloning). Errors are continuous, an infinity of tiny rotations, not discrete flips. And checking for errors means measuring, which destroys the superposition you were trying to protect. In 1995 Peter Shor found a code that defeats all three objections at once. The trick is to measure only the error, never the data.

On the plate: Shor’s nine-qubit code. Damage it any way you like and watch it heal.

The nine-qubit code

Shor nested two repetition codes. Against bit flips, copy the basis state (not the amplitudes) across three qubits: ∣0⟩→∣000⟩|0\rangle\to|000\rangle, ∣1⟩→∣111⟩|1\rangle\to|111\rangle; this is entanglement, not cloning. Against phase flips, do the same in the ∣±⟩|\pm\rangle basis. Together:

∣0L⟩=122(∣000⟩+∣111⟩)⊗3∣1L⟩=122(∣000⟩−∣111⟩)⊗3\begin{gathered}|0_L\rangle = \tfrac{1}{2\sqrt2}(|000\rangle + |111\rangle)^{\otimes3}\\ |1_L\rangle = \tfrac{1}{2\sqrt2}(|000\rangle - |111\rangle)^{\otimes3}\end{gathered}

Eight stabilizers check the code: six of the form ZiZjZ_iZ_j within each block, comparing neighbours’ bit values, and two of the form X⊗6X^{\otimes6} across pairs of blocks, comparing block signs. Every codeword is a +1 eigenstate of all eight, so measuring them reveals nothing about α\alpha and β\beta. Established

Break it, heal it

Encode a logical state with angle θ=\theta = 1.20. Choose an error: , on qubit 5, then .

The plate measures all eight stabilizers on the full 512-amplitude state, applies the correction the syndrome names, and reports the fidelity with the original. It is 1 every time. For the rotation, measuring the syndrome projects the continuous error onto either “no error” or a full Pauli flip, each of which is then fixed: continuous errors are discretized by measurement. A Y error is both an X and a Z, and both repetition layers fix their part. Established

The general conditions, and CSS codes

Emanuel Knill and Raymond Laflamme gave in 1997 the condition for a code with projector PP to correct a set of errors {Ea}\{E_a\}:

PEa†EbP=cabPPE_a^\dagger E_bP = c_{ab}P

Errors must not distinguish codewords, and must map the code to spaces that are either orthogonal or identical in effect. Correcting X, Y and Z on a qubit suffices to correct any error on it, because every operator is a combination of those four. Robert Calderbank and Shor, and Andrew Steane, independently in 1996, showed how to build quantum codes from two classical codes C2⊂C1C_2\subset C_1: use one for bit flips and the dual of the other for phase flips. From the [7,4,3] Hamming code (Chapter 136), which contains its own dual, comes the Steane code [[7,1,3]]: seven qubits, one logical qubit, distance 3, with the Hamming checks used once as X-type stabilizers and once as Z-type. Established

What you now have

Why the three objections fail, the Shor code and its stabilizers, syndrome measurement and discretization of errors, the Knill–Laflamme conditions, CSS codes and the Steane code.

· · ·

Three questions to carry with you.

How does the Shor code avoid cloning?tap
It entangles basis states, |0⟩ → |000⟩, rather than copying an unknown state.
Why doesn’t syndrome measurement destroy the data?tap
Every codeword has the same stabilizer eigenvalues, so the outcome carries no information about α, β.
How are continuous errors handled?tap
Measuring the syndrome projects them onto discrete Pauli errors, which are corrected.

And one to work out. The Shor code has 8 stabilizer generators, each giving one ±1 outcome. How many distinct syndromes are possible?

syndromes =

Sources: P. W. Shor, Phys. Rev. A 52, R2493 (1995). A. M. Steane, Phys. Rev. Lett. 77, 793 (1996). A. R. Calderbank and P. W. Shor, Phys. Rev. A 54, 1098 (1996). E. Knill and R. Laflamme, Phys. Rev. A 55, 900 (1997). Checked by the plate: exact 512-amplitude simulation; fidelity 1 after correcting X, Y, Z and rotation errors on every qubit.

Next · Chapter 138
Stabilizer codes in depth

No code can apply every gate transversally (Eastin–Knill). That one theorem shapes the entire architecture of a quantum computer.

Turn the page