Threshold Fidelity in Quantum Systems

• Threshold fidelity is a critical boundary that determines if quantum operations, such as Bell-pair or gate fidelities, meet quality standards for viability.
• It appears in various contexts including quantum repeater networks, fault-tolerant computation, teleportation, steering, and surface-code analyses.
• Crossing the threshold enables effective error correction and purification, serving as a benchmark for genuine quantum performance in diverse systems.

Threshold fidelity denotes a critical fidelity boundary used to separate acceptable from unacceptable operation, but the quantity being thresholded varies with the operational setting. In quantum repeater networks it is an application-level minimum usable Bell-pair fidelity $F_{\min}$; in fault-tolerant computation it is the minimum physical-gate or resource-state fidelity above which error correction or distillation suppresses errors rather than amplifying them; in teleportation and steering it is a benchmark that separates classical or unsteerable behavior from genuinely quantum performance; and in surface-code analyses with correlated environments it appears as a critical point of one-cycle logical fidelity under microscopic noise models (Ercetin et al., 10 Feb 2026, Noiri et al., 2021, Roa et al., 2015, Jouzdani et al., 2014).

1. Core meanings and mathematical forms

Across the cited literature, fidelity is not a single object but a family of task-specific figures of merit. For a two-qubit state $\rho$ and a target Bell state $|\psi_{\mathrm{tar}}\rangle$, the standard state fidelity is

$$F=\langle \psi_{\mathrm{tar}}|\rho|\psi_{\mathrm{tar}}\rangle.$$

In fault-tolerant gate analysis, if a gate is modeled as an ideal unitary followed by a depolarizing channel with error probability $p$, the average gate fidelity is approximately $F \approx 1-p$ for small $p$. In teleportation, the benchmark is the average output fidelity $f$, with $f=2/3$ marking the best purely classical measure-and-prepare strategy. In steering, the relevant quantity is an averaged fidelity $F_{\mathrm{avg}}$, compared with a nonsteering threshold $\rho$0. In correlated-noise surface-code analyses, the threshold is often expressed not directly as a fidelity value but as a critical coupling $\rho$1 or $\rho$2 beyond which the logical fidelity no longer approaches 1 in the thermodynamic limit (Ercetin et al., 10 Feb 2026, Noiri et al., 2021, Roa et al., 2015, Wu et al., 2019, Novais et al., 2012).

Domain	Fidelity object	Threshold criterion
Quantum repeater networks	End-to-end Bell-pair fidelity $\rho$3	Usable only if $\rho$4
Fault-tolerant QC	Physical gate fidelity	Surface-code target about $\rho$5; error threshold $\rho$6 to $\rho$7
Magic-state distillation	Logical magic-state fidelity	$\rho$8 for H-type and $\rho$9 for T-type
Teleportation	Average fidelity $|\psi_{\mathrm{tar}}\rangle$0	Quantum if $|\psi_{\mathrm{tar}}\rangle$1
Steering	Averaged fidelity $|\psi_{\mathrm{tar}}\rangle$2	Steerable if $|\psi_{\mathrm{tar}}\rangle$3
Correlated-noise surface code	One-cycle logical fidelity $|\psi_{\mathrm{tar}}\rangle$4	Threshold at critical coupling $|\psi_{\mathrm{tar}}\rangle$5 or $|\psi_{\mathrm{tar}}\rangle$6

Taken together, these formulations show that threshold fidelity is best understood as an operational boundary: a state, gate, route, decoder, or measurement protocol is counted as successful only when it lies on the correct side of a problem-specific threshold.

2. Application-level threshold fidelity in quantum networks

In quantum repeater networks, threshold fidelity is an explicit service constraint. Using a Werner-state approximation, the Werner parameter is

$|\psi_{\mathrm{tar}}\rangle$7

and after entanglement swapping along a path $|\psi_{\mathrm{tar}}\rangle$8 with links $|\psi_{\mathrm{tar}}\rangle$9,

$$F=\langle \psi_{\mathrm{tar}}|\rho|\psi_{\mathrm{tar}}\rangle.$$0

A delivered pair is usable only if $$F=\langle \psi_{\mathrm{tar}}|\rho|\psi_{\mathrm{tar}}\rangle.$$1; otherwise it is treated as a failure or discarded. In this formulation, “delivering a pair that has decohered below $$F=\langle \psi_{\mathrm{tar}}|\rho|\psi_{\mathrm{tar}}\rangle.$$2 is equivalent to a delivery failure” (Ercetin et al., 10 Feb 2026).

This threshold is built directly into the usable-delivery indicator and into the Fidelity-Age (FA) metric. FA measures the interval since the last delivery whose fidelity exceeds $$F=\langle \psi_{\mathrm{tar}}|\rho|\psi_{\mathrm{tar}}\rangle.$$3; only such deliveries reset the age process. The renewal identity

$$F=\langle \psi_{\mathrm{tar}}|\rho|\psi_{\mathrm{tar}}\rangle.$$4

shows that stricter thresholds worsen age by reducing the usable success probability. Simulations on slotted repeater grids show that FA-aware scheduling preserves throughput while reducing extreme-age events by up to two orders of magnitude, and in a $$F=\langle \psi_{\mathrm{tar}}|\rho|\psi_{\mathrm{tar}}\rangle.$$5 baseline with $$F=\langle \psi_{\mathrm{tar}}|\rho|\psi_{\mathrm{tar}}\rangle.$$6 the FA-aware policies keep throughput essentially unchanged while driving mean and tail age from catastrophic values to near-zero starvation (Ercetin et al., 10 Feb 2026).

A related but distinct network-level use appears in distributed routing with purification. Q-GUARD and Q-GUARD-WS treat each request as $$F=\langle \psi_{\mathrm{tar}}|\rho|\psi_{\mathrm{tar}}\rangle.$$7 and enforce the request-specific threshold in a distributed protocol where nodes exchange link-state information only with their $$F=\langle \psi_{\mathrm{tar}}|\rho|\psi_{\mathrm{tar}}\rangle.$$8-hop neighbors. Q-GUARD converts the global threshold into per-hop Werner targets using an equal-split rule, builds purification cost tables from realized Bell pairs, and ranks candidate spans with the segment-local expected-goodput metric

$$F=\langle \psi_{\mathrm{tar}}|\rho|\psi_{\mathrm{tar}}\rangle.$$9

On synthetic 100-node topologies, Q-GUARD raises the qualified success rate from under $p$0 to over $p$1 on 4-hop paths and nearly doubles the qualified service radius in Euclidean distance relative to throughput-only and naive-purification baselines; Q-GUARD-WS adds further gains under high hardware heterogeneity by allocating purification non-uniformly across hops (Gatti et al., 30 Apr 2026).

3. Fault-tolerance, gate thresholds, and distillation thresholds

In fault-tolerant quantum computing, threshold fidelity is the operational form of the threshold theorem. The theorem states that there exists a critical error rate $p$2 such that, if the physical error probability per gate and measurement is below this value, larger codes can make logical error rates arbitrarily small. Under a small-error depolarizing approximation, this becomes a gate-fidelity threshold $p$3, placing the familiar surface-code target at about $p$4 (Noiri et al., 2021). For superconducting qubits, the surface code is described as having a per-step fidelity threshold of only about $p$5, while for spin-qubit fault tolerance the surface-code threshold is quoted as $p$6 to $p$7 in error-rate language (Barends et al., 2014, Xue et al., 2021).

Several experiments in the supplied literature are organized explicitly around this threshold. A superconducting five-qubit processor reported an average single-qubit gate fidelity of $p$8 and a two-qubit gate fidelity up to $p$9, placing Josephson quantum computing at the fault-tolerant threshold for surface-code error correction (Barends et al., 2014). In silicon spin qubits, fast electrical control produced single-qubit primitive gate fidelities $F \approx 1-p$0, two-qubit primitive gate fidelity $F \approx 1-p$1, and interleaved-RB CNOT fidelity $F \approx 1-p$2, explicitly beyond the fault-tolerance threshold (Noiri et al., 2021). A spin-based silicon processor characterized by gate-set tomography reported all single- and two-qubit gate fidelities above $F \approx 1-p$3, with average single-qubit fidelity in the full two-qubit space of $F \approx 1-p$4, and framed this as computing at the surface-code error threshold (Xue et al., 2021). In a different platform, robust single-qubit Clifford gates protected by composite pulses and dynamical decoupling reached average fidelities of up to $F \approx 1-p$5, exceeding the threshold value for some quantum error-correction schemes (Souza et al., 2015).

Threshold fidelity also governs resource-state preparation. In logical magic-state distillation, the threshold is the minimum input fidelity above which distillation improves the state. For the protocol studied on the rotated surface code, the threshold fidelity is $F \approx 1-p$6 for H-type magic states and $F \approx 1-p$7 for T-type magic states. The experiment reports logical fidelities $F \approx 1-p$8 for $F \approx 1-p$9, $p$0 for $p$1, and $p$2 for $p$3, all above their relevant distillation thresholds even when state-preparation and measurement errors are included (Ye et al., 2023).

A recurrent point in these works is that threshold fidelity is necessary but not sufficient for scalable fault tolerance. Crossing the threshold establishes that error correction or distillation can in principle suppress errors; further requirements include local connectivity, parallelism, leakage control, readout quality, and decoder performance (Noiri et al., 2021, Ye et al., 2023).

4. Teleportation and steering benchmarks

In quantum teleportation, threshold fidelity is a direct boundary between classical and quantum performance. For an unknown pure qubit $p$4, the best purely classical measure-and-prepare protocol yields average fidelity

$p$5

A teleportation protocol displays quantum features only if its average fidelity exceeds this value (Roa et al., 2015).

For noisy X-state channels, the average teleportation fidelity is

$p$6

Expressed in terms of concurrence, the channel’s entanglement is necessary but not sufficient for $p$7. The paper introduces a threshold concurrence

$p$8

so that quantum teleportation requires $p$9, not merely $f$0. With probabilistic and unambiguous state extraction (USE), fidelity can be redistributed across outcomes: the filtered successful branch can have normalized average fidelity above $f$1, and its threshold concurrence $f$2 can be smaller than $f$3, while the failure branch has average fidelity below $f$4 (Roa et al., 2015).

In EPR steering, the benchmark is not $f$5 but the nonsteering threshold $f$6. For Bob’s target projectors $f$7, the averaged fidelity is

$f$8

The nonsteering threshold is the maximum value of this averaged fidelity compatible with any local-hidden-state model. If

$f$9

the shared state is steerable from Alice to Bob, and Alice’s measurements are verified to be incompatible (Wu et al., 2019). In the qubit equatorial case, the geometric nonsteering threshold is $f=2/3$0, and for two and three Pauli settings the construction recovers the familiar optimal linear steering criteria.

These two settings—teleportation and steering—use threshold fidelity as an operational witness rather than a hardware specification. The threshold is a certification boundary: above it, the observed behavior cannot be reproduced by the relevant classical or nonsteering model (Roa et al., 2015, Wu et al., 2019).

5. Surface-code fidelity thresholds under correlated environments

A separate line of work studies threshold fidelity as a phase transition of one-cycle logical fidelity under correlated microscopic noise. For a surface code initially in $f=2/3$1, one paper defines the post-cycle logical fidelity under a no-error syndrome as

$f=2/3$2

with $f=2/3$3 and $f=2/3$4. The critical parameter $f=2/3$5 is defined as the value of $f=2/3$6 that separates the regime where $f=2/3$7 from the regime where $f=2/3$8 in the thermodynamic limit. For nearest-neighbor interactions the approximate critical value is

$f=2/3$9

where $F_{\mathrm{avg}}$0 is the effective nearest-neighbor coupling and $F_{\mathrm{avg}}$1 is the lattice connectivity constant. Longer-range interactions reduce $F_{\mathrm{avg}}$2, and if all qubits lie in the “causality cone” there is never a threshold (Novais et al., 2012).

A more general exact formulation writes the effective code evolution as

$F_{\mathrm{avg}}$3

with

$F_{\mathrm{avg}}$4

This framework goes beyond stochastic single-qubit error models and incorporates correlated errors and inhomogeneous couplings. For homogeneous nearest-neighbor couplings, the threshold is mapped to the 2D Ising critical point

$F_{\mathrm{avg}}$5

and Monte Carlo simulations give $F_{\mathrm{avg}}$6, close to the analytical prediction $F_{\mathrm{avg}}$7 (Jouzdani et al., 2014).

At finite temperature, a threshold still exists. The finite-temperature analysis states that for regimes relevant to current experiments, quantum error correction works well even in the presence of environment-induced, long-range inter-qubit interactions; that a threshold always exists at finite temperatures; that for the super-Ohmic case the critical coupling constant separating high- from low-fidelity decreases with increasing temperature; that for both Ohmic and super-Ohmic cases the dependence of the critical coupling on temperature is weak; and that for the sub-Ohmic case it also depends strongly on the duration of the QEC cycle (Novais et al., 2016). Numerical work on the zero-temperature non-Markovian case shows that different interaction ranges set different intrinsic bounds on the fidelity of the code and that these bounds are unrelated to error thresholds based on stochastic error models (Jouzdani et al., 2014).

Within this literature, threshold fidelity is not a percentage target such as $F_{\mathrm{avg}}$8. It is a critical point of a many-body statistical model that separates a high-fidelity phase from a low-fidelity phase. This suggests a different but closely related use of the term: threshold fidelity can denote a collective property of the code-plus-environment system, not only a property of individual gates or states.

6. Readout, decoding, and broader methodological usages

Threshold dependence also appears in measurement and decoding. In single-shot spin readout for semiconductor quantum dots, the conventional analysis depends on a threshold voltage $F_{\mathrm{avg}}$9 and readout time $\rho$00. Readout visibility is

$\rho$01

and the measured spin-up probability obeys

$\rho$02

with dark count $\rho$03. The proposed threshold-independent method extrapolates

$\rho$04

and reports an effective area 60 times larger than that of the most commonly used threshold-based method, while emphasizing that the extrapolated-probability error cannot be neglected without constraints on $\rho$05 and $\rho$06 (Hu et al., 2022). Here threshold fidelity is not a physical phase boundary but a dependence of inferred fidelity on analysis thresholds.

A complementary measurement-oriented usage appears in superconducting transmon readout. A frequency-multiplexed readout scheme that combines shelving, two-tone resonator excitation, and machine-learning post-processing reaches $\rho$07 assignment fidelity for two-state readout and $\rho$08 for three-state readout in $\rho$09 ns, explicitly positioning readout fidelity at the threshold for quantum error correction without a quantum-limited amplifier (Chen et al., 2022).

In topological decoding, a GAN-based toric-code decoder is assigned a logical fidelity threshold operationally through crossings of logical-fidelity curves for different code distances. The paper reports a decoder threshold of about $\rho$10, compared with $\rho$11 for the classical decoding model, and teleportation fidelity improvements within the depolarizing-noise threshold range $\rho$12 for $\rho$13 and $\rho$14 for $\rho$15 (Li et al., 2024). This is another task-specific threshold: the maximum physical noise rate below which encoding and decoding improve effective fidelity.

A distinct methodological usage appears in multi-fidelity stochastic simulation, where fidelity denotes tunable simulation quality rather than quantum-state quality. The target quantity is the probability that a scalar output exceeds a critical threshold at the highest fidelity level,

$\rho$16

and the Maximum Speed of Uncertainty Reduction strategy chooses both physical input and fidelity level to maximize uncertainty reduction per unit cost (Stroh et al., 2017). This suggests that “threshold fidelity” can also denote a design problem coupling a fidelity level with threshold-exceedance estimation, rather than a threshold on fidelity itself.

Across these usages, the unifying feature is operational selectivity: threshold fidelity formalizes when a measured, routed, decoded, or simulated object is counted as acceptable for the task at hand.