Quantum Locally Testable Codes
- Quantum locally testable codes (qLTCs) are quantum error-correcting codes that quantify how well a state adheres to the codespace via measurement of constraint violations.
- The codes are assessed via locally testable constraints that relate to the average number of constraint violations for detecting state deviations.
- Despite similarities with classical expander codes, stabilizer qLTCs face unique quantum-fine factors obstructing increased soundness in local testability.
Quantum locally testable codes (qLTCs) are quantum error-correcting codes equipped with local constraints such that the fraction of violated constraints lower-bounds the distance of a state from the codespace. For a CSS code with stabilizer generators , local violation projectors are , and quantum local testability is expressed by an operator inequality of the form
where is the distance observable from the codespace and is the soundness. A qLTC is therefore distinguished from a quantum LDPC code: LDPC sparsity constrains check weight and qubit incidence, whereas local testability requires low local-test energy to imply proximity to the codespace. The central parameters are block length, encoded dimension, rate, distance, locality, and soundness. The main problem is to realize constant rate, constant relative distance, constant locality, and constant soundness simultaneously.
1. Definitions and structural framework
A quantum code on qudits is represented as the common zero eigenspace of local projectors,
For a state , the qLTC condition with soundness is
With normalized Hamiltonian, the average constraint energy is
0
For stabilizer codes, the constraints are commuting Pauli generators. If 1 is the Abelian stabilizer group, the code space is
2
A Pauli operator in 3 is a logical operator, and the code distance is the minimum weight of such an operator. For an error 4, the relevant effective weight is often weight modulo the centralizer,
5
The stabilizer-specific notion of local testability, sometimes called an sLTC, requires that every Pauli error of effective weight at least 6 anticommute with at least an 7 fraction of the generating set:
8
For stabilizer codes, the qLTC and sLTC definitions coincide with the same query complexity and soundness (Aharonov et al., 2013). In CSS codes, quantum soundness reduces, up to constant factors, to classical local testability of the two associated parity-check systems.
A CSS code is specified by matrices 9 satisfying
0
The 1- and 2-type logical operators arise from the corresponding kernel quotients, and the quantum distance is the minimum of the two distances. Locality may refer to maximum check weight, maximum qubit incidence, or both. Constant locality and bounded incidence constitute the quantum LDPC condition; they do not imply qLTC soundness.
2. Expansion, commutation, and the first limitations
Interaction graphs encode which qudits participate in which constraints. For a bipartite interaction graph 3, small-set expansion means that every sufficiently small qudit set has nearly the maximum possible number of neighboring constraints. If 4 is a set of qudits, an 5-small-set expander satisfies
6
for 7. Consequently, all but an 8 fraction of constraints touching 9 meet it in exactly one qudit.
Classically, this unique-neighbor property supports robust local testability: a constraint seeing exactly one erroneous bit generally rejects it. In stabilizer codes, however, a constraint touching one erroneous qudit may still commute with the corresponding single-qudit Pauli. This distinction produces the characteristic quantum obstruction.
Aharonov and Eldar showed that for stabilizer codes with 0-local generators and an 1-small-set-expanding interaction graph, sufficiently small-proximity soundness satisfies
2
where relative soundness is normalized by the maximum number of checks incident on the error support (Aharonov et al., 2013). Thus, in the stabilizer setting,
3
The proof selects a separated collection of local constraints and places a single-qudit Pauli error inside each selected support. Expansion supplies a qudit whose incident constraints usually overlap the selected support only once. The Pauli can then commute with most of those single-qudit restrictions while retaining large effective weight modulo the centralizer. The resulting error is far from the codespace but violates only an 4 fraction of the relevant checks. The phenomenon is associated with local noncommutativity and monogamy of entanglement.
The same work established an alphabetic upper bound
5
for 6-dimensional qudits, giving 7 for qubits. For 8, a stronger graph-independent separation below this alphabetic bound was proved under distance and succinctness assumptions. These results apply primarily to stabilizer codes and do not constitute general no-go theorems for arbitrary noncommuting qLTCs.
3. Early constructions and classical-to-quantum contrasts
The first systematic study of qLTCs defined quantum local testability, established the equivalence between qLTCs and sLTCs for stabilizer codes, and formulated a quantum analogue of PCPs of proximity (Aharonov et al., 2013). It also highlighted the contrast with classical expander codes. Classical lossless-expander constructions can have constant rate, constant relative distance, constant locality, and relative soundness approaching one. Stabilizer qLTCs on excellent small-set expanders instead have soundness tending to zero as the expansion error tends to zero.
The toric code provides a distinct low-soundness mechanism. A string-like Pauli error of length 9 may violate only its two endpoints, so the error weight is large while the number of violated checks is constant. This reflects topological order and the concentration of energy at defects rather than throughout the error support.
Hemicubic codes established an explicit global qLTC with one logical qubit, logarithmic locality, minimum distance of order 0, and soundness
1
improving the previously known 2 scale (Leverrier et al., 2019). The construction places qubits on 3-faces of the hypercube with antipodal faces identified. Its CSS structure derives from the boundary and coboundary maps
4
For an error represented by a chain and a cochain, the syndrome is 5. Filling and cofilling inequalities yield the logarithmic soundness. An efficient recursive decoder corrects adversarial errors up to
6
up to the stated polylogarithmic factors.
The generalized hemicubic construction quotients the cube by an arbitrary binary linear code. It gives explicit length, dimension, and distance formulas, but local testability of the generalized family remains unresolved. Its significance is conditional: improved soundness for the one-qubit hemicube could yield qLTCs with growing logical dimension and potentially exponential or polynomial length in the code dimension.
4. Constant-soundness constructions and parameter transformations
Constant soundness does not require large quantum distance. Check-product constructions demonstrate that soundness and distance are logically distinct. Duplicating every bit of a classical LTC produces a code of distance 7 while preserving local testability. Applying the same construction to a CSS code yields a quantum LDPC code with distance 8 and constant soundness (Cross et al., 2022).
More generally, the check product of classical codes
9
satisfies
0
with distance 1. Applied to a quantum CSS code and a classical LTC, it can preserve constant soundness while increasing locality. Taking an auxiliary length 2 yields constant rate, logarithmic distance, logarithmic locality, and constant soundness.
A second construction takes a hypergraph product with a repetition code modified into a star geometry. Unlike a path, whose interior error set can have only two boundary defects, the star gives boundary proportional to the number of edges. It yields constant soundness, distance up to 3, and constant average locality, but maximum locality grows with the repetition length. The distinction between average and worst-case locality is essential: a vanishing fraction of large checks does not satisfy the usual bounded-locality LDPC condition.
Distance balancing with a classical code of length 4 generalizes repetition-code balancing. For a qLTC with parameters 5, the construction gives
6
with soundness
7
provided the auxiliary classical code has independent checks and suitable LDPC behavior (Wills et al., 2023). Balancing both directions yields
8
This increases logical dimension and distance while preserving locality when the auxiliary code is LDPC, at the cost of soundness.
Tradeoff constructions supply three further transformations (Wills et al., 2023). Weight reduction converts high-weight checks into constant-locality checks with polynomial losses in soundness and distance. Soundness amplification adds redundant stabilizers obtained as products of existing stabilizers; it preserves dimension and distance while raising soundness to a constant, with locality increasing polynomially in the inverse original soundness. AEL distance amplification converts sufficiently large relative distance into linear distance, again at the expense of locality and soundness. Applied to asymptotically good quantum LDPC codes, soundness amplification produces asymptotically good testable quantum codes with
9
but without constant locality.
5. Geometric, tensor, and high-dimensional constructions
Tensor-product robustness is a central classical ingredient in qLTC constructions. For codes 0, the tensor product consists of matrices whose rows belong to one code and columns to the other. A pair is 1-robust if
2
Tensor products of algebraic-geometry codes are robustly locally testable when
3
where 4 is constituent dimension and 5 is genus (Garg et al., 2024). This gives an explicit family with super-constant dual distance, but the direct row-and-column tester examines an entire row or column and therefore does not itself provide bounded-query quantum locality.
Quantum Tanner codes combine tensor codes with left-right Cayley complexes. The same local-peeling theorem yields asymptotically good quantum LDPC codes and recovers local testability of an associated classical Tanner code (Leverrier et al., 2022). The quantum code has constant rate and linear minimum distance for fixed underlying degree, but the paper does not prove that the quantum CSS code is itself a qLTC. It establishes quantum LDPC sparsity and quantum distance, while the local-testability theorem concerns the related classical code.
Higher-dimensional cubical and cosheaf constructions provide a direct route to qLTCs. A 6-dimensional cubical complex is equipped with local coefficient spaces and local codes. Product expansion of the local codes yields local filling inequalities; spectral expansion of the direction graphs then gives global cycle and cocycle expansion. These imply both quantum distance and qLTC soundness.
For 7, an explicit construction gives binary CSS codes with constant-weight checks,
8
using arithmetic cubical complexes and two-way product-expanding local codes (Dinur et al., 2024). The construction improves the dimension of earlier geometric qLTCs from constant to linear while retaining constant check weight, at the cost of inverse-polylogarithmic distance and soundness.
The subsequent good-qLTC construction uses arithmetic quotients of products of Bruhat–Tits trees and compatible projective Reed–Solomon local codes. Under a variant of the Bafna–Vyas product-expansion conjecture, non-Abelian cubical complexes with constant directional degree yield binary CSS qLTCs with
9
constant soundness, and constant check weights (Bafna et al., 22 Sep 2026). The construction is conditional: spectral expansion, compatible symmetries, and the CSS construction are established, but the required product expansion for the specified Reed–Solomon families is assumed rather than proved.
6. Asymptotically good qLTCs and current status
An explicit unconditional construction of asymptotically good binary qLTCs is given by higher-dimensional cubical and cosheaf complexes (Gay et al., 17 Sep 2026). The resulting CSS codes have constants 0 and bounded locality such that
1
Every stabilizer check has bounded weight and every qubit participates in boundedly many checks. The construction uses arithmetic quotients of products of Bruhat–Tits trees, compatible local Reed–Solomon codes, and a cosheaf differential. Restriction of scalars from a fixed field 2 to 3 produces qubit codes without changing the asymptotic parameters.
The local-to-global proof has three components: universal product expansion of local codes, spectral expansion of parallel-face graphs, and a double-complex argument transferring cohomological expansion to both 4- and 5-type errors. The stated fixed numerical constants are extremely small, but they are independent of block length. The construction therefore establishes constant rate, constant relative distance, constant-weight local tests, constant soundness, and bounded qubit incidence simultaneously.
A related 2026 construction adds nontrivial transversal multi-controlled-6 gates to good qLTCs (Li et al., 22 Sep 2026). It combines the good qLTC family with cup-product invariants on arithmetic cubical complexes. For every fixed number 7 of code blocks, it gives
8
qLDPC codes and qLTCs with constant soundness supporting a nontrivial transversal logical controlled-9 gate. Earlier almost-good constructions achieved such gates with polylogarithmic losses; the arithmetic covering-space method preserves a nonzero cup-product pairing while allowing the complexes to grow.
The status of the main parameter regimes can be summarized as follows:
| Construction or result | Rate and distance | Soundness and locality | Status |
|---|---|---|---|
| Stabilizer expander limitations | Upper bounds at small proximity | Good expansion can force 0 | Unconditional no-go result for the stated stabilizer regime |
| Hemicubic codes | One logical qubit, 1 | 2 soundness, logarithmic locality | Explicit and unconditional |
| Constant-soundness constructions | Various scalable distances | Constant soundness with growing worst-case locality or reduced rate | Explicit and unconditional |
| Higher-dimensional cosheaf codes | 3, 4 | 5, constant check weight | Explicit and unconditional |
| Product-expansion construction | 6, 7 | Constant soundness and locality | Conditional on Reed–Solomon product expansion |
| Arithmetic cubical qLTCs | 8, 9 | Constant soundness and bounded locality | Explicit and unconditional |
| Transversal-gate extensions | 0, 1 | Constant qLTC soundness and bounded locality | Explicit, with additional algebraic-topological gate construction |
The principal unresolved issues have consequently shifted. The existence of asymptotically good qLTCs is established by the arithmetic cubical construction, but the constants and implementation complexity are severe. Open questions include improving quantitative parameters, simplifying the geometry and local codes, constructing practical decoders, understanding the relationship between qLTC soundness and spectral gaps, and determining how broadly the arithmetic and product-expansion mechanisms generalize. The conditional Reed–Solomon route remains relevant because it may provide alternative constructions with simpler local algebra once the product-expansion conjecture is resolved.
qLTCs also remain connected to the quantum PCP and NLTS programs. A qLTC Hamiltonian converts distance from the codespace into local energy, while constant rate and distance constrain shallow-circuit preparation. However, qLTC existence alone does not prove the quantum PCP conjecture, and prior commuting-Hamiltonian results show that classical expander intuition cannot be transferred directly to stabilizer systems. The modern theory therefore combines high-dimensional expansion, tensor-product robustness, chain complexes, sheaves, and quantum CSS structure rather than relying on ordinary interaction-graph expansion alone.