Extended validations on photon number resolving detector based Gaussian boson sampling with low noises

Basic Information

• Paper ID: 2510.06300
• Title: Extended validations on photon number resolving detector based Gaussian boson sampling with low noises
• Authors: Yang Ji, Yongzheng Wu, Shi Wang, Jie Hou, Zijian Wang, Bo Jiang
• Classification: quant-ph (Quantum Physics)
• Institutions: The 32nd Research Institute of China Electronics Technology Group Corporation, Shanghai Quantum Science Research Center
• Paper Link: https://arxiv.org/abs/2510.06300

Abstract

Gaussian boson sampling (GBS) is a variant of boson sampling that overcomes the difficulties of the latter in stable single-photon preparation. However, like the original version, noise in GBS leads to deviations in output patterns and reduced classical simulation complexity. This paper extends the pattern recognition validation method and combines it with correlation-based methods for comparison in photon number resolving detector-based GBS, addressing photon loss and distinguishability noise to quantitatively assess noise levels. For the noisy classical simulation used in validation, which is actually a mixed-state simulation, existing photon-pair strategies are employed to achieve local polynomial acceleration. Additionally, an output binning strategy is implemented to accelerate validation. Simulation results demonstrate that the pattern recognition protocol is robust in GBS noise assessment even under sufficiently low noise conditions.

Research Background and Motivation

Problem Description

1. Challenges in Boson Sampling: Original Aaronson-Arkhipov boson sampling (AABS) is difficult to scale due to the lack of indistinguishable single-photon sources
2. Noise Issues in GBS: Although GBS using squeezed vacuum states as input sources is easier to prepare, physical noise (such as photon loss, photon distinguishability, dark counts) affects claims of quantum computational advantage
3. Limitations of Existing Validation Methods: Current validation methods face accuracy challenges at low noise levels and require more sensitive and accurate assessment approaches

Research Significance

• Quantum Computational Advantage Verification: Accurate noise level assessment is crucial for verifying quantum computational advantage
• Experimental Guidance: Provides quantitative noise assessment tools for practical GBS experiments
• Theoretical Refinement: Deepens understanding of noise effects on GBS output data structure

Limitations of Existing Methods

• Bayesian Methods: Require extensive hafnian computations, applicable only to small-scale experiments
• Low-Order Correlation Methods: Insensitive to high-order multi-photon interference, potentially unable to accurately handle photon distinguishability noise
• Output Binning Methods: Unclear accuracy when noise is sufficiently low but still amenable to efficient classical simulation

Core Contributions

1. Extended Pattern Recognition Validation: Extends pattern recognition techniques to validation of noisy GBS, particularly targeting photon loss and distinguishability noise
2. Precise Noise Simulation: Implements exact simulation of noisy GBS using chain rule methods with photon-pair strategies
3. Output Binning Strategy: Proposes output binning strategy to improve validation efficiency and overcome sample inefficiency
4. Data Structure Analysis: Provides in-depth analysis of noise effects on intrinsic data structure of GBS output, including probability distribution and norm distance imbalance
5. Quantitative Noise Relationships: Establishes quantitative relationships between validation results and noise levels, maintaining sensitivity even under extremely low noise conditions

Methodology Details

Task Definition

Input: Noisy GBS system parameters (squeezing parameter r, number of modes m, number of input SMSS K, interferometer matrix T) Output: Quantitative assessment of noise levels (transmission rate ηt or indistinguishability probability ηind) Constraints: Maintain sensitivity and accuracy of validation method under low noise conditions

GBS Fundamental Theory

For ideal GBS, the probability of output pattern s is:

pr(s) = haf(A_s) / (s₁!...sₘ!√det(Q_out))

Where:

• A_s is the submatrix of core matrix A
• Q_out is the covariance matrix of output state
• haf(·) denotes the hafnian function

Noise Models

Photon Loss Model

Under balanced loss conditions, the output state covariance matrix is:

V_loss = η_t V₀ + (1 - η_t)I_{2m}

The output probability of lossy GBS is calculated through ideal probability:

pr_loss(s) = Σ_{s'} {pr_ideal(s') ∏ᵢ [(s'ᵢ choose sᵢ) η_t^{sᵢ}(1-η_t)^{s'ᵢ-sᵢ}]}

Photon Distinguishability Model

Output probability with partially distinguishable photons is:

pr_pd(s) = Σ_{s⁽⁰⁾,...,s⁽ᴷ⁾} [pr_actual(s⁽⁰⁾) ∏ᵢ pr_virtual(s⁽ⁱ⁾)]

Where s⁽⁰⁾ + ... + s⁽ᴷ⁾ = s, containing one actual part and K virtual parts.

Exact Simulation Methods

Chain Rule Method

Implements simplification from heterogeneous measurements using auxiliary variables α = (α₂, ..., αₘ):

pr(s) ∝ ∏ᵢ pr(sᵢ|s₁,...,sᵢ₋₁,αᵢ₊₁,...,αₘ)

Photon-Pair Strategy

Employs greedy algorithm to rearrange photons in PNRDs into pairs, achieving acceleration through:

lhaf_mix(C, ν̄, n) = (1/2^{N/2}(N/2)!) Σ_z {...}

Pattern Recognition Validation

Cluster Construction

• Constructs k clusters using K-means++ algorithm
• Establishes feature clusters through training on real samples
• Evaluates cluster quality through imbalance in sample number distribution

Validation Metrics

Computes test value χ²:

χ² = Σᵢ Σⱼ [(Nᵢⱼ - Eᵢⱼ)²/Eᵢⱼ]

Where Eᵢⱼ = NᵢNⱼ/k, establishing linear relationship between Gaussian peak center Xc and noise level.

Output Binning Strategy

Groups output photons by subsets for measurement, reducing Hilbert space dimension from (n_cutoff + 1)^m to ∏ᵢ(m_{sub,i}n_cutoff + 1), significantly improving validation efficiency.

Experimental Setup

Simulation Parameters

• Small-scale case: r = 0.5, K = m = 5, n_cutoff = 4
• Large-scale case: r = 0.2, K = m = 10, n_cutoff = 3
• Sample numbers: 10⁴ samples for small-scale, 10⁵ for large-scale

Validation Parameters

• Number of clusters: k = 150 for small-scale, k = 700 for large-scale
• Training samples: 3000 for small-scale, 2×10⁴ for large-scale
• Noise range: η_t ≥ 0.9, η_ind ≥ 0.9 (low noise conditions)

Comparison Methods

1. Correlation-based Validation: Uses 1st to 4th order correlation functions for comparison
2. Other Mockups: GBS variants with thermal, coherent, and squeezed state inputs

Experimental Results

Main Results

Photon Loss Validation

• Within η_t ≥ 0.9 range, Gaussian peak center Xc shows near-linear relationship with loss rate
• Small-scale case displays clear monotonic increasing trend
• Large-scale case maintains same validation effectiveness

Photon Distinguishability Validation

• η_ind exhibits curved relationship with validation results, with tail rise suggesting importance of high-order multi-photon interference
• Pattern recognition method is sensitive to photon distinguishability noise, particularly near indistinguishability

Output Binning Effects

• Significantly reduces parameter requirements: k reduced from 700 to 100, samples from 10⁵ to 10⁴
• Maintains monotonicity of validation results without disrupting data structure imbalance

Data Structure Analysis

Probability Distribution Imbalance

• Sum of high-probability patterns varies monotonically with noise level
• Both photon loss and distinguishability suppress output imbalance

Norm Distance Analysis

• Average 2-norm distance L̄₂ exhibits linear relationship with noise level
• Probabilities of short-distance and long-distance patterns increase with noise

Comparison with Correlation Methods

• Higher-order correlation functions are more sensitive to noise
• Computational complexity grows rapidly with order and pattern number
• Sensitivity at low noise is inferior to pattern recognition method

Related Work

GBS Validation Methods

1. Bayesian Methods: Universal but computationally intensive, suitable for small-scale systems
2. Correlation Methods: Computationally efficient but insensitive to high-order interference
3. Graph-Theoretic Methods: Effective when output photon number is much less than mode number
4. Output Binning Methods: Provide semi-quantitative assessment with polynomial-time computation

Noise Model Research

• Matrix product state approximation of photon loss models
• Equivalent descriptions of partially distinguishable photons
• Classical simulation methods for mixed states

Conclusions and Discussion

Main Conclusions

1. Method Validity: Pattern recognition validation maintains robustness and sensitivity in low-noise GBS
2. Data Structure Insights: Noise affects GBS output by suppressing imbalance caused by multi-photon interference
3. Practical Value: Provides practical tools for quantitative noise assessment applicable to real GBS experiments

Limitations

1. Sample Efficiency: Pattern recognition method depends on Hilbert space dimension, with relatively low sample efficiency
2. Computational Complexity: Exact simulation still requires exponential time, limiting system scale
3. Noise Types: Primarily focuses on photon loss and distinguishability, not addressing other noise sources like dark counts

Future Directions

1. More Efficient Algorithms: Develop more efficient validation algorithms for larger-scale systems
2. Multiple Noise Sources: Extend to comprehensive models including multiple noise sources
3. Experimental Verification: Validate theoretical predictions in actual GBS experiments

In-Depth Evaluation

Strengths

1. Theoretical Rigor: Based on rigorous mathematical framework providing precise noise simulation
2. Methodological Innovation: Successfully extends pattern recognition techniques to GBS noise verification
3. Practical Value: Output binning strategy effectively addresses computational efficiency issues
4. Deep Analysis: Provides profound insights into intrinsic mechanisms of data structure

Weaknesses

1. Scale Limitations: Computational complexity constraints make handling truly large-scale GBS systems difficult
2. Noise Range: Primarily validated in high transmission rate/high indistinguishability ranges, requiring more exploration at extremely low noise
3. Missing Experimental Verification: Lacks comparative validation with actual GBS experiments

Impact

1. Theoretical Contribution: Provides new theoretical tools for GBS noise assessment
2. Experimental Guidance: Offers quantitative standards for practical quantum advantage verification
3. Method Generalization: Validation methods potentially applicable to other quantum computing models

Applicable Scenarios

• Noise assessment in small to medium-scale GBS experiments
• Verification of quantum computational advantage claims
• Guidance for GBS experimental parameter optimization
• Benchmark testing for quantum noise theory research

References

The paper cites 49 related references covering boson sampling theory, GBS implementation, noise models, validation methods, and other aspects, providing solid theoretical foundation for the research.