--- title: Directional Codes in Quantum Systems url: https://www.emergentmind.com/topics/directional-codes type: topic --- # Directional Codes in Quantum Systems Directional codes are a class of constructions in which directional, orientational, or anisotropic structure is made explicit in the code definition rather than treated as incidental. In current arXiv literature, the term is used in several technically distinct ways. The most specific usage denotes a family of hardware-motivated quantum low-density parity-check (qLDPC) codes defined by stabilizer-measurement routes on square- and hex-grid hardware [2507.19430]. Closely related work studies the algebraic structure of such route-generated CSS codes [2602.19057], directionally informed decoding for quantum codes [2601.07240], and topological-code modifications with explicit anyonic confinement along one spatial direction [2506.22025]. In another line of work, “directional codes” denotes product-structured finite alphabets for preserving vector direction under low-precision quantization [2605.07662]. A related geometric theme also appears in spatio-directional neural encodings on $S^2$ [2603.05079]. ## 1. Directional Codes as qLDPC codes on local-connectivity hardware The qLDPC family called “Directional Codes” is defined on the infinite square lattice $\mathbb Z^2$, with data qubits on $$ \mathbb Z^2_{\rm data}=\{(x,y)\colon x+y\equiv0\pmod2\} $$ and ancilla qubits on the complementary checkerboard sublattice. A code is specified by a direction sequence $(D_1,D_2,\dots,D_w)$ with $D_j\in\{N,E,S,W\}$, a layout $\alpha:\mathbb Z^2_{\rm anc}\to\{X,Z\}$ assigning each ancilla to an $X$-type or $Z$-type plaquette, and two lattice vectors $\vec v_1,\vec v_2$ used to wrap the plane onto a torus [2507.19430]. Syndrome extraction is implemented by $w$ simultaneous layers of controlled-Pauli-then-swap gates. For an $X$ plaquette ancilla $A$, the elementary operation is $$ \mathrm{CXSWAP}(A,Q)=\bigl(\mathrm{CX}(A,Q)\bigr)\,\mathrm{SWAP}(A,Q), $$ and for a $Z$ plaquette ancilla the CX is replaced by CZ. Up to single-qubit Cliffords, both CXSWAP and CZSWAP are equivalent to the native iSWAP gate. In each round, ancillas are reset in $\lvert+\rangle$, the $w$ directional layers are applied, and ancillas are measured in the $X$ basis. Each ancilla thereby “slides” across the lattice and measures a weight-$w$ plaquette operator supported on the data qubits it visits. Under mild nondegeneracy conditions on the torus identification, this yields a finite CSS stabilizer code on $\mathbb Z^2/\langle \vec v_1,\vec v_2\rangle$ [2507.19430]. Several small-weight families were isolated explicitly. The weight-4 NEEN sequence yields a linear transform of the distance-$d$ toric code. The weight-5 $NE^3N$ family uses only hexagonal-grid connectivity, degree-$3$ per qubit. The weight-6 $N^2E^2N^2$ family admits a rectangular torus on the square grid, and the weight-7 $N^2E^3N^2$ family admits a layout on square or hex grid [2507.19430]. The principal finite families studied numerically encode $4$, $6$, or $12$ logical qubits. | Family | Parameters | Connectivity or performance claim | |---|---|---| | $NE^3N$ | $n=\tfrac32 d^2+3d$, $k=4$, distance $d$ | Hexagonal-grid connectivity; comparable to four RPC copies at $37.5\%-56.25\%$ of the total qubit count at $p=10^{-3}$ | | $N^2E^2N^2$ | $n=4d^2$, $k=6$, distance $d$ | Beats six RPC copies using $41.15\%-66.67\%$ of qubits | | $N^2E^3N^2$ | $n=6d^2-12d$ or $6d^2$, $k=12$, distance $d$ | Beats twelve RPC copies using $25\%-50\%$ of qubits at equal distance $d\le10$ | The numerical evaluation used the superconducting-inspired SI-1000 circuit-level Pauli noise model over five full QEC rounds, with BP-OSD decoding. At $p=10^{-3}$, the reported logical error probabilities were favorable relative to repeated rotated planar code instances. A representative $d=6$ comparison gave RPC$_{d=6}\times4$ with $480$ total qubits and $p_L=0.012$, versus $NE^3N$ $\llbracket72,4,6\rrbracket$ with $288$ total qubits and $p_L=0.0088$, $N^2E^2N^2$ $\llbracket144,6,6\rrbracket$ with $216$ total qubits and $p_L=0.0065$, and $N^2E^3N^2$ $\llbracket144,12,6\rrbracket$ with $264$ total qubits and $p_L=0.0071$ [2507.19430]. ## 2. Word-first structural theory of route-generated codes A later structural treatment places these qLDPC constructions into a “word-first” framework for route-generated, translation-invariant CSS codes on rectangular tori [2602.19057]. The starting object is a direction word $$ W=d_1d_2\cdots d_w,\qquad d_j\in D, $$ over the alphabet $$ D=\{N=(0,1),\,E=(1,0),\,S=(0,-1),\,W=(-1,0)\}\subset\mathbb Z^2. $$ With partial sums $$ S_0=(0,0),\qquad S_j=\sum_{t=1}^j d_t, $$ the route-to-support lemma defines midpoint offsets $$ Q_j=S_{j-1}+S_j=2\sum_{t=1}^{j-1}d_t+d_j. $$ On a checkerboard torus $G=\mathbb Z_{L_x}\times\mathbb Z_{L_y}$, an ancilla at $a\in\Lambda_A$ interacts with data qubits at $a+Q_j\pmod{(L_x,L_y)}$, and the support pattern is $$ P(W)=\{Q_1,\dots,Q_w\}\subset\mathbb Z^2. $$ This supplies a closed-form route-to-support map [2602.19057]. The central commutation invariant is the odd-multiplicity difference lattice. From the multiset of pairwise differences $$ \Delta(W)=\{Q_j-Q_i:1\le i\max_{v\in D(W)\cup[D(W)\pm D(W)]}|v_x|, \qquad L_y>\max_{v\in D(W)\cup[D(W)\pm D(W)]}|v_y|. $$ Under these inequalities, no nonzero relevant offset vanishes modulo $(L_x,L_y)$, so the infinite-lattice schedule remains collision-free on the torus [2602.19057]. The theory also quotients the search over direction words by dihedral lattice symmetries $D_4$, reversal plus inversion, and cyclic shifts. The equivalence class has size at most $16w$, and canonicalization chooses the lexicographically smallest representative in expanded-letter form. An inverse problem is solved as well: a translation-invariant support pattern $P$ is realizable by a single route iff, for some ordering $Q_1,\dots,Q_w$, one has $Q_1\in D$, each difference $\Delta_j=Q_{j+1}-Q_j$ lies in $$ \Sigma=\{u+v:\,u,v\in D\}=\{\pm(2,0),\pm(0,2),\pm(1,1)\}, $$ and the recursion $$ d_1=Q_1,\qquad d_{j+1}=\Delta_j-d_j $$ stays in $D$ at every step. Failure certifies non-realizability [2602.19057]. As a case study, the word $W=NE^2NE^2N$ has $$ P(W)=\{(0,1),(1,2),(3,2),(4,3),(5,4),(7,4),(8,5)\}, $$ $$ \Delta_{\rm odd}(W)=\{(2,2),(6,2),(8,4)\}, \qquad L(W)=\mathrm{span}_{\mathbb Z}\{(4,0),(2,2)\}. $$ For thin rectangles $(L_x,L_y)=(2d,d)$ with row alternation, the exact criterion is $$ k= \begin{cases} 4,&6\mid d,\\ 0,&6\nmid d. \end{cases} $$ This makes the dependence of code dimension on boundary conditions completely explicit [2602.19057]. ## 3. Decoding anisotropy and directional confinement in related quantum-code constructions The directional-code program in quantum error correction extends beyond code construction to decoding and topological phase engineering. In directionally informed BP decoding for CSS codes, one assigns nonnegative orientation weights $D_Z(i,j)$ to Tanner-graph edges and aggregates them into per-qubit weights $$ w_i=\sum_{j\in N_Z(i)}D_Z(i,j). $$ For a fixed syndrome $s_Z$, each degeneracy class is scored by the minimum weighted Hamming cost $$ \Delta_{\bm w}(E)=\sum_{i=1}^n w_iE_i, \qquad \Delta_*([e])=\min_{u\in S_X}\Delta_{\bm w}(e+u), $$ leading to the directional degeneracy enumerator $$ \Gamma_X(s_Z;\beta)=\sum_{[e]\in\mathcal D_X(s_Z)}\exp\bigl(-\beta\,\Delta_*([e])\bigr),\qquad \beta\ge0. $$ A single bias parameter $\beta$ maps the weights to site-dependent log-likelihood ratios through $$ p_i(\beta)=\frac{p_0\,e^{\beta w_i}}{\tfrac1n\sum_{j=1}^n e^{\beta w_j}}, \qquad \ell_i(\beta)=\log\frac{1-p_i(\beta)}{p_i(\beta)}, $$ which plug directly into standard BP$\rightarrow$OSD decoders without altering the code itself [2601.07240]. The framework yields weighted-distance bounds, $$ w_{\min}d\le d_{\bm w}^{\rm L}\le w_{\max}d, \qquad w_{\min}d_{\mathrm S}\le d_{\bm w}^{\rm S}\le w_{\max}d_{\mathrm S}, $$ and a MacWilliams-type expression for the global directional enumerator. Finite-length code-capacity simulations were reported for a planar NE3N code $\llbracket36,4\rrbracket$ on an $18\times4$ lattice and a toric code $\llbracket2L^2,2,L\rrbracket$ with $L=9$. For both families, a modest bias strength $\beta\approx1\text{--}3$ reduced the logical error rate by one to two orders of magnitude at physical error rates in the $10^{-3}$–$10^{-2}$ range, with a clear optimum around $\beta\approx1.5$–$2$ in the toric-code example [2601.07240]. A different use of directional structure appears in twisted toric codes with explicit anyonic confinement along one spatial direction. There, vertical plaquette terms are twisted by a group $2$-cocycle through projective representations of $G=\mathbb Z_2\times\mathbb Z_2$. A vertical string $$ S^{\rm vert}_\alpha=\bigotimes_{j=1}^{L_y}X^\alpha_g(j) $$ excites $L_y+2$ plaquettes, so its energy cost grows linearly with vertical length, whereas horizontal strings of $X_g$ remain deconfined. Binding a left-confined string and a right-confined string produces $$ S^{\rm dipole}_g=\bigotimes_{j=1}^{L_y}\bigl(X^{\bar\alpha_g}(j)\otimes X^\alpha_g(j)\bigr), $$ which creates dipolar excitations that can move vertically without extra plaquette cost but cannot move horizontally without breaking the bound state. On a torus, the logical-operator structure depends on the parity of $L_y$: when $L_y$ is odd, there are two logical operators and one logical qubit; when $L_y$ is even, there are four logical operators and two logical qubits [2506.22025]. These two developments are technically distinct. One alters priors and degeneracy accounting for a fixed quantum code [2601.07240]; the other alters the Hamiltonian itself to induce anisotropic excitation structure [2506.22025]. This suggests that “directionality” in quantum coding theory now functions at three levels: code construction, decoder design, and phase engineering. ## 4. Direction-preserving number representations In low-precision numerical computing, directional codes are product-structured codes induced by a finite scalar alphabet $A\subset\mathbb R$ common to all coordinates [2605.07662]. For block dimension $n$, the achievable directions are $$ {}_n(A):=\{x/\|x\|_2:\,x\in A^n\setminus\{0\}\}\subset S^{n-1}, $$ and worst-case angular fidelity is measured by the spherical covering radius $$ F_n(A):=\sup_{u\in S^{n-1}}\min_{c\in{}_n(A)}\arccos\langle u,c\rangle. $$ Equivalently, $$ \alpha_n(A):=\inf_{u\in S^{n-1}}\max_{x\in A^n\setminus\{0\}}\frac{\langle u,x\rangle}{\|x\|_2}, \qquad F_n(A)=\arccos \alpha_n(A). $$ For alphabet size $|A|=q$, the best achievable product-code coverage is $$ w_{n,q}:=\inf_{A:|A|=q}F_n(A). $$ The theory compares these product codes with unconstrained spherical codes of the same total size. In dimension $n=2$, the exact planar optimum is $\rho_{\rm sph}(2,m)=\pi/m$. If $A=\{-a,+a\}$, then $F_2(A)=\pi/4=\rho_{\rm sph}(2,4)$. Otherwise, $$ F_2(A)>\rho_{\rm sph}(2,q^2)=\pi/q^2. $$ Except for the antipodal binary case, every two-coordinate product code is therefore strictly worse than the best unconstrained $2$-D spherical code with the same number of points [2605.07662]. In high dimension with fixed alphabet size $q$, the gap persists asymptotically. A harmonic-witness lower bound gives $$ F_n(A)\ge \arccos\min\{1,\,2\sqrt{\lfloor q/2\rfloor/H_n}\}, $$ where $H_n=\sum_{i=1}^n (1/i)\to \log n+\gamma$. A sphere-covering upper bound via Wyner’s theorem shows that for any $\theta>\arcsin(1/q)$ there exists $N$ such that for all $n\ge N$, $$ \rho_{\rm sph}(n,q^n)\le \theta<\pi/2. $$ Combining the two yields $$ \rho_{\rm sph}(n,q^n)