1D-Local Sparse Pauli-Lindblad Channel

• 1D-local sparse Pauli-Lindblad channels are quantum noise models defined by generators built from a sparse set of local Pauli strings, ensuring tractable simulation in 1D qubit arrays.
• They leverage both spatial locality and limited parameterization to simplify learning, tomography, and error correction in quantum systems.
• Efficient numerical techniques such as maximum-likelihood estimation and belief propagation facilitate the scalable reconstruction and simulation of these channels.

A 1D-local sparse Pauli-Lindblad channel is a CPTP semigroup $e^{t\mathcal{L}}$ on a one-dimensional qubit array whose generator is built from a sparse set of Pauli strings supported on single sites or short contiguous intervals. In the Pauli-diagonal specialization, one writes

$$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$

with $P_k\in\{I,X,Y,Z\}^{\otimes n}$, $\lambda_k\ge 0$, and $\mathcal{K}$ restricted by geometry and locality; for site-local jumps this reduces at finite time to a tensor product of single-qubit Pauli channels, while for bounded-range terms it defines a quasi-local Markovian noise model adapted to a 1D chain (Jaloveckas et al., 2023, Darmawan et al., 2022).

1. GKSL form and Pauli-diagonal specialization

The ambient continuous-time framework is the Lindblad master equation

$$\frac{d\rho}{dt}=\mathcal{L}(\rho)=-i[H,\rho]+\sum_j\Big(L_j\rho L_j^\dagger-\tfrac12(L_j^\dagger L_j\rho+\rho L_j^\dagger L_j)\Big),$$

with simulation error conventionally measured in diamond norm in algorithmic treatments of sparse Lindbladians (Childs et al., 2016). A Pauli-Lindblad model is the Hermitian-unitary specialization in which the jump operators are Pauli strings $P_k$, so that $P_k^2=I$ and the dissipator collapses to $\lambda_k(P_k\rho P_k-\rho)$ (Jaloveckas et al., 2023).

For a 1D chain, the standard nearest-neighbour two-local form is

$$\mathcal{L}(\rho)= \sum_{i=1}^{N}\sum_{\sigma\in\{X,Y,Z\}} \lambda_{i,\sigma}(\sigma_i\rho\sigma_i-\rho) + \sum_{i=1}^{N-1}\sum_{\sigma,\tau\in\{X,Y,Z\}} \lambda_{i,i+1}^{\sigma,\tau}(\sigma_i\tau_{i+1}\rho \sigma_i\tau_{i+1}-\rho),$$

with only polynomially many nonzero coefficients (Jaloveckas et al., 2023). This is the canonical 1D-local sparse Pauli-Lindblad generator: locality is enforced by support on $$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$0 or $$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$1, and sparsity by the restriction to a small subset of all $$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$2 non-identity Pauli strings.

A particularly important single-site case is the IID Pauli channel

$$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$3

and, for depolarizing noise,

$$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$4

Its standard Markovian embedding is

$$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$5

so that $$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$6 reproduces the depolarizing channel to first order with $$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$7. For an $$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$8-qubit chain with local jump operators $$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$9,

$P_k\in\{I,X,Y,Z\}^{\otimes n}$0

and $P_k\in\{I,X,Y,Z\}^{\otimes n}$1 is exactly a product of single-qubit Pauli channels of the type used in code-capacity analyses (Darmawan et al., 2022).

2. One-dimensional locality and sparsity

In this setting, “local” and “sparse” occur in two distinct senses. First, sparsity in the Pauli basis means that the set $P_k\in\{I,X,Y,Z\}^{\otimes n}$2 of nonzero coefficients is small compared with the full Pauli basis. Second, sparsity in the locality graph means that only low-weight Pauli strings compatible with the hardware topology are kept. For a topology graph $P_k\in\{I,X,Y,Z\}^{\otimes n}$3, a two-local model takes

$P_k\in\{I,X,Y,Z\}^{\otimes n}$4

and in a 1D chain $P_k\in\{I,X,Y,Z\}^{\otimes n}$5, so weight-2 terms act only on nearest neighbours (Jaloveckas et al., 2023).

For the linear topology used in sparse Pauli-Lindblad learning and PEC, the allowed 1D two-local model contains all single-qubit Paulis and all two-qubit Paulis on consecutive sites,

$P_k\in\{I,X,Y,Z\}^{\otimes n}$6

so the parameter count is $P_k\in\{I,X,Y,Z\}^{\otimes n}$7 (Berg et al., 2022).

A more abstract formulation uses the interaction hypergraph $P_k\in\{I,X,Y,Z\}^{\otimes n}$8, with

$P_k\in\{I,X,Y,Z\}^{\otimes n}$9

each $\lambda_k\ge 0$0 supported on at most $\lambda_k\ge 0$1 qubits. In the Pauli-GKSL basis, $\lambda_k\ge 0$2-locality implies $\lambda_k\ge 0$3 whenever $\lambda_k\ge 0$4, and 1D bounded-range interactions have bounded intersection degree and bounded weighted interaction strength $\lambda_k\ge 0$5 (Möbus et al., 22 Jun 2026). This bounded-intersection structure is what turns many 1D learning and reconstruction problems from exponential to polynomial, or even polylogarithmic in $\lambda_k\ge 0$6, once the support is supplied.

3. Effective channels and correction on a 1D chain

A central operational use of 1D-local sparse Pauli-Lindblad channels is as the physical noise model for 1D quantum coding. In the code-capacity setting of low-depth random Clifford encoders, qubits are laid out on a 1D chain, the encoding circuit is nearest-neighbour and logarithmic-depth, and each physical qubit experiences IID Pauli noise after a noiseless encoding circuit. Because Clifford conjugation maps Pauli operators to Pauli operators, the logical noise remains Pauli, but its support is confined by the encoder light cone; for a depth-$\lambda_k\ge 0$7 1D encoder, each check satisfies $\lambda_k\ge 0$8, and the induced logical channel has error strings of length $\lambda_k\ge 0$9 with numerically observed short-range logical failure correlations (Darmawan et al., 2022).

For an IID Pauli channel,

$\mathcal{K}$0

random stabilizer codes achieve the hashing-bound rate

$\mathcal{K}$1

and for depolarizing noise this becomes

$\mathcal{K}$2

The numerics of low-depth 1D Clifford encoding show that, with a tensor-network maximum-likelihood decoder, thresholds $\mathcal{K}$3 are very close to the hashing bound even when the encoding depth is only $\mathcal{K}$4 (Darmawan et al., 2022).

$\mathcal{K}$5	$\mathcal{K}$6 (TN decoder)	$\mathcal{K}$7 (hashing)
$\mathcal{K}$8	$\mathcal{K}$9	$$\frac{d\rho}{dt}=\mathcal{L}(\rho)=-i[H,\rho]+\sum_j\Big(L_j\rho L_j^\dagger-\tfrac12(L_j^\dagger L_j\rho+\rho L_j^\dagger L_j)\Big),$$0
$$\frac{d\rho}{dt}=\mathcal{L}(\rho)=-i[H,\rho]+\sum_j\Big(L_j\rho L_j^\dagger-\tfrac12(L_j^\dagger L_j\rho+\rho L_j^\dagger L_j)\Big),$$1	$$\frac{d\rho}{dt}=\mathcal{L}(\rho)=-i[H,\rho]+\sum_j\Big(L_j\rho L_j^\dagger-\tfrac12(L_j^\dagger L_j\rho+\rho L_j^\dagger L_j)\Big),$$2	$$\frac{d\rho}{dt}=\mathcal{L}(\rho)=-i[H,\rho]+\sum_j\Big(L_j\rho L_j^\dagger-\tfrac12(L_j^\dagger L_j\rho+\rho L_j^\dagger L_j)\Big),$$3
$$\frac{d\rho}{dt}=\mathcal{L}(\rho)=-i[H,\rho]+\sum_j\Big(L_j\rho L_j^\dagger-\tfrac12(L_j^\dagger L_j\rho+\rho L_j^\dagger L_j)\Big),$$4	$$\frac{d\rho}{dt}=\mathcal{L}(\rho)=-i[H,\rho]+\sum_j\Big(L_j\rho L_j^\dagger-\tfrac12(L_j^\dagger L_j\rho+\rho L_j^\dagger L_j)\Big),$$5	$$\frac{d\rho}{dt}=\mathcal{L}(\rho)=-i[H,\rho]+\sum_j\Big(L_j\rho L_j^\dagger-\tfrac12(L_j^\dagger L_j\rho+\rho L_j^\dagger L_j)\Big),$$6
$$\frac{d\rho}{dt}=\mathcal{L}(\rho)=-i[H,\rho]+\sum_j\Big(L_j\rho L_j^\dagger-\tfrac12(L_j^\dagger L_j\rho+\rho L_j^\dagger L_j)\Big),$$7	$$\frac{d\rho}{dt}=\mathcal{L}(\rho)=-i[H,\rho]+\sum_j\Big(L_j\rho L_j^\dagger-\tfrac12(L_j^\dagger L_j\rho+\rho L_j^\dagger L_j)\Big),$$8	$$\frac{d\rho}{dt}=\mathcal{L}(\rho)=-i[H,\rho]+\sum_j\Big(L_j\rho L_j^\dagger-\tfrac12(L_j^\dagger L_j\rho+\rho L_j^\dagger L_j)\Big),$$9
$P_k$0	$P_k$1	$P_k$2

Below threshold, the bulk logical error obeys

$P_k$3

and for $P_k$4 the fitted thresholds match the hashing values to within a few $P_k$5 (Darmawan et al., 2022). Since a site-local Pauli-Lindblad generator at finite time induces exactly the same class of single-qubit Pauli channels, this directly carries over to the finite-time discretization of 1D-local Pauli-Lindblad noise used in that analysis (Darmawan et al., 2022).

4. Learning, tomography, and probabilistic error cancellation

Sparse Pauli-Lindblad models were introduced precisely to make learning and inversion of correlated noise tractable. In the two-local model

$P_k$6

with

$P_k$7

the forward map from Pauli fidelities to $P_k$8 is linear in log-space, and on a 1D chain the model has only $P_k$9 parameters in the full two-local nearest-neighbour case (Berg et al., 2022).

For cycle benchmarking on low-degree topologies that include linear chains, a constant number of global measurement bases suffices: there exist nine Pauli strings such that for each edge $P_k^2=I$0, the substrings at locations $P_k^2=I$1 and $P_k^2=I$2 exactly cover $P_k^2=I$3. This is the structural reason that learning two-local sparse Pauli-Lindblad models is particularly efficient in 1D (Jaloveckas et al., 2023).

The same locality also makes full maximum-likelihood estimation tractable. For a 1D-local sparse Pauli-Lindblad channel, the likelihood of observed bit strings under product-state preparation, a 1D layer of CZ gates, and product Pauli-basis measurement reduces to an efficiently-evaluable Bayesian network. The latent variables are Bernoulli error events attached to local Pauli generators, and the observed bits depend only on the parity of the local errors that touch them; exact evaluation is then performed by belief propagation on a bounded-width factor graph (Belkin et al., 2 Jun 2026). In numerical studies, MLE and EPF both showed $P_k^2=I$4, but MLE reached a given MSE with roughly one third as many samples as EPF in the high-precision regime, and in a 10-qubit Trotterized Ising example this improved channel learning extended accurate PEC from about $P_k^2=I$5 to $P_k^2=I$6 (Belkin et al., 2 Jun 2026).

Short-time, local-control learning results complement this layerwise picture. For a general unknown $P_k^2=I$7-local Lindblad generator on $P_k^2=I$8 qubits, product-state preparation, short-time evolution, and single-qubit Pauli measurements suffice to estimate all Hamiltonian and dissipative Pauli-GKSL coefficients to entrywise accuracy $P_k^2=I$9 with probability at least $\lambda_k(P_k\rho P_k-\rho)$0 using

$\lambda_k(P_k\rho P_k-\rho)$1

samples, and a semidefinite projection yields a valid $\lambda_k(P_k\rho P_k-\rho)$2-local Lindblad generator with diamond-norm error at most $\lambda_k(P_k\rho P_k-\rho)$3 using

$\lambda_k(P_k\rho P_k-\rho)$4

samples. When the exact support has bounded intersection degree, as in a 1D chain, the dependence on system size can improve to

$\lambda_k(P_k\rho P_k-\rho)$5

for coefficient learning (Möbus et al., 22 Jun 2026).

5. Efficient simulation and channel reconstruction

The algorithmic theory of 1D-local sparse Pauli-Lindblad channels has two complementary strands: direct simulation of $\lambda_k(P_k\rho P_k-\rho)$6 and scalable reconstruction of the resulting channel. In the simulation direction, sparse Lindbladians with Pauli-string structure fall simultaneously into the “sparse Lindblad operator” and “local Lindbladian” settings. For $\lambda_k(P_k\rho P_k-\rho)$7-qubit Lindbladians presented as linear combinations of polynomially many Pauli strings, there is a quantum algorithm implementing a channel $\lambda_k(P_k\rho P_k-\rho)$8 with

$\lambda_k(P_k\rho P_k-\rho)$9

using gate complexity

$$\mathcal{L}(\rho)= \sum_{i=1}^{N}\sum_{\sigma\in\{X,Y,Z\}} \lambda_{i,\sigma}(\sigma_i\rho\sigma_i-\rho) + \sum_{i=1}^{N-1}\sum_{\sigma,\tau\in\{X,Y,Z\}} \lambda_{i,i+1}^{\sigma,\tau}(\sigma_i\tau_{i+1}\rho \sigma_i\tau_{i+1}-\rho),$$0

when the Pauli specification is polynomial in $$\mathcal{L}(\rho)= \sum_{i=1}^{N}\sum_{\sigma\in\{X,Y,Z\}} \lambda_{i,\sigma}(\sigma_i\rho\sigma_i-\rho) + \sum_{i=1}^{N-1}\sum_{\sigma,\tau\in\{X,Y,Z\}} \lambda_{i,i+1}^{\sigma,\tau}(\sigma_i\tau_{i+1}\rho \sigma_i\tau_{i+1}-\rho),$$1 (Cleve et al., 2016). Childs–Li gave complementary sparse/local algorithms and proved a no-fast-forwarding theorem for sparse Lindbladians in black-box models, so generic sublinear-in-$$\mathcal{L}(\rho)= \sum_{i=1}^{N}\sum_{\sigma\in\{X,Y,Z\}} \lambda_{i,\sigma}(\sigma_i\rho\sigma_i-\rho) + \sum_{i=1}^{N-1}\sum_{\sigma,\tau\in\{X,Y,Z\}} \lambda_{i,i+1}^{\sigma,\tau}(\sigma_i\tau_{i+1}\rho \sigma_i\tau_{i+1}-\rho),$$2 simulation is excluded even though 1D-local Pauli structure makes the concrete constants favorable (Childs et al., 2016).

In the characterization direction, local-to-global reconstruction methods exploit the fact that 1D channels generated by local Lindbladians often have Choi states with exponentially decaying conditional mutual information. Under this approximate Markov condition, local process-shadow tomography on contiguous windows plus locally optimal recovery maps reconstructs an MPO representation of the global Choi state with sample complexity polynomial in $$\mathcal{L}(\rho)= \sum_{i=1}^{N}\sum_{\sigma\in\{X,Y,Z\}} \lambda_{i,\sigma}(\sigma_i\rho\sigma_i-\rho) + \sum_{i=1}^{N-1}\sum_{\sigma,\tau\in\{X,Y,Z\}} \lambda_{i,i+1}^{\sigma,\tau}(\sigma_i\tau_{i+1}\rho \sigma_i\tau_{i+1}-\rho),$$3 and the desired global reconstruction error (Liu et al., 7 Mar 2026). The protocol was demonstrated numerically for a 50-qubit channel generated by a local Lindbladian with Heisenberg Hamiltonian

$$\mathcal{L}(\rho)= \sum_{i=1}^{N}\sum_{\sigma\in\{X,Y,Z\}} \lambda_{i,\sigma}(\sigma_i\rho\sigma_i-\rho) + \sum_{i=1}^{N-1}\sum_{\sigma,\tau\in\{X,Y,Z\}} \lambda_{i,i+1}^{\sigma,\tau}(\sigma_i\tau_{i+1}\rho \sigma_i\tau_{i+1}-\rho),$$4

and dephasing jumps

$$\mathcal{L}(\rho)= \sum_{i=1}^{N}\sum_{\sigma\in\{X,Y,Z\}} \lambda_{i,\sigma}(\sigma_i\rho\sigma_i-\rho) + \sum_{i=1}^{N-1}\sum_{\sigma,\tau\in\{X,Y,Z\}} \lambda_{i,i+1}^{\sigma,\tau}(\sigma_i\tau_{i+1}\rho \sigma_i\tau_{i+1}-\rho),$$5

recovering global diagnostics such as the process fidelity, the Choi state purity, and Pauli-weight-resolved process matrix elements (Liu et al., 7 Mar 2026).

These two strands are complementary rather than redundant. Simulation algorithms assume an explicit $$\mathcal{L}(\rho)= \sum_{i=1}^{N}\sum_{\sigma\in\{X,Y,Z\}} \lambda_{i,\sigma}(\sigma_i\rho\sigma_i-\rho) + \sum_{i=1}^{N-1}\sum_{\sigma,\tau\in\{X,Y,Z\}} \lambda_{i,i+1}^{\sigma,\tau}(\sigma_i\tau_{i+1}\rho \sigma_i\tau_{i+1}-\rho),$$6; reconstruction algorithms infer $$\mathcal{L}(\rho)= \sum_{i=1}^{N}\sum_{\sigma\in\{X,Y,Z\}} \lambda_{i,\sigma}(\sigma_i\rho\sigma_i-\rho) + \sum_{i=1}^{N-1}\sum_{\sigma,\tau\in\{X,Y,Z\}} \lambda_{i,i+1}^{\sigma,\tau}(\sigma_i\tau_{i+1}\rho \sigma_i\tau_{i+1}-\rho),$$7 from local data. In a 1D-local sparse Pauli-Lindblad setting, both benefit from the same structural facts: bounded support, bounded effective width, and Pauli-basis compatibility.

6. Markovianity, generalized rates, and deeper structure

The phrase “Pauli-Lindblad” is frequently used as shorthand for a Pauli channel with nonnegative generator rates, but this equivalence is exact only at the level of channel semigroup Markovianity. For a Pauli channel,

$$\mathcal{L}(\rho)= \sum_{i=1}^{N}\sum_{\sigma\in\{X,Y,Z\}} \lambda_{i,\sigma}(\sigma_i\rho\sigma_i-\rho) + \sum_{i=1}^{N-1}\sum_{\sigma,\tau\in\{X,Y,Z\}} \lambda_{i,i+1}^{\sigma,\tau}(\sigma_i\tau_{i+1}\rho \sigma_i\tau_{i+1}-\rho),$$8

real nonnegative Pauli-Lindblad parameters are necessary and sufficient for channel semigroup Markovianity; equivalently, a Pauli channel is non-Markovian if and only if at least one of its Pauli-Lindblad parameters is negative (Kattemölle et al., 9 Feb 2026). This matters because Pauli twirling does not preserve Markovianity: physically Markovian channels often become non-Markovian after twirling, so an accurate effective description may require negative Pauli-Lindblad parameters even when the underlying laboratory noise is generated by a Lindbladian (Kattemölle et al., 9 Feb 2026).

The single-snapshot analysis of Pauli channels strengthens this point. Every Pauli channel admits a generator of Pauli-Lindblad form, but the rates may be negative or complex; in that case the appropriate object is a Pauli pseudo-Lindblad generator rather than a GKSL generator. Random Pauli channels are almost always non-Markovian, with the probability of encountering a negative rate converging doubly exponentially to unity with the number of qubits, and experimentally learned two-local nearest-neighbour models on superconducting devices are more accurate when negative rates are allowed than when one imposes $$\mathcal{L}(\rho)= \sum_{i=1}^{N}\sum_{\sigma\in\{X,Y,Z\}} \lambda_{i,\sigma}(\sigma_i\rho\sigma_i-\rho) + \sum_{i=1}^{N-1}\sum_{\sigma,\tau\in\{X,Y,Z\}} \lambda_{i,i+1}^{\sigma,\tau}(\sigma_i\tau_{i+1}\rho \sigma_i\tau_{i+1}-\rho),$$9 (Seif et al., 13 Feb 2026).

A different structural refinement arises in stabilizer-based 1D Pauli-Lindblad models with commuting frustration-free Hamiltonians and Pauli-string jumps. There the relevant decomposition is not only spatial but operator-theoretic: the Liouvillian bond algebra and its commutant induce an exponential fragmentation of operator space into dynamically disconnected fragments. In the 1D cluster chain, $$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$00 jump operators lead within each fragment to an effective non-Hermitian transverse-field Ising model and free-fermion solvability, whereas $$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$01 jumps produce non-integrable fragments with spectral chaos and PT-symmetric spectral statistics (Paszko et al., 19 Jun 2025). This establishes that even within the class of 1D-local sparse Pauli-Lindblad channels, locality and sparsity do not imply uniform dynamical behavior; integrable, chaotic, and exceptional-point regimes can coexist across different fragments.

Standard caveats remain model-dependent. Code-capacity results near the hashing bound assume IID Pauli noise on physical qubits, noiseless encoding gates and stabilizer measurements, strictly 1D nearest-neighbour Clifford encoders, and numerical rather than rigorous threshold evidence (Darmawan et al., 2022). Learning guarantees assume fixed $$\mathcal{L}(\rho)=\sum_{k\in\mathcal{K}} \lambda_k\big(P_k \rho P_k^\dagger-\rho\big),$$02, bounded weighted interaction strength, and short-time access; extensions to long-range interactions, higher-dimensional geometries, and non-Pauli or non-Markovian error models are treated only partially or heuristically in the present literature (Möbus et al., 22 Jun 2026, Belkin et al., 2 Jun 2026).