The subfield metric and its application to quantum error correction (Q6580193)

This paper introduced a distance notion for algebraic quantum error correcting codes (QECC) which is tailored to capture the behaviour of biased errors. The authors use this metric to derive classical bounds that tradeoff code parameters, such as the Singelton- and Gilbert-Varshamov bounds.\N\N\textbf{Quantum error correction.}\NA Quantum error correcting code embeds a \(k\)-dimensional (complex projective) Hilbert space \(\mathcal{K}\) into a \(n\)-dimensional ``physical'' Hilbert space \(\mathcal{H}\). We identify the physical Hilbert space with \({(\mathbb{C})^q}^{\otimes n}\), where the prime \(q\) describes the local dimension of the system.\NEach local system \(\mathbb{C}^q\) admits an operator basis generated by\N\[\NX,\, Z: \quad XZ=\omega ZX,\quad X^q=Z^q=1,\N\]\Nwhere \(\omega:\, \omega^q=1\) is a root of unity.\NThe operator basis extends over the tensor powers\N\[\N\mathcal{P}=\{X^{\mathbf{a}}Z^{\mathbf{b}},\; (\mathbf{a}, \mathbf{b}) \in \mathbb{F}_q^{2n}\} / \langle\omega\rangle=\langle X, Z\rangle^{\otimes n}/\langle\omega\rangle\N\]\Nand forms a natural basis for the error model.\N\N\textbf{Errors and distance.} A physical assumption is that errors act independently and locally on each tensor factor of the Hilbert space. Denoting by \(\mathrm{wt}_H(P)\) the Hamming weight of the operator, that is the number of tensor factors \(P\) acts on non-trivially, the probability for an error \(E\in \mathcal{P}\) is\N\[\NP(E)=(1-p)^n\left(\frac{p}{1-p}\right)^{\mathrm{wt}_H(P)}.\N\]\NThis error model motivates the definition of the distance of a code as the minimal Hamming weight of an error \(E\in \mathcal{P}\) that yield a non-trivial automorphism of \(\mathcal{K}\).\N\N\textbf{Algebraic quantum error correction.}\NAn algebraic formulation of the above situation maps the defining vectors \(\mathbb{F}_q^{2n}\) to \(\mathbb{F}_{q^2}^n\) by identifying vectors\N\[\N\Phi: \mathbb{F}_q^{n}\times \mathbb{F}_q^{n} \to \mathbb{F}_{q^2}^n\N: \;(\mathbf{a}, \mathbf{b})\mapsto \mathbf{a}\gamma + \mathbf{b}\N\]\Nfor some \(\gamma \in \mathbb{F}_{q^2}\setminus \mathbb{F}_q\).\NUnder this mapping, we associate with a (linear) code \(\mathcal{C}\subset \mathbb{F}_{q^2}^n\) a subset of vectors that represent Paulis\Nwe have\N\[\N\mathrm{wt}_H(X^{\mathbf{a}}Z^{\mathbf{b}})=\|\mathbf{a}\gamma + \mathbf{b}\|_0\N\]\Nand the distance \(d\) is given by the smallest value \(\mathrm{wt}_H(E), E\in \mathcal{P}/\langle \omega^{1/2}\rangle\).\N\N\textbf{Sphere packing bounds.}The definition of the distance allows the use of sphere-packing bounds to infer code properties. The intuition behind this link is the following: if we were to choose codes \(\mathcal{C}\subset \mathbb{F}_{q^2}^n\) at random, the average code \(\mathcal{C}\) is expected to evenly distribute over \(\mathbb{F}_{q^2}^n\), such that for each \(\mathcal{C}\) we obtain a partition of \(\mathbb{F}_{q^2}^n\) into\N\[\N\mathbb{F}_{q^2}^n = \bigcup_{c\in \mathcal{C}} B(c, d-1)\N\]\Neach centered at points \(c\in \mathcal{C}\), where the number of Balls \(|\mathcal{C}|\) can be bounded by their size\N\[\N|\mathbb{F}_{q^2}^n| \leq |\mathcal{C}| | B(0, d-1)|.\N\]\NThe distance metric defines the balls \(B\) and implies a tradeoff between the number of logical operators \(|\mathcal{C}|\), the size of the physical space \(n\) and the distance \(d\). This idea is the origin of the Singelton- and Gilbert-Varshamov bounds in coding theory.\N\N\textbf{Biased noise.} Physical experiments find that the error model we started with is not always natural, it is common that errors are actually biased and errors of, e.g., \(Z\)-type appear more frequently than those of \(X\)-type. To develop a coding theory and derive similar bounds as above, this article defines a new metric distance notion that the authors call the \textit{subfield metric} on \(\mathbb{F}_{q^m}\), which is defined using a weighting parameter \(\lambda \geq 1/2\)\N\[\N\mathrm{wt}_{\lambda}(x):=\N\begin{cases}\N0\quad \text{if}\quad x=0, \\\N1\quad \text{if}\quad x\in \mathbb{F}_q\setminus\{0\}, \\\N\lambda\quad \text{if}\quad x \in \mathbb{F}_{q^2}\setminus \mathbb{F}_q.\N\end{cases}\N\]\NUsing this new metric, the authors define the adapted distance notion \(d_{\lambda}\) and revisit the classical coding theoretic notions and rederive in this setting sphere packing bounds, like the Gilbert-Varshamov bound\N\NTheorem 1. Let us denote by \(g(d)=\log_{q^m}(|B_{d,\lambda}(\mathbb F_{q^m}^n)|)\).\NLet \(\varepsilon >0\) and \(\mathcal{C}\subseteq\mathbb F_{q^m}^n\) be a randomly chosen linear code of dimension \(k= \lceil(1- g(d)/n - \varepsilon)n \rceil\). Then the probability that \(\mathcal{C}\) achieves the sphere covering bound, i.e., that its minimum \(\lambda\)-subfield distance is \(d\), is at least \(1-q^{m(1-\varepsilon n)}\).\N\Nand the Singelton bound.\N\NTheorem 2. Let \(\lambda \geq 1\) and \(\mathcal{C} \subseteq \mathbb{F}_{q^m}^n\) be a code (not necessarily linear) with minimum \(\lambda\)-subfield distance \(d_\lambda\). Then\N\[\N\left\lfloor \frac{d_\lambda -1}{\lambda}\right\rfloor \leq n-\lceil \log_{q^m}(\mid \mathcal{C}\mid)\rceil.\N\]\NThus for linear codes of dimension \(k\), we get\N\[\N\left\lfloor \frac{d_\lambda -1}{\lambda}\right\rfloor \leq n-k.\N\]\NOther notions like Plotkin-type bounds, MacWilliams identities and existence of optimal families relative to the proven bounds are also derived.\N\NWith these developments, the authors have initiated a rigorous study of coding theory tailored towards biased noise. This work represents an onset of noise-tailored coding theory, which is a timely and practically useful endeavor.