Spacetime Code in Quantum Error Correction

• Spacetime code is a quantum error-correction framework where circuit processes and measurement outcomes merge into a single object spanning qubits and timesteps.
• It employs methods that lift Clifford circuit measurement outcomes to stabilizer operators, enabling decoding through standard error-correction techniques.
• Applications include dynamical codes like Floquet systems and tensor-network constructions, with performance evaluated via fault distance and LDPC properties.

In the circuit-centric quantum error-correction literature, a spacetime code is a representation in which a fault-tolerant process is treated as a single object distributed over qubits and timesteps rather than as a static code supplemented by a separate syndrome-extraction circuit. One formulation defines it as a subsystem stabilizer code on “spacetime qubits,” with one physical qubit degree of freedom for each qubit$\times$timestep in the circuit, and with stabilizers identified with the circuit’s detectors (Pesah et al., 11 Sep 2025). A complementary formulation starts from the set of all possible measurement outcomes of a Clifford circuit, organizes those outcomes as a linear code, and lifts that outcome code to a stabilizer “spacetime code” on virtual qubits (Delfosse et al., 2023). Related work extends the notion to Floquet and other dynamical codes, tensor-network descriptions, chain-complex models, and gauge-theoretic constructions (Blackwell et al., 7 Oct 2025).

1. Definition, scope, and contrast with static codes

A static $[[n,k,d]]$ stabilizer code is ordinarily specified by an Abelian stabilizer group together with a fixed syndrome-extraction circuit. In the spacetime-code viewpoint, the distinction between “the code” and “the circuit that protects it” is collapsed: code and syndrome extraction become one object, so that the fault distance is intrinsic to the spacetime code itself (Pesah et al., 11 Sep 2025). This is the sense in which dynamical codes, including Floquet codes and subsystem protocols with time-dependent gauges, fit naturally into the framework (Pesah et al., 11 Sep 2025).

In this formulation, gauge operators encode trivial circuit-level propagations, such as the placement of a Pauli before a gate versus after a gate, while stabilizers or “detectors” arise from redundancies among measurement outcomes or from input-stabilizer postselections (Pesah et al., 11 Sep 2025). A common misconception is that a spacetime code is merely a static code with repeated syndrome measurements written in an extended notation. The literature instead assigns it its own logical content, detector structure, and decoding problem at circuit level (Pesah et al., 11 Sep 2025).

The same broad perspective appears in dynamical-code work on periodic Pauli measurements. There, spacetime codes are described as a natural generalization of static Pauli stabilizer codes to settings in which quantum information is protected by an ongoing sequence of Pauli measurements rather than by a fixed stabilizer group (Blackwell et al., 7 Oct 2025). This suggests that the term is best understood not as a single construction, but as a family of closely related formalisms whose common feature is the promotion of time from external schedule to encoded structure.

2. Outcome codes and the stabilizer construction from Clifford circuits

For a depth-$\Delta$ Clifford circuit $\mathcal C$ on $n$ qubits with $m$ Pauli measurements, every run produces an $m$-bit outcome string $o\in\mathbb F_2^m$. The set of all possible outcome strings is an affine subspace of $\mathbb F_2^m$, and after a trivial sign redefinition it may be taken to be a linear subspace $C_{\mathrm{out}}\subseteq\mathbb F_2^m$ (Delfosse et al., 2023). Equivalently, one computes a full set of checks $[[n,k,d]]$0, writes the parity-check matrix $[[n,k,d]]$1, and characterizes the outcome code as

$[[n,k,d]]$2

The spacetime code $[[n,k,d]]$3 is then obtained by lifting checks of $[[n,k,d]]$4 to Pauli operators on $[[n,k,d]]$5 virtual qubits. For each $[[n,k,d]]$6, one defines a “pre-fault” operator $[[n,k,d]]$7 by placing each measured Pauli $[[n,k,d]]$8 immediately before its measurement when $[[n,k,d]]$9, and then applies the back-cumulant or backward-propagation map $\Delta$0 to obtain the spacetime-check operator

$\Delta$1

Proposition 4.2 and Theorem 4.3 show that $\Delta$2 is a commuting set of Pauli operators without $\Delta$3, and hence the stabilizer generators of a $\Delta$4 code with

$\Delta$5

This construction extends earlier circuit-to-code correspondences to include intermediate and multi-qubit measurements (Delfosse et al., 2023).

The decoding statement is equally central. For a product-fault model on spacetime Pauli faults, a most-likely-error decoder for $\Delta$6 can be transformed into a most-likely-fault decoder for the original circuit (Delfosse et al., 2023). In effect, fault correction for the circuit is reduced to error correction in a stabilizer code. This is operationally significant because the latter problem admits standard decoders and established complexity analyses.

The construction is algorithmic rather than merely existential. Outcome-code checks are obtained by a modified stabilizer-simulation routine in the Heisenberg picture, and the stated implementation runs in $\Delta$7 time using tableau or Gauss-Jordan techniques (Delfosse et al., 2023). The paper also gives an algorithm to generate low-weight spacetime checks by searching bounded-radius balls in a spacetime graph and extracting local stabilizers by restricted Gaussian elimination (Delfosse et al., 2023).

Locality properties can be inherited from the circuit. If the family of circuits is “bounded,” meaning that the checks of $\Delta$8 can be chosen to have $\Delta$9 weight and $\mathcal C$0 depth, and if every gate acts on $\mathcal C$1 qubits, then each spacetime check has $\mathcal C$2 weight and the resulting spacetime code is LDPC. If, in addition, gates are geometrically local in $\mathcal C$3 spatial dimensions, then the spacetime graph is a $\mathcal C$4-dimensional lattice and the spacetime checks are local in $\mathcal C$5 dimensions (Delfosse et al., 2023).

3. Dynamical codes, benign errors, and spacetime distance

For dynamical codes built from periodic sets of commuting Pauli measurements, the spacetime formalism begins with a sequence $\mathcal C$6, with $\mathcal C$7 measured simultaneously at time $\mathcal C$8. If $\mathcal C$9, the code is a Floquet code of period $n$0 (Blackwell et al., 7 Oct 2025). The associated spacetime Pauli errors form the direct-sum space

$n$1

where $n$2 is inserted immediately after the measurements at time $n$3. The natural weight is

$n$4

Detectability is defined through detectors and ancestries of stabilizers. Measuring $n$5 at time $n$6 yields a detector, and an error is undetectable precisely when it commutes, in the spacetime sense, with every detector (Blackwell et al., 7 Oct 2025). This separates dynamical decoding from the simpler static picture: the relevant observables are not just instantaneous stabilizers, but spacetime parity relations generated by the measurement history.

A crucial distinction is the class of benign errors. These are generated by two operations: inserting a measurement operator $n$7 at the time it is measured, and inserting identical Pauli operators on adjacent time slices around a commuting measurement, forming a “sandwiching” error (Blackwell et al., 7 Oct 2025). Every benign error is undetectable, and every instantaneous stabilizer is itself benign (Blackwell et al., 7 Oct 2025). Equivalent errors are defined by quotienting by benign errors.

The main bounded-inference result states that, in any bounded-inference dynamical code, every undetectable error in the steady stage is equivalent to a one-time-slice error $n$8 commuting with the instantaneous stabilizer group at time $n$9; equivalently, it is benign plus a logical operator of the instantaneous code (Blackwell et al., 7 Oct 2025). The resulting dichotomy is sharp: an undetectable error is correctable if and only if it is benign (Blackwell et al., 7 Oct 2025). Accordingly, the code distance is defined as the minimal weight of an undetectable non-benign spacetime Pauli error in the steady stage (Blackwell et al., 7 Oct 2025).

This distance need not coincide with any instantaneous code distance. The planar honeycomb example has instantaneous surface-code distance $m$0 but admits a constant-weight spacetime logical of weight $m$1 (Blackwell et al., 7 Oct 2025). By contrast, for the Floquet Bacon–Shor code on a $m$2 grid, one finds $m$3 (Blackwell et al., 7 Oct 2025). A recurrent misunderstanding is therefore that large instantaneous distance automatically implies large spacetime distance; the cited examples show that the dynamical schedule can introduce much shorter non-benign undetectable histories.

4. Algebraic, graphical, and gauge-theoretic formalisms

One algebraic formulation models a spacetime code by a length-2 chain complex

$m$4

with $m$5 the gauge space, $m$6 the error space, and $m$7 the detector space (Pesah et al., 11 Sep 2025). In this language,

$m$8

is the space of logical errors, while

$m$9

is the space of logical correlations (Pesah et al., 11 Sep 2025). The distance becomes

$m$0

and minimum-weight decoding is the constrained minimization of $m$1 subject to $m$2 (Pesah et al., 11 Sep 2025).

This complex formalism supports transformations between spacetime codes. A weak chain map induces a map on $m$3, and a fault-tolerant map is defined as a weak quasi-isomorphism that preserves the number of encoded qubits, fault distance, and minimum-weight decoding problem (Pesah et al., 11 Sep 2025). The equivalence of spacetime codes is then characterized by the existence of fault-tolerant maps in both directions (Pesah et al., 11 Sep 2025). The same framework extends foliated cluster-state constructions from stabilizer codes to arbitrary spacetime codes and shows that any Clifford circuit can be transformed into a measurement-based protocol with the same fault-tolerant properties (Pesah et al., 11 Sep 2025).

A complementary graphical formalism represents any Clifford circuit with Pauli measurements as a three-colored tensor network built from blue, red, and green “spider” projectors corresponding to the Pauli bases (Fuente et al., 2024). Its basic invariants are Pauli flows: assignments of highlights to internal edges and measurement labels satisfying the projective symmetries of each tensor. Flows form an $m$4-vector space with color addition given by the $m$5 fusion rules, such as red+blue=green and $m$6any=none (Fuente et al., 2024). Detector flows, stabilizer flows, and logical flows then encode, respectively, detector constraints, commuting stabilizer groups on open legs, and logical Pauli action through the network (Fuente et al., 2024).

From detector flows one builds a decoding hypergraph or, equivalently, a factor graph with parity-check matrix $m$7. Decoding becomes a MAP or belief-propagation-plus-OSD problem on the induced classical code: find the lowest-weight fault configuration consistent with the measured detector syndrome, then infer logical flips from an observable matrix $m$8 (Fuente et al., 2024). An important claim of this framework is that a well-defined decoding problem can be derived from the tensor network and its Pauli flows alone, independent of any stabilizer code or fixed circuit (Fuente et al., 2024).

A further reformulation gauges the spacetime code into a $m$9 lattice gauge theory. Starting from the ECO matrix of a Clifford circuit, one introduces a gauge-field bit $o\in\mathbb F_2^m$0 for each elementary-circuit operator and imposes Gauss laws

$o\in\mathbb F_2^m$1

for each matter site $o\in\mathbb F_2^m$2 (Lee, 4 Jun 2026). In this gauged theory, Gauss laws encode fault-equivalence under multiplication by elementary-circuit operators, while Wilson loops

$o\in\mathbb F_2^m$3

for redundancies $o\in\mathbb F_2^m$4 are in one-to-one correspondence with detectors (Lee, 4 Jun 2026). The same paper identifies three applications: foliated MBQC, classical memory in topologically ordered mixed states, and learning of Pauli noise, where the learnable observables coincide with the gauge-invariant detector degrees of freedom (Lee, 4 Jun 2026).

5. Families, compilation methods, and concrete protocols

The spacetime-code viewpoint has been used not only to analyze existing protocols but also to generate new ones. One example is the XYZ ruby code, a family of Floquet codes derived from topological subsystem codes and described by the three-colored tensor-network calculus. Its instantaneous stabilizer groups alternate among three phases; at $o\in\mathbb F_2^m$5 one has the hexagonal color-code stabilizer group, while after $o\in\mathbb F_2^m$6 and $o\in\mathbb F_2^m$7 measuring steps one obtains phases locally equivalent to three decoupled toric codes (Fuente et al., 2024). Over each length-3 cycle, the protocol applies a $o\in\mathbb F_2^m$8 automorphism to the logical Pauli group, and circuit-level benchmarks on a torus report thresholds $o\in\mathbb F_2^m$9, $\mathbb F_2^m$0, and $\mathbb F_2^m$1 (Fuente et al., 2024).

A different construction, spacetime concatenation, treats a dynamical code as an instrument $\mathbb F_2^m$2 connecting an incoming and an outgoing stabilizer code while preserving logical information up to a logical unitary (Xu et al., 11 Apr 2025). The map is decomposed into local Clifford gadgets connected by internal “bond” legs, with binary encoding matrices $\mathbb F_2^m$3 and $\mathbb F_2^m$4 satisfying the symplectic condition $\mathbb F_2^m$5, or $\mathbb F_2^m$6 in the CSS case (Xu et al., 11 Apr 2025). Two sufficient conditions for fault tolerance are isolated: the Bond-Kernel-Rank Condition, which prevents measurement of logical operators, and the strict-$\mathbb F_2^m$7-locality-preservation condition, which controls Pauli-web spreading. Under these conditions, Theorem 2 gives $\mathbb F_2^m$8 for macroscopic static distance $\mathbb F_2^m$9 (Xu et al., 11 Apr 2025). Explicit examples include the dynamical bivariate bicycle code and a dynamical Haah code, and local defect adaptations are described for broken connectors and qubit dropout, both with a reported distance reduction to $C_{\mathrm{out}}\subseteq\mathbb F_2^m$0 (Xu et al., 11 Apr 2025).

The same circuit-level perspective also supports the design of specific logical gates. For the surface code, a twist-defect braiding protocol has been proposed that implements a logical $C_{\mathrm{out}}\subseteq\mathbb F_2^m$1 gate in spacetime volume

$C_{\mathrm{out}}\subseteq\mathbb F_2^m$2

compared with $C_{\mathrm{out}}\subseteq\mathbb F_2^m$3 and $C_{\mathrm{out}}\subseteq\mathbb F_2^m$4 for earlier schemes (Hirai et al., 15 Apr 2026). The protocol proceeds by expanding the patch, creating a triangular $C_{\mathrm{out}}\subseteq\mathbb F_2^m$5 domain-wall “tent,” braiding one twist defect diagonally, performing $C_{\mathrm{out}}\subseteq\mathbb F_2^m$6 rounds of sequential Pauli-$C_{\mathrm{out}}\subseteq\mathbb F_2^m$7 measurements, and then shrinking the patch back to its original size (Hirai et al., 15 Apr 2026). Circuit-level realizations are given both with constant-length non-local gates and with nearest-neighbor two-qubit gates on a square grid, without additional two-qubit gate depth beyond standard syndrome extraction (Hirai et al., 15 Apr 2026).

The reported fault distances for this $C_{\mathrm{out}}\subseteq\mathbb F_2^m$8-gate block are $C_{\mathrm{out}}\subseteq\mathbb F_2^m$9 for the Bombín protocol, $[[n,k,d]]$00 for the Gidney protocol, $[[n,k,d]]$01 for the proposed non-local version, and $[[n,k,d]]$02 for the proposed local version (Hirai et al., 15 Apr 2026). Monte Carlo with PyMatching yields logical error rates at $[[n,k,d]]$03 for $[[n,k,d]]$04, and the proposed protocols are described as having logical error rates comparable to existing methods at large code distances $[[n,k,d]]$05 and physical error rates near $[[n,k,d]]$06 despite the reduced fault distance (Hirai et al., 15 Apr 2026). A plausible implication is that spacetime volume, rather than static code distance alone, can be the dominant optimization variable in compiled fault-tolerant operations.

6. Broader and distinct usages of the term

A distinct line of work uses “spacetime code” in a finite-geometric setting. There the boundary is a blown-up Gibbons–Hoffman–Wootters phase space $[[n,k,d]]$07, whose totally isotropic $[[n,k,d]]$08-planes correspond to maximal commuting sets of $[[n,k,d]]$09-qubit Pauli observables, while the bulk is obtained from the Grassmannian of these planes via the Plücker embedding into a projective space $[[n,k,d]]$10 with $[[n,k,d]]$11 (Lévay et al., 2018). Choosing an isotropic spread gives a constant-dimension subspace code in $[[n,k,d]]$12 over $[[n,k,d]]$13 with parameters $[[n,k,d]]$14 (Lévay et al., 2018).

In that model, messages are complete sets of commuting observables associated with Lagrangian subspaces, corrupted messages are lower- or higher-dimensional subspaces, and decoding is geometric: one maps an error subspace $[[n,k,d]]$15 to a Schubert variety $[[n,k,d]]$16, then intersects its Plücker image with a distinguished code slice $[[n,k,d]]$17 to recover the unique codeword (Lévay et al., 2018). For $[[n,k,d]]$18, the bulk is the Klein quadric $[[n,k,d]]$19, and line intersection on the boundary corresponds to light-like separation in the bulk (Lévay et al., 2018). This usage is related to error correction, but it is not the same as the circuit-level spacetime-code formalism of the recent quantum fault-tolerance literature.

An even looser, explicitly metaphorical usage appears in a paper deriving the Lorentz and Poincaré algebras from the commutator and anticommutator structure of the Pauli matrices. There, one may regard the Pauli-matrix algebra as the “code” underpinning spacetime symmetry: rotations arise from commutators, boosts from anticommutators, and translations from appropriately chosen vector matrices in a reducible representation (Shurtleff, 2010). This is conceptually suggestive, but it is not a coding-theoretic spacetime code in the quantum-error-correction sense.

Taken together, these usages show that “spacetime code” is not a monosemous term. In present arXiv usage, its dominant technical meaning is the QEC notion in which a protected dynamical process is encoded directly in spacetime, with detectors, logical action, and decoding defined at circuit level. Broader finite-geometric and algebraic usages retain the same intuition—that spacetime structure can itself be encoded—but deploy it in mathematically distinct ways (Delfosse et al., 2023).