Break-Even Iso-Fidelity Analysis

• The paper introduces a quantitative framework where encoded or multi-fidelity methods reach a break-even point by equating logical and physical fidelity through error propagation and cost analysis.
• Break-even iso-fidelity analysis is a method that determines when advanced encoding or multifidelity approaches equal traditional methods by using precise error metrics and resource allocation models.
• It applies across quantum computing and surrogate modeling by benchmarking logical versus physical processes, guiding optimal strategies for error mitigation and resource efficiency.

Break-even iso-fidelity analysis quantitatively demarcates the regime in which operations using encoded (logical) entities—such as qubits in quantum error correction, predictions from multifidelity surrogate models, or simulation runs at varying simulator fidelities—first match or surpass the fidelity of the best corresponding unencoded (physical, single-fidelity, or low-level) counterpart. Across fields, the break-even or iso-fidelity point is precisely defined by equality of logical (or multi-fidelity) and physical (or baseline) fidelity metrics under well-characterized error models, resource constraints, and postselection/gain overhead—enabling systematic certification of whether and when advanced error mitigation, encoding, or simulation design truly delivers a quantitative advantage.

1. Fundamental Definitions: Iso-Fidelity and Break-Even Criteria

The break-even iso-fidelity point is defined as the parameter locus (e.g., physical error rate $p_\text{phys}$, circuit depth, simulation budget, or process time) where the fidelity of the encoded, error-mitigated, or multi-fidelity scheme equals that of the physical or baseline process,

$F_\text{log}(p, ...) = F_\text{phys}(p, ...)$

using consistent, operationally meaningful definitions of fidelity (Dasu et al., 25 Feb 2026, Riffel et al., 14 Apr 2026, Hong et al., 2024, Gupta et al., 2023, Ni et al., 2022, Hao et al., 26 Nov 2025, Joshi et al., 4 Mar 2026). In quantum computing, $F_\text{phys}$ and $F_\text{log}$ are typically derived from error rates per qubit, gate, or logical block, with

$$F_\text{phys} = 1 - p_\text{phys}, \quad F_\text{log} = 1 - p_\text{log}$$

where $p_\text{log}$ accounts for syndrome extraction, postselection, and concatenation effects. In multifidelity modeling, fidelity may correspond to the mean squared error (MSE) of a prediction, mutual information in measurement theory, or process-specific metrics. The iso-fidelity condition may then be expressed as equality of cost-vs.-error frontiers:

$$C_\text{multi-fid}(\epsilon) = C_\text{single-fid}(\epsilon) \Rightarrow \epsilon = \epsilon^*$$

the smallest tolerable error $\epsilon^*$ where the two approaches become equivalent in accuracy-to-cost ratio (Chen et al., 2024, Renganathan et al., 2022, Baheri et al., 2023).

2. Analytical Frameworks: Scaling Laws and Break-Even Modeling

Break-even analysis relies on explicit error propagation and cost/fidelity scaling models:

• Quantum Codes: For $[[n,k,d]]$ codes, logical error rates obey $p_L \sim A p^{d/2}$ for even $F_\text{log}(p, ...) = F_\text{phys}(p, ...)$0 under depolarizing noise and postselection (Dasu et al., 25 Feb 2026). For concatenated codes, $F_\text{log}(p, ...) = F_\text{phys}(p, ...)$1.
• Resource Rate: The code rate $F_\text{log}(p, ...) = F_\text{phys}(p, ...)$2 controls the per-logical-qubit overhead required to achieve $F_\text{log}(p, ...) = F_\text{phys}(p, ...)$3.
• Costs in Multifidelity: Under Gaussian process and autoregressive (ARGP) models, multi-level (ML) cost to error $F_\text{log}(p, ...) = F_\text{phys}(p, ...)$4 scales as $F_\text{log}(p, ...) = F_\text{phys}(p, ...)$5, while any single-fidelity (SL) design must pay $F_\text{log}(p, ...) = F_\text{phys}(p, ...)$6 (Chen et al., 2024). The iso-fidelity break-even $F_\text{log}(p, ...) = F_\text{phys}(p, ...)$7 is the unique $F_\text{log}(p, ...) = F_\text{phys}(p, ...)$8 where these frontiers cross:

$F_\text{log}(p, ...) = F_\text{phys}(p, ...)$9

• Quantum Memory/Decay: Lifetime fits obey $F_\text{phys}$0, and break-even times $F_\text{phys}$1 solve $F_\text{phys}$2 (Ni et al., 2022, Joshi et al., 4 Mar 2026).
• Surface Code Logical Error: $F_\text{phys}$3, with the iso-fidelity threshold for code distance $F_\text{phys}$4 given by $F_\text{phys}$5, $F_\text{phys}$6 being the physical error rate reference for the best unencoded circuit (Hao et al., 26 Nov 2025).

3. Experimental Protocols and Methodologies

Break-even iso-fidelity points are operationally characterized by parallel benchmarking of the unencoded and encoded/multi-fidelity process under matched resource and control conditions:

• Quantum Experiments: Simultaneous measurement of $F_\text{phys}$7 and $F_\text{phys}$8 under identical physical error rates, circuit depths, and postselection thresholds across primitives such as SPAM, memory, gate operations, entanglement generation (e.g. GHZ states), and application-level circuits (e.g. simulation of quantum magnetism) (Dasu et al., 25 Feb 2026, Riffel et al., 14 Apr 2026, Hong et al., 2024, Gupta et al., 2023, Ni et al., 2022, Joshi et al., 4 Mar 2026).
• Multifidelity Surrogate Modeling: MLGP vs. SL designs are compared by empirical or analytically derived error-vs.-cost curves, and the location of the crossing (iso-fidelity) is extracted by plotting or solving for budget/equivalent accuracy (Chen et al., 2024, Renganathan et al., 2022).
• Simulation Validation: In safety-critical system validation, iso-fidelity arises from equalized cost-to-error (sample complexity scaling) across simulator fidelities (Baheri et al., 2023).

Representative data protocols:

• Independent measurement of physical and logical fidelities per benchmark as a function of error rate, depth, or cost.
• Extraction of code/gadget/detection overheads, such as gate count expansion, postselection acceptance ratio (AR), or syndrome extraction resources.
• Reporting of crossing points (physical error rate, circuit depth, or simulation cost) where $F_\text{phys}$9 surpasses $F_\text{log}$0, sometimes at >$F_\text{log}$1 standard deviation or higher statistical confidence.

4. Empirical Results: Iso-Fidelity Crossing and Regime Maps

Experimental and simulation results consistently exhibit distinct break-even or iso-fidelity points, contingent upon circuit complexity, code distance, and resource overhead:

• Trapped-Ion Quantum Codes: At typical two-qubit error rates $F_\text{log}$2–$F_\text{log}$3, $F_\text{log}$4 and $F_\text{log}$5 iceberg codes achieve beyond break-even performance across SPAM ($F_\text{log}$6 at $F_\text{log}$7), memory cycles ($F_\text{log}$8), logical gates ($F_\text{log}$9), and large-scale GHZ states (iso-fidelity crossing at $$F_\text{phys} = 1 - p_\text{phys}, \quad F_\text{log} = 1 - p_\text{log}$$0) (Dasu et al., 25 Feb 2026).
• Surface Code Compilation: For QAOA and QPE on grid architectures, algorithmic break-even is achieved for $$F_\text{phys} = 1 - p_\text{phys}, \quad F_\text{log} = 1 - p_\text{log}$$1 ($$F_\text{phys} = 1 - p_\text{phys}, \quad F_\text{log} = 1 - p_\text{log}$$2) at $$F_\text{phys} = 1 - p_\text{phys}, \quad F_\text{log} = 1 - p_\text{log}$$3, or for $$F_\text{phys} = 1 - p_\text{phys}, \quad F_\text{log} = 1 - p_\text{log}$$4 ($$F_\text{phys} = 1 - p_\text{phys}, \quad F_\text{log} = 1 - p_\text{log}$$5) at $$F_\text{phys} = 1 - p_\text{phys}, \quad F_\text{log} = 1 - p_\text{log}$$6 (Hao et al., 26 Nov 2025).
• Nonlocal LDPC Codes: A $$F_\text{phys} = 1 - p_\text{phys}, \quad F_\text{log} = 1 - p_\text{log}$$7 code yields $$F_\text{phys} = 1 - p_\text{phys}, \quad F_\text{log} = 1 - p_\text{log}$$8–$$F_\text{phys} = 1 - p_\text{phys}, \quad F_\text{log} = 1 - p_\text{log}$$9 vs. $p_\text{log}$0–$p_\text{log}$1, with break-even statistically confirmed at $p_\text{log}$2 (Hong et al., 2024).
• Adaptive Magic State Encoding: Four-qubit error-detecting circuits achieve $p_\text{log}$3–$p_\text{log}$4, below the best physical $p_\text{log}$5 (Gupta et al., 2023).
• Multifidelity Surrogates: Break-even cost in MLGP vs. SL strategies is observed at substantially lower resources for the former, with iso-fidelity points determined for prescribed $p_\text{log}$6 across synthetic and real tasks (e.g., gas turbine blade, wing simulations, $p_\text{log}$7 for small $p_\text{log}$8) (Chen et al., 2024, Renganathan et al., 2022).
• Quantum Memory Lifetime: Discrete-variable QEC in cavity QED achieves $p_\text{log}$9 enhancement in process-fidelity lifetime over bare qubits, with explicit break-even times derivable from exponential fit parameters ($$C_\text{multi-fid}(\epsilon) = C_\text{single-fid}(\epsilon) \Rightarrow \epsilon = \epsilon^*$$0s for two-layer QEC) (Ni et al., 2022, Joshi et al., 4 Mar 2026).

5. Overhead, Concatenation, and Practical Regimes

Resource overhead, code concatenation, and postselection acceptance rates critically affect iso-fidelity trade-offs:

• Postselection and Concatenation: For distance-2 codes, postselection acceptance AR$$C_\text{multi-fid}(\epsilon) = C_\text{single-fid}(\epsilon) \Rightarrow \epsilon = \epsilon^*$$1 after 8 cycles, while distance-4 achieves AR$$C_\text{multi-fid}(\epsilon) = C_\text{single-fid}(\epsilon) \Rightarrow \epsilon = \epsilon^*$$2. Logical error rate and acceptance both scale exponentially with code distance: $$C_\text{multi-fid}(\epsilon) = C_\text{single-fid}(\epsilon) \Rightarrow \epsilon = \epsilon^*$$3 (d=2), $$C_\text{multi-fid}(\epsilon) = C_\text{single-fid}(\epsilon) \Rightarrow \epsilon = \epsilon^*$$4 (d=4); thus, concatenation shifts break-even to lower physical errors (Dasu et al., 25 Feb 2026).
• Gate/Gadget Overhead: Each increase in code distance or gadget FT-ness adds substantial circuit depth and gate-count. Overhead minimization (e.g. exploiting transversal gates for H, fold-swap automorphisms for nonlocal codes) is essential for net iso-fidelity gain, especially in intermediate-scale NISQ circuits (Riffel et al., 14 Apr 2026, Hong et al., 2024).
• Simulation Overhead: In multifidelity simulation, allocation of samples across fidelity levels according to closed-form allocation laws (e.g., Equation 3.11 of (Chen et al., 2024)) yields iso-fidelity cost minima, while in safety validation the per-sample cost $$C_\text{multi-fid}(\epsilon) = C_\text{single-fid}(\epsilon) \Rightarrow \epsilon = \epsilon^*$$5 and achievable accuracy $$C_\text{multi-fid}(\epsilon) = C_\text{single-fid}(\epsilon) \Rightarrow \epsilon = \epsilon^*$$6 determine the optimal fidelity or mix (Baheri et al., 2023).

6. Impact and Implications for Fault-Tolerance and Efficient Experimentation

Break-even iso-fidelity analysis provides rigorous, architecture- and problem-specific certification that advanced methods (error correction, multifidelity design, quantum circuit compiling) quantitatively outperform baseline or single-fidelity approaches:

• Quantum Computation: Demonstrated iso-fidelity crossing marks the transition to "beyond break-even" quantum error correction—for both memory and logic—on near-term devices; code and hardware co-design, including nonlocality and high connectivity, is pivotal for achieving these gains at scale (Dasu et al., 25 Feb 2026, Hong et al., 2024, Gupta et al., 2023, Ni et al., 2022).
• Simulation and Optimization: Iso-fidelity-based allocation and budget splitting enable rigorous, provably optimal use of expensive high-fidelity simulators in engineering, safety, and reliability; mixture or hierarchical approaches can be strictly preferable for tight error tolerances (Chen et al., 2024, Renganathan et al., 2022, Baheri et al., 2023).
• Guidelines for Practice: Minimizing gate expansion, matching code parameters to native error profiles, scheduling postselection and syndrome extraction to optimize AR, and explicit reporting of all relevant fidelities and overheads are requirements for transparent break-even analysis (Riffel et al., 14 Apr 2026, Dasu et al., 25 Feb 2026).
• Contingency on Error Models: All scalings and empirical crossing points depend on the validity of stochastic and uncorrelated noise models; correlated errors or drift may shift iso-fidelity thresholds in practice.

Summary Table: Typical Break-Even Benchmarks

Domain	Iso-Fidelity Comparison	Break-Even Regime
Trapped-ion QC (high-rate QEC)	Encoded vs. physical SPAM/gate/memory	$$C_\text{multi-fid}(\epsilon) = C_\text{single-fid}(\epsilon) \Rightarrow \epsilon = \epsilon^*$$7–$$C_\text{multi-fid}(\epsilon) = C_\text{single-fid}(\epsilon) \Rightarrow \epsilon = \epsilon^*$$8
Nonlocal LDPC code GHZ	Logical GHZ vs. physical GHZ fidelity	$$C_\text{multi-fid}(\epsilon) = C_\text{single-fid}(\epsilon) \Rightarrow \epsilon = \epsilon^*$$9 by $\epsilon^*$0
Surface code QAOA/QPE	FT ($\epsilon^*$1=9–11) vs. physical, $\epsilon^*$2	$\epsilon^*$3–$\epsilon^*$4, 1,700–2,500 qubits
Magic-state encoding	Encoded $\epsilon^*$5 vs. best pair	$\epsilon^*$6
Multifidelity surrogate	MLGP cost vs. SL cost at fixed $\epsilon^*$7	$\epsilon^*$8
Quantum memory/QEC	Logical vs. physical lifetime	$\epsilon^*$9 enhancement, $[[n,k,d]]$0s

Break-even iso-fidelity analysis has become a central benchmark across quantum computation, quantum simulation, multifidelity surrogate modeling, and experimental validation. Its predictive and empirical realization governs both present-day scaling demonstrations and the ultimate feasibility of efficient, fault-tolerant, or cost-effective implementation. Continued advances depend on optimization of code parameters, error model calibration, resource allocation, and architecture–algorithm co-design (Dasu et al., 25 Feb 2026, Chen et al., 2024, Riffel et al., 14 Apr 2026, Hong et al., 2024, Gupta et al., 2023, Ni et al., 2022, Hao et al., 26 Nov 2025, Joshi et al., 4 Mar 2026, Renganathan et al., 2022, Baheri et al., 2023).