Erasure-Biased Noise Channel

• Erasure-Biased Noise is a model where most errors convert to detectable erasures, flagging out-of-subspace states for easier decoding.
• This approach is engineered in platforms like neutral atoms and superconducting circuits to leverage hardware co-design for improved error thresholds.
• Utilizing heralded erasures, advanced decoders reduce qubit overhead and boost logical error suppression in quantum error correction schemes.

An erasure-biased noise channel is a quantum or classical error model in which the overwhelming majority of errors take the form of “erasures”—transformations that map a system out of its computational subspace into a detectable orthogonal state (the “erasure flag”), such that the location of the error is known to the decoder. This bias can be engineered in several quantum platforms, notably in neutral atom, superconducting, and spin-optical devices, by converting dominant physical noise (e.g., amplitude damping, atomic loss, leakage) into heralded erasures via auxiliary state detection. The resulting error structure yields profound advantages for information transmission, quantum error correction, metrology, and even the fundamental thermodynamics of memory erasure.

1. Mathematical Definitions and Error Channel Structure

In the quantum setting, the canonical erasure-biased channel acts on the computational Hilbert space $$\mathcal H_{\rm comp} = \mathrm{span}\{\lvert0\rangle, \lvert1\rangle\}$$ with probability $1-p_e$ as the identity and with probability $p_e$ replaces the state by a flagged “erasure” state $\lvert e\rangle$ orthogonal to $\mathcal H_{\rm comp}$: $$\mathcal E(\rho) = (1-p_e)\,\rho + p_e\, \mathrm{Tr}_{\mathcal H_{\rm comp}}(\rho) \lvert e \rangle \langle e \rvert$$ with Kraus operators

$$K_0 = \sqrt{1 - p_e}\, I_{\mathcal Q}\,,\quad K_e = \sqrt{p_e}\, \lvert e \rangle \langle \psi |, \forall |\psi\rangle \in \mathcal Q$$

Residual non-erasure errors (Pauli $X,Y,Z$ or depolarization) are typically strongly suppressed: $p_X, p_Y, p_Z \ll p_e$ (Violaris et al., 5 Jan 2026, Niroula et al., 2023, Dakis et al., 9 Oct 2025).

In the more specialized “biased-erasure” model, transitions to $\lvert e\rangle$ occur predominantly from a particular computational basis state (e.g., only from $1-p_e$0, with $1-p_e$1 and $1-p_e$2), parameterized by an erasure bias $1-p_e$3 (Sahay et al., 2023, Mansouri et al., 11 May 2026).

The classical analog replaces transmission errors (substitutions) by erasures (“?”), which can be detected at the receiver and corrected with different code constructions and round complexity (Grossman et al., 2018). More generally, Lattice Erasure Codes and LDPC constructions leverage erasure structure to maximize noise margins and code rate (Vaishampayan, 2018, Pecorari et al., 27 Feb 2025).

2. Physical Implementation and Mechanisms

Erasure-biased noise arises naturally or can be engineered in several platforms:

• Neutral atom qubits: Dominant two-qubit gate error is leakage from $1-p_e$4 to a detectable auxiliary $1-p_e$5, with negligible leakage from $1-p_e$6 ($1-p_e$7) (Sahay et al., 2023).
• Superconducting dual-rail cavity or transmon qubits: Logical subspace is the single-excitation sector of two modes; amplitude damping maps $1-p_e$8 or $1-p_e$9 to $p_e$0 (heralded erasure), with erasure checks performed via dispersive coupling to an ancilla (Violaris et al., 5 Jan 2026, Dakis et al., 9 Oct 2025).
• Solid-state spin systems: In strain-coupled $p_e$1-manifolds, crystallographic symmetry filters noise into detectable auxiliary states (A$p_e$2 sector; erasure) or echo-suppressed dephasing (E sector; residual Z error) (Mansouri et al., 11 May 2026).
• Spin-optical and photonic links: Qubit loss (photon loss) in distributed architectures directly implements an erasure channel (Galimova, 20 Feb 2026).
• Gate error conversion: Error-correction circuits implement mid-circuit erasure checks using ancilla, projectors, or syndrome extraction to flag and immediately reset or reload erased qubits (Violaris et al., 5 Jan 2026, Li et al., 2024).

3. Impact on Quantum Error Correction: Thresholds and Overhead

The crucial feature of erasure-biased noise is that error locations are known, decoders can exploit this by using zero-weight edges in matching algorithms, and the effective correctable weight is doubled relative to unknown errors.

Key implications for codes and performance include:

Noise Model / Code	Threshold (approx.)	Overhead Reduction / Scaling
Surface code, Pauli	$p_e$3	Distance-$p_e$4: logical $p_e$5
Surface code, pure erasure	$p_e$6	$p_e$7
Biased erasure, XZZX	$p_e$8 Pauli (up to $p_e$9) (Sahay et al., 2023)	64% fewer qubits for fixed $\lvert e\rangle$0 (Mansouri et al., 11 May 2026)
LDPC codes (La-cross), erasure-biased	$\lvert e\rangle$1	Orders of magnitude lower $\lvert e\rangle$2 at $\lvert e\rangle$3 (Pecorari et al., 27 Feb 2025)
Hyperbolic codes, erasure	$\lvert e\rangle$4	Supports large-scale distributed QEC (Galimova, 20 Feb 2026)

Biased-erasure channels result in a strong Z-dephasing bias, which XZZX and Clifford-deformed LDPC codes exploit to approach their optimal erasure threshold. Decoders incorporate erasure flags to dramatically reduce matching weights, allowing exponential scaling in distance $\lvert e\rangle$5 for logical errors. For instance, under the biased-erasure model extracted from symmetry-protected $\lvert e\rangle$6-gates, XZZX decoding on the channel $\lvert e\rangle$7, $\lvert e\rangle$8, $\lvert e\rangle$9 achieves a 64% reduction in data qubits at fixed $\mathcal H_{\rm comp}$0 relative to a generic depolarizing baseline (Mansouri et al., 11 May 2026).

Imperfect erasure checks (efficiency $\mathcal H_{\rm comp}$1) decrease thresholds but maintain more than a $\mathcal H_{\rm comp}$2 increase over Pauli for realistic parameters. The effective error distance per code increases from $\mathcal H_{\rm comp}$3 to $\mathcal H_{\rm comp}$4 for perfect checks, yielding exponentially lower logical error rates at fixed $\mathcal H_{\rm comp}$5 (Chang et al., 2024).

4. Metrological and Information-Theoretic Implications

Erasure-biased noise fundamentally changes the information capacity and metrological performance of both classical and quantum systems.

• Metrology and Quantum Sensing: For a fixed erasure rate $\mathcal H_{\rm comp}$6, measurable erasure errors permit post-selection, yielding quantum Fisher information scaling $\mathcal H_{\rm comp}$7 versus the quadratic penalty $\mathcal H_{\rm comp}$8 for unheralded dephasing (the Cramér–Rao bound yields smaller estimation variance) (Niroula et al., 2023, Arieli et al., 12 Mar 2026). Continuous erasure detection protocols approach within a factor $\mathcal H_{\rm comp}$9 of the ultimate sequential bound and attain Heisenberg scaling in fully biased cases (Arieli et al., 12 Mar 2026). Any platform with detectable leakage can implement this metrological enhancement.
• Interactive and Broadcast Communication: Erasure channels allow higher noise tolerance than substitution errors—interactive protocols can tolerate up to $$\mathcal E(\rho) = (1-p_e)\,\rho + p_e\, \mathrm{Tr}_{\mathcal H_{\rm comp}}(\rho) \lvert e \rangle \langle e \rvert$$0 erasures (large alphabet) and $$\mathcal E(\rho) = (1-p_e)\,\rho + p_e\, \mathrm{Tr}_{\mathcal H_{\rm comp}}(\rho) \lvert e \rangle \langle e \rvert$$1 (binary), exceeding the substitution-error bound (Efremenko et al., 2015, Grossman et al., 2018). Efficient algorithms achieve $$\mathcal E(\rho) = (1-p_e)\,\rho + p_e\, \mathrm{Tr}_{\mathcal H_{\rm comp}}(\rho) \lvert e \rangle \langle e \rvert$$2 (binary) or even $$\mathcal E(\rho) = (1-p_e)\,\rho + p_e\, \mathrm{Tr}_{\mathcal H_{\rm comp}}(\rho) \lvert e \rangle \langle e \rvert$$3 (large alphabet) round complexity for global function computation in erasure-biased settings, breaking classical lower bounds (Grossman et al., 2018).

5. Thermodynamic and Fundamental Entropy Costs

Erasure-biased noise connects directly to the thermodynamics of information. In the strong erasure setting (reset with erased, environment-agnostic macrostate), the minimal entropy cost is $$\mathcal E(\rho) = (1-p_e)\,\rho + p_e\, \mathrm{Tr}_{\mathcal H_{\rm comp}}(\rho) \lvert e \rangle \langle e \rvert$$4 per bit, augmented by an additional entropy $$\mathcal E(\rho) = (1-p_e)\,\rho + p_e\, \mathrm{Tr}_{\mathcal H_{\rm comp}}(\rho) \lvert e \rangle \langle e \rvert$$5 required to suppress thermal fluctuations and ensure reliability $$\mathcal E(\rho) = (1-p_e)\,\rho + p_e\, \mathrm{Tr}_{\mathcal H_{\rm comp}}(\rho) \lvert e \rangle \langle e \rvert$$6 (Norton, 24 Feb 2025). Weak erasure (where the environment retains a trace of the pre-erasure state) can be dissipationless ($$\mathcal E(\rho) = (1-p_e)\,\rho + p_e\, \mathrm{Tr}_{\mathcal H_{\rm comp}}(\rho) \lvert e \rangle \langle e \rvert$$7). The entropy cost is thus governed directly by the statistical bias and structure of the noise, not merely by “missing information” in the Gibbs formalism.

6. Code Design, Hardware Co-Design, and Engineering Techniques

The erasure-bias principle motivates co-design of hardware and error-correction codes. For instance, crystallographic symmetry can structure noise such that parity-filtered (A$$\mathcal E(\rho) = (1-p_e)\,\rho + p_e\, \mathrm{Tr}_{\mathcal H_{\rm comp}}(\rho) \lvert e \rangle \langle e \rvert$$8) erasures become optically distinguishable and tailored Z-dephasing (E-sector) can be echo-suppressed (Mansouri et al., 11 May 2026). Dual-rail qubits with dynamical error-reshaping pulses achieve erasure-bias ratios $$\mathcal E(\rho) = (1-p_e)\,\rho + p_e\, \mathrm{Tr}_{\mathcal H_{\rm comp}}(\rho) \lvert e \rangle \langle e \rvert$$9, leveraging pulse shaping (echo sequences, dynamically corrected gates) to suppress ancilla-induced non-erasure errors by up to three orders of magnitude (Dakis et al., 9 Oct 2025).

Quantum LDPC codes, especially Clifford-deformed and high-rate bicycle codes, exploit the steeper logical error suppression in the erasure-dominated regime, yielding up to orders of magnitude improvements over the surface code (Pecorari et al., 27 Feb 2025).

7. Practical Performance, Limitations, and Open Problems

Hardware demonstrators across superconducting, atomic, and spin platforms have measured heralded erasure rates $$K_0 = \sqrt{1 - p_e}\, I_{\mathcal Q}\,,\quad K_e = \sqrt{p_e}\, \lvert e \rangle \langle \psi |, \forall |\psi\rangle \in \mathcal Q$$0–$$K_0 = \sqrt{1 - p_e}\, I_{\mathcal Q}\,,\quad K_e = \sqrt{p_e}\, \lvert e \rangle \langle \psi |, \forall |\psi\rangle \in \mathcal Q$$1, with residual $$K_0 = \sqrt{1 - p_e}\, I_{\mathcal Q}\,,\quad K_e = \sqrt{p_e}\, \lvert e \rangle \langle \psi |, \forall |\psi\rangle \in \mathcal Q$$2, demonstrating erasure-biased operation sufficient to pass early fault-tolerance thresholds in surface and LDPC codes (Violaris et al., 5 Jan 2026). Code thresholds under erasure bias saturate the 50% bound for surface codes and exceed 30% for distributed hyperbolic codes (Galimova, 20 Feb 2026).

Limitations center on imperfect erasure detection (finite efficiency $$K_0 = \sqrt{1 - p_e}\, I_{\mathcal Q}\,,\quad K_e = \sqrt{p_e}\, \lvert e \rangle \langle \psi |, \forall |\psi\rangle \in \mathcal Q$$3), hardware integration challenges for rapid mid-circuit detection, and ensuring that induced Pauli faults from undetected leakage do not degrade effective code distance. There is active investigation into optimizing detection protocols, decoder toolchains, and hardware design for scalable erasure-biased quantum processors.

Open questions involve optimal code design for hybrid erasure–Pauli models, the fundamental limits of round complexity and erasure correction in both classical and quantum networks, and the role of erasure bias in extending quantum computation and sensing to the fault-tolerant regime (Violaris et al., 5 Jan 2026, Pecorari et al., 27 Feb 2025, Grossman et al., 2018).