CSS Codes

The general stabilizer formalism is flexible but can be unwieldy. CSS codes, named for Calderbank, Shor, and Steane, are a structured subclass that builds a quantum code out of two classical linear codes. They separate $X$ -errors and $Z$ -errors cleanly, inherit decades of classical coding theory, and include many of the most important quantum codes.

Recap: classical linear codes

A classical $[n, k]$ binary linear code $C$ is a $k$ -dimensional subspace of $\mathbb{F}_2^n$ , specified by an $(n-k) \times n$ parity-check matrix $H$ : a vector $v$ is a codeword iff $Hv = 0$ . The dual code $C^\perp$ is the set of vectors orthogonal to every codeword of $C$ . Its minimum distance $d$ — the smallest weight of a nonzero codeword — sets how many bit errors it corrects.

The CSS construction

Pick two classical codes $C_X$ and $C_Z$ with parity-check matrices $H_X$ and $H_Z$ , satisfying the CSS orthogonality condition

H_X\,H_Z^{\mathsf T} = 0 \pmod 2.

Then define a stabilizer code whose generators come in two flavors:

For each row $r$ of $H_X$ , an $X$ -type stabilizer $\prod_{j: r_j = 1} X_j$ .
For each row $r$ of $H_Z$ , a $Z$ -type stabilizer $\prod_{j: r_j = 1} Z_j$ .

Because every generator is purely $X$ -type or purely $Z$ -type, two same-type generators automatically commute. An $X$ -type and a $Z$ -type generator commute iff the supporting rows of $H_X$ and $H_Z$ overlap in an even number of positions — which is exactly the condition $H_X H_Z^{\mathsf T} = 0$ . So the orthogonality condition is precisely what makes the stabilizer group abelian.

Code parameters

If $H_X$ has rank $r_X$ and $H_Z$ has rank $r_Z$ , the code uses $r_X + r_Z$ independent generators on $n$ qubits, encoding

k = n - r_X - r_Z

logical qubits. The distance is governed by the classical codes: $Z$ -errors are detected by the $X$ -stabilizers and corrected up to the distance properties of $C_X$ , and vice versa for $X$ -errors.

Why CSS codes matter

CSS codes are the foundation of fault-tolerant quantum computing for several reasons:

Decoding splits in two. $X$ - and $Z$ -syndromes are decoded independently with classical decoders, halving the problem.
Transversal gates. Many CSS codes admit transversal logical Clifford gates (e.g. transversal CNOT between two code blocks), a key fault-tolerance property.
They scale. The surface code and other topological codes are CSS codes; their $H_X$ and $H_Z$ are sparse, enabling efficient decoders and a high error threshold.

Try it

Verify the CSS orthogonality condition for the Steane code, which sets $H_X = H_Z = H$ where $H$ is the parity-check matrix of the classical $[7,4]$ Hamming code. Compute $H H^{\mathsf T} \bmod 2$ and confirm it is the zero matrix — proving the stabilizer group is abelian.

Run your code to see the quantum state.