Evaluating Noises of Boson Sampling with Statistical Benchmark Methods

Basic Information

• Paper ID: 2510.00056
• Title: Evaluating noises of boson sampling with statistical benchmark methods
• Authors: Yang Ji, Yongjin Ye, Qiao Wang, Shi Wang, Jie Hou, Yongzheng Wu, Zijian Wang, Bo Jiang
• Classification: quant-ph (Quantum Physics)
• Institutions: Shanghai Research Center for Quantum Sciences; The 32nd Research Institute of China Electronics Technology Group Corporation
• Paper Link: https://arxiv.org/abs/2510.00056

Abstract

Boson sampling lacks inherent error correction codes, which impedes its development toward large-scale and robust implementations. Therefore, understanding noise levels is crucial for carefully demonstrating quantum computational advantage or achieving specific tasks. Based on statistical benchmark methods originally developed to distinguish boson sampling from other simulations (such as correlators and cloud methods), this paper quantitatively evaluates photon partial distinguishability and photon loss noise compensated by dark counts. This approach is feasible because the imbalance in output distributions is suppressed by noise, which is actually a consequence of multi-photon interference. This explains why better performance is achieved when using higher-order correlators or corresponding cloud methods. The results demonstrate that statistical benchmark methods can also be employed for noise evaluation in boson sampling.

Research Background and Motivation

Problem Background

1. Noise Challenges in Boson Sampling: As an important candidate for optical quantum computing, boson sampling lacks inherent error correction codes, making it susceptible to realistic noise in large-scale implementations
2. Reliability of Quantum Advantage: Quantum computational advantage in boson sampling can only be guaranteed under extremely low noise conditions; otherwise, output deviations completely destroy multi-photon interference pathways
3. Importance of Noise Evaluation: Accurate understanding of noise levels is necessary to determine whether results originate from quantum processes

Primary Noise Types

• Photon Partial Distinguishability: Arising from intrinsic variations in single-photon source preparation (such as position and frequency differences)
• Photon Loss: Due to imperfections in practical devices (such as linear interferometers)
• Dark Counts: Dark counts occurring in detectors may mask photon loss

Limitations of Existing Methods

• Low computational efficiency in noise evaluation through calculating total variation distance across all output modes
• Existing verification methods primarily provide binary judgments, lacking quantitative noise assessment capabilities

Core Contributions

1. Extension of Statistical Benchmark Methods: Extends correlator and cloud methods, originally designed to distinguish boson sampling from other simulations, for quantitative noise evaluation
2. Multi-type Noise Assessment: Systematically evaluates noise types including photon partial distinguishability, photon loss, and dark count compensation
3. Advantages of Higher-Order Correlators: Demonstrates superior performance of higher-order correlators in noise evaluation due to their enhanced sensitivity to multi-photon interference
4. Computationally Efficient Assessment Scheme: Provides noise evaluation methods that do not require learning interferometer matrix information

Methodology Details

Boson Sampling Fundamentals

In ideal boson sampling, the probability of output mode $|T\rangle = |T_1, T_2, \ldots, T_m\rangle$ is determined by:

$P_{\text{ideal}}(T) = \frac{|\text{Perm}(\mathbf{M}_{S,T})|^2}{\prod_{i=1}^m S_i! \prod_{i=1}^m T_i!}$

where $\mathbf{M}_{S,T}$ is a submatrix selected according to input and output states, and $\text{Perm}(\cdot)$ denotes the matrix permanent.

Noise Modeling

1. Photon Partial Distinguishability

Using tensor collection to gather pairwise noise information, the output probability is:

$P_{\text{rpd}}(T) = \frac{\sum_\sigma (\sigma \prod_{j=1}^n a_{\sigma_j j}) \text{Perm}(\mathbf{M}_{S,T} \odot \mathbf{M}_{1,\sigma}^{S,T*})}{\prod_{i=1}^m S_i! \prod_{i=1}^m T_i!}$

where $a_{ij} = x_{\text{ind}} + (1-x_{\text{ind}})\delta_{ij}$, and $x_{\text{ind}}$ is the distinguishability parameter.

Virtual Beam Splitter Model: Partial distinguishability is simulated by introducing a virtual beam splitter, where the reflectivity relates to $x_{\text{ind}}$ as: $\cos^2\omega = x_{\text{ind}}$

2. Photon Loss and Dark Counts

• Balanced Loss Model: Constant photon transmission rate $\eta_t$ for each mode
• Dark Counts: Dark count probability $p_{\text{dc}}$ for each detector
• Combined Noise: In experiments, $1-\eta_t = p_{\text{loss}} = p_{\text{dc}} = p_{\text{noise}}$

Statistical Benchmark Methods

Correlator Method

The $t$-th order correlator is defined as:

$\kappa(n_{o_1}, \ldots, n_{o_t}) = \sum_\pi [(|\pi|-1)!(-1)^{|\pi|-1} \prod_{B \in \pi} \langle \prod_{i \in B} n_{o_i} \rangle]$

Evaluation Metrics

1. Ratio Parameter: $\gamma = \frac{\sum_i \kappa_{\text{test},i}}{\sum_i \kappa_{\text{comp},i}}$
2. Pearson and Spearman Correlation Coefficients
3. Cloud Method: Based on coefficient of variation (CV) and skewness coefficient (CS)

Experimental Setup

Simulation Parameters

• Number of photons $n$: 2-10
• Number of modes $m$: 10-20
• Sample size: $10^4$ - $10^6$
• Distinguishability parameter $x_{\text{ind}}$: 0-1
• Noise level $p_{\text{noise}}$: 0-0.5

Algorithm Implementation

Classical simulation using the Clifford-Clifford algorithm, extended through virtual optical networks to handle partial distinguishability noise.

Experimental Results

Photon Partial Distinguishability Noise Evaluation

Correlator Analysis

• Trend Observation: As $x_{\text{ind}}$ increases, correlator points gradually approach the $x=y$ line of the ideal case
• Higher-Order Advantage: Higher-order correlators (3rd, 4th order) show more pronounced variation trends compared to lower-order correlators (2nd order)
• Continuous Improvement: The $\gamma$ parameter provides more continuous assessment compared to Pearson and Spearman correlation coefficients

Scale Effects

• Photon Number Impact: Increasing $n$ or decreasing $m$ makes deviations between noisy and ideal samples more apparent
• Robustness: Correlator methods maintain robustness even in large $m$ scenarios

Cloud Method Results

• Sensitivity Differences: Average CV is more sensitive to high $x_{\text{ind}}$, while average CS is more sensitive to moderate $x_{\text{ind}}$
• No Matrix Information Required: Cloud methods can operate without learning interferometer matrix information

Photon Loss and Dark Count Noise Evaluation

Performance Comparison

• Low Noise Region: Assessment performance is poor with severe fluctuations when noise levels are very low
• High Noise Region: When noise levels are higher, $\gamma$ shows a declining trend, and CS and CV changes in cloud methods are continuous

Distribution Analysis

• Probability Deviation: Probability deviations become apparent only when $p_{\text{noise}}$ is relatively high
• Total Variation Distance: In low noise regions, enhancement of total variation distance is suppressed, explaining the poor performance of statistical methods

Related Work

Boson Sampling Verification Methods

1. Bayesian Methods: Bayesian verification methods proposed by Bentivegna et al.
2. Pattern Recognition: Pattern recognition techniques by Agresti et al.
3. Statistical Benchmarking: Correlator and cloud methods by Walschaers et al.

Noise Modeling Research

1. Classical Simulation Algorithms: Renema's truncated interferometer algorithm, Clifford-Clifford algorithm
2. Noise Thresholds: Theoretical studies on how noise affects quantum advantage

Conclusions and Discussion

Main Conclusions

1. Method Feasibility: Statistical benchmark methods can successfully quantitatively evaluate boson sampling noise
2. Higher-Order Advantage: Higher-order correlators are more sensitive to multi-photon interference, providing better noise evaluation performance
3. Noise Type Differences: Different noise types affect statistical characteristics through different mechanisms

Limitations

1. Low Noise Sensitivity: For photon loss and dark counts, assessment performance is poor in low noise regions
2. Sample Requirements: Cloud methods require large sample sizes to obtain reliable results
3. Extreme Condition Assumptions: The assumed condition $p_{\text{loss}} = p_{\text{dc}}$ in experiments is unrealistic in practice

Future Directions

1. Improve assessment methods for low noise regions
2. Extend to noise evaluation in other quantum computing systems
3. Develop more efficient statistical metrics

In-Depth Evaluation

Strengths

1. Innovative Extension: Innovatively extends verification methods into noise evaluation tools
2. Systematic Research: Comprehensively considers major noise types in boson sampling
3. Practical Value: Provides computationally efficient noise evaluation schemes without requiring additional experimental information
4. Theoretical Insights: Reveals the relationship between higher-order correlators and sensitivity to multi-photon interference

Weaknesses

1. Theoretical Analysis Depth: Lacks in-depth theoretical analysis of the sensitivity mechanisms of statistical methods
2. Experimental Validation: Based solely on numerical simulations, lacking actual experimental verification
3. Method Comparison: Insufficient systematic comparison with other noise evaluation methods
4. Applicable Scope: Insufficient discussion of parameter ranges for method applicability

Impact

1. Academic Contribution: Provides new perspectives for noise evaluation in quantum computing systems
2. Practical Value: Has direct application value for noise monitoring in boson sampling experiments
3. Method Generality: Statistical benchmark methods may be applicable to noise evaluation in other quantum systems

Application Scenarios

1. Boson Sampling Experiments: Real-time monitoring of noise levels in experiments
2. Quantum Advantage Verification: Determining the reliability of quantum computational advantage claims
3. Device Characterization: Evaluating performance and noise characteristics of optical devices

References

The paper cites 37 related references covering multiple aspects including boson sampling theory, noise modeling, verification methods, and classical simulation algorithms, providing a solid theoretical foundation for the research.