Realizing Logical Diagonal Gates via Transversal Physical $Z$-Rotations in CSS Codes

Abstract: Calderbank-Shor-Steane (CSS) codes, constructed from nested classical codes $C_2 \subseteq C_1$, are typically optimized for good code parameters. However, practical quantum computing equally demands fault-tolerant logical gates. In this work, we characterize nested pairs $(C_1, C_2)$ whose resulting CSS codes realize a target logical diagonal gate via transversal physical $Z$-rotations. In doing so, we recover a result of Camps-Moreno et al. that CSS codes can realize only logical single-qubit $Z$-rotations and multi-qubit controlled-$Z$ rotations via transversal physical $Z$-rotations. Building on our characterization, we develop the ''appending construction'', that takes as input an $[[n',k']]$ CSS code $Q'$ and a target logical $Z$-rotation (single-qubit or multi-controlled) $U_L$, and extends $Q'$ by systematically appending $n''$ physical qubits to obtain an $[[n,k]]$ CSS code $Q$ with $n = n'+n''$ and $k=k'$. The target logical gate $U_L$ is realized in $Q$ by applying a well-chosen physical transversal $Z$-rotation to the $n''$ appended physical qubits. The CSS code $Q$ may incur a loss in minimum distance, but the loss can be controlled through the parameter choices made in the construction. By repeatedly applying the appending construction, we can extend any CSS code $Q'$ to obtain a CSS code $Q$ that supports fault-tolerant implementations of multiple desired logical $Z$-rotations. The cost to be paid for this is the increased physical qubit overhead as the number of target logical gates grows.

• The paper establishes necessary and sufficient modular conditions for transversal physical Z-rotations to implement logical single-qubit and multi-controlled-Z gates in CSS codes, including the bound p ≥ ℓ + m − 1 for addressable (m−1)-controlled rotations.
• The appending construction extends any primary CSS code with dedicated auxiliary blocks for arbitrary target gates while preserving logical-qubit count and controlling distance loss, including a worked [[15,7,3]] Steane-code example producing a [[43,7,≥2]] code for an addressable logical T gate.
• The resulting asymptotic families achieve [[n,Θ(n),Ω(n)]] codes for fixed addressable S gates, while higher-level rotations face sublinear-distance, non-LDPC, and auxiliary-code limitations tied to open questions about divisible classical codes.

This paper develops a theory and a constructive framework for CSS quantum error-correcting codes that implement target logical diagonal gates fault-tolerantly via transversal physical $Z$-rotations (2608.19094). The work proceeds in three stages: an exact characterization of nested classical code pairs $(C_2 \subseteq C_1)$ whose associated CSS codes realize a prescribed logical diagonal gate; a modular "appending construction" that extends any primary CSS code to support an arbitrary set of such gates; and explicit asymptotic code families realizing addressable logical $S$ gates and higher-level $Z$-rotations.

Characterization of realizable logical diagonal gates

The central technical result is a necessary-and-sufficient condition for a dyadic transversal physical $Z$-rotation $$U(p,w) = diag\left(\exp{\iota \frac{\pi}{2^p}(w \cdot x)} : x \in \mathbb{F}_2^n\right)$$ to realize a target logical gate $$U_L = diag\left(\exp{\iota \frac{\pi}{2^{\ell}} f(a)} : a \in \mathbb{F}_2^k\right)$$ on a CSS code $(C_1, C_2)_{\mathrm{CSS}}$. The condition requires $p \ge \ell$ together with three families of modular equations involving the dot products of $w$ with codewords of $(C_2 \subseteq C_1)$0, with coset representatives of $(C_2 \subseteq C_1)$1, and with their Schur products. A key structural consequence, recovered independently by Camps-Moreno et al., is that the function $(C_2 \subseteq C_1)$2 cannot be arbitrary: it decomposes over the coset basis $(C_2 \subseteq C_1)$3 and its $(C_2 \subseteq C_1)$4-fold Schur products, so CSS codes can realize only logical single-qubit $(C_2 \subseteq C_1)$5-rotations and multi-controlled-$(C_2 \subseteq C_1)$6 rotations through transversal physical $(C_2 \subseteq C_1)$7-rotations. This is a hard restriction on what any CSS construction of this type can achieve.

For addressable multi-controlled-$(C_2 \subseteq C_1)$8 rotations, the paper strengthens the level requirement: realizing an addressable $(C_2 \subseteq C_1)$9-controlled rotation $S$0 forces $S$1, proved by contradiction using the parity structure of the address tensor. The proof technique reduces global modular constraints on the full coset space to equivalent constraints on basis vectors and their Schur products, which is what makes the subsequent constructive framework tractable. A relaxed variant of the characterization is also derived, in which the strict modulo-$S$2 conditions are weakened to modulo-$S$3; the resulting vector can be lifted to one satisfying the original equations, so the relaxed conditions characterize codes realizing the target gate up to physical Pauli $S$4 corrections. The authors note this relaxation cannot be pushed further without introducing bilinear terms that would require non-transversal two-qubit gates.

The appending framework

The constructive contribution exploits the locality of the characterization: satisfying the modular equations depends only on the appended coordinates on which the rotation vector $S$5 is supported. Given a primary $S$6 CSS code and target gates $S$7, the construction appends dedicated block-columns to the generator matrices of $S$8 and $S$9. Each block is assigned to one target gate, and the corresponding transversal rotation acts exclusively on that block's qubits. Because $Z$0 becomes a direct sum, the derived code's parameters satisfy $Z$1 and $Z$2, so the number of logical qubits is preserved and the distance loss is controlled by the auxiliary codes used to build the blocks.

The appending matrices themselves are extracted from an auxiliary CSS code $Z$3 in which the physical transversal $Z$4 realizes the logical transversal $Z$5: rows of its coset generator matrix supply the required weight congruences. When the target address vector has more than $Z$6 nonzero entries, the target gate is factored into sub-gates each supported on at most $Z$7 logical qubits, and the corresponding physical rotations are composed. For multi-controlled-$Z$8 rotations, the construction is inductive in the number of controls: the base case is the single-qubit construction, and each $Z$9-controlled gate is built from lower-order matrices, with residual logical phases from lower-order Schur-product terms cancelled by appending matrices realizing the inverse gates (inverses being obtained simply by negating $Z$0). A worked example starting from the $Z$1 Steane code and a punctured Reed–Muller auxiliary code yields a $Z$2 code realizing an addressable logical $Z$3 gate on three specified logical qubits.

Asymptotic code families

For $Z$4 (logical $Z$5), combining asymptotically good doubly-even self-dual codes as both primary and auxiliary components yields an asymptotically good family $Z$6 realizing any fixed sequence of addressable logical $Z$7 gates; using explicit asymptotically good self-orthogonal codes with good duals makes the family explicit via the relaxed characterization. For $Z$8, the bottleneck is the auxiliary code: punctured Reed–Muller codes give $Z$9 codes realizing logical transversal $$U(p,w) = diag\left(\exp{\iota \frac{\pi}{2^p}(w \cdot x)} : x \in \mathbb{F}_2^n\right)$$0, with $$U(p,w) = diag\left(\exp{\iota \frac{\pi}{2^p}(w \cdot x)} : x \in \mathbb{F}_2^n\right)$$1. Instantiating against an asymptotically good primary family produces:

Regime	Parameters	Gate realized
Fixed addresses, $$U(p,w) = diag\left(\exp{\iota \frac{\pi}{2^p}(w \cdot x)} : x \in \mathbb{F}_2^n\right)$$2	$$U(p,w) = diag\left(\exp{\iota \frac{\pi}{2^p}(w \cdot x)} : x \in \mathbb{F}_2^n\right)$$3	$$U(p,w) = diag\left(\exp{\iota \frac{\pi}{2^p}(w \cdot x)} : x \in \mathbb{F}_2^n\right)$$4
Fixed addresses, $$U(p,w) = diag\left(\exp{\iota \frac{\pi}{2^p}(w \cdot x)} : x \in \mathbb{F}_2^n\right)$$5, support $$U(p,w) = diag\left(\exp{\iota \frac{\pi}{2^p}(w \cdot x)} : x \in \mathbb{F}_2^n\right)$$6	$$U(p,w) = diag\left(\exp{\iota \frac{\pi}{2^p}(w \cdot x)} : x \in \mathbb{F}_2^n\right)$$7, $$U(p,w) = diag\left(\exp{\iota \frac{\pi}{2^p}(w \cdot x)} : x \in \mathbb{F}_2^n\right)$$8	$$U(p,w) = diag\left(\exp{\iota \frac{\pi}{2^p}(w \cdot x)} : x \in \mathbb{F}_2^n\right)$$9
Any address, $$U_L = diag\left(\exp{\iota \frac{\pi}{2^{\ell}} f(a)} : a \in \mathbb{F}_2^k\right)$$0	$$U_L = diag\left(\exp{\iota \frac{\pi}{2^{\ell}} f(a)} : a \in \mathbb{F}_2^k\right)$$1	any $$U_L = diag\left(\exp{\iota \frac{\pi}{2^{\ell}} f(a)} : a \in \mathbb{F}_2^k\right)$$2
Any address, $$U_L = diag\left(\exp{\iota \frac{\pi}{2^{\ell}} f(a)} : a \in \mathbb{F}_2^k\right)$$3, $$U_L = diag\left(\exp{\iota \frac{\pi}{2^{\ell}} f(a)} : a \in \mathbb{F}_2^k\right)$$4	$$U_L = diag\left(\exp{\iota \frac{\pi}{2^{\ell}} f(a)} : a \in \mathbb{F}_2^k\right)$$5	any $$U_L = diag\left(\exp{\iota \frac{\pi}{2^{\ell}} f(a)} : a \in \mathbb{F}_2^k\right)$$6
Any address, $$U_L = diag\left(\exp{\iota \frac{\pi}{2^{\ell}} f(a)} : a \in \mathbb{F}_2^k\right)$$7, $$U_L = diag\left(\exp{\iota \frac{\pi}{2^{\ell}} f(a)} : a \in \mathbb{F}_2^k\right)$$8	$$U_L = diag\left(\exp{\iota \frac{\pi}{2^{\ell}} f(a)} : a \in \mathbb{F}_2^k\right)$$9	any $(C_1, C_2)_{\mathrm{CSS}}$0

The tradeoff between target-support size and minimum distance in the second row is explicit: larger $(C_1, C_2)_{\mathrm{CSS}}$1 degrades $(C_1, C_2)_{\mathrm{CSS}}$2. Analogous families for multi-controlled-$(C_1, C_2)_{\mathrm{CSS}}$3 rotations follow from the general appending-matrix constructions.

Limitations and open questions

Several limitations are stated plainly by the authors. First, the derived families are not LDPC: guaranteeing $(C_1, C_2)_{\mathrm{CSS}}$4-distance by bounding the weight of the entire logical-$(C_1, C_2)_{\mathrm{CSS}}$5 space forces dense $(C_1, C_2)_{\mathrm{CSS}}$6-stabilizers, and constructing LDPC auxiliary codes realizing logical transversal $(C_1, C_2)_{\mathrm{CSS}}$7 remains open. Second, the sub-linear scaling of the best known CSS codes realizing logical transversal higher-level $(C_1, C_2)_{\mathrm{CSS}}$8-rotations caps the achievable parameters; the paper poses the underlying classical question—whether asymptotically good $(C_1, C_2)_{\mathrm{CSS}}$9-divisible codes with asymptotically good duals exist for $p \ge \ell$0—whose resolution would directly improve all Reed–Muller-based families here. Third, the appending construction incurs physical overhead proportional to the number of target gates, since each gate receives dedicated coordinates; length-optimized direct constructions would sacrifice the framework's flexibility. Finally, the relaxed characterizations have so far been exploited only for $p \ge \ell$1 and $p \ge \ell$2 gates, where punctured self-orthogonal codes substitute for doubly-even ones; whether they yield improved codes more broadly is unresolved.

Conclusion

The paper provides both a complete algebraic characterization of when transversal physical $p \ge \ell$3-rotations realize target logical diagonal gates in CSS codes and a systematic, composable construction—appending—that converts any primary CSS code into one supporting an arbitrary prescribed set of addressable single-qubit and multi-controlled-$p \ge \ell$4 rotations at bounded distance loss. The resulting asymptotic families demonstrate that diverse diagonal gate sets, including non-Clifford levels beyond $p \ge \ell$5, can be consolidated within single CSS codes, complementing code-switching approaches that need only supply the logical Hadamard. The principal quantitative limits—sub-linear distances for $p \ge \ell$6, non-LDPC stabilizers, and per-gate overhead—are tied to specific open problems in divisible-code theory rather than to intrinsic barriers of the framework itself.